Date post: | 26-Feb-2023 |
Category: |
Documents |
Upload: | khangminh22 |
View: | 0 times |
Download: | 0 times |
te
XWX5-BR—
VOL I
ANAISDO SIMPÓSIO
LATINO-AMERICANODE FÍSICA DOS
SISTEMAS AMORFOS
EDITOR: ENRIQUE V. ANDA
CENTRO LATINO-AMERICANO DE FÍSICA — CLAFUNIVERSIDADE FEDERAL FLUMINENSE — UFF
ANAIS DO SIMPÓSIOLATWO-AMERICANO DE BSKA
DOS SISTEMAS AMORFOS
NITERÓI, 27 OE FEVEREIRO A 2 DE MARÇO DE 1984
EDITADO POR
ENRIQUE V. ANDAUNIVERSIDADE FEDERAL FLUMINENSENITERÓI - RJ
V O L U M E I
PUBLICAÇÃO DO CENTRO LATINO-AMERICANO DE FÍSICARIO DE JANEIRO - BRASIL
MAIO - 1985
SIMPÓSIO LATINO-AMERICANO DE FlSICA DOS SISTEMAS AMORFOS
NITERÓI - BRASIL, 27 DE FEVEREIRO A 2 DE MARÇO DE 1984
Organizadores
- Centro Latino-Americano de Física (CLAF)- Universidade Federal Fluminense (UFF)
PatrocinadoresCentro Latino-Americano de Física (CLAF)Universidade Federal Fluminense (UFF)Conselho Nacional de Desenvolvimento CientíficoTecnológico (CNPq)Fundação de Amparo â Pesquisa do Estado de SãoPaulo (FAPESP)Comissão de Aperfeiçoamento de Pessoal de NívelSuperior (CAPES)Sociedade Brasileira de Física (SBF)
Comissão Organizadora Nacional- Antonio Ferreira da Silva (INPE)- Francisco César de Sá Barreto (UFMG)- Frank Missel (USP)- Ivon Fittipaldi (UFPE)- Mario Baibich (UFRS)- Reiko Sato-Turtelli (UNICAMP)- Enrique V. Anda (UFF)- Mucio Amado Continentino (UFF)
Comissão Organizadora Local- Elisa Saitovich (CBPF)- Maria Augusta Davidovich (PUC)- Roberto Nicolsky (UFRJ)- Álvaro Ferraz Filho (UFF)- Fernando A. Oliveira (UFF)- Luiz Carlos S. do Nascimento (UFF)- Norberto Majlis (UFF)- Paulo Murilo C. de Oliveira (UFF)- Sergio S. Makler (UFF)- Silvia Selzer (UFF)
SUMÁRIO
V O L U M E I
INTRODUÇÃO i
CONFERÊNCIAS:
THEORY OF GIASS 1N. Rivier
PHASE TRANSITIONS IN GLASSY MATERIALS 92M. F. Thoipe
METALLIC GLASSES: ST; JCTURAL MODELS 110E. Nassif
AMORPHOUS MfcTAL? - FABRICATION AND CHARACTERIZATION . 128F. P. Missell
STRUCTURAL RELAXATION LOW TEMPERATURE PROPERTIES ... 144F. de la Ciuz
AMORPHOUS SUIERCONDUCTORS 161F. P. Missell
ON THE SUPERCONDUCTIVITY OF VANADIUM BASED ALLOYS ... 176F. Brouers ?nd J. Van der Rest
DENSIDADE DE ESTADOS ELETRÔNICOS EM VIDROS METÁLICOS. 186S. F. Pessoa
REAL SPACE RENORMALIZAfMN TECHNIQUES FOR DISORDEREDSYSTEMS 201
E. V. Anda
RESEARCH ON HYJ;ROGENATED AMORPHOUS SILICON AND RELATEDALLOYS FOR PHOT»')VOLTAIC APPLICATIONS 221
I. E. Chamboui'iyron
A GENERALIZED SLATER-PAUL]N3 CONSTRUCTION FOR MAGNETICAMORPHOUS ALLOTS 232A.P. Malozemoff, A.R. Wil/iams and V.L.Moruzzi
AMORPHOUS MAGrfETJfM '. 245H. R. RechenbfTg
REENTRANT FERROMAÒNETISM 258M. A. Continentino
INTRODUÇÃO
A realização do Simpósio Latino-Americano de Fí-sica dos Sistesas Aaorfos surgiu como conseqüência do interesse Manifestado por pesquisadores latino-americanos nessa área (eu crescente desenvolvimento na América Latina),eque já se constitui nua campo onde operam alguns gruposconsolidados (que trabalham ativamente) e outros que estãoem formação.
Os materiais anorfos apresentam uma rica fenomenologia ainda sem explicação satisfatória e possuem grandes implicações tecnológicas, o que representa um estímulopara cientistas e tecnõlogos. Frente a essa realidade, ecom o auspício do Centro Latino*Americano de Física, o Grupo de Física do Estado Sólido da Universidade Federal Fluainense (UFF) pensou na possibilidade de reunir os pesqiúsadores da América Latina engajados na área para facilitaro intercâmbio e a colaboração mais articulada e estreita entre os aesaos, além de permitir a realização de una avaliação das perspectivas do tema, tanto nos seus aspectos basi^cos como aplicados, dando um ênfase especial na integraçãoteõrico-experimental.
A oportunidade da realização do Simpósio nos pareceu particularmente apropriada como uma forma de promover o trabalho científico num momento político e econômicodifícil pelo qual atravessa a região, com manifestações ne_gativas nas atividades de pesquisa de nossos países, ativi^dades essas fundamentais para a resolução de muitos dosproblemas que a realidade latino-americana apresenta.
ii
O Simpósio teve lugar nas instalações do Institu
to de Química da Universidade Federal Fluminense, na cida
de de Niterói, durante o período de 27 de fevereiro a 2 de
•arco de 1984, e contou COB a participação de 144 pesquisa
dores. Entre as atividades do Simpósio, destacamos sessões
de painéis com apresentação de 56 trabalhos e a realização
de 9 palestras convidadas e 4 cursos sobre diferentes a£
pectos da área dos Sistcaas Anorfos Desordenados. Foram or
ganizados também dois grupos de trabalho sobre Semiconduto
res Amorfos e Metais Amorfos, o que permitiu fazer una ava
liação sobre os recursos humanos e de infraestrutura dispo
níveis na América Latina nessas duas áreas.
Nesta publicação incluem-se os cursos, as pales
trás e os trabalhos apresentados durante o Simpósio, cujos
textos foram enviados ao Comitê Organizador.
0 Simpósio foi realizado graças ao apoio finance^
ro do programa conjunto UNESCO/CLAF, do Conselho Nacional de
Desenvolvimento Científico e Tecnológico (CNPq), do UNIBAN
CO, da Universidade Federal Fluminense (hospedeira do even
to), da FAPESP, e de instituições latino-americanas que.co
brindo despesas de passagens, permitiram aumentar a part_i
cipação de colegas de outros países.
Gostaríamos de agradecer especialmente o apoio
fin .,.ceiro e institucional prestado pela Sociedade Brasilei^
ra de Física, bem como aos professores e funcionários do
Centro de Estudos Gerais e dos Institutos de Física e Qu£
nica da UFF, cujo trabalho e solidariedade tornaram possjf
vel a realização do Simpósio, colocando ã nossa disposição
a infraestrutura dessas unidades universitárias.
A secretaria administrativa esteve a cargo do
Vera Rosenthal e Humberto Teixeira, que muito contribui ram
para o êxito do evento.
THEORY OF GLASS
N. RttierCenter for Nonlinear SUdies
Los Alamos National LaboratoryLos Alamos, NM 87545, USA
Institute for Theori'ical PhysicsUniversity of California
Santa Barbara, CA 03106, USA
and (permanent address)
Blackett LaboraoryImperial College
London SW7 2BZ, Great Britain
OUTLINE
I Structure, from a topoáyg, ». VM. vpointD Gauge invarianceIII Tunneling modes
IV Supercooled liquid and glass transitionV Statistical crystallography
I - STRUCTURE OF GLASS FROM A TOPOLOGICAL VIEWPOINT
1.1 Introduction
Glass spans nearly 60 centuries of human activity, from early glazed objects
(4000 BC) and glassy beads which a p p e a l in Egypt between 3200 and 2500
yean BC (depending on the encyclopedia), to metallic glasses, discovered and
developed in the (19)sixties and seventies. Yet, from a condensed matter
physicist's poiBt of view, it b an ill-understood material, fall of fascinating and
surprising physical properties.
Tk* Methodology of condensed matter physics consists in identifying, and relat-
ing to taen other, the physical properties, structure, and constitutive dements of
a daw of materials. Glasses hare several physical properties which are both
universal and specific to disordered condensed matter (cf section 1.2), but
emphasis on ', and demonstration of (see eg. *) their universality have only
recently been made. To relate these properties to the structural constituents
requires an understanding of the structure of glass, which, at Erst sight, xtnsists
almost exclusively of negative statements: no metric geometry, trivial space
group, no generative symmetry, no Bloch states, no single ground state, no
Unique best structure, etc.
By contrast, perfect crystalline materials have easily discernible structure (space
group) and constituents (atoms, electrons, holes, unit cell). However, rare are the
physical properties of perfect crystals which depend specifically on their space
group. But imperfect crystals have properties directly and crucially affected by
extended structural constituents, or defects (vortices, dislocations, flux lines),
whose definition, label and existence (structural stability) is granted by the struc-
ture of the material3. We shall see that the physical properties of glasses are also
governed by a single, extended constituent, the odd line or 2T-disclinatR>n \
which is the only structural element surviving the absence of generative homo-
geneity and the triviality of the space group.
The main purpose of this lecture is to identify those universal properties and
structural constituents, both at the lowest level of specificity. The level of
understanding of glass is similar to that of crystalline solid state physics before
1910-301, Le. before quantum statistics, Bloch theorem, electron bands and dislo-
cations were introduced. General concepts are required, howrrer ovetsimplifi:..*
rather than the solution of a particular problem from fist principles. It wiD tun
out that even such a simplified description yields non-trivial results.
1.2
We shall list those properties which occur so widely in glasses that they can be
regarded as universal, and so rarely or never in crystals that they can be viewed
as specif c to disordered condensed matter.
a) Tunneling modes
At low temperatures, the glass is an elastic solid with overall homogeneity,
capable of supporting phonons of wavelength long enough to average over any
«homogeneity. Glass can ring, as Mozart, and probably others found out long
ago. Surprisingly, this is not all. At low temperatures, the specific heat is linear in
temperature, and this contribution dominates the T* phonon contribution below
IK. The thermal conductivity is lower than that of the corresponding crystalline
material (eg. quartz) and goes as 7*. The specific heat betrays the presence of
additional localized elementary excitations (MO per 10* atoms) , which can
absorb phonons and thereby reduce the thermal conductivity K It was suggested
by Anderson, Halperin and Varroa, and independently by W.A. Phillips, that
these additional excitations were tunneling modes between potential minima, or
valleys, distributed at random in configuration space *. These valleys are few
and far away m configuration space, so that tunneling occurs between pain,
thetuby forming two-level systems. The main evidence for tunneling, and for *
Inite nnmber of levels, comes from the fact that the ultrasonic absorption can he
•afurufeé'. Abo, echo spectroscopy can be performed, exactly as for an assembly
of spins 1/2 7.
The concept of tunneling modes has been entirely successful both m explaining
existing experiments, and in suggesting new ones. Tunneling modes occur m
Detalbc and covakut glasses •. But, what is H that mores, let alone tunnel ? How
are we to label the valleys in configuration space and determine their distances ?
(The height of the saddle points between valleys sfaouM be of the order of the
glass transition temperature). Answers in lecture ID.
b) Viscosity, relaxation rates, entropy and the Kauzmann paradox
Traditionally, glass transition b defined to coincide with the change in the
thermal expansion coefficient from a value characteristic of a liquid to that of a
solid. This corresponds to a viscosity of the order of 10u poises. The transition is
smooth and its temperature Tf depends on the cooling rate and on the thermal
history of the system. The viscosity (or any inverse transport relaxation rate)
increases smoothly and rapidly with decreasing temperature, and follows the
empirical Vogel-Fulcher formula11W
over n wide range of temperatures. Equation (1.1) is used by the National Bureau
of Standards for calibrating reference glasses *. The various names under which it
it known in different fields (WLF, Cohen-Turnbull, Tammann, Doolittle, etc., for
glasses, polymers, oik, etc.), should support its claim to universality, even
though, for some glasses, inite viscosities have been measured below T0, and the
viscosity seer» to cross over from (1.1) to Arrhenins behaviour as the tempera-
ture B lowered *. Thus, roughly, glass behaves like a supercooled Said above Tm,
and an elastic, random solid below T0. T, B well above room temperaUre for
window glass (a-SiOs), whereas T.=\V \K for glyeerol. Slow modes in the
sapertoolcsl fluid become frozen (qoencbed) below 7*#.
A measure of the entropy of a glass is obtained by integrating under the specifc
heat corre, with a known value for the entropy of the liquid as initial condition-
Its value depends on the cooling rate, but, if extrapolated to zero cooing rate,
the entropy becomes negligibly small at and below the same Enite temperature
Tf, at which the extrapolated viscosity (1.1) diverges '*. This vanishing entropy
at a finite temperature is called Kaiizmaan's paradox, even though it was prob-
ably known to Simon. A good discussion of the meaning of entropy of glasses as
measured by this method, can be found in a recent paper by Jaeckle ".
Unlike ÇWP5, whose viscosity is proportional to the atomic diffusion rate and
increases with increasing temperature, viscous or supercooled fluids have fluidity
1/iy - ability to yield to a shearing stress - proportional to the mobility of n
diffusing object or "defect", i.e. to its diffusion rate />, by Einstein's relation. One
would expect D to be activated in condensed matter, and 1/n to decrease with
decreasing temperature as in eq. (1.1), but with 7*,=0 I2. The other problem »:
which "defect" ? Answers in IV.
c) Energy gap
Window glass is transparent, and amorphous semiconductors have a gap.
Despite the absence of long-range order and of Bragg reflection, the existence of a
gap was shown by Weaire a to be a consequence of the fixed valence in amor-
phous semiconductors, i.e. to the uniformity in vertex coordination z, or regular-
ity of the graph describing their structure (cf section 1.3). Furthermore, for a
simple but realistic model, Weaire and Thorpe M have shown that the energy
spectrum consists of degenerate states and of the spectrum of eigenvalues of the
connectivity, or adjacency matrix (a matrix describing the topology of the
graph). The gap is filled by localized defects (eg. dangling bonds) and modified by
many-body effects (polarons, excitons). The subject is treated in M.H. Gphen's
lectures.
d) Hall effect
The Hall effect is doubly anomalous in amorphous semiconductors: "...two ord*
era of magnitude less than expected..., almost invariably it has the wrong
sign" lb. The magnitude of the effect suggests interference of the electron wave
packet, due to the fact that the space in which it propagates is not simply con»
netted. The cores of line "defects" puncture the space available to the charge car*
riers.
Structure
See references 16 and 17 for details.
Glasses belong to either of two classes; covalent glasses (like window glasses or
â-Si) or random packings (like metallic glasses). In both cases, their structure
can be described by a regular graph.
ID covalent glasses (Pl.l), vertices and edges of the graph are atoms and bonds,
respectively, (non-planar) faces are rings, and cells do not have a direct physical
interpretation. The atoms have fixed valency, so that the graph has fixed vertex
coordination z ( z = 4 for Si), and is said to be regular. Dangling bonds are per-
mitted. This graph is also called a continuouê random network (CRN).
In metallic glasses, only vertices have direct physical interpretation (atoms), and
neighbourhood (edges) must be defined precisely. This is done by
Voronoi construction, a partition of space whereby any point belongs to the ter-
ritory of the nearest atom. Every atom is thus ascribed a (convex) polyhedron or
cell, and the packing is a space-filling assembly of such cells. The assembly of
Voronoi polyhedra, with their (planar) faces, (straight) edges, and vertices, form a
graph, called Voronoi froth, which is regular (z—4 in 3D and r = 3 in 2D)
because any vertex with higher coordination can be split into several regular ver-
tices by an infinitesimal deformation of the packing (through transformation Tl
of section 1.5). They are not topologically or structurally stable, and occur with
negligible probability. Thus the Four Corners boundary between Colorado, Utah,
Arizona and New-Mexico, is a cartographer's conspiracy and is not geographically
stable. Similarly, in 3D, only edges where 3 faces or cells meet are stucturally
stable. Structural stability is what Lewis 2 I calls "random avoidance of the
niceties of adjustement", and the structural elements (vertices, edges, etc.) are
simply territorial boundaries of a random packing. By contrast, CRN (covalent
glasses) have vertex coordination given by chemical, not territorial considerations,
and more than 3 (non-planar) faces can meet on & single edge, even though their
P|.J. <3o»tintto«» rsndon} network, exbibiiicg homogeneity, non-coliirtearity and odd line» (thethread thronsh odd riugs exclti»ive!y)
FI2. Mndcracking
vertices are still - in SiO2 or a-Si - all tetrarcoordinated. The crystalline analo-
gues to the Voronoi construction are the Wigner-Seitz cells, or, in reciprocal
space, the Brillouin zones.
Let us return to the original packing. Each atom corresponds to a Voronoi
polyhedron or cell. Two atoms are neighbours, and will be joined by an edge, if
their Voronoi polyhedra share a face. Thus, two graphs, related by duality,
describe the same packing, the original atomic packing, and the Voronoi froth.
Only the Voronoi froth b regular. The packing has high, fluctuating vertex coor-
dination ( < z * > « 13.4, increasing with anisotropy of the Voronoi polyhedra,
and decreasing if the polyhedra have different volumes 18, as happens in real
metallic glasses made of at least two different elements). The self-dual condition
is 2 = z * = 6 in 3D and 4 in 2D (cubic and square lattices, respectively).
The essential feature is tLe considerable variety of shapes of Voronoi polyhedra.
Matzke w has identified 100 different kinds in soap bubble froth, out of which 20
occur frequently. Thus, there is no single unit cell in glasses, not even smooth
deformations thereof. The 20 Voronoi polyhedra are all topologically different,
and the structure of one metallic glass is a member of a statistical ensemble (of
most probable distributions - see lecture V). Moreover, glasses belong to a
different class of packings than the aperiodic tilings of Robinson and Penrose,
which have only a few elementary geometric tiles 20. The situation has been sum-
marized in 1043 by the botanist F.T.Lewis: "The average 14-hedral shape
observed in massed bodies of diverse surface tension may be due to random
avoidance of the niceties of adjustment. Failure to arrange the bodies so that 5 or
6 meet at a mathematical point, or form intersection where 4 meet along a
mathematical line is sufficient to account for promiscuous, vnoritntotcd
10
pçlfkeárm having an average of 14 facets" 21. This failure, associated with struc-
tural stability, gives rise to a substantial variety of cell shapes, and to statistical
equilibrium.
Voronoi froths and continuous random networks are topologicalrjr equivalent:
they have the same, low vertex coordination z=4 , but CRN's have bent edges,
nonplanar faces , nonconvex cells and two faces sharing more than one common
edge, whereas Voronoi froths have staight edges, planar faces, convex cells, but
unequal edge lengths. The ring statistics is accordingly different: Voronoi froths
have a majority of 5-bond rings, CRN's 7-bond ringsn .
The Voronoi partition of space can be extended to packings of different atoms (as
befits real metallic glasses) and unequal cells. This is not as elementary as it
sounds: Consider 3 atoms in 2D; they form the vertices of a triangle, whose per-
pendicular bissectors are concurrent, by an elementary theorem of geometry. If
the atoms are unequal, which perpendicular division retains concurrence, and
thus space-filling ? There is only one known answer, the radical «rú, a straight
line (a plane in 3D) wich is the locus of points with equal tangents to two circles
(spheres) representing the atoms. It is obvious that radical axes are concurrent,
loaf so that they are straight lines, and the proof of this last statement** (an
elegant exercise in inversion geometry) strongly suggests that the radical froth is
the only generalization of the Voronoi froth to unequal atoms. It has been applied
to metallic glasses by Gellatly and Finney 2*.
The Voronoi construction has been used to describe not only glasses, but also
ecological and geographical problems, soap bubbles froths, metallurgical aggre-
gate», convective cells, etc.17.
11
1.4 The need for topology «ad its consequences
Manifestly, amorphous structures have neither global length scale (ruler), nor
nor unique global reference frame orientation (compass). In fact, amorphous or
non-crystalline materials are characterized, at first sight, by what they are or
have not: No generative symmetry (translation or rotation), trivial space- or
point groups, MO unique ground state see tunneling modes), no Bloch
theorem,..., so that non-negative concepts are required, which are structurally
stable under those transformations wr;;h keep tje structure unchanged.
At a glance, one distinguishes the three mzn structural features, not only of
glasses, but of all large, space-filling random structures (metallurgical aggregates,
undifferentiated biological tissues, geological jointings, etc.):
i) Non-collinearity of local reference frames (or variety of cell shapes)
ii) Overall, but non-generative homogeneity
iii) Odd lines.
These three features are characteristic of the amorphous state, and can be taken
as its stuctural definition. If either (i) or (iii) are missing, one may still have a
random structure, but without topologieal disorder.
Homogeneity implies that atoms are distinct, but not physically distinguishable.
Glasses share this property with crystals. However, unlike in crystals, this homo-
geneity is not a generative symmetry, but only the fact that any objective (physi-
cal) statement (one which does not contain T , "this", "here",...) about one par-
ticular atom can equally well be made about any other, even though their local
environment (tetrapods attached to the Si atoms in a-Si or vitreous silica, Voro-
noi cells in metallic glasses) are manifestly different and non-collinear. (See Pl.l).
r» MtNã. I
f\%\. Topologicai invariuice apd transformation, iüusliated by Deni» We»irt (based on a realevent). ©Dcnii Weair«r 1982
13
A CRN construction can be continued ai nt/tntt«m. This is the kind of homo-
geneity experienced by getting lost in & forest, with trees taking the part of
atoms. This homogeneity of glasses manifests itself at a microscopic level, to the
surprise of early X-ray crystallographers: "...one of the most interesting
discoveries made in the comparatively early history of X-ray analysis was the fact
that silk and even paper are more crystalline than glass." (Dame Kathleen Lons-
dale, quoted in ref.[25], p.26). Thus the scale of homogeneity in glass is 3 order of
magnitude smaller than the smallest mkrocrystallites (10 versus 1000 atomic dis-
tances).
The automorphisms probing homogeneity are permutations of the atoms (or the
trees) and their surroundings, effected, for example, by local rotations of the
tetrapods. The physical properties must be unchanged under these local transfor-
mations, and homogeneity of glasses is a genuine, gauge symmetry. We shall dis-
cuss gauge invariance in detail in lecture II. At the structural level, the most ele-
mentary such automorphisms are automorphisms of the graph, that is permuta-
tions of the vertices which preserve adjacency. At a dynamical level, local rota-
tions must also be included.
In the absence of a global metric and reference frame orientation, the only invari-
ant under automorphisms probing the homogeneity of amorphous structure is
their connectivity or the spatial relationship imparted by, eg., Voronoi construc-
tion. (We shall see in lecture V that some transformations can switch neighbours
(Tl), but still leave tile statistical properties of the structure invariant). Thus,
the relevant geometry for amorphous condensed matter is topology (rubber
geometry), which replaces metric geometry of claasical crystallography. By the
•ame token, the 230 metric space groups are replaced by homotopy groups which
14
describe connectivity, k. states and Bloch theorem give way to topological sectors
atd to a theorem, also formulated by Bloch for superconductors n, stating that
lhe free energy is a periodic function of the flux triggering the gauge transforma-
tion;. The period corresponds to a large, or non-trivial gauge transformation,
which brings the system from one topological sector to another. The best illustra-
tion of topological invariance is given in a cartoon by Denis Weaire (Fig.})-
Consequences of this enforced retreat from metric to topology are that
a) One cannot distinguish structural constituents by their sizes alone.
b) The existence, and the definition of a structural constituent depends on its
structural stability, so that it cannot be made to disappear by small, continuous
deformations.
e) Its only distinctive feature fa its shape. Hence randomness and topological
disorder imp/f that cells have many different shapes, like the various soap bub-
bles identified by Matzke w . Glasses cannot be made of only one type of cell, as a
result of "random avoidance of the niceties of adjustement" 21. The problem of
describing the structure is a statistical one, and the methods of statistical
mechanics yield the average features (its "equation of state", which is called
l/ewis's law in the case of cellular stuctures, see refs.[27],[57]f or lecture V) of the
ideal random, space-filling structure (the class of most probable members of an
m$mN« of structures). This equation of state is a correlation between sizes and
shape» of the constituting cells, and implies a medium-range correlation (one
should not say order) in the amorphous structure.
1.6 Elementary structural transformations
There are three, locai, elementary transformations of a random structure
under small, continuous deformations 17>28.
i) Tl , or neighbour-switching. (Fig.2).
ii) T2, or cell disappearance. (Fig.3).
if) Mitosis, or cell divbion. (Fig.4).
All of which occur in 2D and in 3D. In addition, in 3D, one has
iii) Face disappearance.
All these processes (or their inverses) occur in froths, foams or emulsions, in rock
or mud crackings, and in convection cells. Voronoi froths have a conserved
number of cells, and do not accomodate (ii) or (ii'). Covalent networks (CRN)
only have Tl transformations, which is there a local valence or bend exchange.
(The term "valence alternation" is now used to describe a particular class of
non-local transformations, accompanied by charge transfer and change in vertex
coordination, which control the electronic and optical properties of amorphous
chalcogenides n). The T l or neighbour-switching process is also that which res-
tricts topological stability to 3- (in 2D) and 4-coordinated vertices (in 3D), only.
Note that it conserves the vertex coordination of the froth, but not that of the
dual graph. In this respect, it affects adjacency, but we shall see in lecture V that
the statistical properties of the structure remain invariant.
There has been several recent attempts to construct random networks from by
applying successive, random T l transformations on an initially regular (hexago-
nal or diamond) lattice in 2D " " o r 3 D " .
I-is. 2 1 lt-;ikin.ir\ i.K.i| KMi
TI process
incnl of ivlk
T2 process
X. Vanishing of ;i cell.
mitosis
-I (VII division.
l-'tj!. *. A s ? p;iir nf ivIK. m.ikmir up ;i
These transformations change the number of edges of the faces involved. In 3D,
this corresponds to the introduction of topologiral disclinations (rotation disloca-
tions). Hexagons tile the plane (or the floor of a kitchen). Removing or adding
one edge to one hexagon can be done by making a cut in the tiling, removing or
adding a wedge (segment) of material and reglueing. One recovers a perfect hex-
agonal tiling apart from the cell at the end of the cut which is now pentagonal or
heptagonal, and the fact that the tiling is now warped. It has positive, or nega-
tive curvature. The non-hexagonal cell is a disclination, and imparts curvature,
through the Gauss-Bonnet theorem of geometry. 12 pentagonal cells transform
your kitchen floor into a football, an elementary consequence of Euler's theorem
to be discussed in the next Section.
This geometrical picture no longer holds completely in 3D, where there is no
Gauss-Bonnet theorem. In relaxed networks or froths, faces have on average 5.1
edges (Voronoi froth 18) or ^ 7 edges (CRN a ) so that there is no unique regu-
lar, flat packing. More to the point, disclinations exist, and are labelled in amor-
pbous materials whose space group is trivial, by the fundamental group of rota-
tions 3'4 nx{SOD). In Z>=2, 7|(502) = #> an<* disclinations, labelled by sign and
intensity, exist, whereas KX(SOD) = Z2 for all D>S, and the only topologically
stable line "defects" are odd lines (see below). Nevertheless, even in 3D, curvature
associated with the number of edges per face, remains an useful concept to
describe the local strain.
A dipole of neighbouring 5- and 7-sided faces form a dislocation (Fig.5). (Again,
this is only a local concept in 3D, without topological stability). The dislocation
has the following physical properties in random networks:
IS
1) It can glide (by u^ng T! rans'ormatioos). This determines the elastic amá
plastic properties of commercial foanis tt. It also constitutes a very efficient «ay
of dissipating shear energy. This property finds an airusing illustration in the
structure of daisies, y>>aecones or pineapples (phyUotaxb) M .
2) It can climb. A mitosis vor a T l tracsformaticn) creates a pair of dislocations,
and subsequent mitoses of the larger cells, corresponding to a glide of the two
dislocations away from each other, tearing an additional layer of cells inbetwecn
(Fig.6). Again, as far as elastic energy is concerned, this is a very efficient method
of solving the problem of copii.3 with addition»* material (growth of biological
tissues M or of metallurgical grains s&). Hillerts mechanism is slightly more com-
plicated as it involves creation of 3-sided celb (in 2D) only (a combination of T l
and T2 transformations, exclusively).
ii1 Ire lion ;i::^ «li-. iM;t!:on »rf .1 iji-l«K.ition pair by SUCCCIMVC cell divnior».
Morral and Ashby 3e uave given a very clear rrpresentation of these transforma-
tions and defects in 3D ordered foams or grains packings. In particular, the dislo-
cation core forms one single odd line (cf. fig. Ha of rcf. (36J), which shows thai
topological concepts applicable io 2D cannot always be transposed verbatim to
3D. The glide of the dislocation occurs through succesive Tl transformations, as
m JD. It is abo emphasised that a l tkcsc ar* bat
3) Dbiocataos» scree» the stiam d«e to discMsmtiosa (cwratmre). For example,
stnKtwaBy, a radnbria (mietoseopie mame aainal) h a huge footbaB, and ases
dishxatioas to screea the steam from the 12 dwctiaatioa repaired by topology
(football » sphere). The same «rn ramg of (topolopcaBy stable) disdiaatioas by
distocatioas (which are aot topofopeaBy stable ia passes) oetar ia aa elastic
4) Disfecatioas Jiable disdiaatioas to more. (For example, a Tl toaasformatioa
aezt to a disdmatioB creates a disfecatioa aad mores the disdiaatioa).
AD these properties emphasize the efmtmicd role of dislocations, which is partic-
ularly efecthre m raadom stractvres, where dislocations are not topologieaDy
stable, aad caa be regarded as local strain lactvatioas falfiUiag a specific, physi-
cal pvrpose.
Although these traasfonnatioas have beea iatrodaced here from a purely mechaa-
khl point of view, one can show (lectare V) that they abo leave the statistical
properties of the raadom structure (average shapes and their correlations) invari-
ant, so that they caa occar independently of each other, anywhere in space or
time, without affecting the statistical equilibrium of the structure. They play the
same part in statistical crystallography that meroremr$Mlitf plays in statistical
thermodynamics.
20
1.6 Topological conservation laws in random structures
Topologically stable constituents of disordered condensed matter are those
which retain their identity (and their existence) under small, continuous deforma-
tions of the system (including its motion). For example, a dislocation in a crystal
is topologically stable, an arbitrary elastic distortion, usually not. Topological
stability can be expressed as a conservation law or a continuity equation. In ran»
dom networks or froths, there are two conservation laws:
1) Euler's relation
F-E + r = X (2Z>)
- C + F - E + V = O (ZD) (1.2)
for a space-filiing graph with V vertices, E edges, F faces and C cells (including
the cell at oo). The Euler-Poincare characteristics x is a topological invariant of
the manifold (space) containing the network or the froth. It is an integer of order
1 ( « V,E,F in systems with a large number of elements). Euler's relation is
easily proven by induction, adding elements to a given network. In particular,
one can verify that the left-hand sidM of equations (1.2) are invariant under the
elementary transformations of section 1.5. Indeed, their physical properties sets
the time scale for topological evolution, which is much longer than that of non-
topological deformations (the latter, typically of order a/t, where a is the aver-
age spacing between atoms and e, the speed of sound).
2) Odd lines
The second conservation law is even simpler, and applies to 3D structures: The
presence of five- or seven-sided faces in a structure is symptomatic of its non-
crystalünity, being incompatible with simultaneous rotation and translation sym*
metrics. Odd-membered rings (faces with an odd number of edges) are not found
in isolation, but are threaded through by uninterrupted lines, which form closed
loops or terminate on the surface of the material 4 (cf. the ribbons of Pl.l)
Thus, odd faces are not isolated elements, but linear objects which are topologi-
cally stable. They are characterized by oddness, rather than intensity. However,
because these objects are lines, they remain difficult to eliminate despite their
modulo 2 algebra.
There are two, similar proofs of this conservation law. Both rely on the construc-
tion of an arbitrary, closed surface S, homeomorphic to a sphere for simplicity,
which intersects the random network at its vertices. S contains therefore some
vertices, edges and faces of the network, which triangulate S (this can always be
arranged by small deformation of $). The assertion is proven if, for any S, there
is an even Dumber of odd faces on S (providing an exit for any odd line entering
S).
i) Ascribe a weight Jt—{ 1) to every edge e on S, and an index
* / = I I Jf ~ ± 1 (even/odd face), to every face / on S. Then,
JJ 4> = f | JeJ = l, (Je enters twice in the product because every edge
ft 5 t ' li
belongs to 2 fact» on S). q.e.d. 37.
ii) Let «,/,- denote the number of edges, i-sided faces on S. The incidence rela-
tion between edges and faces, 2e = £]»'/,, implies that ][] í /,- = even, and,
because í /,• and / , have the same parity for i odd, VJ /,- = even. q.p.d. *.
The second proof suggests a generalization of the odd line conservation law to 2D
surfaces of arbitrary EuJer-Poincare characteristics and networks of constant ver-
tex coordination 38, and also to the structure of polyhedral networks (a model of
amorphous packings with tetrahedra and octohedra of atoms as elements, pro-
posed by Ninomiya M to describe medium-range "order" in amorphous packings).
For some amorphous materials, like polymer glasses, it may be too difficult or
complicated to construct a random network describing the structure. The sim-
plest, and least specific description of any glass is as an elastic continuum, with a
trivial space group (ie. without any infinitesimal translations! and rotational sym-
metries). (A crystal can also be represented as an elastic continuum, but with
either full (VoUerra continuum), or discrete (one of the 230 space groups) sym-
metry). Odd line is the only stable structural constituent surviving the transition
from discrete, cellular network, to continuum. (Cells,..., ie. simplices have disap-
peared). It takes then the form of 2ir-disclination. This can be shown very sim-
ply 40: Consider a configuration at a point, and take it for a walk in space, while
maintaining its orientation relative to the local reference frames (parallel tran-
sport) (Fig.7). Upon returning to the starting point, the configuration, which
determines the physical properties (strain, etc.) of the system, must either be
restored to its original orientation, or any mismatch must be physically
irrelevant. The latter situation applies to crystals, where possible mismatches are
elements of the space group, which label the Burgers vectors of dislocations and
disclinations. In glasses, the space group is trivial and the configuration must be
restored to its original orientation. Nevertheless, there are still two possible
transformations of the local configuration upon circumnavigation: a rotation by
4ÍT is homotopic (continuously deformable) to the identity, but rotation by 2ir
entangles connections of the local configuration with the rest of the system. Thus,
one remains with one single structurally stable constituent in continuous amor-
phous, condensed matter, the 2ff-disclination which has the same algebra as odd
lines in networks. Both have cores puncturing space (there is no way of filling an
odd ring because the relationship between any two vertices depends on the path
chosen around the ring - zero is an even number), and both are sources of non-
collinearity (see PI.l). They are therefore the same specific and universal consti-
tuents of glasses which we were seeking. They are also the only ones: transla-
tions are homotopically trivial, so that dislocations of any kind are not structur-
ally stable. Neither are point defects.
In summary:
Group of all possible local transformations (excitations) G = T3 A S 0 3
Space group H — 1 A 1 (trivial)
Topologically stable constituents (labelled by non-trivial homotopy groups of
G/H 3:
Odd lines - *i{SO3) = Z2
Non-singular textures - JT3(S03) = Z (yet to be identified)
Odd lines are therefore what count as configurations in the glass (eg. in the resi-
dual entropy *''"), and also as slow modes in a viscous liquid or in the glass
above To. Indeed, as long as the time scale for topological transformations is
longer than that for non-topological deformations (~a /c ) , the flow of topologi-
cally stable objects is a slow- or hydrodynamic mode, and any time-independant
conservation law has a time-dependant correspondant. Such conservation laws are
identities expressing topological stability, called Bianchi identities. [In elec-
tromapetism, the first couple of Maxwell equations include a time-independant
(div J9 = 0) and a time-dependant (0(D + curl E = 0) conservation laws for
24
the density B of topological objects. In elasticity, the topological objects are dis-
clinations in general (dislocations in crystals, where disclinations cost a prohibi-
tively high strain energy, and are excluded by an ad hoc hypothesis (distant
parallelism)), and inclusion of time-dependance constitutes the standard,
phenomenological generalization of elasticity to rheology and visco-elasticity 43.]
Thus, Bianchi identity plays two parts. It grants the object its topological stabil-
ity. But also, it is a topological conservation law which constraints the motion of
the object and of the fluid containing it.
Note that disclinations also appear in Kleman and Sadoc 4* description of the
structure of glasses, as perfect crystals in an "ideal" (in the sense of Plato),
curved space, which are projected into our usual, Euclidean space (the cave).
Overall homogeneity is guaranteed by construction in the curved space, as is the
best local packing (tetraledral or icosahedral, and therefore incompatible with
Euclidean space-filling requirements), and disclinations lines appear as a result of
the projection ib. Whereas this approach and our own agree in 2D, the disclina-
tions of Kleman and Sadoc conserve a sign and intensity in 3D, and are
prevented to cut across each other by topological obstruction (they are labelled
by & non-abelian group). Consequently, glass above To and supercooled liquid
cannot be described by the same curved space model (where Sow would be akin
to solving the Rubik cube), unless the tetrahedron SiO4 (in covalent glasses), or
the icosahedral cluster (in metallic glasses) order parameters, vanish at To.
J.C.Phillips has proposed a cluster model for chalcogenide and oxide glasses
glasses **. The size of the cluster (radius & 30 A) fits with homogeneity discussed
in section 1.4, but its surface represents a wall-like defect, which is not
2S
topologically stable (?o('903) = 1). It can be healed into line defects, as demon-
strated explicitely in the case of metallic clusters by Sadoc and Mosseri *7>4&, and
also in actual crystalline phases like Wp Cu*>Mg, Mna, Up 48. This lack of topo-
logical stability does not mean, of course, that local strains are not concentrated
on internai surfaces in glasses, and several experimental properties are best
explained by Phillips's model. Our topological point of view analyzes glasses at a
lower level of sophistication. Even then, the odd lines, which are the only
ingredients at that level, are not physically irrelevant but can account for non-
trivial, specific and universal properties of glasses.
n - GAUGE INVARIANCE
2.1 Discrete gauge in variance in spin glasses
Gauge «variance in disordered condensed matter was first mentioned in a
paper by Toulouse4fl dealing with spin glass on a lattice, described by the
Ed wards-Anderson Hanultonian,
where S ,=± 1 (Ising spins on lattice points i). and tbe coupling between nearest
neighbour spins is Jjj—± .!. according to some probability distribution given
a priori. Hamiltonian (2.1), and thus the physics of the system, are invariant
under tbe lotai transformation
S{' ^TtS, , J;i> ^TtJij7j (2.2)
parametrized by r,—±1 {Z* gauge transformation). This is a local transforma-
tion, involving variables which live on vertices (dynamical variables) and on edges
(connections); it is therefore a ga»ge transformation. Transformation (2.2) is
exact. It is also not very useful, since it involves two variables of physically
different status: Tho spins 5,- arc dynamical variables, allowed to reach thermo-
dynamic equilibrium within a canonical ensemble (Boltzmann distribution). By
contrast, the couplings J l ; are fixed (quenched) when sample under investigation
was made. They arc only random in the ensemble of different, bu! physically
equivalent, realizations of similar spin glasses. It is tbe free energy of every par-
ticular spin glass realization which is averaged over the coupling». In short-, half
of the gauge transformation J—*J' is not physically realizable on a pive.n sample.
Invariants under transformation (2.2) include not only the Hamiltonian, but abo
the face (or plaquette) index • / = f l ^ • an<l therefore the frustration (or odd-«7
ness in the language of lecture I) tig <fr = - 1 . (It b easy to show that frustration
forms dose loops or lines terminating on the surface of the material, using, eg.,
the proof (i) of the theorem of section 1.6). Frustration is a geometrical property,
independant of the "matter" field 5,, so that geometry (and essential, topologkal
disorder) is preserved under gauge transformation. Moreover, the partition func-
tion for a given sample, £[{ J}], and thus its physics, b independant of all the
details of the distribution of couplings {J} apart from those which are gauge
invariant **:
Z[{J'}) = Z[{J}} (2.3)
In other words, the tiling (warping, curvature, oddness,...) {$} completely charac-
terizes the geometry of the system, and is gauge invariant. As a corollary, the
unfnistrated or Mattis model (J,y=J, Jy so that all $ig • = 1 ), has the same
statistical mechanics as a ferromagnet.
[ It B amusing to see what happens to gauge invariance within the replica formal*
ism **. For simplicity, consider the infinite range model for N spins M, with gaus-
sian distribution of couplings. Then,
> = £ exp[/J*/V(2iV) Y,Y.S?S?SfSf),
which is invariant under two types of transformations: Either a local spin flip
S*' = TjSf, in all replicas, or a global rotation of each replica independently,
S°' = R ( l a ) Sa. The Edwards-Anderson order parameter ? „ £ = <5<o rS/> is
not invariant under the latVer transformation, which expresses the fact that the
symmetry between replicas is broken (<^a ; ?>^0). The system is trapped in one
of tbe many valleys (replicas) in configuration space. 2nd hydrodynamic modes
«re associated with this broken symmetry S3&1. ]
Transformation (2.2) cannot meaningfully be generalized to continuous (XY or
Heisenberg) spins, because the couplings J,y are essentially real numbers. We
shall see that, by going from a lattice to the continuum, and from a microscopic
(spins) to a semi-macroscopic description of the matter field, a full exploitation of
gauge invariance in disordered condensed matter (Yang-Mills theory) becomes
possible.
This simple example (2.2) of a spin glass on a lattice emphasize? the main charac-
teristics of gauge invariance in disordered systems:
a) Gauge invariance is an exact symmetry of / /
b) It preserves essential geometrical ingredients (frustration or odd lines)
c) It enables us to recover as much generativo homogeneity as is compatible with
(b), for example, by treating J as dynamical variables, and including (b) as con-
straints (source terms)50: From eq (2.3), Z -> £ Z[{J}\ 6[frustration).
An integral representation of the delta function restores the full gauge invariance,
at the price of a more complicated effective Hamiitonian.
2.2 Gauge invariance at a semi-macroscopic scale as a genuine sym-
metry
It was argued in section 1.4 that the non-gener.ilive homogeneity of glasses,
29
associated with randomness (non-collinearity of local reference frames), is a
genuine, 50(3) gauge symmetry. Indeed, as Jaynes has aptly put it in a
different, but relevant context (the Bertrand "paradox9 of probability theory),
"Every circumstance left unspecified in the statement of a problem (here, the
frames' orientations) defines an invaríance property which the solution must have
if there is to be any definite solution at all. The transformation group, which
expresses these invariances mathematically (here, the gauge group of local rota-
tions), imposes definite restrictions on the form of the solution, and in many cases
fully determines it" 81. Let us now be more specific.
The seminal 1078 paper of Dzyaloshinskii and Volovik (DV) M begins with the
statement that the temptation to use the concept of local exchange invaríance to
describe the spin glass state, is difficult to resist. Most readers did, I believe,
agree immediately with this preamble, albeit for their own, different reasons, even
though they did not accept at face value the specific modol of DV.
Here were my own reasons: Elementary excitations are deviations of the spins,
SiOt tetrapods,... from the orientation of the local reference frame, and their
(exchange, twist,...) energy is given by comparing deviations at different points,
each with a different frame orientation. The ordinary derivative dS, which meas-
ures these deviations in uniform magnets, must be replaced by a covariant
derivative DS, with D = d + iA, where A is the connection or gauge field,
which defines parallelism in the frames at two different points. The gauge field
provides the answer to the technical problem of connecting points with different
frames. The next step is to show that the particular orientation of the frame at a
given point has no physical importance (gauge invaríance).
30
Hertz51 argues by analogy with ferro- and antiferromagnets: The exchange
energy is given, typically, by | (#-«$)# |2t where ^ is, for example, the angle of a
XY spin, and <? is the wave vector of the magnetic modulation, 3 = 0 in fer-
romagnets, $5^0 in antiferromagnets (Ginzburg-Landau free energy). In spin
glasses, (${2) is a random variable with a given distribution; it is the gauge field
or connection.
In fact, gauge invariance should occur whenever one can define noncollinear, local
reference frames. Connection between neighbouring frames is defined arcwise, by
requiring that two overlapping neighbourhoods have local frames fitting together
without any rotation. However, a finite circumnavigation docs not necessarily
restore the frame to its original orientation; it only does so in the absence of cur-
vature, or of disclinations (Fig.7).
Fig.7. a) Example of parallel transport and non-unkity of the local reference frame in the pres-eaee of curvature. Tbe frame ia A has been rotated upon circumnavigation. The cone, tangent tothe trajectory, can be flattened on a plane to define parallel transport.
b) la a crystal, rotation of tbe reference frame mast be an element of the space group.
31
Consequently, the orientations of the local reference frames cannot be defined
uniquely everywhere in the presence of disclinations. It is locally arbitrary, but
the physical properties of the system are independant of this arbitrariness, ie.
invariant under a local rotation of the reference frame (given a connection
between neighbouring frames), which is precisely a SO3 gauge invariance.
Local frames occur in continuum elasticity theory. Glass can be regarded, on a
semi-macroscopic scale (whenever all relevant length scales are longer than the
interatomic distance a), as an isotropic, elastic continuum, with frozen-in internal
stresses 40>57>58. In metallic glasses, the stresses are due to the fact that Euclidean
space cannot be filled by atomic configurations miniming locally the energy (or
maximizing packing: tetrahedra, icosahedra) 44>4*. In covalent glasses, there is an
entropy barrier preventing crystallization (and inducing non-coUinearity of tbe
SiO4 tetrapods, and stresses), when cooled from the melt. Thus, stresses are asso-
ciated with Don-collinearity of the frames, which are in turn related to disclina-
tions or curvature 40>S7^. Continuum elasticity is, accordingly, a gauge
theory w.**.57.58.*9. Unfortunately, almost no three-dimensional crystals have any
disclination (because their strain energy is prohibitively high), so that, histori-
cally, the full gauge invariance of tbe theory was lost by the introduction of an
ad hoe, distant parallelism (zero curvature, global reference orientation)
hypothesis, justifiable in crystals, but not in glasses. This hypothesis breaks most
of the gauge invariance of the theory at the onset. In glasses, dislocations screen
tbe strain energy of disclinations (odd lines), so that they can axist on energy, as
weft as on topologccal grounds *°.
Comtet has found an explicit relation between integrability condition (connec-
tion) for frames on a minimal surface and the Euler-Lagrange equations of a class
32
of two-daaeasioaal 5Ü, gauge field theories n . Gauge iavariaace of the field
theory correspoads to rotatioa of the frames in the tangent plane of the surface.
Tab relation caa probably be generalized to three-dimensional, Yang-Milb {SOJ
field theory and the frames of a three-dimeasioaal hypenurface, thereby estab-
hshmg explieitery the correspoadaace betweea aoa-coBiaear frames aad SOt
gauge mvariaace at the core of thb lecture. As a bonus, the correspoadaace of
Comtet requires the exbtaace of aa underlying curved hypersurface, which may
tun out to be aa explieK reaHiatioa of the ideal, curved space of Klemaa aad
Sadoe".
Gauge theory b therefore the proper method to go from a discrete lattice to the
continuum, aad in particular, to generalize tike coupling Jtj to all points betweea
i aad / (which b easy to do if J>0, but not so if / < 0 ) . So, roughly, we have
some matter field, coupled through a covariant derivative to a gauge field
representing the coupling between spins. Specific questions remain:
i) What b the matter field ?
ii) What b the gauge field, which replaces {/} in the continuum ?
in) How should the disorder be quenched ?
