+ All Categories
Home > Documents > c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S.,...

c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S.,...

Date post: 23-Aug-2020
Category:
Upload: others
View: 3 times
Download: 0 times
Share this document with a friend
171
c Copyright by Shashank Misra, 2004
Transcript
Page 1: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

c© Copyright by Shashank Misra, 2004

Page 2: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

SCANNING TUNNELING MICROSCOPY OF THE

CUPRATE SUPERCONDUCTOR BSCCO

BY

SHASHANK MISRA

B.S., University of Wisconsin - Madison, 1998

DISSERTATION

Submitted in partial fulfillment of the requirements

for the degree of Doctor of Philosophy in Physics

in the Graduate College of the

University of Illinois at Urbana-Champaign, 2004

Urbana, Illinois

Page 3: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

SCANNING TUNNELING MICROSCOPY OF THECUPRATE SUPERCONDUCTOR BSCCO

Shashank Misra, Ph.D.Department of Physics

University of Illinois at Urbana-Champaign, 2004Ali Yazdani, Advisor

The cuprate superconductors remain an enigma after nearly 20 years of research. Al-

though most of their properties in the superconducting state can be attributed to d-wave

superconducting order with a short coherence length, very little else about them is under-

stood. In particular, the nature of the electronic state outside the superconducting phase

remains controversial. Whether the anamolous ’normal’ state properties of the cuprate super-

conductors are the result of some remnant of superconductivity or some competing electronic

order, and whether these properties influence superconductivity, remain perhaps the most

important unanswered questions in the field. Because the superconducting phase and many

of the proposed normal phases have variations in their electronic structure on atomic length

scales, a local probe with atomic spatial resolution could be very useful in understanding the

electronic properties of the cuprates.

In this thesis, we will exploit the ability of Scanning Tunneling Microscopy (STM) to

measure the density of states on an atomic length scale to study both the superconduct-

ing and normal state of the high temperature superconductor Bi2Sr2CaCu2O8+δ. In the

superconducting state, we have found a novel signature of d-wave superconductivity- cer-

tain extended defects nucleate a one dimensional zero-energy bound state. The existence of

the bound state can be used as a characterization tool for other defect structures. In the

pseudogap regime, we find an incommensurate, fixed wavelength modulation in the spatial

structure of the density of states that correlates with the pseudogap energy scale. Unlike

similar structures found in the superconducting regime, which are believed to result from

iii

Page 4: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

scattering interference, the modulations in the pseudogap regime appear to be some form

of local ordering. Finally, we present the first conclusive identification and density of states

measurements on a single CuO2 plane at the surface of a high temperature superconductor.

Although the strongly correlated electron behaviour of the cuprates is thought to originate

on the CuO2 planes common to these materials, most spectroscopic measurements have

been made on samples terminated with different crystal planes until now. This experiment

demonstrates a different way with which we can access the physics of a doped CuO2 system.

In total, the experiments in this thesis are meant to demonstrate how many of the compelling

mysteries of a highly correlated material are uniquely accessable with an atomic scale probe.

iv

Page 5: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

To my fiancee, Julie Shuler.

v

Page 6: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Acknowledgments

It is my pleasure to acknowledge invaluable conversations with Ali Yazdani, Dan Hornbaker,

Michael Vershinin, and Philip Phillips. My group is indebted to the research groups of

Jim Eckstein and Yoichi Ando for providing wonderful high quality samples to work with.

Finally, this work was made possible by support from my parents, Jennifer Tate, Akilan

Palanisami, Richard Hasty, Jonathan Kaufman, Martin Holt, Tim Kidd, and Herve Aubin.

I would also like to acknowledge the following agencies for funding the research presented

in this thesis: NSF through grants DMR-98-75565 and DMR-03-1529632, US-DOE through

grant DEFG-02-91ER4539 via the Frederick Seitz Materials Research Laboratory, and the

ONR through grant N000140110071. All of the research was performed in the Frederick Seitz

Materials Research Laboratory at the University of Illinois. Some basic debugging work was

done in the Center for Microanalysis of Materials, which is partially funded by the US-DOE

under grant DEFG02-91-ER45439. Finally, I would like to thank the University of Illinois

for keeping me fed and clothed my first two years of graduate school through the University

Fellowship and the GAANN Fellowship.

vi

Page 7: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Table of Contents

Chapter Page

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Why study the high temperature superconductors with scanning tunneling

microscopy? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2 Scanning Tunneling Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.1 Theory of Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Scattering in the Superconducting State . . . . . . . . . . . . . . . . . . . . . . . 21

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 The Andreev Bound State . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3.3 Local DOS Modulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4 Local Ordering in the Pseudogap Regime . . . . . . . . . . . . . . . . . . . . . . . 50

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

4.1.1 Precursor Superconductivity Experiments . . . . . . . . . . . . . . . 55

vii

Page 8: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

4.1.2 Spin Order Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 58

4.1.3 Theoretical Ideas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

4.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

4.3 Scattering Interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4.4 Local Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86

5 The CuO2 Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

5.3 Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110

6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

Appendix

A STM Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

A.1 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

A.2 Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

A.3 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

A.4 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

B Elastic Scattering Interference Calculations . . . . . . . . . . . . . . . . . . . . . . 145

B.1 Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

B.2 Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

B.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

Curriculum Vitae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

viii

Page 9: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 1

Introduction

1.1 Why study the high temperature superconductors

with scanning tunneling microscopy?

Some of the most exciting discoveries in condensed matter physics in the last fifty years

have centered on how electrons behave when subjected to strong interactions. In most

materials, the electrons involved in conduction delocalize into a set of plane waves with

a well-defined momentum (wavevector) and energy, and behave like free electrons with a

renormalized mass. Interactions are relatively weak, and can be treated as perturbations to

the free-electron behaviour. However, the ideas we use to understand conventional materials

fail spectacularly in describing materials with strong interactions, which lead to electron-

electron correlations that are responsible for the basic behaviour of the system.

One storied example of a strongly correlated system is superconductivity. [1] The so-

called BCS (Bardeen-Cooper-Schrieffer) superconductors are normal metals for temperatures

above the superconducting transition. However, when these materials are cooled, phonons

mediate an attractive potential between electrons of opposite spin and momenta near the

Fermi energy, and they pair into bosons. This pairing manifests itself as an energy gap in the

1

Page 10: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A CB

Figure 1.1: (A) The High Temperature Superconductors (HTS) are layeredmetal oxides. This schematic drawing shows the positions of the metal ions inBi2Sr2CaCu2O8+δ. (B) Shown here is a schematic diagram of the CuO2 plane,showing the Cu sites (red) and O sites (blue). The doping level of the HTSrefers to the concentration of holes (δ) on each plaquette (green region) in thisplane. (C) The phase diagram of the HTS has two well-established regions:the antiferromagnetic insulator (below TN ≈ 320K and for hole dopings δ ≤0.02), and the superconducting state (below TC ≈ 100K and between δ =.06 − .26). Although its boundaries aren’t well-established, the enigmaticpseudogap regime exists below optimal doping (δ ≈ 0.16), where TC reachesits maximum, and for T ∗ > T > TC .

density of states; simply, you have to pay the energy cost of breaking a so-called Cooper pair

before you can remove an electron from the system. Simultaneously with their formation, the

Cooper pairs condense into a single macroscopic quantum state. This quantum state flows

without dissipation, leading to zero resistance, the first experimental hallmark of supercon-

ductivity. Moreover, the state has a macroscopic phase, with the Cooper pairs remaining

coherent over long length scales. A non-trivial consequence of this is the expulsion of exter-

nally applied magnetic fields, the second major experimental hallmark of superconductivity.

The theoretical explanation of superconductivity, BCS theory, is one of the landmark devel-

opments in the field of condensed matter physics: not only did it completely describe the

effect, it demonstrated that the solution to strongly correlated problems lies in determining

the novel states that correlations can force electrons into. [2]

The discovery of High Temperature Superconductors (HTS) understandably came as a

2

Page 11: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

shock, as their superconducting transition temperatures are significantly higher than pre-

dicted by the BCS theory. [3] However, the HTS are more than simply high transition

temperature analogues of their low temperature counterparts. [4] Instead of being normal

three dimensional metals, the parent compounds of the HTS are layered metal oxides, with

superconductivity occuring in dopable CuO2 planes common to these materials (Fig. 1.1a).

In a phenomenological sense, we understand the strongly correlated phases at the extremes

of the doping-temperature phase diagram fairly well- the undoped insulator appears to be

a Heisenberg antiferromagnet [5], and the heavily overdoped superconductor appears to be

a quasi-two-dimensional metallic BCS superconductor (Fig. 1.1b). [4] However, stemming

from the poor understand of the microscopic details of either these phases, we have a poor

understanding of the strongly correlated behavior in between the two extremes. [6] The HTS

have become the poster child for the study of strongly correlated problems, as many of the

issues in understanding the middle of the cuprate phase diagram are general.

In their undoped state, the traditional theory of metals would predict that the CuO2

plane would have one half-filled band of states at the Fermi energy. However, there is a

strong Coulomb repulsion between these electrons (one electron per unit cell), and they

localize on the Cu sites, forming an insulator. The system lowers the total energy of this

configuration by forcing neighboring sites to have anti-aligned spins. Thus the undoped,

parent state of the CuO2 plane itself has strong correlations which lead to long-range anti-

ferromagnetic order. Adding holes (or removing electrons) from the CuO2 plane suppresses

this ordering much more quickly than would be expected from the simple dilution of the spin

lattice (Fig. 1.2a). [7] The phenomenological reason for this is relatively simple- the addition

of a very small number of holes (.02 per unit cell) introduces itinerant states at the Fermi

level. [8] While ordering in the undoped compound is well-understood theoretically, we have

a poor understanding of the relationship between short-range (10 − 100 A) in-plane anti-

ferromagnetic correlations, which survive the suppression of global antiferromagnetic order,

3

Page 12: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A BA

⟨π,π⟩

⟨0,π⟩

Figure 1.2: (A) This is a schematic diagram of the CuO2 plane at half-filling (adoping of δ = 0). The removal of a single spin on this 5×5 grid, correspondingto a doping of δ = .04, is enough to destroy antiferromagnetic order in theHTS. (B) A schematic diagram of the amplitude (radial distance) and phase(sign) of the d-wave order parameter of the HTS as a function of angulardirection. Here, 〈0, π〉 refers to the Cu-O bond direction of the CuO2 plane,and 〈π, π〉 is 45 rotated from the bond direction.

and these electronic states. [5] This problem is generic to the study of strongly correlated

systems- it remains unclear what happens when electron kinetic energy and correlation ener-

gies are of similar magnitude. What is clear is that these electronic states strongly interact

with some mode of the system, and will superconduct at sufficiently low temperatures with

the addition of only a few more holes.

Superconductivity sets in at sufficiently low temperatures upon the addition of between

.06 and .26 holes per CuO2 unit cell. This superconducting order is understood to be

BCS-like with two significant modifications of the BCS theory. The superconducting wave

function has a phase that changes sign upon π/2 rotations, with a corresponding node in

the superconducting gap between oppositely signed lobes of the order parameter (Fig. 1.2b).

[9] The coherence length for superconducting correlations is also many orders of magnitude

smaller than in BCS superconductors– just 16A in-plane and 3A out-of-plane. [10] A modified

BCS theory that includes these details has proven remarkably successful at explaining most

experimental observations in the superconducting state. [4] However, key issues, like the

reason why the superconducting phase forms a dome in the phase diagram, remain open

4

Page 13: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

questions. For overdoped samples, the energy gap in the density of states is seen to shrink,

which should result in falling TC values. However, for underdoped samples, the value of the

energy gap is increasing while TC is falling, suggesting that another energy scale is involved

in the formation of the superconducting state. The appearance of two energy scales that

separately influence the ground state of a system is, too, a recurrent theme in the study of

strongly correlated electron systems.

The importance of this second energy scale is magnified by the curious properties of

the state above the underdoped half of the superconducting dome- the so-called pseudogap

regime. [6] The energy gap in the density of states evolves smoothly through the super-

conducting transition, with the gap filling in with states upon reaching a temperature T ∗.

Because the pseudogap is symmetric about the Fermi energy, it intuitively appears to result

from correlation effects, which might determine the electronic phase of the system. One

prominent proposal posits that fluctuating superconducting correlations survive the global

phase transition. The pseudogap is then a signature of this fluctuating superconductivity,

where global phase coherence has been lost, but Cooper pairs still exist. The other energy

scale is then associated with phase coherence of the Cooper pairs. The shape of the super-

conducting dome results from this energy scale shrinking in the underdoped regime, and the

pairing energy scale shrinking in the overdoped regime. The idea that a mysterious energy

scale might be controlling global phase coherence is not as far-fetched as it sounds. In purely

two dimensional systems, off-diagonal long-range order should set in only at T = 0. However,

most real systems are quasi- two dimensional, and off-diagonal long-range order sets in below

a temperature related to the anisotropic nature of the interactions in the sample, like the

energy scale for out-of-plane interactions. Meanwhile, vestiges of local order can survive up

to much higher temperatures, associated with the energy scale for in-plane interactions. Such

effects are common in planar correlated systems, and are present in the antiferromagnetic

insulating regime of the HTS. [7]

5

Page 14: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Another set of proposals posits that the pseudogap results from some exotic order, such

as charge and/or spin order. In the HTS, given the complex interplay of charge and spin

degrees of freedom and the proximity to the antiferromagnetic insulator, the most obvious

form of ordering would be one derived from residual antiferromagnetic correlations. Any

number of ordered states have been proposed for the pseudogap regime, some competing

with superconducting order, and others either facilitating or remaining indifferent to it. [4]

Most of these proposals, however, involve modulations of either spin or charge degrees of

freedom on atomic length scales. The energy gap observed in the underdoped part of the

phase diagram is then associated with this ordering, and the superconducting gap masks

the energy gap associated with this order for overdoped samples. Finally, the shape of the

superconducting dome is then determined solely by the energy gap associated with Cooper

pairing. The idea of competing order parameters, or of two different kinds of order being the

manifestation of a multi-dimensional order parameter, is also a popular concept in describing

many correlated electron systems.

After more than fifteen years of investigation, significant questions remain unanswered

about the cuprate superconductors, clearly requiring new experimental tools to elucidate

them. Because the relevant length scales throughout the cuprate phase diagram are of

the order of tens of Angstroms, Scanning Tunneling Microscopy (STM) is one of the most

promising such tools. STM allows us to measure both topographic features, allowing us

to get a snapshot of the local surface structure, and spectroscopic features, allowing us to

determine the local density of states, each with sub-Angstrom spatial resolution. In this

thesis, we will investigate the electronic density of states on the atomic scale in each of the

regions of the phase diagram. We hope to demonstrate how this atomic-scale view leads

to both new basic information about the cuprates, and tools which can be used in future

research.

6

Page 15: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

1.2 Outline

The superconducting state of the HTS is well-described as a d-wave BCS superconductor with

a short coherence length. In chapter 3 of this thesis, we will examine a novel consequence of

both these characteristics. When the superconducting state scatters strongly off structural

defects, a novel one-dimensional state at zero energy forms within a coherence length of the

defect as a consequence of the d-wave phase of the pair potential. Because the zero bias state

should only appear near structures that strongly scatter superconducting quasiparticles, its

appearance can be used as a test of the strength of scattering structures about which little is

known. We will also examine the spatially-resolved density of states far from these structures.

We find that the in-gap electronic states form dispersing standing waves in real space. Many

of their characteristics are shown to be consistent with weak scattering, which is sensitive to

the amplitude of the pair potential. Although neither result is a new discovery by itself, they

each have useful application for the study of the cuprate superconductors. First, the zero-

bias state is a novel consequence of having d-wave superconducting correlations. It can thus

be used to identify whether or not a sample has superconducting correlations using density

of states measurements, which are a single-particle probe, even when the phase of the sample

is unknown. Second, the Born formalism used to describe the modulated patterns can be

used to model any modulated density of states pattern. In situations where the electronic

structure of a sample is unknown, how well different candidate electronic states match the

measured modulated patterns could be used to distinguish between the candidate states.

The pseudogap in the density of states, on the other hand, remains an enigma after a

decade of intense research into its origin. Whether it results from residual superconducting

correlations, from some other form of ordering, or some other phenomenon entirely remains

an open question. In chapter 4 of this thesis, we will describe how we find that the electronic

states inside the pseudogap form a static, incommensurate standing wave pattern. While

7

Page 16: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

similar patterns have been found in samples below TC , the patterns found in the pseudogap

regime are unique in that their wavelength is fixed as a function of energy. We find that the

modulations cannot be described by simple Born scattering, which suggests that the patterns

in the pseudogap state result from local ordering. Furthermore, it now appears that when-

ever superconductivity is suppressed, either by decreasing doping, increasing temperature or

applying a magnetic field, local electronic order develops in the density of states.

All the correlated electronic behaviour of the cuprates, from superconductivity to the

pseudogap to the antiferromagnetic insulator, originate on the CuO2 planes common to

these materials. However, electronic density of states measurements have typically been made

using surface sensitive techniques on samples terminated by different crystallographic planes.

In Chapter 5 of this thesis, we will describe how we have made the first definitive density

of states measurements on a single CuO2 plane at the surface of a cuprate superconductor.

The unexpected density of states measured on samples terminated by the CuO2 plane will

require us to consider the possibility that exposing this plane at the surface has changed

its doping level, and hence its phase. Although more measurements, such as searching for

zero-bias states and looking for spatial structures in spatial maps of the density of states,

will be required before we can identify its electronic phase, this discovery opens up a new

avenue with which to examine cuprate physics.

The main result of this work is to demonstrate that Scanning Tunneling Microscopy, with

its ability to correlate local structural and electronic information, is capable of revealing

information about the cuprate superconductors that is simply inaccessible to other probes.

In general, using methods similar to the ones used in this thesis, STM appears poised to

make significant contributions to the study of any correlated electronic system with short

characteristic length scales.

8

Page 17: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

1.3 References

[1] J. Schrieffer and M. Tinkham, Rev. Mod. Phys. 71, 313 (1999).

[2] J. Bardeen, L. Cooper, and J. Schrieffer, Phys. Rev. 108, 1175 (1957).

[3] J. Bednorz and K. Muller, in Nobel Lectures, Physics 1981-1990, edited by T. Frangsmyr

and G. Ekspang (World Scientific Publishing, ADDRESS, 1987), Chap. Perovskite-Type

Oxides- The New Approach to High-Tc Superconductivity.

[4] M. Norman and C. Pepin, Rep. Prog. Phys. 66, 1547 (2003).

[5] M. Kastner, R. Birgeneau, G. Shirane, and Y. Endoh, Rev. Mod. Phys. 70, 897 (1998).

[6] T. Timusk and B. Statt, Rep. Prog. Phys. 62, 61 (1999).

[7] B. Keimer et al., Phys. Rev. B 46, 14034 (1992).

[8] T. Yoshida et al., Phys. Rev. Lett. 91, 27001 (2003).

[9] D. V. Harlingen, Rev. Mod. Phys. 67, 515 (1995).

[10] U. Welp et al., Phys. Rev. Lett. 62, 1908 (1989).

9

Page 18: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 2

Scanning Tunneling Microscopy

2.1 Theory of Measurement

Many tools already exist that allow scientists to measure surface structure, and the electronic

properties of these surfaces. Over the last two decades, scanning electron microscopy has

become a workhorse in measuring local surface structure, and angle-resolved photoemission

has become the de−facto standard for measuring the electronic density of states as a function

of energy and momentum. What makes Scanning Tunneling Microscopy unique is its ability

to measure the density of states locally with sub-Angstrom resolution, and then to correlate

these measurements with simultaneously-acquired topographic information. Although the

methods discussed in this chapter have originated from the study of normal metals, they

offer the tantalizing ability to investigate short-length-scale correlations in any number of

systems, which we will demonstrate explicitly for the HTS in the rest of this thesis. A more

complete discussion of the experimental methods and sources of noise appears in Appendix

A.

In an STM experiment, a sharp metallic wire is brought close enough to the surface of a

sample that electrons can tunnel between them (Fig. 2.1a). [1] If the sample is biased at a

10

Page 19: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

voltage V with respect to the tip, the Golden Rule dictates that a net tunneling current

I(~r, V ) =4πe

h

∞∫

−∞

dE|T 2(d, E, E − V )| ×

[ρT (E)f(E)][ρS(~r, E − V )(1 − f(E − V ))] −

[ρT (E)(1 − f(E))][ρS(~r, E − V )(f(E − V ))] (2.1)

flows from the tip to the sample [2, 3, 4], where T (d, E, U) is the tunneling matrix element for

an electron between the states at energy E in the tip and the states at energy U in the sample,

d is the tip-sample separation, ρT (E) is the density of electronic states (DOS) in the tip at

an energy E, ρS(~r, E) is the DOS of the sample at a position ~r and an energy E, and f(E)

is the Fermi function. Two reasonable assumptions simplify Equation 2.1 into something

more intuitively understandable. First, the tip used in most STM experiments is a simple

metal with a flat density of states over the energies explored by the experiment, allowing us

to set ρT (E)=ρT (0). Second, we will assume, for now, that none of the transitions between

states in the tip and states in the sample are more favorable than others. The tunneling

matrix element will then be given by the overlap of the tip and sample wavefunctions. If

the energy height of the barrier, given by the average of the two work functions (typically

3 − 5eV ), is significantly greater than the energy curvature of the barrier (given by the

difference of the two work functions added to the sample bias, typically 0 − 1eV ), the

tunneling matrix element can be treated with a simple WKB approximation. This gives

T (d, U, E) = T (d) = exp(−d√

8mφh2 ), where m is the mass of an electron, and φ is the average

of the tip and sample work functions. The simplified expression,

I(~r, V ) =4πρT (0)e

he−d

8mφ

h2 ∞∫

−∞

dEρS(~r, E)[f(E + V ) − f(E)], (2.2)

reveals that the the tunneling current changes proportionally to the integral of the local

density of states of the sample between EF and EF + V as a function of voltage, and

exponentially as a function of the tip-sample separation.