The answer to (i) was an important idea of DVW . The matter field #(! )
represents, not one single «pin, but a group of spins within aa elementary cube of
the discrete lattice centered at 1 (semi-macroscopic representation). Because frus-
tration induces non-cottinearity of the spins, 4 b no longer a vector of ghrea
length (with values on a sphere (Hebenberg spins) or a circle (XY spins)), but a
hedgehog, or hirsute object, whose manifold of states b that of the fuD rotation
group S09 (Hebenberg) or SOt (XY). Similarly, in glasses, •<*) b an object br-
ing in the fall rotation group SOit rather than a single tetrapod (< SOrfT, where
7* b the tetrahedral group), for eovalent glasses, 01 a single kosahedroa
(c SOJI, where / b the kosahedral group), for metalbc glasses. Thb b because
the field • must be defined everywhere in a continuous space, and not only at the
centre of the tetrapod or ieosahedroa.
In fact, 4(i) must be chosen so that tt yields the correct "defects* (vortices, frus-
tration, odd fines) **. Hebeuberg spas would not have given any frustration
(*t(Sj) — I). Similarly, single tetrapod», tetrahedra or ieosahedra would have
given rise to line defects which would have been topotogkalry entangled (since
their lespective *, are aon-abeiian •*), thereby preventing the material to flow, so
that the Mine model of the structure of glass could cot describe the dynamics of
Ms supercooled or viscous liquid, even though both states have the same struc-
ture. (At aay rate, there b no generative tetrahedral or kosahedral symmetry in
the glass, hence no reason to restrict the manifold of • to a coset of SOS]
Hydrodynamics (irrotational fluids have for field a scalar potential 4(7), with
velocity 7 = v ^ , to be replaced by 7 itself when the field b rotational (with «3=
curl 7 as vortex density)), dectromagnetbm without, or with magnetic monopoks
[A replaced by 5= carl à as the field), and continuara elasticity without, or
with dbclinations **, provide classical examples of the overriding influence of the
defects on the selection of the proper matter field. Cf. abo our introductory
remarks (section 1.1).
Consequently, the matter field can be expressed in terms of a rotation operator
•(*) . X(*) 0(7) (2.4)
(Stueckelberg decomposition). Whereas in spin glasses, the amplitude X(7) can
34
be regarded as the order parameter M which vanishes at the spin glass transition,
in glasses, it represents the size of the tetrapods and can be taken as a constant,
X,, everywhere except at the core of the odd lines where it vanishes. The charac-
teristic length associated with fluctuations in X is #»« , the interatomic spacing
in the network or the size of a plaquette. \jíQ corresponds to the chemical pro-
perties of the constituting atoms, and is taken for granted at all temperatures
relevant for the glassy, liquid or solid states. This approximation is sometimes
referred to as the nonlinear <r model.
(ii) We postpone discussion of the precise meaning of the gauge field AJfi) until
next lecture (eq.3.2), apart from remarking that it should appear naturally and
automatically in the free energy density through the covariant derivative of the
matter field, and that it is directly related to the non-collinearity of the tetra-
pods. In fact, introduction of a gauge field A^H) is necessary to make the
derivative covariant (eq. 2.6 below). As it is its only physical purpose, one can
assume that the covariant derivative also provides the only coupling between
gauge and matter fields (minimal coupling). The length scale of the gauge field, /,
(penetration depth in superconductors) is the range of non-collinearity. Thus (see
Pl.l), t»&&a, and glasses are type II gauge materials (in the superconductivity
terminology). They exhibit vortices rather than full Meissner effect.
(iii) From our experience with discrete lattices (end of section 2.1), it seems evi-
dent that gauge fields should be treated as dynamical variables (and full gauge
invariance or homogeneity maintained) as much as possible. Only the source of
gauge fields (the frustration or odd lines), ie. the essential disorder, need be
quenched. This remark is in contrast with the assumptions of DV and of Hertz,
who both felt that all disorder terms should be quenched. Hertz M quenches the
35
gauge field Â(2), and DV5 5 give it a mass term in the free energy. Both pro-
cedures break gauge invariance. Ours is gauge invariant. Furthermore, the
sources of gauge field - the (geometrical) odd lines, quenched below T0 in glasses,
and at all temperatures in spin glasses - can be treated simply as punctures of the
space E into which matter and gauge fields are put, exactly like flux lines in
extreme type II superconductors. Boundary conditions on the puncture are free.
This allows gauge and matter field (with amplitude X=c«f) to take up
configurations (eg. rotated by 2* around the puncture), which would have been
forbidden if the space had been simply-connected. Thus, we replace » simply-
connected, Euclidean space by a punctured one (E). This complicates slightly the
geometry but simplifies enormously the algebra, notably by letting
X = | $ | = est everywhere in E, and by restoring full gauge invariance to the
system within E.
For example, consider a superconducting ring E. The free energy within E is
gauge invariant, the magnitude of the matter field (superconducting order param-
eter) is uniform within E, and the new configurations associated with multiple-
connectivity are those of quantized fluxoid. The degeneracy between these
configurations is only lifted if the (electromagnetic) free energy outside £ (chiefly
inside the ring's hole) is included.
In summary, sources of gauge field are puncture», with free boundary conditions
(except that every configuration must be single-valued, and rotate by a multiple
of 2* about every puncture). The free energy is fully gauge invariant in punc-
tured space E, where the magnitude of the matter field can be taken as constant.
The multiple-connectivity of E implies that there are several possible ground
states (valleys), all degenerate because they are related to each other by gauge
36
transformations (see lecture III), instead of the unique ground state in simply-
connected space, which is one of the cornerstones of classical solid state physics.
2.3 Gaage invariant model free energy for glass
Without further ado, we can now write down the model free energy for
glass87.
F = / « * / ( * )E
Âp\ (2.5)
where E is the Euclidean space, punctured by odd lines, with free boundary con*
ditions.
/(?) is invariant under gauge transformations, the /oco/ rotations
 ; = üÀpü-1 + - í / ^ c r 1 ) (2.8)
under which the derivative is covariant,
37
Fpy is gauge covamnt,
F¥J = ÜF^ÍT1 (2.7)
(unlike its abelian counterpart, the electromagnetic induction S — curl Ã, which
is gauge invariant), and is therefore only related to the non-collinearity density
Fpy, a physical observable which must be gauge invariant (eq.3.2).
Free energy (2.5) was written down as early as 1954 by Yang and Mills M in the
completely different context of elementary particles. This suggests that there is
little arbitrariness in the selection of a gauge invariant free energy.
Gauge invariance imposes the introduction of a new field, the gauge field. On the
other hand, it severely restricts the possible free energy densities Consequently,
eq. (2.5) has very little arbitrariness: Only,
1. Minimal coupling between gauge (À^ and matter (0) fields (solely through the
covariant derivative)
2. One energy (density) scale t = (i\jg)2 = l/{g2l4) = kB TJl*
3. One length scale (the penetration depth) / = l / f v ^ X , )
4. The distance ç between punctures (l»ç»^ a), and their configuration
(semi-dilute)
have or will be chosen on physical grounds.
Odd lines, and gauge invariance will enable us to construct explicitely the many
valleys and two-level systems responsible for the low-temperature properties of
glasses
38
m - TUNNELING MODES
8.1 Many potential valleys in configuration space
The anomalous properties of glasses at low temperatures, briefly reviewed in
section 1.2.a, and in detail in refs.[2] and (5], are properly described by the con*
cept of tunneling modes ' , which are the elementary excitations of systems with
several deep potential minima in configuration space. In this lecture, I shall use
the model free energy (2.5), ie. odd lines and gauge invariance, to locate precisely
the many valleys in configuration space, to label them, and to calculate the tun-
neling rate between different valleys.
It is elementary to show that a distribution JV(Atf) of two-level systems (2LS)
split by an energy A, (proportional to the tunneling rate), yields a specific heat
increasing linearly with temperature,
(assembly of Fermi-Dirac oscillators), where kB is Boltzmann's constant, as long
as the density of two-level systems remains finite as A,-+0. The coefficient
C/Tfal^Wêif^K'2 is typically an order of magnitude smaller than that of a
good metal (fvidr* W^K'* for Cu), suggesting ÍO^-IO"6 2LS / atom active at#
« IK, with a size of «40.4, since resonance (T2) measurements indicate that
the 2LS are roughly independant excitations 2-105.
Three remarks can be made at the onset:
39
1) In tunneling modes, the higher and broader the potential barrier, the smaller
the tunneling rate and the level splitting. In glasses, splittings down to
£&{10~2K)kg are observed (no departure from linearity, or time dependance of the
specific heat down to these temperatures), suggesting high (^kBT0) and broad
potential barriers, that is a lot of atoms moving very little. The size of 2LS indi-
cates that, as far as quantum coherence goes, they are semi-macroscopic: consid-
erably larger than a single electron or alpha particle, but still smaller than a
SQUID or Schroedinder's cat.
2) The classical potential barrier may even be infinite (and will be so for our con-
tinuum model free energy (2.5)), but, as long as W is integrable, the tunneling
rate remains finite. Then, the 2LS have no classical equivalent. This may explain
why the explicit nature of tunneling modes has remained elusive for 10 yean.
3) Saturability 5, and coherence 7 of the 2LS indicate that the elementary excita-
tions within each potential we'll are thermally inaccessible below &1K. They lie
at much higher energies than the splitting A0 between ground states of different
wells. Thus, only the ground state within each well is relevant to the physics ol
glass at low temperatures. The situation is reminiscent of the tight-binding, oi
LCAO model of electronic band structure of metals and molecules, where only
one orbital per atom is relevant to the physical or chemical properties in a lim-
ited energy range, from which core levels, and higher atomic excited states, are
inaccessible.
These remarks are sketched in Fig.8, which is also an adequate summary of the
conclusions of this lecture.
40
CONR SPACE
UO 2 3 D)
!27T>
_mm CONF. SPACE(ID/OOO LINE)
Fig.8. Tbe many valleys in configuration space, giving rise to tunneling modes.
a) Old picture (ref.|6j).
b) New picture (this lecture).
Note that the lO^dimensional conflgoration space has been redwed to ID per odd line; the vat-
leys (topological sectors) can be labelled ( | 0 > , |2f>), and their distance (or tbe energy splitting
A,) calculated.
Dotted Unes: excited states in each sector (irrelevant at low T)
Foil line»: 2LS
Only a rough topography of tbe valleys is required to obtain groand state and elementary excita-
tions of a glass.
41
The specific heat is approximative)? linear in the range Q.Q25K-IK. This sug-
gests a constant density of states for 2LS, N(A#) = N(0), in thb energy range.
In fact, Lasjauias ef «I9* hare suggested that N(At) has a gap below 16 mK.kB,
thereby giving a limit for the slowest tunneling rate observable experimentally. It
must be emphasized that several problems of detail remain with the tunneling
mode concept (only approximate linearity of C, 7 s contribution to C
significantly above that expected from Debye phonons, response of the system to
short thermal pulses, etc " A 1 * ) , bat the major theoretical challenge is univer-
sality: "...no plausible argument has been presented yet why all amorphous sub-
stances have approximately the same density of states of tunehng defects which,
and that b probably even more puzzling, scatter the phonons with almost equal
strength" a (US, not HM spelling). The experimental situation will abo be
clarified, as new and direct methods of investigation and comparison are
developed (effects of high pressure w , or, as in epoxy resins, of the size of the net*
work constituents n ) .
3.2 Classical, ground state configurations
According to Fig.8, it b sufficient to identify and label the ground state or
metastable configuration of every valley. These ground state configurations
{À0t(l} are classical solutions of the Euler-Lagrange (EL) equations obtained by
minimizing the free energy (2.5) in punctured space E (the punctures being put in
• priori - quenched at random below T0), with free boundary conditions on the
puncture. Free energy, boundary conditions, and EL equations, are all gauge
invariant. The EL equations look frighteningly complicated 37, but we shall not
need An explicit solution.
42
The matter field ft is obviously parametrized by rotations, and one anticipates
two ground state configurations per puncture, one corresponding to a rotation of
Ò upon circumnavigation around the puncture by 0 or a multiple of 4ir, the other
to rotation by an odd multiple of 2*. (The latter is the new configuration per-
mitted by the multiple-connectivity of £). All other configurations can be con-
tinuously deformed into these two. (Recall our discussion of section 1.6, and the
fact that the rotation group is not simply connected (jr,(S0s) = Z2), necessary to
justify quantization of the electron's spin in half integral multiples of h/2x). We
must show that the gauge field Âf is aho parametrized by rotations. Then, we
will have proven that there are two ground state configurations per puncture,
parametrized by rotations.
To do so we construct a linear combination of matter and gauge fields which is
gauge invariant, so that a covariant gauge transformation (rotation) of the
matter field ft induces a contravariant rotation of the gauge field. This procedure
is familiar in superconductivity and electromagnetism, where the phase 9 of the
order parameter ^(7) = p(?)e2p [#(?)] forms with the vector potential X[t), the
gauge invariant combination <5(a?) =
A' = A +
0' =0-{hc
(This constitutes also the simplest derivation of the Higgs mechanism (massive
vector boson) in minimally coupled, abelian gauge theories).
The construction proceeds in a few steps: 37>68
43
i) A given configuration of the matter in £ can be written in terms of its orienta-
tion at some point ?#, Ò# = Ô(?#), and a rotation operator W(t) c SOS), which
is independant of the path from % to ?, apart from a winding number (rotation
by 2*n, n=O,l) around every puncture. Then, tl{t) = W(2)ilt W~l(&), and the
"phase gradient" (the phase orientation density x* =s (dflW)W~1) are path-
independant.
ii) One then goes to a rotated frame, defined by à = WTlkWt for any operator
A. The new covariant derivative Õjb. = ÚT^D^W, involves the phase £„ « d
the gauge field Â^, in a linear combination C^, (Stueckelberg decomposition)
= gÀ,(t) + xfi) - \Cfi)\' (3.1)
which is gauge invariant, as is any operator à in the rotated frame. (Under gauge
transformation ÜÇ2), W'{?) — C/(7)^(aT), and Â' — Ü\ÍTl impües Ã' = Ã ) .
This includes F^,
which, being gauge invariant, measures the physical density of non-collinearity.
Similarly, the free energy density (2.5) can be written in term of the gauge invari-
ant C,.,
— • F F t SSS I ITI lasol
iii) The combined gauge field C^ is accordingly gauge invariant, and therefore
unique. The rotation WÇZ), which parametrizes the matter field Xf, or 0, also
44
parametrizes the gauge field À0. Because there are only two possible
configurations per puncture of the matter field, up to continuous deformations,
corresponding to rotation by 0 or 2JT, there are also two configurations of the
gauge field, and therefore, two ground state configurations per puncture {tl^p},
solutions of the EL equations for the free energy F in punctured space E. q.e.d.
Invariance of the free energy under a non-trivial gauge transformation (rotation
by 2x), implies that F is periodic in he fiux triggering the gauge transformation,
as long as the source of non-colliacarity vanishes in E (cf. superconducting ring,
where F is periodic in the applied magnetic flux, as long as the applied magnetic
induction vanishes inside the superconductor). This generalizes Bloch's theorem
for superconductors **, to any minimally coupled gauge theory.
Similarly, the restriction on the matter field (which must be uniform for a given
configuration, ie. returned to the same orientation after circumnavigation),
translated into a restriction on the gauge field, corresponds to fluxoid quantiza-
tion in, superconductors. It is a direct consequence in minimally coupled gauge
theories, of the non-triviality of some gauge transformations.
3.S Tunneling states, two-level systems
We have identified and labelled two ground state configurations per puncture
(odd line), which will serve as the "atomic orbitais* of our LCAO. These two
"tunneling states" | 0 > and | 2?> are characterized by rotations of 0 and 2JT
upon circumnavigation about the puncture. They are degenerate in energy (by
gauge invariance of F). But neither classical configuration is gauge invariant,
45
and can qualify as the true ground state of the glass, which must be gauge invari-
ant. (There is no reason to assume that gauge invariance should be broken in
glasses, us we have argued in lecture II). Tunneling, however slow, must take
place to restore gauge invariance. The true ground state and elementary excita-
tion, per odd line, are the gauge invariant combinations,
l ± > = ^ l | O > ± \2x>] (3.4)
which are the 2LS, split by (fc/2x) times the tunneling rate.
3.4 Tunneling rate
We shall assume that the frozen configuration of punctures is semi-dilute, that
is, intermediate between dilute (alphabet soup), and melt (dense spaghetti bun-
dles) (The appellation is borrowed from polymer physics and may not correspond
to gastronomic states). In this case, only one length describes the frozen, isotropk
configuration, the distance ç between non-adjacent segments of odd lines (which
may equally well belong to the same or to different odd lines). Thus, to transform
configuration | 0 > into |2*> about a given odd line, it is sufficient for a quan-
tum flux of rotation of length ç to tunnel through a distance ç only. Further away
lies the region influenced by other odd lines.
Justification for a semi-dilute distribution relies either on the dynamics of odd
lines when the glass is cooled from the melt (lecture IV), or on maximizing the
entropy (most probable distribution) &7. Qualitatively, the most probable distri-
bution of odd lines is that which maximizes the density of low-energy states, per
unit volume and energy interval. This suggests the semi-dilute distribution, in
46
which the loops are concentrated enough for the odd lines to intertwine and the
nearest non-adjacent segment to belong to another loop (maximum namber of
2LS per unit volume), but not concentrated enough that their cores overlap
(maximum number of 2LS per energy interval). This picture has not yet been
confirmed experimentally.
The energy splitting A# is given by the transition amplitude,
- 1 ^ - = <012r> = / [DÜ] [DÂJ cjp {-{2x/h)Jdr L (r) ) (3.5)
for all paths in imaginary time r connecting | 0 > and |2x>, where L(T) is the
Lagrangian, and the energy has been adjusted to vanish at the initial and final
configurations. The leading contribution to A# is given by the classical paths
from 10> to | 2JT> minimizing the exponent (action) in (3.5). One faces, at once,
two major problems: 1) To obtain a Lagrangian from the free energy (2.5), ie.
dynamics from thermodynamics. 2) There are no continuous, classical paths from
|0> to 12x>, since the configurations are not homotopic. A trick must be used:
For a semi-dilute distribution of odd lines / » ç » a , the free energy (2.5) or
(3.3) can be linearized (the free energy is dominated by terms with the highest
power in /, that is the linear terms in C^ in Fpin because the length scale of dC
is now ç rather than / as for a single odd line or in the dilute limit (>/) . The
linearized free energy is similar to that of an extreme type II superconductor,
with penetration depth / and constant magnitude of the order parameter. A# will
therefore be given by the tunneling rate of a flux quantum across a superconduct-
ing ring of radius f, height ç and radius of puncture a. This rate can then be
evaluated by opening a thin slit (width a) across the cylinder. This trick pro-
vides at a stroke 1) a Lagrangian, 2) a classical path.
4?
One has, in fact, the geometry of a Josepkson transmission hne, along which the
magnetic lux 4 satbfes a sine-Gordon equation n , whose Lagrangian b well-
known
where X/ b the Josephson penetration depth and «, the warefront Telocity. For a
thin slit of width c, \j^l, and c"«<:\/«7(*fj
A typical superconducting ring has height L, radhts & and L > ç » X j , so that a
tall, thin solitoa propagates along a long slit. The cosine term dominates the
dynamics, and the tunneling amplitude goes as expf- cat Lç\ which is utterly
negligible, so that superconducting magnets can hold their magnetic field for long
enough to be useful.
By contrast, in glass, the rotation flux quantum is a very fat X/^/ , short (height
f) object, which has to more very little (distance ç«l). The [cos* - l]/\j term
b negligible, and jir L{i) «s the energy of a 2D vortex. The tunneling ampli-
tude is much larger **,
A, = ^II. exp \-D (Ç/B) lnfr/«) ] (3.7)
as long as ç is not macroscopic (semi-dilute limit). Here D =
Typically, ç/« « 40, T.(*sdO0K, (kc/2wkB 7 » » 10 , yielding an exponent
Abo, a dbtribution of distances Ç,
(f/a)l exp [-
48
corresponds to a flat and broad density of 2LS energies /V(A,)«(t&T*)"1 for
At<kBTó. Numbers obtained from (3.7) are therefore reasonnable. Note that
the density of 2LS N(Aa), is inversely proportionnal to the only energy scale of
the problem kg To (see eq. (2.5)). Thus, in glass, tunneling does occur, and yields
the dominant excitations below some low temperature proportionnal to
1/y/kgT,. This temperature may be anomalously low if the elastic energy kB T,
is large, as is the case in a-Ge, or if the distribution of odd lines is not semi-
dilute. On the other hand, no odd lines, and no tunneling modes are expected in
glasses which can be described by a bichromatic model containing even rings
only 71, or its continuous equivalent, a pure gauge model. a-GaAs may be a sys-
tem without odd member rings, as long as wrong bonds {As-As or Ga-Go) are
excluded by electronegativity.
The density of 2LS is inversely proportionnal to the only energy scale of the
problem kB To. (Cohen and Grest also obtain this relation in their free volume +
percolation model72). This should be directly observable by comparing viscosity
and low T properties of the same system under pressure. Measurements by Bar-
tell and Hunklinger M in vitreous Si02 show that the low temperature ultrasonic
attenuation (ie. the density of 2LS) surprisingly increases with increasing pres-
sure. This is consistent with the anomalous pressure dependance of the viscosity
(which decreases as p increases) observed in GeO2 (and Si0 2 is expected to follow
the same behaviour 73, to the delight of geologists, worried about the viscosity of
the earth's mantle) through Tt decreasing with pressure (although why this is so,
is not clear microscopically). On the other hand, neutron irradiated silica
apparently follows the opposite, less unexpected, trend: Increase of the mass den-
sity upon irradiation is accompanied by a decrease in the density of 2LS, N(A0),
ie. an increase of the thermal conductivity and a decrease of the specific heat u.
49
However, irradiated and fused silica have strikingly different structural charac-
teristics, even if they are both amorphous. (The intensity of the 600 cm ~* (sharp]
Raman line of fused silica increaes by an order of magnitude upon irradiation 76).
8*S Can one see what it fa t ha t tunnels ?
A superconducting ring can be prepared in one of its free energy minima.
r>ecause an external magnetic flux is readily available irom a magnet. Tunneling
between energy minima is then observed as Josephson effect. In glass, by con-
trast, rotation flux does not correspond to a commercially available knob. How-
ever, there exists a method to prepare the system in one of its classical ground
states, which takes advantage of the coherence of tunneling between two levels,
the phonon or electric (when the 2LS have a dipole moment) echoes7, the 2LS
counterparts of spin (5=5=1/2) - echo spectroscopy in magnetic resonance.
In order to find out precisely what it is that tunnels, one would like to tune a
structural probe (pulsed neutron beam) to the coherently oscillating tunneling
modes, so that neutrons take successive snapshots of the 2LS when they are in
the classical state | 0 > , say. Constructive interference between successive
snapshots can be arranged by splitting the neutron beam, scattering beam 1 on
the sample at time t—0, when the 2LS are in the classical state
| 0 > = (l/ \ /2)[ |+ > + | -> ] i then delaying the scattered beam, while beam 2
is first delayed, then scattered by the s; .nple at time t~2t0 when the 2LS are
again in state | 0 > . The two beams are combined at some later time t°>2t0, and
superposition of neutron diffraction of the same 2LS in the same, classical state
| 0 > , should show constructive interference78. (Fig.Q). Varying the delay time
50
detunes neutron and 2LS, and leads to destructive interference between scattered
beams. Varying the time of scattering of the first beam (taken above to be f =0),
will find the 2LS outside the classical state, within the barrier, and the construc-
tive interference of the scattered beams destroyed. (Even though neutrons are
quantum probes, which could couple to 2LS through the off-diagonal operators ax
or o$, as well as through the diagonal ones, i or <rx, the 2LS involve too many
atoms in a disordered configuration, to be detected by neutrons as anything else
than density fluctuations (through operator i). In this case, echo (coherence) is
observed only for the classical states 10> and 12?> (eq.3.8)). Combination of
constructive and destructive interferences should fingerprint what it is that tun-
nels.
Phonon echoes are obtained as follows: One applies to the 2LS, described by the
Hamiltonian
( | ± > are eigenvectors of <rz), two high amplitude , A,, electrical or acoustic
pulses of duration r (SET pulse) and 2r (REVERSE pulse), respectively. The fre-
quency A = w(A/2ff) of the pulses should be ^ A < ( ( « A 1 ) , for resonance.
Hi(t) = — Ai(<
The SET pulse induces a phase shift
S = ^ ( A - A J 2 * A,2 T (2ir/A) « A, r (2TT/A) = JT/4
The evolution of the 2LS is evaluated in a rotating frame, where the effective
Hamiltonian is time-independant. The time-table of the tuned echo-neutron pulse
51
system is sketched in Fig.9. A r/4 pulse prepares the 2LS in classical state jO>
at f=0. A JT/2 puke at t=t, reverses the 2LS state, from a | + > + b | - > to
« | - > + 6 | + >- The echo is observed at f =2* ,+ 2r^2^, when the 2LS is again
in the classical state | 0 > . The two pulsed neutron beams (*) measure the state
of the system at i = 0 and 2t,+ 2r.
n. nn . nSET REVERSE ECHO
T <- t0 - 2r - t, —
Fig.9. Time-table of the tuned 2LS echo-neutron system
A distribution of 2LS {A,,} stills shows coherence (echo), as long as the reverse
pulse is short compared to the time of free propagation Ar«A t f i < ) , and the 2LS
are initially in a classical state:
| + / ->(a i f=0)-> C\-/+> + t,-deptndant term {at f= / # +2r) (3.8)
Thus,
| + > ± \->(at t—O)-+±C(\+>± j - > ) + incoherent terms {at t—to+1r)
A,where C=-isin(2S)sin0 and 0 — tan"1 •-• •. Maximum coherence occurs at
resonance A « Í A P , where 2S=0—ic/2.
In this lecture, we have exhibited explicitely the many potential valleys in a
52
continuous model of glass. The situation is summarized in Fig.8. Some progress
has recently been made towards repeating the same analysis for discrete random
networks, and obtaining again two classical configurations per odd lines 7I.
53
IV - SUPERCOOLED LIQUID AND GLASS TRANSITION
4.1 Introduction
This lecture will be concerned with the regime T>T9 above the glass transi-
tion temperature, where the structure can be deformed under shear, and glass
behaves like a supercooled, highly viscous liquid instead of a structurally invari-
ant, elastic solid below Tf. Viscosity, or any inverse structural relaxation rate,
follows the empirical Vogel-Fulcher relation (1.1), at least approximatively, as
described in detail in section 1.2b.
Above To, odd lines can move as the structure is deformed. However, they obey
a topologkal conservation law (section 1.6), so that they can expand or shrink,
but remain uninterrupted, assuming as before that bonds are broken and recon-
structed over a much shorter time scale than that associated with the fluidity.
Odd lines are the slow - or hydrodynamic modes of the system. Viscosity, or any
structural relaxation rate, is associated with the diffusion of some "defect", in the
crude but adequate model of Glarum, and Phillips, Barlow and Lamb 78. The
"defect" must be of sufficient generality to account for the universality of the
relaxation process in glass, and of sufficient stability to retain its identity and
avoid disintegrating during (slow) diffusion. Odd lines, or 2^-discJinations, as the
only structurally stable ingredients of any glass, clearly fill the bill on both
counts. Free volume would not, on its own, be stable enough.
The glass transition at T0 corresponds to the freezing or "condensation" of the
odd lines. We know already (section 2.2), that glass transition does not
54
correspond to a vanishing amplitude of some order parameter (unlike the transi-
tion in spin glasses), nor to the breaking of some (gauge) symmetry. A mean field
calculation of the density of free, or mobile odd lines will yield a condensation
similar to the Kosterlitz-Thouless transition, but in 3D, which may be frustrated
close to 7*# by a crossover to Arrhenius behaviour (see section 4.5).
Ideally, one would like to start from the free energy (2.5), and allowing the punc-
tures to move freely, like vortices in hydrodynamics and flux lines in supercon-
ductivity. Unfortunately, to the best of my knowledge, no method has been found
as yet to isolate the "vortices" as elementary excitations of a field theory in the
continuum, similar to the Villain method on lattices ". [ In electromagnetism,
thb would mean partitioning the integral over all gauge field configurations A (7)
into a discrete sum over configurations of different, quantized flux £ A{2) & ].
We must therefore calculate the energy E and entropy 5 of an assembly of free
(mobile) odd lines of density pj. Minimization of the free energy
F\pf] = E[pf]-TS[pf] , 6F/6pf =0, yields the density of free odd lines in
thermodynamic equilibrium Pj(T). A semi-dilute distribution of odd lines, has
their dimensionless density related to their shortest mutual distance ç by
P = (a/çf (4.1)
This is a pedestrian treatment of the model free energy (2.5) at high tempera-
tures when the punctures are mobile, but it is the best we can do. The argument
was first sketched by Anderson 1, and worked out in ref.[86].
This method has the free odd lines as the only (slow) dynamical variables or
degrees of freedom, through their density pj. The other (frozen) odd lines do not
contribute to the dynamics and thermodynamics as a first approximation. They
cannot be displaced or adjusted to lower the free energy of the sytem. At T4, all
odd lines are frozen pf{TJ=O (eq.4.5). Just above 7*#, there are so few free odd
lines that they form a dilute distribution, even though the configuration of all
odd lines, frozen and free, b semi-dilute at all temperatures, as discussed in sec-
tion 3.4.
All the results of this lecture have also been obtained from continuum elasticity
theory, allowing for (2JT-) disclinations, which are, as we have seen in section 1.6,
the only topologically stable, line constituents in a medium with trivia] space
group. Dislocations are not topologically stable, and screen the strain energy of
disclinations, which are then allowed to exist in 3D glasses on energy grounds
(their strain energy is unacceptably high in crystals, where it increase» linearly
with the distance between two disclinations). Most of the continuum elasticity
results on glass are due to Duffy, and have been published extensively 40-80'57.
4.2 Energy of an assembly of odd lines
We start with the fields and geometry of free energy (2.5) or (3.3). Assuming
(to be confirmed a posteriori) a semi-dilute distribution of free odd lines at
T»T0, ç«l (where / is the length scale of the gauge field, which measures
the range of non-collinearity), we can linearize the theory: The quartic terms in
the free energy density (2.5) or (3.3) are negligible when compared to the qua-
dratic terms involving spatial derivatives , because, in terms of rcscaled gauge
fields <?„ - /.
idC « //f » C
56
The free energy density (3.3) reduces to
(4.2)
which is that of eiectromagnetism of current loops (apart from the fact (taken
care of in the trace) that C is a tensor rather than a vector). In particular, the
Euler-Lagrange equations become simply
curl curl C =
where are the odd loops, sources of gauge field C. Consequently, the energy E
of the loops takes the Ampere form,
dTdT K* ~ < S/J -rr-fr + E W i: = « E
I ' 1t) I ' . - ' / I í i -1
E — F\-Apf In pj + Bpf] (4.3)
for all, semi-dilute loop geometries 57. Here, it is convenient to refer sums or
integrals to an underlying random network of edge length a, with F faces, L odd
loops making up a total of K segments of of length a (or threading through K
odd faces), core energy ecort, so that pj =s= KJF — (a/f)2. r;- is the distance
from segment / to the nearest, non-adjacent one. In eq. (4.3), the parameters
take the values
A = i i / 3 , B ^ etm + £ <ln(r í/í)>
The leading term in (4.3) is proportionnal to the only energy scale c, and only B
depends (weakly) on the microscopic structure of the glass through the core
structure. This is, of course, the result of the semi-dilute distribution of free odd
57
loops: A semi-dilute distribution has been assumed twice in the derivation of the
energy (4.3). First in the linearization of the field theory (2.5-3.3), and second in
the evaluation of the Ampere integral (4.3). When the distribution of free odd
lines becomes dilute (when T approaches To as we shall see in section 4.5), the
logarithmic contribution in (4.3) vanishes, and the energy of each loop adds up,
E — B' fij (dilute).
Elasticity theory yields the same result (4.3) 4°.80-*7. Outside the cores of odd
lines, the strains and connections are small and the theory can be linearized.
Dislocations (which are not topologically stable) screen the strain energy of decli-
nation down to the Ampere form. They also enable disclinations to move, as dis-
cussed in section 1.5.
4.3 Entropy of an assembly of free odd loops
The calculation of the previous section, despite starting from a free energy
density, kept the odd lines frozen in order to calculate their energy. The entropy
of free odd lines, which can move, expand or shrink, cannot be evaluated by the
same method, but must be calculated directly.
Let us calculate the number of configurations fi of K odd faces (in an underlying
random network with a total of F faces), making up an arbitrary number L of
closed loops. The first loop threads through ni odd faces, the second through n2,
etc. Lm is the maximum possible number of loop, n0 the minimum number of
odd faces threaded through by a single loop, thus X=Lm/K=l/n^ — 0(1). Let
C(n)/n be the number of configurations of a single n-stcp loop with arbitrary
58
starting point. Typically,
C(n)/n
with fl&3 in 3D, but actual numbers are of little relevance to the final result for
the entropy (4.4). The number of configurations is given by
The binomial factor is responsible for the entropy of mixing between loops, and
C(n)/n, for the configurational entropy. In mixtures of loops of arbitrary length,
the entropy of mixing of the smaller loops ( ^ \aK\ *=» K ln/T) is expected to
dominate the configurational entropy ^ K In(z-l).
One introduces the generating function
in terms of which ft is expressed as
Whenever l « L m « F , the most probable distribution has overwhelmingly
more configurations than all the others, as is well-known in statistical mechanics,
thus,
since Lm <<(1/2)F, and the entropy is obtained in terms of pj^
5 = kB\nÇl = kBFi~Xpf lapf + Ypf) + O(\nK) (4.4)
It is dominated, as anticipated, by the binomial coefficient ie. by the entropy of
mixing {X=Lm/K). The second term in (4.4), l^YV*- Yit has both mixing
(Y^X-XlnX) and confipurational {Y2~ln{z-l)-Xln(n^) origins.
4.4 Density of free odd Unes in thermodynamie equilibrium
Although glass is not strictly in thermodynamic equilibrium, and the observed
"glass transition" is essentially kinetic in character (the fluid-solid crossover is
smooth, and occurs at a temperature above T, which depends on the cooling rate
and on the thermal history of the sample), the time scale for topological
modification and for change in the physical properties is much longer than that
relevant to the propagation of elementary excitations (eg. phonons) about a
metastable state. Thus, slow modes (the density of free odd lines) can be
obtained from a thermodynamic derivation, and describe the physical behaviour
of the system, even if its linear response is so sluggish as to be inaccessible experi-
mentally very close to To.
The density of free odd lines at a temperature T, Pf{T), is that minimizing the
free energy F — E ~ TS, with E(pf) and S(pf) given by eq. (4.3) and (4.4),
respectively. One obtains,
, / ( D « » e - í ' í r - T - ) (4.5)
which has the Vogel-Fulcher form (1.1) because both energy and entropy have
the same functional dependance, pf\npf. Here kBT0=A/X^ c is the only
energy scale of the problem, as expected (section 3.4). 6=exp(yr/X)-l and
60
kBc={B-AY/X)/X is a positive number because et#re>«, (B>A).
The density of free odd lines vanishes at Tt, where all odd lines have been frozen
by their mutual interaction. Similarly, their entropy abo vanishes exponentially
at Tt (Kauzmann paradox),
Below T#, the odd lines arc frozen, and their entropy (which could then be calcu-
lated from the free energy density /(7) (3.3)) is not accessible thermodynamicaUy
(eg. from the specific heat). This condensation of odd lines at 7*# is a genuine
phase transition, the extrapolation to zero cooling rate of the kinematic 'glass
transition". However, the assumption of a semi-dilate distribution of odd loops,
responsible for the pflnpf dependance of the energy (4.3), is not valid close to
r#, and the pj[T) crosses over from Fulcher to Arrhenius behaviour, as we shall
discuss in secton 4.5.
The viscosity, or any structural relaxation time, then follows the Vogel-Fulcher
law (1.1), since its inverse, fluidity, is due to the diffusion of some topologically
stable, universal "defect" in glass 78>12'1, here, the odd lines. By dimensional
analysis, the length of free odd line per unit volume is pj a~2, the average dis-
tance between any site in the glass and the nearest odd line is pj1^*, and the
average time 7 for the odd line to diffuse to that site in order to relax it, is pro-
portionnal to pjla2, so that the fluidity is itself proportionnal to the density of
free odd lines in thermal equilibrium
61
4.6 Crossover to Arrbenfos
The Vogd-Fulcber condensation of free odd lines is only a mean field resvlt,
valid as long as the distribution of free odd lines remains scmi-dihrte (cf. deriva-
tion of the energy (4.3)). This is not valid dose to T#l where pj is very amafl.
How close will be discussed in this section.
The semi-dilate regime corresponds to
where the crossover temperature T, b given by
and eq (4.5). In this r^me. the Tree energy can be linearized, one has a similar
situation to electrodynamics of ~urrent loops, &nd E a
However, when
T.<T<T., p , <{a/!f, ç > í
the distribution of <«\i Hu<*s is dilute, the energy is the sum of individual loop
contributions E a pf. A, ic. lhe r Té" in eq.(4.5) vaol«b, and the density of free
odd lines crosses ovrr to Arrhenius behaviour, pj is the larger than it would have
been in tb^ semi-dilate regime at the same temperature (4.5), and the viscosity
accordingly falls below the Vogel-Fulcher curve.
Whether the crossover to Arrhenius is observable in a given glass, depends on the
non-collinearity length /= l / ( / i (X , i ) . If / is larg*, 7,«* T,, and n{T.) is large,
the crossover is not observable in an experiment lasting a finite time. On the
other hand, the crossover is obervable if / is small. The status of several glasses
is reviewed in Cohen and Grest9.
Although the behaviour of a glass above To can be described by the motion of
odd lines, which show a tendency to freeze in a Vogel-Fulcher fashion, the
existence of a thermodynamic phase transition is offset by crossover to dilute dis-
tribution of odd lines, and to non-linear elastic behaviour. Even within the semi-
dilute regime (4.3), the precise nature of the condensation (its universality class)
is not known, and cannot be obtained from our rough mean field calculation.
For completeness, and without discussion, let me mention different approaches to
the glass transition:
i) All kinematics (no transition)
ü) Statistical mechanics o* long polymer molecules fô. This approximate calcula-
tion leads to a phase transition of a very particular type (the continuous relation
between energy and entropy cannot persist below a finite temperature), but a
counter-example for networks of coordination 2D has been given by Gujrati and
Goldstein M .
iii) Free volume + percolation theory 72>0.
iv) Floppy (underconstrained) and rigid (overconstrained) regions (first discussed
by J.C.Phillips 4t) + percolation theory M.
The experimental situation is reviewed in ref.[85].
The present approach is by far the most economical, both in the number of
parameters and of specific ingredients. We have shown in these four lectures that
the combination of odd lines (as essential structural constituents), and gauge
63
invariance (describing the non generative homogeneity of glass) can explain and
control directly and specifically the universal properties of glasses and super-
cooled liquids.
64
V - STATISTICAL CRYSTALLOGRAPHY
5.1 And now for something completely different
In this lecture, I shall return to the structure, not only of glasses, but of ran-
dom, space-filling, cellular structures in general, which will be referred to as ran-
dom tissues, froths, or mosaics in the case of a 2D pattern. Glasses are
represented in this class of structures by the Voronoi froths of amorphous pack-
ings and metallic glasses. But such structures abound in nature and in inrolun-
tary art (Fig. 10). In 3D, metallurgical aggregates, foams, soap bubble froth,
undifferentiated biological tissues, lead-shot packings (Bernal model for the struc-
ture of a liquid); in 2D, geological jointings (cracked lava flows like the Giant's
Causeway), mud crackings (P1.2), photographic emulsions, etc., all are random
froths.
Fig.10. Wiodow áedm of an archaic, but common variety in England. "The Bell", Narborough,Leici.: Gentlemen lavatory, left window.
65
To a first approximation, all these structures are indistinguishable, apart from an
obvious scaling factor. This structural similarity is exhibited dramatically in
Fig.l of Dormer's book 87I in Figs.2-7 of Weaire's essay w , and in a recent review
of 2D random patterns 17. One notices also (as befits random structures - section
1.3) a considerable variety of cell shapes. It is improbable that specific physical,
or biological forces should be responsible for such identical but variegated archi-
tectural style, so that one looks again for universality: An ideal random space*
filling structure, determined solely by inescapable, mathematical constraints and
the fact that the structure is the most probable one. It is in statistical equili-
brium, in that any topological rearrangement of the cells leaves its "arbitrariness"
invariant, the arbitrariness being measured precisely by the entropy or informa-
tion contained in the structure. One sees immediately the analogy of this pro-
gramme wih statistical mechanics, and with the Maximum Entropy formalism of
probability theory Wi9°. If such an ideal structure exists, it is the (most probable)
representative of an ensemble of structures. It is not unique. Accordingly, criteria
for ideality will be relations between averaged, measurable properties of the
structure, like the ideal gas law in thermodynamics, rather than geometrical data
like unit cells or Bragg spots. One is looking for the statistical analogue of the
simple cubic structure in crystallography, or of the ideal gas law in thermo-
dynamics, from which departures can be measured to identify the forces which
may differentiate between structures at a second order of approximation.
[ A few words of apology: To the addict, random patterns are a cause of many
delightfully wasted hours. By others, it is regarded as an amusing, and rather
trivial hobby, harmless, but without artistic pretensions or cosmic significance. In
this respect, it suffers from its very universality: a subject which, spans beer froth,
crazy paving, cucumber skin and ideal partition of Ireland may not be
66
immediately associated with deep mathematics or ethereal beauty. It is rather
vulgar (Fig. 10). It is also littered with empty experiments (eg. measuring Euler's
theorem, as quoted in ref.[87|), false statements repeated blindly over centuries
(eg. that in 3D, cells are 12 -, or, later on, 14-sided on average lfl>17), and irredu-
cible positions stated in dramatic language: Contrast "There are aspects of tissue
geometry so obvious that they can hardly escape the attention of any person who
seriously considers the question at all. The appreciation that cells are polyhedral
figures came with the very first histological report ever published" (ref.{87), p.7-
8), with a referee's report from Nature "...the paper, which deals with cells as if
they were polyhedra, which they are not...". Incidentally, Dormer is right, but the
paper was rejected. The subject started (in the 17th century) as useless,
because w the standard activity at the time consisted in packing cannon balls,
and random packing of these is unlikely to make one the Ruler of the Queen's
Navy, but (in the 1940's), botanists have taken the trouble to compare plant tis-
sues with the structures of soap bubble» froths, and lead-shot packings, in order
to find out which physical force (surface tension or bard-sphere repulsion) was
responsible for the structure of biological tissues. Without success as all three
types of structures are roughly identicall0. ]
There are also several beautiful examples of man-made and natural foams in
ref.[104]. M.F.Asbby's conclusions (on the mechanical properties of cellular solids)
are very similar to those of this lecture: Importance of the structure in determin-
ing the mechanical properties of cellular solids, which are roughly independant of
the chemical or physical properties of their constituting materials, and universal-
ity of the structures, which are well described by a few parameters: cell aniso-
tropy, open or closed cells, and relative density.
6?
6.2 The random tissue or froth
Random froth labels the class of structures with which we shall be concerned.
It is a maximally random, space-filling, cellular structure. Maximal randomness
(or "random avoidance of the niceties of adjustment" 21) means that all vertices,
edges and faces are structurally stable (their connectivity is unchanged under a
small deformation), ie., in 3D, 4 edges, 6 faces and 4 cells meeting at every ver-
tex, every edge shared by 3 faces and cells, besides the general property that
every face separates 2 cells. In 2D, vertices have 3 incident edges and faces,
besides every edge separating 2 faces. Exceptional vertices with more than 4
edges are not structurally stable in 3D: They can be split into 2 normal vertices
by infinitesimal transformation (half of T l of Fig.2). Probability of their
occurrence in a random packing is negligible. The same holds for exceptional
edges with more than 3 faces. CRN are excluded from this restricted class of
random structures, because their edges may share more than 3 (non-planar) faces,
even if their vertices still have 4 edges (albeit for chemical rather than topological
reason). Consequently, the 3D relation (eq.(5.2) below) between average numbers
of faces per cell and of edges per face, should be modified for CRN. Indeed, it has
already been remarked that <n>£& 7 in CRN 22.
The structural stability of random froths was discussed in lecture I, with Voronoi
froth as paradigm. They exhibit all the vocabulary (dislocations, disclinations)
and the grammar (elementary transformations) of section 1.5.
There are two topological random variables for the 3D froth, the number / of
faces per cell, and the number n of edges per face. For the 2D froth, n is the
only topological random variable. They are not independant: Euler's relation
6S
(1.2), and the valence relations between incident edges and vertices, etc., dis-
cussed above, yield immediately tbe followL •£ topological identities for the ran-
dom froth
< n > = 6 (2D) (5.1)
^ (5-2)
which relate the expectation values of the topological random variables. Eq.(5.2)
is an identity, which holds both for any individual cell in the froth, and for the
froth as a whole. Indeed, there are two statistical problems in 3D, one at the
froth level, described by the random variable / , and the other at the cell level,
where n can still fluctuate (with < " > f t a determined by eq.(5.2)). The two prob-
lems are related by (5.2). All this is summarized in the considerable variety of cell
shapes observed lfl.
Equations (5.1) and (5.2) are the only topological constraints on the statistical
structure of the froth.
In 2D, the expectation value of the topological random variable < n > , is fixed
by eq. (5.1). In 3D, it is found empirically that most random froths have
< / >:=» 14, but it is, emphatically, not an exact result, or even a limit, despite
repeated statements in the literature to the countrary (cf. reis.[19] and [17]). À>
random froth with isotropic cells of equal volumes has < / >=13.40, correspond-
ing to the impossible feat (in Euclidean space) of packing 5.1 equal, regular
tetrahedra in the dual graph (cf. section 1.3). Fluctuations in the volumes of the
celb reduce < / > , whereas fluctuations in the angles (anisotropy of the cells)
increase < / > l8, as observed ««.«.«.«.«. (The high values of < / > observed for
lead-shot packings w , are probably due to uniaxial compression causing cell
b9
aaisotropy, as D.E.G.Williams has aptly remarked). For a Voronoi froth with
centres at random (Poisson distribution), Meijering81 obtains </>p=15.54
exactly. The cells appear indeed highly anisotropic.
We now turn to the correlation between cell shapes (ie. between the topological
variables in neighbouring cells). One knows from experience that large cells tend
to have small neighbours, and vice versa. The precise formulation of this result is
due to Aboav (empirically)*4, and was made plausible by Weaire *5.
6.3 Topological correlations, Aboar-Weaire law and mkroreversibility
In this section, we shall investigate shape correlations between neighbouring
cells of the random froth. We shall use a derivation of Aboav-Weaire's relation
due to Blanc and Mocellin M, mainly because it investigates the random froth
under the elementary structural transformations of section 1.5. In other, fashion-
able words, it follows the froth under structural renormalization. As a bonus, we
shall obtain the same recursion relation (5.3) for shape correlation under all ele-
mentary transformations, Tl, T2, mitosis and their inverses. This implies, first,
microrevcrribility: all these transformations can occur independently of each
other in space or time, without affecting the statistical equilibrium of the struc-
ture. Second, êtatiêtieal equilibrium itself: a random froth is already very close to,
or at a fixed point under structural transformations, with both expectation value
(hi 2D, through (5.1)) and correlation of its topological variable invariant under
these transformations.
Let us start with the froth in 2D. The random variable is n, its expectation value
70
is fixed by eq.(5.1), and the topological correlations are given by m(n), the aver-
age shape (number of sides) of the nearest neighbours of a n-sided cell. For short,
label each cell by the number of its sides. Consider transformation T2 first
(Fig.3): triangular cell d, neighbour to cell a, disappears. Among cell fl's neigh-
bours, cells b and e lose one side in the process (as does cell a). The other cells
(«) remain unaffected. Thus
a m(a)=6+ e+d+t -+ (a-1) m{a-l)—{b-l)+ (c-l)-f t
Since d=3, one has the recursion relation,
a m{a) = {a-1) m(o-l) + 5 (5.3)
Similarly, for a Tl transformation (Fig.2): a - m ' = a - l , 6f =6-1, e ' = e + l ,
rf'=rf+l, «'=«, hence,
a m(a) = (tf-1) m(a-l) + < 6 ' > - 1
But cell b' is no longer nearest neighbour to a (whereas b was, prior to the
transformation), so that, assuming no correlation beyond nearest neighbours, we
have <b' > = 6 (eq.5.1), and obtain the same recursion relation (5.3) under T l as
under T2. Under the same assumption, one obtain the r'th iterate of (5.3) under
mitosis where cell a splits into cells r and a-r+ 4,
a m(a) = r m(r) + 5(a-r)
Since the same recursion relation holds under Tl , T2, mitosis and their inverses,
all these transformations can occur independently in space and time (roicrorever-
sibility), without affecting statistical equilibrium.
7 !
The recursion relation (5.3) is readily solved:
() 5+n
(5.4)
which is the Aboav 84-Weaire °5 law. It is well obeyed experimentally 17. (It was
originally a purely empirical relation M, like the other equation of (the statistical
equilibrium) state, Lewis's law discussed in the next section). A sum rule due to
Weaire,
< n m(n)> = < n 2 > = / i 2 + < n > 2 (5.5)
relates the parameter B to the variance //2 of the distribution of {n}.
In 3D, there are aow two random variables n and / , and there is no relation
involving only / and the average number of faces of cells neighbouring cell / , as
far as I am aware. However, one obtains the same results a» in 2D (Aboav rela-
tion, unique recursion relation, and microreversibility) if one considers what hap-
pens to a n-sided face of a f-faceted cell, and its neighbours under structural
transformations Tl, T2 and mitosis *7, ie. the statistical description of the froth
at the cell level. Consider a given, f-faceted cell. Denote as before, by w»/(n), the
average number of sides of the neighbouring faces to face n on / . Transforma-
tion Tl conserves / , but T2 reduces it: / ' = / - J (T2). Also, there is now a corre-
lation between a' and b' f, despite the fact that they are DO longer nearest
peighbours (cf, Tl), because they belong to the same, finite cell where eq.(5.2)
must hold.
the dust settles, one obtain the 3D Aboav relation °7
^= 5/ - 11 - C{f-\-n) (5.6)
72
where C is a parameter of the froth, independant of / . There has been, so far, no
experimental verification of eq.(5.6). Weaire's sum rale (5.5) (for a given cell) and
eq.(5.2), inserted into (5.0), yield an interesting relation between the variance of
the distribution of sides on a f-sided cell, /»2,/=<[«-<» > / J 2 > / ? *&d f
Calculating bounds for /i2>/ for small cells (/ =4,5), one obtains - 1 < (7<2.
5.4 Ideal random froth, most probable distribution and Lewis's law
Let us define statistical equilibrium and universality of random structures.
We begin with the remark that ail random, space-filling structures enumerated in
section 5.1 are, roughly, identical. They are therefore unlikely to depend on the
^particular physical, biological or chemical properties of their constituting materi-
als, except for their single length scales. Indeed, the very fact that random struc-
tures occur (at least in 2D where there is uo conflict (frustration) between local
packing and global ordering), suggests that specific, short-ranged, directional
forces are less important in -framing the structure1 than the inescapable,
mathematical and .-universal constraints ($1-2, 5.8-0)' pertaining 'to the space
which the cells are filling. (Shortoanged foroes in 2D gite rise to triangular pack-
ings or honeycombs, possibly with a few dislocations). f
The miracle is that these two mathematical constraints are sufficient to frame in
a preei$e, ob$ervable fashion the structures generated under their sole or overrid-. f
ring influence. This is due to randomness and to the fact, well-known in statistical
thermodynamics, that the most probable distribution of cells is overwhelmingly
more probable than any other (cf. comments preceding eq.(4.4)).
The complete statistical problem involves two random variables (in 2D), the
number n of sides per cell - the topological variable, and the area .A of a cell -
the metric variable. To obtain an equation of state, it is sufficient to concentrate
on the topological variable n only. A full calculation of the most probable distri-
bution p(n,A) will be published elsewhere97. Correlation between shapes of
neighbouring cells, discussed in the last section, is an automatic consequence of
statistical equilibrium, and not an additional constraint.
One looks for the most probable distribution, {/>„}, of shapes of the cells in the
structure, where p. is the probability of finding a n-sided cell.
[ p. = / dA p{n,A) , Ãm = f iA p(n,A) A \
It is that distribution which maximizes the entropy or information 8fl
(5.7)
subject to constraints corresponding to our prior knowledge of the system. For
00 random space-filling structures, the constraints are
H = 1 (normalization)
».û = A*IF (spacefilling) (5.8)
$ > . i i = Ô (topology) (5.1)
in 2D. In 3D, (5.8) and (5.1) are replaced by
? = V,/C (3D space-filling) (5.9)
7-1
(3D topology) (S.2)
respectively. Here .4m is the average area of a it -sided cell, and A,, the total area
available to the F cells io the 2D mosaic. Similarly, mvtatis mutandis, in 3D.
This is all. The problem, as formulated, b entirely mathematical. The constraints
are also all mathematical. Physics (or biology,...) is absent at this level, so that
the resulting structures arc universal. The only subjective step is in the coding of
the structure by the sole , topological parameter n, and the requirement that it b
in statistical equilibrium.
We do not need to evaluate the entropy, or the most probable distribution {pn};
the "equation of state" can be obtained by the following argument of Lissowski
and myself v (fully confirmed by a full calculation *7<s7). The constraints are a
linear systems of equations between p%, so that the smaller the dimensionality of
the space of constraints (<3), the larger that of the space of possible solutions
f=(P3tP4,P^f ' ' ' ). and the more probable one such solution will be. The most
probable distribution is obtained by reducing as much as possible the dimen-
sionality of the space of constraints, by making them linearly dependant,
K = (. 1,/F)X[n-(6-l/X)l » {AjF)\Mn 6) + 1} (5.10)
The averaged area of a n sided cell, Ant is linearly related to the number n of its
sides. [The intercept n# = (6-1/X)}. This relation was suggested empirically in
1928 by Lewis "a, on the basis of observations on cucumber epidermis, human
annion and the pigmented epithelium of the retina. Most undifferentiated biolog-
ical tissues obey i t M . It is abo obeyed by Voronoi froth generated from Poisson-
distributed centres M, albeit with a smaller intercept (n#£&0 instead of « 1 2 in
biological mosaics), but not, usually, by metallurgical grain aggregates, as will be
discussed below. It is a relation bet wees averaged, observable parameters of the
random froth, ie. the equation of state of statistical crystallography describing an
ideal froth (in the same sense as "ideal gas" in thermodynamics).
The effective elimination of one constraint, leading to lewis's law (5.10),
increases the entropy further. This is because, as in statistical thermodynamics,
the most probable entropy 5(X,) (5.7) is a convex function of its variables, the
right-hand sides of the constraint equations (S.I) and (5.8), written generally
y,f»cai=Xi, and imposed by Lagrange multipliers X,- n ,
(specific heats are positive). Let i=s be the constraint we wish to eliminate,
imposed by Lagrange multiplier X,. As a function of A,, 5 is maximum when
dS/dX, = 0. But dS/dX, = -X#, and X, = 0 states precisely that constraint s
b no longer operative, in our case because it b no longer independant of the oth-
ers57.
This derivation of Lewis'9 law goes therefore one step beyond the standard appli-
cations of the Maximum Entropy formalbm n>M, in which the functional form of
the constraints b known a priori. Here, we have taken advantage of the liberty
to tijust the functional form of the constraints in order to maximize further the
entropy.
There exbts, at least in principle, an alternative to Lewis's law in decreasing the
76
dimensionality of constraint space: It relates linearly the space-filling constraint
to the normalization, instead of the topology constraint, and yields ,4B=cst,
independant of n, that is no correlation between cell shapes and sizes. Even
though this solution also maximizes further the entropy, it does not occur in
natural froths or mosaics, but for no mathemetical reason as far as I know. This
remarks shows that further increase of the entropy corresponds to a choice
between discrete alternatives.
The parameter X in (5.10) is the undetermined Lagrange multiplier imposing the
linearly dependant constraints (5.1) and (5.8). (The other multiplier is eliminated
because of the linear dependance of the constraints). It is related to the slope
and intercept of Lewis's relation, and is therefore an important descriptive
parameter of the structure. Moreover, Lagrange multipliers have a habit in ther-
modynamics and in mechanics of possessing a physical meaning of their own.
They are not merely arbitrary mathematical factors. What is, therefore, the
meaning of X ? We shall see in next section that it measures the ageing of the
structure.
Lewis's law is not obeyed by 2D metallurgical grain aggregates, whether experi-
mentally, or in computer simulations by the EXXON group 10°, where it is the
gram's radius (or its perimeter) Wn, which is proportional to n, rather than its
area as in Lewis's law. According to the methodology of Maximum Entropy for-
malism, this fact demonstrates the existence of new constraint, besides the
mathematical ones. Clearly, in the computer simulations of grain growth and
statistics by the EXXON group, and as demonstrated experimentally 1 0 \ it is the
energy associated with grain boundary length which is the driving mechanism for
grain growth and statistical equilibrium. [ The EXXON group model grain
statistics and dynamics by a (/=64^oo-state) ferromagnetic Potts model (a gen-
eralization of the Ising model to more than 2 states), on which they carry Monte-
Carlo simulations. Every grain is characterized by a different orientation of the
Potts spin, ie. by a different value of /. All the energy is carried by the inter-
faces, where the "spins" are not "parallel", so that the energy is proportional to
the perimeter (ie. /?„) of the cell ]. The additional, physical constraint is simply
the energy
E P . S . = E (5.H)
This provides an alternative to Lewis's law in reducing by 1 the dimensionality of
the constraint space, namely *7
Rn=a(n-n,') (5.12)
as observed experimentally and in the simulation. One can show that at
(ie. with the Lagrange multiplier imposing the energy constraint (5.11) ^oo), the
maximal entropy using alternative (5.12) is larger than that using Lewis's alter-
native (5.10) ®7. This example emphasizes the diagnostic power of the Maximum
Entropy formalism. A relevant physical constraint has been uncovered by notic-
ing the discrepancy between observed and ideal equations of state.
The most drastic reduction (by 2) of the dimensionality of constraint space, and
the largest entropy, would occur in a polycrystalline aggregate where all con-
straints are linearly dependant, ie. WH ~ An ~ n, or its equivalent in 3D. Such
scaling beween perimeter and area of cells is clearly impossible in 2D, but would
be possible in principle in 3D, if the cell boundaries were fractal (dendritic). But
then, reduction in the energy would no loager be the driving force for grain
growth and statistical equilibrium.
78
It is elementary to generalize Lewis's law to 3D random froths 1S
V> = (VJC)\\f-«f >-l/X)] (5.13)
but this relation has not yet been confirmed experimentally.
5.6 Evolution of a random froth. Von Neumann's law
Many random cellular structures evolve (slowly) in time. In soap bubble
froths, gas diffuses across the interface between bubbles. Biological tissues
undergo growth and cell divisions. In metallurgical aggregates, large grains grow
at the expense of small ones, in a controlled process called sintering, etc. The
time scale for this evolution is much longer than thbt associated with mechanical
response of the structure, so that it can be assumed that the random froth, once
in statistical equilibrium, remains in statistical equilibrium at all times. Conse-
quently, Lewis's law (5.10-13) is expected to hold throughout the evolution of the
structure. The parameters Ao, F, and especially the hitherto undetermined
Lagrange multiplier X, have their own, specific evolution.
The simplest case is the evolution of a 2D froth with constant total area Ao and
constant number of cells F. It refers to time intervals not long enough for bub-
bles to disappear (no T2 process) in a soap bubble froth, or for a cell to divide in
a biological tissue. Differentiating Lewis's law with respect to time, one obtains
the rate of growth of an averaged n-sided cell,
This result was actually derived by von Neumann 102, for 2D soap bubble froths
79
exclusively. (Actually, von Neumann's result for soap bubbles is slightly stronger
than eq.(5.14), as it involves the rate of growth of every individual bubble rather
than its average for all n-sided cells as in (5.14)). Von Neumann's derivation
relied on the physical mechanism of evolution of a soap bubble froth and on
three assumptions specific to this particular sytem, even though the final result
(5.14) is topological: A pentagonal cell loses area at the same rate as a heptagonal
cell gains it, and at half the rate of an octogonal cell, etc., regardless of their
geometrical shapes or areas. Our derivation 103 is completely general and topolog-
ical, so that von Neumann's law (5.14) is applicable to any evolving mosaic, and
generalizable to 3D random froths, since Lewis's law, and statistical equilibrium,
can also be generalized to 3D.
[ Von Neumann's derivation of (5.14) required a) a specific physical mechanism
for the evolution of a soap bubble froth: diffusion of the (incompressible) gas
across the interface between two bubbles, at a rate proportional to the pressure
difference between the bubbles, ie. to the radius of curvature of their interface,
and b) three, crucial assumptions: i) incompressible gas, ii) interfaces meeting at
120' on vertices, and iii) 2D system. Yet, the result is topological. ]
For a soap bubble froth, (A^/F)\ — 7 = (2x/3)&r > 0, where a is the surface
tension of the interfacial liquid, and 6, its diffusivity 102. 7 is therefore a constant
of the froth, invariant in time, at least as long as the interfacial liquid does not
change appreciably in thickness. This provides the physical interpretation of the
Lagrange multiplier X:
X measures therefore the ageing of the - -ructure (or, simply, the time). This is so
80
even if F, A4 and 7 evolve with time (mitosis, T2 process, etc.),
dt (5.15)
but then, the cells which do not evolve in time are no longer hexagonal as in
(S.14), but n |-sided, with n t given by
ft 1 d{AJF)
Computer-generated Voronoi froths with Poisson centres m are "as-quenched",
younger structures (X=l/6 f Lewis intercept »f l=0) than biological tissues 27>M,
which are "aged" (X=l/4, n,=2). Ageing affects also the appearance of the
structure, where cells become more isotropic and diflerentiated in sizes as time, ie.
X or n0 increases.
As Lewis's law, and statistical equilibrium can be extended to 3D froths, von
Neumann's law can also be generalized to 3D structures t03. In particular, over
short periods when C and < / > are constant in time, one has
with < / > given by eq.(5.2). The Lagrange multiplier in Lewis's law still meas-
ures ageing of the structure.
We end with a direct check of statistical equilibrium. From Lewis (5.10), and
<n>=6(5 .1 ) ,
* dt ~ it
81
From the space-filling constraint (5.8),
Si ~ ^p-rfT + ^ A- I T
Since Aa > 0 for all n, one obtains
that is, a stationnary distribution or statistical equilibrium.
6.7 Concrasiona
We have seen that statistical crystallography, a method to describe and clas-
sify the structures of .amorphous materials, can usefully be set up, along the
guidelines of established statistical thermodynamics M. It actually goes one step
beyond statistical thermodynamics, as one can take advantage of the arbitrari-
ness in the functional dependance of one of the constraints to maximize further
the entropy. The constraints (ie. the statistical ensemble) are easily selected:
They arc purely mathematical, pertaining to the topology of the space which the
sructure is filling. There may be additional constraints, which are then physical
or biological. In the absence of these additional, specific constraints, the random
structure is called ideal, by analogy with ideal gas in thermodynamics. Its equa-
tion of state (Lewis's law) is experimentally observable, as is the ideal gas law.
Deviations from Lewis's law betray the presence and structural relevance of
specific (physical, biological,...) forces, which are obviously worth identifying,
exactly as departures from the ideal gas law yield information, through the virial
expansion, on interatomic forces and, ultimately, on other phases of matter. As in
82
statistical thermodynamics, statistical equilibrium, and Lewis's law, are conse-
quences of the balance between entropy (most probable distribution), and the
lowest level of organization (space-filling, territorial partition, encoded by the
constraints).
Acknowledgements
Denis Weaire has kindly let me use hb cartoon (Fig.l) to illustrate topological
equivalence. I am very grateful to him also for all the arguments we have had
over the years about many points raised in these lectures. Figs.2-6 have been
taken from ref.[17], Fig.7 from ref.[S7].
Dorothy Duffy, Andrei Lissowski and Helena GUchrist have made substantial con-
tributions to the material presented in these lectures. Discussions with D.Bovet,
AComtet, N.R.DaSilva, F.Dowell, J.-M.Dubois, C.Eilbeck, H.L.Frbch,
J.P.Gaspard, M.Goldstein, O.J.Greene, H.Guentherodt, J.Jaeckle, R.Kemer,
M.Kleman, J.Malbouisson, R.Mosseri, T.Ninomiya, J.F.Sadoc, J.Sethna, J.Spalek,
G.Toulouse, M.F.Thorpe and D.E.G.WiUiams are gratefully acknowledged.
I am grateful to the CNLS and CMS, Los Alamos National Laboratory, and to
the organizers of the Latin-American Symposium on the Physics of Amorphous
Systems, Niterói 1084, for the opportunity to present and prepare these lectures.
Financial support by the British Council, CNPq, SERC and CNLS-CMS is ack-
nowledged with thanks.
83
References
[1] P.W.Anderson, in III-Condensed Matter, R.Balian, R.Maynard and G.
Toulouse, eds., North Holland (1979), 159
[2] J.L Black, in Amorphou$ Metals / , H. J Guentherodt and H.Beck, eds.,
Springer (1981), 167
[3] G.Toulouse and M.Kleman, J.Physique Lettres, 57 (1976), L-149
G.E.Volovik and V.P.Viineev, Zh.Eksp.Teor.Fiz. Pis'ma, 23 (1976), 647
N.D.Mermin, Rev.Mod.Phys., 61 (1979), 591
L.Michel, Rev.Mod.Phys., 52 (1980), 617
G.Toulouse, Phys.Rep., 48 (1979), 267
For an excellent introduction, see
G.Toulouse, in Modern Trends in the Theory of Condensed Matter,
A.Pekalski and J.Przystawa, eds., Springer (1980), 189
[4] N.Rivier, Phil Mag.A, 40 (1979), 859
[5] W.A.Phillips, Amorphous Solids. Low-Temperature Properties, Topics
in Current Pbysics, 24 , Springer (1981)
[6] P.W.Anderson, B.I.Halperin and C.M.Vanna, Phil.Mag., 25 (1972), 1
WAPhillips, J.Low Temp.Phys., 7 (1972), 351
(7) B.Golding and J.E.Graebner, in ref.[5], (1981), 107
(8) HEHagy, in Introduction to Glass Seienee, L.D.Pye, H.J.Stevens and
84
W.CLaCourse, eds., Plenum (1072), 343
[9] See, for example,
D.TurnbuIl and B G.Bagley, in Treatise on Solid State Chemittri,
N.B.Hannay, ed., Plenum, B (1075), 513
M.H.Cohen and G.S.Grest, Phjs.Rev.B, SO (1079), 1077
[10] C.A-Angell and KJRao, J.Chem.Phys., S7 (1072), 470
C.A.Angell and J.C.Tucker, J.Phys.Chem., 78 (1074), 278
[11] JJaeeUe, PhiLMag.B, 44 (1081), 533
J.Jaeckle, Physica B, (1084), to appear
[12] J.Frenkel, Kinetic Theory of Liquids, Dover (1055), ch.IV.2
[13] D.Weaire, Phys.Rev.Letters, 26 (1071), 1541
[14] D.Weaire and M.F.Thorpe, Pbys.Rev.B, 4 (1071), 2508
A very readable account is given in the lecture notes:
M.F.Thorpe, Some Aspect» of Disordered Solid», UFF Publ, Niterói (1080)
[15] EA.Davb, Pail.Mag., 38 (1978), 463
[16] R.Zallen, in Fluctuation Phenomena, E.W.Montroll and J.L.Lebowitz, eds.,
North Holland (1079), ch.3
[17] D.Weaire and N.Rivier, Contemp.Phys., 26 (1084), 59
[18] N.Rivier, J Physique (Coll.), 43 (1082), C9-01
[10] E.B.Matzke, Am.J.Botany, 33 (1046), 50
See also the "detective story":
E.B.Matzke, BulLTorrey Botan.Club, 27 (1050), 222
85
[20] B.Gruenbaum and G.C.Shephard, Tiling and Patterns, to be published
by W.H.Fretman, ch.iO
R.Mosseri and J.F.Sadoc, in Structure of Ncn-CrustaBine Materials 1082,
RHGaskell, EADavis and J.M.Pwker, eds., Taylor and Francis
(1983), 137
M.Gardner, Scientific Amer., MS / I (1077), 110
[21] F.T.Lewis, Am. J Botany, SO (1043), 74
[22] See for example
P.Steinhardt and P.Chaudhary, Phil.MagJV, 44 (1081), 1375
[23] H S.M.Coxeter, Introduction to Geometry, WUey (1061), ch.6.5
[24] BJ.Gellatly and J.L.Fmney, J.Non-cryst.SoUds, 60 (1082), 313
J.L.Finney, B.J.GeUatly and J.P.Bouquiere, in
Amorphous Material*: Modeling of Structure and Properties, V.Vitek,
ed., TMS-AIME (1983), 3
[25] Glas$ and W.E.S.Turner, E.J.Gooding and EMcigh, eds., Soc. of Glass
Tecbn.(1951)
[26] F.Bloch, Phys Rev B, 2 (1970), 109
[27] F.T.Lewis, Anat.Rec, 38 (1928), 341
N.Rivier and A.Lissowski, J.Phys.A, 16 (1982), L143
[28] Valence alternation pairs have been suggested by Mott, ftreet and Davis,
and independently by Kastner, Adler and Fritsche. See
R.Zallen, The Pkf$ic$ of Amorphous Solids, Wiley (1983), 105
86
[29] Hue and M.F.Thorpe, JNoo-ciystSofids, (1984), to appear
[30] D.Weaire«ad J.P.Kermode, Phil.Mag.B, 48 (1983), 245
[31] F.Wooten and D.Weaire, INoft-eryst-Solkb, (1984), to appear
[32] H.M Princen, XCoUokUnterface^ci., 91 (1983), 180
[33) N Jthrier, R Occdli, J.Pantalom and AAbsowski, J.Pkysique, 46 (1984), 49
[34] F.TLewis, Am J.Botany, M (1943), 786
M.B PyshnoT, JTheorBiol. 87 (1980), 189
[35] MHillert, ActaMet., IS (1965), 227
[36) JEMorral and M.F.Ashby, AcU Met., 22 (1974), 567
[37] N.Rivier and D.M.Duffy, J Physique, 43 (1982), 293
(38) JP.Gaspard, R.Mosseri and J.F.Sadoc, PhU.Mag.B, (1984), to appear
[39] T.Ninomiya, in Topologieil Duoritr m Condensed Matter, F.Yonezawa
and T.Ninomiya, eds., Springer Series in Solid State Sciences 48 (1983),
40
T.Ninomiya, in Structure of Non-Crystalline Material» 1982, ref.(20],
(1983), 558
[40] D.M.Duffy and N.Rivier, J.Physique (Coll.), 43 (1982), CO-475
[41] N.Rivier, J.Physique (Coll.), 39 (1978), C6-984
[42] N.Rivier and D.M.Duffy, J.Phys.C, IS (1082), 2867
[43] D.Bovet, SM Archives, 4 (1979), 31
[44] M.Kleman and J.F.Sadoc, J.Physique Lettres, 40 (1979)4^569
[45] J.P.Gaspard, R.Mosseri and J.F.Sadoc, in Structure of Non-Crystalline
87
MêUnd» 1982, ref.|20j, (1983), 550
D-R.Nekon, P. J-Stenhanlt and MRoechettí, PhysJterB, 28 (1983), 784
R Mosseri, These, Uniy.de Parâ-Sod, Orsay, (1983)
J.P.Sethna, PhysJlerXetten, SI (1083), 2108
[46] J.CPhifflps, Solid State Phys., 37 (1082), 03
J.CPhilhps, Phys.Today, 36 (1081), 27
(47) JTSaáoc and R.Mosseri, in Tofdoficd Disorder m Contented Matter,
ref. [30], (1083), 30
[48] J.F.Sadoc, JPhysiqne Lettres, 44 (1083), L-707
See also
DR.Nebon, Phys.Rey.Letters, 60(1083), 083
[49] G.Toulouse, Commun.Phys, 2 (1977), 115
and in Modern Trend* in the Theory of Condented Matter, ref.(3],
(1980), 195
[50] EFradkin, B.A.Huberman and S.HShenker, Phys.Rev B, 18 (1978), 4789
[51] MKac,Ark.Det.FysbkeSem.Trondheim, 11 (1968), 1
S.F.Edwards and P.W.Anderson, J.Phys F, 7 (1975), 965
The trick has been used to average a logarithm in
G.HHardy, J.E.Littlewood and G.Polya, Inequalitie*, CambridgeUniv.Press
(1934), 6.8
[52] D.Sherrington and S. Kirk pat rick, Phys.Rev.Letters, 86 (1975), 1792
[53] C.L.Henley, H.SompoIinsky and B.IHalpenn, Phys.Rev B, 26 (1982), 5849
83
EMGoWkson, DR.Fred km and S.Sehultz, PhysRerLetters, SO (1083), S37
A-Fert and F Hippert, Phys.RevJLetteis, 4 t (1982), 1508
H Sompolinsky, G.Kotliar and A.Zippehus, (1084), to be published
[54] See for example
AHoughton, S Jain and A.P.Young, J Phys.C, lft (1983), L 375
[55] I.E.Dzy*k»hinskii and CE.Volovik, JPhysique, 5» (1978), 693
(56j JA-Hertz, Phy.RevB, 18 (1978), 4875
[57] N.Rivier, in Structure of No*-Cru$taBitu Maicriol* 1982. ref.[20j,
(1983), 517
NRirier, in Amorpkout Materials, ref.[24], (1983), 81
[58] N.Rivier, mTopologicol Disorder in Conieiuei Matter, ref.{39], (1983), 14
[59] A.Kadic and D.G.B.Edelca, A Gauge Theoru of Dislocation* and
DUchnations, Springer (1983)
[60] EKroeaer «as the first to exploit gauge invariance in elasticity. See
E.Kroeoer, in Physic» of Defect», R.Balian, M.Kleman and J.P.Poirier,
eds., North Holland (1981), ch.3
(61] A.Ck>mtet, Phys.Rev.D, 18 (1978), 3890
[62] B.Julia and G.Toulouse, J Physique Lettres, 40 (1979), L-395
[63] VPoenara and G.Toulouse, J.Physique, 38 (1977), 887
[64] Natanael.R.DaSilva, PhD Thesis, Imperial College, London University (1880)
N.Rivier, ID Theory of Magnetic Alloys, J.Morkowski and S.KIama, eds.,
89
IFM-PAN Poznan, (1980), 67
[651 C.N.Yang and R.L.Mills, Pbys.Rev., 06 (1954), 191
[66] J.C.Lasjaunias, R,Maynard and M.Vandorpe, J.Physique Coll., 80 (1978),
C6-973
[67] R.O.Pohl, in ref.[5], (1981), 27
[68] U.Bart ell, Dr.rer.nat. thesis, Univ. Konstanz (1983)
U.Bartell and S.Hunklinger, J.Physique Coll., 12 (1982), C9-489
[69] H.M.Rosenberg, (1984), to be published
[70] M.Tinkham, Introduction to Superconductivity, McGraw Hill, (1975)
[71] G.AN.Connell and R.J.Temkin, Phys.Rev. B 0 (1974), 5323
See also the spin glass equivalent
D.CMattis, Phys.Lett., 66 A (1976), 421
[72] M.H.Cohen and G.S.Grest, Ann.NY Acad.Sci., 371 (1981), 199
[73] A.R.Cooper, J.Physique (Coll.), 43 (1982), C9-369
[74] T.L.Smith, P.J.Anthony and A.C.Anderson, Pbys.Rev.B, 17 (1978), 4997
[75] R.H.Stolen, J.T.Krauso and C.R.Kurkjian, DiscFaraday Soc, SO (1970), 103
[76] N.Rivier, LT-17 Karlsruhe (1984), and to be published
[77] N.Rivier and H.Gilchrist, (1984), in preparation
[78] SH.Glarum, J.Chem.Phys., 33 (1960), 639
M.C.Phillips, AJ.Barlow and ALamb, Proc.Roy.Soc, A 320 (1972), 193
[79] R.Savit, Rev.Mod.Phys. 62 (1980), 453
[80] D.M.Duffy, PhD Thesis, Imperial College, University of London (1981)
90
[81] E.T.Jaynes, "The Well-Posed Problem", Foundations of Physics, 3 (1973),
477
[82} JH.Gibbs and EA-DiMarzio, J.ChemPhys., 28 (1958), 373
E.A.DiMarzio, Ann.NY Acad.Sci., 371 (1981), 1
[83] P.D.Gujrati and M.Goldstein, J.Chem.Phys., 74 (1981), 2596
[84] M.F.Thorpe, J.Non-cryst.Solids, 57 (1983), 355; and this volume
[85] A.E.Owen, Trieste Lectures (1982)
[86] N.Rivier and D.M.Duffy, in Studies in Critical Phenomena, J .Delia-Dora,
J.Demongeot and B.Lacolle, eds., Springer Series in Synergetics, Q
(1981), 132
[87] K.J.Dormer, Fundamental Tissue Geometry for Biologists, Cambridge UP
(1980)
D.Weaire, in Topological Disorder in Condensed Matter, ref.[39], (1983), 51
E.T.Jaynes, Phys.Rev., 106 (1957), 620; 108 (1957), 171
[90] E.T.Jaynes, in The Maximum Entropy Formalism, R.D.Levine and
M.Tribus, eds., MIT Press, (1979), 15
[91] J.L.Meijering, Philips Res.Rep., 8 (1953), 270
[92] J.W.Marvin, Am.J.Botany, 28 (1939), 280
[93] E.B.Matzke and J. Nest lor, Am.J.Botany, 33 (1946), 130
[94] D.A.Aboav, Metallogr.. 3 (1970), 383
[95] D.Weaire, Metallogr., 7 (1974), 157
[96] M.Blanc and A.Mocellin, Acta Met., 21 (1979), 1231
91
|97] N.Rivier, in preparation (1984)
[98] I.K.Crain, Computers and Geosc, 4 (1978), 131
[99] V.V.Smoljaninov, Mathematical Models of Tissues, Nauka, (IflSO), ch.3
(in Russian)
[100]D.J.Srolovitz, M.P.Anderson, P.S.Sahni and G.S.Grest, Acta Met., (1984),
to appear
P.S.Sahni, D.J.SroIovitz, G.S.Grost, M.P.Andcrson and SASafran,
Phys.Rev.B, 28 (1983), 2705
[101)J.EBurke? Trans.AIME, 180 (1949), 73
[102]J.von Neumann, in Metal Interfaces, Amer.Soc.for Metals, (1952), 108
[103]N.Rivior, Phil.Mag.B, 47 (1983), L-45
(lOi)M.F.Ashby, Metall.Trans.A, 14 (1983), 1755
[105]J..Ioffrin, in III Condensed Mattter, ref.[l], (1979), Gi
92
Phase Transitions in Glassy Materials
M.F. THORPE
Department of Physics and AstronomyMichigan State UniversityEast Lansing, HI 48824
U.S.A.
We examine the mechanical properties of covalent random networks
with high and low mean coordination. It is shown that networks with
high mean coordination (amorphous solids) have elastic constants de-
termined by the covalent forces whereas networks with low mean coor-
dination (polymeric glasses) have elastic constants determined by the
longer range molecular forces.
These ideas can be made rigorous by considering the number of
continuous deformations (i.e. zero frequency modes) allowed within
the network. In the transition from one kind of network to another,
rigidity percolates through the system. This leads to a picture in
which polymeric glasses have large floppy or spongy regions with a
few rigid inclusions. On the other hand in amorphous solids, the
rigid regions have percolated to form a rigid solid with a few floppy
or spongy inclusions.
1. Introduction
In this lecture we will summarize some ideas concerning the
nature of random networks. Although these ideas have not been firm-
ly established either theoretically or experimentally, there is
strong circumstantial evidence that they are correct. We present
these ideas in this paper. Much of this work has been published pre-
viously [1] although some comments relating to elasticity are new.
It is quite reasonable to consider a network consisting of long
polymer chains with a few cross links [2] to be quite different in
terms of its rigidity from a random network describing an amorphous
solid [3] like Si. The former can be deformed easily whereas the
latter is rigid. Ideas somewhat along these lines have led Phillips
[4-6] to postulate the notion of "overconstrained" and "undercon-
strained" glasses. His arguments attempt to delate the entropy and
strain at the glass forming temperature T with the average cuordi-
nation. While we have found his general ldoas to be insightful and
physically applealing, two objections must be raised.
The first Is that any discussion of entropy and strain at T is
bound to be imprecise as the processes taking place are complex and
poorly understood. However the important physical insights of
Phillips caj» be incorporated into arguments that are rigorous if the
following viewpoint is adopted.. Given a network structure at some
reasonably low temperature (< T_ where T_ is the Debye temperature
and T_. < T ), let us enumerate the number of wavs M in which theD g o
network can be continuously deformed with no cost in energy. This
is equivalent to asking for the number of zero frequency modes and
is a well posed mathematical problem. In this paper we argue that
94
for low scan coordination <r> the network is a polymeric glass in
which the rigid regions are isolated as sketched in fig. 1. As the
•can coordination <r> increases, these rigid regions increase in vol-
ume until <r> - r when a percolation transition takes place to a
rigid network or amorphous solid also sketched in fig. 1. Although
this «ay seem at first sight to be just another simple variant of the
percolation problem, albeit with a rather unusual quantity (rigidity)
doing the percolating, in fact the situation is rather more subtle
and complex. This is because the floppy regions contain both rigid
and floppy modes. Indeed the rigidity, associated with finite fre-
quency modes, and the floppiness, associated with zero frequency
modes, have many similarities with extended and localized modes
found in Anderson localization [7]; the finite frequency modes being
extended and the zero frequency modes localized. However, we prefer
to focus on the percolation aspects of the problem as it is the per-
colation of rigidity that drives the transition. In the next sec-
tion we develop a mean field theory which leads .to the transition
taking place at
r p-2.4 (1)
as previously obtained by Dohler et. al. [8].
The second objection is that whilst these ideas lead inevitably
to the concept of floppy and rigid regions and rigidity percolation,
they do not lead inevitably to the formation of molecular clusters
14-6]. What we are saying is that, ideas based on constraints can-
not be used as an agrument to support.the existence of molecular
clusters. On the other hand, the existence of such clusters is not
inconsistent with these ideas either.
95
The layout of this paper is as follows. In Che next section ve
focus attention on the number of zero frequency modes and rederive
the result r =2.4. We note that chis sane result can also be de-P
rived from a mean field theory of the cross linking of polymer chains.
In §3 we argue that this is really a percolation problem. The count-
ing of linearly independent constraints is crucial and not always
easy. In section four we set up a simple soluble model, that con-
tains more some of the features of interest. We show that a third
order phase transition takes place in this model. We comment on the
related problem of spin wave stiffness and infer the behaviour of the
elastic constants in networks.
2. Zero frequency modes
Imagine that a particular network with N atoms has been con-
structed with n atoms having r bonds
rSmall vibrations about this (equilibrium structure are described by a
1/2 £ ^ ( - ^ i , ) 2 + 1/2 Z e <A6 ) 2 , (3)
where Ar.. is the change in the nearest neighbour bond length be-
tween atoms <ij> and A9 . is the change in the bond angle between
two adjacer* bonds <ij> and <jk>. The a.., 6 are force constants
whose precise value will not concern us (an alternative form for the
angular part of eq.(3) such as that used by Keating [9] would not
affect the argument). These forces are known to be the most impor-
tant in covalent molecules and solids [10] and all the comments a-
part from those at the end of section four assume that the network
is described by the potential (3). We seek to calculate the number
96
M of vibrational nodes with zero frequency. These correspond to
ways In which the network can be continuously defomed at no cost in
energy.
In general the number of Bodes with zero frequency M is giveno
by the difference between the number of degrees of freedom 3N and the
number of linearly independent constraints N
M - 3N - N . (4)o c
These constraints are just the number of eigenvectors of the dynami-
cal matrix formed from eq.(3) that correspond to non-zero frequencies.
Put another way N is the rank of the dynamical matrix [11]. The
number of zero frequency modes N could be ascertained directly by
diagonalizing the dynamical matrix or by determining its rank. An-
other way is to estimate the number of constraints . These are;
(A) one per bond that is associated with the first term in
eq.(3) and
(B) for a 2 coordinated atom there is a single angular constraint
associated with the second term in eq.(3). Adding each additional
bond gives two more constraints because the angles with two existing
bonds must be specified. As we are only concerned with continuous
deformations, other discrete possibilities are ignored. This gives a
total of 2r - 3 linearly independent constraints around an atom with
r 2 bonds [8]. This has previously been overestimated [4,8] to be
r(r - 1)11 which gives 1, 3, 6 for r » 2, 3, 4 rather than the
Similar arguments have been used before. In the context of randomnetworks these go back to Weaire and Thorpe [12]; as applied to thevibrational problem see Weaire and Alben [13] and Sen and Thorpe [14]and particularly Thorpe and Galeener [15] where it was shown thatM - 3N - N, for nearest neighbour central forces only in networksWith r • 2, 3 and 4 where N. is the number of nearest neighbour bonds.
97
correct values 1, 2, 5.
Using the constraints associated with both terms in the potential
(3) we have
Mo « 3 j > r - £ n r [ r / 2 + <2r - 3 ) ] ,r r
which setting Mn = 0 yields
rp Z r £ r <>r r
Note that in applying eq. (6), dangling bonds (r = 1) are excluded.
They take no part in the connectivity of the network and can be pro-
gressively removed until no more exist. They are relevant in reduc-
ing the average co-ordination of the remaining network (e.g. H in
amorphous Si, see for example ref. 16).
For <r> > r , the network, is rigid and we refer to it as an amor-
phous solid; whereas for <r> < r , the network is not rigid and can be
macroscopically deformed. This is referred to as a polymeric glass.
Because r is not too much greater than 2, it is instructive to re-
examine an isolated polymer chains with N' atoms where,
M = N + 1.5n , (7)o a e
and n are the number of ends (two/chain). We can imagine a (math-
ematical) cross-linking in which 3 and A co-ordinated sites are form-
ed by fusing together pairs of atoms as indicated in fig. 2. In order
not to have any free ends, the number of 3 co-ordinated atoms n_ = n .
Each time a 3 co-ordinated atom is made, the total number of atoms is
reduced by one (costing 3 constraints to coalesce the atoms) and 2 ad-
ditional angular constraints are required. The situation is similar
when a 4 co-ordinated atom is made, except that 3 rather than 2 addi-
tional angular constraints are required. It there are N atoms when
this process is complete
98
• " Na -»3 ~ V (8)
and
i- 2(N - n, - n.)]/H, (9)
and M Q given in eq.(8) has been reduced to
Mô " Na + 1*5n3 " ° + 2)n3 " ° * 3) V
At this point it is convenient to introduce the quantity f,
which is the fraction of zero frequency modes in the system
f - MJ/3K (11)
and from eqs.(9)-(ll) we find that
f - (12 - 5<r>)/6, (12)
which is shown as the dashed line in fig. 3. The quantity f goes to
zero at <r> • r - 2.4 as before. Of course f cannot be negative and
so vithin this simple mean field scheme we have f = 0 for <r> > r .
This particular way of achieving the final result is somewhat arbi-
trary and other schemes could be used. Using polymer language,
<r> « 2.4 implies that 40Z of all atoms are cross linked if the cross
linkages are all of the type 3 shown in fig. 2 whereas only 20Z are
cross linked if they are all of the type marked 4.
3. Rigidity percolation
The ideas of the previous section argue that covalent networks
can be divided into two classes. Those of type I for which f > o
(polymeric glasses) and those of type II for which f • 0 (amorphous
solids). It is clear that by definition f cannot be negative. If,
in attempt to count the number of constraints, f should appear to be-
come negative, then the constraints are not all linearly Independent
constraints that reduce the number of zero frequency modes.
For a single long polymer chain, there is one bond and one angle
99
constraint per site so that f " i for <r> « 2. Note* that dangling
bonds, if they ever existed, are removed by progressive elimination
as mentioned in the previous section. For <r> close to 2, the ini-
tial slope df/d<r> * - ^ as <r> •» 2 is also given correctly be mean
field theory. However as <r> increases, mean field theory become in-
creasingly unreliable as <r> approaches r . This is because, mean
field theory overestimates the number of linearly independent con-
straints.
In fig. 3 we show in the solid curve the conjectured behaviour
of f against <r>. As <r> is increased from 2, the small rigid re-
gions grow in size until they percolate [17] at r . The quantity f
is not zero at r because snail floppy inclusions exist in the perco-
lated rigid region. The whole solid is rigid for <r> > r . The be-
haviour for <r> - r is harder to conjecture and for this we rely
heavily on the simple soluble model described in section four. This
would suggest that f and its first two derivatives are continuous at
r but the third derivative is discontinuous,P 3
d fr discontinuous at <r> • r . (13)
d<r>3 p
One of the main deficiencies of the simple model in S4 is that
it is completely random, whereas the network situation is much less
so especially when chemical ordering is taken into account (i.e. 2
co-ordinated atoms avoiding each other, etc.). It is harder to say
where the tail terminates and f may well be zero for <r> > 3. This
tall is somewhat akin to a Lifshitz tail [18] depending upon how
random the configurations are above r .
4. A simple soluble nodel
In order to understand better how the effects that we have been
100
discussing occur, consider the following simple model system. Mass
points m are placed at the sites of a (crystalline) lattice and their
motion is described by the Lagrangian
L -SiEaK:.)2 - !j£ a(z. - z.)2 Pii, (14)
j> J J
where z. is the displacement from equilibrium of the mass at site i in
a direction orthogonal to the dimensions that define the lattice geom-
etry. The sum over <ij> is over nearest neighbour pairs. The quanti-
ties p., = 1, 0 with probability p, 1 - p so that we have a bond per-
colation problem [17]. Note that every site is always occupied by a
•ass point. An example would be a square lattice defined in the xy
plane with particle motion in the z direction. Again we ask for the
number of zero frequency modes M . It is easy to see that there is
exactly one such Coldstone mode associated with each isolated cluster
[19-21] so that
Mo = W P ) > + N(1 " P)Z' (15)
where <Y (p)> is the average number of distinct connected clusters per
bond for bond percolation, N is the number of bonds and N(l - p) isb
the number of isolated sites. Therefore we have
* - z/2 <YC(P)> + (1 - P)Z (16)
and che problem of finding f is reduced to that of knowing <y (p)> for
this simple model. An identical result to eq.(16) would be obtained
for the fraction of zero frequency spin wave modes in a bond dilute
*Heisenberg ferromagnet .
In order to link up with our previous notation,, we will define
Another interpretation for f is that it is the ratio of the entropyat zero temperature to that at infinite temperature for a dilute bondspin h Ising model [22]. Each cluster behaves like a spin with twoorientations at zero temperature.
101
<r> • zp. (17)
The quantity <Y (p)> is not known analytically for any real lat-
tice although excellent numerical simulations exist [20J and its over-
all behaviour and critical behaviour is well understood. <y (p)>
plays the role of a free energy [19-21J for the percolation problem
and so f defined in eq.(16) also behaves like a free energy analogue
although of course there is no temperature defined in this problem.
In general f and its first two derivatives are continuous at p so
we may regard the transition at p as being third order [23] (in the
Ehrenfest sense) with
3 3d f/d<r> discontinuous at r . (18)
P
Put another way, a "specific heat" that can be formed from <Y (p)> by
taking two derivatives is continuous, but its derivative is discontinu-
ous. The specific heat has critical exponents [20],
o • a1 • - 0.60 ± 0.1 in two dimensions,(19)
a » a1 » - 0.58 ± 0.11 in three dimensions,
which correspond to a cusp.
In order to have an analytic expression for f for this model we
have examined it on a Bethe lattice. Using the formalism of Fisher
and Essam [24] we find that for p < Pc(i.e. <r> < r )
f - 1 - <r>/2, (20)where zp - r • z/(z - 1). The form (20) is universal for all z.
c pFor p > p , (i.e. <r> > r ) the result depends on z. For z * 3,
f - [(3 - <r>)/<r>]2 (<r> - l)/2 (21)
and for z • 4,
f • [(4 - <r>)/<r»]4 { <r> - 4 + [<r>(16 - 3<r>)]>i } /2. (22)
These results are shown in fig. 4, where the values of r are marked
102
by crosses as a discontinuity in the third derivative is not apparent
to the eye!
For a real lattice vith rings etc. like the square net, it is
possible to count the small clusters and show that for <r> < rP
f - 1 - <r>/2 + (nz/2s)(<r>/z)s + 0(<r>s"1), (23)
where the first two terms are universal and the leading corrections
depend on the size of the smallest rings which have s bonds with n of
them through each edge (2 for the square net). For <r> < z, we have
f - [(2 - <r>)/<r>]Z, (24)
which is obtained by isolating a single site by removing the z bonds
around it. For the square net the value of <Y (p)> at the critical
point is known (25] and from eq.(24) leads to f - 0.16 at r . We have
used this and the results (23) and (24) to sketch the result for the
square net shown in fig. 4.
Adding the next bond may either reduce the number of zero fre-
quency modes by 1 or 0. For small <r>, it will almost always be 1 but
as the density of bonds gets larger it may be 0 if the two sites, as-
sociated with the bond, already belong to a connected cluster. This
means that
- h <_ df/d<r> £ 0, (25)
where the lower value (— h) is attained for small <r> and the slope
gradually goes to 0 as <r> •* z.
The important aspect of this simple model is that there is a
phase transition of the percolation kind 117]; that this phase transi-
tion is third order; and that the behaviour of f with <r> is rather
universal. It is interesting to note that the result for the linear
chair, eq.(20) for <r> <l, is linear all the way like the mean field
103
result of §2 whereas the Bethe lattice results show more curvature
as z increases. The value of f at r lies between 0.16 and 0.33 forP
the two-dimensional networks shown in fig. 4.
It is clear that it would be almost impossible c- determine the
high order derivative in f experimentally, even if some method of
measuring f were to be found. However, fortunately there is another
quantity that manifests the phase transition much more directly and
is easy to measure experimentally. The model described in this sec-
tion is isomorphous to the Heisenberg ferromagnet which has a dis-
persion relation for spin wave excitations
« - Ak2 (26)
where u is the frequency, k the wave vector and A is the spin wave
stiffness. The spin wave stiffness is zero for <r> < r as the sys-
tem is made up of finite islands that cannot communicate with each
other. For <r> > r , the spin wave stiffness increases until it ob-
tains its pure system value at <r> « 4.
By analogy we expect similar behaviour for the elastic constants
in the rigidity percolation problem, with the elastic constants beingzero for <r> < r and finite for <r> > r . This has been demonstrated
P P
in computer experiments on a triangular net with bonds removed at
random [26], The transition takes place at p = — which is well be-
fore the lattice falls apart which is at the percolation concentration
p - 0.3473 where p is the fraction of bonds present. This is be-
cause certain configurations that are effective in carrying current
are ineffective in transmitting rigidity information. We expect sim-
ilar effects in 3D random networks.
Of course in reality the elastic constants in the low coordina-
104
tion polymeric glasses will not be zero but will be determined by
the weaker force constants of longer range. These have values typi-
cal of molecular crystals. In the high coordination amorphous solids,
the elastic constants will be higher and determined by the covalent
forces [i.e., the a and @ terns in equation (3)].
5. Conclusions
We have presented a new framework within which to look at the
differences between random networks with high and low mean co-
ordination. The original ideas of Phillips 14-6] focused on the free
energy F » E - TS at the temperature T - T at which the network is
formed. This is an extraordinarily complex quantity involving both
the energy E and the entropy S. We have adopted a different and
simpler viewpoint in this paper. Given a network at some tempera-
ture, we do not worry about how it was formed or what is its energy
or entropy. We ask only if it can be continuously deformed. This
is a well posed mathematical problem.