11

Page 20: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Ψ

Ψ

Tip

Sample

overlap

region

Tip

VI

Sample

Vacuum

Energy EF

EF

+V

De

nsi

ty o

f S

tate

s (x(t),y(t))Z(t)

I

Itunnel

ref-A B

Figure 2.1: (A) (left) This is a schematic diagram of an STM junction. Theexponential dependence of tunneling current on tip-sample separation derrivesfrom the overlap of the evanescent wavefunctions of the tip and sample in thevacuum barrier. (right) Applying a voltage bias to the tip relative to thesample yields a net tunnel current flowing from the tip to the sample (sampleto the tip) proportional to the number of occupied states in the tip(sample)convoluted with the number of empty states in the sample (tip). (B) A feed-back loop ensures that the tunnel current remains constant by controlling thetip-sample separation z(t). Recording z(t) while rastering the tip across thesurface (x(t),y(t)) yields a height-field of the surface that is called a topograph.

A number of theorists, most prominently Harrison, [5] have objected to this formulation

of the tunneling current. They point out that the tunneling matrix element should be a tun-

nelling attempt frequency multiplied by the probability for successful electron transmission

through the potential barrier, with the former having been omitted by the above treatment.

This attempt frequency can be estimated as the particle flux incident on the tunnel barrier,

which, from dimensional arguments, is inversely proportional to the velocity of the parti-

cles. In a three-dimensional metal, the attempt frequency cancels out the density of states

term in the tunneling current. However, for the materials investigated here, which cannot

be approximated as simple three dimensional metals, we do not expect such a cancellation

to occur. Moreover, as demonstrated by the pioneering work of Giaever [6], tunneling does

sensitively probe the density of states in a superconductor. Consequently, we proceed with

the formulation presented in Equation 2.2 above.

The tunneling current’s exponential dependance on tip-sample separation can be ex-

12

Page 21: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A

B

C

D

Figure 2.2: These are STM topographs of the same 700A × 350A area of aAu(111) surface taken under two different tunneling conditions: I = 200pAand V = +100mV for (A) and I = 200pA and V = −100mV for (B). Tohighlight the modulated pattern, we have taken a spatial derrivative of (A),shown in (C), and of (B), shown in (D). The location of the crests of themodulated pattern have been designated by the yellow lines in (C). Plottedin (D) are the same lines, which no longer match up with the crests of themodulations.

ploited to learn about the topography of the surface (Fig. 2.1b). [7] To take a topographic

image of the surface, a feedback loop is used to keep the tunneling current constant while

the tip is rastered across the surface. The output of the feedback loop, which changes the

height of the tip in response to changes in the tunneling current, is recorded as a function of

position to produce an image. If the sample has the same integrated density of states every-

where on the surface, the feedback loop will change the height of the tip solely in response

to structural changes at the surface. In this way, heights of features such as step edges can

be exactly determined. However, the density of states is almost never homogeneous across

a sample, and the feedback loop also changes the tip’s height in response to these changes.

Thus topographic images contain a convolution of both structural and electronic informa-

tion, and must be considered contours of constant electron density, not merely local maps of

surface structure.

13

Page 22: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

The canonical examples of this convolution of structural and electronic information are

topographic images taken on certain surfaces of noble metals, such as Au(111) (Fig. 2.2). The

STM topographs of the Au(111) surface contain three main features: line defects separating

two terraces of different height [7], a herringbone feature [8], and a modulated pattern found

near the line defects. [9, 10] The structural nature of the line defect can be confirmed by

taking topographs at different voltages. Changing the voltage should change the integrated

density of states, with the feedback loop responding to this change in the tunneling current

by changing the tip height. If the line defect is a structural feature, the height of the two

terraces should be the same in both topographs. Averaging all the lines in each topograph

together to remove the effects of the other two topographic features, the height separating

the two terraces is found to be equal (2.4A), proving that the line defect is a structural

feature. What’s more, it agrees precisely with the distance between two adjacent atomic

planes of the crystal, indicating that the step edges are each one-atom high. The herringbone

pattern also does not change significantly between the two topographs, and hence can also

be identified as having a structural origin. The modulated pattern, on the other hand, has

a different period of modulation for the two biases (Fig. 2.2c & d), indicating its origin is

primarily electronic. Spectroscopic information, which can be acquired simultaneously with

topographic information, can be used to further elucidate the origin of these features.

Scanning tunneling microscopy can also be used to measure the local density of states.

Starting from Equation 2.2, the differential conductance of the tunnel junction is given by

dI(~r, V )

dV=

4πe|T 2(d)|ρT

h

∞∫

−∞

dEρS(~r, E)d

dV[f(E + V )]. (2.3)

In the scheme discussed here, positive voltages correspond to energies above the Fermi energy

(empty states), and vice versa. At zero temperature, the differential conductance as a

function of voltage is then directly proportional to (only) the sample density of electronic

states at energies relative to the Fermi energy. [3, 4] This result has been confirmed by

Yazdani and coworkers, who demonstrated that the conductance as a function of bias voltage

14

Page 23: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A

B C

D E

F

Figure 2.3: (A) This is an 560A × 560A STM topograph of Au(111) takenwith a I = 100pA and V = 200mV junction at T = 100K. Conductancemaps were acquired simultaneously with this topograph at a range of voltagesusing lockin parameters f = 513Hz, dV = 3mVrms, and τ = 20mS. Shownare subsections of these conductance maps taken in the square region in (A)at (B) V = −415mV , (C) V = −288mV , (D) V = −161mV , and (E) V =−34mV . Also shown is the conductance map of the entire region in (A) takenat V = 93mV (F).

on Nb(110), a BCS superconductor, exactly matches the BCS density of states. [11] This

spectroscopic ability can be used to examine the spatial arrangement of the density of states

at a given energy, or the energy-dependence of the density of states at a given position. To

take either measurement, the tip is moved to a particular location on the sample with the

feedback loop closed. The feedback loop is then opened, the bias is changed to the voltages

of interest, and the differential conductance of the tunnel junction is recorded. The feedback

loop is then closed, and the tip is moved to the next point where density of states information

will be measured. This method of taking the data introduces the (unavoidable) complication

that the feedback loop, in an attempt to keep the tunneling current constant, is normalizing

the integrated density of states at every point that spectroscopy data is taken. However,

if the local variation of the density of states does not result in features that show up in

topographic images, this normalization has a minimal effect.

15

Page 24: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A D

0.00

0.10

0.20

0.30

0.40

-400 -300 -200 -100 0 100 200

Wav

en

um

be

r (1

/Å)

Energy

CB

Figure 2.4: (A & B)These are false-color images of the Fourier amplitudeof the data in Fig. 2.3c&e. The white arrows point at a dispersing featurecorresponding to the standing wave pattern seen near the step edge. (C)The wavevector can be determined for a set of energies, mapping a dispersionrelation for the modulated patterns (closed circles). The dispersion of anelastic scattering interference model based on photoemission data, as discussedin the text, is shown as open circles. (D) ARPES data from Ref. [12] clearlyresolves the dispersion of a surface state. Note that ARPES has resolved asplitting of the dispersion. The width of the peaks in Fourier transforms of theSTM data (≈ .06A−1) are approximately equal to this splitting, and precludeit from being resolved in the STM data.

Returning to the simple example of the Au(111) surface, local density of states maps can

be used to characterize the modulated electronic pattern seen in topographs next to step

edges. From topographs taken at different voltages, the modulations were hypothesized to be

electronic in origin. The DOS maps confirm this- the modulated pattern next to step edges

has a wavelength that changes with energy (Fig. 2.3). [10] To more thoroughly characterize

the wavelength of these periodic modulations, we can take Fourier transforms of the DOS

maps (FTDOS, Fig. 2.4a&b). [13] The FTDOS maps taken at several energies can then be

used to compile a dispersion relation for this feature (Fig. 2.4c). ARPES data on Au(111)

shows that a two-dimensional surface band is located at these energies (Fig. 2.4d), but

at a wavevector exactly half that seen by STM. The two sets of data are connected by a

phenomenon known as elastic scattering interference.

In picture terms, there are two roughly equivalent ways to understand elastic scattering-

16

Page 25: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

q

k=(0,0)

k=(π,π)

k=(0,π)

k=(0,0)

k=(π,π)

k=(0,π)q

q=(0,0)

q=(π,π)

q=(0,π)

A CB

Figure 2.5: Here we illustrate both the wave-based and particle-based toymodels for scattering interference. (A) In a wave-based picture, electron wave

functions at a given energy have a set of resitricted ~k values (blue circle). Here,the thin black square extends between (±π,±π). Density of states modulations

resulting from interaction with defects have a wavevector ~q = 2 × ~k (redcircle). (B) Alternatively, we can take a particle-based view. Electrons existon contours of constant electron energy, which is the same blue circle as in (A).Density of states modulations are expected for the set of wavevectors ~q joiningthe points on these curves (red). The largest density of states contributionswill be for the wavevectors connecting parallel parts of the curves of constantelectron energy. (C) Compiling the full set of these wavevectors ~q at a givenenergy yields a scattering map which agrees with the wave-based picture in(A).

17

Page 26: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

a wave-based picture [13] (Fig. 2.5a) and a particle-based picture [14, 15] (Fig. 2.5b). The

density of states over regions with no scattering centers contains electron wave functions

with many different momenta (~k) and phases incoherently superposed with one another. A

defect can pin the phase of electron wave functions at the scattering center, making incoming

and outgoing waves interfere with one another to form standing waves. Electrons at a given

energy have a set of allowed ~k vectors; at the Fermi energy, the locus of allowed ~k vectors is

the Fermi surface. The electron density, however, is given by the square of the wavefunction,

and thus has a wavevector twice as long as the wavevector of the wavefunction. The density

of states patterns will then occur for a set of wavevectors ~q = 2 × ~k. Equivalently, we can

consider scattering in particle terms. Electrons at a given energy exist on curves of constant

electron energy; at the Fermi energy, this curve is the Fermi surface. An electron with a

given ~ki from this set can scatter elastically off a defect to a different wavevector from this

set ~kf . The density of states correction from scattering will occur for the set of wavevectors

~q = ~kf − ~ki. As can be seen in Figure 2.5a&c, the wavevectors comprising the density of

states correction from elastic scattering produced by these two pictures is identical. Applying

either these two procedures to the ARPES data from Au(111), we can calculate a dispersion

curve that agrees with the dispersion of the standing wave pattern seen near step edges using

STM (Fig. 2.4c). The modulated patterns seen in STM can thus be understood to result

from the elastic scattering of the Au(111) surface state off the step edge.

The herringbone pattern, which we suspected to be primarily structural in origin from

the STM topographs, also has an electronic component (Fig. 2.6). [16] Taking conductance

measurements as a function of energy both in the small flat region between two herringbone

corrugations and between two pairs of herringbone corrugations shows that there is an ad-

ditional contribution to the density of states at the herringbone centered around −425meV .

At this energy, elastic scattering of the surface band has a real-space period of ≈ 63A, which

is precisely the spacing between pairs of the herringbone corrugations. The herringbone pat-

18

Page 27: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

3

4

5

6

7

8

9

-0.6 -0.4 -0.2 0 0.2 0.4 0.6

Spectra on Au 111 surface

hcp regionherringbone crestfcc region

dI/

dV

(a

rb.

un

its)

Energy (V)

fcc regionhcp region

herringbone crestA B

Figure 2.6: (A) An STM topograph of a 270A × 270A region (I = 130pAand V = −100mV ) rendered in 3D. The long flat regions (38A) betweenherringbones has an fcc structure, and the short region (25A) in between thetwo ridges has an hcp structure. [8] (B) Taking point spectra on top of anfcc region, an hcp region, and the ridge separating the two (the crest of theherringbone pattern) reveals an increased number of states accumulate in thehcp region at about −425mV and an increased number of states accumulateat the crests at about −375mV and −275mV . Data were taken at T = 80K.

tern is effectively acting like a Fabry-Perot resonator for the surface state. As a further test

of this hypothesis, spectra were also taken on the crest of the herringbone corrugation, which

revealed an additional contribution to the density of states at both −375mV and −275mV .

At these energies, elastic scattering of the surface band should produce a real-space period of

38A and 25A, respectively. These two periods are exactly the size of the the large flat region

between two sets of crests and the small flat region between two crests. [8] Although the data

set isn’t conclusive, it appears that the herringbone pattern acts like a weak scattering cen-

ter for the surface state band, with an enhanced density of states at energies (wavelengths)

where the surface state can resonate between parts of the herringbone structure.

19

Page 28: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

2.2 References

[1] G. Binnig and H. Rohrer, in Nobel Lectures, Physics 1981-1990, edited by T. Frangsmyr

and G. Ekspang (World Scientific Publishing, ADDRESS, 1993), Chap. The Develop-

ment of the Electron Microscope and of Electron Microscopy.

[2] J. Bardeen, Phys. Rev. Lett. 6, 57 (1961).

[3] J. Tersoff and D. Hamann, Phys. Rev. Lett. 50, 6858 (1983).

[4] J. Tersoff and D. Hamann, Phys. Rev. B 31, 805 (1985).

[5] W. Harrison, Solid State Theory (McGraw-Hill, ADDRESS, 1970).

[6] I. Giaever, Phys. Rev. Lett. 5, 147 (1960).

[7] G. B. (sic), H. Rohrer, C. Gerber, and E. Weibel, Phys. Rev. Lett. 49, 57 (1982).

[8] J. Barth, H. Brune, G. Ertl, and R. Behm, Phys. Rev. B 42, 9307 (1990).

[9] M. Crommie, C. Lutz, and D. Eigler, Nature 363, 524 (1993).

[10] Y. Hasegawa and P. Avouris, Phys. Rev. Lett. 71, 1071 (1993).

[11] A. Yazdani et al., Science 275, 1767 (1997).

[12] M. Hoesch et al., Phys. Rev. B 69, 241401 (2004).

[13] L. Peterson et al., Phys. Rev. B 57, 6858 (1998).

[14] J. Hoffman et al., Science 297, 1148 (2002).

[15] K. McElroy et al., Nature 422, 592 (2003).

[16] W. Chen, V. Madhavan, T. Jamneala, and M. Crommie, Phys. Rev. Lett. 80, 1469

(1998).

20

Page 29: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 3

Scattering in the Superconducting

State

3.1 Introduction

The superconducting state of the HTS is referred to as being unconventional since, although

a modified BCS theory has been remarkably successful in describing their properties, their

measured properties are often fundamentally different than their BCS counterparts. Per-

haps the most significant differences arise from the pair potential in HTS having a d-wave

symmetry, compared with s-wave symmetry for BCS superconductors (Fig. 3.1a). This un-

conventional pairing symmetry directly impacts most measurements through the density of

states, which is sensitive to the amplitude of the pairing potential. In the HTS, the pairing

potential has nodes arising from its d-wave nature. Averaging the density of states equally

for all angles produces a V-shaped energy gap with in-gap states, in contrast to the s-wave

case, which has no in-gap states (Fig. 3.1b). More spectacular differences between s and

d-wave superconductors arise in experiments sensitive to the phase of the pairing potential.

The symmetry of the pair potential was originally established in experiments probing

21

Page 30: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

-2 20 -2 20

1

2

1

2

Ε/∆ Ε/∆

Ν (

Ε)/

Ν (

Ε )

s

0

F

Ν (

Ε)/

Ν (

Ε )

s

0

F

<0,π>

<π,π>

<0,π>

<π,π>

A B

C D

++

-

-

Figure 3.1: The s-wave pair potential is isotropic as a function of angle (A),while the d-wave pair potential changes sign upon π/2 rotations (B). Theamplitude of the pair potential affects many measurements through the densityof states, shown here for both the s-wave (C) and d-wave (D) pair potentials.The latter was calculated by averaging the density of states equally for allangles.

the phase of the superconducting order parameter, which has the same phase as the pair

potential. These experiments found novel electronic effects arising from the sign change of

the d-wave order parameter upon π/2 rotation, which simply don’t exist for isotropic s-wave

order parameters. [1, 2] In one of these, a superconducting loop was formed by joining

orthogonal edges of a superconducting HTS sample with a BCS superconductor (Fig. 3.2).

The single quantum state that the phase-coherent Cooper pairs have condensed into can

be described by a pseudo-wavefunction that, thermodynamically, is the order parameter for

superconductivity. The square of the amplitude of this wave function is the density of Cooper

pairs in the condensed state, and the phase of the condensate has the symmetry of the pairing

potential. When two superconductors are connected to one another, a current flows between

them in response to any change in the phase of the superconducting order parameter across

22

Page 31: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B C

Figure 3.2: The dc-SQUID experiment was used to show that the order pa-rameter in HTS changes sign upon rotation by π/2. A square block of testsuperconductor is connected into a loop by an isotropic s-wave BCS super-conductor. If the phase of the test sample does not change sign at the twojunctions, the currents from the two junctions will cancel one another (A andB), and there is no net current around the loop. However, a corner junction fora d-wave superconductor has an additive contribution from the two junctionsbecause the order parameter changes sign on π/2 rotation. (Figures adaptedfrom Ref. [1]).

the junction (the dc Josephson effect), even in the absence of an applied field. If HTS had

an order parameter that did not change sign upon π/2 rotation, the current flow from one

junction of the corner SQUID would cancel the current flow from the other. In fact, there is

a net current flow around the loop, confirming that that the order parameter does not have

s-wave symmetry. Further details in this work showed that the order parameter not only

changes phase on π/2 rotations, but has a strictly d-wave character.

Potential scattering is another sensitive probe of the d-wave pair potential, and produces

novel phenomena which have no analog for s-wave superconductors. [3] In an s-wave su-

perconductor, there are no available single-particle states at low energies for potentials to

scatter unless they break a Cooper pair apart. Consequently, potentials that do not break

time-reversal symmetry, like magnetic impurities, have no effect, invariant of the strength

23

Page 32: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

of scattering. [4] In a d-wave superconductor, on the other hand, there is a finite density

of low-energy quasiparticle states because the magnitude of the pair potential goes to zero

at certain angles. [5, 6] For strong potentials, including non-magnetic impurities [7, 8, 9],

surfaces, and interfaces [10, 11, 12, 13, 3], the scattering of these low-energy states can lead

to the suppression of unconventional order parameters. [14, 15] Quasiparticles inside this

region of suppressed superconductivity can scatter in one of four ways off the pair potential

of the superconducting region (Fig. 3.3a). Two of these are intuitively obvious- reflection or

transmission of the injected quasiparticle. Because BCS superconductivity involves a mixing

of electron-like and hole-like quasiparticles, a third possibility is that the injected quasiparti-

cle is transmitted through the boundary as a quasiparticle of the opposite type. Finally, the

most interesting of these is the so-called Andreev process, in which a quasiparticle injected

into the superconductor pairs with another quasiparticle in the superconductor to form a

Cooper pair, and a quasiparticle of the opposite charge is retro-reflected from the interface.

This retro-reflected quasiparticle then reflects off the strong scattering potential, and can

subsequently be Andreev scattered again by the pair potential of the superconductor (Fig.

3.3b). In superconductors with higher pairing symmetries (non s-wave), the pair potential

that scatters the quasiparticle can be different for the two Andreev processes. In the case

that the lobes of the d-wave order parameter are oriented at a π/4 angle to the scatter-

ing interface, the pair potential has opposite sign (and the same magnitude) for successive

Andreev reflections for all possible quasiparticle trajectories. As a direct consequence of un-

conventional pair potential symmetry, the back-scattered quasiparticles interfere coherently

to form in-gap states that are bound to the scatterer.

This so-called Andreev bound state, which is a novel signature of the sign change in

the pair potential, should appear as a peak in the in-gap single particle density of states.

Indeed, a zero bias conductance peak has been seen in a large number of both planar and

STM tunnel junctions on 110 surfaces, a geometry that should produce an Andreev bound

24

Page 33: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

injected quasiparticle

reflected quasiparticle

1

injected quasiparticle Electron-like quasiparticle2

Normal Superconducting

Hole-like quasiparticle3 injected quasiparticle

4 injected quasiparticle

Hole-like quasiparticle

Cooper Pair

A B

++

-

-

++

-

-

1. Injected electro

n

2. Retro-re

flected hole

3. Reflected hole

4. Retro-reflected electron

Normal Superconducting

θ

sc

att

eri

ng

in

terf

ac

e

Figure 3.3: (A) A quasiparticle in a region of normal material can scatter inone of four ways off the pair potential of a superconductor. (B) The region ofsuppressed superconductivity (normal) can support quasiparticles (1) whichundergo Andreev retro-reflection from the normal-superconductor interface(2), are then reflected by the scattering interface (3), and are finally retro-reflected again by at the normal-superconducting interface (4) in a closed loop.

state at the Fermi energy (Fig. 3.3c, with θ = π/4). [16, 17, 18, 19, 20, 21, 22, 23, 24, 25]

Although the in-gap states seen in these experiments could arise from other sources, the

accumulated evidence [3], particularly the state’s sensitivity to disorder [26, 27], suggests

that the effect originates from Andreev Scattering. Still, to date, a direct observation of

neither the suppression of the order parameter near such an interface, nor of the spatial

confinement of the state to within a coherence length of the interface, has been made. STM

investigations of the scattering from both non-magnetic [28, 29, 30, 31] and magnetic [32]

pointlike impurities have also shown in-gap resonances in the density of states within a

coherence length of the scattering center. Although the theoretical understanding of these

in-gap resonant states has largely been in the context of either Born or unitary scattering

[7, 8, 9], their presence has also been posited to be an extention of the Andreev mechanism.

[33]

In this chapter, we will extend both bodies of work by measuring the local density of states

25

Page 34: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

both near and far away from spatially extended defects. [34] We find that step edges running

along the 110 direction strongly suppress superconducting order- the superconducting gap

in the density of states fills in upon approaching the step. Furthermore, we find a novel

zero-bias state that exists only within a coherence length of the step, and extends for several

coherence lengths along it. This results from Andreev scattering at the step edge, which

is sensitive to the phase of the d-wave pair potential. This behavior is in contrast to our

finding that 110 twin boundaries show no suppression of the order parameter, nor any

bound state. This suggests, contrary to published theory [35, 36], that twin boundaries are

not strong scatterers, and may result from a more gradual atomic-scale strain field. [37]

Finally, the low energy electronic states far away from defect structures are seen to form

spatially modulated patterns in real space. [38, 39, 40, 41] Although no scattering center

can be identified in the data, it has been suggested that these patterns result from weak

scattering, which probes the amplitude of the pair potential. [42, 43, 44, 45, 46, 47] We

conclude by proposing how the results presented in this chapter can be applied to situations

where the electronic ground state of the system is not well-understood.

3.2 The Andreev Bound State

To examine the effects of spatially extended scattering structures in the HTS, we performed

STM measurements on the superconducting state of 1000A thick Bi2Sr2CaCu2O8+δ thin

films grown using molecular beam epitaxy. The slightly underdoped samples had a resistive

superconducting transition at 84K and a transition width of 4K. The samples were mechan-

ically cleaved along the c-axis direction in UHV at room temperature prior to insertion into

a home-built UHV STM operating at liquid helium temperatures (LTSTM). The cleaving

process exposes flat BiO terraces at the surface, as the inter-unit-cell BiO-BiO bond is the

weakest one in the crystal structure. [48] Figure 3.4 shows a typical STM topograph and

26

Page 35: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

conductance spectrum taken on this surface, both of which are comparable to data reported

in the literature for cleaved single crystal samples. [28, 29, 30, 31, 32, 49] Notably, the

topograph shows a clear atomic corrugation, and a b-axis superlattice distortion known to

run throughout every plane of the bulk. [50] The spectra show an inchoate energy gap and

pronounced peaks at about ±40mV , both characteristic of the density of states in the super-

conducting state of Bi2Sr2CaCu2O8+δ. However, cleaving thin films requires significantly

more mechanical force than cleaving single crystals, most likely because thin films contain a

higher density of defects. Whether the consequence of the thin film growth having more step

edges, and hence requiring more forceful cleaves, or of the more forceful cleaves resulting in

stepped surfaces themselves, the samples were found to BiO terraces 100− 1000A in extent

containing extended defect structures such as step edges and twin boundaries.

Cleaved thin film samples often contain BiO terraces terminated by step edge boundaries,

such as the one shown in the STM topograph in Figure 3.5a. Because tunneling conductance

spectra taken far away from this step edge (Fig. 3.6a) resemble the spectra taken on cleaved

single crystal samples, which are terminated by BiO surfaces, we can identify the terrace

on the top side of the step edge as being the BiO plane. From comparing the measured

distance between the top and bottom terrace to the known separation of the crystallographic

planes [50], the bottom terrace is shown to be the lower of the two CuO2 planes of the

Bi2Sr2CaCu2O8+δ crystal structure (Fig. 3.5b), which we will focus on in Chapter 5. This

means that the step edge terminates the upper of the two CuO2 planes of the unit cell.

To determine the angle that the step edge makes relative to the pair potential, a careful

registry of the atomic corrugations in the topograph are made. These corrugations most

likely represent Bi atoms in the surface BiO plane, and the Cu atoms of the CuO2 plane sit

directly beneath them. The step edge is seen to run at roughly a π/4 angle to the nearest

neighbor direction of the corrugations, which corresponds to a π/4 angle relative to the

Cu−O bond direction. Angle resolved photoemission spectroscopy has established that this

27

Page 36: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

0

1

2

3

4

-100 0 100

16Å

24Å

32Å

40Å

48Å

56Å

Co

nd

ucta

nce (

pS

)

BA

Figure 3.4: (A) STM topographs of the BiO surface of a cleavedBi2Sr2CaCu2O8+δ thin film far from any defects resemble topographs onpristine single crystals (100A × 100A area imaged with I = 200pA andV = −200mV ). (B) Spatially-resolved conductance spectra taken along theline indicated in (A) resemble published data taken on single crystals (dc junc-tion impedence Rj = |−200mV |/200pA = 1GΩ, the data are offset for clarity).[28, 29, 30, 31, 32, 49] Both figures have been adapted from Ref. [34].

28

Page 37: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

+

-

(0,π

)

(π,π)

11

0

ste

p e

dg

e

Normal Superconducting

02468

0 25 50 75 100

12

Å2

37

Å5

75

Å

A B C

Lateral Distance (Å)

He

igh

t (Å

)

(0,π

)

(π,π)

BiO

SrOCuO

Ca

Figure 3.5: (A) This STM topograph shows a step edge oriented at a π/4 tothe underlying lattice (I = 50pA and V = −200mV ). It has been renderedin a way to bring out the atomic corrugations on both the high (right) andlow (left) sides of the step edge. (B) This graph shows a single line scanfrom the topograph in (A) overlayed with the known separation between thecrystallographic planes of the Bi2Sr2CaCu2O8+δ. The spectra in Figure 3.6awere taken at the positions indicated by the arrows in (B) and the boxes in(A). (C) A schematic diagram of the subsurface CuO2 plane can be createdfrom the atomic positions in the topograph. All figures have been adaptedfrom Ref. [34].

corresponds to the nodal direction of the pair potential. [51] We thus have a situation similar

to the schematic diagram in Figure 3.5c, where a 110 step edge terminates the edge of a

CuO2 plane.

The interruption of the CuO2 lattice at the step edge should serve as a strong pertur-

bation to the superconducting states which exist on this plane. First, this should suppress

the superconducting order parameter strongly within a coherence length of the step edge.

[14, 15] Intuitively, the spectra should thus resemble the V-shaped spectra typical of conduc-

tance spectra taken over pristine samples at temperatures well above the superconducting

transition. [49] Although this spectrum has been interpreted to result from either inelastic

scattering [52] or a non-Fermi liquid ’normal’ state [53], superconducting order is unambigu-

ously suppressed in these spectra. What’s more, the suppression of the order parameter

29

Page 38: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

0

1

2

3

50Å

37Å

25Å

12Å

-100 0 1000

1

2

3

4

5

6

12Å

25Å

37Å

50Å

62Å

-100 0 100

Energy (mV)

A B

Co

nd

ucta

nce (

pS

)

Figure 3.6: Spatially resolved STM spectra taken along a line perpendicularto (A) and along (B) the 110 step edge in Figure 3.5 are shown. For both(A) and (B), the dc junction impedence was Rj = | − 200mV |/200pA = 1GΩ,and the lines indicate the zero conductance level for the spectra, which havebeen offset for clarity. The figures have been adapted from Ref. [34].

near the interface raises the possibility that quasiparticles in this region can undergo An-

dreev reflection from the pair potential of the superconductor. [3] For a step edge oriented

at a π/4 angle with respect to the pair potential, quasiparticles reflected by the step edge

will be retro-reflected by a pair potential of opposite sign for successive Andreev reflections.

Moreover, for this orientation, there should be constructive interference between successively

Andreev reflected quasiparticles for all possible quasiparticle trajectories. We therefore ex-

pect scattering from the step edge in Figure 3.5 to give rise not only to a suppression of the

order parameter, but also to the appearance of an Andreev bound state.

We used STM conductance spectra to examine the effects of quasiparticle scattering on

30

Page 39: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

the local density of states near the step edge. Spectra taken perpendicular to the step edge

(Fig 3.6a) clearly demonstrate both the consequences arising from quasiparticle scattering at

the step edge we had anticipated. Starting about 30A from the step edge, the conductance

peaks at about ±40mV begin to decrease in amplitude, and, concomittently, the in-gap con-

ductance increases. These spectra as a function of distance are remarkably similar to spectra

taken over pristine samples at temperatures well above the superconducting transition. [49]

The similarity of these two sets of spectra indicate that superconducting order is completely

suppressed near the step edge. In addition, within a coherence length of the step edge,

a well-defined zero bias peak appears in the tunneling conductance. Conductance spectra

taken alongside the step edge (Fig. 3.6b) confirm that this state persists for several coher-

ence lengths next to the step edge. This observation confirms that a novel one-dimensional

zero-bias state is bound to step edges running along the nodal directions of the pair poten-

tial. The spatial mapping of the suppression of superconducting order and the formation of

the Andreev bound state near an extended 110 defect are the central new results reported

in this section.

The results reported here on the density of states near 110 step edges bear a strong

resemblance to tunneling results on 110 surfaces. [3] Although the geometry for quasi-

particle scattering is similar in both sets of experiments, the geometry of the experiments

themselves is different. Here, we are tunneling into the states at a 110 interface in the

out-of-plane direction, whereas previous experiments have tunneled in-plane into a 110

oriented surface. Our results confirm that the Andreev scattering modifies the states at

the interfaces, rather than only influencing tunneling into the sample on in-plane directions.

Our results also add considerable detail to earlier reports of the Andreev bound state’s sen-

sitivity to disorder. [26, 27] In previous experiments on planar tunnel junctions, the zero

bias conductance peak was shown to be suppressed in junctions where the 110 surface

had been disordered from ion bombardment. [26] Here, the sensitivity to disorder manifests

31

Page 40: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

itself in the spatial inhomogeneity of the peak. As seen in Figure 3.6b, although the peak

extends for many coherence lengths along the step edge, it vanishes at other locations. The

data does not, however, show a clear correlation between disorder in the topography and the

suppression of the zero bias state.

Because it can find Andreev bound states at strongly scattering defects, STM can be

used to characterize extended defects about which little is known. Consider the 110 twin

boundary shown schematically in Figure 3.7a. This interface has the same qualities that led

to the formation of the Andreev bound state at the 110 step edge, provided, as has been

suggested [35, 36], that it strongly scatters quasiparticles. In fact, this geometry is similar

to aligned grain boundary junctions, where a zero-bias conductance peak has been seen. [18]

In Figure 3.7b, we show an STM topograph of a twin boundary that has this geometry.

However, spectra taken near this twin boundary (Fig. 3.7c) show no evidence either for the

suppression of superconducting order, nor of a zero bias conductance state. This naively

suggests that twin boundaries do not have a strong defect potential. An alternative view of

the twin boundary is that it is the visible symptom of a gradual atomic-scale strain field.

[54] Self-consistent calculations of the order parameter near a strain-induced twin boundary

show that d-wave superconducting order is not strongly suppressed nearby. [37] This view

is supported by the conductance spectra taken near the twin boundary in Figure 3.7.

Finally, previous theoretical works have raised the question whether secondary super-

conducting order parameters can stabilize at extended scattering interfaces. [55, 35, 36, 56]

Previous experiments on 110 surfaces have observed a splitting of the zero bias conduc-

tance peak below T ≈ 7K under zero applied field, which was interpreted to result from

the formation of a subdominant complex order parameter at the interface. [16, 22] Within

our experimental resolution (±3mV ), we have not seen a splitting of any of the zero bias

conductance peaks we have found near step edges at T = 4.2K. The width of the zero-

bias conductance peaks in this work (±15mV ) exceeds both the width of the unsplit peaks

32

Page 41: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

B

-100 0 100

1

2

3

25Å

50Å

Energy (mV)

Co

nd

uc

tan

ce

(p

S)

C

+

-

110 twin boundary

Normal

Superconducting

b

a(0,π)

a

b(0,π)

A

Superconducting

Figure 3.7: (A) This is a schematic representation of a 110 twin boundary.A sample quasiparticle trajectory, and the pair potential, is also shown. (B)This is an STM topograph of a twin boundary similar to the one sketchedschematically in (A). It is identifiable as a 110 twin because of the per-pendicular b-axis supermodulation on both sides of the twin (I = 200pA andV = −200mV ). (C) These STM spectra were taken along the twin boundaryevery 25A (Rj = | − 200mV |/200pA = 1GΩ). There is some evidence for theformation of an in-gap kink in the conductance at some (denoted by arrows),but not all (the middle trace) locations near the twin boundary. The zero con-ductance level of the spectra, which have been offset for clarity, is indicatedby the bar. These figures have been adapted from Ref. [34].

33

Page 42: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

(±5mV ) and the splitting of the peak (±2mV ) seen in the planar tunnel junction experi-

ments, and may thus be masking the splitting. The width of the zero bias conductance peak

has been proposed to be related to the roughness of the interface. [27] However, calculations

show that a roughness much larger than we observe would be required to explain the mea-

sured width. We believe a more likely explanation is that the width of the peak seen here

arises from a reduced lifetime for the Andreev bound state, caused by scattering from the

atomic-scale disorder at the step edge.

The possibility of nucleating a sub-dominant order parameter at 110 twin boundaries

has also been raised, although most models have assumed a strong suppression of supercon-

ducting order near the twin boundary. [35, 36, 56] However, calculations treating the twin

boundary as an atomic-scale strain field have also shown that a real subdominant order pa-

rameter with s-wave symmetry mixes with the dominant d-wave order parameter near twin

boundaries. [37] Although one would expect that the admixture of a subdominant s-wave

order parameter, which introduces a finite amplitude energy gap along the nodal direction of

the d-wave gap, would produce a hard energy gap in the density of states, calculations have

shown the low-energy density of states to remain unperturbed in this situation. [37] However,

they find the subdominant order parameter producing in-gap kinks in the density of states.

[37] Consistent with the calculations, the STM conductance spectra show no evidence for a

hard energy gap at low energies, and do show evidence for in-gap kinks at some positions

near the twin boundary. More definitive spatial mapping of the tunneling conductance will

be required to determine why the in-gap kinks disappear at other locations along the twin

boundary, and, ultimately, whether stronger evidence can be found for this sub-dominant

order parameter.

34

Page 43: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

0.1

0.2

0.3

-0.4 -0.2 0 0.2 0.4

dI/

dV

(n

S)

Energy (V)

A B

Figure 3.8: (A) STM topographs of the BiO surface of a cleavedBi2Sr2CaCu2O8+δ single crystal taken at 40K (380A × 380A area imagedwith I = 20pA and V = −200mV ) resemble those taken at liquid heliumtemperatures. (B) Typical conductance spectra also show the same features-sharp peaks at ≈ ±40mV and an inchoate energy gap- as those taken at liquidhelium temperatures (dc junction impedence Rj = | − 150mV |/40pA).

3.3 Local DOS Modulations

In the previous section, we have examined how the density of states changes on the atomic

length scale near spatially extended defects. We found that a novel zero-energy state forms

as a consequence of the phase of the d-wave pair potential near extended defects that strongly

scatter quasiparticles. Here, we will examine the spatially-resolved density of states far away

from these defects. Surprisingly, we find that the low energy states form dispersing periodic

patterns in real space. [38, 39, 40, 41] This dispersion is found to be the consequence of weak

scattering, which is sensitive to the amplitude of the pair potential. [42, 43, 44, 45, 46, 47]

We have examined the spatial structure of the density of states far away from extended de-

fects in the superconducting state. [38] Slightly underdoped single-crystal Bi2Sr2CaCu2O8+δ

samples (TC = 85K) were mechanically cleaved along the c-axis direction in ultra-high vac-

uum at room-temperature prior to performing measurements with a home-built variable-

temperature STM (VTSTM) at T = 40K. Single crystals were used for this study because

cleaved samples rarely contain extended defects, and flat, undisturbed terraces can extend

35

Page 44: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

for > 10000A. Atomic-resolution topographs of the cleaved surface show the atomic corruga-

tions and the b-axis superlattice distortion familiar from low-temperature measurements (Fig.

3.8a). Energy resolved conductance spectra taken at a fixed position on the sample reveal a

density of states typical of that measured in the superconducting state of Bi2Sr2CaCu2O8+δ

(Fig. 3.8b). We will focus on the tunneling conductance measured at a fixed energy as a

function of position, which maps the spatial arrangement of the electronic states at a given

energy. The local density of states (LDOS) maps measured for energies inside the energy

gap, shown in Figure 3.9, contains periodic structures, including an unexpected modulation

running at 45 to the b-axis supermodulation.

To analyze periodicity of the modulated structures seen in the LDOS maps, a two-

dimensional Fourier analysis was performed. Limits on the tunneling conductance were

imposed to limit the effects of the two low-energy resonances in the data in the FFTs

(25 − 55pS for 7mV , 50 − 175pS for 13mV , 75 − 175pS for 20mV , and 90 − 190pS for

13mV . The power spectrum (Fig. 3.9b) contains three main features. The broad feature

near the center corresponds to long-wavelength changes in the density of states associated

with random dopant fluctuations. [57] The two sharp peaks flanking the center peak at

a distance of 2π/6.8a0 along the 〈π, π〉 direction are associated with the structural b-axis

supermodulation. Finally, the two broad peaks along the 〈0, π〉 (the CuO bond direction)

correspond to the unexpected incommensurate modulation we will focus on. Analysis of

maps acquired simultaneously at different in-gap energies shows that the wavevector of the

modulated patterns changes as a function of energy (Fig. 3.10a). Similar modulations have

been seen by other STM studies performed at low temperatures (T = 4K) for a wide range

of dopings (Fig. 3.10b&c). [39, 40] The low temperature experiments see a feature along

the 〈0, π〉 direction that disperses in a fashion similar to the dispersion we have measured at

T = 40K. (Fig. 3.10a) However, these T = 4K studies also see a dispersive feature along

the 〈π, π〉 direction, and a high-resolution data set [40] has identified a total of 5 dispersive

36

Page 45: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

42mV 42mV

q

s

(0,0) <π,π>

<0,π>

Figure 3.9: (A) Spatially-resolved conductance maps (Rj = |−150mV |/75pA),were acquired simultaneously at four energies between 7− 23mV on the same380A × 380A region as Fig. 3.8a. The greyscale extends from 200pS (white)to 25pS (black). The 42mV conductance map, which contains modulationsarising only along the b-axis supermodulation, was taken on a different areameasuring 550A× 550A (Rj = | − 150mV |/40pA, greyscale map extends from36pS to 136pS). (B) FFTs were performed on the data in (A) to highlight thedifferent wavevectors contributing to the modulated patterns. The greyscaleextends from 1.88pS (black) to 0pS. Highlighted are the contribution fromthe b-axis supermodulation (S) and the contribution oriented at 45 to thisdirection (Q).

37

Page 46: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

3

4

5

6

0 20 40Energy (mV)

Wav

ele

ng

th (

a0)

A B C

7

1

45

3

6

Figure 3.10: (A) The Fourier analysis performed on the data in Figure 3.8 wasused to determine the energy-dependence of the wavevector Q running alongthe 〈0, π〉 direction. Plotted here is the wavelength of the patterns measuredby us at T = 40K (black squares) and the dispersion measured by Ref. [39] atT = 4K (grey). The wavelength is in units of a0 = 3.8A, the nearest neighborCu−Cu distance. (B) The data at T = 4K, shown here at V = 16mV , showtwo sets of strong peaks- along the 〈0, π〉 (blue) and 〈π, π〉 (red) directions-and four weaker peaks (black). (C) All the peaks seen at T = 4K were foundto disperse. Parts (B) and (C) of this figure were adapted from Ref. [40].

features in the power spectrum of the LDOS not seen at T = 40K (Fig. 3.10d). The features

seen at T = 4K have been proposed to result from a simple elastic scattering interference

model. [43, 39, 40] We will now describe this model, and demonstrate how it applies to our

data at T = 40K.

Electron waves elastically scattering off defects lead to periodic modulations in the density

of states. A well-studied example of this is the modulated pattern found near step edges

and point defects on certain noble metal surfaces, as discussed in Chapter 2. [58, 59, 60,

61, 62] Recall that the set of wavevectors ~q where there is some correction to the density of

states arising from elastic scattering interference can be determined very simply using toy

models. The easiest one to illustrate for the case of Bi2Sr2CaCu2O8+δ is the particle-based

toy model. Quasiparticles exist on curves of constant electron energy in ~k-space, which,

in the superconducting state of Bi2Sr2CaCu2O8+δ, are a set of “banana”-shaped curves

38

Page 47: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

1

5

7

3

4

6 (0,π)

(π,π)

Figure 3.11: This is a model representation of the curves of constant elec-tron energy in ~k-space in the superconducting state of Bi2Sr2CaCu2O8+δ atω = ∆/3 (red) and ω = 2∆/3 (blue). Also shown are the 6 wavevectors ~qthat connect points on these curves where Ref. [40] sees peaks in the Fouriertransform of the STM density of states.

centered around the 〈π, π〉 direction in the first Brillouin zone (Fig. 3.11a). To determine

the wavevectors ~q where we expect modulations arising from elastic scattering, we find the set

of wavevectors ~q which connect points on these curves in ~k-space. We would naively expect

that the wavevectors ~q joining the points on these curves with the highest joint density of

states would have the largest contribution to elastic scattering. Since the total density of

states at a given energy E is given by∫

ε(~k)=Ed~k

|5ε(~k)|, we expect the points on the curves where

the integrand is largest to contribute the most. Thus the largest contribution from elastic

scattering should arise at the wavevectors connecting the ends of the “bananas”. [39] Indeed,

the 6 unique vectors predicted by this method are the same ones where high-resolution STM

measurements at 4K have seen density of states modulations. [40] However, the toy model

cannot simply describe why only one of these 6 remains in our measurements at 40K.