We have shown that a mean field theory of rigidity percolation
leads to a linear dependence off on <r> and gives r « 2.4. The
network is divided into two kinds of regions that are designated
rigid and floppy as shown in fig. 1. For <r> < r , we have a poly-
meric glass whereas for <r> > r we have an amorphous solid.
We have shown that the key quantity if f, the fraction of zero
frequency modes. This can never be negative and avoids the mathe-
matically undefined notion of an "overconstrained network" {4]. In-
stead the apparent constraints become more and more linearly depend-
ent as r is approached from below. Indeed it is untrue [4] to say
2that strain builds up like R (where R is the radius) when hand built
105
models are constructed; instead the strain is held within the small
rings and does not propagate. Thus it is possible for both a poly-
mer with a few cross links and a tight random network like amorphous
Si to exist. The discussion then passes to the zero frequency modes.
Whether or not such infinite random networks do actually exist in
nature is beyond the scope of the arguments given in this paper.
We have examined a simple model for which f can be calculated
as a function of <r>. The model shows that there is indeed a phase
transition but that it is the third derivative of f that is dis-
continuous. If this result transfers to random networks, it means
that attempts to locate r by numerically diagonalising dynamical
matrices, or by finding their rank, are doomed. It will be necessary
to examine other quantities such as the elastic constants that are
expected to be more singular at r .
I should like to thank the O.N.R. for supporting this research.
I should also like to thank the organisers of the Latin American
Symposium on Amorphous Materials for a very pleasant week.
106
References
til M.F. Thorpe, J. Non-Crystalline Solids _57_, 355 (1983).
[2] P. Flory, Statistical Mechanics of Chain Molecules (Wiley,Mew York, 1969)-
[3] D.E. Polk, J. Non-Crystalline Solids 5 (1971) 365.
[4] J.C. Phillips, J. Non-Crystalline Solids 34 (1979) 153; 43(1981) 37.
[51 J.C. Phillips, Phys. Stat. Sol. (b) 101 (1980) 473.
{6] J.C. Phillips, Phys. Today 35 (1981) 27.
17) P.W. Anderson, Phys. Rev. 109 (1958) 1492.
18] G.H. flbhler, R. Dandoloff and H. Bilz, J. Non-CrystallineSolids 42 (1980) 87.
[9] P.N. Keating, Phys. Rev. 145 (1966) 637.
[10} K. Nakamoto, Infrared and Raman Spectra of Inorganic andCoordination Compounds (Wiley, New York, 1963).
[11] R.A. Frazer, W.J. Duncan and A.R. Collar, Elementary Matrices(Cambridge, England, 1938) p. 18.
[12] D. Weaire and M.F. Thorpe, Phys. Rev. B4 (1971) 2508.
[13] D. Weaire and R. Alben, Phys. Rev. Lett. 29 (1972) 1505.
[14] P. Sen and M.F. Thorpe, Phys. *ev. B15 (1???) ^ O .
[15] M.F. Thorpe and F.L. Galeener, Phys. Rev. B22 (1980) 3078.
[16] M.H. Brodsky, M. Cardona and J.J. Cuomo, Phys. Rev. B16(1977) 3556.
[17] J. Essam, in: Phase Transitions and Critical Phenomena, Vol.2, eds., C. Domb and M.S. Graen (Academic Press, New York,1974) p. 197.
[18] I.M. Lifshitz Adv. Phys. 13 (1969) 483.
[19] C M . Fortini and P.W. Kastelyn, Physica 57 (1977) 536.
[20] S. Kirkpatrick, Phys. Rev. Lett 36 (1976) 69.
[21] M.F. Thorpe and S. Kirkpatrick, J. Phys. A12 (1979) 1835.
[22] A.R. McGurn and M.F. Thorpe, J. Phys. C 12 (1979) 2363.
107
[23] A.B. Pippard, Classical Thermodynamics (Cambridge, England,1966).
124] M.F. Fisher and J.W. Essaj, J. Math. Phys. 2 (1961) 609.
[25] H.N.V. Temperley and E.H. Lieb., Proc. Roy. Soe. A322 (1977) 251.
[26] S. Feng and P.N. Sen, Phys. Rev. Lett. 52,, 216 (1984).
108
X Polymeric Gloss
Amorphous Solid
Fig.l — The rigid and floppy regions in networks of type I
(polymeric glass) and type II (amorphous solid).
Fig.2 - Isolated polymer chains can be cross linked to form
3 and 4 co-ordinated atoms by fusing pairs of atoms
together.
109
40
Fig.3 - The fraction of zero frequency nodes f versus mean
co-ordination <r>. Rigidity percolation occurs at r .
The dashed curve iV a result of a mean field calcula-
and the solid curve is a sketch of the conjectured be-
haviour with the transition taking place at the cross.
os
. 04 •
Fig.4 - The behaviour of f with <r> for the simple model of
§4. The dashed line is for the linear chain. The
transitions in the solid curves take place at the
crosses. The Bethe lattice results are exact and the
result for the square lattice is estimated (see text).
METALLIC GLASSES: STRUCTURAL MODELS
E. Massif, Depto. de Física, Facultad de Ingenierla, U.B.A.
Paseo Colon 850 (1063) BUENOS AIRES, ARGENTINA
INTRODUCTION: Many models have been proposed in the last
few years to describe the structure of metallic glasses.
These models can be regarded as microscopical descriptions
of the atomic arrangement in the system considered, in
which the fundamental unit is formed by a single atom or a
group of atoms. In order to check the validity of any of
these models, the ability of the proposed structure to re-
produce the main macroscopical features of the analized
system must be tested. From a structural point of view our
most important macroscopical parameters in the case of a
disordered system are basically two: the structure factor
S(Q) and any of the atomic distribution functions [RDF(r) ,
g(r) or G(r)]. To define therefore the efficiency of a
definite kind of structural model, it is necessary to know
accurately the experimental values of those functions, as
well as their partial components from which they can be ob-
tained as a linear combination.
As we shall see this has been possible only in a few
experimental investigations up to now, making difficult to
draw conclusions about the fit of the different known mo-
dels to the atomic structure of the studied system. The
aim of this talk is to give a summary of the attempts made
up to the present in order to describe by structural models
the atomic arrangement in metallic glasses, showing also
why the structure factors and atomic distribution functions
cannot be always experimentally determined with a reasonable
accuracy.
STRUCTURE FACTORS AND DISTRIBUTION FUNCTIONSj The Structure
factor for a one-component system can be defined as:
S(0) * Icoh(q)/Nf2(Q) (1)
Ill
where Q= |{J'-§0|=4ir senÔA is the absolute value of the sca-ttering vector, X is the radiation wavelenght, 20 the sca-ttering angle and ICoh(Q)/N the coherently scattered in-tensity per atom.
- < |A(Q)|2> - f2(Q) < X 2 exp{-i(5(rrrk)>>J *
(2)
where f(Q) is the atomic scattering factor, A(Q) is the am-plitude of the scattered radiation and rj the position ofthe atom j. (The brackets < > denote the statistical avera-ge) .
In the case of a two-component system, there are threedifferent definitions for the structure factor, followingthree different formalisms. Considering the definition ofAshcroft and Langreth (1) we have, for the total structurefactor of a binary alloy:
iíüLílíül S-LÍ» (3,N<f2(Q)> TJ I j <f2(Q)>
where <f2(Q)> = Cjff(Q) + c2f|(Q) being CJJ the atomic con-centrations of the atoms of the species i,j respectively andSjj(Q) the partial structure factors, which verify:
Mm S^(Q) » 0 ; 11m sft(Q) -Mm. S^Q) = 1r ^ ti»j ^^
In the Faber-Ziman formalism (2) the structure factor
can be written as follows: r
SFZ(Q)N
- «f2(Q)> -
22 CfCt f f(Q) fj(Q) . -
- i t J J sff(Q)• J <f(Q)>2 J
with <f(Q)> = Cif^QJ + CafziQ) and where {<f2(Q)>-< f(Q)> 2} *" C^ifi-fa)2 i s a term usually known as the "Laue monotonediffraction", and the partial structure factors verify:
112
"m S^(Q) - 1 Yi.j 1.J - 1,2
Comparing both formalisms one can easily state that(SAL(Q) -i)/(sFZ(Q) -i)<i and TJin SFZ(Q) = TJn SAL(Q) = 1 whichmeans, that the Faber-Ziman formalism is more sensitive tothe oscillations of the structure factor. The Aahcroft-Langreth formalism,1s, on the other hand, simpler, moresensible to concentration fluctuations and with the advan-tage that it does not diverge when <f> * 0 , which occurs inthe case of negative scattering lengths usual in some neu-tron diffraction experiments.
A third description of the structure factor is thatgiven by the Bhatia-Thornton formalism (3) in terms ofwhich the structure factor can be written as follows:
<f2(Q)>
<f(Q)>2 , ^ i j W ^ Q ) }SNM(Q) + SCC(Q)
<f2(Q)> NN <f2(Q)> "
ifi(Q)-f2(Q)>
where SNN(Q) , SCC(Q) are, respectively, the partial struc-ture factors of the correlations between number densityfluctuations and of the correlations between concentrationfluctuations; SNç(Q) is the partial structure factor of thecross correlations between concentration fluctuations anddensity fluctuations. They can be expressed in the form»
SNN(Q) " f <N*(Q) «(Q)> ; SCC(Q) » Ç <C*(Q) C(Q)>
SNC(Q) - Re <N*(Q) C(Q)>
being N(Q) and C(Q) the Fourier transform of the local de-viation in the total number density and In the concentra-tion, respectively. It can be shown that:
Mm SBT(Q) - 1 • 11m SNNCQ) - 11m SCC(Q) ; 11m SNC(Q) • 0
113
Comparing the Bhatia-Thornton formalism with those pre-viously seen, it can be stated that the total structure fac-tors are numerically identical in the BT and the AL forma-lisms, being however the interpretation of the structure interms of the partial structure factors completely different.The Bhatia-Thornton formalism is particularly useful to des-cribe the structure of substitutional alloys, in which theatomic size of the components is almost equal (R1âfR2) andcan be verified SNC«0 . Since the partial structure factorof the correlations between concentration fluctuations. Sec»is extremely sensitive to those fluctuations,these formalismis also particularly effective in those systems with a strongchemical short range order (CSRO). (In the Paber-Ziman for-malism, on the other hand, one has for substitutional alloys
sH-sH-sB) .Whereas the total structure factors are almost directly
available from the experimental data through Icoh(Q) (whichis indeed the measured difracted intensity of the radiation,subsequently corrected for sample absorption, polarization,incoherent or Compton scattering and finally normalized (4)),the distribution functions have a more complicated mathema-tical treatment from experimental data but their physicalmeaning remain much more evident than that of the structurefactors. The so called radial distribution function RDF(r)defined as:
RDF(r) * 4*r2p(r) (6)
(where p(r) is the local atomic density at a distance r
from a given atom), measures the number of atoms in a sphe-
rical shell of radius r and thickness unity, and verifies:
Mm RDF(r) * 47rr2p0
(with PQ being the mean atomic density of the glass). Byintegration of the RDF(r) between two consecutive minima,one can compute the coordination number, that means, thenumber of atoms located between distances R1 and Ru from
114
a given atom:
2 P » Í " RDF(r) dr (7)JR1
where Rl ,Ru define usually the minima just by a Maximumof the RDF(r) determining the p-th shell and therefore thep-th coordination number Zp . Although the RDF is quiteuseful, the pair probability function g(r) is more closelyrelated to the scattering pattern (as we shall see later)and is defined as follows:
g(r) - (8)
where Mm g(r) = 1 .Thr most frequently used distribution function is, ho-
wever, the so called pair correlation function (also calledthe reduced atomic distribution function) G(r), which canbe expressed, following the Faber-Ziman formalism, as:
6(r) = 4*rP0lg(r) -1} = 4wr U>(r) -/>„] (9)
verifying 11m 6(r) » 0 .This atomic distribution function has the advantages,
In comparison with those previously seen, that the densityfluctuations can be more clearly appreciated and that hisexpression as function of the structure factor is simpler,which can be shown in the following:
6(r) « 4rr»0íg(r) - 1] = \ Í"QÍS(Q) - 1] sen (Q.r) dQ (10)Jo
S(Q) • 1 • **/><> frtg(r) -1] s e nW' r) dr
H d r (11,
The total pair correlation function can be writtenalso as a function of the partial terms like:
115
(12b)
<f>2 ^ j t j z )GNN(r) + ~ ~ S — 6 c c ( r )
(fx-f2)GNC(r) (12c)(f2>
in the Faber~Ziman, Ashcroft-Langreth and Bhatia-Thornton
formalisms, respectively.
Two important features must be remarked:
i) the expressions of the total structure factor and the
total pair correlation function as function of the partial
contributions in the Ashcroft-Langreth formalism are asy-
mmetric, and
ii) the partial distribution functions ( gf.(r-) ,/t>;j(r) and
Gjt(r)) are the same in the Faber-Ziman and the Ashcroft-
Langreth formalisms. This is not true, however, for the
partial structure factors.
The GNN(r) partial pair correlation function in the
BT formalism represents the topological short range orde-
ring (TSRO), whereas Gcc(r) are related to the chemical
short range ordering (CSRO) and Gtyçfr) represents the size
effect, which is caused by different atomic volumes of the
components (see, for example, Wagner and Ruppersberg (5a)
or Chieux and Ruppersberg (5b) ) , and equals zero in the
case of substitutional alloys.
Unfortunately, the distribution functions, being one
dimensional, are not capable of discriminating between clo-
sely related models of the structure of metallic glasses,
such as the dense random packing of hard spheres or the
polyhedral model (that we shall see in the second part of
this talk), and more details about the structure must be
determined. For this reason, it is of the greatest impor-
116
tance to calculate the partial structure factors (whichmeans also the partial distribution functions), in orderto determine the local atomic arrangement in metallic gla-sses, In the case of binary alloys, three partial func-tions, SU(Q) , S12(Q) and S22(Q) (°* their correspondingSfw(Q) * SNC(Q) andSçç(Q) ) must be calculated. This can bedone quite readily, in principle, by measuring three inde-pendent total scattering functions S(Q) and solving forthe three unknowns S|j(Q) or S ^ Q ) , since S(Q) is theweighted sum of the three partial functions» The weigh-ting factors depend upon the atomic concentration c; andthe scattering factor fj of the element i in the alloy,as we can see following, for example, eg. (4):
(13)
with;
wfj c | C j
(Faber-Ziman formalism).
in matrix notation like:
with:
IT(Q)J
Equation (13) can be re-written
(14)
Sn(Q)
S22ÍQ)Si2(Q)
MQ)J«22 »Í2
W3W12
where [R(Q)J corresponds to the three partial structure
factors, (w(Q)] represents the weighting factors for di-
fferent scattering abilities, and [T(Q)1 are the total
structure factors observed experimentally for each indepen-
dent event. The solution for the partial structure factors
can be written as:
1R(Q)I - (w(Q)J-1 [T(Q)] = [V(Q)] tT(Q)] (15)
117
A unique solution is found for [R(Q)1 only if the de-terminant of [w] is different from zero. A measure of howgood the equations system is conditioned is given by |w|f, ,the normalized determinant of the system of three equationswhose value can be, at most 1. Unfortunately, in most ex-perimental cases, the normalized determinant for metallicglasses is very small, giving thus an indication of the di-fficulties to be expected in order to solve equation (15).That means, even very small experimental uncertainties leadto drastic uncertainties in the resulting partial func-tions, making therefore very difficult to decide whethera definite structural model describes the atomic arrange-ment of a glassy metal in a better way than others.
In order to separate the partial structures measu-ring the total structure factors by three different expe-riments, several methods are used. In principle they canbe divided into four kinds:
a) the three different radiation techniques using X-rays,neutrons and electrons (see, for example, Paasche (6));
b) the isotope-enrichment technique for neutron diffrac-tion in which the scattering power of the constituents arevaried by using different isotopes (see Sperl (7));
c) the polarized neutron technique which is applicable tomagnetic materials; d) the anomalous scattering techniquefor both X-rays and neutrons (see Waseda (8)) and of course;e) any assortment of the above techniques as, for example
a combination of X-rays with the isotope-enrichment tech-nique (Nold (9)) or a combination of X-rays with the pola-rlzed-neutron technique (see, for example, Sadoc and Dix-mier (10)). From all these attempts to separate the par-tial structures the most succesful was that performed bySperl (7) for the NÍ8iB19 metallic glass, using the isoto-pe enrichment technique for neutron diffraction, who had avalue of 0.51 for the determinant of ]wjn •
With the improvement of the experimental techniques,better data can be obtained because of the new develop-ments of more intense sources and efficient detectors.
118
Some of this techniques (energy dispersive X-ray diffrac-
tion (EDXD) or tine of flight measurements with pulsed neu-
trons) enable us to carry out measurements of the scatte-
red intensities up to Q-40A'1, which permits the evalua-
tion of high resolution Fourier transforms which can pro-
vide a more detailed description of nearest neighbour in-
teractions in binary metallic glasses. Measurements of
extended X-ray «absorption fine structure (EXAFS) and Moss-
bauer spectroscopy provide us a rather accurately informa-
tion about the local atomic structure. For a more detai-
led description of the separation methods and of these new
techniques see, for example, Waseda (8), Hafner (11) or
Egami (12).
STRUCTURAL MODELS: A comparison between the structure of
the amorphous state and that of the liquid state shows
(see, for example, Waseda (8)) that the general features
of the structure of the former are similar to those of
the latter, except for a shoulder on the second peak ob-
served in both S(Q) and g(r) in the amorphous diagrams.
From the similarity of the gross features of the S(Q)
and g(r) for metallic glasses, the fundamental configura-
tion of atoms should be considered as liquid like. On
the other hand, the ratio (r2/rj) between the positions of
the second peak (r2) and of the first peak (r\) in the
amorphous state (about 1.67), is rather similar to the va-
lues obtained for the c/a ratio in close packed hexagonal
structures (1,63) and for the (r$/r\) ratio in a fee
structure (1.73), falling rather far away from the values
of (r2/»*i) and {r$/r\) for the liquid state (about 1.86
and 2.70 respectively). This Implies that the short ran-
ge order of near neighbours In metallic glasses is affec-
ted more or less by the atomic arrangement of the crysta-
lline state.
From these considerations the two basic tendences
that exist in the formulation of structural models for me-
tallic glasses can be well understood: on one side (1) the
mlcrocrystalllne disorder models (also called stereocheml-
119
cally defined models), on the other side (ii) the topologi-cal disordered models.
i) Microscrystalline models
These models are based on the similitudes that have
been observed between the short range order in metallic
glasses and the corresponding crystalline phase at the sa-
me concentration or in the same concentration range. Hama-
da et al (13) proposed a structural model for the Fe-B me-
tallic glass, consisting on "crystalline-embryos" with a
bcc structure (similarly to crystalline Fe3B) surrounded
by a statistical atomic distribution in the boundary re-
gions between the embryos.
The total structure factor S(Q) calculated with
this model reproduces acceptably some structural features
seen with the experimental S(Q) like the reduction of the
shoulder on the right hand side of the second maximum of
the structure factor with increasing boron content.
A good agreement between the model and the measured
data was also observed in Aglf8Cu52 (14) and Fe75P25 (15)
metallic glasses, proposing in both cases a fee structure
for the crystalline embryos.
Kuhnast et al (16) presented a microcrystalline model
to explain the structure of amorphous NiçgB3l, , starting
from the crystalline phase NÍ3B with which he could success-
fully described the local ordering for r<6A . Although a
good agreement could be found in all these cases, efforts
to obtain agreement for other materials have been unsuc-
cessful. Cargill (17) has showed that in order to fit the
experimental structure factors with these models, too ma-
ny parameters (such as sizes and strain distributions of
the mlcrocrystals) must be varied.
We can now summarize the basic properties of the mi-
crocrystalline models, following Gaskell (18):
i) In a A. B alloy, one of the possible local configura-
tion of A and B atoms is preferred basically on energetic
grounds.
ii) This arrangement thus represents the dominant coordi-
nation polyhedron over a wide range of concentrations CA
120
and radius ratios IVR8 ' being the coordination number of
the B species (the smaller atoms) relatively independent
of these quantities.
iii) Differences in structure with concentration and cha-
racter of A and B atoms, and the distinction between amor-
phous and crystalline phases, are described by variations
in the way local structural units are interconnected,
iv) There is a potential local structure equivalence of
the crystalline and amorphous phases.
v) In this kind of models, the long range order dissapears
because of the random orientation of the microcrystals.
Historically, the validity of such "embryo-models" was
supported by some X-ray small angle scattering and trans-
mission electron microscopy measurements that showed the
existence of a number of inhomogeneities in the amorphous
state.
ii) Topological disorder models
These structural models (often called homogeneous di-
sordered models) can be separated in two main groups: a)
dense random packing of hard spheres (DRPHS); b) dense ran-
dom packing of coordination polyhedra (DRPCP).
a) DRPHS Modelst They are the natural extension of the
structural models originally employed in the description
of monoatomic liquids and are based on the general simila-
rities of S(Q) and g(r) between the liquid and amorphous
states. The starting point of all these models is the hard
sphere model first proposed by Bernal (19). The spheres
are dense in the sense that no "internal holes" great
enough to accomodate another sphere can be found, and
they are at random because only weak interactions are pre-
sent between spheres separated by four or five atomic dia-
meters. That means, there are no regions of long range or-
der as in the crystalline structures.
In Bernal1s model, the topology is described in terms
of polyhedra (also called "Bernal holes") in which the ver-
tices are defined by the sphere centers. Bernal found 5
different types of polyhedra: 1) tetrahedta, 2) octahedra,
121
3) archimedean trigonal prisms, 4) archimedean antiprisms,
5) tetragonal dodecahedra. The introduction of archimedean
polyhedra prevents from any long range order and from the
realization of the ideal structure precisely. This model
was later improved by J.L. Finney (20), who obtained the
radial distribution function for a dense random packing of
about 8000 hard spheres with much better resolution than
previously available.
Finney's model was used to fit the experimental va-
lues of the reduced radial distribution function G(r) ob-
tained by Cargill (17) for the NiygP^ metallic glass.
The results were rather satisfactory since we take into
account that only one parameter namely the atomic size
(R^jsRp) was varied.
Polk (21) proposed a model consisting of a Bernal
structure (DRPHS) which is primarily metallic, with the
metalloid (the smaller atoms) filling some of the larger
holes inherent in the random packing. Polk suggested
that it is this special relationship between the metal
and the metalloid which could lead to the stabilization
of the amorphous structure. Unfortunately, none of the
holes are as large as originally believed by Polk, not
allowing, therefore, their filling up by the metalloid
atoms. Polk (22) later generalized his view of the DRPHS
void-filling model to allow the metal atoms to occupy
random packing structures somewhat less dense than those
of Bernal and Finney, which should provide more larger
holes to accomodate the metalloid atoms.
Ichikawa (23) constructed a DRPHS model suggesting
some modifications of Bennett's criterion (24). Bennett's
algorithmus consisted on adding spheres to 'an initial equi-
lateral triangle formed by three spheres, enumerating all
possible sites for which an added sphere would be in hard
contact with three spheres already in the cluster, but
would not overlap with any of them and selecting among
them the nearest to the center of the cluster. In his mo-
del, Ichikawa introduced a measure of the tetrahedral per-
fection of the pocket formed by three spheres, defined as:
T = r^/ÍRj+Rj) (16)
where rf*x is the maximum distance between the centers of
spheres i and j in the distribution of three spheres, and
R} and Rj are their radius respectively. In this model
the assembly of hard spheres with diameter T is conside-
red, in which the tetrahedron of four spheres is first
constructed. Then the other spheres are arranged in a
growing process of tetrahedral clusters, under the condi-
tion that the desired sphere occupies only the position of
having been in contact with three spheres of surface within
the distance T - Thus, deviation from the condition T=1.0
allows the formation of a slightly deformed tetrahedron.
Ichikawa suggested that the model calculation with the pa-
rameter T=1.2 (which corresponds to a relatively rigid
packing of tetrahedrons) is more compatible with the expe-
rimental data for the amorphous state. He showed also that
the packing became obscure for T=2.0 , that is the funda-
mental tetrahedra are distorted, and in this case the spli-
tting of the second peak in both S(Q) and g(r) is no
longer observed, a fact that is in good agreement with the
experimental results for the liquid state. This means
that the amorphous structure requires something to be
added to the simple model structure of hard sphere liquids.
In fact more rigid configurations of atoms must be consi-
dered, this corresponding to the introduction of the para-
meter for tetrahedral perfection.
The main problem of this kind of model is that if
atomic arrangements in the amorphous metal-metalloid
alloys were truly like those of spheres in such dense
random packings, then the metalloid atoms would have the
same average surroundings as the noble or transition metal
atoms, a fact that, as showed by several experimental obser-
vations, is unlikely. Binary DRPHS models with smaller
spheres representing metalloid atoms rand with no metalloid-
metalloid nearest neighbours is probably a more realistic
structural model for the metal-metalloid amorphous alloys,
than simple dense random packing of equal size hard spheres
123
with metal and metalloid atoms occupying random sites. Such
a model was proposed by Sadoc etal (25) who used Bennett's
algorithm to construct hard sphere packings with two sizes
of spheres in which the smaller spheres, representing meta-
lloid atoms, were not allowed to occupy adjacent sites*
The introduction of small spheres requires the large sphe-
res to form a looser random packing to provide enough lar-
ge holes to accomodate the small spheres. This tendences
could be even observed in the liquid state, as suggested
by this author (26) for the Ni-B melt.
Cargill and Kirkpatrick (27) obtained a rather good
fitting of the experimental RDF(r) with a binary DRPHS
model using Bennett's global criterion for rare earth-
transition metal alloys, namely Tb3 3Pe6 7,Gd36Fe6it and
GdigCo62* Waseda et al (25,28) used a binary model based
on the previous works from Ichikawa (23) and Cargill and
Kirkpatrick (27), to describe the experimental behaviour
of the partial structure factors Sjj(Q) and pair distri-
bution functions g|j(r) for amorphous Fe-P and Cu-Zr.
An important improvement of binary DRPHS models was that
made by Connell (29) and von Heimendahl (30) who relaxed
the model through energy minimization of the assumed pair
potential. It is important to note that the relaxation
does not conserve the original topology, which means that
one has to make a topological analysis after the structu-
re was relaxed. Boudreaux (31) relaxed a DRPHS model using
a Lennard-Jones potential for the Fe-B system at different
concentrations finding a coordination number of 6.6 around
the boron atoms. This leads Finney (32) to propose a mo-
del based on local structural units (trigonal prisms) with
six iron atoms around a boron one, like in crystalline
Fe3B (cementite structure) with the rest of the boron
atoms distributed as a "statistical adhesive" between the
molecular units. Some investigators (see, for example,
Gaskell (18)) questioned this criterion, pointing out
that the model gives a poor information about that "statis-
tical cement". Another improvement of the DRPHS models
has been carried out through softening of the hard spheres,
allowing them to oscillate around their equilibrium sites
with a defined frequency which corresponds to a given tem-
perature. This leads to a broadening of the first coordi-
nation shell in tha pair correlation functions. With such
dense Random packing of about 5000 soft spheres, Blétry
(33) obtained rather satisfactory results to fit the very
accurately experimental results of the partial structures
determined by Sperl (7) with the NiaiB19 amorphous
alloy. Other calculations based on the same experimental
data were made by Beyer and Hoheisel (34) using molecular
dynamics and taking a Iennard-Jones potential for a sys-
tem under conditions similar to those of the liquid state.
Though this model made reasonable predictions for the va-
lues of the GNf_f|f(r) and GNJ_B(r) partial pair correla-
tion functions, the values of the experimental Gg_g(r)
show considerable differences with those predicted by the
model. Furthermore, the assumption that the glassy alloy
could be handled as a binary liquid alloy on thermal equi-
librium at high temperatures seeias to be unlikely,
b) DRPCP models; The DRPHS models seen before start from
the similarity of the pair correlation functions in the
liquid and in the amorphous states. While this remains
obviously true for the long and medium range structure,
the more recent diffraction experiments support the point
of view that the short range order in the amorphous phase
may be similar to the crystalline structure. Thus one
may begin with larger local units having the desired coor-
dination and a topology borrowed from the crystalline
structure. The model is built by a random stacking of tho-
se local units. In this senss some investigators (see,
for example, Gaskell (13)) supported this kind of models
because they allow a better description of the CSRO (che-
mical short range order) than the DRPHS models. The
opposite argument is that the danse random packing of
coordination polyhedra models doesn't take into account
satisfactorily the random elements , namely the disorder.
Two of the most important D?J?CP models are those proposed
by Takeuchi (35) and Wright (36) for monoatomic systems
125
based on geometrical considerations. As showed by Hoare
and Pal (37), the icosahedron is the I3-atoms cluster con-
figuration with the lowest energy. The icosahedron can be
separated in 20 almost regular tetrahedra. Considering all
atoms of the icosahedron as equivalent, the whole amorphous
structure can be idealized as an array of tetrahedral units.
The model is constructed in such c. way that each new tetra-
hedron is put above one of three faces of the previous te-
trahedron so that the vertex of the new tetrahedron comes
to the farthest distance from the origin. The average
coordination number obtained in this way for the first
shell is about 13.4 which agrees rather well with the va-
lues obtained experimentally for most of the metallic gla-
sses. Sperl (7) presented a very good agreeement between
the normalized atomic distances (rj/n)f»if2... , obtained
from his experimental partial pair correlation function
GNj_Nf(r) and the values taken from Wright's (36) and Ta-
keuchi's (35) models. Other DRPCP models considering lar-
ger fundamental units have been presented, such as that of
Kuhnast et al (16), who proposed a dense random packing of
icosahedral units to describe the structure of amorphous
Ni71B2g. Finally, another important DRPCP model, succes-
fully employed for the Pd-Si amorphous alloy, is that pre-
sented by Gaskell (38) who takes as fundamental units a tri-
gonal prism (with 6 Pd atoms around a Si atom) as well as a
trigonal prism capped with three half-octahedra (9 Pd atoms
around a Si atom) with a similar topology as the cementlte
structure (crystalline Fe3C). The model was relaxed using
Lennard -Jones potentials and supplementary conditions to
maintain the coordination after the relaxation procedure.
This structural model was also employed to reproduce the
above mentioned experimental data of Sperl (7) . The results
were not very satisfactory: a better agreement between the
model and the measured data for the pair correlation func-
tion would require a greater diameter of the metalloid
atoms. Furthermore, the first boron-boron atomic distance
calculated with this model is somewhat smaller than the one
obtained from the experimental curves.
126
The essential features of the homogeneous disordered
models (DRPHS and DRPCP) can be thus summarized as follows:
i) In a Aj_xBx amorphous alloy, the smaller atoms B, are
mainly surrounded by A atoms. This has been experimentaly
demonstrated for some glassy metals and it seems to be a
general phenomenon in these materials.
ii) The atomic arrangement in local coordination polyhedra
is essentially at random and dictated by geometrical (radius
ratio) and compositional factors alone. There is no prefe-
rence, energetic or otherwise, for a particular local sy-
mmetry or coordination number so that the average local
geometry becomes a function of the radius ratio and the
concentration of each species (as pointed out by Jansen
(39)).
CONCLUSIONS: We can summarize the former discussions saying
that three main descriptions can be found for the structure
of metallic glasses: a) Stereochemically defined models; b)
Dense random packing of hard or soft spheres; and c) Dense
random packing of polyhedral units.
There are obvious similarities between these three ty-
pes of models. In each case the structure consists of a
dense, space filling arrangement of local structural units.
The unit, however, can range from a single atom through a
group containing, say, five or seven atoms to a cluster of
several hundreds. The topologlcal disordered models, how-
ever, seem to describe the structure of glassy metals in a
better way than the microcrystalline models and to provide
a more realistic framework in which one can analize the me-
chenical, magnetic and electronic properties of these mate-
rials. In view of the increased Information contact it is
of the greatest importance that not only total but also
partial distribution functions might be accurately determi-
ned for these amorphous alloys and in this way there would
be a lot of experimental work to do.
127
REFERENCES
I . Ashcroft N. and Langreth D.. Phys. Rev. 159,500 (1967)2. , Faber T. and Ziman J.. Phil. Mag.. 11,153 (1965)3. Bhatia A. and Thornton D.. Phys. Rev., B2.3004 (1970)4. Wagner C . Journal of Non Cryst Solids, 31,1 (1978)5a. Wagner C and Ruppersberg H., Atomic Energy Review, 1.101 (1981)5b. Chieux P. and Ruppersberg H., Journ. de Physique, C8.41,145 (1980)6. Paasche F.. Doktorarbeit, Univers. of Stuttgart (1981)7. Spert W., Ooktorarbeit. Univ. of Stuttgart (1982)8. WasedaY., 'The structure of norxrysUiline materials". M e Graw-Hill (1980)9. Nold E., Doktorarbeit. Univ. of Stuttgart (1981)10. Sadoc J, and DixmierJ., Mat Sci.Eng. 23,187(1976)II. Hafner J., •'Glassy Metals I" ed. by H. Guntherodt and H. Beck, Springer Vertag (1981)12. Egami T., in "Glassy Metals I"13. Hamada et al. Proceedings of the RQM IV (4th. International Conference on rapidly quenched
metals) Sendai (Japan) 1981.23 (3)14. Wagner C , Light T.B., Haider N. and Lukens W., Journal of Appi. Phys., 39,3960 (1968)15. Waseda Y.. Okazaki H. and Masumoto T . , Sci. Rep. Res. Inst Tohoku Univers., 26a. 202 (1977)16. Kuhnast F., Machizaud F. and Ftachon J., Journ. de Physique, C8,41,250 (August 1980)17. Cargill Ml G., Solid State Advances in Research and Applications, Vol. 30,227 (1975) Academic
Press, New York18. Gaskell P., Proceedings of the RQM IV, Sendai (Japan) 1981, p. 24719. Bemal J.. Proc R. Soe. A280, 299 (1964)20. Finney J., Proc R. Soe., A319,479 (1970)21 . Polk D., Src Metall, 4,117(1970)22. Polk D., Acta Met, 20,485 (1972)23. Ichikawa T., Phys. Stat. Sol. (a). 29,293 (1975)24. Bennett C , J. Appl. Phys., 43,2727 (1972)25. Sadoc J., Dixmier J., Guinier A., Journal of Non-Cryst Solids, 12,46 (1973)26. Nassif E., Lamparter P., Sedelmeyer B., Steeb S., Z. fur-Naturforsch, 38a, 1098 (1983)27. Cargill DIG., Kirkpatrick S.. AtP Conf. Proc., 31,339 (1976)28. Waseda Y., Masumoto T., Tomizawa S., Can. Met. Qurt, 17.142 (1977)29. Connell G., Solid State Communic, 16,109 (1975)30. von Heimendahl K., H. Phys. F., 5, L141 (1975)31. Boudreaux D., Phys. Rev., B18,4039 (1978)32. Finney J., Wallace J., Journal of non-cryst. solids, 43,165 (1981)33. Blétry J.. Z. fur-Naturforsch. 33a. 327 (1977)34. Beyer O., Hoheisel C , Private Communication, Univers. Bochum (1983)35. TakeuchiS.,KobayashiS.,Phys.Stat.Sol.(a)65,315(1981)36. Wright J., Inst. Phys. Conf. Ser., number 30,251 (1977)37. HoareM.. Pat P., Adv. Phys., 20,161(1971)38. ' Gaskell P., Journal of non-crystalline solids, 32,207 (1979)39. Jansen H.f Boudreaux D. and Snljders H., Phys. Rev., B21, 2274 (1980)
128
AMORPHOUS HETALS - FABRICATION AND CHARACTERIZATION
Frank P. Hissell
Instituto de Física, Universidade de São Paulo,
C P . 20516, Sio Paulo, S.P.
We present a review of the principal methods used to
prepare metallic alloys in the amorphous phase, focussing on
two which have come into widespread use: melt-spinning and
sputtering. Methods of characterizing the amorphous state
«re discussed. Finally we discuss briefly a kinetic approach
to understanding the formation of metallic glasses.
•*••• •'•"•• • • • • • ' • • - • ' • • . I
Probably the earliest appearance of amorphous metals
was in the form of films of Ni-P and Co-P, obtained by elec-
troless deposition. Electroless plating or deposition is the
controlled autocataiytic deposition of a continuous film by
the interaction of a metal salt in solution with a chemical2
reducing agent . The Ni-P alloy possesses good corrosionresistance and hardness and was used as a replacement for
hard chromium. Thus, the initial interest in these materials
was exclusively for engineering purposes and only later did
their amorphous nature become an object of study.
Films of pure metals were prepared in the amorphous
state by Bucket and Hi lsch* Who evaporated metals in high
vacuum and deposited the vapor onto a substrate at He-temp-
eratures. The atoms lose their energy rapidly upon striking
the substrate and the formation of a periodic crystal is
avoided. In this was, the pure metals Bi and Ga can be ob-
tained in the amorphous phase, while other metals like Pb,
Sn, In, Tl, etc. can be forced into the amorphous phase by
adding 10-20% of another component . The effects of prepar-
ing metals in this manner are dramatic: crystalline Bi is
not a superconductor, while amorphous Bi possesses a super-
conducting transition temperature T "6.1 K. The low temp-
erature electrical resistivity also shows large variations
from the crystalline state. However, amorphous Bi cannot
129
be s t u d i e d at room t e m p e r a t u r e b e c a u s e , at a t e m p e r a t u r e of
about 20 K, it u n d e r g o e s an i r r e v e r s i b l e t r a n s i t i o n to the
crystal 1i ne state .
A s i g n i f i c a n t a d v a n c e in the s t u d y of a m o r p h o u s m e t a l s
was m a d e in 1959 when D u w e z , W i l i e n s and Klement d e v e l o p e d
a t e c h n i q u e for the " s p l a t " q u e n c h i n g of m a t e r i a l s . T h e i r
" g u n " t e c h n i q u e d e v e l o p e d c o o l i n g rates g r e a t e r than 10 K / s e c
by f o r c i n g a d r o p of liquid a l l o y to impinge v e r y r a p i d l y
o n t o a h i g h l y c o n d u c t i n g s u b s t r a t e . S o o n a f t e r w a r d , t h e s e
a u t h o r s p r o d u c e d a thin ( M O y ) f l a k e of Au_,.Si2l- (all c o m p o -
s i t i o n s are a t o m i c %) a l l o y w h o s e x - r a y d i f f r a c t i o n p a t t e r n
s h o w e d no s t r o n g peaks a s s o c i a t e d w i t h B r a g g s c a t t e r i n g . The
m a t e r i a l p r o v e d to be u n s t a b l e at room t e m p e r a t u r e , a n d , a f -
ter Ik hours d e c o m p o s e d into p h a s e s w i t h c o m p l e x c r y s t a l
s t r u c t u r e s . The w o r k of D u w e z e_£ ^J_. , h o w e v e r , g e n e r a t e d m u c h
interest b e c a u s e they p r o d u c e d an a m o r p h o u s metal w h i c h could
be e x a m i n e d at room t e m p e r a t u r e . T h e i r s was a l s o the f i r s t
use of x-ray d i f f r a c t i o n to p r o v i d e e v i d e n c e of the a b s e n c e
of long range c r y s t a l l i n e o r d e r in a rapidly q u e n c h e d m e t a l .
In 1 9 6 7 , D u w e z and Lin p r o d u c e d a t e r n a r y F e - P - C a l l o y n e a r
the e u t e c t i c c o m p o s i t i o n w h i c h p r o v e d to be f e r r o m a g n e t i c ,
s t a b l e , and t o t a l l y a m o r p h o u s . The p o s s i b i l i t y of p r o d u c i n g
c h e a p , F e - b a s e d , a m o r p h o u s m a g n e t s p r o v i d e d f u r t h e r impetus
for w o r k in this a r e a .
A n o t h e r s i m i l a r m e t h o d for o b t a i n i n g the rapid s o l i d i f i -Q
c a t i o n o f m e t a l s is t h e p i s t o n a n d a n v i l t e c h n i q u e . I n t h i s
c a s e , a d r o p o f m e t a l i s f o r c e d f r o m t h e b o t t o m o f a c r u c i b l e
i n s u c h a w a y a s t o t r i p a p h o t o c e l l a n d a c t i v a t e a p i s t o n
w h i c h q u e n c h e s t h e d r o p a g a i n s t a n a n v i l . T h e r e s u l t i s s i m i -
l a r t o t h a t o f t h e g u n t e c h n i q u e , a n d o n e o b t a i n s a s p l a t o f
a c e n t i m e t e r o r t w o in d i a m e t e r . A r a t h e r s o p h i s t i c a t e d v a r i -
a t i o n o f t h i s m e t h o d h a s r e c e n t l y b e e n u s e d t o p r o d u c e h i g hq
q u a l i t y s a m p l e s f o r U P S a n d X P S m e a s u r e m e n t s . In A r g e n t i n aa p i s t o n a n d a n v i l a p p a r a t u s is i n u s e a t U n i v e r s i d a d d e
B u e n o s Ai r e s .
T h e f o r m a t i o n o f c o n t i n u o u s m e t a l l i c f i l a m e n t s o f a m o r -
p h o u s a l l o y s p r o b a b l y h a s i t s o r i g i n i n t h e w o r k o f L a n g
w h o m a d e s o l d e r w i r e o n t h e p e r i p h e r y o f a r o t a t i n g d r u m .
130
M o r e r e c e n t l y , the m e l t - s p i n n i n g p r o c e s s of Pond e m p l o y e d
the inner s u r f a c e of a r o t a t i n g c o n c a v e d i s h to r a p i d l y cool
a j e t o f l i q u i d m e t a l . C o n t a c t w i t h the s u b s t r a t e w a s e n h a n -
ced by the c e n t r i f u g a l f o r c e . P o n d ' s i n v e n t i o n w a s i m p r o v e d12by P o n d a n d M a d d i n w h o u s e d the inner s u r f a c e of a r o t a t i n g
d r u m to cool t h e l i q u i d . T o a v o i d the s u p e r p o s i t i o n o f l a y e r s ,
the c r u c i b l e w a s t r a n s l a t e d p a r a l l e l to the a x i s of the d r u m ,
p r o d u c i n g a c o n t i n u o u s h e l i c a l r i b b o n . It w a s w i t h t h i s d e v i c e
that M a s u m o t o and M a d d i n p r o d u c e d a l e n g t h ('VjjO c m ) o f r i b -
bon o f a m o r p h o u s P d o o S * 2 0 'n " 7 1 «C o n t i n u o u s f a b r i c a t i o n o f m e t a l l i c f i l a m e n t s w a s a c c o m -
1 Lp l i s h e d by P o l k a n d Chen , u s i n g a C h e n - M i l l e r t w i n r o l l e r
d e v i c e . In t h i s c a s e , a c o n t i n u o u s a m o r p h o u s r i b b o n w a s
f o r m e d by q u e n c h i n g a j e t of l i q u i d metal b e t w e e n two r o t a t i n g
r o l l e r s . T h i s d e v i c e w a s u s e d to- s u r v e y the g l a s s f o r m i n g1 ka b i l i t y o f a n u m b e r of Fe a n d / o r Ni a l l o y s . T h ? s t e c h n i q u e
w a s a d e q u a t e f o r m a k i n g s c i e n t i f i c s a m p J e s , but it w a s s o m e -
w h a t d i f f i c u l t to c o n t r o l t h e p o s i t i o n of the j e t o f l i q u i d
m e t a l and to o b t a i n a d e q u a t e q u e n c h r a t e s . For this r e a s o n ,
B e d e l l and W e l l s l a g e r , at A l l i e d C h e m i c a l , t r i e d d i r e c t i n g
the l i q u i d m e t a l a g a i n s t the o u t e r s u r f a c e of a c o p p e r d i s c ,
r o u g h l y 10 cm in d i a m e t e r , r o t a t i n g w i t h a s u r f a c e v e l o c i t y
o f a b o u t 30 m / s e c . A l t h o u g h t h i s d e v i c e did n o t b e n e f i t f r o m
c e n t r i f u g a l f o r c e to m a i n t a i n g o o d c o n t a c t b e t w e e n t h e s u b -
s t r a t e a n d the m e l t , it w a s , in f a c t , s u c c e s s f u l a n d is t h e
b a s i s of the t e c h n i q u e n o w k n o w n as m e l t - s p i n n i n g . L a t e r ,
B e d e l l s h o w e d h o w the c o n t a c t b e t w e e n w h e e l and m e t a l c o u l d
be e n h a n c e d a n d the p o i n t of d e p a r t u r e of the r i b b o n c o n t r o l -
led.
In F i g . 1 we s h o w a s c h e m a t i c d i a g r a m of a m e l t - s p i n n e r
s u c h a s t h a t w h i c h m i g h t be u s e d to p r o d u c e s a m p l e s f o r r e -
s e a r c h p u r p o s e s . T h e d i s c is f r e q u e n t l y of c o p p e r (The a p p r o -
p r i a t e f i g u r e o f m e r i t f o r the c h i l l s u b s t r a t e is kdC , the
p r o d u c t of the t h e r m a l c o n d u c t i v i t y , d e n s i t y a n d s p e c i f i c1 8h e a t . S i n c e c o p p e r h a s a h i g h e r kdC p r o d u c t than o t h e r
m e t a l s , its u s e is j u s t i f i e d . ) and r o t a t e s w i t h a s u r f a c e
v e l o c i t y w h i c h Is c o n t i n u a l l y v a r i a b l e up to a b o u t 50 m / s e c .
131
The sample is heated in a quartz tube with an induction coil
to the melting point. H o w e v e r , the high surface tension of
the liquid metal prevents it from leaving the crucible. The
application of a pressure of Ar gas above the me!t forces it
out a small hole in the crucible onto the rotating disc, where
it forms a ribbon of the amorphous alloy.
The cooling rate of the metal during melt spinning is
easily estimated. For e x a m p l e , a F e . - N i . ^ .B, alloy melts at
a temperature of about 10009C and is cooled to approximately
room temperature. The ribbon maintains contact with the wheel
during an angle of about 109, which, for a surface velocity
of 30 m/sec, corresponds to a time of 10 sec. Thus the cool-
ing rate is d T / d t M O K/sec.ARGON
PRESSURE.
FUSEDSILICATUBE-
MOLTEN ALLOY
AMORPHOUSALLOYFILM
COIL
ROTATING DISC
Fig. 1. Schematic diagram of a m e l t - s p i n n e r .
A discussion of ribbon formation in a melt-spinner was1 a
g i v e n by l i e b e r m a n n a n d G r a h a m p in t e r m s of the B e r n o u l l i
e q u a t i o n . T h e s e a u t h o r s o b t a i n e d an e q u a t i o n f o r the r i b b o n
c r o s s - s e c t i o n a l a r e a A in t e r m s o f P, the a p p l i e d g a u g e p r e s -
s u r e of the e j e c t i o n g a s , d, t h e d e n s i t y of t h e l i q u i d m e t a l ,
D , t h e d i s c d i a m e t e r , S, the f r e q u e n c y of r o t a t i o n , a n d 9, the
l i q u i d j e t d i a m e t e r , t a k e n e q u a l to the o r i f i c e d i a m e t e r :
132
A »r
The above equation was shown to be in quantitative agreementwith experimental data for fretnN'li0B20 rib')ons* T n e u s e
Bernoulli equation determines the conditions under which acontinuous ribbon will be formed in the melt-spinner. It has
20also been used to discuss ribbon formation by Takayama and 01 .
A study of the thermal and energetic constraints on rib"18bon formation in a melt-spinner has been made by Kavesh f who
also obtained equations relating the ribbon dimensions to the
process parameters. In this case, the fundamental parameters
were taken to be the chill surface velocity V and the flux Q
of material incident on the chill surface. Kavesh considered
two different possible modes of ribbon formation:(1) a mode
in which the ribbon is formed by transport of momentum and/or
heat between the melt puddle and the chill surface, and (2)
the case where the ribbon is formed by wetting the substrate
and, therefore, the ribbon dimensions are determined by the
equilibrium of energy at the chill surface. In the first case,
Kavesh concludes that heat transport is dominant over momen-
tum transport and that the ribbon width w and thickness t are
given by: w»c"Q n/V 1" n and t-Q 1" n/c"V n, with n-0.75- In the
case where energy balance is the important mechanism for rib*
bon formation, the ribbon dimensions are given by wMi and
t M / V . Experimental data for Fe. nNi . nP,. B, were found to be In
agreement with the following equations: w»0.$253 Q ' /V '
and t-1,585 Q O è 1 7 / V 0 ' 8 3 . Thus it is difficult to decide which
mechanism Is dominant in ribbon formation. For the same mater-21iai, H M I m a n and Hiizinger also observed w and t to decrease
with increasing velocity as W W ' and tM/ , in substan-
tial agreement with Kavesh.