To more accurately determine the density of states correction arising from elastic scatter-

ing interference, we have adopted the perturbative treatment of Ref. [63]. A more complete

explanation appears in Appendix B. In general, many-body problems in condensed matter

39

Page 48: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

physics are often solved by finding the Green function that solves a Hamiltonian equation

of motion. We take the total Green function G(~r, ω) near an impurity to be the sum of

the Green function over pristine regions G0(~r, ω), plus some correction due to scattering

off the impurity. To leading order, the correction to the Green function arising from elas-

tic scattering takes the form∫

d2~r1G0(~r − ~r1, ω)V (~r1, ω)G0(~r1 − ~r, ω). This is the limit of

weak scattering, or Born Scattering (Fig. 3.12). The density of states in a superconduc-

tor is n(~r, ω) = − 1πImG(~r, ω) − F (~r, ω), where G is the single-particle Green function

and F is the anamolous Green function that contains the coherence factors responsible for

superconductivity. Elastic scattering interference thus introduces a correction

δn(~r, ω) = − 1

πIm

d2~r1G0(~r − ~r1, ω)V (~r1, ω)G0(~r1 − ~r, ω)

±F0(~r − ~r1, ω)V (~r1, ω)F0(~r1 − ~r, ω) (3.1)

to the density of states, where G0 is the single particle Green function, F0 is the anamolous

Green function (F0 = 0 in the absence of particle-hole correlations), V is a weak, finite-

range scattering potential, and the minus (plus) sign corresponds to potential (magnetic)

scattering. [46, 43, 44, 64, 45, 65] The Fourier transform of the Born correction δn(~q, ω) =

− 1πImV (~q, ω)Λ(~q, ω) separates into a part

Λ(~q, ω) =∫

d2~kG0(~k, ω)G0(~k + ~q, ω) ± F0(~k, ω)F0(~k + ~q, ω) (3.2)

that contains all the wave interference information, and a part

V (~q, ω) =∫

d2~xe−i~q·~xV (~x, ω) (3.3)

that acts like a static structure factor. [44] Assuming that the scattering potential is a single

point impurity in real-space, the structure factor acts like an all-pass filter, and the density

of states correction depends only on the wave interference term Λ. Following previous works,

[43, 44, 64, 45] we model the superconducting state with the Green functions,

G0(~k, ω) =ω + iδ + ε~k

(ω + iδ − ε~k)(ω + iδ + ε~k) − ∆2~k

40

Page 49: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B C

Figure 3.12: (A) This is the diagram corresponding to (weak) Born scattering.(B and C) These types of diagrams have been neglected in the treatmentpresented.

F0(~k, ω) =∆~k

(ω + iδ − ε~k)(ω + iδ + ε~k) − ∆2~k

(3.4)

where ε~k = 120.5mV − 595.1mV × (cos kx + cos ky)/2 + 163.6mV × cos kx cos ky − 51.9mV ×

(cos 2kx+cos 2ky)/2−111.7mV ×(cos 2kx cos ky +cos kx cos 2ky)/2+51mV ×cos 2kx cos 2ky is

the Bi2Sr2CaCu2O8+x band structure from ARPES [66] for slightly underdoped (x = .12)

samples, ∆~k = 45.0 × (cos kx − cos ky)/2 is the superconducting gap function, and δ is a

broadening term.

The density of states corrections arising from scattering interference provide a promising

explanation for the modulations seen in the STM data at both T = 4K and T = 40K. As

shown in Figures 3.13a&b, the formalism yields density of states corrections with sharp peaks

in ~q-space along the 〈π, π〉 (〈0, π〉) direction for potential (magnetic) scattering. The presence

of the sharp peaks is thus understood to be a consequence of the anamolous Green function,

arising from the coherence factors for the superconducting state. [46] By comparison, the

STM studies performed at T = 4K [39, 40] contain peaks along both the 〈π, π〉 and 〈0, π〉

directions (Fig. 3.13c), implying that a combination of magnetic and potential scattering

are required to reproduce the observed data. In total, the calculations observe all 6 features

identified in the high resolution data taken by Ref. [40]. At T = 40K, however, the STM data

(Fig. 3.13f) [38] shows peaks only along the 〈0, π〉 directions. Scattering interference patterns

calculated with δ = 7mV , appropriate for the experimental conditions under which the data

41

Page 50: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

q=(0,0) (0,π)

(π,π)

A B DC

Figure 3.13: We calculated the power spectrum of δn(~q, ω) in the first Brillouinzone for the superconducting state. Dark regions correspond to a larger densityof states correction than lighter regions. The power spectrum of δn(~q, ω =16mV ) in the first Brillouin zone shows big peaks (denoted by black arrows)along the 〈π, π〉 directions for pure potential scattering (A), and along the〈0, π〉 directions for magnetic scattering (B). In addition, the four additionalweaker peaks seen in the STM data at 4K are identified here by the blackcircles. These patterns were calculated using a broadening of δ = 1mV . Thecalculation was repeated for an energy broadening of δ = 7mV , and is shownhere in (C) for pure potential scattering and in (D) for magnetic scattering.The only feature which remains strong is the 〈0, π〉 feature, denoted by theblack arrow.

was measured, demonstrate that the other peaks have washed out at this level of energy

broadening, while the 〈0, π〉 mode remains (Fig. 3.13d). The reason is relatively simple- two

of the modes, one along the 〈0, π〉 direction and the other along the 〈π, π〉 direction, have

a significantly larger amplitude than the others. The 〈π, π〉 mode gets washed out because

it disperses faster than the 〈0, π〉 mode, meaning that a smaller level of energy broadening

leads to a more significant broadening in ~q-space for the former in comparison to the latter.

The most distinctive characteristic of the peaks in the STM data at T = 4K and T =

40K, and the calculations in the superconducting state, is that their wavelength changes as

a function of energy. Because the peak along the 〈0, π〉 direction appears at both T = 4K

and T = 40K, we focus on its dispersion here. As can be seen in Figure 3.14a, the calculated

dispersion agrees with the measured dispersion for higher energies (|ω| > 15mV ). The

dispersion can be understood on qualitative grounds by returning to the particle-based toy

model discussed earlier (Fig. 3.14b). Quasiparticles of a given energy exist on banana-

42

Page 51: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

shaped curves of constant electron energy in the superconducting state. The characteristic

wavevector for scattering interference are the ones joining the tips of these “bananas”. At

different energies, the shape of these “bananas” changes predominantly because of the pair

potential, as changes in the band structure tend to be gradual over small energy ranges.

As a result, the wavevector for scattering interference can be used to map the amplitude

of the pair potential. Indeed, the dispersion of the peaks seen in the data at T = 4K has

been shown previously to be consistent with the amplitude of the d-wave pair potential (Fig.

3.14c). [39] However, the disagreement between the data and the scattering interference

model at low energies (|ω| < 15mV ) is significant. The size of the scattering wavevector at

the Fermi energy is essentially independent of the details of the scattering, as the available

phase space for scattering is restricted to the nodal points of the Fermi surface, where the

energy gap has zero amplitude. This is exactly where the data and the scattering interference

model disagree on the size of the scattering wavevector the most. It has been suggested in

the literature [41, 67, 68] that the disagreement stems from the fact that the modulated

patterns no longer disperse at low energies, and that this lack of dispersion represents some

form of ordering. In this scenario, the dispersion seen at higher energies, or along different

directions, is due to scattering off this order. Although more work is needed to interpret the

almost non-dispersing low-energy data, it now appears clear that the dispersion seen in the

superconducting state data arises from elastic scattering interference.

Here, we have presented a case for quasiparticle scattering being consistent with the

periodic modulations seen in real-space maps of the density of states by STM in the super-

conducting state. The energy dependence of the wavelength of these modulations appears

to be sensitive to the amplitude of the d-wave pair potential for the HTS. Although the

data taken by us at T = 40K contains defects, as evidenced by the low-energy resonances

in the data of Figure 3.9, the modulated patterns in the density of states do not appear to

be pinned by these defects. Moreover, data taken by other groups at T = 4K show these

43

Page 52: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

B

2

3

4

5

6

-40 -20 0 20 40Energy (mV)

Wav

ele

ng

th (

a0)

A C

k=(0,0) (0,π)

(π,π)

q

Figure 3.14: (A) The dispersion for the calculated 〈0, π〉 mode (solid line)compares favorably with the dispersion measured for this feature at T = 40K(squares, from Ref. [38]) and T = 4K (triangles, diamonds and circles, fromRef. [39]) for energies ω > 15mV . (B) Here we show the curves of constantelectron energy for Bi2Sr2CaCu2O8+δ in the superconducting state at ω =∆/3 (blue) and ω = 2∆/3 (red). The characteristic wavevector for scatteringat the two energies along the 〈0, π〉 direction is also shown. (C) This dispersionhas been used by Ref. [40] to reconstruct the angular dependence of the energygap, and is shown to be consistent with a d-wave form (Figure from Ref. [40]).

modulated patterns, but show no obvious scattering site. We suggest this might result from

the scattering potential, which causes these modulations, being either very weak or extended

in space. [69] True bound states, which would produce a discernable resonance in the LDOS

maps, are expected only in the unitary, or strong, scattering limit. [63] Conversely, weak

scattering might produce modulated LDOS patterns without producing any in-gap resonance

feature.

3.4 Applications

Although the experiments in this chapter do not lead to a new understanding of the su-

perconducting state of the HTS, they present us with a set of tools that will be useful in

examining parts of the phase diagram where the electronic state is not well-understood. In

particular, examining the density of states close to scattering sites has been demonstrated

to sensitively probe the electronic state of the system.

44

Page 53: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

The ability to detect Andreev bound states locally has practical value as a phase-sensitive

test for d-wave, BCS-like superconducting quasiparticles. [25] Here we have shown that An-

dreev bound states should be bound to 110 step edges on a sample that was known to have

d-wave superconducting order. In many situations throughout the cuprate phase diagram,

it is unclear whether local, fluctuating superconducting correlations are present, even when

global superconducting order is not. For example, local, fluctuating superconductivity has

been postulated as an explanation for the so-called pseudogap in the density of states for

underdoped samples at T > TC , the subject of the next chapter. In addition, the electronic

phase of the subject of the last chapter of this thesis, a bare CuO2 plane at the surface of a

HTS, is undeterminable from a simple examination of the density of states. In both cases,

a careful search for a zero bias conductance peak in the density of states near a 110 step

edge could be used to establish whether d-wave superconductivity is present.

The scattering interference scenario can be used to characterize any modulated pattern

seen in the density of states by STM. Here, we have shown that modulated patterns in

the LDOS measured in a superconducting sample disperse. This dispersion arises from

weak scattering interference, which probes the amplitude of the d-wave pair potential in the

superconducting state. In many situations, such as the pseudogap regime of the HTS, the

electronic state of the system is not known. Comparison of the modulated patterns measured

in STM data with calculations using the candidate Green functions for an electronic state

can help determine which of the candidates is consistent with the data. [46] Candidate Green

functions often differ only by correlations that subtly affect the density of states, but produce

completely different scattering interference patterns. In the next chapter of this thesis, we will

discover that STM experiments see LDOS modulations in the enigmatic pseudogap regime

of the HTS, and will attempt to compare the Green functions of various proposed electronic

states with the measured patterns using the scattering interference formalism developed in

the superconducting state.

45

Page 54: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

3.5 References

[1] D. V. Harlingen, Rev. Mod. Phys. 67, 515 (1995).

[2] C. Tsuei and J. Kirtley, Rev. Mod. Phys. 72, 969 (2000).

[3] S. Kashiwaya and Y. Tanaka, Rep. Prog. Phys. 63, 1641 (2000).

[4] D. Markowitz and L. Kadanoff, Phys. Rev. 131, 563 (1963).

[5] P. Lee, Phys. Rev. Lett. 71, 1887 (1993).

[6] P. Hirschfeld and N. Goldenfeld, Phys. Rev. B 48, 4219 (1993).

[7] A. Balatsky, M. Salkola, and A. Rosengren, Phys. Rev. B 51, 15547 (1995).

[8] M. Salkola, A. Balatsky, and D. Scalapino, Phys. Rev. Lett. 77, 1841 (1996).

[9] M. Flatte and J. Byers, Phys. Rev. Lett. 80, 4546 (1998).

[10] L. Buchholtz and G. Zwicknagl, Phys. Rev. B 23, 5788 (1981).

[11] L. Buchholtz, M. Palumbo, D. Rainer, and J. Sauls, J. Low Temp. Phys. 101, 1099

(1995).

[12] C.-R. Hu, Phys. Rev. Lett. 72, 1526 (1994).

[13] Y. Tanaka and S. Kashiwaya, Phys. Rev. Lett. 74, 3451 (1995).

[14] V. Ambegoekar, P. de Gennes, and D. Rainer, Phys. Rev. A 9, 2676 (1974).

[15] J. Hara and K. Nagai, Prog. Theor. Phys. 74, 1237 (1986).

[16] M. Covington et al., Phys. Rev. Lett. 79, 277 (1997).

[17] L. Alff et al., Phys. Rev. B 55, 14757 (1997).

46

Page 55: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[18] L. Alff et al., Phys. Rev. B 58, 11197 (1998).

[19] J. Wei, N.-C. Yeh, D. Garrigus, and M. Stasik, Phys. Rev. Lett. 81, 2542 (1998).

[20] M. Aprili, E. Badica, and L. Greene, Phys. Rev. Lett. 83, 4630 (1999).

[21] R. Krupke and G. Deutscher, Phys. Rev. Lett. 83, 4634 (1999).

[22] Y. Dagan and G. Deutscher, Phys. Rev. Lett. 87, 177004 (2001).

[23] A. Sharoni, G. Koren, and O. Millo, Europhys. Lett. 54, 675 (2001).

[24] A. Sharoni et al., Phys. Rev. B 65, 134526 (2002).

[25] H. Aubin, L. Greene, S. Jian, and D. Hinks, Phys. Rev. Lett. 89, 177001 (2002).

[26] M. Aprili et al., Phys. Rev. B 57, 8139 (1998).

[27] A. Poenicke, Y. Barash, C. Bruder, and V. Istyukov, Phys. Rev. B 59, 7102 (1999).

[28] A. Yazdani et al., Phys. Rev. Lett. 83, 176 (1999).

[29] E. Hudson et al., Science 285, 88 (1999).

[30] S. Pan et al., Nature 403, 746 (2000).

[31] E. Hudson et al., Physica B 329, 1365 (2003).

[32] E. Hudson et al., Nature 411, 920 (2001).

[33] I. Adagideli, P. Goldbart, A. Shnirman, and A. Yazdani, Phys. Rev. Lett. 83, 5571

(1999).

[34] S. Misra et al., Phys. Rev. B 66, 100510 (2002).

[35] W. Belzig, C. Bruder, and M. Sigrist, Phys. Rev. Lett. 80, 4285 (1998).

47

Page 56: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[36] D. Feder, A. Beardsall, A. Berlinksy, and C. Kallin, Phys. Rev. B 56, 5751 (1997).

[37] J.-X. Zhu et al., Phys. Rev. Lett. 91, 57004 (2003).

[38] M. Vershinin et al., Science 303, 1995 (2004).

[39] J. Hoffman et al., Science 297, 1148 (2002).

[40] K. McElroy et al., Nature 422, 592 (2003).

[41] C. Howald et al., Phys. Rev. B 67, 14533 (2003).

[42] J. Byers, M. Flatte, and D. Scalapino, Phys. Rev. Lett. 71, 3363 (1993).

[43] Q.-H. Wang and D.-H. Lee, Phys. Rev. B 67, 20511 (2003).

[44] L. Capriotti, D. Scalapino, and R. Sedgewick, Phys. Rev. B 68, 14508 (2003).

[45] L. Zhu, W. Atkinson, and P. Hirschfeld, Phys. Rev. B 69, 60503 (2004).

[46] T. Pereg-Barnea and M. Franz, Phys. Rev. B 68, 180506 (2003).

[47] S. Misra, M. Vershinin, P. Phillips, and A. Yazdani, Phys. Rev. B in press (2004).

[48] P. Lindberg et al., Phys. Rev. B 39, 2890 (1989).

[49] C. Renner and O. Fischer, Phys. Rev. B 51, 9208 (1995).

[50] H. Heinrich, G. Kostorz, B. Heeb, and L. Gauckler, Physica C 224, 133 (1994).

[51] B. Wells et al., Phys. Rev. B 46, 11830 (1992).

[52] J. Kirtley and D. Scalapino, Phys. Rev. Lett. 65, 798 (1990).

[53] C. Varma et al., Phys. Rev. Lett. 63, 1996 (1989).

[54] K. Ahn, T. Lookman, A. Saxena, and A. Bishop, Phys. Rev. B 68, 92101 (2003).

48

Page 57: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[55] M. Fogelstrom, D. Dainer, and J. Sauls, Phys. Rev. Lett. 79, 281 (1997).

[56] M. Zhitomirsky and M. Walker, Phys. Rev. Lett. 79, 1734 (1997).

[57] S. Pan et al., Nature 413, 282 (2001).

[58] Y. Hasegawa and P. Avouris, Phys. Rev. Lett. 71, 1071 (1993).

[59] M. Crommie, C. Lutz, and D. Eigler, Nature 363, 524 (1993).

[60] L. Burgi, O. Jeandupeux, H. Brune, and K. Kern, Phys. Rev. Lett. 82, 4516 (1999).

[61] M. van der Wielen, A. can Roij, and H. van Kempen, Phys. Rev. Lett. 76, 1075 (1996).

[62] L. Peterson, P. Laitenberger, E. Laegsgaard, and F. Besenbacher, Phys. Rev. B 58,

7361 (1998).

[63] M. Salkola, A. Balatsky, and D. Scalapino, Phys. Rev. Lett. 77, 1841 (1996).

[64] A. Polkovnikov, M. Vojta, and S. Sachdev, Phys. Rev. B 65, 220509 (2002).

[65] C.-T. Chen and N.-C. Yeh, Phys. Rev. B 68, 220505 (2003).

[66] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 52, 615 (1995).

[67] S. Kivelson et al., Rev. Mod. Phys. 75, 1201 (2003).

[68] K. McElroy et al., cond-mat/0404005 (unpublished).

[69] D. Scalapino, T. Nunner, and P. Hirschfeld, cond-mat/0409204 (unpublished).

49

Page 58: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 4

Local Ordering in the Pseudogap

Regime

4.1 Introduction

No other issue in the physics of the HTS exposes our lack of understanding of the microscopic

mechanisms underlying the entire phase diagram more than the so-called pseudogap regime

(Fig. 4.1a). This part of the phase diagram gets its name from a depression in the density

of states at the Fermi energy, which slowly fills in upon increasing temperature through an

ill-defined T ∗ for underdoped samples (Fig. 4.1b). Because this depression in the density

of states is symmetric about the Fermi energy, it intuitively appears to result from some

correlation effect. Whether this correlation originates from residual antiferromagnetism,

remnant superconductivity, or some other source remains a crucial open question. The

answer not only has bearing on the origin of the pseudogap, but also has the potential to

alter our view of the doped insulator, the superconducting state, and the whole cuprate

phase diagram itself.

The pseudogap in the density of states in the underdoped HTS above TC strongly in-

50

Page 59: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 4.1: (A) The doping-temperature phase diagram of the cuprates con-tains a superconducting dome centered at a doping of x = 0.16, where TC

reaches its maximum value (≈ 95K in Bi2Sr2CaCu2O8+δ). A symmetric de-pletion of electronic states at the Fermi energy called the pseudogap (B) isseen for underdoped samples (x ≤ 0.16) above TC . As temperature is raisedtowards T ∗, the pseudogap in the density of states diminishes, as states fill inthe gap. Although it remains unclear where exactly the T ∗ line lies, both thepseudogap and T ∗ increase monotonically with underdoping (from Ref. [1]).The only feature in the density of states which disappears at TC are the sharppeaks flanking the energy gap.

fluences any measurement that probes quasiparticle degrees of freedom. The most direct

measurement of gapped electronic states is perhaps provided by angle-resolved photoemis-

sion spectroscopy (ARPES), which measures the occupied part of the density of states as

a function of energy and momentum. For optimally doped samples at low temperatures,

ARPES sees that the density of states is gapped away from the Fermi energy with a d-wave

angular dependence. [2] For overdoped samples, the size of the energy gap scales with TC at

low temperatures, and the energy gap disappears above TC . This behavior is consistent with

the view that it is a manifestation of the amplitude of the pair potential, and that increasing

temperature destroys superconductivity by breaking Cooper pairs apart. Contrary to ex-

pectation, for underdoped samples, the energy gap continues to grow while TC is shrinking

51

Page 60: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

(Fig. 4.2a). [3, 4] Even more mysteriously, as the temperature is raised through TC , ARPES

finds that the energy gap near the (0, π) points persists- the so-called pseudogap (Fig. 4.2b).

[5, 6] The size of the pseudogap decreases linearly with increasing temperture or hole con-

centration. Meanwhile, the locus of momenta with states at the Fermi energy widens from a

single point in the superconducting state, into a so-called Fermi arc in the pseudogap regime

((Fig. 4.2c)). [7, 8] As temperature is further raised, this arc widens to meet the zone edge,

and the pseudogap is destroyed. Upon including the observation that the quasiparticle states

in the pseudogap regime become ill-defined in energy and momentum [9], the ARPES data

appears to be a key ingredient in explaining the properties of underdoped HTS above TC , as

it can be phenomenologically tied to their thermal, electronic [10], and optical [11, 12, 13]

properties.

The underlying cause of the pseudogap in the density of states, however, remains elusive.

Complicating matters is the fact that the electronic state of the HTS in the pseudogap regime

may not even contain quasiparticles. The well-defined features measured by photoemission

as a function of energy and momentum in the superconducting state have been replaced

with very broad features in the pseudogap regime (Fig. 4.3). The ill-defined nature of the

electronic states in the pseudogap regime suggests that a fundamentally different picture

from that used to describe conventional metals is required.

The basic ideas underlying the current debate actually predate the discovery of the pseu-

dogap in the density of states. The first experiment to detect an abnormal signal in under-

doped samples above TC was nuclear magnetic resonance (NMR). [15, 16] In these exper-

iments, the nuclear spins of a particular species were first perturbed by a magnetic pulse,

and then the relaxation rates of the excited spins were measured. The relaxation rate of

Cu spins in overdoped samples drops from finite values above the superconducting tran-

sition, to near zero in the superconducting state. For underdoped samples, on the other

hand, there is a gradual decrease of the relaxation rate starting at temperatures well above

52

Page 61: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

C

Figure 4.2: (A) In the superconducting state, ARPES measures an energy gapwhose maximum value increases while TC decreases for underdoped samples(from Ref. [3]). (B) As underdoped samples are warmed up through TC , theenergy gap persists to higher temperatures. Shown is the density of statesalong the gap maximum (〈0, π〉) direction as a function of energy (from Ref.[6]). (C) This is a schematic diagram of the Fermi surface measured at differenttemperatures in an underdoped sample. Here, Γ − M refers to the 〈0, π〉direction and Γ−X and Γ− Y refer to the 〈π, π〉 directions. The left panel isfor T < TC , the middle panel is between TC and T ∗, and the right panel is forT > T ∗ (from Ref. [7]).

53

Page 62: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A

A

B

B

Figure 4.3: Shown is the photoemission intensity as a function of electronenergy on the Fermi surface along the 〈π, π〉 direction (A) and the 〈0, π〉 direc-tion (B). In the superconducting state of a slightly underdoped sample (top),angle-resolved photoemission measures sharp quasiparticle features. In con-trast, above TC (bottom), the peak-like features have become broad enoughthat their position is now comparable to their width. (Figures adapted fromRef. [14]).

54

Page 63: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

the superconducting transition. The discovery of this effect led to the claim that the effect

was due to some form of precursor superconductivity [15]. Almost immediately afterwards,

another group claimed the Cu relaxation rates are enhanced by local fluctuating spin order

[16]. Either phenomenon could produce a pseudogap in the density of states. The pseudo-

gap could be a signature of Cooper pairing occuring above TC , with the pairs not becoming

phase coherent until the temperature is decreased below TC . Pseudogap-like depressions in

the density of states can also result from various forms of ordering. Since the initial NMR

experiments, compelling experimental evidence has suggested that both these phenomena

are present above TC for underdoped samples. Simultaneously, various theoretical proposals

have been raised to explain the abnormal behavior of the underdoped HTS above TC .

4.1.1 Precursor Superconductivity Experiments

Although neither zero resistivity nor the Meissner effect have ever been seen in HTS samples

in the pseudogap state, other probes indicate that some fluctuating form of superconductivity

might be present in the pseudogap regime. In Nernst effect measurements, a magnetic field

B is applied along the z direction, a thermal gradient −5 T along the x direction, and the

electric field E develops along the y direction is measured as a function of temperature and

doping (Fig. 4.4a). [17] Normal metals have a very small Nernst voltage, originating from the

motion of charged carriers. The thermal gradient causes a diffusion of electrons from the hot

end of the sample to the cold end. Because there can be no net current flow, this sets up an

electric field that causes a perfectly matching counterflow of electrons in the other direction.

Application of a vertical magnetic field causes a portion of these currents to deflect sideways.

These sideways currents almost cancel, leading to a small voltage being established to ensure

no net current flow sideways. In contrast, in type-II superconducting samples, like the HTS,

an applied magnetic field generates vortices, regions of normal material that thread magnetic

flux and carry no charge. If the vortices are mobile, the thermal gradient makes them move

55

Page 64: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

from the hot end of the sample to the cold end. The motion of these vortices generates a

relatively large electric field perpendicular to both the direction of vortex motion and applied

magnetic field from the Josephson effect (Fig. 4.4b). The data indicates that the Nernst

signal is very large in the superconducting state, and drops to a value one hundreth of its

value in the superconducting state at high temperatures. However, its value does not drop

off steeply at TC for underdoped samples- it remains anomalously large into the pseudogap

regime (Fig. 4.4c). The authors interpret their results as meaning that vortices, and hence

fluctuating superconductivity, exist above TC for underdoped samples even though the entire

sample is not superconducting. Although the temperature above which the Nernst signal

drops below a certain level does not track T ∗ throughout the phase diagram [18], fluctuating

superconductivity appears to exist even outside the superconducting dome for underdoped

samples.

Optical conductivity measurements have been used to measure the time scale for this

fluctuating superconductivity. A fast probe might be required to see properties of a rapidly

fluctuating superconductor, which might be inaccessable to DC probes like resistivity or

magnetization. [19] The frequency dependent conductivity of a superconductor σ(ω) =

σQkBTΘ(T )/ω, where TΘ is a measure of the stiffness of the phase of the order parameter,

σQ = e2/d, and d is the distance between superconducting planes. In the pseudogap state,

optical conductivity should approach the DC conductivity at low frequencies, and if fluc-

tuating superconductivity is present, the superconducting form at high frequencies. The

frequency where one behavior crosses over into the other one, Ω, is a measure of the time

scale for superconducting fluctuations (Fig. 4.5a). Upon normalizing the frequency to Ω, the

normalized conductivity |σ| = σQkBTΘ(T )/Ω(T ) was seen to collapse to a universal curve for

all frequencies and at temperatures both above and below the superconducting transition.

From analyzing a set of conductivity curves taken as a function of temperature at different

frequencies, both the phase stiffness energy Θ and the lifetime of the superconducting fluc-

56

Page 65: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

∇ΤΕ

Β

V

∇Τ

Β

V

A C

B

e

Figure 4.4: This is a schematic diagram (top view) of a Nernst effect experi-ment in a normal metal (A) and a superconductor (B). The transverse voltageis small for (A), but can be comparitively large for (B). (C) The temperaturedependence of the Nernst signal ν = Ey/Bz|5T | minus the background signalfrom the motion of charged carriers at elevated temperatures is shown here fora wide range of dopings. The arrows indicate the TC for the given sample. Forthe overdoped sample (x = 0.17), the Nernst sigal becomes undetectable atTC , while for underdoped samples (0.06 < x < 0.12) the Nernst signal persistswell above TC (Figures from Ref. [17]).

57

Page 66: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 4.5: (A) The optical conductivity of a HTS has a universal scalingbehavior both above and below the superconducting transition (TC = 74K forthis sample). (B) From this scaling analysis, the temperature dependent phasestiffness energy TΘ and the lifetime for superconducting fluctuations τ = 1/Ωcan be determined as a function of temperature. Both these quantities remainfinite for temperatures above TC = 74K, but vanish below the temperatureT ∗ ≈ 170K (Figures adapted from Ref. [19]).

tuations τ = 1/Ω can be extracted as a function of temperature (Fig. 4.5b). For underdoped

samples, the phase stiffness energy does not abruptly fall to zero above the superconduct-

ing transition, but rather persists to higher temperatures. Furthermore, the time scale for

these superconducting fluctuations is small, but detectable, to similarly elevated tempera-

tures. This seems to support the idea that superconductivity exists on short time scales well

into the pseudogap regime, although once again, the characteristic temperature where these

effects disappear was identified to be smaller than T ∗.

4.1.2 Spin Order Experiments

Neutron scattering experiments have been used to characterize the spin structure of the

HTS throughout the phase diagram, and see remnants of antiferromagnetism above TC for

underdoped samples. Neutron scattering measures the energy shift and wavevector change

of a neutron that scatters off the atomic scale spin structure of a sample, which encodes

the momentum and energy of spin excitations in the material. The elastic signal (no energy

58

Page 67: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

BA C D

Figure 4.6: (A) This is a schematic diagram of the peaks in ~k-space measuredby elastic neutron scattering in the undoped insulator. The Bragg peaks arelabeled as squares, and the peaks resulting from antiferomagnetic alignmentare labeled as circles. (B) This is the real-space schematic diagram correspond-

ing to the ~k-space picture in (A). Here, each bright spot corresponds to a Cusite, with red sites being spin up, and blue sites being spin down. (C) This

is a ~k-space diagram of the inelastic spectrum at higher energies. (D) This is

the real-space picture corresponding to the ~k-space diagram in (C). Here, redsites correspond to up spins, and blue sites correspond to up spins.

transfer) encodes the spin configuration of the ground state. In the undoped antiferromag-

netic insulator, neutron scattering sees the Bragg peaks for the Cu sites, which have localized

spins on them. A second set of peaks, at ((2n + 1)π, (2m + 1)π), arises from the breaking

of the square lattice symmetry due to the antiferromagnetic alignment of these spins (Fig.

4.6a&b). [20, 21, 22] In the inelastic spectrum, the peaks centered at (π, π) split, with in-

creased splitting at higher energies (Fig. 4.6c&d). [23, 24] This corresponds to a real-space

modulation of the spins when the neutron transfers energy to the spin lattice- the excita-

tion of spin waves on the antiferromagnetic ground state. For underdoped samples, neutron

scattering sees remnants of these features above TC in both the elastic and inelastic channels.

Data from one family of the HTS compounds, the Lanthanum cuprates, shows evidence

for ordered spins in the ground state for underdoped samples above TC . In this part of

the phase diagram, elastic neutron scattering shows that the elastic peaks present in the

undoped antiferromagnetic insulator at (π, π) are split along the nearest neighbor direction

59

Page 68: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

δ(r.l.u.)

A

B C

Figure 4.7: (A) This is a schematic diagram of the elastic neutron scatteringspectrum on the Lanthanum compounds. (B) This is the real-space picture

complementary to the ~k-space diagram in (A). Here, black sites correspondto down spins, white sites correspond to up spins, and the grey regions inbetween are spinless. (C) The splitting of the peaks centered at (π, π) scaleslinearly with doping for underdoped samples. This corresponds to a decreasingdistance between the antiferromagnetic domains in (B) as holes are doped intothe system (Figure adapted from [25]).

(Fig. 4.6a,b). [30, 28, 25] This corresponds to a modulation in real space of antiferromagnetic

domains, reminscent of a spin wave, in the ground state of these compounds. Moreover, the

splitting of the peaks increases linearly with doping for underdoped samples, suggesting

that the size of the antiferromagnetic domains shrinks as holes are added (Fig. 4.6c). [25]

However, although the amplitude of the elastic signal disappears for overdoped samples above

TC [31], there is no correlation between the elastic signal measured by neutron scattering

and the pseudogap temperature scale T ∗. [32] More importantly, no elastic signal has yet

been detected in any of the non-Lanthanum HTS at superconducting dopings.

Inelastic neutron scattering data from all the HTS compounds shows evidence for residual

antiferromagnetic correlations above TC for underdoped samples. In the undoped insulator,

the elastic peaks at ((2n+1)π, (2m+1)π), characteristic of having an antiferromagnetically

aligned lattice, split at higher energies. [23, 24] Upon increasing doping to superconducting

dopings, the peaks at ((2n + 1)π, (2m + 1)π) move to a higher energy Er that scales with

TC . [33, 34] In the superconducting state, at energies both above and below Er, the peak

splits in a fashion similar to the spin wave in the undoped antiferromagnet (Fig. 4.8a).

60

Page 69: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

δ(r.l.u.)

δ(r.l.u.)

hω(meV)

CA B

Figure 4.8: (A) This is the dispersion of the mode centered at (π, π) mea-sured by inelastic neutron scattering on heavily underdoped Y Ba2Cu3O6+x

below TC (Figure adapted from Ref. [26]). (B) This is the dispersion ofthe same mode, measured on a nearly optimally-doped Y Ba2Cu3O6+x aboveTC (Figure adapted from Ref. [27]). Allthough there is a peak in the data,the features have become too broad to identify a dispersion. (C) This is thedispersion of the mode centered at (π, π) measured by inelastic neutron scat-tering on La2−xBaxCuO4 above TC (Figure adapted from Ref. [28]). In otherLanthanum-based compounds, there is no appreciable change in dispersionabove and below TC . [29]

61

Page 70: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[29, 27, 26] In the pseudogap state, the features near the antiferromagnetic points persist,

[28] although their dispersing nature becomes washed out in some materials (Fig. 4.8b &

c). [27] Because the spin ground state is not known in general, as there is no elastic neutron

scattering signal in the non-Lanthanum-based compounds, the inelastic neutron scattering

data cannot be directly interpreted as proof of spin waves, but can be interpreted as general

evidence for fluctuating antiferromagnetic spin correlations. However, no clear connection

has as yet been made between the inelastic neutron data and the pseudogap temperature or

energy scales in an experiment.

4.1.3 Theoretical Ideas

Innumerable theoretical attempts have been made to understand the pseudogap in the den-

sity of states, and, more broadly, the behavior of the HTS in the pseudogap regime. At-

tempting to segregate them into well-defined groups is a somewhat artificial procedure, as

the theories often attempt to explain both the experimental data supporting precursor su-

perconductivity and fluctuating spin order, and tend to borrow ideas from one another.

However, for the purposes of discussion, they can be roughly divided into two categories:

those using the superconducting state as their starting point, and those using exotic ordered

states as their starting point.

One set of theories ascribes the pseudogap in the density of states to preformed Cooper

pairs existing above the temperature (TC) where they become phase-coherent. Such behavior

is not unusual for planar materials. True off-diagonal long-range order should, thermody-

namically, only occur at T = 0 in two- dimensional systems. Global phase transitions in

quasi-two dimensional materials occur when thermal fluctuations become as large as (often

hidden) interactions that make the system not perfectly two dimensional, like out-of-plane

interactions. Meanwhile, local correlations can have significantly higher energy scales associ-

ated with them because they are associated with in-plane interactions. In the HTS, however,

62

Page 71: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

both the phase stiffness energy and the anomalously large Nernst signal disappear at temper-

atures below where the pseudogap in the density of states does. [18, 19] Still, the pseudogap

could well be tied to a pairing energy scale, the Nernst and optical conductivity temper-

atures to the establishment of fluctuating (in-plane) phase coherence between these pairs,

and the superconducting transition to the global phase coherence of these fluctuating super-

conducting patches. The precursor superconductivity scenario for the HTS is qualitatively

analogous to the behavior of the antiferromagnetic insulator at low doping. [35, 20] Here,

three-dimensional antiferromagnetic order survives up to a temperature TN , which is associ-

ated with the in-plane magnetic anisotropy and the out-of-plane magnetic exchange energy.

Thermal fluctuations become a limiting factor for in-plane correlations only at tempertures

well above TN . Finally, at experimentally unobtainable temperatures, thermal fluctuations

reach the strength of the nearest neighbor exchange energy and the local antiferromagnetic

’pairing’ of nearest neighbor spins vanishes. However, there is as yet no solid experimental

proof that the observed pseudogap is actually connected with Cooper pairing.

Other proposals seeking to explain the pseudogap in the density of states have focused

on the possibility that some form of ordering is responsible for the measured behavior of the

underdoped cuprates above TC . Pseudogap-like depressions in the density of states associated

with charge ordering have been seen in other layered metal oxides, such as K0.9Mo6O17 [36],

and with both charge and spin order in the dichalcogenides [37].

One set of theoretical efforts proposing exotic order in the pseudogap regime treats the

microscopic details of doping an antiferromagnetic insulator into the pseudogap regime. In

the undoped insulator, which has one electron per Cu atom, Coulomb repulsion localizes

electrons on the Cu sites of the CuO2 plane. These electrons can lower their total energy

by being allowed to hop from one site to another. Because this would be prevented by the

Pauli Exclusion Principle for aligned nearest neighbor spins, the undoped insulator develops

antiferromagnetic spin alignment. (Fig. 4.9a). One proposal suggests that doping this

63

Page 72: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

antiferromagnetic lattice is best thought of as turning the lattice into a liquid of spin pairs

(Fig. 4.9b). [38] Although it is known that the undoped lattice has long-range (crystalline)

antiferromagnetic order, this order is destroyed with the removal of a very small concentration

of spins (just one in 25). However, the spin at a given site can be thought of as bound to a

nearest neighbor, forming a spin singlet. Because it can be equivalently bonded to any one

of its nearest neighbors in the antiferromagnetic state, a situation analogous to the single-

bond, double-bond resonance of the Carbon atoms in a Benzene ring, this state is called

the resonant valence bond state (RVB). The RVB scenario suggests that the destruction

of antiferromagnetic order arises from the ability of the spin pairs to slide around, as if

comprising a liquid. Theoretical works have shown that the hole-doped spin liquid state

has the remarkable property that charges and spins can be treated separately. [39] The

pseudogap would then arise from the energy cost of having to remove a real electron, which

requires breaking apart a spin pair. In addition, these spin pairs have been conjectured to be

analogous to the Cooper pairs in BCS superconductivity. [39] This is appealing as well, as

the shape of the superconducting dome in the phase diagram would then be limited by the

low number of charged carriers for underdoping (despite a growing pair potential for spins),

and by the shrinking pair potential for spin pairing with overdoping (despite a growth in the

number of holes).

Another prominent proposal, that of ordered circulating currents, has come out of mi-

croscopic ideas developed in understanding the resonant valence bond scenario. The spin

liquid has instabilities to various kinds of ordered states, [40, 41, 42] including some involving

currents circulating either between the Cu atoms [43], or between the Cu and O atoms (Fig.

4.10a). [44] An experimental hallmark of such a state is that it breaks time-reversal sym-

metry. Signatures of this broken symmetry might have been seen in a recent photoemission

experiment, although the data remains highly controversial. [45] Moreover, no microscopic

observation of circulating current order has been made. In both scenarios, the circulating

64

Page 73: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 4.9: (A) This is a schematic diagram of the antiferromagnetic latticepresent in the undoped insulator. (B) This is a schematic diagram of thespin-pairs in the resonant valence bond picture. Removal of less than one spin(grey) in 25 is enough to destroy the crystalline antiferromagnetic state, andallow the spin singlets to form a liquid. Figures (A) and (B) have been adaptedfrom Ref. [32].

current order competes with superconductivity. Thus, the suppression of TC at low dopings

arises from the increase in size of the circulating current order parameter, with the suppres-

sion of TC at high dopings arising from the decrease in the size of the order parameter for

superconducting order.

The concept of spin-charge separation, often discussed in the context of the resonant

valence bond scenario, has played a prominent role in another ordering proposal for the

pseudogap that also derives from the microscopic physics of doping an antiferromagnetic

insulator- the so-called stripe model. [47, 48] The stripe model suggests that holes added to

the parent antiferromagnet organize into one-dimensional spinless charge structures separat-

ing antiferromagnetic domains. [30] The separation between neighboring stripes, and hence

the size of the antiferromagnetic domains, is then directly related to the doping level of

the sample. This evolution of the antiferromagnetic spin structure is consistent with elastic

neutron scattering experiments on the Lanthanum family of the HTS, which reveal that the

size of antiferromagnetic domains shrinks as holes are added (Fig. 4.7). [25]

Another set of theories have examined the doped antiferromagnetic insulator from a more

global perspective, not starting from microscopic details. They suggest that the inelastic

65

Page 74: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Figure 4.10: Two proposed types of circulating current order for the HTSare shown. On the left is the so-called d-density wave state, which involvescurrents circulating between Cu sites (from Ref. [43]). On the right is a stateproposed by Varma, where currents circulate between a Cu site and the O itsbonded to (from Ref. [46]).

neutron data of Figure 4.8 indicates that fluctuating spin order survives even when the

undoped insulator is doped across the antiferromagnetic phase boundary. These theories

postulate that fluctuating spin order is responsible for the pseudogap behavior, [49, 50]

and competes with superconducting order, [51, 52] resulting in the suppression of TC for

underdoped samples.