The melt-spinner has become a widely used technique for
preparing metallic glasses from the melt and is used by many
experimental groups. In Brazil, melt-spinners are In use at
Universidade de São Paulo and Universidade Federal do Rio
Grande do Sul, and, in Argentina, at Centro
Atômico Bariloche . Several other institutions are In the
133
p r o c e s s of i n s t a l l i n g this d e v i c e .
W h i l e c o n t i n u o u s r i b b o n s can be p r o d u c e d by the above
p r o c e s s , the w f d t h of the ribbon is limited to about 5 mm.
T o p r o d u c e w i d e r r i b b o n s it is n e c e s s a r y to e s t a b l i s h a rec-
t a n g u l a r melt p u d d l e on the chill s u r f a c e and this cannot be
p r o d u c e d by a c y l i n d r i c a l j e t . A l s o , r e c t a n g u l a r j e t s d e g e n *
o r a t e rapidly d u e to the low v i s c o s i t y and high s u r f a c e ten-
sion of the m e l t . T h e p r o b l e m of p r o d u c i n g w i d e r ribbons w a s22solved by M, C, N a r a s i m h a n w h o invented the p l a n a r flow
c a s t i n g p r o c e s s - A s c h e m a t i c view of the p l a n a r flow casting
p r o c e s s is shown in Fig. 2, In this c a s e , the m o l t e n metal is
Substrate
Fig. 2 S c h e m a t i c v i e w of the p l a n a r flow c a s t i n g p r o c e s s .
forced through a r e c t a n g u l a r o r i f i c e in the b o t t o m of a c r u -
cible in close p r o x i m i t y to the chill s u b s t r a t e - Flow is be-»
sically p r e s s u r e c o n t r o l l e d but is c o n s t r a i n e d by orifice
w i d t h * s u b s t r a t e s p e e d and o r i f i c e - s u b s t r a t e d i s t a n c e . This
p r o c e s s or s i m i l a r v a r i a t i o n s have been used to p r o d u c e
ribbons w i t h w i d t h s up to 10 cm. Commercial p r o d u c t i o n of Fe-
based wide a m o r p h o u s ribbons is c u r r e n t l y u n d e r w a y by the
A l l i e d C o r p , , V a c u u m s c h m e l z e GmbH, and Hitachi M e t a l s Ltd,
A n o t h e r t e c h n i q u e w h i c h has come into w i d e s p r e a d use for
the p r o d u c t i o n of a m o r p h o u s alloys is s p u t t e r i n g . This p r o c e s s23
w a s d i s c o v e r e d • by Sir W i l l i a m Robert Grove in 1852 when he
n o t i c e d that the w a l l s of the d i s c h a r g e tube that he was
S t u d y i n g b e c a m e coated w i t h metal from the e l e c t r o d e s . Later
F. Stark gave the first c o r r e c t e x p l a n a t i o n of the p h e n o m e n o n
«nd J. vJ. T h o m p s o n p r o d u c e d the name " s p l u t t e r i n g " to d e s c r i b e
134
the process. Sputtering was used commercially as early as
by the Western Electric Co. to deposit a thin conducting
metallic film onto wax phonograph records. The mechanism of
dc (direct current) sputtering was studied by Wehner e± al.
With the development of rf (radio frequency) sputtering ,
applications for this technique expanded greatly. Recent
Improvements In vacuum technology have turned sputtering into
a major industry.
The basic mechanism of sputtering is illustrated in
F|g. 3» where the diode sputtering configuration is shown.
An expendable target is attached to the cathode, while the
«node supports the substrates onto which the target material
I» to be deposited. The process is carried out in a vacuum
chamber, which is first evacuated to high vacuum and then
filled with a low pressure of argon gas. The cathode voltage
TARGET M*T£Bl*l
ÍUISTSATES
Fig. 3 Pi ode sputtering configuration.
Is adjusted to - 500-600 V and a plasma is created in the
vacuum chamber. The ionized Ar are attracted to the cathode,
and., upon collision with the target, the kinetic energy of
the; Ar ions is transferred to atoms of the target material.
The target material is then deposited onto the substrates.
Thç system functions in either a dc or rf mode when the tar-
get material is a metal. However, the rf excitation mode must
be used when the target material is a dielectric in order to
«void charge accumulation on the cathode, which would reduce
the sputtering rate,
135
O t h e r s p u t t e r i n g c o n f i g u r a t i o n s h a v e been used in o r d e r
to i n c r e a s e s p u t t e r i n g r a t e s . For e x a m p l e , the t r i o d e c o n f i g -
u r a t i o n has s i g n i f i c a n t rate a d v a n t a g e s o v e r the d i o d e m o d e
but at a c o n s i d e r a b l e i n c r e a s e in the c o m p l e x i t y o f the s y s -26tem . High s p u t t e r i n g rates are a l s o o b t a i n e d w i t h m a g n e t r o n
s p u t t e r i n g . In this c a s e , a m a g n e t i c field is a l i g n e d p e r p e n -
d i c u l a r to the e l e c t r i c e x c i t a t i o n f i e l d o f the s p u t t e r i n g
s y s t e m . T h e m a g n e t i c field t r a p s and c o n f i n e s the e l e c t r o n s
in the p l a s m a to a r e g i o n v e r y c l o s e to the target s u r f a c e ,
r e s u l t i n g in high ion c u r r e n t s and h i g h s p u t t e r i n g r a t e s . T h e
dc m a g n e t r o n m o d e is used to s p u t t e r m e t a l s , w h i l e the rf
m a g n e t r o n e x c i t a t i o n m o d e is used for d i e l e c t r i c s .
The s p u t t e r i n g t e c h n i q u e has a s e r i e s of a d v a n t a g e s w h e n
c o m p a r e d to o t h e r thin f> 1m d e p o s i t i o n p r o c e s s e s . For e x a m p l e
c h e m i c a l v a p o r d e p o s i t i o n (CVO) p r o c e s s e s can be u s e d to p r o -
d u c e a b r o a d s p e c t r u m of m a t e r i a l s . H o w e v e r , a m a j o r d i s a d -
v a n t a g e of the C V D p r o c e s s is that m a j o r c h a n g e s m u s t b e m a d e
in the a p p a r a t u s in o r d e r to p r e p a r e d i f f e r e n t m a t e r i a l s .
V a c u u m e v a p o r a t i o n u s i n g r e s i s t i v e or e l e c t r o n beam h e a t i n g
of the melt is a r e l a t i v e l y s i m p l e and e a s i l y c o n t r o l l e d m e -
thod. H o w e v e r , the d e p o s i t i o n rates vary g r e a t l y w h e n the v a r
ious c o m p o n e n t s of an a l l o y have d i f f e r e n t vapor p r e s s u r e s ,
and so it is d i f f i c u l t to d e p o s i t c e r t a i n a l l o y s . T h e s e d i s -
a d v a n t a g e s are not p r e s e n t in s p u t t e r i n g . The e q u i p m e n t is
v e r s a t i l e in that o n l y the target need be m o d i f i e d to p r o d u c e
a d i f f e r e n t a l l o y . F u r t h e r m o r e , target m o d i f i c a t i o n s m a y be
a c c o m p l i s h e d s i m p l y by c o v e r i n g part of the target w i t h a n o -7 7 9ft
ther m a t e r i a l ' . D e p o s i t i o n rates in the s p u t t e r i n g p r o -
cess are a l s o m u c h m o r e u n i f o r m than for e v a p o r a t i o n .
The e f f e c t i v e q u e n c h rate d u r i n g s p u t t e r i n g may be e s t i -
m a t e d from the o b s e r v a t i o n that the d e p o s i t e d atom will lose
its k i n e t i c e n e r g y in a time c o m p a r a b l e to a few ionic v i b r a -
t i o n s . T h u s , if the atom is in e q u i l i b r i u m w i t h the p l a s m a
at ^ 5 0 0 ? C , and loses its e n e r g y in a time 5 x 10 s e c , then1 U
the e f f e c t i v e q u e n c h rate will be d T / d t ^ 10 K / s e c . Thei n c r e a s e d q u e n c h rate d u r i n g s p u t t e r i n g c o m p a r e d w i t h p r o -
c e s s e s such as m e l t - s p i n n i n g has a p r a c t i c a l c o n s e q u e n c e :
136
amorphous alloys are formed over a much wider composition
range using sputtering. For example, for the 2r-Rh system»
the eutectic occurs around 25(atomic)$ Rh and melt-spinning
may be used to produce amorphous alloys containing 20-301 Rh.
On (he other hand, with sputtering, amorphous alloys
containing as much as 43% Rh may be obtained . The properties
of alloys made by sputtering are generally comparable to those
produced by melt-spinning .
Sputtering has been gsed to produce amorphous alloys at
Universidade de $ao Paulo and Universidade Estadual de Cam-
pinas in Brazil. In Argentina, a sputtering system is being
installed at Centro Atômico Bariloche.
Another technique with a very high effective quench rate
is ion implantation, in this case, beams of ions ar^ acceler-
ated by an electrostatic particle accelerator and are direc-
ted onto a substrate. Using accelerating voltages pf kOO kV,
it is possible to implant ions to a depth of around 3000A.
Thus it is possible to form thin films of amorphous alloys
on the surface of a substrate. A variation of this technique,
known as ion beam mixing, uses a beam of rare gas tons such
as Ar or Kr to mix atoms deposited in a series of very thin
(MQ0 8) layers. The resultant mixing produces an amorphous
alloy , In Brazil, ion implantation techniques are in use
at Universidade de Sao Paulo and Universidade Federal do Rio
Grande do Sul.
Finally, we mention that amorphous alloys have recently
been formed by solid state reactions. For example, an amor-
phous Zr-Ni phase has been produced by annealing vapor depo-32sited crystalline layers of elemental Zr and Ni . Thus the
already lengthy list of techniques for producing amorphous
metals fs constantly growing.
Characterization of the amorphous state begins with an
x-ray diffraction sc^n. The graph of x-ray intensity vs,
scattering angle shows no strong narrow peaks, such as those
which result from Bragg scattering. Instead, there are a num-
ber of weaker broad maxima, the most intense of which
137
g e n e r a l l y o c c u r s at the angle c o r r e s p o n d i n g to the most inten-
se Bragg p e a k . In Fig. k w e show x-ray intensity v s . angle
28 for three samples of ^ o n B ^ o * The uppermost trace c o r r e s -
ponds to a sample which is totally a m o r p h o u s , while the m i d d l e
trace belongs to a sample showing slight signs of crystal U n -
ity. In the lower t r a c e , the Bragg peaks are already becoming
e v i d e n t , for this sample is much more c r y s t a l l i n e than the
o t h e r two. The first indication of an a m o r p h o u s s a m p l e , t h e r e -
fore, is the absence of Bragg p e a k s .
Fig. k X-ray intensity v s . angle 28 for three samples ofF e 8 0 B 2 O '
For the u p p e r m o s t trace of Fig. k, one can use the
S c h e r r e r e q u a t i o n to make a simple e s t i m a t e of the m a x i m u m
dimension 0 of the region of the sample from which there is a
coherent s c a t t e r i n g c o n t r i b u t i o n to the broad x-ray maximum.
If B is the full width at half maximum of the intensity curve
and X is the w a v e l e n g t h of the radiation used, then
0-0.9 A/ B c o s 8 . For the upper trace of Fig. k, D ^ 2 o 8 . More
detailed structural information, such as the radial d i s t r i b u -
tion f u n c t i o n , mav be obtained from a detailed numerical
analysis of the x-ray diffraction v
A rapid way to test samples to determine if they are
amorphous is to make use of the 180? bend test. This test,
138
w h i c h w o r k s for m a n y F e - b a s e d a l l o y s , is b a s e d on the f a c t
that m a n y a m o r p h o u s a l l o y s a r e m o r e d u c t i l e than t h e i r
c r y s t a l l i n e c o u n t e r p a r t s . For e x a m p l e , the s a m p l e c o r r e s p o n d -
ing to the u p p e r m o s t t r a c e o f F i g . k w a s a b l e to s u s t a i n an
a c u t e 180? bend w i t h o u t b r e a k i n g . The m i d d l e s a m p l e a l m o s t
s u s t a i n e d the s h a r p b e n d , w h i l e the l o w e r s a m p l e b r o k e b e f o r e
It w a s bent t h r o u g h an a n g l e o f 9 0 ? . T h i s t e s t w o r k s for
r e l a t i v e l y d u c t i l e m a t e r i a l s , but m u s t be u s e d in c o n j u n c t i o n
w i t h x - r a y d i f f r a c t i o n m e a s u r e m e n t s .
A n o t h e r p r o p e r t y w h i c h is q u i t e d i f f e r e n t in the c r y s -
t a l l i n e and a m o r p h o u s s t a t e s is the e l e c t r i c a l r e s i s t i v i t y
and the t e m p e r a t u r e d e p e n d e n c e of the r e s i s t i v i t y . D u e to the
I n c r e a s e d a t o m i c d i s o r d e r in the a m o r p h o u s s t a t e , the e l e c -
t r o n i c m e a n f r e e p a t h is g r e a t l y r e d u c e d in t h e s e m a t e r i a l s .
T h u s , the e l e c t r i c a l r e s i s t i v i t y of m e t a l l i c g l a s s e s is h i g h
( 1 0 0 - 2 0 0 pfi-cm) and v e r y s i m i l a r in m a g n i t u d e to that of the
liquid s t a t e . S i n c e the t e m p e r a t u r e c o e f f i c i e n t of the r e s i s -
t i v i t y is a l s o v e r y small ( ( 1 / p ) d p / d T ^ 1 0 ~ 5 / K ) , as in the
liquid s t a t e , one m i g h t e x p e c t that the t h e o r y of liquid
m e t a l s w o u l d be u s e f u l in u n d e r s t a n d i n g r e s u l t s in m e t a l l i c
g l a s s e s . S u b s t a n t i a l p r o g r e s s has b e e n m a d e in t r e a t i n g the
teer.36
r e s i s t i v i t y o f liquid m e t a l s , and t h e s e r e s u l t s h a v e been
v e r y u s e f u l in i n t e r p r e t i n g d a t a in m e t a l l i c g l a s s e s '
One must e x e r c i s e a c e r t a i n c a u t i o n , h o w e v e r . Early e x p e r i -
m e n t a l w o r k s s o m e t i m e s a f f i r m e d that a n e g a t i v e t e m p e r a t u r e
c o e f f i c i e n t o f the e l e c t r i c a l r e s i s t i v i t y w a s in i t s e l f a n
i n d i c a t i o n of the a m o r p h o u s s t a t e . M o r e recent t h e o r e t i c a l
w o r k , h o w e v e r , has s h o w n that a m o r p h o u s m a t e r i a l s m a y h a v e
e i t h e r p o s i t i v e or n e g a t i v e t e m p e r a t u r e c o e f f i c i e n t s o f r e -
s i s t i v i t y , and t h e s e c o n c l u s i o n s h a v e been c o n f i r m e d by e x -
p e r i m e n t .
G l a s s e s a r e u n s t a b l e r e l a t i v e to the c r y s t a l l i n e s t a t e
and c r y s t a l l i z e w h e n h e a t e d b e y o n d the c r y s t a l l i z a t i o n t e m p -
e r a t u r e T • T is r o u g h l y one h a l f of the m e l t i n g t e m p e r a t u r ex x
T and is u s u a l l y s l i g h t l y h i g h e r than the g l a s s t e m p e r a t u r eT . T h e c r y s t a l l i z a t i o n p r o c e s s o c c u r s as an e x o t h e r m i c9
r e a c t i o n and is c l e a r l y v i s i b l e in the s p e c i f i c h e a t , in
139
c o n t r a s t t o t h e g l a s s t r a n s i t i o n , w h i c h is u s u a l l y d i f f i c u l t
t o o b s e r v e . T w o t e c h n i q u e s w h i c h a r e u s e d t o s t u d y t h e t h e r -
m a l b e h a v i o r o f g l a s s e s a r e d i f f e r e n t i a l t h e r m a l a n a l y s i s
( D T A ) a n d d i f f e r e n t i a l s c a n n i n g c a l o r i m e t r y ( D S C ) . D S C w a s
f i r s t a p p l i e d t o t h e s t u d y o f t h e g l a s s t r a n s i t i o n in t h e3 7a m o r p h o u s a l l o y A u _ _ G e 1 _ /;Si Q . by C h e n a n d T u r n b u l l '.
B o t h a l l o w T to b e e a s i l y d e t e r m i n e d . A s t u d y o f T a s aX X
f u n c t i o n of the s a m p l e h e a t i n g r a t e a l l o w s a d e t e r m i n a t i o n
of the a c t i v a t i o n e n e r g y for c r y s t a l l i z a t i o n .
V a r i o u s t h e o r i e s for the f o r m a t i o n o f m e t a l l i c g l a s s e s1,0
h a v e b e e n p r o p o s e d . B e n n e t t , P o l k , and T u r n b u l l d i s c u s s e d
the r o l e of c o m p o s i t i o n in m e t a l l i c g l a s s f o r m a t i o n , in p a r t ,
b a s e d on a s t r u c t u r a l m o d e l o f a m o r p h o u s a l l o y s w h i c h had
b e e n p r o p o s e d e a r l i e r by P o l k . N a g e l and T a u c c o n s i d e r e d
the r o l e o f the e l e c t r o n gas in g l a s s f o r m a t i o n a n d s t a b i l i t y
of a m o r p h o u s a l l o y s . T h e s e t h e o r i e s tend to e q u a t e s t a b i l i t y
w i t h e a s e of g l a s s f o r m a t i o n . On the o t h e r h a n d , in a k i n e t i c
a p p r o a c h to g l a s s f o r m a t i o n , o n e a t t e m p t s to d e t e r m i n e t h e
q u e n c h r a t e n e c e s s a r y to o b t a i n the a m o r p h o u s p h a s e by
c o o l i n g f r o m the l i q u i d p h a s e . T h u s o n e a t t e m p t s to c a l c u l a t e
a c r i t i c a l c o o l i n g r a t e R for the s y s t e m u n d e r c o n s i d e r a t i o n ,c hi
T h i s k i n d of a p p r o a c h has b e e n a d o p t e d by D a v i e s and by
U h l m a n n and o t h e r s . A r e v i e w of t h e s e t r e a t m e n t s h a s b e e n
g i v e n by D a v i e s . We d e s c r i b e b r i e f l y the k i n e t i c a p p r o a c h
w h i c h s e e m s to be the m o s t r e l e v a n t to t h e e x p e r i m e n t a l fact
that h i g h c o o l i n g r a t e s are n e c e s s a r y for the f o r m a t i o n of
m e t a l 1Ic g l a s s e s .
U h l m a n n a d o p t e d the c l a s s i c a l J o h n s o n - M e h i and A v r a m i
t r e a t m e n t of t r a n s f o r m a t i o n k i n e t i c s , r e l a t i n g the f r a c t i o n
of the t r a n s f o r m e d p h a s e to the h o m o g e n e o u s n u c l e a t i o n f r e -
q u e n c y , t h e g r o w t h r a t e and the t i m e . P h y s i c a l l y , the f o r -
m a t i o n of c r y s t a l l i n e s o l i d s from l i q u i d s t a k e s p l a c e by a
n u c l e a t i o n and g r o w t h p r o c e s s b e c a u s e the a t o m i c s t r u c t u r e
of a s o l i d , w i t h t r a n s 1 a t i o n a l and p o i n t s y m m e t r y , is i n c o m -
m e n s u r a t e w i t h the s t r u c t u r e of the l i q u i d . T h e i n t e r f a c e
b e t w e e n the c r y s t a l l i n e and l i q u i d p h a s e s c o r r e s p o n d s to
140
Lo..;,o time
Fig. 5 Time-temperature-transformation curve.
locally distorted atomic arrangements and the energy neces-
sary to produce the interface is positive. Thus, even below
the solidification temperature, where the energy of the
crystalline phase is lower than the liquid phase, it takes
a finite time for crystalline nuclei to reach a critical size
for growth into larger grains. Using accepted theories of
crystal growth and homogeneous nucleation, as well as esti-
mated values for the liquid viscosity between T and T ,
Uhlmann constructed time-temperature-transformation curves
for a detectable fraction of a crystal. The crystallization
time becomes shorter at first as the driving force due to
cooling below the solidification temperature is increased.
As the temperature is reduced, however, the diffusivity of
the liquid phase is reduced, which, in turn, slows down the
growth of crystalline nuclei, and the time necessary for
crystallization increases. Thus one obtains a curve such a^
that of Fig. 5 above. The critical cooling rate for glass
formation is then' approximately given by the cooling curve
necessary to avoid the nose of this time-temperature-trans^
formation curve: R « (T -T )/t , where T and t are respec-c m n n n n
tively the temperature and time corresponding to the nose of
the curve. As shown by Davies and Uhlmann, the kinetic
approach gives good agreement with experiment (depending upon
the accuracy of the viscosity data) for glasses ranging from
SiO, to P d 8 2 S ' i g to amorphous Ni. The kinetics of glass for-
141
mat!on by cooling from the melt is governed largely by the
reduced glass temperature T /T . Through this kinetic theory,
the formation of metallic glasses has been approximately
rationalized in terms of theories of nucleation and crystal
growth.
REFERENCES
1. A. Wurtz, C.R.Acad.Sci.Paris ±&, 702(18**);2±, 1*9(
A.Brenner and G.Riddell, U.S.Patent 2,532,283 (Dec.5, 1950).
2. G. A. Krulick, in Encyclopedia of Semiconductor Technology,
ed. M. Grayson (John Wiley, New York, 198*)p. 1*5.
3. W. Buckel and R. Hilsch, Z. Physik V3_8, 109 (195*).
*. G. Bergmann, Phys. Rep. 27, 159 (1976).
5- P. Ouwez, R.H. Wi liens, and W.Klement, J. Appl. Phys. 31,
1136 (I960).
6. W. Klement, R. H. Willens, and P. Duwez, Nature Jjtt, 869
(I960).
7. P. Duwez and S.C.H.Lin, J.Appl. Phys. }£, *096 (1967).
8. I. S. Miroshnickenko and I. V. Sal 1 i , Ind. Lab. 2J5, 1*63
(1959).
9. P. Oelhafen, in Glassy Metals ll,_ed. H. J. Guntherodt
and H. Beck (Springer, New York, in press).
10. E. M. Lang, U. S. Patent 112,50* (Feb. 21, 1871).
11. R. B. Pond, U. S. Patent 2,825,108 (March *, 1958).
12. R. B. Pond and R. Maddin, Trans. Met. Soe. AIME 2*5,
2*75 (1969).
13. T. Masumoto and R. Maddin, Acta Met. JMi, 725 (1971).
1*. D.E.Polk and H.S.Chen, J.Non-Cryst. Solids J_5, 165 (197*).
15. H.S.Chen and C.E.Miller, Rev.Sci . I nstrum.jn_, 1237(1970).
16. See L. A. Davis, N. J. DeChristofaro, and C. H. Smith, in
Proc. Conf. Met. Glasses:Sci. and Tech., Budapest, 1980.
17- J. Bedell, U.S.Patent 3,862,658 (Jan. 28, 1975).
18. S. Kavesh, in Metal 1i c Glasses, ed. J. J. GiIman and H.J.
Leamy (Amer. Soc. Metals, Metals Park,OH, 1978)p. 36.
19. H. H. Llebermann and C. D. Graham,Jr.,IEEE Trars. Mag.
1-12, 921 (1976).
142
20. S. Takayama and T. Oi , J. Appl. Phys. J50.4962 (1979).
21. H. Hilman and H. R. Hilzinger, in Rapidly Quenched Metals
III, ed. B. Cantor (Metals S o c , London, 1978)vol . 1 ,p.3O.
22. H. C. Narasimhan, U.S.Patent 4,142,571 (Mar. 6, 1979).
23. W. R. Grove, Phil. Tran. Roy. Soc. J_42, 87 (1852).
2^. S. K. Wehner, Advances in Electronics and Electron Physics,
VII I (Academic, New York, 1955)p. 239-
25. G. S. Anderson, W. N. Hayer and G. K. Wehner, J. Appl.
Phys. JJ), 2991 (1962).
26. Sputter Deposition and Ion Beam Processes, publication
of the Education Committee of the American Vacuum Society.
27. J. J. Hanak, J. Vac. Sci. Technol . £, 172 (1970.
28. A. K. Ghosh and D. H. Douglass, Phys. Rev. Lett. 37' 32
(1976).
29. F. P. Missell, S. Frota-Pessôa, J. Wood, J. Tyler, and
J. E. Keem, Phys. Rev. J3£7, 1596 (1983).
30. F. P. Missell, R. Bergeron, J. E. Keem and S. R. Ovshin-
sky, Solid State Commun. ^ 7 , 177 (1983).
|1. See Metastabie Materials Formation by Ion Implantation,
eds. S. T. Picraux and W. J. Choyke(EIsevier, Mew York,
1982).
32. B. M. Clemmens, Buli. Amer. Phys. Soc. 2_9, 506 (1984).
33. B. D. Cullity, Elements of X-ray Diffraction, 2 ed.,
(Addison-Wesley, Reading,MA, 1978)p. 102.
J4. G. S. Cargilt, M l , in Solid State Physics, Vol. 30.eds.
H. Ehrenreich, F. Seitz, and D. Turnbuil (Academic, New
York, 1975) p. 227.
35. G. Busch and H. Guntherodt, in Solid State Physics,Vol.29,
eds. H. Ehrenreich, F. Seitz, and D. Turnbu11(Academíc,
New York, 1974) p. 235-
36. H. J. Guntherodt and H.U.Kunzi, in Metal Ii c Glasses, ed.
J. J. GiIman and H. J. Leamy (Amer. Soc. Metals, Metals
Park, OH, 1978)p. 247.
37. H. S. Chen and D. Turnbull, J. Appl. Pbys.J_0, 284 (1967).
38. H. E. Kissinger, J. Res. Natl. Bur. Stan. $]_, 217 (1956).
39. P. C. Boswell, J. Therm. Anal. )£> 3 5 3 (1980).
kO. C. H. Bennett, D. E. Polk, and D. Turnbull, Acta metal 1.
143
19, 1295 (1971).
41. S. R. Nagel and J. Tauc, Phys. Rev. Lett.3^, 380 (1975).
J. Tauc and S. R. Nagel, Comments Sol. St. Phys. 2»^9(1976)
42. H. A. Davies, Phys. and Chem. Glasses JJ, 159 (1976).
43. D. R. Uhlmann, J. Non-Cryst. Solids 7, 337 (1972).
Work supported by FAPESP, CNPq and FINEP (Brazil).
144
STRUCTURAL RELAXATION: LOW TEMPERATURE PROPERTIES
Francisco de la Cruz
Centro Atômico Báriioche
8'+00 - S.C. de Bariioche (R.N.), Argentina.
ABSTRACT
We discuss the changes in transport and superconduct-
ing properties of amorphous Zr70Cu30, induced by thermal
relaxation.
The experimentai results are used to investigate the
relation between the microscopic parameters and the
observed physical properties. It is shown that the density
of electronic states determines the shift in T as well as
c
the variation of the electrical resistivity.
It is necessary to assume strong hybridization between
a and d bands to understand the eiectrodynamic response of
the superconductor.
145
INTRODUCTION
In this lecture I will refer to the research made In the
Low Temperature Lab in Bariioche, during the last four
years, concerning the normal and superconducting properties
of metallic amorphous systems. The title of the talk is
misleading. I am not an expert in relaxation and you will
see that the heat treatment is only used to induce changes
in the physical properties of our samples, in order to
study the behaviour of the microscopic parameters of these
metals.
The materials investigated are lOum thick ribbons of
Zr 7 0Cu 3 0 and La 7 0Cu 3 0 alloys, obtained by melt spinning.
In most of this talk we will refer to the results obtained
from the Zr70Cu30 system.
Before we start to show and discuss the experimental
results I will remark some properties chat are common to
all the transition metals amorphous alloys:
a) High electrical resistivity: p » 200uftcm.
b) Non-validity of Matthiessen's rule when applied to the
temperature dependence of the electrical resistivity.
This result is known as Mooij's criterium.
c) The thermodynamic and transport properties of the
amorphous materials at low temperatures are characteriz-
ed by the presence of the low energy excitations, TLS.
d) Since the electron mean free path l, is of the order of
interatomic distances the heat is mainiy carried by
phonon3. As a consequence, amorphous systems are ideal
146
materials to investigate the phonon thermal conduction
in metals.
We will now focus our attention in some questions
related to the properties we have indicated:
i) Is the high electrical resistivity of these amorphous
metals due to the d-eiectron contribution in
transition metals?
ii) Which is the origin of the negative temperature
coefficient of P(T)?
iil) Assuming that in this amorphous metals it is possible
to define a Fermi wave vector, it is found that
kp«i=«l. Is the BCS-Gorkov theory adequate to describe
superconductivity in this extreme dirty limit?
iv) Another question related to the previous one is: are
the Gorkov equations valid when the d and s electrons
contribute to the transport properties and super-
conductivity?
v) Is there any dependence between the superconducting
critical temperature, T , and the density of TLS,
n(0)?
vi) Is the Matthiessen's rule valid when applied to the
phonon thermal conduction in amorphous metals?
THEORETICAL AND EXPERIMENTAL BACKGROUND
Following the BCS scheme, superconductivity arises
from the competition between an attractive phonon-electron
147
*i n t e r a c t i o n , c h a r a c t e r i z e d by a parameter X end a
r e p u l s i v e Coulomb i n t e r a c t i o n » u . As a r e s u l t the
c r i t i c a l temperature should be a function of these two
parameters
Tc = f(X M ) (1)
Due to the lack of tunneling data in amorphous metals
the electron-phonon parameter can be approached by X «
N(0)I2/92, where the symbols are those typically used in
the literature. The parameter y » 0-1 is usually accepted
for transition metals.
A review of the critical temperature behaviour of
amorphous metals can be found in ref.2. There it is
indicated that relation (1), with X and u as described
previously, is enough to understand most of the
experimental data. Nevertheless, I believe there are some
questions that have no definite answer. One of the
problems is the possibility that the TLS contributes to
T . If this is the case, T should also be a function ofc c
n(0) and expression (1) shouid then be generalized. It is3
also important to remark that if the TLS can modify the
* '+effective x , recent calculations indicate that disorder
could Increase v . This is an important result since it
indicates that T , H . and p could be correlated through
the degree of electron localization and would indicate that
in the extreme dirty limit, the critical temperature should
also be a function of the electron mean free path. From
the experimental point of view there are no answers co
148
these questions.
Measurements in the Zr Cu, system indicate that the
increase in x induces a rise in T together with a decrease
in p . The behaviour of T has being explained taking into
account the measured behaviour of N(0). The decrease in
9 10p has also been related ' to the increase in the density
of states due to contribution of the d-Zr band. In those
experiments it is difficult to separate the contribution of
TLS (if any) and/or, of localization. There is not enough
systematic investigation of a possible direct correlation
between T and n(0). In this lecture we will discuss some
results related to this topic.
Other superconducting parameter related to the
electronic properties of the material is the upper critical
field, H 2* Within the Gorkov theory and for the dirty
limit
H c 2 - Vk f N(0)pf(T). (2)
To obtain (2) it has being used the Ginzburg-Landau
coherence length in the dirty limit, given by Ç2(0,0 »ÇQ4,
with ç0 - 0.18 hvF/kTc , vF » k2S/6hy, p-1-(2/3)e2vFN(0)£,
and Y-(2/3)(wk)2N(0). Here 40 is the BCS coherence
lengths, S is the area of the Fermi sphere, I the electron
mean free path and y the coefficient of the electron heat
capacity. Ail these expressions have been obtained
assuming that kp* ~» 1. As was mentioned before this limit
is not adequate for the amorphous samples used in our
experiments.
149
Within the same approximation the superconducting
response to the presence of a low magnetic field Is
determined by the superconducting penetration depth
X(l,t) « ^(OMCo/Oi'Zfit). (3)
For T near Tc
where the London penetration depth x1(0) « 3h*l/2Yl/2/ekS.
It Is Interesting to recall that A. (0) Is only related to
the normal properties of the material and that expressions
(3) and (31) indicate a correction to *T(0) through the
square root of the ratio of two distances. Superconducti-
vity only appears through the definition of ÇQ.
Expressions (2) and (3') can be verified since ail physical
quantities that appear in them are experimentally
accesible. Although the verification of expressions (2)
and (31) is interesting from the point of view of the
effects induced by an extreme short i, we believe that
there is other related point that has to be considered when
studying transition metals. It was realized by Bergmann
that expression (2) should not be valid when applied to
metals that can be characterized by the presence of two
bands (d and s, in our case). Following the same arguments
we will see that it is difficult to Justify the validity of
expression (3'). The coherence length ç0 is strongly
associated to the interaction energy necessary to form the
Cooper pairs (kTc). The critical temperature in a d and s
150
band superconductor, is believed to be determined by the d
band density of states, N,. On the other hand in an
independent two band model the s electrons contribution
7 8should at least be competitive with the d electrons ' .
9 10Recent work ' gives the experimental results of the
Theresistivity of Zr Cu._ as a function of xo
concentration dependence has being explained , on the basis
of a two band model, where the contribution to p from d and
s electrons are found to be comparable. We believe that if
this is the correct explanation expression (31) should not
be applicable.
Until here we have referred to changes in the physical
properties of the amorphous material induced by changing
concentration. We have other available experimental
technique to change the behaviour of the material at
constant concentration. It has being shown in the last
years that thermal heat treatment modifies the normal
and superconducting properties of these materials. In the
case of Zr Cu, it has being suggested chat the supercon-
ducting critical temperature is determined by the
electronic density of states, in agreement with Varraa and
Dynes. This result has being obtained from the analysis
of the variation of Tc and N(0) with concentration . If
the analysis is correct the change in T Induced by thermal
relaxation should also be determined by a corresponding
change in N(0).
151
EXPERIMENTAL RESULTS AND DISCUSSION
We have measured the thermal conductivity of Zr70Cu30
amorphous ribbons, in the range of temperature between
O.i°K and 7°K. The results are shown in Fig.l.
Experimental details can be found in ref. 12. It is
clearly seen that the thermal conductivity below the
critical temperature of the alloy is monotonically
increased with annealing. Further annealing is not
possible because the sample starts to crystallize as
indicated by X-ray diffraction analysis and electrical
resistivity measurement . The critical temperature is
11 'decreased when annealing, as it is also indicated by the
12structure of the thermal conductivity plot in Fig.l.
The T2 dependence of the thermal conductivity at low
temperature is characteristic of phonon-TLS resonant
scattering. Since annealing does not change the
temperature dependence but increases the thermal
conductivity we conclude that annealing increases the
coefficient of the T2 dependence. That is to say, it
decreases the product of the number of scattering cento's,
n(0), times the square of the coupling matrix between the
phonon and TLS. Considering only thermal conductivity
measurements that is all we can say. In any event, these
measurements indicate that these mild heat treatments can
modify considerably the TLS behaviour. In Fig.2 we plot
Che thermal conductivity of the amorphous sample taken at
T-0.5K, normalized by the value of the as quenched one, as
152
a function of the critical temperature, also normalized by
the critical temperature of the as quenched sample. From
these results it is tempting to say that there is a
correlation between the TLS behaviour and the critical
temperature. We will see later that this is not necessari-
ly true and that the change in T can be explained without
involving the assistance of the TLS.
In Fig.3 we show the effect of annealing in the
critical temperature and electrical resistivity. In the
plot of T vs pj normalized by the respective values of the
npn annealing sample, we can clearly distinguish two
regions. First, the critical temperature decreases at
almost constant p, later there is a rapid decrease in p
without major changes in T . This indicates two thermally
Induced processes. To detect structure changes during
annealing we have investigated the X-ray diffraction
pattern. In the first region, we were not able to
distinguish any change within our experimental error, in
the second when p('4k)/ p^C+k)- 0.8 it was detected a weak
structure typical oi crystallization. A n the results we
discuss here, including the thermal conductivity
measurements, correspond to thermal heat treatment in the
first region.
As was mentioned in the Introduction, if there Is only
one microscopic parameter that determines T , the change In
the parameter that correaponoa to a given AT should be
independent of the method used to vary T .
153
From specific heat and H » measurements the change in
density of states as a function of concentration is known.
From these data we obtain that the change we should expect
from the AT induced by annealing is only a few percent,c
Since it ia very' difficult to achieve the necessary
precision by measuring specific heats we decided to use
the H 2 And p measurements, together with expression (2),
to determine the relation between N(0) and T when
5 18annealing. There is experimental evidence ' indicating
that this expression is valid when applied to splat cooled
samples. In this work we assume the validity of expression
(2) and we will discuss later some related experimental
results, obtained in our laboratory.
Figure % shows the results of N(0) obtained from H j
as a function of T c. The dotts correspond to the
variation of N(0) with T , induced by annealing and the
full line is an interpolation from the data obtained by
changing concentration. We see that the data obtained by
thermal heat treatment are quite similar to that obtained
from the change in concentration. As a consequence, the
correlation suggested by Fig. 2 is not more than spurious,
indicating that there is no direct relation between T and
n(0). It would be intereting to understand why the thermal
relaxation changes the electronic density of states as well
as that of the TLS.
154
Let us focus our attention on the behaviour of the
electrical resistance. We have found that the resistivity
increases when the sample is annealed. Since we will not
discuss the kinetics of the relaxation process and it is
found that T is strongly correlated with the hehaviour of
the resistivity, we plot the resistivity change as a
function of the variation of T , see Fig. 5. The increase
in resistivity found for these alloys seems co be
characteristic of amorphous transition rcetais and, to my
knowledge, there is no explanation for such behaviour.
It is interesting to remark that in the range of
9 10concentration ' we investigate T decreases with N(0) and
p increases when N(0) is diminished. Since we know the
experimentai relation between concentration and N(0) we can
determine & P / A K ( 0 ) in the range of concentration of
19interest. It is found that the Ap/iN(0) obtained from
Figs. '* and 5 is smaller by a factor between 1.3 and 2.2
when compared with that obtained from the change ino jo
concentration'' . The range in the slope values is due to
the difference between the experimentai values of refs. 9
and 10. Considering the difficulties in determining the
geometrical factor of amorphous ribbons, we think chat çhe
19similarity found between the Ap/AN(0) obtained from the
change in concentration and annealing experiments, is
strongly indicating that N(0) is also the fundamental
parameter determining the behaviour of p.
Let us now discuss the results obtained from the
155
penetration depth measurements. Details on the
experimentai technique used to measure x(t) can be found in
reference 20. We wiii not discuss here the temperature
dependence of A(t). We will refer only to the relations
between X(0), p and T , as given by expression (3). We
19 20have measured ' Ã ( 0 ) , T and p for different amorphous
alloys, the results are shown in table I. We see that
expression (3) is verified within a 10% error. Since the
error in the geocetricai factor is not less than 10X we
find che agreement surprising and good. These results are
important since until now we hav~ indicated that N(0) is
the main microscopic parameter determining the behaviour of
several properties of the Z r ^ C u ^ systems-
In a two band model the density of electronic states
should be maimy related to the d contribution. As we said
in the introduction, expression (3) seems to be
incompatible with a two band model since the correction due
to a finite mean free path is given by a ratio of two
lengths, one characteristic of the superconducting state,
ÇQ, che other, l, related to the transport properties in
the normal state. In a two band jnodei T is determined by
c J
the d density of states but che t that appears in (3)
should not be the one that determines the measured
electrical conductivity.
The experimental verification of expression (3) is
strongly suggesting that in thest; transition amorphous
metal there is a single type of carriers contributing to
156
Che chernodynamic and transport properties. These results
5 18are in agreement with those ' supporting the verification
of expression (2). We believe that the suggestion made byo
tenBosch and Bennemann concerning to hybridization of d
and 8 bands is of fundamental importance for a correct
understanding of transport properties in amorphous
transition metals.
We have not been able to complete the discussion
proposed at the introduction but I hope that future work
will serve to verify the ideas exposed previously and will
clarify the rest of the remarks made at the beginning of
this lecture.
ACKNOWLEDGEMENTS
I want to thank M.E. de la Cruz for her help during
the preparation of the manuscript. The results that have
been discussed here were obtained at the Low Temperature
Group in Bariloche. I gracefully acknowledge many
suggestions and discussions with the staff of che group.
Many points discussed in che lecture become clearer after
useful conversations with 0. Balseiro.
157
REFERENCES
1 H. Tuczauer, P. Esquinazi, M.E. de ia Cruz and F. de
ia Cruz, Rev. Sci. Instrum. 51, 5'+6 (1980).
2 W.L. Johnson, in Glassy Metais I, Vol. '+6 of Topics in
Appiied Physics, edited by H.J. Güntherodt and H. Beck
(Springer, N.Y. 1981).
3 R. Harris, L.J. Lewis and M.J. Zuckermann, J. Phys. F
13, 2323 (1983), S.V. Maieev, Sov.Phys. JETP, 57, l'+9
(1983).
% P.W. Anderson, K.A. Muttalib, and T.V. Raroakrishnan,
Phys. Rev. B 28, 117 (1983); Liam Coffey, K.A.
Muttaiib and K. Levin, Phys. Rev. Lett. 51, 783
5 F.P. Misseii, S. Frota-Pessoa, J. Wood, J. Tyier and
J.E. Keens, Phys. Rev. B 27, 1596 (1983); K. Samwer
and H.v. Lohneysen, Phys. Rev. B 26, 107 (1982).
6 G. Bergmann, Phys. Rev. B 7, '+850 (1973).
7 G.F. Weir and G.J. Morgan, J. Phys. F. U, 1833
(1981).
8 A. cen Bosch and K.H. Bennemann, J. Phys. F. 5, 1333
(1975).
9 D. Pavuna, J. of Non Cryst. Sot. - in print.
10 M.N. Baibich, W.B. Muir, Z. Aitounian and Tu Guo-Hua,
Phya. Rev. B 2_7, 619 (1983).
11 J. Guimpei and F. de ia Cruz, Solid State Commun. '*\,
1045 (1982).
158
12 P. Esquinazi, M.E. de ia Cruz, A. Ridner and F. de la
Cruz, Solid Stace Commun. j-^, 9'4l (1982).
13 H.J. Schink, S. Grondey and H.v. Lohneysen, in
Phonon Scattering in Condensed Matter, Ed. W.
Eisenmenger and S. Dõttinger (Springer Series in
Solid State Sciences, Vol. 51, 1984).
1'* J.C. Lasjaunias, A. Ravex, and 0. Bêthoux, in Phonon
Scattering in Condensed Matter, Ed. V. Eisenmenger and
S. D'dttlnger (Springer Series in Solid State
Sciences, Vol. 51, 198'4).
15 L. Civale, F. de ia Cruz and J. Luzuriaga, Solid State
Cornmun, Vb, 389 (1983).
16 P.H. Kes and C.C. Tsuei, Phya. Rev. B J29, 5126 (1983).
17 Ç.M. Varma and Dynes, in Superconductivity in d and f-
band Metals, edited by D.H. Douglass (Plenum, N.Y.,
1976).
18 M.G. Karkut and R.R. Hake, Phys. Rev. B 28, 1396
(1983).
19 F. de ia Cruz, M.E. de la Cruz, L. Civale and R. Arce,
to be published.
20 R. Arce, F. de la Cruz and J. Gulmpel, Solid State
Cotnmun. '+7, 885 (1983).
O
w.
159
' "2T ! K )
Fig.l - Thermal conductivity, k, as a function of tenr-erature,T, for amorphous Zr7,.Cu_c, for different heat treat-ments. See ref.12.
1 - a A
Õ.9 Tc/Tc,
Fig.2 - Thermal conductivitv nt 0.5k as a function of the
change in T_, induced by annealing..
160
OHi
oi
Fig.3 - Variation of the critical temperature as a function
of the chanpe in resistivity, induced by annealing.
L...27 / "
Fig.4 - Density of states, N(0), as a function of T^. Open
circles correspond to the value obtained by annea-
ling, the full curve is an interpolation from ref.5.
F \
us!
Fip.5 - Variation of the electrical resistance
as a function of the critical temera-
turc.
161
A M O R P H O U S S U P E R C O N D U C T O R S
Frank P. M i s s e l l
I n s t i t u t o de F f s i c a , U n i v e r s i d a d e de S ã o P a u l o ,
C P . 2 0 5 1 6 , S ã o P a u l o , S . P .
W e d e s c r i b e b r i e f l y the s t r o n g c o u p l i n g s u p e r c o n d u c t i -
v i t y o b s e r v e d in a m o r p h o u s a l l o y s b a s e d u p o n s i m p l e m e t a l s .
For t r a n s i t i o n m e t a l a l l o y s w e d i s c u s s the b e h a v i o r of the
s u p e r c o n d u c t i n g t r a n s i t i o n t e m p e r a t u r e T , the u p p e r c r i t i -
cal f i e l d H , and the c r i t i c a l c u r r e n t J . A s u r v e y of c u r -
rent p r o b l e m s is p r e s e n t e d .
T h e f i r s t a m o r p h o u s s u p e r c o n d u c t o r w a s p r e p a r e d ín 1951»
by B u c k e l and H M s c h . T h e y e v a p o r a t e d Bi in h i g h v a c u u m
and d e p o s i t e d the v a p o r o n t o a s u b s t r a t e at l i q u i d H e t e m p -
e r a t u r e . T h e e f f e c t o f p r o d u c i n g B; in this m a n n e r is d r a -
m a t i c : in t h e n o r m a l c r y s t a l l i n e p h a s e Bi is not a s u p e r -
c o n d u c t o r , w h i l e a m o r p h o u s Bi has a s u p e r c o n d u c t i n g t r a n s i -
tion t e m p e r a t u r e T = 6 . 1 K. P u r e Ga can a l s o be p r e p a r e d in
the a m o r p h o u s p h a s e , as can m e t a l s s u c h as P b , S n , T l , In,
e t c . by the a d d i t i o n of 1 0 - 2 0 % of a n o t h e r c o m p o n e n t . O n e
g e n e r a l l y f i n d s that T i n c r e a s e s a b o v e the c r y s t a l l i n ec 2v a l u e for t h e s e a m o r p h o u s a l l o y s . In a d d i t i o n . from t u n n e l -
ing e x p e r i m e n t s o n e f i n d s that the v a l u e of 2 A / k T is of the
o r d e r of k.5 for t h e s e m a t e r i a l s ( w h e r e 2A is the z e r o
t e m p e r a t u r e e n e r g y g a p ) . T h i s v a l u e c o n t r a s t s w i t h the v a l u e
of 3 * 5 2 , a p p r o p r i a t e for a w e a k - c o u p l i n g 8 C S s u p e r c o n d u c t o r ,
and s u g g e s t s that t h e s e s u p e r c o n d u c t o r s a r e s t r o n g - c o u p l i n g ,
that i s , t h e y h a v e a s t r o n g e 1 e c t r o n - p h o n o n c o u p l i n g . In
T a b l e I, w e p r e s e n t v a l u e s of T and 2 A / k T for s e v e r a l
a m o r p h o u s a l l o y s . A l s o s h o w n in t h i s t a b l e a r e v a l u e s of >.,
the e l e c t r o n - p h o n o n c o u p l i n g c o n s t a n t , as d e t e r m i n e d f r o m
t u n n e l i n g e x p e r i m e n t s . A s t r o n g e l e c t r o n - p h o n o n c o u p l i n g
l e a d s to an e n h a n c e m e n t of the e l e c t r o n i c d e n s i t y of s t a t e s
by a f a c t o r 1+X, w h e r e >.«2/oi2F(w)d(i}/u). The E l i a s h b e r g f u n c -
162
T a b l e I. P r o p e r t i e s o f a m o r p h o u s s u p e r c o n d u c t o r s b a s e d o n2
s i m p l e tneta 1 s .Al l o y
Bi
Ga
S n 9 0 C u 1 0P b 9 0 C u 1 0P b75 B 525l n 8 0 S b 2 0T 1 9 0 T e 1 0
T c(K)
6.1
8.4
6.76
6.5
6.9
5-6
4.2
2A/kTc
4.60
4.60
k.k6
4.754.98
U.XO
i. .6
2
1
X.2-2.