The final proposal we will discuss here has culled together many ideas attributed to

the other models discussed here- the so-called SO(5) theory. This theory proposes that

superconducting order and antiferromagnetic order are degenerate, and can be described by

a single order parameter with five degrees of freedom- the real and imaginary parts of the

superconducting order parameter, and the three orthogonal degrees of freedom associated

with a spin vector. [53] At a high temperature scale, spins pair to form singlets, giving this

five dimensional order parameter a finite amplitude. Meanwhile, at lower temperatures, the

degeneracy between the superconducting and antiferromagnetic degrees of freedom can be

lifted by relatively small energy scales, which selects a phase for the order parameter and

determines whether the pairs have antiferromagnetic and/or superconducting order. The

antiferromagnetic state is essentially a crystalline lattice of Cooper pairs, which, when holes

66

Page 75: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

are doped into the system, melts into a superfluid of Cooper pairs at low temperatures.

The pseudogap in the density of states is then associated with the initial formation of the

pairs (the finite amplitude SO(5) order parameter, with undefined phase). This amplitude

decreases with the addition of holes, consistent with the behavior of the pseudogap in the

density of states. The shape of the superconducting dome is limited by increased fluctuations

due to the proximity of antiferromagntism for low doping, and the decrease in pair amplitude

at overdoping. Although starting from global considerations, the microscopic picture that

emerges for the pseudogap regime is that of spin singlets that increasingly freeze into a

crystalline lattice as doping decreases. [54]

To date, no experimental technique has made a definitive identification of the physics

underlying the pseudogap regime. Although the proposals outlined here have been grouped

into separate categories, many share common features. All the proposals share the common

feature that the characteristic length scales are on the order of the lattice constant. The

precursor superconductivity scenario has a characteristic length scale set by the supercon-

ducting coherence length, just 16A in-plane for the HTS. [55] The scenarios based on various

proposed exotic order have characteristic length scales on the order of the lattice constant.

Because STM can measure the electronic density of states as a function of location on these

length scales, and directly measures the pseudogap, it could be an ideal tool to study pseudo-

gap physics. In this chapter, we use STM to look for signs of electronic ordering on a sample

in the pseudogap regime for the first time. [56] We find a dispersionless, incommensurate

standing wave pattern in the spatially resolved density of states at fixed energies inside the

pseudogap. The energy-independent wavelength of these patterns indicate that they cannot

be the result of scattering interference, and suggest they are the signature of some form of

local ordering in the pseudogap regime. Finally, we review how the measured patterns relate

to some of the theoretical proposals for ordering in the pseudogap regime.

67

Page 76: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

4.2 Results

To search for signs of local electronic ordering in the pseudogap regime of the HTS, we used a

home-built variable temperature STM to map the spatial dependence of the density of states

at several energies. A slightly underdoped single crystal Bi2Sr2CaCu2O8+δ sample with 0.6%

Zn impurities (TC = 80K) was first cleaved in ultra-high vacuum at room temperature, then

thermalized in the STM at T = 100K. Topographic images of the surface (Fig. 4.11b)

show an atomic corrugation and b-axis superlattice distortion similar to that seen at low

temperatures (Fig. 3.4a). A tunneling conductance curve taken at an arbitrary position

on the sample (Fig. 4.11a) shows a pseudogap of ∆ ≈ 35 − 40mV centered at the Fermi

energy, in agreement with previous measurements made on slightly underdoped samples

above TC . [1] We will focus on the spatial distribution of electronic states at fixed energies

inside this pseudogap (Fig. 4.11c-f) taken simultaneously with the topograph in Figure

4.11b. The spatially resolved density of states at high energies shows the expected structural

features- the atomic corrugation and the b-axis superlattice distortion. In addition, there

are bright and dark patches in the local density of states, thought to originate from local

dopant inhomogeneity. [57] Upon lowering the energy below the pseudogap energy scale, a

CuO bond-oriented modulation appears, and becomes more prominent upon decreasing the

energy to the Fermi energy. Similar results have also been acquired on slightly underdoped

samples containing no Zn (Fig. 4.12).

To characterize these new modulations, we performed a two-dimensional Fourier analysis

of the conductance maps. The Fourier transformed density of states maps (Fig. 4.13a)

typically contain three features (shown schematically in Fig. 4.13b): one, a set of four peaks

along the 〈0, π〉 directions corresponding to the atomic corrugation (A); two, a set of two

peaks along the 〈π, π〉 directions corresponding to the b-axis superlattice distortion (S); and

three, a set of four sharp peaks along the 〈0, π〉 directions (the Cu − O bond direction)

68

Page 77: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

150 pS

35 pS

10 n

m

dI/dV

(nS

)

Voltage (mV)

0.3

0.1

0.2

00−150 150

A B

C D

E F

6 mV12 mV

24 mV41 mV

Figure 4.11: Underdoped Bi2Sr2CaCu2O8+δ samples with 0.6%Zn impuri-ties (TC = 80K) were investigated using an STM operating at T = 100K.(A) STM topographic data taken on a representative 450A × 195A field ofview with a tunneling current I = 40pA and a sample bias of −150mV re-semble topographic maps taken at low temperatures. (B) A representativeSTM conductance spectrum taken at a fixed position as a function of energy(Rj = | − 150mV |/40pA = 3.75GΩ) exhibits a pseudogap of ≈ 35 − 40mVcentered at the Fermi energy. Spatially resolved conductance maps were ac-quired simultaneously with the topograph in (A) at (C) 41mV, (D) 24mV,(E) 12mV, and (F) 6mV. All STM conductance spectra were taken using astandard ac lockin technique with a bias modulation of 4mVrms. Figures havebeen adapted from Ref. [56]

69

Page 78: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B C

Figure 4.12: Underdoped Bi2Sr2CaCu2O8+δ samples with no Zn impuri-ties (TC = 85K) were shown to produce qualitatively similar STM data atT = 100K as samples with Zn impurities (Fig. 4.11). (A) An STM topo-graph of a typical region taken with I = 40pA and V = −150mV showsthe expected atomic corrugation and superlattice distortion. Simultaneouslyacquired spatially-resolved conductance maps contain primarily structural fea-tures at (B) 30mV , and the emergence of a new bond-oriented modulation at(C) 10mV .

corresponding to the new modulation (Q). The new modulation behaves very differently

than the structural peaks. To see this, the intensities of the peaks were first scaled to their

value at 41mV (Q, 70.3pS; S, 227.7pS; A, 35.5pS), and then tracked as a function of energy

(Fig. 4.13c). The scaled amplitudes of the peaks associated with structural features (A and

S) decrease inside the pseudogap as the energy approaches the Fermi energy. This mirrors

the overall drop of the density of states inside the pseudogap, and is consistent with the

behavior of structural features in STM density of states maps taken on other surfaces. In

contrast, the new modulation (Q) only appears for energies inside the pseudogap, and its

scaled amplitude grows as the energy approaches the Fermi energy, suggesting that the new

modulation has an electronic origin, and not a structural one.

Analyzing the structure of the new modulation Q in Fourier transform space reveals

precise quantitative information about the real-space structure of the modulations. The

locations of the Q and A peaks can be tracked by taking two-pixel-averaged line-cuts of the

FFT, shown as the dashed line in Figure 4.13b. From determining their peaks’ location (Fig.

4.13d), we find that these modulations have a real-space incommensurate period of 4.7a0,

where a0 = 3.8A is the nearest neighbor Cu − Cu distance, in these slightly underdoped

70

Page 79: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A

k (2π/ao)

Fo

urie

r a

mp

litu

de

(p

S)

D Q −15 mv

0 mv

15 mv

100

50

0

150

0 0.2 0.4 0.6 0.8 1

A

Q

S

B <0,π

><π,π>

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

2

2.1

2.2

2.3

0 20 40 60 80 100

<π-0> peak (Q)

b-axis peak (S)

atomic peak (A)

DO

S (n

S)

DOS

No

rma

lize

d p

ea

k in

ten

sity

Voltage (mV)

<0,π

><π,π>

A

C

Figure 4.13: (A) The Fast Fourier Transform of a conductance map (I = 40pAand V = −150mV ) acquired on a 380A × 380A area with 200 × 200 pixels at15mV has four basic components. (B) These components are labeled in thisschematic diagram as A for the atomic peaks, S for the superlattice distortionpeaks, and Q for the new bond-oriented modulation. The broad peak in thecenter (unlabeled) corresponds to long-wavelength changes in conductance as-sociated with dopant inhomogeneity in the sample. (C) The relative intensityof the structural peaks (A and S) decreases inside the pseudogap, while that ofthe new modulation (Q) increases. Also shown is the energy-resolved densityof states (green). (D) Shown are line-cuts from tranformed STM data acquiredsimultaneously at 15mV (shown in (A)), 0mV and -15mV. Notably, the po-sition of the nearly resolution-limited Q peaks does not shift as a function ofenergy.

71

Page 80: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

samples containing Zn. Importantly, the positions of the Q peaks does not shift for all the

energies where these peaks are seen on a given sample to within our experimental resolution

of ±0.2a0. Measurements on different samples, including those not containing Zn, find

a variation in this period for a given sample between 4.5a0 and 4.8a0. This supports an

electronic origin to the modulations as well, since the atomic lattice peak (A) and the b-axis

supermodulation peak (S) had the same wavevector in all the samples. Finally, the narrow

width of the Q peaks indicates that the modulations have a correlation length of about 4−5

periods, which is corroborated by the somewhat disordered appearance of the modulations

in the real-space data. The appearance of a non-dispersing, incommensurate standing wave

pattern in the real-space density of states inside the pseudogap is the main experimental

result of this chapter.

The distinguishing feature of the modulations in the pseudogap regime is their energy-

independent wavelength. Although they appear similar to the modulations in the real-

space density of states measured by STM in the superconducting state (Chapter 3), the

modulations in the superconducting state were found to disperse. A potentially simple

explanation for the lack of dispersion seen in the pseudogap regime is that the patterns do

have an underlying dispersion, but the large level of energy broadening (±16mV for the

measurements presented here) would simply average over a range of wavevectors specified

by the underlying dispersion. Although this could produce modulations with a seemingly

fixed wavevector, the energy broadening would also result in a broadening of features in the

Fourier transforms of the data. We measure the modulations to have a wavelength of 4.7a0

with an uncertainty of ±0.2a0, where the uncertainty is due to the observed width of the Q

peaks. Thus we need to identify whether some mechanism can produce either dispersionless

modulations in the density of states, or modulations that disperse less than ±0.2a0 over the

range of energies (−15mV to ≈ 40mV ) where we find the modulated patterns.

72

Page 81: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

4.3 Scattering Interference

Recall that the modulated patterns found in the superconducting state were thought to arise

from scattering interference, as described in Chapter 3. [58, 59, 60, 61, 60, 62, 63] We can

attempt to extend the scattering interference scenario into the pseudogap regime to describe

the dispersionless modulations seen here. [64] Although angle-resolved photoemission spec-

troscopy does not find well-defined quasiparticles for T > TC , the electronic states at a given

energy can still be thought to have a broad range of available wavevectors, and should be

able to scatter elastically between them. In the Born approximation, scattering interference

introduces a correction

δn(~r, ω) = − 1

πIm

d2~r1G0(~r − ~r1, ω)V (~r1, ω)G0(~r1 − ~r, ω) (4.1)

to the density of states, where G0 is the single particle Green function, and V is a weak,

finite-range scattering potential. [65, 60, 66, 67, 63, 68] The Fourier transform of the Born

correction δn(~q, ω) = − 1πImV (~q, ω)Λ(~q, ω) separates into a part

Λ(~q, ω) =∫

d2~kG0(~k, ω)G0(~k + ~q, ω) (4.2)

that contains all the wave interference information, and a part

V (~q, ω) =∫

d2~xe−i~q·~xV (~x, ω) (4.3)

that acts like a static structure factor. [66] For now, we assume that the structure factor

acts like an all-pass filter, and the density of states correction depends only on the wave

interference term Λ.

Before addressing the various electronic states proposed for the pseudogap regime, we

treat the case of the Fermi liquid normal state for didactic purposes. Assuming that the

electronic state of the sample can be described as a metal, the Green function is given as

G0(~k, ω) = (ω + iδ − ε~k)−1 (4.4)

73

Page 82: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

where δ is a broadening term, and ε~k is the Bi2Sr2CaCu2O8+x band structure from ARPES

[69] for slightly underdoped (µ = 120.5meV , corresponding to a doping of x = .12)) samples,

[66] The resultant correction to the density of states contains dispersing caustics, in contract

to the sharp peaks in ~q-space for the superconducting state (Fig. 4.14a&b). The equivalence

of all points in ~k-space in the Green function leads directly to an absence of sharp peaks

in ~q-space. Furthermore, the dispersion is solely the consequence of the band structure

(Fig. 4.14c). Although the Fermi liquid picture helps to illustrate the effect of the band

structure on scattering interference, it is not directly applicable to the STM data measured

in the pseudogap regime. We must still account for the two key phenomenological features

of the pseudogap regime measured by photoemission: the pseudogap in the density of states

n(~k, ω) = − 1πImG0(~k, ω), and the ill-defined nature of the electrons as a function of both

energy and momentum.

Extending the scattering interference scenario into the pseudogap regime requires choos-

ing a Green function with which to model the enigmatic pseudogap electronic state. Recall

that in the pseudogap regime, angle-resolved photoemission spectroscopy (ARPES) finds

that electronic states form extended arcs centered along the 〈π, π〉 directions in ~k-space at

the Fermi energy. [70] As the energy approaches the pseudogap energy scale, these arcs

extend towards the Brillouin zone boundary, and form a closed contour for ω > ∆ (Fig.

4.15a). Finally, the ARPES lineshape is quite broad as a function of both energy and mo-

mentum, calling into question whether true quasiparticles exist at all. [14] Features of the

measured behavior have been modeled by numerous theoretical works, which have typically

either followed a phenomenological approach, or proposed exotic electronic states. We start

by focusing on the former.

Norman et al. [71] have captured many details of the ARPES data with the Green

function

G0(~k, ω) = (ω − ε~k − Σ~k)−1 (4.5)

74

Page 83: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

DC

q=(0,0) (0,π)

(π,π)

k=(0,0) (π/a,0)

(π/a,π/a)

2

3

4

5

6

7

8

-40 -20 0 20 40Energy (mV)

Wav

ele

ng

th (

a0)

Figure 4.14: We calculated the power spectrum of δn(~q, ω) using the Fermiliquid Green function with δ = 1mV . Shown are the ~q-space patterns in thefirst Brillouin zone at ω = 0 (A), and ω = 32mV (B). The dispersion of thefeature along the 〈0, π〉 direction, indicated by the arrows in (A)&(B) is plottedin (C) as a solid line. Also shown are is the dispersionless STM data takenat T = 100 > TC on a slightly underdoped sample (closed circles). [56] (D)The reason for the dispersion is that the contours of constant electron energy(dark blue, 0mV ; light blue, 32mV ) change shape in ~k-space, as dictated bythe band structure.

75

Page 84: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

where Σ(~k, ω) = −iΓ1 +∆2~k/(ω + ε~k + iΓ0) is the self-energy, Γ1 is a single particle scattering

rate, Γ0 is a measure of decoherence, and ∆~k is a gap function of amplitude ∆ = 45mV with

a ~k dependence which matches the Fermi arcs. The most promising aspect of the Fermi arc

model is that, for unknown reasons, the wavevector that joins the tips of the Fermi arcs at

the Fermi energy has the same wavelength (4.6a0) as the patterns seen by STM.

The scattering interference patterns calculated using this phenomenologically-motivated

Green functions are unlike the measured STM data (Fig. 4.15). First, the calculated pat-

terns contain caustics, while the measured patterns contain sharp peaks along the 〈0, π〉

directions. This can be addressed by assuming, as has been suggested in the literature,

that tunneling occurs preferentially along the 〈0, π〉 directions. Second, the features seen

in the calculation along the 〈0, π〉 direction disperse, unlike the dispersionless peaks seen in

the data. As indicated by the arrow, which is of the same length in both Fig. 4.15c&d,

there is a measurable dispersion for the slowest dispersing mode along the 〈0, π〉 direction.

The calculated dispersion resembles the dispersion calculated for the superconducting state

(see Fig. 4.16e) because the strongest wavevector for scattering interference is of a similar

length at the gap energy for both cases, but is shorter in the pseudogap regime at the Fermi

energy. However, the shallower dispersion, which results in a change in wavelength between

7a0 at the Fermi energy and 4.6a0 at the pseudogap energy, should still be easily resolved

in the data, which sees a fixed wavelength of 4.7 ± 0.2a0 over the entire energy range (Fig.

4.15d). [56] Unphysically small values of Γ0 = 1mV and Γ1 = 1mV were chosen to produce

sharper features in the calculations shown. However, larger (≈ 10mV ) values of Γ0 lead to

the gradual suppression of the observed mode in favor of a mode that disperses similar to

the band structure (Fig. 4.14). Larger values of Γ1 yield pictures that average over some

part of the dispersion for a particular mode, leading to features significantly broader than

the sharp peaks seen in the data.

We next calculated scattering interference patterns using various exotic electronic states

76

Page 85: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

B

2

3

4

5

6

-40 -20 0 20 40

Wav

ele

ng

th (

a0)

Energy (mV)

E

k=(0,0) (π/a,0)

(π/a,π/a)

A C

q=(0,0) (0,π)

(π,π)D

q=(0,0) (0,π)

(π,π)

Figure 4.15: (A) This is a model representation of the electronic density of

states in ~k-space measured by ARPES for underdoped samples above TC .Shown are the curves of constant electron energy for ω = 0mV (light blue)and ω = 32mV (dark blue), and the charcteristic scattering vectors ~q along the〈0, π〉 direction at both energies (light red, dark red). The power spectrum ofδn(~q, ω) calculated using a Fermi arc Green function is shown here at ω = 0mV(C) and ω = 32mV (D).

77

Page 86: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

2

3

4

5

6

7

-40 -20 0 20 40Energy (mV)

Wav

ele

ng

th (

a0)

A

EDC

B

q=(0,0) (0,π)

(π,π)

Figure 4.16: The power spectrum of δn(~q, ω) calculated using the preformedpairs Green function of Chen et al. [72] is shown here at ω = 32mV forΓ0 = 1mV (A) and Γ0 = 10mV (B). The power spectrum of a QED3 Greenfunction [65] is shown here at ω = 32mV for η = 0.4 (C) and η = 1.2 (D). Thevalue of the energy gap was chosen to be ∆ = 45mV for both calculations.(E) The dispersion of the mode identified by the arrow in both (A) and (C) isshown here as the grey line. The dispersion of the mode in (B) is also shown,and matches the dispersion of the band structure. Finally, the solid circlesdenote the dispersion measured by STM in the pseudogap regime. [56]

78

Page 87: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

proposed for the pseudogap regime. Any number of proposals have been put forth for the

pseudogap electronic state, but we will focus on four of these exotic Green functions- two

based on precursor superconductivity, and two based on other forms of order. Chen et al.

[72] have proposed that the pseudogap is due to preformed pairs, and can be described by a

modified Fermi arc Green function that has a d-wave energy gap. The calculated scattering

interference pictures based on this Green function resemble those of the superconducting

state for Γ0 small, and a broad version of the Fermi liquid state for Γ0 large (Fig. 4.16a&b).

In either limit, the patterns disperse, and should be resolved easily by STM (Fig. 4.16e).

Pereg-Barnea and Franz [65] propose a different Green function to describe preformed Cooper

pairs,

G0(~k, ω) = (ω + ε~k)/(ω2 − ε2~k− ∆2

~k)1−η/2, (4.6)

with η serving as a tunable parameter controlling the level of decoherence. As seen in Figure

4.16c&e, the scattering interference patterns generated using this Green function produce a

dispersion similar to the superconducting state (Chapter 3) for η < 0.5. Larger values of η

produce increasingly diffuse features in the calculated density of states correction until, by

η = 1.0, no features are discernable in ~q-space whatsoever (Fig. 4.16d).

Scattering interference calculations based on other proposed exotic electronic states in the

pseudogap regime also contain an unacceptable amount of dispersion. In Figure 4.17a&b, we

show scattering interference patterns for the marginal Fermi liquid scenario with circulating

current order. [73] The Green Function for this phase,

G0(~k, ω) = (ω − ε~k ± D(~k) − Σ~k)−1 (4.7)

with the plus (minus) sign for k > kF (k < kF ) , Σ(~k, ω) = λω log x/ωc+iπx/ cosh (D(~k)/x),

x = (ω2 + π2T 2)1/2 and D(~k) = D0(cos(kxa) − cos(kya))2, produces patterns that disperse

in a fashion similar to the band structure (Fig. 4.17a&b). Thus, it also fails to match the

dispersionless feature seen in STM in the pseudogap regime (Fig. 4.17d). It is replaced by a

79

Page 88: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

faster dispersing mode at energies |ω| < 2D0 for finite values of D0, although the large energy

broadening of the Green function washes this mode out too much to track its dispersion. The

last proposal we illustrate in this chapter is the scattering interference patterns calculated

using the d-density wave proposal (DDW). [65, 74] As discussed in the literature [65], the

density of states correction arising from scattering interference in the DDW proposal is given

by

δn(~q, ω) = − 1

πImV (~q, ω)

d2~k G0(~k, ω)G0(~k + ~q, ω)

+F0(~k, ω)F0(~k + ~q, ω)

+G0(~k, ω)F0(~k + ~q, ω)

+F0(~k, ω)G0(~k + ~q, ω)

(4.8)

where

G0(~k, ω) = (ω − ∆µ − ε~k+ ~Q)/((ω − ∆µ − ε~k)(ω − ∆µ − ε~k+ ~Q) − D2~k), (4.9)

F0(~k, ω) = D~k/((ω − ∆µ − ε~k)(ω − ∆µ − ε~k+ ~Q) − D2~k), (4.10)

D~k = D0(cos kx−cos ky)/2 is the DDW gap, ∆µ is a chemical potential shift, and ~Q = (π, π).

As can be seen in Figure 4.17c&d, the patterns disperse through a range of wavelengths

(∆λ = 0.8a0 between −15mV and 35mV ), which should be easily resolved in the STM

experiment (maximum ∆λ = 0.4a0 over the same range of energies).

The inability of the proposed electronic states to produce dispersionless scattering inter-

ference density of states corrections points to a fundamental problem: any Green function

will result in a wave contribution to Born scattering that disperses. This happens because

we have placed too many restrictions on the Green function. The occupied density of states

(the imaginary part of the Green function times the Fermi function) must disperse in order

to match ARPES measurements. On the other hand, the density of states correction from

80

Page 89: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

scattering interference, which is just the imaginary part of the Green function convoluted

with itself, cannot disperse if it is to match the STM data in the pseudogap regime. In

order for scattering interference to simultaneously match both the ARPES and STM data,

the structure factor, assumed to be an all-pass filter until now, could be chosen to pass only

contributions near ~q = 2πa0

(0, 1/4.7). [66] However, this corresponds to an incommensurate,

bond-oriented square lattice of scattering centers in real-space. Although this can be justified

on various physical grounds, [67, 68] it amounts to an ad hoc assumption that some form of

ordering exists in the material, and electrons are scattering off the ordered scattering sites

to produce dispersionless patterns.

These qualitative observations have been made by calculating density of states corrections

in the Born approximation (weak scattering). In the Born approximation, we have considered

only the first order correction to the Green function arising from elastic scattering: G =

G0 +G0V G0, where G0 is the unperturbed Green function, and V is the scattering potential.

However, the arguments also apply in the strong scattering limit, where we consider the

same type of correction to all orders: G = G0 + G0V G0 + G0V G0V G0 + .... Starting from a

more general form for elastic scattering,

δn(~r, ω) =∫

d2~kG0(~k, ω)T (~k,~k + ~q, ω)G0(~k + ~q, ω), (4.11)

where

T (~k,~k + ~q) = V (~k, ~k + q) +∫

d2~k′V (~k, ~k′)G0(~k′)T (~k′, ~k + ~q). (4.12)

The behavior of δn will be determined by the convolution of G0 with itself if T has no poles

at a given energy. In order for T to have poles as a function of ~k and ~k + ~q, V must contain

poles itself. Thus, in order to overcome the dispersion in δn arising from G0, we must assume

that V has poles in it, corresponding to an ordered real-space array of scattering sites. We

are left to conclude that the modulations seen in the density of states measured by STM in

the pseudogap regime simply cannot be described by scattering interference.

81

Page 90: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

q=(0,0) (0,π)

(π,π)A B

C

2

3

4

5

6

-40 -20 0 20 40

Wav

ele

ng

th (

a0)

Energy (mV)

D

Figure 4.17: The power spectrum of δn(~q, ω) was calculated with the circulat-ing current Green function using ∆0 = 10mV (A) and ∆ = 0 (B), shown hereat ω = 16mV . Values of λ = .27, T = 10mV , and ωc = 500mV were usedin both calculations. The mode pointed to by the arrow in (B) disperses likethe Fermi liquid case. (C) The power spectrum of δn(~q, ω) calculated usingthe d-density wave formalism [74, 65] is shown here at ω = 16mV . The valueof the energy gap was chosen to be ∆ = 45mV , and the chemical potentialshift was chosen to be ∆µ = 40mV . (D) The dispersion of the 〈0, π〉 modescalculated using the circulating current Green function with D0 = 0 is shownhere in grey. Also shown is the slowest dispersing 〈0, π〉 mode for the DDWGreen function (black), and the faster dispersing mode identified by Ref. [75](open circles). Finally, the solid circles denote the dispersion measured bySTM in the pseudogap regime. [56]

82

Page 91: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

4.4 Local Order

The most obvious explanation for the modulated patterns seen by STM in the pseudogap

regime would be that they result from some form of ordering. In general, the patterns seen

in the STM data could result from local charge ordering itself, from scattering off a static

ordered state, or from scattering off fluctuating order pinned by defects in the sample. One

obvious candidate for the modulations is charge stripes, as the patterns seen in STM have a

periodicity close to that expected from the stripe ordering scenario. However, the patterns

seen here naively appear to be two-dimensional, which would require one-dimensional stripes

to be either fluctuating or highly disordered in two orthogonal directions.

Another promising explanation is the local ordering of spins predicted by numerous the-

ories. [47, 53, 51, 52] These theoretical predictions have often been found to be consistent

with either the elastic spin structure or the inelastic spin fluctuations measured by neu-

tron scattering (or both). Recent theoretical work has attempted to connect the real-space

periodicity seen in STM conductance measurements and the neutron scattering data. [67]

This work postulates that spin fluctuations with wavevector 2π/λ can be pinned by de-

fects. Scattering interference off this pinned spin wave structure leads to electronic density

of states modulations with wavevector 4π/λ. In the superconducting state, the agreement

between the wavevector measured by neutron scattering as a function of energy, and half the

wavevector of the DOS modulations measured by STM, is quite remarkable (Fig. 4.18). [76]

However, neutron scattering measures the same dispersion in the Lanthanum compounds

above TC (Fig. 4.18b), [76] and only very broad features in other compounds (Fig. 4.8c).

[27] In contrast, STM sees a comparatively sharp, non-dispersing feature in the electronic

density of states. What’s more, there is no measured correlation between the spin informa-

tion measured in neutron scattering and the pseudogap energy or temperature scales, while

there is a correlation between the appearance of the dispersionless modulations seen here

83

Page 92: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 4.18: (A) The wavevector of the spin wave measured in inelastic neutronscattering (K is the distance away from (π, π) measured in reciprocal latticeunits) is plotted as a function of energy for optimally doped LSCO (diamondsand sqaures) and underdoped YBCO (circles) in the superconducting state.Also shown as the red symbols is half the wavevector of the bond-orientedmodulations measured by STM in the superconducting state. [58] (B) Shownis the wavevector of the spin wave (δ, measured relative to the (π, π) point) asa function of energy for samples in the superconducting (blue) and pseudogap(red) states. The data was acquired on optimally doped LSCO (squares anddiamonds) and underdoped LSCO (triangles). In contrast, the modulated pat-terns seen in STM data are non-dispersing in the pseudogap regime. (Figureadapted from Ref. [76]).

with STM and the pseudogap energy scale (Fig. 4.13).

The STM data presented here could also be compatible with precursor superconductivity

scenarios for the pseudogap. [77] Recall that these proposals link the pseudogap to Cooper

pair correlations above TC . Theoretical works have shown that a pair liquid with a fluc-

tuating phase, consistent with the loss of global superconducting order and the persistence

of superconducting fluctuations, can form modulated patterns in the density of states. [65]

However, these modulations should disperse in a manner similar to the quasiparticle scat-

tering picture. Alternatively, proposals have been made that hole pairs can localize into a

disordered lattice. [78, 79] These proposals, which combine elements of the ordering sce-

nario with precursor superconductivity, yield a density of states modulation with the right

periodicity, and whose appearance correlates with the pseudogap energy scale.

Accompanying the issue of what drives the local electronic ordering is whether the local

84

Page 93: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

order competes with, coexists with, or promotes superconductivity. Here, we have shown that

in-gap states form dispersionless periodic modulations when superconductivity is suppressed

by raising temperature above TC . In an earlier study, STM was used to examine the in-gap

states when superconductivity is suppressed at low temperatures by applying a magnetic

field. Near vortices, where magnetic flux threads the sample, STM sees what appears to

be a low-temperature version of the pseudogap in the density of states (Fig. 4.19a). [80]

A modulated spatial pattern with a wavelength of ≈ 4.3 ± 0.5a0 is found upon integrating

the measured density of states between the Fermi energy and 10mV (Fig. 4.19b). [81]

Unfortunately, the lack of energy- resolved data near vortices prevents determination of

the amount of dispersion for these patterns. Another recent study has searched for spatial

modulations in the density of states on samples where superconductivity has been suppressed

by heavy underdoping. [82] Although the samples used were at superconducting dopings

(δ = 0.1 and above), they find that the density of states has a large (δ ≈ 100mV ) pseudogap

(Fig. 4.19c), and shows little evidence for any feature associated with superconductivity.

The spatially-resolved density of states maps of the surface show a dispersionless modulated

pattern, with a commensurate period of 4a0 (Fig. 4.19d). In all three experiments, whenever

superconductivity is suppressed, whether by temperature, magnetic field, or doping, and

a pseudogap is seen in the electronic density of states, STM finds similar dispersionless

modulations in the local density of states.

It must be emphasized that more experiments must be done before concluding that the

modulated density of states patterns actually cause the pseudogap in the density of states.

First, it remains to be seen whether the appearance of the patterns correlates with the

appearance of the pseudogap throughout the phase diagram. Although we show here that

this correlation exists for slightly underdoped samples at T = 100K, and the correlation

has also been demonstrated for heavily underdoped samples at T = 0.1K, experiments

remain to be done on samples which do not exhibit a pseudogap, and it remains to be

85

Page 94: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

C DA B

Figure 4.19: (A) This is the energy-resolved density of states measured ata vortex core on a nearly optimally doped Bi2Sr2CaCu2O8+δ sample at lowtemperatures in an applied field(figure adapted from [80]). (B) The spatially-resolved, energy-integrated density of states of a typical region also show abond-oriented modulation near the vortex cores with a period of 4.3a0 (figureadapted from Ref. [81]). (C) This is the energy-resolved density of states inheavily underdoped (x = 0.12) Ca2−xNaxCuO2Cl2 measured at low tempera-tures. (D) The spatially resolved density of states on the same surface, shownhere at 24mV , shows a bond-oriented modulation with a period of 4a0. Boththese figures have been adapted from Ref. [82].

seen whether it is possible to find samples with a pseudogap and no modulated patterns.

Independent of whether the appearance of the patterns is correlated with the presence of

a pseudogap, it could still be unclear whether the local ordering is actually causing the

pseudogap. Recent theoretical work has shown that proximity to the antiferromagnetic

insulator could independently create a pseudogap in the density of states, and an instability

towards forming ordered states. [83] The important physics of the pseudogap would then be

encapsulated in residual antiferromagnetism, the antiferromagnetic correlations not involved

in the formation of the ordered states.

4.5 References

[1] C. Renner et al., Phys. Rev. Lett. 80, 149 (1997).

[2] H. Ding et al., Phys. Rev. B 54, 9678 (1996).

86

Page 95: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[3] J. Harris et al., Phys. Rev. B 54, 15665 (1996).

[4] A. Ino et al., Phys. Rev. B 65, 94504 (2002).

[5] A. Loeser et al., Science 273, 325 (1996).

[6] H. Ding et al., Nature 382, 51 (1996).

[7] M. Norman et al., Nature 392, 157 (1998).

[8] F. Ronning et al., Phys. Rev. B 67, 165101 (2003).

[9] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 57, 11093 (1998).

[10] T. Yoshida et al., cond-mat/0401565 (unpublished).

[11] A. Millis and H. Drew, cond-mat/0303018 (unpublished).

[12] T. Timusk, cond-mat/0303383 (unpublished).

[13] A. Kaminski et al., Phys. Rev. Lett. 84, 1788 (2000).

[14] A. Kaminski et al., Phys. Rev. Lett. 84, 1788 (2000).

[15] W. W. Jr. et al., Phys. Rev. Lett. 62, 1193 (1989).

[16] P. Hammel et al., Phys. Rev. Lett. 63, 1992 (1989).

[17] Z. Xu et al., Nature 406, 486 (2000).

[18] Y. Wang et al., Phys. Rev. B 64, 224519 (2001).

[19] J. Corson, R. M. amd J. Orenstein, J. Eckstein, and I. Bozovic, Nature 398, 221 (1999).

[20] M. Kastner, R. Birgeneau, G. Shirane, and Y. Endoh, Rev. Mod. Phys. 70, 897 (1998).

[21] P. Bourges, H. Casalata, A. Ivanov, and D. Petitgrand, Phys. Rev. Lett. 79, 4906 (1997).

87

Page 96: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[22] J. Tranquada et al., Phys. Rev. Lett. 60, 156 (1988).

[23] S. Hayden et al., Phys. Rev. Lett. 67, 3622 (1991).

[24] S. Shamoto et al., Phys. Rev. B 48, 13817 (1993).

[25] K. Yamada et al., Phys. Rev. B 57, 6165 (1998).

[26] C. Stock et al., cond-mat/0408071 (unpublished).

[27] P. Bourges et al., Science 288, 1234 (2000).

[28] J. Tranquada et al., Nature 429, 534 (2004).

[29] N. Christensen et al., cond-mat/0403439 (unpublished).

[30] J. Tranquada et al., Nature 375, 561 (1995).

[31] S. Wakimoto et al., cond-mat/0312648 (unpublished).

[32] M. Norman and C. Pepin, Rep. Prog. Phys. 66, 1547 (2003).

[33] H. He et al., Phys. Rev. Lett. 86, 1610 (2001).

[34] P. Dai et al., Science 284, 1344 (1999).

[35] B. Keimer et al., Phys. Rev. B 46, 14034 (1992).

[36] P. Mallet et al., Phys. Rev. B 60, 2122 (1999).

[37] Z. Dai, Q. Xue, Y. G. C. Slough, and R. Coleman, Phys. Rev. B 48, 14543 (1993).

[38] P. Anderson, Science 235, 1196 (1987).

[39] P. Anderson, The Theory of Superconductivity in the High-TC Cuprates (Princeton Uni-

versity Press, ADDRESS, 1997).

88

Page 97: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[40] I. Afflek, Z. Zou, T. Hsu, and P. Anderson, Phys. Rev. B 38, 745 (1988).

[41] X.-G. Wen and P. Lee, Phys. Rev. Lett. 76, 503 (1996).

[42] P. Lee, N. Nagaosa, T.-K. Ng, and X.-G. Wen, Phys. Rev. B 57, 6003 (1996).

[43] S. Chakravarty, R. Laughlin, D. Morr, and C. Nayak, Phys. Rev. B 63, 94503 (2001).

[44] C. Varma, Phys. Rev. Lett. 83, 3538 (1999).

[45] A. Kaminski et al., Nature 416, 610 (2002).

[46] C. Varma, Phys. Rev. B 61, 3804 (2000).

[47] J. Z. and. O. Gunnarsson, Phys. Rev. B 40, 7391 (1989).

[48] V. Emery, S. Kivelson, and J. Tranquada, Proc. Natl. Acad. Sci. USA 96, 8814 (1999).

[49] I. Vekhter and A. Chubukov, Phys. Rev. Lett. 93, 16405 (2004).

[50] A. Abanov et al., Phys. Rev. Lett. 89, 177002 (2002).

[51] Y. Zhang, E. Demler, and S. Sachdev, Phys. Rev. B 66, 94501 (2002).

[52] J. Schmalian, D. Pines, and B. Stojkovic, Phys. Rev. Lett. 80, 3839 (1998).

[53] S.-C. Zhang, Science 275, 1089 (1997).

[54] H.-D. Chen et al., Phys. Rev. Lett. 89, 137004 (2002).

[55] U. Welp et al., Phys. Rev. Lett. 62, 1908 (1989).

[56] M. Vershinin et al., Science 303, 1995 (2004).

[57] S. Pan et al., Nature 413, 282 (2001).

[58] J. Hoffman et al., Science 297, 1148 (2002).

89

Page 98: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[59] K. McElroy et al., Nature 422, 592 (2003).

[60] Q.-H. Wang and D.-H. Lee, Phys. Rev. B 67, 20511 (2003).

[61] J. Byers, M. Flatte, and D. Scalapino, Phys. Rev. Lett. 71, 3363 (1993).

[62] L. Capriotti, D. Scalapino, and R. Sedgewick, Phys. Rev. B 68, 45120 (2003).

[63] L. Zhu, W. Atkinson, and P. Hirschfeld, Phys. Rev. B 69, 60503 (2004).

[64] S. Misra, M. Vershinin, P. Phillips, and A. Yazdani, Phys. Rev. B in press (2004).

[65] T. Pereg-Barnea and M. Franz, Phys. Rev. B 68, 180506 (2003).

[66] L. Capriotti, D. Scalapino, and R. Sedgewick, Phys. Rev. B 68, 14508 (2003).

[67] A. Polkovnikov, M. Vojta, and S. Sachdev, Phys. Rev. B 65, 220509 (2002).

[68] C.-T. Chen and N.-C. Yeh, Phys. Rev. B 68, 220505 (2003).

[69] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 52, 615 (1995).

[70] M. Norman et al., Nature 392, 157 (1998).

[71] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 57, 11093 (1998).

[72] Q. Chen, K. Levin, and I. Kosztin, Phys. Rev. B 63, 184519 (2001).

[73] C. Varma, Phys. Rev. Lett. 83, 3538 (1999).

[74] C. Bena, S. Chakravarty, J. Hu, and C. Nayak, cond-mat/0311299 (unpublished).

[75] C. Bena, S. Chakravarty, J. Hu, and C. Nayak, cond-mat/0405468 (unpublished).

[76] N. Christensen et al., cond-mat/0403439 (unpublished).

[77] V. Emery and S. Kivelson, Nature 374, 434 (1995).

90

Page 99: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[78] H.-D. Chen, O. Vafek, A. Yazdani, and S.-C. Zhang, Phys. Rev. Lett. 93, 187002 (2004).

[79] P. Anderson, cond-mat/0406689 (unpublished).

[80] S. Pan et al., Phys. Rev. Lett. 85, 1536 (2000).

[81] J. Hoffman et al., Science 295, 466 (2002).

[82] T. Hanaguri et al., Nature 430, 1001 (2004).

[83] T. Stanescu and P. Phillips, Phys. Rev. Lett. 91, 17002 (2003).

91

Page 100: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 5

The CuO2 Plane

5.1 Introduction

Although the materials that exhibit high temperature superconductivity vary widely in com-

position and structure, they are all layered compounds with one or more perovskite CuO2

plane per unit cell (Fig. 5.1a&b). As a result, the interesting highly-correlated electronic be-

havior of the HTS is thought to originate on these CuO2 planes. Their electronic structure

derives from Cu 3dx2−y2 and O 2px and 2py orbitals, with the anti-bonding configuration

of these orbitals sitting nearest the chemical potential (Fig. 5.1b). In its parent state,

one spin-half electron occupies every CuO2 plaquette, with strong electron-electron correla-

tions resulting in the formation of an antiferromagnetic insulator. Once some of the other

planes of the crystal structure remove electrons (create holes) on the CuO2 plane, the rest of

the strongly correlated behavior, including superconductivity and the pseudogap behavior,

emerges. It is thus of fundamental importance to make measurements that focus on the

properties of the CuO2 plane.

Both photoemission and scanning tunneling spectroscopy have been workhorses in diag-

nosing this correlated electronic behavior, in particular for their ability to measure the energy

dependence of the density of states. However, both techniques are surface-sensitive, and thus

92

Page 101: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 5.1: (A) This schematic drawing (side view) shows the positions ofthe metal ions in Bi2Sr2CaCu2O8+δ. The interesting many- body electronicbehaviour occurs on the pair of CuO2 planes (red) in every half-unit cell. Theother planes act as either dopant layers containing the extra oxygens (BiO,blue) or as spacer layers (Ca, green and SrO, yellow). (B) This is a schematicdrawing (top view) of the structure of the CuO2 plane common to all the HTSmaterials. Here, the red dots correspond to Cu atoms and the small greendots to O atoms. The unit cell of this plane, called a plaquette, is a squarecontaining four Cu atoms at the corners and four O atoms on the sides. Alsoshown are the Cu 3d orbitals and O 2p orbitals.

rely on having flat, clean surfaces on which to perform measurements. For this reason, these

techniques rely on mechanically cleaving samples before making measurements, a process

that breaks apart the two crystallographic plane in the crystal most weakly bonded to one

another. In general, nearly all the STM and photoemission studies on the HTS have been

performed on samples terminated by planes other than the CuO2 plane as a consequence of

this cleaving process. Most studies have focused on the compound Bi2Sr2CaCu2O8+δ, be-

cause it cleaves easily and exposes macroscopically flat regions for measurement. However,

in this case, cleaving exposes the BiO layer, [1] which sits 4.5A above the pair of CuO2

planes in the unit cell (Fig. 5.1a). This layer contains the excess oxygens used to dope

Bi2Sr2CaCu2O8+δ, [2, 3] and thus the BiO plane has been thought of as a doping plane,

93

Page 102: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

D

A CB

Binding Energy (V)

Figure 5.2: (A) The tunneling conductance measured by STM on an under-doped Bi2Sr2CaCu2O8+δ sample is shown here as a function of temperature(adapted from Ref. [4]). (B) This is the spectrum measured by photoemis-sion near the (0, π) point of the Brillouin as a function of temperature on anunderdoped sample with TC = 83K. (adapted from Ref. [5]) (C) A phase dia-gram can be constructed from either measurement which resembles the phasediagram constructed from in-plane resistivity measurements (D). Figures (C)and (D) have been adapted from Ref. [6]

with the SrO and Ca planes acting as spacer layers.