.94-2
1.84
2.0
2.76
1.69
1 .70
46
• 25
t i o n a 2 F ( c o ) i s g i v e n in t e r m s o f a t h e a v e r a g e e l e c t r o n -
p h o n o n m a t r i x e l e m e n t a n d F ( u > ) , t h e d e n s i t y o f p h o n o n s t a t e s .
In t h e n o r m a l s t a t e C X 2 F ( O J ) is p r o p o r t i o n a l t o t h e p r o b a b i -
l i t y t h a t a n e x c i t e d e l e c t r o n c a n e m i t a p h o r t o n w i t h f r e q u e n -
cy to. T h e f u n c t i o n O I 2 F ( Í J Ü ) c a n b e d e t e r m i n e d e x p e r i m e n t a l l y
b y i n v e r t i n g d a t a f r o m t u n n e l i n g e x p e r i m e n t s .
F i g . 1 T h e E ' l i a s h b e r g f u n c t i o n a z F i u ) f o r c r y s t a l l i n e a n d
d i s o r d e r e d P b a n d a m o r p h o u s P b n C u j ( , .
In F i g . 1 w e s h o w t h e E l i a s h b e r g f u n c t i o n f o r c r y s t a l -
l i n e P b , f i n e - g r a i n e d c r y s t a l l i n e P b , a n d a m o r p h o u s P b ^ / ^ C u ^
d e t e r m i n e d f r o m t u n n e l i n g e x p e r i m e n t s b y K n o r r a n d B a r t h .
F o r c r y s t a l l i n e P b , v.he p e a k s c o r r e s p o n d i n g t o t h e t r a n s -
v e r s e a n d l o n g i t u d i n a l p h o n o n s a r e c l e a r l y v i s i b l e , w h i l e
t h e c u r v e f o r a m o r p h o u s P b C u . s h o w s a "large c o n t r i b u t i o n
a t l o w f r e q u e n c i e s . E v i d e n t l y t h i s l a r g e i n c r e a s e at l o w
163
frequencies is responsible for the increased values of X in
the amorphous alloys. It i? possible to show experimentally
from Mossbauer measurements of the Debye-Waller factor that
this increase is due primarily to an enhanced electron-
phonon interaction. The Mossbauer effect furnishes
/F(w)du/w at zero temperature, a quantity which weights the
phonon density of states in the same manner as the integral
defining A. Bolz and P o b e M measured this quantity for both
crystalline and amorphous Sn, obtaining the same result to
within 10%. Thgs the large differences observed in X between
then crystalline and amorphous states must be related to an
increased electron-phonon interaction a 2.
To calculate the electron-phonon coupling constant, it
is necessary to have some idea of the electronic structure
of these amorphous metals. There is experimental evidence
thst the amorphous superconducting alloys may be treated as
free electron metals. Measurements of the Hall effect for
n ç w t r a n s i t i on metals have yielded, with few exceptions,
the free electron value of the Hall constant . Furthermore,
the measured values of the optical surface resistance for
amorphous Ga agree with the predictions of the free elec-gtron model to '-;ithin e x p e r i m e n t a l u n c e r t a i n t y . Similar
9
agreement is o b t a i n e d for BÍ . A calculation of the electron-
phonon interaction, based on the free e l e c t r o n m o d e ! and
neg l e c t i n g c o n s e r v a t i o n of m o m e n t u m , has been carried out
by Bergmann , He showed that the increased phase space for
electron and phonon interactions could e x p l a i n the increased
values of a 2 in the amorphous s t a t e . Thus, it appears that
the increased values of X and 2A/kT can be explained in
terms of the changes in a 2F(u>), w h i c h , in turn, can be
related to d i s o r d e r . Other c o n s e q u e n c e s of strong coupling,
such as a modified temperature d e p e n d e n c e of the energy gap
and critical field as compared with the BC3 theory, have2
been observed and are discussed by Bergmann ,The first s y s t e m a t i c study of amorphous transition
meta,l alloys was carried out by Collver and Hammond w h o
employed e l e c t r o n - b e a m e v a p o r a t i o n onto cryogenic s u b s t r a t e s
Io4
to o b t a i n a large n u m b e r of m e t a s t a b l e m a t e r i a l s b a s e d on
the ki and 5d transition m e t a l s . T h e s e data are n o t e w o r t h y
?n that the v a r i a t i o n of T w i t h e l e c t r o n / a t o m (e/a) ratio
is m a r k e d l y different than in c r y s t a l l i n e m a t e r i a l s . For
the a m o r p h o u s t r a n s i t i o n metal a l l o y s , T varies slowly w i t h
e/a and e x h i b i t s a broad m a x i m u m a r o u n d e / a ^ 6 . 5 . In c o n t r a s t ,
T in c r y s t a l l i n e a l l o y s s h o w s two sharp peaks near e/a v a l u e s
of k.S and 6-5 ( M a t t h i a s ' r u l e ) . T h e s e peaks have been
a t t r i b u t e d to structure in the d-band density of s t a t e s for
c r y s t a l l i n e m a t e r i a l s . T h i s c o n t r a s t is shown in Fig. 2 .
P r e v i o u s l y , Crow et a_l . had m a d e similar o b s e r v a t i o n s for
c e r t a i n t r a n s i t i o n metal a l l o y s and a t t r i b u t e d the i n c r e a s e d
T^ values in the a m o r p h o u s s t a t e to m o d i f i c a t i o n s in the
e l e c t r o n i c density of s t a t e s due to the p r e s e n c e of d i s o r -
d e r . If T is d e t e r m i n e d m a i n l y by the e l e c t r o n i c d e n s i t y
of states N ( E p ) aná if N ( E p ) p o s s e s s e s a single b r o a d m a x -
imum for a m o r p h o u s alloys b a s e d on the Ad t r a n s i t i o n m e t a l s ,
then the Collver and H a m m o n d result could be a r e f l e c t i o n
of a s t r u c t u r e l e s s d band for these a m o r p h o u s t r a n s i t i o n
m e t a ! a 1 loys.
20
16 -
12 -
2'I
1 I
-
-
i f •
A>
§•
h\4 -
3 4 S 6 7 8 9(V) (Zf) INb) (Mo) (Tc) (Rui (Rh)
i/m
F i g . 2 T h e v a r i a t i o n o f T f o r kà t r a n s i t i o n m e t a l s in t h e
c r y s t a l l i n e ( d a s h e d c u r v e ) and a m o r p h o u s ( s o l i d c u r v e )
s t a t e s , f r o m C o l l v e r a n d H a m m o n d .
165
Th e a m o r p h o u s n a t u r e of the films o f C o l l v e r and Ham ~
m o n d w a s d e d u c e d i n d i r e c t l y from the sharp drop in the e l e c -
trical r e s i s t a n c e w h e n the fi l m s w e r e h e a t e d to room t e m p e r -
a t u r e . In a few c a s e s , such as for al l o y s of M o - R u o r N b - Z r ,
the a m o r p h o u s p h a s e wa.s s t a b l e at room t e m p e r a t u r e and it
wa s p o s s i b l e to v e r i f y the a m o r p h o u s n a t u r e by m e a n s of
t r a n s m i s s i o n e l e c t r o n m i c r o s c o p y . The first m e t a l l i c g l a s s e s
e x h i b i t i n g s u p e r c o n d u c t i v i t y w e r e La-Au a l l o y s , P r e p a r e d by
splat q u e n c h i n g . Many o t h e r s u p e r c o n d u c t i n g t r a n s i t i o n
m e t a l a l l o y s h a v e s i n c e been p r e p a r e d w h i c h a re s t a b l e at
room t e m p e r a t u r e . The T v a l u e s of some of the s e a l l o y s are
given in Ta b l e II. We note that the h i g h e s t t r a n s i t i o n temp-
e r a t u r e is T = 9 . 0 0 K for M o 8 o P i O B 1 0 ' T h e s e a 1 1 o v s f a " i n t o
two b r o a d c a t e g o r i e s : e a r l y t r a n s i t i o n m e t a l s ( Z r , Ti , N b )
wit h late t r a n s i t i o n m e t a l s (Rh, Cu, Pd) or t r a n s i t i o n m e t a l s1 k 15wi t h a n o n - t r a n s í t i o n metal or a m e t a l l o i d . J o h n s o n '
has p r e s e n t e d reviews of the p r o p e r t i e s of these m a t e r i a l s .
T a b l e II, T v a l u e s for some a m o r p h o u s s u p e r c o n d u c t o r s
Al loy T XiSi
( M o 0 . 8 R u 0 . 2 ) 8 0 P 2 C 7 .31
*VioBio 9 - 0 0
( M o 0 . 8 R e 0 . 2 ) 8 0 P 1 0 B 1 0 8 ' 7 1
L a 8 0 A u 2 0 3 ' 5
L a 8 0 G a 2 0 3 - 8 4
7 5 5
Z r70P d30 2 .807 3
Nb,Ge 3-6
NbjSi 3-9
Mo 6 8 Si 3 2 7.4«°80N20 8 ' 3
T h e s u p e r c o n d u c t i v i t y of a m o r p h o u s t r a n s i t i o n metal
a l l o y s p r e s e n t s d i f f e r e n t f e a t u r e s than that of a l l o y s based
1Ó6
upon simple m e t a l s . For exa m p l e , results of tunneling exper-
iments furnish values of 2A/kT which are in good agree-
ment with the value obtained from BCS theory. Thus these
alloys are weak coupling superconductors with X < 1 , and we
expect that many of their superconducting properties will
be in agreement with the BCS or Ginzb u r g - L a n d a u - Abrikosov-
Gor'kov (GLAG) theory. The situation is more complex,
however.
Early m e a s u r e m e n t s of the upper critical field H (T)
for amorphous transition m e t a l s showed rather large values
for this quantity, consistent with a small coherence
length £ and large penetration depth \., characteristic of
materials possessing a short mean free p a t h . Furthermore,
the field slope at T , (dH _ / d T ) _ , is also very largec ci i
(20-l»0 kG/K) , cons i s t en t w i t h the r e s u l t of the extended
GLAG t h e o r y 1 7 ' 1 8 :
T = " 6 J & £ £ pN { E F ) ( 1 >c TT
where p »s the normal resistivity, N (E_) is the electron-
phonon dressed density of states at the Fermi level for one
spin direction, and 8 is an enhancement factor (of order 1)
for strong coupled superconductors. Since the amorphous
transition metal alloys have large normal resistivities and
large densities of states, it is reasonable that the field
slope is also large.
The temperature dependence of H „ was observed to be
linear over a substantial portion of the temperature range
below T '"15> 9 |.ater, Tenhover et al . made a carefulc -
comparison of H ?(T) to the GLAG theory for the amorphous
alloys M o 3 0 R e 7 ^ ( H o ^ R u ^ ^ B , v and (Mo0 s*u Q J ^? n,
produced by the hammer-and-anvi1 method. These authors found
the experimental values of H « to be larger than the pre-
dictions of the GLAG theory for low values of the reduced21temperature t»T/T . Carter et al. showed how these enhan-
c «—«——
ced values of H might arise from a fine-scale inhomogen-
efty In the sample.
167
They suggested that this inhomogeneity might result from the
fact that the quench rate in the experiments of Tenhover
et al . (dT/dtM 0 K/sec.)might not have been sufficient to
produce a thoroughly homogeneous sample. As evidence, they
noted that their sputtered films of and M oç 2G el»8
were in complete agreement with the GLAG theory. Recent22
measurements on melt-spun and sputtered samples of Zr.^Rh.
show good agreement of H ,(T) with the GLAG theory Jn both
cases. Thus it appears that the linear behavior observed
for H ~(T) down to low reduced temperatures, may not be
general behavior for all amorphous superconductors, but may
be restricted to certain sample compositions or to certain
sample preparation techniques. Further experiments are need"
ed in this area.
0
Fig. 3 The upper critical field H . vs temperature T for
sputtered ( Z r7^ R h24^ an(* "^it'sP" 0 (Zr _Rh )material,
Solid lines are predictions of the GLAG theory in the
absence of paramagnetic limiting. Data can be brought
into agreement with theory by considering spin-orbit
scattering and paramagnetic limiting.
The homogeneous nature of amorphous superconductors
becomes quite evident when we consider the critical current
density J The J values encountered are very low in thecas-quenched material. In Fig.
6 R u0 i»^80B10^'i0
we compare J for two samplescby the hammer-and-anv i I
16S
IO4
^. IO1
10
I ' I • I • I
"A.-nofphoo«"
T-I.6K
K
«T-4.2K
.• I • I M , 1O 1 2 3 4 5
10' _
!0* -
101 -
10»
r
-
T M
UI
;
ran9e totJei0)
, L ...
i | r f i i . j
Airwphoui withcryitallinc inclusions _
_ . T-2.57 K -
x :T - 4 . I B K :
•
! i I • • • ' •0 1 2 3 * 5
ll.T
F i g . í» C r i t i c a i c u r r e n t d e n s i t y v s . m a g n e t i c f i e l d f o r t w o
s a m p l e s o f ( M o 0 ^ Ü Q } 8 ( JS i , Q 3 1 fl ,
t e c h n i q u e . In o n e c a s e , the s a m p l e h a d b e e n p a r t i a l l y r e -
c r y s t a l H z e d by m e a n s o f a n n e a l i n g a n d the r e s u l t a n t c r y s -
t a l l i t e s s e r v e as f l u x - p i n n i n o s i t e s . A s a r e s u l t , J h a s3 c
i n c r e a s e d by a f a c t o r of 10 at 20 k G . A p r o m i s i n g m a t e r i a l
in t e r m s o f i is t h e a l l o y ( Z r Q 7 H f Q . ) , n V , . . In the a m o r -
p h o u s p h a s e , t h i s m a t e r i a l e x h i b i t s g r e a t d u c t i l i t y b u t Is
not s u p e r c o n d u c t i n g . A f t e r an a p p r o p r i a t e h e a t t r e a t m e n t ,
the b r i t t l e C - 1 5 p h a s e f o r m s w i t h a J in e x c e s s o f lOOkA/cir/
at 1 5 0 k G . P o s s i b l e a p p l i c a t i o n s o f a m o r p h o u s m a t e r i a l s h a v ebeen d i s c u s s e d e l s e w h e r e25
As in the case of H -,(í), the T behavior* of amorphousc2 c
transition metal alloys sti1 1 presents some d i f f i c u l t i e s .
The suggestion by Crow et a 1., that disorder-induced changes
ín the density of electronic states N ( E _ ; may be responsible
for the differences in T behavior between •rystalline and
amorphous stater., has been discussed extent vely by
Johnson ' in terms of the McMillan equal on . The Mc-
Millan equation gives T_ in terms of <w z> the mean square
phonon frequency, u , t^e Coulomb pseudopotentfa 1, and >,
the e1ectron-phonon coupling constant a s :
169
r i.oi(i+\)
L" I^+TnõTr2xTTc = TTfT exp
McMillan showed that X,as defined previously, can be rewrit-
ten as A»N(E F)<I2>/M<a) 2>, where N ( E F ) = N
+ ( E F ) / O + X ) is the
bare density of electron states at the Fermi level, <\2> is
an average squared electronic matrix element, with the
average taken over all electron-phonon scattering processes,
and M is the Ionic mass.
Now in order to apply this equation to the problem of
amorphous transition-metal superconductors, Johnson employed
a tight-binding analysis of the electron-phonon coupling27constant X due to Varma and Dynes . These authors used a
simple model of the d band density of states, in which s
electrons and s-d hybridization were neglected. For the case
in which there is one orbital at the Fermi surface, Varma
and Dynes argued that the constant X is given by X=N(E )W(1+5),
where W is the width of the d band and S is the overlap in-
tegral of atomic d orbitais on neighboring sites. The factor
(ITS) takes into account the nonorthogonal ity of atomic d
functions on different sites. The mi nus(p1 us) sign is appro-
priate for E_ in the lower(upper) half of the d band des-
cribed by N ( £ _ ) . To the extent that the width V is a constant
for a given transition metal series, then X^N(E_) within
the upper and lower (antibonding and bonding) halves of the
d band. Thus, N(E_) is the principal parameter governing
superconductivity in these materials.
In order to compare the Varma-Dynes predictions with
experiment, Johnson obtained N(E_) from the low temperature
specific heat coefficient y and obtained X by inverting the
McMillan equation. The ratio X/N(E_) was compared for several
alloys and was found to be constant, as one might expect
from the Varma-Dynes calculation . In an attempt to study
the Varma-Dynes mode) for a greater number of alloys, Flasck
et a l / c plotted T c vs. N+(F.p) for (Mo, - x K % ) 8 o
p2 0 .
("V^oVi-xV ("VeKV^i^V and (Hoi-yVi-*sV
where T is a transition metal. N (E p) was determined from
170
Eq.(i), using measured values of p and dH _/dT. Although
Eq.(i) was derived for free electrons, a number of experi-28 +ments show that N (E_), derived from this equation and Y»
agree to within 5-10?. Thus, if N+(E ) is the most important
parameter in determining T , a graph of T vs. N (E_)
should have a positive slope. In fact, a correlation was
observed between T and N (E_), but this correlation depends
upon the metalloid involved in the alloy. For the alloys
involving Si, the slope of the curve T vs. H (E_) turns out
to be negative, i.e., larger values of N (Ep) result in
smaller T . Since the electronic structure of these alloys
is not well understood, this result does not necessarily
represent a contradiction of the Varma-Dynes model. It is
possible, for example, that hybridization effects might
invalidate some of the assumptions of the model.
In an attempt to study the Varma-Dynes model for a si-
tuation where its assumptions should be valid, alloys with a29simple electronic structure were considered . For example,
Zr-Cu alloys should satisfy the assumption of one orbital
at the Fermi level, since calculations performed using the
; I29
recursion method'' have shown that the Cu 3d states lie about
3-5 eV below E _ . Zr-rich alloys of Zr-fl.h also seem 7 to have
only Zr states near E_. In these two cases, the curve
T v s . N (E ) has a positive slope and thus larger values of
M (E-) result in larger values of T . Thus it appears that
the Varma-Dynes model might have a certain limited validity,
in cases where its assumptions are clearly satisfied. This
does not mean to say that all amorphous alloys based on
early/late transition metals satisfy the assumptions of the
model. Recent measurements of N (E_) in Zr-Pd and (slightly
contaminated)Zr-Ni alloys indicate that, in these m a t e r i a l s ,
N (E-) is not the dominant factor in determining T .
The question then arises as to whether the Collver-
Hammond curve does indeed result simply from a single broad
maximum in a featureless d band. It is known, for example,
that calculations of the d band for amorphous Co, simulated
by a dense random packing of hard spheres relaxed through a
171
lennard'Jones potential, show two broad peaks, one for the32
bonding and the other for the antibonding states . Thus a
simple rigid-band approach to this double-peaked band would
not explain the Col 1ver-Mammond data, in order to a n s w e r
this question, calculations are under way, using the recur-
sion methpu, of N(E_) for amorphous Mo.. Ru alloys . Since
e/a"6 for pure Mo and e/a^tJ for pure Ru, these c a l c u l a t i o n s
will furnish H(E-) «round the peak of the Col 1ver-Hammand
curve and allow a test of the arguments which have so often
been put forth to explain these d a t a 1 1 ' 1 2 ' ' **.
Recent studies of the density of states N (E p) In amor-
phous alloys have raised another problem related to the
experimental determination of this quantity. For e x a m p l e ,
Altounian and Strom-Olsen- have recently determined N (E_)
for melt-spun Zr-Ni alloys using E q . ( 1 ) . Their results are
in reasonable agreement with those of Kroeger et a l . who
studied melt-spun Zr-Ni alloys and determined N (E-) from
specific heat m e a s u r e m e n t s , On the other hand, the N (E_)
values obtained for sputtered Zr-Ni alloys are in agreement
with results from melt-spun material for Zr-rich c o m p o s i t i o n s ,
but there is some disagreement as the Zr fraction Is redu-
c«d . This disagreement may .esult from the different
quench rates obtained In m<?lt spinning and s p u t t e r i n g . T h u s ,
the melt-spun m a t e r i a l , produced at a lower cooling rate
( d T / d t M O K / s e c , ) , may show a slight chemical short range
ordering, which would tend to decrease N(E_) in these materials
and which would not be present in the sputtered m a t e r i a l s1 kbecause of the higher quench rate ( d T / d f U O K / s e c ) . Indeed
Krocger et a 1 .have seen evidence of chemical ordering in
their Zr-Ni #)loys. W h e t h c or not this is the e x p l a n a t i o n
for these discrepancies is difficult to decide since there
has been very little in the way of structural studies on
these materials. It would be Interesting to repeat these
measurements on melt-spun and sputtered samples w h i c h have
been thoroughly characterised by x-ray and neutron s c a t t e r -
ing experiments.
Another possible e x p l a n a t i o n for the different v a l u e *
172
of N + ( E p ) observed in these experiments lies in the use of
two different methods (specific heat and Eq. (?)) for deter-
mining N + ( E p ) . Recently, Ravex et al. determined N + ( E _ )
for sputtered Zr-Cu alloys using both specific heat measure"
msnts and Çq, (1) . The value obtained from Eq.(l) was ^ 0 %
smaller than that obtained directly from the low temperature
specific heat- This is the first and only report of a serious
disagreement between these two methods of measuring H (E p) .
I % must be teken seriously,however,sSnce Eq.(l)was derived
for free electrons.We note that the results on m e l f s p u n34 35Zr-Ni material generally agree among themselves, even
though in one case Eq.(1) was used to obtain N + ( E _ ) while in
pec i f131,36
the o t h e r çaçe t h i s q u a n t i t y was o b t a i n e d f rom the s p e c i f i c
h e a t . A s i m i l a r s i t u a t i o n h o l d s f o r s p u t t e r e d m a t e r i a l
F u r t h e r m o r e , many o t h e r examples e x i s t o f good agreement
between t besç two methods o f d e t e r m i n i n g N ( E , . ) . Nevertheless,
i t wou ld be u s e f u l t o have f u r t h e r t h e o r e t i c a l s t u d i e s o f
t he a p l i c a b i j i t y o f the GLAG t h e o r y to amorphous s y s t e m s .
F i n a l l y we men t i on the e f f e c t o f a n n e a l i n g on t he super38 ~"
c o n d u c t i n g p r o p e r t i e s . Drehman and Johnsor r s t u d i e d the e_f
f e e t on t h e r m a l r e l a x a t i o n on the s u p e r c o n d u c t i n g propert ies
o f an amorphous Zr -Rh a l l o y . These a u t h o r s obse rved a M 5 &
decrease i n T r , as w e l l as o t h e r changes upon a n n e a l i n g a t
t e m p e r a t u r e s w e l l be low the c r y s t a l i z a t i o n t e m p e r a t u r e . M o r e
r e c e n t l y Qaroche e t a i . • s t u d i e d the e f f e c t o f t h e r m a l r e l £
x a t l o n on amorphous Zr -Cu and no ted the changes w h i c h oc -
c u r r e d i n T-, p, O. , and A. These r e s u l t s y/ere c o n s i s t e n t l y
i n t e r p r e t e d |n terms o f the M c M i l l a n e q u a t i o n f o r T - . l n s p f -
t© o f t h i s , however , the T« r e d u c t i o n sugges ts a r e d u c t i o n
i n t he e f f e c t i v e e l e c t r o n - e l e c t r o n i n t e r a c t i o n . At t h e same
t i m e , a n n e a l i n g has been sliown t o reduce the low t e m p e r a t u -
re s p e c i f i c heat , a r e d u c t i o n wh ich has been a t t r i b u t e d to
a r e d u c t i o n |n the number o f t w o - l e v e l «.ystems ( T L S ) . The
TI.5 a re g e n e r a l l y accep ted as the e x p l a n a t i o n f o r the low
t e m p e r a t u r e the rma l and u l t r a s o n i c anoma l i es o b s e r v e d in
g l a s s e s . A model f o r TLS i n m e t a l l i c g l a s s e s has been d e -h]
v e l o p e d by B?nvi l le and Harr is , and models fo r the enhancement of T-by the TLS in m e t a l l i c g l a s s e s have been proposed . However,
173
recent experiment by Grondey et a l , * In amorphous Zr-Cg
Has thrown doubt on this e x p i r a t i o n of the annealing
Vior of the superconducting properties of amorphous alloy?.
These authors measured the low temperature heat capacity and
thermal conductivity, for several well characterized alloys,
as a function of annealing. Large change? in the thermal çqn.
d.uctivity were clearly observed, but the corresponding ch$n
ge$ in the specific heat were much smaller than expected.
Thus there appears to be some doubt whether the reduction
Of TLS upon annealing really causes the changes in the heat
capacity and thermal conductivity, or if it only occasionally
a,ççompanys these changes.
REFERENCES
1, W. Buckel and R. Hiisch, Z. Physik r3_£, 109 O 9 5 1 » ) ,
3, G. Bergmann, Phys. Reports (Ph/s fetters C) ,2_7, 15?
(1976).
3, N.W, Ashcroft and N-0. Mermin, Solid State Physics (Holt,
Rifiehart, Winston, New York, 1976) p. 520.
hi W.L, McMillan and J.M. R o w e H , Superconcjuct i yi ty , Vol . I ,
edT R. Parks (Marcel Dekker, New York, 1969) p, UH$.
5, K. Knorr and N. Barth, J. Low Temp. Phys. k_t i»69 (1971)»
6, J. 0olz and F. Pobell, Z. Physik ^ 2 0 , 95 (1975).
7, G. Busch and H.J. Guntherodt, in Sol 1 d State Physics,
Vplr 29i eds. H. Ehrenreich, F. Sçítz, and Turnbull
(Academic, New York, 1971») p. 235,
8, 0. HMnderi and R. Ryberg, J. Phys. F k, 2096 n ^ ) .
9, 0. Hunderi, J. Phys. F 5, 2 2 U (1975).
10, G. Bçrgmann, Phys. Rev. B£, 3797 (1971).
U , M.M, Collyer and R.H. Hammond, Phys. Rev. Lett. 30« 92
(1973).
12, J.Et Crow, M. Strongin, R.S, Thompson, and Q.F. K«mmer(?r,
Phys, Lett. 3£A» 161 (1969).
13, W.L, Johnson, S.J. Poon, and P. Duwez, Phys. Rev. 811 ,
150 (1975).
H , W.L, Johnson, J. Appi. Phys. 5£, 1557 (1979),
174
1JJ. W.L. Johnson, in Glassy Metals I , Vol. 46 of Topics In
AppIi ed Phys i çs, ed. H,J. Guntherpdt and H, Peck
(Springer, New York, 1981),1$. C.C. Tsuçl, Ví.L. Johnson, R.B. Laibowitz and J,M, Vigglano,
Solid State Commun. 2£, 615 (1979); D.H, Ktmhi and T. H.Geballe, Phys. Rev. Lett. <r5, 1039 (1980).
17- 6. Eilenbergep and V. Ambegaokar, Phys, Rev. 11 jj>j, 332
(1967).1$. 0. Rainer» G. Bergmqnn, and U. Eckhardt, Phys. Rev, B8,
5324 (1973>-19. E. Bomb «rid W.L. Johnson, J. Low Temp. Phys. 3JJ,, 29
(1979).
20. K. Tçnhpyçr, W.L. Johnson, and C.C. Tsuei, Solid StateCommun, ^ 8 , 53 (1981),
21. W.L. Carter, S.J. Poon, J.W. Hull, and T.H, GebaHi»,Solid St*te Commun, 3.9, i»! (1981).
32. F.P. Mis*el1, R. Bergeron, J.E. Keem and S,R. Qvshinski,Solid St*te Cftmmun, jtj, 177 0 9 8 3 ) .
23. B.H. Clemens, W.L. Johnson, and J. Bennett, J. Appl .Phys, U., 1116 (1980).
2*. M. Tenhover, IEEE Trao*. Meg. Mag-17, 1021 (1911).25. F.P. Migfell and B,B. Schwartz, in Ki rk^Othmer Encyclo-
pedia of Chemical Technolofiy, Vol. .22, ed. M. 6ray»on(John W i U y , New York, 1983) p. 298.
26. W.L. McMillan, Phys. Rev. ^67_, 33 T (1968).27« C M . Varnui and R.C. Dynes, in Super conduct:jyi ty In d-and
f-band Metals, ed. O.H. Douglass (Plenum, New York, 1976).29. J. Fiasck, J. Wood, A.S. Edeistetn, J. Keam, and F, P,
Mis3«ll, Solid State Common. j*±, 649 (i982).29. F.P, MIsjeii, S. FrQta"Pes6Ôa, J. Wood, J. Ty'cr and J.
E. Kçem, Phys, Rev. _B2^t 1Ç96 (1983).
3Q. S. Frota-PessÔa, Phys. Rev. £2j8, 3753 (1983).31. F.P, Misfell and J.E. Keem, Phys. Rev. H O , (19«<i),
32. S.N, Khanna and F. Cyrot-Lackmann, Phys. Rev. I% 1, \h\2
(1980).
33- W, Mlyak^wa and S. Frota-Pessôa, private communication.
$<». Z, AUounian and J,Q. Stroin-Olsen, Phys. Rey. B?7, V H 9
175
(1983).35. D.M. Kroeger, C.C. Koch, J.O. Scarbrough, and C.G.
McKamey, Phys. Rev. 879, 1199 (1984).
D.M. Kroeger, C.C. Koch, C.G. McKamey, and J.0. Scarbrough,
J. Non-Cryst. Solids 61+62, 937 (1984).
36. A. Ravez. J.C. Lasjaunias, and 0. Béthoux, Physica
(Utrecht) 107 B+C, 397 (1981); A. Ravex, J .C. Lasjaunias,
and 0. Béthoux, Solid State Commun. k±, 649 (1982).
37. A. Ravex, J.C. Lasjaunias and 0. Béthoux, Submitted to
J. Phys. F.
38. A.J. Drehman and W.L. Johnson, Phys. Stat. Solidi (a)
II, 499 (1979).39* P. Garoche, Y. Calvayrac, W. Cheng, and J.J. Veyssié, J.
Phys. V\l, 2783 (1982).
40. S. Hunklinger and W. Arnold, in Physical Acousti cs Vol .12,
(Academic, New York, 1976) p. 155.
41. M. Banvilie and R. Harris, Phys. Rev. Lett. _44_, 1136
(1980).
42. R. Harris, L.J. Lewis and M.J. Zuckermann, J. Phys.
F Jl, 2323 (1983).43. S. Grondey, H.V. Lühneysen, H.J. Schink, and K. Samwer,
Z. Physik 35±, 287 (1983).
Work Supported by FAPESP, CNPq and FINEP (Brazil)
176
ON THE SUPERCONDUCTIVITY OF VANADIUM BASED ALLOYS
F. Brouers
Freie U&iversitat Berlin, Inst i tut für Theoretische Physik
Arnimallee lkt D-1000 Berlin 33
and
Department of Physics, University of the West Indies
Mona, Kingston 7, Jamaica W. I .
and
J. Van der RestInst i tut de Physique, Université de Liege,Belgium
Abstract
We have computed the electron density of states of solid
solutions of vanadium based transition metal alloys V 5Í by using
the tight-binding recursion method for degenerate d-bands in order
to calculate the alloy superconducting transition temperature with
the McMillan formula. As observed experimentally for X on the left
hand side of V in the periodic table one obtains an increase of T
while for X on the right hand side of>V the cr i t ical temperature
decreases.
The detailed comparison with experiments indicate that
when the bandwidths of the two constituents are different, one can-
not neglect the variation of the electron-phonon interactions.
Another important conclusion is that for alloys which are
in the split-band limit like VAu, VPd and VPt, the agreement with
177
experimental data can be obtained only by assorting that these alloys
have a short-range order favouring clusters of pure vanadium.
1. Introduction
There i s currently an active interest in the effect of
disorder on the superconducting properties of materials. Superconduc-
tivity in alloys and amorphous metals and alloys has been the subject
of a great number of experimental and theoretical studies.
From these efforts, one expects to obtain not only a better
understanding of the basic nature of superconductivity in highly dis-
ordered metallic systems but also some important technological appli-
cations of new class of materials. This interest is related among
others to the search for superconductors which show minimum degrada-
tion of superconducting properties by neutron irradiation.
To analyze the variation of the critical superconducting
temperature with disorder, experimentalists use the McMillan formula
(McMillan 1968)
( ! )X -u* - 0.62Xu*
6. i s the Debye temperature and X is the Fermi surface averaged
enhancement of the electron mass which corresponds to a reduction
of the electron velocity due to electron-phonon scattering by a
factor (1 + A) McMillan formula (1) is an extension of the BCS
formula
Tc = 1.13 8D exp | £• | (2)
178
obtained by introducing an electron mass enhancement in the two-
particle propagator of the "vertex function", which has the effect
of replacing the exponent A~l by (1 + A)/A, by taking account of
the effective Coulomb repulsion u* between the electrons and by
considering the T-dependence of X.
Moreover, McMillan (1968) showed that X can be factorized
into electronic and primarily phonon dependent factors by writing
H(Er) < I2 >A = i (3)
M < o»2 >
where H(E_) i s the bare density of states at Fermi level, <I2> the
Fend surface average of the square of the electron-phonon matrix
elements and
« 2
/dw o2 (u)F(w)w~1
and average over the frequency distribution fmction multiplied by
a factor a(u) depending on the phonon induced electron-electron
scattering.
For transition metals, the calculated values of < Iz> and M< (o2>
show a maximum near the middle of the series.
It is not surprising that the electron-phonon interaction should be
strong near the center of the transition metal series. Assuming the
atomic orbitais to decay exponentially as 0(r)exp(-qQr), one has
< I2 > * q* W2 (5)
where w, is the d-bandwidth which peaks near the middle of the
transition series.
179
Although the variation in < I2> overcomes the variation of
M < w2 > , the parallel variation of < I2> and M <u>2 > with atomic
number has led to the conclusion that one can consider the ratio con-
stant and the effect of disorder on T has been mostly discussed by
considering only the variation of N(Ep)/pynes and Vanna 1976/.
However, the comparison of the superconductivity of crystalline and
amorphous phases of transition metals has shown deviations with respect
to this simple rule.
Recent comparative studies of crystalline and amorphous phase
of Nbjàe suggest that there is no strong correlation between N(E_) and
T in strong coupled A15-superconductors /Tsui 1980/. In particular,
the high T superconductivity of Nb.Ge does not appear to stem from
an unusually high N(E_), but rather from a relatively strong electron-
phonon coupling enhanced by some localized electron states. Hanke et
al. (1976), Cowan and Carbotte (1978).
These recent works show the interest of a systematic study
of the variation of T with N(Ep) and electron-phonon interaction in
disordered systems.
In the present paper we want to analyze the experimental
data of Daumer et al (1982) who have studied the superconductivity
of vanadium solid solution VQ X where X is a transition metal.
These authors have applied the McMillan formula assuming that A varies
only with N(Ep). The agreement is good for the alloys where specific
heat measurements are available. For the others these authors have
estimated N(E_) from the measured values of T .r C
180
We have made a systematic calculation of the density of
states at Fermi-level of these solid solutions as well as other for
which data are not yet available* We have used the technique of
calculating the S-fold degenerate d-band density of states froa a
continuous fraction expansion of the tight binding Green's function
with the help of the recursion method of Haydock, Heine and Kelly
(1980). We did not try to f i t the experimental data. We have chosen
the alloy and disorder parameters used by Van der Rest et al. (1975)
to calculate alloy energies of formation.
From these results, we can conclude that, when the bandwidths
of the two constituents are different, one cannot neglect the variation
of the electron-phonon interaction as i t was assumed by Da user
(19B2).
2. Method of Calculation
We have calculated the density of states using the recursion
method of Haydock, Heine and Kelly (1980).
For each alloy, we have calculated the local densities of
states on a V and on a X atom for the seven most probable configurations
of the first neighbors. The occupation of more distant atoms i s
chosen randomly. For a concentration of 90% of V, we have treated
exactly 96% of the first neighbors possible configurations. The
average density of states i s obtained by averaging over these partial
densities of states.
We have used the BCC structure and we have considered the
first and second neighbors hopping integrals and calculated the first
181
seven steps of the continuous fraction. This corresponds to clusters
of 1695 atoms. We have only considered the d-density of states. The
s-density of states i s supposed constant aid equal to the inverse of
the s-bandwidth.
The energy levels E. , the electron nushers N. as well as
the d-bandwidths of the pure metals (W.) are given in Table 1.
The hopping integrals between two V neighboring atoms are
taken to be (in eV)
dd o1 = - 1.05, dd TTj = 0.5, dd 6± = - 0.065
The hopping integrals between second neighbors are taken to be
(for a discussion of these values see Derenne et al (1982))
dd o2 = - 0.i»3, dd Tt2 ? 0.16, dd 62= - 0.016.
The hopping integrals between the X atoms are obtained by multiply-
X V
ing the V-hopping integrals by the ratio W./W.. The hopping in-
tegrals between a V-atotn and a X-atom are obtained by taking the
geometric average of V- and X- hopping integrals. When the s-density
of states i s added to the d-density of states, the f i l l ing of the
d-band i s fitted in order to obtain the pure vanadium N(E-) calculated
by Glotzel et al. (1979).
The energy levels and charge transfers have been calculated
self-consistently.
3. Numerical Results
The results of our calculations are summarized in Table 1.
182
The e and W parameters are the ones used by Van der Rest et al.
(1975). To calculate T we have used McMillan formula (1). lhec
Debye temperature is the average of the pure &.. T is the
critical temperature assuming that X depends only on N(Ep). The
proportionality constant i s chosen to f i t the experimental T of(2)pure vanadium. T is the calculated critical temperature assuming
that the electron-phonon interaction varies as the averaged bandwidth
following Eq. (S).
t . Conclusions
From the observation of Table 1 , one can conclude that
1) when X i s on the lef t hand side of the periodic table with
respect to V, T i s enhanced, when X is on the right hand
side i t decreases (Zr and Hf are between brackets because
they do not form 10at% solid solutions);
2) when the d-bandwidths are different, one cannot neglect
the variation of the electron-phonon interactions. It i s
interesting to note that such an effect appears already
for alloys of relatively small (10at%) X-concentration.
3) The calculations (Brouers and Van der Rest 1981) reveal
that the three alloys which exhibit a large discrepancy
with experimental data (AuV, PdV, PtV) are in the spl i t
d-band limit and have partial densities of states the
shape of which varies strongly with local environment.
Cohesive energy calculations indicate that the short-range
order favours the formation of pure V-clusters in the alloy.
183
In that case, the density of states H(E_) of Pt and Att
are reduced respectively to 1.11 and 1.12 and the respective
Tc to 0.<t and 0.3 °K which i s in agreement with experimental
data.
Our calculation show that more correlated experimental data (specific
heat and critical teaperature measurements) are needed for nore alloys
and for different concentrations i f one wants to have a better under-
standing of the effect of disorder on superconductivity of alloys.
184
TABLE 1
V
Cr
Ma
(Zr)
Mb
Mo
Pd
(Hf)
TaW
Re
Pt
A.U
Ed(eV)
0
-0.5
-1
0.6
-0.1
-0.8
-3.6
0.6
-0.1
-0.8
-1.5
-3.6
-5
Wd(eV)
7
6.5
6
7
9
9
6
8
11
11
10.5
7.5
1.5
N(Ep)(StateseV/at)
1.80
1.51
1.45
2.01»
1.75
1.15
1.25
2.00
1.72
1.50
1.17
1.22
1.32
T (Dc
5.3
2.6
2.1
7.6
4.2
2.3
0.8
7.2
4.1
2.6
0.6
0.7
1.2
T c( 2 )
5.3
2.5
1.9
7.6
5.9
3.1
0.7
7.5
6.3
4
1.1
0.7
0.9
5.3a)b)
2.818) - 2.166)
2.1ia)
».08a)" - 5.6C)
2.85a) - 3.1b)
0.37a)
-
3.59a) - 4.47C)
3.5«*a) - 4.08d)
0.76a)
0.37a)
0.30a)
a) Daumer et ai (1982)
b) Andres et ai (1969)
c) Corsan and Cook (1970)
d) Shivkov et ai (1975)
e) R. Kuentzler (1982)
185
References
- Andres, K., Bucker, E . , Maita, J .P. and Sherwood, R.C (1969)
Phys. Rev. 178_, 702.
- Brouers, F. and Van der Rest, J. (1984) t o be published.
- Corsan, J.M. and Cook, A.J. (1970) Phys. Stat . Sol. W,, 657.
- Cowan, N.B. and Carbotte, J.P. (1978) J. Phys. C 11^ L 265.
- Dauner, W., Kettschau, A. , Khan, H.R., Lviders, K., Raub, Ch.J.»
Rieseneier, H. and Roth, 6. (1982) Phys. Stat . S o l . ( b ) . 112_, K67.
- Derenne, M., Brouers, F. and Van der Rest, J. (1982) Phys. Rev.
B26, 1048.
- Glotzel , D. , Rainer, D. and Schrober, H.R. (1979) Z.Physik
B35, 317.
- Hanke, W., Hafher, J. and Bi lz , H. (1976) Phys.Rev.Lett. £7 , 1560.
- Haydock, R., Heine, V. and Kelly, M.J. (1980) Solid State Physics
35, (Acad. Press) .
- Shikov, A.A., Chemoplekov, N.A., Panova, G.K., Samoilov, B.N.
and Zhernov, A.P. (1975) JETP 4_2_, 927.
- Tsuei, C.C. (1980) in "Superconductivity in d- and f-band metals,
ed. by H.Suhl and M.B. Maple, Acad.Press p.233.
• Van der Rest, J . , Gautier, F. and Brouers, F. (1975) J. of Phys. F5_,
2283.
186
DENSIDADE DE ESTADOS ELETRÔNICOS EM VIDROS METÁLICOS
SÔNIA FROTA-PESSOA
Instituto de Físisa, Universidade de São Paulo, CP.20516»
São Paulo, SP.
INTRODUÇÃO - O conhecimento da densidade de estados eletrô-
nicos em geral, e da densidade de estados na energia de Fe_r
ml em particular» ê crucial para um melhor entendimento do
comportamento dos materiais. Freqüentemente a diferença e£
tre as propriedades de uma mesma liga, na forma cristalina
e na forma amorfa pode ser atribuída a diferentes densida-
des de estados eletrônicos para a liga nessas duas formas.
Assim, dentro do curso de vidros metálicos, essa parte será
dedicada ao estudo de densidade de estados eletrônicos em me
tais amorfos, com especial ênfase para ligas amorfas de me-
tal de transição.
1. Comentários gerais - Um cálculo exato da densidade de es_
tados pára -mate ri a is amorfos, ao nível de sofisticação atí£
gtdo em materiais cristalinos não é factível. A própria es_
trutura amorfa é de difícil determinação. Além disso os m£
todos mais poderosos para o cálculo de densidade de estados
são implementados no espaço recíproco que no amorfo é esseji
cialmente perdido. Felizmente, devido a diversidade de ar-
ranjos locais existentes no amorfo, a curva da densidade de
estados como função da energia é suave (ver fig. I) e pode-
mos usar com bons resultados métodos de pouca resolução em
energia, que não seriam úteis no caso cristalino.
Antes de prosseguir gostaria de dividir os metais em
duas categorias; os metais simples cujos elétrons de valên-
ela são s_ e £, e metais mais complicados envolvendo metais
de transição.
1.a. Metais Simples - Para esses sistemas a densidade de e$
tados é usualmente bem representada pelo modelo do elétron
livre* . Esse modelo funciona melhor para metais simples
amorfos que para cristalinos, devido a ausência de planosli.)de Bragg. Várias teorias de sucesso* ' (como a teoria de
Zlman para condutividade) tomam por base o modelo de elétron
livre para representar metais simples amorfos.
187
(b)
1.b. Sistemas envolvendo metais de transição - Como elétrons
_d não são bem representados por um modelo de elétrons livres
a situação neses materiais é mais complicada e um cálculo
de bandas mais
sofisticado se
faz necessário
A fig. I i lus- (o)
tra a densida-
de de estados
N ( E ) , tomada
por átomo do
material, para
a Landa d em
um material
cristalino
(fig. la) e em
um material a-r
morfo (fig.1b
A área sob as
curvas é a me_s
ma uma vez que em ambos os casos cabem 10 elétrons por át£
mo na banda d^ Para ilustrar o comportamento de uma liga
binaria, na fig. 2 mostramos uma comparação entre a densi-
dade de estados cal-
0.1 Ryd 0.1 Ryd
£o.» 2
Ii
m
]uiZ
1
ii
|
11/'
i— Zr3 Cu
—1 -
-5
Fig. 2 E (eV)
photoelectron espectroscopy). ,6 ,7 ,8)
culada para Zr-Cu nu-ma estrutura cristalj_na densa ( l inhacheia) e para um sis-tema amorfo<x-Zrg7Cu-, * ' (1 inhaponti1hada).2.a. Medidas de H(E)-2?_ XPS (X-ray photo-electron espectroscopy)e UPS (u l t rav io let
As técnicas de XPS e UPS dão uma idéia da forma deN(E)
como função da energia para energias menores que a energia
de Fermi E^. As duas técnicas diferem essencialmente pela
188
energia utilizada, a primeira usando raio X e a segunda uti-
lizando luz na região do ultra-violeta. Embora os resulta-
dos de XPS e UPS retratem a grosso modo a densidade de esta
dos do material, existem distorções pois os elementos de n»£
triz para transições óticas podem ser fortemente dependen-
tes da energia. 5 Normalmente espectros XPS são mais con-
fiáveis pois distorções que favorecem energias na vizinhan-
ça da energia de Fermi sio comuns quando são usados fotons
de mats baixa energia.
Por outro lado a res£
luçio (possibilidade
de distinguir estrutu
ras finas em N(E)) ê
melhor em medidas fej_
tas por UPS. Na fig.3
mostramos resultados
de UPS*6* para várias
concentrações da liga
amorfa de ZrCu.
2.b. Medidas de N(Ef) - Calor específico a baixas tempera-
turas, ' ' medidas de H , contra temperatura T em super-(11 in) c í
condutores* ' ' e medidas de susceptibi 1 idade de Pauli.i) Calor específico e baixas temperaturas - Em metais nor-
mais a baixas tempera
turas o calor espec£
fico C é da forma:
8 7 e sFig. 3
3 2 1 0 eV
BINDING ENERGY
C - Y T • 0 TJ (DPodemos avaliar
á densidade de esta
dos na superfície die
Fermi N(Ef) através
da constante y. 0
coeficiente 8 nos dá
a temperatura de
Debye 8Q, relaciona,
do com as freqüên-
cias de fonos no nu»
terial. Usualmente
V
10OO
IOO0
v •o
m
4
•
* : '
Fia
0W
5f
. 4
i
to i<
-Vi,
•
189
os dados experimentais sao arranjados em um gráfico de C/T
contra T como na fig. 4. Para um metal normal os pontos
em geral se alinham em uma reta de coeficiente angular B. 0
valor da constante y é dado pelo ponto de intersecçlo da re
ta com o eixo vertical. Na fig. k mostramos dados experimen(9) para uma amostra de a-2r7,Ni2.. Notetais de Ravex et.ai
que além da reta esperada, observamos um pico a baixas tern
peraturas. Esse pico está associado a uma transiçio super-
condutora, com temperatura de transição T ^3K, sofrida pe-
lo material. Para obter y e 6, ignoramos esta estrutura e e>c
trapolamos a reta observada na regiio normal T > T , para re
giões de T < T (parte pontilhada na figura). Se y é medido2 c
em mJ/mole-K podemos obter a densidade de estados na supejr
ffcie de Fermi N (E.) em estados/eV-ãtomo-spin através da
relação N*(Ef) = 0,212Y. A densidade de estados N*(E~) é
chamada densidade de estados vestida e incluí efeitos ass£
ciados com a dinâmica da rede. A densidade de estados
N(E-) obtida a partir de cálculo de bandas se relaciona com
N (E,) através da constante de acoplamento eletron-fonon ^
na forma dada abaixo:
M*(Ef) = N(E f). ( 1 + X) (2)
t interessante notar que ^ ê en geral a mesma constante
de acoplamento que aparece em supercondutividade.
ii) Medidas de JíC2_contra T - Medidas de calor especffico
são bastante trabalhosas, e es-
pecialmente difíceis de obter
em filmes finos. No entanto, no
caso de materiais superconduto-
res, N*(Ef) pode ser obtida de
forma simples através de medi-
das do campo crítico superior
Fig.5
H 2 como funçio da temperaturana vizinhança de T ' usand
a relação obtida a part i r de
teoria de Ginzburg-Landau gen£
ral i z a d a ( U ) dada abaixo:
x-0.13 \ V0.41V0.37 \0.24
T(K)
. N * ( E f ) ( 3 )
190
A resistividade normal pode ser estimada se a geometria da
amostra for conhecida. Como a constante de Boltzmann, a
carga e e a velocidade da luz c são constantes bem conheci-
das, H(E,) pode ser facilmente determinada. Note que esse
procedimento não fornece o valor de 9_. Na fig. 5 ntostra-(12)mos curvas de H . contra T amostras de ot-ZrRh
iíi) Medidas de susceptibi1idade - A contribuição de Paulí
para a susceptibi1 idade magnética fornece diretamente N(E_}
No entanto medidas de suscept i bi 1 idades são pouco usadas pj»
ra a determinação de N(E,) por ser extremamente difícil se_
parar a contribuição de Pauli das outras contribuições.
3. Densidade de estados local - Muitas vezes no caso de ma_
teriais amorfos, é* conveniente definir uma densidade dê es_
tados local N. (E) em torno de um átomo, de forma que N. (E)
dE dê o número de estados eletrônicos entre E e E+dE asso-
ciados com o átomo considerado. Sob esse ponto de vista a
densidade de estados N(E) do material pode ser considerada
como uma média sobre as densidade de estados locais dos vã
rios átomos que constituem o material. Se temos um número
ji de átomos na liga podemos escrever:
N(E) = l N[(E)/n
i « t o d o sos átomos
Para uma liga binaria, onde temos n. átomos do tipo J
e n . átomos do tipo 2^, a densidade de estados N(E) pode ser
escrita em termos das densidade de estados locais de cada
átomo como:
n.N(E) - l N[(E) + l N[(E) - n ^ U ) + n2N2(E) (5)
i 3 átomos i « átomostipo 1 tipo 2
onde, em analogia com a definição anterior, tomamos N . ( E ) ,
a densidade de estados para átomos do tipo 1 como uma média
sobre as densidades de estados locais dos n^ átomos desse
tipo.
Num composto cristalino onde todos os átomos do tipo I
são equivalentes N. (E) é dado pela própria densidade de es_
tados local. No caso amorfo no entanto, átomos de um mesmo
191
tipo podem ter densidades de estados locais distintas (ver
fig. 8), por se encontrarem cercados por arranjos locais dj^
ferentes.
Notando que c,«n,/n ê a concentração de átomos do tipo
1 na liga binaria podemos finalmente escrever:
N(E) - CjM^E) + c2N2(E) (6)
onde N.(E) e N_(E) são definidos como uma média sobre densj[
dades de estados locais para átomos do tipo 1 e do tipo 2
respect ivãmente.
k. Modelos de banda rígida para sistemas binários - Os mo-
delos mais simples usados para representar a densidade de
estados numa liga binaria são baseados no modelo de banda
rígida* 1". Nesse modelo a densidade de estados da liga me
tâlica ê escrita em termos das densidades de estados dosrae
tais puros N° (E) e N^ (E) como:
N(E) » CiNf(E)+c2Nj(E) (7)
por exemplo, se estamos estuda'-?* ^ W g a de ZrCu, N^u(E)
seria a densidade de estados do Cu puro e N*° (E) a densi-
dade de estados do Zr puro.
0 modelo de banda rígida
embora simples ê inapro-
priado, principalmente no Fig.6
caso em que as bandas dos
materiais puros estão ceji
tradas em torno de ener-
gias E° e E°, (ver figura) *__
muito afastada entre si, Ej' EJ
caso bastante comum em li-
gas amorfas. Para salien-
tar as falhas do modelo dividiremos, por uma questão de cl£
reza suas origens em duas partes:
I- o modelo falha porque ignora interação entre átomos de
tipos diferentes presentes no composto.
li- mesmo que essa interação fosse nula o modelo continua
a fornecer resultados espúrios uma vez que não considera o
fato de que um átomo tem um menor número de vizinhos de sua
192
espécie no bínirio que no metal puro.
Podemos agora tentar ver como seria modificada a densi-
dade de estados se tentássemos corrigir as falhas do modelo
acima para obter uma banda mais realística para a liga
Vamos primeiramente ignorar a interação entre átomos distin^
tos (isso pode ser feito se E ]> > E «) e tentar corrigir a
segunda falha ( i í ) . Podemos visualizar as bandas d_ de um
metal dè transição dentro de um formalismo tipo "tight-bin-
ding". Nesse formalismo a largura da banda depende das in-
tegrais de "overlap" entre
os elétrons do átomo consj^
derado e seus vizinhos. C£ F i g . 7
mo os átomos de um dado tj_
pò no binârio tem menos v_i
zinhos que no metal puro
(lembre que só estamos con
siderando interação entre
átomos do mesmo tipo e po_r
tanto vizinhos do tipo opojs
to não conta!) as bandas
associadas a cada tipo de átomo no binário devem ser mais
estreitas que as obtidas pelo modelo de banda rígida (ver
figura 7 ) . Queremos agora descrever o efeito da interação
entre os átomos de espécie oposta sobre as densidades da es_
tados N.(E) e N 2 ( E ) . A inclusão dessa interação resulta n £
ma hibridízação que permite qie existam elétrons localiza-
dos em átomos do tipo 1 (portanto contribuindo para N.(E))
com energias características do átomo de, tipo 2. Na figura
£ Ilustramos a densidades de estados Nj (E) e N_(E) de um bj_
nãrio usando o modelo de banda rígida. He figura 7 ê feito
um esboço do que se espera para essas densidades de estados
para o mesmo binário num modelo.mais realístico. Alguns rno
dêlos mais sofisticados para tratar o amorfo são discutidos
a segui r. ......-.,...,.•
5. Métodos teóricos roajs usuais - Os métodos teóricos usar
dos para calcular densidades de estados em metais amorfos
podem ser arbitrariamente divididos em dois grupos princi-
pais* 0 primeiro deles engloba métodos que simulam a liga
amorfa através de estruturas mais tratáveis (cristais, pe-
193
quenos aglomerados etc..) Entre esses se destacam:
i) Os que simulam o amorfo através de estruturas periódicas
de empacotamento denso e utilizam métodos de espaço recíprc»
co, exatos e confiáveis para achar a densidade de esta-
ii) Os que usam uma rede cristalina e simulam desordem ape_
nas quanto ao tipo de átomos que ocupa cada ponto da ride.
Esses sistemas podem ser tratados através de cálculos tipoCPA.(17,18)
iii) Os que simulam amorfos através de aglomerados com pou-
cos átomos embebidos num meio efetivo usando métodos apro-
priados para a soluçio desse problema.
Por outro lado temos métodos mais simples, baseados n j
ma expansão da função de onda em termos de orbitais locali-
zados (LCAO), que podem ser implementados no espaço real e
procuram tratar a estrutura amorfa de forma mais realfstica.
Dentro desse espírito se destacam dois métodos o método dos
momentos ' ' e o método de recorrência. * Esses mé
todos fornecem a densidade de estados local e são bastante
semelhante em princípio diferindo apenas no processo usado
para, a partir da Hami1 toniana, obter a densidade de esta-
dos. Em ambos os casos, para apl'car o método é necessário
que se conheça a Hamiltoniana do sistema. Normalmente é t£
mada uma Hamiltoniana tipo (LCAO) em forma matricial obtida,
como veremos adiante através de parametrizaçio a cálculos
cristalinos. Para obter a densidade de estados para a ban-
da jd de um "cluster" amorfo de N átomos devemos construir
una matriz Hamiltoniana de dimensões 5Nx5N, uma vez que o
nível d é 5 vezes degenerado. Essas matrizes, normalmente
muito grandes, (é comum encontrarmos cálculos para "clusters"
de até 1000 átomos) são bastante esparças uma vez que den-
tro do formalismo "tight-binding" cada átomo interage ape-
nas com seus vizinhos próximos.
Para ilustrar a construção da matriz Hamiltoniana no
formalismo (LCAO) vamos considerar como exemplo o caso de 3
átomos distribuídos como mostrado abaixo, supondo a exístên
cia de um orbital localizado por átomo. Nesse caso tomamos
por base o conjunto { $. , $ *f <|>, } onde <f>. representa o es_;
tado localizado no átomo i. Considerando apenas interação
194
entre primeiros vizinhos a Hami1toniana fica:\
el
* 3 " "I *" ^ °0 0 e3
(8)
onde e. = < Í>.|H| $.> e t.. = < $.|H|$.> e nessa aproxima-
ção t.. » 0 se i e j nio são primeiros vizinhos. Em geral
podemos identificar os termos diagonais da Hamiltoniana com
energias efetivas associadas com os orbitais local izados ,eji
quanto que os termos não diagonais t.. constituem as chama-
das integrais de "hopping". No caso de estarmos interessa-
dos no calculo de uma banda ji para um ":!uster" de N átomos
para construir a Hamiltoniana devemos tomar por base um co_n
junto de 5N funções localizadas constituidos pelos 5 orbi-
tais d de cada átomo. Ainda assim os elementos da diagonal
serão energias efetivas dos orbitais localizados e os fora
da diagonal integrais de "hopping". São esses elementos que
devemos determinar.(24 25)
6. Parametrizaçao da Hamiltoniana * - Nos cálculos de
densidade de estados pelo método de momentos ou pelo método
de recorrência devemos fornecer a Hamiltoniana do sistema.
Se tentarmos calcular as energias e as integrais de "hopping"
diretamente supondo que as funções localizadas sejam fun-
ções atômicas e o potencial envolvido uma superposição poteji
ciais atômicos, os resultados podem ser insatisfatórios.
Isso porque embora seja apropriado tomar funções bases loca
lizadas para tratar o metal, elas podem não ser estritamen-
te atômicas, o mesmo acontecendo com os potenciais. Por ijs
so, se existe disponível um cálculo das bandas de energias
para um composto cristalino para o mesmo material que quere
mos estudar na forma amorfa, é mais confiável obter a Hamilí 2 5T
toniana através de parametrizaçao como descrito abaixo:
- Supomos que a estrutura de bandas do sistema cristalino
possa ser obtida corretamente, não só através do método exai
to usado no cálculo original, mas também através de um for-
ma li smo LCAO, desde que os orbitais localizados e os poten-
ciais usados na expansão sejam apropriados.
- Achamos então soluções tipo LCAO para as bandas do com-
195
posto cristalino, mas como não conhecemos a forma apropria_
da para os orbitais e potenciais deixamos as energias efe-
tivas e. e as integrais de "hopping" t.. como parâmetros a
serem determinados.
- Estes parâmetros são então escolhidos de forma que os
resultados LCAO para as bandas do composto cristalino repro
duzam os da melhor forma possível resultados do cálculo exa
to.
- As integrais de "hopping" assim obtidas são caracterís-
ticas da rede cristalina considerada e não podem ser usadas
para o material na forma amorfa. No entanto para orbitais
á_ essas integrais podem, em primeira ordem, ser expressas
em termos de 3 integrais moleculares (ddo, ddir e dd6) e da(25)
geometria da rede. Como a geometria da rede cristalina
é conhecida podemos usar as integrais de "hopping" já detejr
minadas para obter as integrais moleculares. Como as inte-
grais moleculares só dependem do tipo de átomos envolvidos
e da distância entre eles, elas podem (com correções apro-
priadas para variação da distância entre os átomos) ser u-
sadas para caracterizar a interação entre os átomos também
no caso amorfo.
- Finalmente conhecendo as integrais moleculares e a es-
trutura (posição dos átomos) do "cluster" amorfo considera
do, podemos construir as integrais de "hopping'1 e assim oj>
ter uma Hamiltoniana para caracterizar a liga amorfa.
Obter a Hamiltoniana pode ser como vimos acima um pro-
cesso trabalhoso. Em alguns casos no entanto (ligas biná-
rias de metal de transição por exemplo) podemos obter bons
resultados para a liga binaria usando apenas integrais mo-
leculares obtidas para os metais puros. Quando isso acon-
tecer o procedimento se simplifica pois podemos em muitos(26)casos usar integrais moleculares estão tabeladas .
7. Alguns resultados - Para finalizar apresentamos aquí al-
guns resultados obtidos pelo método de recorrência para a
densidade de estados eletrônicos em ligas amorfas. ' ' Co
mo ressaltamos anteriormente o método de Recorrência forne-
ce a densidade de estados 1oc.il em torno de cada átomo. No
entanto como vimos na secção 5» conhecendo as densidades de
estados locais podemos obter as densidades de estados N|(E)
196
e N-(E) para
cada um dos
componentes da
11 ga, bem co-
mo a densida-
de de estados
da liga N(E).
Nas figs.
8a e 8b mos-
tramos a de£
sidade de es_
tados local
para dois ã-
tomos de Cu
e dois de Zr
para um clu£
ter com per-
to de 600 !t£
trios simuian-
I>.DC
«5
40
20
1 A
- , i \'VV
I» V/ / v
(a) Cu
-0.4
Fig. 8
-0.3 -0.2 -0.1
E(Ry)
1
•en -0.2
Fig.9ElRy)
Fig.10
197
a-Zr^CuçQ. A diferença entre as densidades de estados p£
ra dois átomos do mesmo tipo no "cluster11 e bastante sign£
ficativa e se deve ao fato de que os átomos em questão es-
tão cercados por diferentes arranjos locais.
Na fig. 9 ê mostrada a densidade de estados N(E) calcu
lada para a-Zr^Cu-,, bem como as densidades de estado
M C u(E) (linha cheia) e N Z r(E) (linha pontilhada) para cada
componente da liga. Resultados para a-Zr^Ni,, são mostra
das de forma análoga na fig. 10. A densidade de estados
N(E) mostrada na parte superior das figuras inclui, além da
banda <1, contribuições da banda js.
Na figura 11 comparamos a densidade de estados N(E) caj
culada para duas composições de oc-ZrCu com dados de UPS .
(b)
Fig.ll
(c)
cQ.
eoo
O)
O)
BVi
UJ
2
1
0
I
I•i
J. f• 1rIji
i
i
i
— i ' i
\ fVI
\'wY >1 \\ \\ \\ \
\ \
i
'41 C U 5 9
%7 CU33
__
—-—r^
I
E(eV)
198
A densidade de estados calculada para o Cu fcc também ê mo£
trada na figura lia.
Finalmente na fig. 12 a densidade de estados na energia
de Fermi calculada para duas composições de a-ZrCu ê compa-
rada com valores de N(E,) obtidos a partir de dados experi-
mentais de Minnige-
rade et ai* 2 7*. Na
figura valores cal-
culados sio indica-
dos por círculos e£
curecidos enquanto
que os dados experj_
mentais sio repre-
sentados por quadra oj . «• - "" cãdos. Aqui convém •%
ressaltar que medi- 2
das de calor especj_
fico e H - contra T
fornecem N*(E C), a
densidade de esta-
dos vestida relacio
nada com N(E-) através da constante de acoplamento eletron-
-fonon A. Uma estimativa do valor de X pode ser feita, u-Í28)
sando a equação de McMillanv se conhecermos a temperatu-
ra de Debye 6. e a temperatura crítica supercondutora T do
material. Os pontos experimentais da figura 12, foram obtj_
dos a partir de N*(E_) usando valores estimados para X.
Devemos ressaltar que nem sempre a concordância entre
resultados teóricos e experimentais, principalmente no que
diz respeito a N(E_) é tio boa quanto a mostrada aqui. No
caso de a-Zr^.N!-. por exemplo a densidade de estados na
calculada energia de Fermi é quase duas vezes maior que a(12)obtida a partir dos dados experimentais de Altounían etar
para fitas fabricadas por "mel t-spinning". 0 problema no eji
tanto não está necessariamente associado ao cálculo de
N(E-). Antes de calcular a densidade de estados devemos e£
colher um "cluster" que seja representativo da estrutura da
liga. A discordância entre os resultados teóricos e experj_
mentais podem ser fruto da escolha de um " cluster" inapro
199
priado. Nesse contexto convém ressaltar que a estrutura
amorfa nio ê única, sendo amorfa qualquer disposição nio
cristalina dos átomos. £ interessante notar que ligas de
mesma composição podem exibir diferentes densidades de es-
tado N*(E_), dependendo da técnica de fabricação. Em a-ZrNi
por exemplo ligas obtidas por "sputtering"*°* ' tem N*'EF'
bastante mais elevado que as obtidas por mel t-spinningv '*" .
REFERENCIAS
1) G. BERGMANN, Phisics Reports (Physics Letters C) 27, 159
(1976).
2) G. BUSCH and H.-J. GUNTHERODT, in "Solid State Physics"
Vol. 29, (197*).
3) V.L. MORUZZI, C D . GELATT, Jr. and A.R. WILLIAMS, "Caleu
lated Properties of Ordered Alloys" - a ser publicado.
k) SÔNIA FR0TA-PESS0A, "Cálculo de Densidade de Estados em
Ligas Binârías Amor fas de Metal de Transição; uma apltc£
ção a a-ZrCu e a-ZrNi" - Tese de Livre-Docência - USP
(1983); Phys. Rev. B2£, 3753 (1983).
5) P. OELHAFEN, in "Glassy Metals II", ed. H. Beck and H.
-J. GUNTHERODT, Topics in Current Physics (Springer
Verlag, New York) - (a ser publicado).
6) P. OELHAFEN, E. HAUSER, H.-J. GUNTHERODT and K.H.
BENNEMANN, Phys. Rev. Lett. j«3., 1 13^ (1979) -
7) A. AMAMOU and G. KRILL, Solid State Co mm. 2£, 957 (1978).
8) A. AMAMOU, Solid State Comm. 3j}, 1029 (1980).
9) A. RAVEX, J.C. LASJAUNIAS and 0. BÉTHOUX, Physica 107B,
367 (1981).
10) D.M. KR0EGER, C.C. KOCH, J.O. SCARBROUGH and C.G.
MCKAMEY, Phys. Rev. JJ29, 1199 (1984).
11) F.P. MISSELL, S. FROTA-PESSÕA, J. WOOD, J. TYLER and
J.E. KEEM, Phys. Rev. £2_7, 596 (1983).
12) Z. ALTOUNIAN and J.O. STROM-OLSEN, Phys. Rev. B27,
J>U»9 (1983).
13) W.L. JHONSON, J. Appl. Phys. ££, 1557 (1979).
200
14) 0. RAINER, G. BERGNANN and U. ECKHARDT, Phys. Rev.
B8, 5321» (1973).
15) A.R. MIEDEMA, J. Phys. F4_, 120 (1974).
16) V.L. MORUZZI , P. OELHAFEN, A.R. WILLIAMS, R. LAPKA,
H.-J. GUNTHERODT and J. KUBLER - a ser publicado.
17) G.M. STOKS, W.M. TEMMERMAN and B.L. GYORFFY, Phys. Rev.
Lett. k\_, 339 (1978).
18) A. BANSIL, Phys. Rev. B££, 4025 (1979); B2J), 4035
(1979).
19) B. DELLEY, D.E. ELLIS and A.J. FREEMAN, Journal de
Physique £8, *»37 (1980).
20) R.H. FAIRLIE, W..M. TEMMERMAN and B.L. GYORFFY, J. Phys.
F: Met. Phy=. JJ2, 1641 (1982) .
21) F. CYROT-LACKMANN, Journal de Physique CJ_, 67 (1970).
22) S.N. KHANMA and F. CYROT-LACKMANN, Phys. Rev. B Z\_,
1412 (1980).
23) R. HAYDOCK, V. HEINE and J. KELLY, J. Phys. C (Solid
State Phys.) j>, 7.345 (1972).
24) R. HAYDOCK, in Solid State Physics, Vol. 35, ed. H.
Ehrenreich, f. Seitz and D. Turnbull (Academic Press,
New York, 1S80).
25) J.C SLATER and G.F. KOSTER, Phys. Rev. 94, 1498
(1954).
26) W.A. HARRISON, "Electronic Structure and Properties
of Solids" (W.H. Freeman and Co., San Francisco, 1980).
27) G. von MINMIGERODE and K. SAMWER, Physica (Utrecht)
108B-C, 1217 (1981).
28) W.L. MCMILLAN, Phys. Rev. J_62, 331 (1968).
29) F.P. MISSELL and J.E. KEEM, a ser publicado.
Este Trabalho foi subvencionado pela FAPESP, CNPq e FINEP.
201
Real Space Renormalization Techniques
for Disordered Systems.
Enrique V. Anda
Instituto de Física,
Universidade Federal Fluminense
CP. 296 Niterói 24210 R.J. Brasil*
Summary
Real Space renormalization techniques are applied to study
different disordered systems, with an emphasis on the under-
standing of the electronic properties of amorphous matter,
mainly semiconductors.
Work partially supported by Brazilian agencies CNPq and FINEP
202
Introduction
Real Space renormalization methods have permitted important
progress in the understanding of critical properties near a
phase transition. They were mainly applied to the study of
thermal properties of a great variety of different physical
systems. In the last years these scaling ideas have been applied
successfully to analyse dynamical properties, through the
calculation of the density of states of various elementary
Í1 2 3}
excitations present in condensed matterv * ' . These tech-
niques have been mainly used to study systems ^escribed by
linear chain tight binding hamiltonians, although they have
recently been extended to treat Bethe and Husimi cactus
lattices<4'5).
Its simplicity makes it suitable to study ill-condensed
matter properties (diagonal, non-diagonal, configurational and
topological disordered systems). It has been shown that the
method permits., the application of different approximated
schemes to treat the disorder which on one hand they are more
accurate that the traditional C.P.A. at least for lattices with
small coordination number Z, and in the other more efficient in
terms of computational time than the recursion method and
cluster numerical simulation.
The intention of this seminar is to analyse the main ideas
involved, the potentialities of this renormalization method
when applied to study dynamical properties of different systems,
with an emphasis on the understanding of the electronic proper-
ties of amorphous matter, mainly semiconductors.
I - One-dimentional systems
The tight binding linear chain is the simplest possible
203
system to which this method has been applied and its description
is particularly usefull to understand the ideas involved.
The local density of states of a linear chain hamiltonian
of identical atoas given by ' :
H - V E CfioCjo (I.I)(ij)
can be obtained decimating the local green function equation
GIo> (1.2)Goo -where
G00 '
A #» K • "
go
< «
+ go V (cio
o
This results in a renormalization of the locator go and the
propagator V at each step.
The imaginary part of the fixed point obtained for the
locator corresponds to the electronic density of states of the
linear chain.
The renormalization process defines a Poincari map for
the variable E ( N ) - » ( N )/V ( N ) given by
E(N) . (EC«-1V_ 2 (I.4)ÍN) fit)
where u and V are the renormalized frequency and hopping
integral respectively.and N labels the iteration.
This mapping processes a chaotic region for |u|<2
which corresponds to the support of the electronic density of
states and a non-chaotic tegion |u|>2 where there are no eigeti
states.
Defining a density of visits D(u) as the number of visits
to the interval (u,u+du>) after N-*» iterations of the map (I.A)
204
it is possible to show that D(u) coincides with the density
(3)of states of the linear chain ,
It is interesting to note that the map defined in (1.4)
is a particular case of a more general map, extensivelly
studied by Feingenbaun within the context of the theory of
chaos(6).
Although the density of states for the ordered linear chain
is a very veil known object, it constitutes an excellent exam-
ple to develop renormalization techniques, with the idea that
this scheme could in principle be extended to more complicated
situations, for instance, two or three dimensional systems.
The more interesting aspects of this formalism are obtained
when they are applied to disorder systems. For the case of the
linear chain there have been several proposals to include the
disorder, which has been in all cases treated approximate-*
ly •-» » .in order to apply the formalism, the renormalization
relations for the locator and the propagator have to be
configurationally independent, which in fact they are not. Ás a
consequence, an approximation has to be done. Some authors have
taken the configurational average over the contribution of t!-e
riecimatec? points to the self energy in the first iteration or
in the second . Better results are obtained averaging the
diagonal green function corresponding to the decimated point
at each stage* '.
Although these results are approximated they are more
accurate than thotse resultant from mean field approximations
like C.P.A. with the additional advantage in terms of numerical
calculation that they are not self consistent. This is a
Consequence of the fact that in low dimensional systems, Z be-
205
ing small, the configurational fluctuations neglected in mean
field theories are very important. The renormalization tech-
niques are capable of reproducing the main features of the
density of states (peaks and gaps) because, even if in each
iteration the information of the decimated points are averaged,
the disorder is maintained at each step» which implies that
fluctuations of all orders are included. Very recently a
method has been proposed which consists essentially in deci-
mating clusters which size is frequency dependent, so as to
permit the density of states for the disorder linear chain to
converge, numerically speaking, to the exact result .
The applicability of real space renormalization techniques
are not restricted to systems described by nearest-neighbour
tight binding hamiltonians. All physical quantities satisfying
an equation with the structure given in (1.2) are good candi-
dates to he studied by this formalism. This is the case of the
(8)Kronig and Penney modelv ' described by a potential given by
MVOO - g I fi(x-nai (1.5)
n«Q
If the functions
n An ex"+Bn e l l tJ t (1.6)
are taken to be the basis to represent the hamiltonian, the
following recursion relation is satisfied
where
' 2(cosk+eo Zjç^) (I.8a)
(2lE/h)1/2 (I.8b)
206
which is similar to equation (1.2) and can be decimated to ob-
tain the eigenfunctions and eigenvalues of the problem.
For the case of an ordered linear chain the renormalization
relation is similar to equation (1.4)
* - a2 - 2 (1.9)
n n-1
If we were to impose periodic boundary conditions to a chain
of M atoms, the last decimation before eliminating all points
of the chain corresponds to the number N * £uM/lu2 with aLi
final result for ax. - 2 which is the fixed point of theNL
mapping (1.9). This relation stablishes that the correct setof starting values for a (k(E)) has to be such as to satisfy
that after N iterations the a •+ a «2. The startinge ° NL
values fulfilling this condition determines the k's and the
eigenvalues of the finite chain. It is easy to realize that all
values of a that reach the fixed point a "2 after a number° NL
N<N_ óf iterations belong to the correct starting set. Theit
distribution of these values T-.'hich are contained within the
interval (-2,2) is equivalent to the density of visits obtained
2 -1/2for the case of the nearest neighbour linear chain D(a)"(4-a ) ,
As a consequence the density of states for a Kronig Penney model
can be written immediately as being
PKp(E) - /Í2<(a-2(cos k(E) • *Q
(1*10)
Generally speaking the existence of boundary conditions in
a finite linear chain determines which is the value for the
a., coefficient. To obtain the starting set a and from it theNL °
eigenvalues and eigenfunctions, it is useful to iterate the
inverse Poincarl mapping N. times
207
taking all possible combinations of + and - signa». This could
be an efficient procedure to study the behaviour of small
particles and its spectroscopic properties as a function of
II - The Bethe Lattice
the extention of the method described in the previous
sections for two and three dimensions presents' an essential
difficulty. For lattices of dimensionality greater than one the
decimation of the nearest-neighbours change the structure of
the equation of motion for the green function not allowing an
assignment between parameters so as to define the renormalization
relations. In the case of a square lattice, for instance, one
point in the lattice linked to its four nearest-neighbours is
after the first decimation linked to eight lattice points.
Although it is possible to define several approximated cell to
cell renormalization approaches the results obtained ar«
essentially incorrect because they are not able to reproduce
adequately the density of states of the ordered system. For the
case of ordered lattices a similar sxfeeme. to the one applied to
the linear ordered chain can be used* operating in the recipr£
cal space. For a square lattice it can be shown that the
density of states can be obtained, just by simply calculating
the density of visits of a superposed mapping of two summed
independent variables . However these ideas can not be
extended to the more interesting situation of a disordered
lattice for which it is not passible to define a reciprocal
s p a c e . • • • - • ' . • • • . : , • • • . . • - , - . •
Within the context of the decimation procedure applied to
208
linear chains, it can be defined a renormalization approach
(4 5)applicable to Bethe lattices * . Even if they are objects far
from any real lattice they can reproduce the local environment
correctly for systems of two or three dimensions. Topologically
speaking the linear chain and the Bether lattice are equivalent
(a linear chain is a Bethe lattice of Z » 2). The localized
density of states of an ordered Bethe lattice is identical to
the density of states at the surface of a semi-infinite linear
chain(10) defined by
where
ts - h V (II.2a)
t - /z-1 V (II.2b)
and V is the non diagonal element of the real Bethe lattice.
The semi-infinite linear chain can be decimated similarly to
the infinite linear chain. In this case two frequencies
u • g,t and u • gt have to be defined satisfying differ-
ent renormalization relations. In terms of these variables a
two dimensional mapping is determined by
-J - 4 ( N" 1 }" 2 (II.3a)
-J-- -g(""l) a»/*"1*" Z/U-l) (II.3b)
Although an extention of the result obtained for the linear
chain, would permit us to suppose thpt the density of visits
of the variable w , D(u ) , corresponds to the density of8 o
states of the Bethe lattice, this is not so. The region where
209
chaotic behaviour it found for the napping (II.3a) coincides
with the support of the density of states of the Bethe lattice
( |«J|</Z-1). However, this is not true for (II.3b).
A lore interesting approach for the decimation of the
Bethe lattice is based on a formalism which permits to treat
the disordered lattice (in principle any kind of disorder)
using the same ideas that were successful for the one dimension-
al case.
The formalism is based on a diagrammatic scheme which re-
constructs an arbitrary Bethe lattice from a conveniently.
defined linear chain, which is taken to be the basic diagram-
(4 5)matic element * . It can be seen that the Bethe lattice is
obtained as a limit of an iterative procedure. At each step of
this procedure all free vertices of an infinite linear chain
are linked to other semi-infinite chain of the same structure.
It can be shown that this renormalization process defines
for the case of a pure Bethe lattice a Poincaré mapping given
by
(II.4)
where »_ is the frequency starting value. The fixed point of
this renormalization process is
Z« • (Z-2)/Ü 2 -
* •
In fig.1 u* is represented as a function of u_. The well known
density of states of the Bethe lattice can be obtained from
l»g(- - 1J*). In fig. 1 the path by the iteration process is
1/2shown as dotted lines. In the region (<D_ |<2(Z-1) <•>* has an
imaginary part. As this renormalization scheme dresses the
210
vertices by linear chains at each step, u gains an imaginary
(N)part only when it satisfies the relation w <2, value from
which it converges to the complex fixed point.
This is very instructive to understand how is the process
of convergence for the more interesting case of a disorder
Bethe lattice.
Ill - Amorphous Semiconductors
An obvious extention of the method analysed in Section II
can be applied for the case of a disordered Husimi cactus ,
It is generally accepted that an important topological proper-
ty of amorphous tetrahedral bonded semiconductor lattice is the
existence, due to disorder, of odd as well as even rings, the
smallest of which involves five bonds, statistically
distributed*12*.
The impossibility of satisfying the antibonding electronic
state for a closed loop of odd number of atoms produces an
erotion of the states near the antibonding band edge. This is
(12 13)an important topological disorder effect, between others ' ,
because it changes the gap value of the semiconductor. It can
be qualitatively studied in a Husimi cactus, mixing with
arbitrary concentration rings of five and six number of bonds
as it is shown in fig. 2, with concentration x and (1-x)
respectively.
The electronic density of states for this system is shown
in fig. 3. It corresponds to a one orbital first-neighbour
tight binding hamiltonian. Two main features are worthwhile
mentioning. For intermediate concentrations of five and six-
fold rings, the density of states is smoother in comparison
with the result gained for a pure cactus. This is consistent
211
with photoemission measurements of amorphous silicon which
shows a curve without the characteristic peaks of the crystal-
(14)line structure . The result suggests that the Bethe lattice,
being a ringless lattice with a smooth density of states, could
be considered as an average, a sort of mean field approximation
for the ring statistics.
There is an erotion of the density states near the low
energy edge. The edge has not a perceptible dependence upon
concentration up to values of x< 0.9, above which, it changes
rapidly. This is related with the presence of a percolating
path of six-fold rings that are the only ones that contribute
to the lower energy tail, being xs0«9 t^ie percolation threshold.
In our case this band edge is not a Lifshiftz point due to the
infinite dímesionality of the Bethe lattice.
This calculation shows that the topological disorder relatec
with the ring statistics can be represented, in a first approxi-
mation by a Bethe lattice. Using this lattice our purpose now
is to study the effect of impurities typically Hydrogen in
Silicon, which introduces a completely different type of dis-
order into the system.
In order to study the real density of states of amorphous
Silicon we include four sp orbitais corresponding to the
valence states of the Silicon atom. A simplified version of a
tight binding hamiltonian which describes some of the main
features of a four orbital first neighbour tight binding
random network has been proposed by Wearie and Thorpe1 . They
have shown that the density of states of their simplified
hamiltonian can be mapped into a density of states of an S band
system with the same topological structure. However this
212
hamiltonian has some important short-comings. Two of the p
atomic orbitais do not hybridized to constitute a molecular
orbital, which in the density of states appears as two delta
functions at th? top of the valence and conduction band. This
problem can be overcome describing the system by an sp
Hamiltonian which includes all possible nearest neighbour
hopping integrals . Unfortunately taking these more general
Hamiltonian, there is no mapping of a four orbital density of
states into a one-orbital density of states. In order to be
able to apply the decimation technique used for the Bethe lattice,
we have to generalize the procedure for the case of four orbitais
per site. This can be done by simply rewriting the renormalized
equations in terms of 4 x 4 matrices
Many different theoretical approaches have been used to
study the electronic properties of hydrogenated amorphous
Silicon. They are essencially two different types of theories:
a) those which strictly speaking are only applicable to systems
with zero concentration of impurities as for instance the Cluster
Bethe lattice^18'19). The Hydrogen, sitted in a different
possible configuration is surrounded by an appropriate environ-
ment of pure material. This structure is treated exactly and
the rest of the lattice is represented by a Bethe lattice.
Similar procedures have been proposed using quantum chemistry
methods , These theories are unable to study the effects
of inter-impurity interference which certainly is present in
Si:H which can have up to 207. of Hydrogen; b) the well known
mean field approaches, typically C.P.A., which have the short-
coming that for systems with low coordination number (Silicon
has Z-4) do not include properly compositional fluctuations.
213
(2This method has been applied to hydrogenated crystalline Silicon
It has the difficulty that for lo« concentration of Hydrogen
the density of states reproduces the Van Hove singularities as
peaks, absent in an amorphous structure.
Defects of various Si-H bonding conformations have been
(22 23)suggested to explain some of the experimental results *
Unfortunately these results remains controversial. Ultraviolet
photo emission spectra for a Si:H reflects the appearence of
C 24)three characteristic peaks C, D, and Ev ' that are not present
in a:Si as it is shown in fig. 4. However this structure has
not been founded in recent measurements which shows other peaks
localized at different values.
We will not discuss which should be the most appropriate
hydrogen conformation inside an amorphous Silicon substrate.
Instead we will adopt the simplest one, within the context of
our formalism, in order to use it as an example to show all the
possibilities of the decimation procedure. We take a monohydride
substitutional four fold site impurity that has been studied
Í21 25)by different authors ' . This hydrogenated vacancy is such
that the four dangling bonds directed into a vacancy are
saturated by a Hydrogen atom. Due to its structure it can be
treated as a substitutional impurity by defining effective sp
orbitais adequately combining the four hydrogenic s orbitais.
(21)Taken the matrix element used by Papaconstantopulos et al ,
applied to a first-neighbour Bethe lattice, it is possible to
obtain the density of states shown in fig. 5 and 6 for different
concentrations of impurities. The results show the well known
gap enhancement produced by the presence of Hydrogen and the
existence of an antibonding peak very near the bottom of the
band. The split off peak at the bottom of the valence band»
corresponding to the bonding level, has been seen recently in
(25)photoemission experiments . A shoulder near the top of the
valence band, is transformed into.a peak with a clear anti-
resonance besides it which increases with hydrogen content,
showing the importance of the interference between the impurities
as concentration increases. No other intermediate peaks appear
in the valence band region.
From these results we conclude that a monohydride substitu-
tional four fold site impurity can not explain the peaks C, D
and E seen in photoemission experiments. As has already be
mentioned the existence of these peaks is controvertial and
probably depends upon the way in which the sample is prepared.
Conclusions
Real space renormalization techniques show to be a powerful
method to calculate the density of states of different elementary
excitations in solids. It is particularly interesting when
applied to disorder systems because it permits the formulation
of a simple approximated scheme to treat the disorder which
incorporates compositional fluctuations at all orders, or even
to obtain exact numerical results in a more efficient way in
terms of computational time than numerical simulation, Unfortunat
ly, up to the moment it has not been possible to extend the
method using a simple formalism to study two or three dimensional
lattices. The method is restricted to linear chains, Bethe or
Husimi cactus lattices which are not able to reproduce a real
system due to its topology. Caskets admit the application of
(24)decimation te .uniques and they have much more conectivity
215
that the Bethe lattice. Research oriented to the application
of these structures to describe amorphous semiconductors is
now in progres.
216
References
01. Gonçalves da Silva CET and Koiller K 1981 Sol. St. Comm.
40, 215-19.
02. Koiller B, Robbins MO, Davidovich MA and Gonçalves da Silva
CET 1983 Sol. St. Coram. 45, 955-9.
03. Oliveira PM, Continentino MA and Anda EV 1983, Phys. Rev. B
£9_, 2808, 1984.
04. Albuquerque J 1984 to be published.
05. Anda EV, Makler SS, Continentino MA and Oliveira PM to be
published in Journal of Phys. C 1984.
06. Feingenbaum. Los Alamos Science 1, 4 (1980).
07. Makler SS and Anda EV to be published in J. of Phys. C 1984.
08. Jose J 1982 Lecture Notes, XIX Latin American School of
Physics Cali, Colombia.
09. Monsivais M and Jose J to be published 1984.
10. Thorpe MF 1981 "Excitation in Disordered Systems" ed Thorpe
MF (Plenum Press NY) 85-107.
11. Balseiro C private coimnunication.
12. Yonesawa F and Cohen M "Fundamental Physics of Amorphous
Semiconductors ed Yonesawa F (Springer-Verlag, New York 1981)
13. "The Physics of Amorphous Solids" R Zallen 1983, John Wiley
and Sons, N.Y.
14. B yon Roedern, L Lexg, M Cordona and FW Smith 1979 Philos.
Mag. B 4_0, 433.
15. Wearie D and Thorpe MF 1971 Phys. Rev. B4 2508-20.
16. Singh J 1981 Phys. Rev. B 23, 4156-4168.
17. Anda EV and Makler JS 1984 Proceedings of 17 International
Conference on the Physics of Semiconductors. San Francisco
1984.
217
18. Joannopoulos JD and Tndurain F 1974, Phys. Rev. B JJ), 5164-73,
19. Tndurain F and Joannopoulos JD 1976 Phys. Rev. B l± 3569-3577.
20. R Barrio and J Tagflena-Martlnez in this Proceedings (Simpósio
Latino Americano de Física dos Sistemas Amorfos).
21. Papaconstantopoulos DA and Economu EN 1981 Phys. Rev. B 24,
7233-7246.
22. Ching VY, Laa DJ and Lin CC 1979 Phys. Rev. Lett. 42,805-808.
23. Allan DC and Joannopoulos JD 1980 Phys. Rev. Lett 44, 43-47.
24. Robertson J 1983 Phys. Rev. B 28, 4658-70.
25. Wesner D and Eberbart W 1983 Phys. Rev. B 28, 7087-93.
26. Domany E, Alexander S, Bensimon D and Kadanoff LP Phys. Rev.
B 28, 3110, 1984
218
ftt«L
I
/
i i i• i
?r
Renormaliration process for
of an ordered Bethe Lattice.
Fig
the
. 1
self
Wl
energy
Fig. 2
Husimi cactus Lattice formed by a random
distribuition of hexagons and pentagons.
219
>'\
Fig. 3
A/ *
! /•
0 +
V
^
Density of states for one orbital binary alloy
Husimi cactus composed by hexagons and pentagons
with concetration x and 1-x respectivelly.
Fig. 4
Ultraviolet photoemission spectra for a-Si(dashed line) and hidrogenated a-Si (solidline) ref.14 .
220
Fig. 5
tf
.10 -C O € «O
Density of states of pure Si.
•.I
Fig. 6
Aft-I* «10
Total density of states for the alloy
at Ch= 0,1 .
221
RESEARCH ON HYDROGENATED AMORPHOUS SILICON
AND RELATED ALLOYS FOR PHOTOVOLTAIC APPLICATIONS
Ivan E. CharobouleyronPhotovoltaic Conversion LaboratoryInstitute of Physics, P.O.Box 6165UNICAMP-Campinas, SP.,13100 Brasil
INTRODUCTION
In this paper we present an overview of the main
activities on amorphous materials of photovoltaic interest
performed at our laboratory. The research group started in
1980 and was supported from the beginning by the Universi-
dade Estadual de Campinas (UNUCAMP), the Companhia Energé-
tica de São Paulo (CESP) and the Financiadora de Estudos e
Projetos (FINEP). The funding level varied from year to
year. A rough estimate gives 100.000 US$ per year approxi-
mately. Since 1980 the group has grown to its actual size-
Researchers, post graduate studentes and technicians were
incorporated during the past years. The minimum infrastructure
necessary to work in amorphous semiconductors was locally
built or imported and, taking advantage of the good existing
facilities at UNICAMP, many research activities were under-
taken in collaboration with other groups. In order to over-
come the difficulties associated to the relative isolation
we experience in the field we established a scientific
exchange program with some foreign laboratories and
specialists. We benefited from visits to these laboratories
and we also received some outstanding experts that stay
between us for periods of varying length. '"
222
THE PHOTOVOLTAIC CONVERSION LABORATORY TODAY
The main research activities of our group concern
amorphous semiconductors, other activities currently
developed relate to poiycrystalline silicon solar cells and
M-I-S structures on single crystal and poly silicon.
PERSONNEL: Prof.: Ivan Chambouieyron (Head of the
Group), Fernando Alvarez, Rene Brenzikofer, Jorge I. Cisne-
ros and Sergio Koehlecke; six post graduate students, two
of them working for their Ph.D. and the remaining four for
their M.Sc. degree and three technicians.
FACILITIES
Two capacitively coupled glow discharge reactors were
locally built. The first one is a stainless steel, 30 cm
diamter, reactor having a parallel field configuration (1).
The second one is a pyrex, 20 cm diameter, reactor possessing
a cross field configuration (2). Each reactor has recently
been equiped with two mechanical pumps and a diffusion pump.
They operate in a way similar to other reactors (mainly
Japanese). One rotary + diffusion pump being used to reach
a good background vacuum and devised to pump only air. The
second rotary pump work when si lane or dopant gases are
present. This system prevents oxygen and silane mixtures
to occur that are harmuful to the oil of the rotary pumps.
Both reactors allow to deposit a-S1:H films in 2Mx2"
areas with good uniformity. More recently an RF sputtering
and electron beam (Leybold Heraus Z-400) system was installed.
It 1s now used for studies on transparent electrodes,
223
Metallization, or growing a-Si:H and other compound amorphous
semiconductor layers. Studies on the properties of amorphous
and polycrystalline SnO2 have been made during the whole
period, the material being deposited by the chemical spray
method. Finally some facilities exist to measure electrical
and optoelectronic properties of the films, to deposit
metallic contacts, to anneal the samples, etc., etc.
RESEARCH ACTIVITIES
a-Si:H
a-Si:H films were grown to characterize the reactors
behavior. The measurements made to determine the properties
and quality of the material include,among others:
•Growth rate as a function of si lane pressure, gas flux
and RF power. Silane is normally helium diluted in a
proportion varying from 25 to 75%. Few studies on film
properties with varying substrate temperature were under»
taken. We chose deposition temperatures in the neighborhood
of 300OC. Our working conditions give growth rates of ca.
lX/sec.•Transport properties. Electrical properties of the
samples are determined by measuring dark conductivities as
a function of temperature in the temperature range 300-500K
and under vacuum conditions. Activation energies for undoped
and doped samples are deduced from °à vs 1/T plots. Photo-
conductivity of amorphous samples is also measured under
vacuum as a function of both photon flux and temperature
(90 - 350K). They provide useful information concerning the
224
nature and density of recombination and trap levels. We are
presently making a system able to mesure thermopower on our
amorphous films.
Doped samples are currently made by adding controlled
amounts of phosphine or diborane to the main si lane flux.
Till now we used mechanical flowmeters. Electronically
controlled mass flowmeters were bought and will be installed
during the present year.
•Optical properties. We determine the optical prpperties
of our films by transmission measurements in a visible-near
IR spectrophotometer (3). Infrared studies are made on
samples deposited onto high purity polycrystalline silicon
wafers* giving useful information concerning the amount of
hydrogen included in the material.
•Devices. Some devices of Schottky type and p-i-n solar
cells were fabricated in 1982. Figure 1 shows the structure
of those devices that possessed low conversion efficiency
Corning glass (7059)
T1 (2500 Jt)
P* (800 %) n* (300 %)
V»O.(75O X)
J-Sl.M(1)(6000Ü)
225
(~1 - 2%). We are presently undertaking a program aiming to
produce p-i-n solar cells having more than 4% conversion
efficiency in areas larger than 1 cm .
a-SiNx_:H
In order to reduce the optical losses produced by the
top layer of p-i-n solar cells we undertook a systematic
study of the properties of off-stoichioraetric silicon
nitride compounds. Samples are prepared by the simultaneous
plasma decomposition of SiH4/N gaseous mixtures. The optical
gap of glow discharge a-SiNY:H samples prepared in that way
increases with r.f. power densities and depends on silane
IN
ü
c
O)
o
it ti "
PHOTON ENERGY
>
FIO. 2 (a * * ) 1 " vs photon energy for different a-Si, N , . , :H layers obtainedfrom the O D of a SiH« and N , mixture of fixed compoaition (Sir!*/(N,) - 0.33. AD sample» are nonintentkmally doped and i f power density isindicated on each curve.
FIO. 3 (a) Room-temperature conductivity vs gaseous doping mixture forsamples grown under conditions identical of those of curve C. Fig. Z Theeffect of optical gap shrinking due to boron doping is alto indicated, (b) Darkconductivity activation energies vs gaseous doping mixture. It is found ex-perimentally that the conductivity varies with temperature according toa « e9 cxp(-£,/kT) in the 300-JOO-K temperature range.
-t
-6
-1
•to
-12
t
1.0
0J0.80.70.10.5
1
p — 1 - *•> |• • '
i
- I'I
"*Air*
* 1 « 1i
- (I
T• I•j
1,1
*
•
•
-
-
i i |
Ib)1 1
- 23
• U
?o
O)1.1 UJ
- 1.7
0 -t .4 .8 - « - » • * 0i ii t
"i rPM,
log gas. imp. ratio
226
to nitrogen gaseous ratio. At low r.f. power densities the
plasma will essentially break silane molecules, the
dissociation energy of N2 molecules being 2 to 3 times
larger than that of the silane molecule. The material does
not differ much from a-Si:H. Higher power densities produce
more nitrogen radicals in the plasma increasing the number
of nitrogen atoms incorporated into the network. Figure 2
shows the effect of increasing r.f. power on the optical
gap of samples produced by a SiH^/N2 gaseous mixture of
fixed composition (4).
Doping experiments were done in samples grown at fixed
silane dilution and r.f. power. The results for phosphorus
and boron doping on samples having an optical gap of nearly
2eV, are shown in figure 3 were room temperature conductivities,
optical gaps and activation energies are plotted as a(4)function of dopant gas concentration in the reactor chamber .
The nitrogen concentration of those layers was AES determined
at SERI, Colorado (USA), and estimated to be 33Í 2 at. %.
The photoconductivity of such samples was studied as a
function of photon flux and under ELH - 100 mW/cmz
illumination ' '. For undoped and lightly boron doped
samples a supralinear dependence of the photocurrent on
light intensity was found within the illumination range
- 1 0 - 10 photons cm s . Light boron doping
unsensitizes the material while heavy doping sensitizes
it again. The opposite behavior is found with phosphorus
doping. The supralinear behavior is interpreted assuming
two types of defect centers having energies that are
227
located above and below the dark quasi Fermi level and that
possess a large difference in their electron capture cross
section.
Figure 4 shows, for a set of samples, the photocurrent
as a function of photon flux for monochromatic light
excitation of -2.2eV. Figure 5 shows dark and photoconduc-
tivities as a function
if.
Fig:4. *8i»x:H •«•ple«. Photocurrenta« • function of photon flux for mono-ehroaatic light excitation of c«. 2.2 eV.
of r.f. power density
for undoped a-SiN :H
samples. Infrared
studies were performed
on samples grown onto
silicon substrate (6).
From them it is possible
to determine» not only
the hydrogen content
in the material, but
the nitrogen content
as well. Independent
AES measurements are
used for calibration
purposes.
SnOg
In solar cells, and opto-electronic devices in general,
1t is necessary to minimize the losses coming from the
optical reflection at the semiconductor air interface. To
that aim transparent layers, having an appropriate thickness
228
b°
and refractive index, are deposited. If the transparent
coating is conductive its use has a twofold purpose: to
minimize optical losses and to allow a thinner top electrode
in p-i-n structures. Tin oxide films deposited by the
chemical spray
method (hydrolysis
of SnCl4) possess
both properties and
we studied the
electrical behavior
of such films as a
function of deposition
temperature and doping.
Low deposition (Td£
280<>C) temperatures
produce amorphous
\3
-19
POWER (WcrrT)
5- Filled squares: roo» temperature dark conductivity
•s • function of RF power density for undoped
staples. Optn squirts: roo» teapereture photo-
conductivity on ELH 111u»1n«t1on (100 «H/c»2).0pen
trUnglcs: ratio of roo» te»p»raturt photoconductivity to
dark conductivity «s • function of RF power density.
films while polycrys^
talline layers are
obtained when the
reaction takes place
at higher temperature^),
In the case of non-intentionally doped films, large
conductivity variations are obtained with deposition
temperatures. They come from oxygen vacancies and chlorine
inclusion into the network liberated during the incomplete
hydrolisis reaction* '. Our work on tin oxide films refers
to conductivity studies^9). Hall efect(10), composition*8)
and the amorphous-polycrystalline transition (See figures
229
'eu
a:
CH
LO
RIN
E C
ON
CE
N1
-
500 400 300 25C)
I I I I
Aboof «tol /
• ,
Kone etoi. /^-
* /
r i gOH >
• • • •
9 ™ #
1 I 1
—
. -
1
- 1 0 s
"Io
- . O 2
1.2 1.4 1.6 1.8 2.0
Fig. 6 Chlorine concenirttion vs. inverse deposition temperature for samples grown at four differenttemperatures. On the same figure measurements by Aboaf ei al. , Carlson and Kane el al. areindicaicd. Relative O H and H contem (arbitrary uniu) are also shown for the same samples.
6 and 7. It is worth to mention that the temperature of the
process for tin oxide film deposition is compatible with
amorphous silicon solar cell technology and that the spray
method has the advantages of easy of operation, reprodudbility,
low cost and no-scaling-up area problems. The properties of
fluorine doped tin oxide layers are presently studied and
the main results will soon be published.
FUTURE WORK
We are presently starting a series of projects concern-
ing amorphous semiconductors and devices:
230
SnO
200 300 400 500
SUBSTRATE TEMPERATURE I°C) 200 300 4 0 0 500
FIG. 7. Variations of room temperatureconductivity, electron Hall mobility, andcarrier concentration as a function of sub-«rate temperature (solution flunes: T 1 m l /min.Q'iml/minJ.
a) Amorphous silicon alloys a-SixGe. :H and a-SixSn-t :H
deposited by RF sputtering techniques;
b) a-SnOg and a-ITO deposited by electron beam techniques;
c) High efficiency, medium area, a-Si:H solar cells
(p-i-n structures obtained by glow discharge);
d) Schottky type and SIS (a-Si :H/SnO2) cells;
e) New characterization techniques including Raman
spectroscopy,photoluminescence and thermopower.
CONCLUSION
The Laboratory of Photovoltaic Conversion at UNICAMP is
actively engaged in a research program concerning amorphous
semiconductors and devices. We presently have the necessary
231
infrastructure and personnel and we hope to contribute in
the near future to this interesting and expanding field
of research.
REFERENCES
1. F. Alvarez, Rev. Bras. FTs., 12, 4, 832 (1982).
2. R. Brenzikofer, unpublished.
3. J.I. Cisneros, E.L. Carpi, F. Alvarez and I. Chambo£
leyron, to be published.
4. F. Alvarez, I. Chambouleyron, C. Constantino and J.
I. Cisneros, Appl. Phys. Letters, 44, 1, 116 (1984).
5. F. Alvarez and I. Chambouleyron, Solar Energy Materials
(in press).
6. I. Chambouleyron, F. Alvarez, C. Constantino and J.I.
Cisneros, Proc. 5th E.C. Photovoltaic Solar Energy
Conference, Athens Greece, 1983 (in press).
7. I. Chambouleyron, C. Constantino, D. Jousse, R. Assumg
ção and R. Brenzikofer, J. Phys. (Paris), £2, C-4,
1009 (1981).
8. I. Chambouleyron, C. Constantino, H. Fantini and M.
Farias, Solar Energy Materials, 9, 2, 127 (1983).
9. 0. Jousse, C. Constantino and I. Chambouleyron, J.
Appl. Phys., £4, 1, 431 (1983).
10. D. Jousse, J. Non-Cryst. Solids, 59 & 60, 637 (1983).
Ill
A GENERALIZED SLATER-PAULING CONSTRUCTION FOR MAGNETIC AMOR-
PHOUS ALLOYS
A. P. Malozemoff, A. R. Williams and V. L. Momzzi
IBM T. J. Watson Research Center
Yorktown Heights, NY 10S98
ABSTRACT:
The concept of magnetic valence and a generalized Slater-Pauling construction permit a
consistent interpretation of tbe ferromagnetism of both transition metal and metalloid alloys
with Fe, Co and Ni. Amorphous alloys like CoB, CoSi, FeB and FeSi are revealed to be
strong ferromagnets with a constant number of majority-spin sp electrons over a certain
composition range.
233
We review our recent progress in developing a coherent picture of the zero-temperature
magnetism of transition-metal alloys.1*7 The essential ideas are extremely simple and apply to
both crystalline and amorphous alloys.
We focus on an AtxBx alloy whose atom-averaged moment in Bohr magnetons is the
difference of the atom-averaged number of spin-up and spin-down electrons:
= ^ - J V * (1)
We define the average valence as the atom-averaged number of electrons outside the last filled
shell:
Za*e - ZA(\ - x) + ZBx = JV* + JV* (2)
Now fr is in most cases less precisely known than N' for reasons to be discussed below; so
eliminating it in Eqs. 1 and 2 we find Friedel's8 simple but profound result
Have = M f ' - Z ^ (3)
Now we can break up Ar into its sp and d components. In two limiting cases, Ny is
known quite precisely. First, in what we call the "common-band" limit of strong ferromagne-
tism, the d-bands of the host (A) and solute (B) merge, and, for the majority-spin electrons,
these "commmon" bands lie entirely below the Fermi level, so that AfJ is precisely five. An
example is Ni^_xCux. Second, in what we call the "split-band" limit of strong ferromagnetism,
the d-bands of the alloy subdivide or "split" into two groups, one with wave functions
concentrated on the host atoms and one with wave functions concentrated on the solute
atoms. The meaning of "strong" for these split-band magnets is that the position of the Fermi
level lies in the gap separating the two sets of band states. An example of such a split-band
system is shown in Fig. 1, where we see that the d states of YCo^ segregate into two groups
separated by a region of very low state density ("gap") of approximately 2 eV. The hybridi-
234
zation of host and solute states results in the almost perfect maintenance of local neutrality.
Nevertheless, as long as the gap between the subbands remains well-defined, quantum
mechanics dictates that the number of states in each subband is conserved. In the case of our
example, YCo^ shown in Fig. 1, the atom-averaged number of up-spin d electrons N* is
therefore simply the number associated with Co, namely five times the concentration factor
1-x with x * 0.25. To summarize, in spite of hybridization and local neutrality, the number of
up-spin d electrons per atom is, to a very good approximation, simply five or zero.
The common- and split-band limiting cases are actually quite often realized in Fe, Co and
Ni alloys, and can usually be predicted from well known band-structure systematics. For
example, the upper edge of the d-bands is known to shift systematically downward in energy
as one proceeds either to the right along a row of the periodic table or up along a column.6
Thus> for early transition-metal alloys1 such as Y, Zr or Ti with Fe, Co or Ni, one can
anticipate the split-band limit (e.g. N'dY » 0, N'dFe « 5), while for late transition-metal
alloys5 involving Cu, Zn or Pd solutes, for example, one can anticipate the common-band limit
(e.g. N\CU = 5). Metalloid alloys,4 involving B, Al and Si, for example, correspond to the
split-band limit, because their d-states lie very high in energy (e.g. N'dB * 0). The number
of up-spin d electrons is an integer multiple of five (and therefore easily predicted) only when
the Fermi level lies above the up-spin common bands or in the gap of the split bands. This is
usually, but not always, the case. It is hot the case, for example, in Fe-rich alloys, where the
intra-atomic exchange interaction is not quite strong enough to fully occupy the Fe-dominated
up-spin d bands. Even for Fe alloys, however, when the solute concentration is sufficient,
intra-atomic exchange again wins over inter-àtomic hopping, and strong magnetism (fully
occupied Fe-derived up-spin d bands) and simply predicted magnetization return. (The origin
of this behavior is discussed in Sec. II of Ref. 4.)- Thus, even in an amorphous alloy with
different degrees of hybridization in different local environments, the simple d-state sum rule
holds, as long as the local density of states has a clear-cut gap or top and the lower-lying d
bands are fully occupied.
235
Now it becomes advantageous to write Eq. 3 in the form
I) (4)
The atom-averaged moment is seen to be a sum of two terms, one being two times the number
of up-spin sp electrons and the other being an atom average over the quantity
2mi - 2tfJ,.-Z, (5)
for atom i . We call this quantity the "magnetic valence".2 Since, as shown above, N'djis
usually an integer, Zmi is usually an integer depending on the valence of each constituent. For
example, it is 2, 1 and 0 for Fe, Co and Ni respectively, reflecting their increasing valence and
their common ability to bind five up-spin d electrons. It is -1 and -2 for Cu and Zn respec-
tively, -3 for Y, B and Al, -4 for Zr, Ti, Si and Ge, etc. These negative numbers reflect a
valence that exceeds the ability to bind d electrons. Negative values for the magnetic valence
measure the tendency of these solutes to reduce the magnetic moment. In certain cases, like
Rh for example, with an electronic valence of 9, the d-band position relative to Ni or Co is
neither clearly in the split-band nor in the common-band limits. Thus, one might expect the
moment to lie somewhere between that predicted from ZmRh = 1 and ZmHh = - 9, as turns
out to be the case.5
What is the physical significance of the magnetic valence? It is just the number of holes
in a low-lying d-band which are left after filling it with the electronic valence. Fe, Co and Ni
give non-negative values because they have no more electrons than d-states. But atoms with
no low-lying d-bands give a negative value because the excess of electrons will lead to a
reduction in the number of holes of the host d-band. It should be noted that this effect does
not imply change transfer from one atom to another, because hybridization compensates for
any apparent charge transfer.
236
The utility of the magnetic-valence concept arises from the consistent treatment of all
solutes, whether transition metal or metalloid. By contrast, the traditional Slater-Pauling
construction9'10 which plots average moment versus average electron-to-atom ratio, cannot
conveniently incorporate the metalloid alloys. Using the magnetic valence concept, we can
generalize the Slater-Pauling construction to all alloys2-4 by simply plotting average moment
vs. average magnetic valence as in Fig. 2 for crystalline TVi,_xCux and NiixSilxl (we have
corrected the moment for a shift in the g-factor away from 2). According to Eq. 4, if 2N]p is
constant, this "generalized Slater-Pauling construction" should yield a line of slope 1 intersect-
ing the moment axis at 2JV .
The crucial question for this approach is whether or not 2N'sp should be constant. For
alloys consisting exclusively of transition metals, this is a plausible assumption. It was
originally explained in terms of the "rigid-band model".910 but it is clearly more general, since
N\p is an integrated quantity not depending necessarily on the shape of the band. Data like
that shown in Fig. 2a on Nii_xCux demonstrate the validity of this assumption. For alloys
containing metalloids, however, it is much less obvious why N\p should be constant as a
function of composition. For while 2N\p is approximately 0.6 (or more precisely O.SS) for
most purely transition-metal alloys like Nix_xCux, one might expect 2N\p to be 3 for trivalent
metalloids, 4 for a tetravalent metalloids, etc., simply because of charge neutrality. And yet
the data of Fig. 2b on Nix_xSix, for example, show that, while not precisely constant, the
deviation in 2N\p at x « 0.1 {Zmavt = -0 .4) is only 0.05 rather than the 0.34 ( « (0.55 x
0.9-l-4 x 0.1)-0.55) one might have expected.
This mystery was solved by Terakura and Kanamori'2 who identified the physical
mechanism that holds 2N\p constant in nickel-metalloid alloys to be a d-sp hybridization
effect. This "Fano anti-resonance" interaction causes a strong depression in the sp-state
density at energies near the top of the d band. This depression acts like a gap, preventing
metalloid states from passing through the Fermi energy and thus from increasing 2N]p. We
237
want to emphasize that the surprising constancy of .v]_ is not inconsistent with local neutrality.
For example, the metalloid Si in magnetic alloys does have four (strongly hybridized) s and p
electrons, and is approximately neutral. The implication of the Fano-gap-induced constancy of
N]p is that local neutrality is achieved in such systems by the polarization of existing states,
(e.g. of sp states on the neighbors) not by the occupation of new ones.
The constancy of JV*, means that magnetic valence continues to be a useful concept in
alloys containing metalloids, and that the magnetization of such alloys can be understood
without a detailed knowledge of the atomic geometry. Because of the importance of the
split-band and the Fano gaps, we have called a theory based on the constancy of 2N]p a
"bandgap theory" for strong ferromagnetism.4
While Terakura and Kanamori accounted successfully for dilute Ni-alloys on the basis of
single impurity calculations, it remained unclear for many years whether the same mechanism
would apply to more concentrated alloys. Indeed, most amorphous alloys were stable only at
large solute concentrations, typically around 20% metalloid.
The discovery that the Terakura-Kanamori effect can persist to higher concentrations
came from our band calculations on 3:1 ordered compounds like that shown in Fig. 3, which
revealed a characteristic minimum in the sp-state-density near the Fermi energy (shown by the
arrow). The appearance of this minimum in both bec and fee calculations suggested an
independence of structure, which might extend to amorphous materials. The presence of such
a minimum is a requisite for holding 2N\p constant, as we have seen above. A second
discovery was that the value of 2N\p deduced from the calculated moments of spin-polarized
calculations for 3:1 structures varied from 0.8 for bec structures to more than 1 for fee
structures.4 This suggested that 2N\p might be different from the value of ~0.6 found for
most transition metal-transition metal alloys.
238
Why have we done ordered-compound calculations rather than using more realistic
amorphous structures?1316 The simplification achieved by treating the periodic replication of a
single local atomic arrangement instead of a random array of atomic arrangements allows us to
perform calculations in which the physical quantities of greatest importance are treated
accurately. In the present context these are the atomic volume and charge. Because our
compound calculations determine both the volume and the self-consistent charge distribution
that minimize the calculated total energy, these effects tend to be accurately described. It
would be enormously more difficult to treat these effects with comparable accuracy for a
random system. Furthermore, as we have seen, crucial to the understanding of alloy magneti-
zation is the constancy of the number of s-p electrons. The compound geometry therefore
plays an important role in not restricting us to models, such as tight-binding theory, in which
the s-p electrons are ignored. Finally, the analysis of photoemission data for several amor-
phous alloys17-1* suggests that the close-packed fcc-like compound geometry used in our
calculations models the amorphous alloy sufficiently well to describe the chemical bonding
responsible for the principal variations of the state density within the d band.
Nevertheless, until both bonding and structural aspects can be treated with equal com-
pleteness, we must turn to experiment as the final arbiter. By making generalized Slater-
Pauling plots, we can identify where 2N]p holds constant and where by implication, the
Terakura-Kanamori "bandgap" mechanism applies. We have reviewed existing data1'5 and
find a large number of systems following Eq. 4 with constant 2/vj. over at least some range of
concentration. A good example is amorphous Cof_xBx shown in Fig. 4. Amorphous CoSi,
CoSn, CoP, FeB and FeSi also show such regions. However, some alloys do not;4 these
include amorphous FeC, FeSn, FeGe and FeP. It will be interesting to trace down the reasons
why 2N]p is constant in certain cases and not in others. Moment insensitivity to metallurgical
conditions can be expected in regions where 2N\p is constant, but not where it varies as a
function of composition. The existence of a minimum in the sp-density also has implications
239
for conduction and structural stability which remain to be explored. These are topics for
further work.
The authors acknowledge the key input and earlier collaboration of K. Terakura in this
work.
240
References
1. A. P. Malozemoff, A. R. Williams, K. Terakura, V. L. Moruzzi and K. Fukamichi, J.
Magn. and Magnetic Materials 35, 192 (1983).
2. A. R. Williams, V. L. Moruzzi, A. P. Malozemoff and K. Terakura, IEEE Trans.
Magn. 19, 1983 (1983).
3. V. L. Moruzzi, A. R. Williams, A. P. Malozemoff and R. J. Gambino, Phys. Rev.
B.28, 5511 (1983).
4. A. P. Malozemoff, A. R. Williams and V. L. Moruzzi, Phys. Rev. B 29, 1620 (1984).
5. A. R. Williams, A. P. Malozemoff, V. L. Moruzzi and M. Matsui, J. Appl. Phys. 55,
2353 (1984).
6. V. L. Moruzzi, J. F. Janak and A. R. Williams, Calculated Electronic Properties of
Metals (Pergamon Press, New York, 1978).
7. A. R. Williams, J. Kübler and C. D. Gelatt Jr., Phys. Rev. B 19, 6094 (1983).
8. J. Friedel, Nuovo Cimento, Suppl. to Vol. Ill, 287 (1958); also in Metallic Solid
Solutions, ed. J. Friedel and A. Guinier (W. A. Benjamin, Inc., New York, 1963).
9. J. C. Slater, Phys. Rev. 49, 537 (1936).
10. L. Pauling, Phys. Rev. 54, 899 (1938).
11. J. Crangle and M. J. C. Martin, Phil. Mag. 4, 1006 (1959).
12. K. Terakura and J. Kanaraori, Prog. Theor. Phys. 46, 1007 (1971).
13. R. P. Messmer, Phys. Rev. B23, 1616 (1981).
241
14. T. Fujiwara, J. Phys. K 12. 661 (19X2).
15. S. S. Jaswa! and W. Y. Ching, Phys. Rev. B26, 1064 (1982).
16. S. Frota-Pessoa, Phys. Rev. B28, 3753 (1983).
17. P. Oelhafen, V. L. Moru/zi, A. R. Williams, D. S. Yee, J. J. Cuomo, U. Gubler, G.
Indlekofer and H.-J. GUntherodt, Sol. St. Comm. 44, 1551 (19H2).
18. V. L. Moruzzi, P Oelhafen, A R. Williams, R. Lapka, H.-J. Güntherodt and J.
Kübler, Phys. Rev. B 27, 2049 (1983).
19. H. Watanabe, H. Morita, and H. Yamauchi, IEEE Trans. Magn. 14, 944 (1978).
20. R. Hasegawa and R. Ray, J. Appl. Phys. 50, 1586 (1979).
21. T. R. McGuire, J. A. Aboaf and E. Klokholm, IEEE Trans. Magn. 16, 905 (1980).
These data lie slightly lower than those of other workers (Refs. 17 and 18). We
tentatively attribute this difference to systematic error in determining sample thickness.
22. M. C. Cadeville and E. J. Daniel, J. Phys. (Paris) 27, 449 (1966).
-5 0 5
ENERGY (IN eV RELATIVE TO Ef )
10
Fig.l - Parameter-free, self-consistent, spin-polarized
energy-band calculation of majority and minori-
ty-spin state densities for Co.Y in the Cu^Au
structure. This is an example of a split-band
strong magnet with a well-defined gap separa-
ting the lower energy Co band from the higher
energy Y band. The calculated moment of 0.8 y_
average over all atoms of the compound agrees
well with the value for amorphous Co_Y, which
also has an fee-like nearest neighbor environment.
243
-1.0 -0.8 -0.6 -0.4 -0.2 0
(a) ATOM-AVERAGED MAGNETIC VALENCE
- l . o -o.e -o.e -0.4 -0.2 0ATOM-AVERAGED MAGNETIC VALENCE
Fig.2 - Generalized Slater-Pauling construction (average
magnetic moment vs average magnetic valence) for
data of Crangle and Martin for crystalline a)
NiCu and b) NiSi alloys. The moment has been co-
rrected for the shift in g-factor from 2 to 2.2,
which is assumed to be constant as a function of
composition. Along the 45° slope, 2N is cons-
tant as a function of concentration. The Si alloy
approximates this condition in spite of the large
difference in the number of sp electrons for pure
Ni (0.55) and pure Si (4) .
to
i<ena.«oI/)
b O 5
tNtHÜY I IN EV RELATIVE TO EF )
10
Fig..3 - Calculated paramagnetic (non-spin-polarized) sp
and d state-densities on the cobalt site of Co_Si4 3
in the Cu-Au fee structure. Note the difference
in scales for the sp and d densities. The arrow
indicates the gap discussed in the text.
IT
>3 2.5
ÊS 2.0O
sc 1 S
Ctt^ 1.0
0.5SO
v fee xtal, Hasegawa and Hay+ cpd, Cadeville and Danielo amorph. Hasegawa and Kayo amorph, UcGuire et al.A amorph, Hatanabr et al.
- 2 - 1 0 1
ATOM-AVERAGED MAGNETIC VALENCE
Fig.4 - Generalized Slater-Pauling plot for CoB alloys
with data from references 19-22. The 45° line
shows that 2N is constant in the amorphous alloy.
245
AMORPHOUS MAGNETISM
H. R. Rechenberg
Instituto de FTsica - USP
1. INTRODUCTION
In these notes we will briefly examine the consequences of dis-
order on the magnetic properties of solids. In this context, the word
"disorder" is not synonimous to structural amorphicity; chemical dis-
order can be achieved e. g. by randomly diffusing magnetic atoms on a
nonmagnetic crystalline lattice. The name Amorphous Magnetism must be
taken in a broad sense.
The immediate consequence of structural and/or chemical disorder
is a great variety of atomic environments, which in turn has three
effects: •
-a distribution of atomic moments 14;;
-a distribution ef exchange interactions J..;
-a distribution of local anisotropy energies and easy axes.
Many types of magnetic behavior result from the Interplay of
these factors. For the convenience of discussion, we shall classify
disordered materials in three major categories, according to the con-
centration x of magnetic atoms:
a) 10 £ x < 1: ferromagnetic or noncollinear spin structures;
b) 10~ ;£ x £ 10 : cluster magnetism (mictomagnetism);
c) x;£ 10 : spin glasses.
From the viewpoint of chemical composition and preparation meth-
od, several broad families of materials have been investigated to date:
246
a) Amorphous transition meta)-metalloid alloys (metallic glasses).
The basic formula is T«.M..t where T stands for a combination of 3d
metals (Fe, Co, Ni, Cr ...) and M is a combination of metalloids (B,
P, Si, C . . . ) , of course including the binary case. They are most often
prepared by the melt-spinning technique, and many are technologically
important.
b) Amorphous rare earth-transition metal alloys, mostly binaries;
prepared by vapor deposition, sputtering, melt spinning, etc.
c) Glasses in the strict sense, containing transition metals or
rare earths as relatively concentrated impurities; most often prepared
by cooling from the melt.
d) Crystalline metallic alloys or insulating compounds, in which
disorder is introduced by substituting a magnetic element by a non-mag-
netic one or vice versa.
e) Amorphous oxides, halides, sul fides, etc. have been less stu-
died because of preparation difficulties. Some interesting materials,
such as ferric hydroxide gel, can be found in nature.
2.FERR0MAGNETISM
As mentioned above, disorder introduces a distribution P(j) of
exchange interactions. If this distribution is relatively narrow ano
concentrated on positive J values, spins will align parallel to each
other at low temperatures. Ferromagnetic order without a periodic lat-
tice presents no conceptual difficulties, although the possibility of
its existence had not been proposed until I960 (by Gubanov). As we
shall see, its properties differ only marginally from those of crys-
talline ferromagnets.
Representative examples are: most 3d-meta!loid metallic glasses;
247
many 3d-W5d alloys; some 3d-4f alloys. As in the crystalline case,
insulators are rarely ferromagnetic. One striking exception is FeF«,
which is an antiferromagnet in crystal form and a ferromagnet in the
amorphous state. This is a good example of the sensitivity of superex-
change parameters (including their sign) to metal-ligand-metal bond
angles.
In amorphous GdCo- and GdFe_, the Gd moments are aligned anti-
parallel to the transition metal moments, which are ferromagneticaily
coupled among themselves. In such cases i t is appropriate to speak of
ferr{magnetism.
Some characteristic features of amorphous ferromagnetIsm are
listed below.
a) There is a well-defined Curie temperature, with "normal"
critical behavior ( f> - 0.32 for Fei,oNll»oPii,B6' °'1'1 f o r
Fe32Ni36Cpi4P12B6 • " ) -
b) Reduced (M/M vs. T/T ) magnetization curves are flattened
with respect to Briliouin or to crystalline Fe or Ni curves. This can
be ascribed to the P(J) distribution, whose width can be roughly esti -
mated from experimental curves.
c) Mttssbauer spectra have broad lines, reflecting the distribution
of atomic moments at low temperatures, otherwise also the P(J). In the
case of Fe, a quadrupole splitting is often observed at T > T but
not in the six-line spectra below T ; this is explained by the vanish-
ing, on the average, of the factor (3 cos S - 1) which appears in the
expression for hyperfine levels when magnetic and electric quadrupole
interactions are simultaneously present. Since 6 is the angle between
the local electrostatic z axis and the magnetization direction, the
fact that {cos 0 > « 1/3 confirms that the z axis varies ran-
248
domly from site to site, as is expected in an amorphous material.
d) Detailed analysis of the magnetization vs. temperature in me-
tallic glasses, in the range 0 ^ T £o.3 T , reveal a T^ behavior,
which is evidence for the existence of spin waves. However, the norma1-
ized coefficient of the 1 term turns out to be l» to 5 times larger
than In crystalline Fe or Ni. This would imply a significant reduction
of the spin wave stiffness coefficient, whose origin is still obscure.
An alternative interpretation has been proposed whose central Idea
is that amorphous magnets can be well described by a Heisenberg model
of short-range exchange interactions between localized spins. As is
well known, such s model is highly inadequate for dealing with crys-
talline metallic ferromagnets.
The last mentioned feature is perhaps the most fundamental differ-
ence between amorphous and crystalline ferromagnets. Of course, there
are also the important differences in permeability, losses etc., which
render metallic glasses so attractive for many applications, but these
will not be discussed here.
3.NONCOLLINEAR SPIN STRUCTURES
Let us now examine the case where exchange interactions are pre-
dominantly negative, i.e., tending to favor antiparallei spin arrange*
ments. In the simplest situation, we can imagine a crystal lattice to
be subdivided Into two interpenetrating sublattices A and B, such that
all nearest neighbors of an A spin are on the B sub lattice and vice
versa. The ground-state spin configuration is then the classical anti-
ferromagnetic one, in which the spins are collinear but point in oppo-
site directions, in a 1:1 proportion.
Such a subdivision is not always possible. In an FCC lattice,
249
nearest neighbors of a given site may be nearest neighbors of each
other; thus there will be three spins on the vertices on an equilateral
triangle, and it is obviously impossible to have three pairs of anti-
para U el spins. Energy minimization necessarily requires a noncollinear
arrangement»
This ts the simplest example of what is nowadays known as "frus-
tration". This concept acquires its deepest significance in the spin
glass problem, to be discussed later on.
Frustration is certainly favored when the exchange interactions
can be randomly positive or negative. This occurs especially in the
case of magnetic impurities diluted in a nonmagnetic metal, since then
the exchange interactions are of the oscillatory RKKY type:
cos (2k_r..)
J (r ) - A E-4J 3
4r ) 3
where k_ is the Fermi wavevector of the electron gas. Computer simula-
tions have shown that P(J) is then essentially symmetric around J = 0.
Another source of directional disorder is the single-site, "crys-
tal-field" anisotropy, which is particularly important for rare earths.
With this term included (in its simplest, uniaxial form), the magnetic
hamiltonian is written as
D.S7 - 2. J(r..)S.«S.
where not only the strength D. but also the z direction of the easy
axis varies from site to site.
When anisotropy is the dominant factor (|D| > /Jj), we expect
that each spin gets locked along its own easy axis, with a +z or -z
orientation determined by exchange interaction with its neighbors.
250
The resulting structure is then, ideally, that named "speroroagnetic"
by Coey: spins freezed in random directions, without long-range corre-
lations. This type of magnetic ordering can also occur with negligible
an I sotropy, as a result of competition between exchange interactions
of both signs (case of metallic spin glasses) or even in the case of
a single, negative J value, as a result of frustration associated with
topological disorder (examples are amorphous FeF. and Fe(OH).). In
this case It can be said that speromagnetism is the amorphous version
of antiferromagnetism, which obviously cannot exist in the absence of
a crystalline lattice.
Other, more complicated noncollinear arrangements are possible.
Fuller discussions can be found in Refs. (2, 3).
4.CLUSTER MAGNETISM
We will now discuss disordered materials in which the concentra-
tion of magnetic atoms is below the percolation limit. This means that
ferromagnetic, speromagnetic etc. order can still exist below a certain
temperature, but the spatial extent of this order will now be limited
to small regions within the sample. Such ordered domains or clusters
behave essentially as independent entities. As a consequence of this
fine subdivision, completely new features will show up in the magnetic
behavior, both in bulk and in microscopic (e. g., MBssbauer) experi-
ments. These are closely related to superparamagnetism, a class of
phenomena known long before the appearence of amorphous magnetism. Work
in this area was initiated by Nêel in 1949. For a recent review on su-
perparamagnetism, see Ref. (4).
Consider an assembly of ferromagnetic particles so small that the
usual subdivision into domains will be energetically unfavorable. For
251
iron, this means a critical diameter of about 15 nm. Each particle will
2 3then have a resultant magnetic moment of 10 -.10 Bohr magnetons. In
an external field, this moment will rotate as a whole, since individual
spins are kept parallel to each other by exchange coupling.
Even if the intraparticle spin coupling is of antiferromagnetic
nature, a net moment will result, since sublattice cancellation or spe-
romagnetic randomness, whatever the case, can never be perfect in a
small particle.
Each particle will usually have an anisotropy energy whose origin
may be magnetocrystal line, dipolar, shape, etc. An energy barrier E
thus has to be overcome if the net spin is to flip between different
easy directions. Due to small ness of the particles, however, E can be
very small, not greatly exceeding kT; spontaneous thermal fluctuations
will then occur, with a relaxation time X that is a very steep function
of temperature.
The outcome of any oper intent will then be governed by the compa-
rison between T and r , the intrinsic time constant of that experi-
ment. (For static magnetization measurements, T 2f1 - 10 sec; for
mAC susceptibility, Xm « f t / 1 £T io"5 - 10"2 sec; for 57Fe Mtfssbauer,
r í IO"' sec; for neutron scattering, t CIO" 1 1 sec.)m ^ m
The temperature at which C« C >s called the blocking tempera-
ture Tg. I t must be emphasized that a T_ is defined for a given experi-
mental technique and for a given particle size ( i . e . , energy E).
I f T y T_, or X 4C T » magnetization or susceptibility experi-
ments wil l exhibit the equilibrium behavior (Langevin, Curie-Weiss . . )
of an assembly of giant spins. In Hbssbauer spectra, on the other hand,
magnetic hyperfine structure wil l appear washed out by too rapid fluc-
tuations. This is the so-called "superparamagnetic regime".
252
If T < L or T » t (the "blocked regime"), a) the magnetiza-B m
tion will respond very slowly to a sudden change of the external field,
giving rise to phenomena like thermoremanence, etc.; b) spins will be
unable to follow an oscillating magnetic field, thus exhibiting an es-
sentially zero AC susceptibility; c) Müssbauer spectra will show normal
hyperfine structure.
The most conspicuous sign of a blocking temperature is a peak in
the AC susceptibility, separating the two regimes. This may sometimes
be confused with an antiferromagnetic transition. In that case, it
suffices to repeat ths measurements at a different frequency, thereby
changing the measurement time; a shift of the peak temperature is an
unequivocal proof of its relaxational nature.
Turning back to amorphous magnetism, superparamagnetic behavior
is frequently observed in a great variety of disordered materials.
Whenever chemical clustering of magnetic atoms occurs, small-particle
effects such as those described above will be present. This happens,
more often than not, in crystalline transition metal-noble metal alloys,
even at high dilution and despite great metallurgical care.
Chemical clustering, however, is not a necessary condition for the
formation of magnetic clusters. As the temperature is lowered, random-
ly positioned spins may progressively build themselves into locally
correlated regions which can then rotate as a whole. This process will
be assisted by the concentration fluctuations that are always present
on purely statistical grounds. Upon further lowering of temperature,
clusters will grow either by incorporation of new spins or by coalescen-
ce of adjacent groups. As a result, the average cluster moment will be
temperature dependent.
Many transition-metal containing glasses are good examples of this
253
behavior. Alumino-silicate glasses ( MO.Al.O-.SiO., M « Mn or Co ) have
been extensively investigated during the last decade . Measurements
of heat capacity, sound velocity, Mbssbauer spectroscopy, muon depolari-
zation rates, frequency and field dependence of AC susceptibility, mag-
netization decay have yielded results consistent with the superpara-
magnetic interpretation. More recently, measurements of AC susceptibili-
ty in small DC fields have provided evidence for a strong temperature
dependence of the average cluster moment . As to the kind of order
within clusters, it is presumably speromagnetic, since exchange inter-
actions in these glasses are strongly negative.
Strictly speaking, Neel's theory of superparamagnetism should
not be applied, for quantitative purposes, to systems other than physi-
cally well-defined and noninteracting magnetic particles. Neglect of
interactions is sometimes a major shortcoming. On the other hand, the
simple Arrhenius1 law used to relate the relaxation time to particle vo-
lume and temperature has proven inadequate in some cases. Nevertheless,
the physical picture provided by Neel's theory is often the only sim-
ple way to rationalize experimental results on amorphous sysitems.
5.SPIN GLASSES
(For a good recent review, see Ref. (7).)
The spin glass problem is a fashionable and fascinating research
topic. This statement raises two questions: What is a spin glass? Why
is there a problem?
In 1972, Cannella and MydoSh made the first observation of a
sharp, cusp-like maximum In the At susceptibility of some AuFe alloys
at low concentrations. Such an anomaly strongly suggested a phase tran-
sition. Conventional types of magnetic ordering could, however, be
254
ruled out because of the low concentration of magnetic impurities (the
critical concentration for ferromagnetism in Au-Fe is \6% Fe). The
"spin-glass" transition was then assumed to be a cooperative freezing
of spins in random directions, analogous to speromagnetism, occurring
at a well-defined freezing temperature T,.
For seme time it was believed that the essential ingredient for
a spin glass was the RKKY interaction, which is capable of coupling iso-
lated spins separated by rather long distances. Various RKKY, or "ca-
nonical", spin glasses have been investigated, including ^uMn, AgMn,
AuCr, AuMn, etc. The same mechanism is responsible for the coupling be-
tween rare-earth moments in La. Gd Al, and similar alloys. However,
spin glass behavior has also been observed in concentrated amorphous
alloys such as Gd.-Al,-, Ni-qP.-Bg, FeqQZr10, etc., where exchange in-
teractions are short-range (cf. Section 2) and spin-spin coordination
is high. Finally, there is the striking case of the insulating com -
pounds Eu. Sr S, which show cluster magnetism for x - 0.13» ferro -
magnetism for x ,> 0.50 and spin-glass behavior in the intermediate
range. For this material, it is by now rather clear that the mechanism
responsible for spin-glass behavior is the competition between posi-
tive (nearest-neighbor) and negative (second-neighbor) exchange inter-
actions.
That the spin-glass transition is one of a rather special nature
became clear when it was realized that the heat capacity has no anoma-
ly at all; the magnetic contribution to it shows, at best, a rounded
maximum well above T-. Magnetic entropy at T, usually has less than
half the total value Nk In (23 + 1), implying considerable short-range
magnetic order above T,.
Another universal feature of spin glasses is the onset of time
2S5
effects (magnetic viscosity, thermoremanence ...) just below T,. In
other words, the ordered state seems to be in metastable equilibrium.
Much theoretical effort has been made during the last decade in
order to clarify the spin glass phenomenon. The question of whether a
a phase transition really occurs has not been settled yet. One essen-
tial point is that the time effects observed below T, are hard to re-
concile with the idea of the ordered phase as a state of thermodynami-
cal equilibrium.
It is not surprising that an alternative school of thought has
emerged, which discards the very idea of a phase transition at T,, in
favor of an interpretation based on Niel's theory of superparamagne-
tism. Indeed, the susceptibility peak, time effects, the absence of a
heat capacity anomaly and so on fit naturally into the simple model of
a progressive blocking of small clusters of correlated spins upon tem-
perature lowering.
The existence of magnetic clusters above T, can indeed be inferred
from neutron scattering, heat capacity and electrical resistivity mea-
surements. The crucial question now is: can the increase of the relaxa-
tion time of these clusters alone explain the observed susceptibility
cusp? It is often argued that rather artificial energy barrier distri-
butions would be needed to produce a sharp cusp. On the other hand,
the actual sharpness of a susceptibility maximum is often hard to as-
sess experimentally, due to the rounding-off effect of a finite mag-
netic field.
The frequency dependence test has not been conclusive either.
Some alloys show a distinct peak shift, but at least for AgMn none
was observed over five decades of measuring frequency. The range of
experimental time constants can be considerably expanded by the inclu-
256
sion of neutron scattering experiments, but their interpretation may
be rather subtle.
Metastabi1ity effects below T f are probably a manifestation of
the high degeneracy of the ground state. It is quite conceivable that
the configuration into which the spins actually freeze at low tempera-
tures is not unique; it surely does correspond to an energy minimum,
but other configurations are equally possible, which are also energy
minima, separated from the chosen one by energy barriers. These are
then responsible for the slow response to a change in external para-
meters, as the magnetic field. The ground state degeneracy has a close
connection with the frustration concept». In Hurd's words , "a frus-
trated system is one which, not being able to achieve a state that sa-
tisfies entirely its microscopic constraints, possesses a multiplicity
of equally unsatisfied states".
The picture that will eventually emerge for the spin glass phe-
nomenon must be more complex than any of the models we have been dis-
cussing, but it will almost certainly include elements of every one.
(8)As an example, it has been suggested that simultaneous freezing
and growth of magnetic clusters may result, at T « T,, in the appear-
ance of an infinite cluster, such a "cluster percolation" being a
true cooperative effect or phase transition. The infinite cluster, in
turn, would still contain weak links due to frustration, so that ther-
mally activated processes would occur, much in the spirit of Neel's
model.
257
REFERENCES
(1) S. N. KAUL: Phys. Rev. B 27, 5761 (1983)
(2) J. M. D.COEY: J. Appl. Phys. 49, 1646 (1978)
(3) C. M. HURD: Contemp. Phys. 23, 469 (1982)
(4) J. L. DORMANN: Rev. Phys, Appllquêe J6, 275 (1981)
(5) H. R. RECHENBERG, A. H. OE GRAAF: J. Phys. C J£, L397 (1980)
(6) E. NUNES FILHO, H. R. RECHENBERG: to be published
(7) P. J. FORD: Con temp. Phys. 23., 141 (1982)
(8) J. A. HYDOSH: J. Hagn. Magn. Mater. Ji-Ji, 99 (1980)
258
Reentrant Ferromagnetism
M. A. Continentino
Instituto de Física
Universidade Federal Fluminense
Caixa Postal 296 - Niterói - RJ - Brasil
We briefly review the problem of reentrant ferromagnetism
with particualr emphasis on the controversy between coexistence
or reentrant behaviour. We discuss a cluster and a glassy
model which are relevant for the understanding of the proper-
ties of these materials. We find that for a ferromagnetic in-
stability to occur it is essential the existence of non-col*
linearity or an anisotropic coupling between spins.
259
Recently much interest has been devoted to the phenomenon
of reentrant ferromagnetism, that is of systems which exhibit
a ferromagnetic to spin glass transition on cooling . In spite
however of experimental evidence, provided- for example through
a scaling analysis of the magnetization as a function of tempe£
ature in small fields , there is still a controversy if all,
or any of, the systems included under this designation do real-
ly have a ferromagnetic instability below a given temperature .
This controversy can be expressed as a duality involving co-
existence versus reentrant behaviour.
From the theoretical side we can get support for both
point pf views.
In fact Gabay 'and Toulouse have studied a vector model of
a spin glass and found a new mixed phase where long range
ferromagnetic order of the longitudinal component of the
magnetization coexists with spin gla^s ordering of the trans-
verse components. On the other hand a random field type of
approach to these systems suggests taat the ferromagnetic
phase should become unstable as the transverse components
freeze with a spin glass ordering .
Cluster Model
Many of the systems which exhibit reentrant type of behaviour-
are dilute ferromagnets with a concentration of magnetic atoms
a little above the percolation threshold . These systems can be
imagined as consisting of an infinite ferromagnetic island,
where spin waves propagate, coexisting with finite clusters
of magnetic atoms randomly distributed in the non-magnetic .
medium ' . For simplicity we shall consider just the two
260
lowest energy configurations of the magnetic clusters and
couple them to,the spin waves of the infinite ferromagnetic
island. The Hamiltonian for this system is
H « I e. a^.a. + E JafJ<yV (1)
k k k k a,6
where e. • Dk is the energy of the spin waves of the in-
finite clusters. The Fauli matrices oa are associated with thea
finite clusters and the S are the components of a spin in
the ferromagnetic agglomerate. We considered an anisotropic
coupling between the infinite and finite clusters for general-
ity. If we look for the effect on the spin waves of this
coupling we find that these excitations have their energy
renormalized and are damped. The renormalized energy E. and
dampi ng T. are given by
E k - efc -
8SN/-T 2^T 2 f x, . it, .«I 8SN _ 2_ f z, > zjj—(J
1 +J2 )Re[x (o)-x (w)J+ s — J2 Re[x (o)-xo o
_ _ 8SN ( . 2^ T 2 . _ x , W 8 S N .T^ • j — CJj- +J 2 )Imx Cu)+]j— ^ 1
o o
where u • e^/h, M and NQ are the number of finite clusters
and spin in the infinite «lusters respectively. xZ(<») and
X*(w) are the longitudinal and transverse dynamic
susceptibilities of the finite clusters and are given in
rsf. 7. We took for simplicity JoB - Jj 6aB and Ja0 - J2 for
a i 0. The theory presented above is in fact a random field
type of theory where we took into account the dynamic nature
of these fields. They are due to the finite clusters and in
concentrated ferrotnagnets with frustration they arise from
261
the "frustration islands" which are agglomerates of spins loosel
connected to the ferromagnetic medium due to frustration effect
In the random field model Ja8 - 0, (J°°)2 - J.2 and
(J ) • J2 for o 4 8. In this case the energy splittings of
the' finite clusters are zero that is the configurations are
degenerate. For this situation the static shift is zero and one
gets for the energy and damping of the spin waves
2SN (J2*2J 2) U)2T2 •l zk
k N 2 2O kT (i) T + 1
- . 2SN <Jl 2^J 22> ,t
O tií T TI
where T - to exp(V/kT) with T>- the inverse of an attempt
frequency and V an activation energy. From the expressions
above we can obtain the lineshift and linewidth measured in a
ferromagnetic resonance experiment on reentrant ferromegnets,
although for detailed comparison they should be averaged over
th.e distribution of activation energies. We point out that
there is no ferromagnetic instability in this random field
model which is in fact consistent with coexistence between
ferromagnetism and freezing of finite clusters. The spin wave
energy shift is always positive and the magnons become stiffer
due to their interaction with the finite clusters. When these
clusters freeze, that is for <DT>>1, appears a gap in the spin
wave spectrum. This gap exists even if the coupling between
the finite and infinite clusters is isotropic (Ja0-O, a + 6;
J°° - J,V a) and may be considered as an isotropic "anisotropj
field acting on the infinite cluster. This result does not
contradict Goldstone's theorem since the w - 0 mode is not
shifted.
262
Glassy Model
The glassy model ic appropriate to describe ferromagnets
with frustration which have a large number of nearly degener-
ate equilibrium configurations. These configurations are local
minima of the classical oagnetic energy and are specified by a
set of angles describing the orientations of the spins. The
metastable ferromagnetic states are accessible to one another
through quantum mechanical tunne tunnellling or thermal
activation and most probably the transition between them
involves the rearrangement i-. the directions of a small number
of spins. The Hamiltcnian describing the glassy ferromagnet
is8
H » t e.aife, + ôcrz + La*: + Kozs"
The first term describe;, the spin waves which are the ex-
citations on the local ninima and furthermore we assume they
have the sane dispersici relation in ell metastable configur-
ations, e represents the energy difference between metastable
states lying on eppogita sides of a double well potential and
A allows for quantum mechanical tunnelling between these
states. The lest term in equation above is the lowest order
term on spin wave operators coupling these excitations to
tunnelling modes. It a.-ises if ve consider an anisotropic
interaction between r.pins nr non-colline&rity between them .
The effect of raetastibiliiy on spin vave propagation is
again to renomalí-c anJ dar? these excitations. In this case
we can obtain a ^errorjssuetic instability due to the dynamic
transverse randca field arising fro.?, the fourth term on the left
hand side of the ec.uatioa above. This transverse field
263
increases as the temperature is lowered due tt> an increase in
the susceptibility of the tunnelling modes. Eventually it
erodes the spin wave gap, due to the dipolar interactions,
and gives origin to a soft spin wave node which causes the
8instability of the ferromagnetic phase .
Conclusions
We have presented two models which we believe are
relevant for the understanding of reentrant ferromagnetism.
Ve found that both situations of coexistence and reentrance
can occur in magnetic materials which are generally called
reentrant ferromagnets. It is essential for the appearance
of a ferromagnetic instability, the existence of an aniso-
tropic coupling between spins.
264
References
1. S.M. Shapiro, C.R. Fincher, Jr., A.C. Palumbo and R.D. Parks,
J. Appl. Phys. _52_, 1729 (1981); J.W. Lynn, R.W. Erwin,
J.J. Rhyne and H.S. Chen ibid. 52_, 1738 (1981).
2. T. Yeshorom, M.B. Salamon, K.V. Rao and H.S. Chen, Phys.
Rev. B24, 1536 (1981).
3. A.P. Morani, Phys. Rev. B28, 432 (1983).
4. M. Gabay and G. Toulouse, Phys. Rev. Lett. 4_7, 201 (1981).
5. K. Motoya, S.M. Shapiro and T. Muraoka, Phys. Rev. B28,
6183 (1983).
6. B.R. Coles, B.V.B. Sarkissian and R.H. Taylor, Phil. Mag.
B37, 489 (1978).
7. M.A. Continentino, J. Phys. C: Solid State Physics 16,
L71 (1983).
8. M.A. Continentino, Phys. Rev. B27, 4351 (1983).