The majority of the energy-resolved density of states information known about the

cuprates has thus been gathered using surface-sensitive techniques Bi2Sr2CaCu2O8+δ sam-

ples with BiO surface termination, and has been assumed to be representative of the prop-

erties of the sub-surface CuO2 planes. This assumption appears valid on a qualitative level

because of the rough level of agreement between the two surface-sensitive techniques, and

between them and bulk probes. The density of states measured by STM shows sharp peaks,

located at the edge of the energy gap, which appear only for samples in the superconduct-

ing state (Fig. 5.2a). [4] Similarly, the sharp peak seen in photoemission data near the

zone boundary, where the energy gap reaches its maximum value, appears only below TC

94

Page 103: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Tip

Sample

d

A B

Figure 5.3: (A) In an STM measurement, the tunneling current depends ex-ponentially on the distance d separating the sample and the tip. (B) In aphotoemission measurement, monochromatic light shined on a sample causesthe emission of a photoelectron. The penetration depth of light into the samplesurface is typically tens of Angstroms. However, the degree of surface sensi-tivity can be enhanced by the escape depth of the photoelectron- the depthinside the material from which a photoelectron can be emitted into vacuumwithout interacting with the material. (Figure adapted from [12])

(Fig. 5.2b). [5, 7] In addition, both techniques see a significant density of states at the

Fermi energy, and a pseudogap, for temperatures between TC and T ∗ in underdoped sam-

ples. [8, 9, 10, 4] The two experimental techniques, despite making measurements using

different quantum processes, have even been shown to be in agreement in the subtle details

of the spectra they measure. [11] Most importantly, the doping-temperature phase diagram

that can be constructed using these surface-sensitive techniques qualitatively agree with the

phase diagram constructured using techniques which probe CuO2 planes in the bulk (Fig.

5.2c&d). [6]

The process of making both measurements, however, is surface-sensitive to a disturbing

degree. In STM, tunneling current between the tip and the sample depends exponentially

on the distance separating them. Because the nearest CuO2 plane lies 4.5A beneath the

common BiO surface in Bi2Sr2CaCu2O8+δ, the matrix element suppresses direct tunnel-

ing into the CuO2 plane by four orders of magnitude in comparison to tunneling processes

involving the BiO plane. In photoemission, calculations based on experimental evidence

95

Page 104: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure 5.4: (A) This is a three-dimensional rendition of an STM topographtaken by Ref. [15] on a single crystal of Bi2Sr2CaCu2O8+δ cleaved in air.Although spectra were taken on these terraces, the dirt present on the samplesurface makes it difficult to determine which layer of the crystal structure thesemeasurements were taken on, or whether the spectra are representative. (B)This figure shows an STM topograph of a single crystal of Bi2Sr2CaCu2O8+δ

cleaved in ultra-high vacuum rendered in a way to show the terraces of thismulti-stepped region. The investigators were unable to take spectra on thissample. Although they were able to determine the identity of the terraces,none of the exposed plains was a CuO2 plane (Figure adapted from Ref. [16]).

from the other crystallographic planes indicate that the escape depth for the emitted pho-

toelectrons is only ≈ 3A, meaning that the other planes must play an important role for

photoelectrons originating from the CuO2 planes (Fig. 5.3). [13, 14] Because the effect of

the other crystallographic layers on both measurements is unknown, performing these mea-

surements on samples terminated by the CuO2 plane is necessary to confirm that data taken

on BiO-terminated samples is representative of the behavior of the CuO2 plane in the bulk.

In this chapter, we will present the first conclusive topographic and electronic density of

states measurements on a single CuO2 plane at the surface of a HTS using STM. [17] There

have been some STM experiments performed on HTS samples terminated by different crys-

tallographic planes. [18, 15, 16, 19, 20] However, none of these has managed to conclusively

identify a CuO2 plane at the surface (Fig. 5.4). Such an identification requires a careful

correlation of STM topographs with crystallographic data, which, in turn, requires exper-

iments to be conducted entirely in an ultra-high vacuum environment to prevent surface

contamination. [19] Clean, reproducable atomically-resolved topographs are also required

96

Page 105: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

to make such an identification, and to confirm the quality of the tunnel junction in making

spectroscopic measurements. [20] Our measurements indicate that the CuO2 plane can form

a stable termination layer on the surface of Bi2Sr2CaCu2O8+δ, with a lattice similar to that

of the plane in the bulk of the sample. The spectroscopic measurements made on this plane

are qualitatively different than those made on the common BiO surface, and, in comparison,

tunneling directly into the surface CuO2 plane appears strongly suppressed near the Fermi

energy. We suggest that the reason for this discrepancy is a change in doping for the surface

CuO2 plane resulting from the absence of the other crystallographic planes of the unit cell.

Although more work is required to determine the electronic phase of the surface CuO2 plane,

for which we consider all the phases of the HTS phase diagram to be a possibility, this work

opens up a new avenue to studying the physics of the high temperature superconductors.

5.2 Results

To search for a surface CuO2 plane, we return to the Bi2Sr2CaCu2O8+δ thin films used

in the first part of Chapter 3. Cleaving thin films requires significantly more force than

cleaving single crystals, most likely because the former contain a higher density of inter-

plane defects. This suggests that, unlike single crystals, the thin films might not cleave at the

weakest bond of the crystal structure (intercell BiO-BiO bond) for all regions of the surface.

The slightly underdoped 1000A thick films were grown using molecular beam epitaxy, and

demonstrated a resistive transition at TC = 84K. They were cleaved mechanically in ultra

high vacuum (UHV) at room temperature prior to making measurements using a home-

built UHV STM operating at T = 4.2K (LTSTM). The majority of the surface contains

wide terraces (100 − 1000A) that yield topographs (Fig. 5.5a) and conductance spectra

(Fig. 5.5b) reminiscent of those previously measured on BiO terminated single crystals.

[21, 22, 23, 24, 25, 4] In addition to these BiO terminated regions, we occasionally find

97

Page 106: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

0

1

2

-200 -100 0 100 200

dI/d

V (

nS

)

V (mV)

A B

Figure 5.5: (A) This is an STM topograph of a 100A × 100A area taken ona BiO-terminated region of Bi2Sr2CaCu2O8+δ (I = 200pA and V = 200mVtunnel junction). (B) Although the conductance spectra measured on thisplane vary from location-to-location, they all feature an inchoate gap at theFermi energy (V = 0) flanked by sharp conductance peaks (dc junction impe-dence Rj = | − 200mV |/200pA = 1GΩ). This Figure has been adapted fromRef. [17].

smaller terraces on the sample that are terminated by other crystallographic layers.

We will focus on STM measurements taken near the step edge in Figure 5.6a, which we

find to have subunit cell height. Tunneling conductance measurements taken far from this

step edge on the top terrace (Fig. 5.6b) resemble those taken on the majority of the surface of

the thin film, and those measured on cleaved single crystals. We can thus conclude that the

BiO plane terminates the top terrace. Contrary to the expectation that the spectra will be

unaffected by the identity of the top layer of the sample, the spectra measured far away from

the step edge on the bottom terrace (Fig. 5.6b) are qualitatively different. To determine the

identity of the bottom terrace, we need to make an accurate determination of the height of

separation between the top (BiO) and bottom (unknown) terraces, and compare the result

to the known separation between the crystal planes of Bi2Sr2CaCu2O8+δ (Fig. 5.6c). [3]

Averaging all the scanlines in the topographic image of the step revealed the height to be

98

Page 107: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

0

2

4

6

8

0 25 50 75 100Distance (Å)

Tip

Heig

ht

(Å)

BiO

SrO

Ca

CuO2

CuO2

topbottomA B

C

BiOCuO2

BiOCuO2

BiOCuO2

D

Figure 5.6: (A) This STM topograph of a 100A × 100A region of the sampleshows a sub-unit cell height step edge (I = 50pA and V = −200mV ). Thedata has been rendered in a way to bring out the atomic corrugation on boththe top and bottom terraces. (B) A line of tunneling conductance spectra wastaken perpendicular to the step edge, at the distances indicated. Both thetopmost and bottommost spectra (red curves) were taken at a distance about50A away from the step edge. (C) The linescans comprising the topographin (A) were averaged together, and compared with the distance of separationbetween the crystalligraphic planes of Bi2Sr2CaCu2O8+δ known from bulkscattering techniques. [3] (D) This is a schematic diagram of the sub-unit cellheight step edge in (A). Parts (A)-(C) of this figure have been adapted fromRef. [17].

99

Page 108: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

0.01

0.1

1

-200 -100 0 100 200

Noise

Floor

0

1

2

-200 -100 0 100 200

BiO

CuO2

dI/dV(nS)

V(mV)

C

V(mV)

dI/dV(nS)

Figure 5.7: (A) This is an STM topograph of a 64A × 64A area taken on aCuO2-terminated region of Bi2Sr2CaCu2O8+δ (I = 200pA and V = 200mVtunnel junction). The inset shows a 8A × 8A topograph of a subset ofthis region (adapted from Ref. [17]). (B) This is a set of typical tunnel-ing conductance spectra, taken on the region in (A) (dc junction impedenceRj = | − 200mV |/200pA = 1GΩ). For comparison, we also show a representa-tive spectrum from a BiO-terminated surface taken under the same conditions.(C) This is the same plot as in (B), except on a log-lin scale. The noise floorof our conductance measurement is also shown.

≈ 8A. Sources of error, arising from the miscalibration of our instrument or a difference in

electronic structure between the two planes leading to enhanced tunneling current, were all

estimated to be much smaller (0.1 − 0.2A) than the separation between any two adjacent

layers in the crystal structure (≈ 2A). Consequently, the correlation between the height

of this step edge and the known separation between the crystallographic layers (Fig. 5.6c)

identifies a single CuO2 plane as terminating the lower terrace. We have used this procedure

to identify several CuO2 terraces at the surface of different Bi2Sr2CaCu2O8+δ samples grown

under the same conditions. Because of the large resistivity anisotropy in Bi2Sr2CaCu2O8+δ

(ρc/ρab ≈ 100), the coupling between CuO2 planes in adjacent half-unit cells is expected to be

very weak in this material. Consequently, we believe our measurements to be representative

of a single CuO2 plane doped only by the crystal planes beneath it.

Having successfully identified a CuO2-terminated region of the sample, we now turn

to measuring its unperturbed electronic and structural properties far from the step edge.

100

Page 109: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Topographic images of this surface, shown in Figure 5.7, show a well-ordered square lattice.

The b-axis superlattice distortion is seen to modulate this atomic structure at 45 to the

Cu − Cu nearest neighbor direction, in agreement with other measurements on bulk CuO2

planes. [3, 26] Spectroscopic measurements made on the CuO2 surface show a density of

states that is qualitatively different than those measured on BiO surfaces in four ways (Fig.

5.7b): the CuO2 spectrum has a wider energy gap, weaker conductance peaks at the edge

of this gap, different high energy features, and, most importantly, a strongly suppressed

tunneling conductance near the Fermi energy. Tunneling into the surface CuO2 plane is so

strongly suppressed near the Fermi energy that the conductance is beneath the noise floor

of our equipment (Fig. 5.7b). Conductance measurements taken over distances between

the atomic corrugation spacing and several hundred Angstroms show qualitatively similar

features, indicating the features in the spectra highlighted above are representative of those

measured on a surface CuO2 plane. The contrast between the density of states measured

on BiO-terminated and CuO2-terminated regions of Bi2Sr2CaCu2O8+δ is the main result

reported in this chapter, and demonstrates the importance of the identity of the surface layer

in STM measurements on this material.

5.3 Interpretation

Although the effect of the other crystallographic planes on STM measurements is still un-

known, we believe that tunneling conductance measurements made on BiO-terminated sam-

ples are representative of the electronic behavior of the subsurface CuO2 planes. As outlined

previously, basic features in the spectra measured by both STM and photoemission agree

with one another, despite the fact that the two techniques probe materials in fundamentally

different ways. These features can be used to construct a phase diagram that agrees with

probes that measure the properties of CuO2 planes in the bulk. In addition, a comparison

101

Page 110: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

of the Fermi surface measured by photoemission [12, 27] and electronic structure calcula-

tions [28] has indicated that only the CuO2 planes in Bi2Sr2CaCu2O8+δ have states near

the Fermi energy. This suggests that the low-energy features in the tunneling conductance

measured with STM on the BiO surface should agree with the expectation that d-wave

BSC-like superconductivity is present on the sub-surface CuO2 planes. To test this idea, we

will model the tunneling conductance as

dI

dV(V ) ∝

d2~k∫

dUn(~k, EF + V )|M(~k)|2e−(EF +V −U)/2σ2

, (5.1)

where n(~k, EF + V ) = n0V/√

V 2 − ∆(~k)2 is the BCS density of states, n0 is the den-

sity of electronic states at the Fermi surface in the absence of superconductivity, ∆(~k) =

∆0[cos kxa − cos kya] is the energy gap for a d-wave superconductor, and σ is a broadening

factor. Assuming that the tunneling matrix element is isotropic, the calculated tunneling

conductance agrees with the STM data measured on the BiO plane at energies near the

Fermi energy. The value of the energy gap, ∆0 = 40mV , is similar to what we would expect

for a nearly optimally-doped sample, and the value for thermal broadening, σ = 6mV , is

consistent with the conditions of our measurement. More detailed models, derived using the

band dispersion measured by photoemission and interaction with the fluctuating antiferro-

magnetism measured by neutron scattering (see Chapter 3), have been shown to reproduce

the precise features of the STM tunneling spectrum in the superconducting state out to

V = ±200mV . [11] Although no definitive experiment has been done to identify the ef-

fect of the other planes, a preponderance of evidence indicates that STM measurements on

BiO terminated Bi2Sr2CaCu2O8+δ samples do probe the properties of the subsurface CuO2

planes.

In contrast to tunneling on the BiO surface, the process of tunneling directly into a

CuO2 plane at the surface is not complicated by the presence of the other crystallographic

planes. However, the tunneling conductance measured on the CuO2 terminated surface

differs significantly from the conductance measured on the BiO terminated surface, particu-

102

Page 111: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

-200 0 200

1

2

dI/

dV

(n

S)

V (mV)

-200 2000

V(mV)

dI/d

V (

arb

. u

nit

s)

1

2

Figure 5.8: (A) For in-gap energies, the qualitative features of a tunnelingconductance spectrum measured on the BiO surface (left) are qualitativelyreproduced using the model discussed in the text (right).

larly for in-gap energies, where tunneling into the CuO2 surface is dramatically suppressed.

Moreover, as we have just shown, the spectrum on the BiO surface qualitatively matches

our expectations that the electronic state of the sample is described as a BCS-like d-wave

superconductor. Thus we must conclude that the process of exposing the CuO2 plane at

the surface dramatically modifies its electronic character compared to when it is beneath

a BiO terminated surface, or when it is in the bulk of the sample. Such a situation could

arise if exposing the plane at the surface leads to a structural modification of the CuO2

lattice. To address this possibility, we can compare the topographic images obtained on

both the BiO and CuO2 surfaces (Fig. 5.5a and Fig. 5.7a). The atomic corrugations seen

in topographs on BiO terminated samples are believed to correspond to Bi atoms, which

sit above the Cu atoms of the CuO2 plane, and those on the CuO2 plane are believed to the

the Cu atoms themselves. A comparison of both topographs reveals the nearest neighbor

corrugation distance to be equal (3.8A), and to be consistent with the Cu − Cu distance

measured in the bulk. [3] Furthermore, as mentioned previously, the b-axis corrugation is

seen to have the same periodicity for both the surface and bulk CuO2 planes. Although the

103

Page 112: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

error in determining both these distances using STM topographs of the surface CuO2 plane

is large (≈ 0.1A) compared to the error in determining them using scattering techniques

on planes in the bulk (typically an order of magnitude better), this agreement leads us to

believe that the CuO2 lattice is not distorted at the surface.

We believe the most likely complication introduced by exposing the CuO2 plane at the

surface is precisely the absence of the other crystallographic planes. In order to expose the

CuO2 plane, we have had to remove many of the crystallographic layers of Bi2Sr2CaCu2O8+δ

on one side of this plane (Fig. 5.6d). Chief amongst these is the second CuO2 plane in each

half unit cell. However, STM conductance spectra from the BiO surface of Bi2Sr2CuO6+δ,

which has a single CuO2 plane per half unit cell, have a similar shape for in-gap energies

as the spectra taken here on the BiO surface of Bi2Sr2CaCu2O8+δ, suggesting that the

removal of the other layers might play a more significant role. [29] Indeed, the other missing

planes- one Ca layer, one SrO layer, and one BiO layer- are crucial for the charge balance

in this compound. Doing simple valence charge counting, the oxygen atoms attract two

electrons apiece, the strontium, copper and calcium atoms give up two electrons apiece, and

the bismuth atoms give up three electrons apiece. Consequently, a single CuO2 plane wants

to attract two electrons to become chemically stable. In a complete half unit cell, the two

SrO layers are charge neutral, the two BiO layers each donate an electron to each of the

two CuO2 planes, and the one Ca layer donates an electron to each of the two CuO2 planes,

resulting in chemically stable CuO2 planes. In the geometry that exposes a single CuO2

plane at the surface, we have a BiO, SrO, and a CuO2 plane, meaning that the CuO2 plane

is not chemically stable because it is missing one electron per plaquette. Because topographic

data of this plane indicate it is both stable and structurally similar to a bulk CuO2 plane,

some complex charge redistribution is most likely happening in the vicinity of this plane.

Therefore, we must consider the possibility that we have dramatically changed the electronic

phase of this plane.

104

Page 113: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

The most obvious possibility for the electronic phase of the surface CuO2 plane is an insu-

lator, with the observed energy gap being associated with injecting or removing electrons into

the system. Our data, however, place some stringent requirements on a possible insulating

phase. The most stringent of these requirements is that the insulator must be particle-hole

symmetric, as the chemical potential is located in the center of the 2∆ = 120mV energy

gap seen in the conductance spectrum. In a simple (band) insulator, the chemical potential

is pinned by surface defects. Because we have seen the same electron-hole symmetric en-

ergy gap on several surface CuO2 planes, with different terrace sizes and defects, we believe

that the surface CuO2 plane is not a simple insulator. Another promising explanation is

that the surface CuO2 plane is itself conducting, but the observed spectrum results from a

Coulomb blockade, as seen in disordered or granular systems with STM. [30] These systems

are typically modeled as a small, isolated conductive element in between an electron source

and an electron drain. In a simple model, the small conductive element must be capacitively

charged, at an energy cost of Ec = e2/2C before it conducts (here, e is the charge of an

electron and C is the capacitance between the conductive element and a source of electrons).

Moreover, this energy cost can be roughly the same for electron injection or removal, cor-

responding to positive and negative energies. Modeling the tip-sample tunneling junction

as a parallel-plate capacitor with a separation of d = 6A, we find that the observed energy

gap corresponds to a grain with diameter 100A. However, we measure the same energy

gap over distances much larger than 100A, without encountering any grain boundaries. In

sum, the requirements that the spectra both contain a particle-hole symmetric energy gap

and remain roughly spatially homogeneous suggest that some strongly correlated electronic

behavior exists on the surface CuO2 plane.

The effect of removing the layers on top of the CuO2 plane could be simply a change in

doping level for the surface CuO2 plane compared to the plane in the bulk. This is supported

by the fact that the lattice structure of this plane appears undisturbed by the process of

105

Page 114: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

exposing it at the surface. However, although our samples were seen to have a resistive

superconducting transition at T = 84K both before and after cleaving, the removal of the

planes could change the local doping level quite dramatically. As a result, we consider the

possibility that any of the phases of the HTS phase diagram might be present on this plane.

One possibility is that the doping level of the surface CuO2 plane has not changed

much, and it is still superconducting. Assuming that the superconducting state is similar

to that of bulk CuO2 planes, we expect the density of states at low energies to have a

BCS form with a d-wave energy gap. As we demonstrated earlier, if we assume that the

STM tunneling matrix element is isotropic, this calculation yields a tunneling conductance

with a far greater amplitude inside the energy gap than what is measured on the surface

CuO2 plane (Fig. 5.7b and Fig. 5.8). However, theoretical efforts have proposed that the

mechanism of tunneling an electron into the CuO2 plane along the c-axis, as in the STM

geometry, results in anisotropic tunneling. [31, 32] These theoretical works have argued that

tunneling along the c axis direction does not occur directly into Cu 3d states, which lie flat

in the CuO2 plane (Fig. 5.9a). Instead, tunneling is facilitated by the orbital that extends

the furthest out of the plane- the 4s orbital of the Cu atoms. In the STM geometry, this

means that tunneling probes the electronic states near the Fermi energy in the CuO2 plane

through a two step process, where electrons at the Fermi energy in the tip first tunnel into

the Cu 4s orbital, which lies above the Fermi energy, and then from this orbital into the Cu

3d orbital, which lies near the Fermi energy. However, the coupling of electrons between the

Cu 4s and Cu 3d orbitals introduces an anisotropic matrix element that has a d-wave form

|M(~k)|2 ∝ [cos kxa−cos kya]. Simply, the tunneling of electrons into the CuO2 planes occurs

preferentially along the directions where the energy gap is at its maximum. We show the

effect of incorporating this matrix element into the calculation of the tunneling conductance

using a BCS density of states and a d-wave order parameter in Figure 5.9b.

Accounting for the anisotropy of tunneling into the CuO2 plane arising from the symme-

106

Page 115: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Cu

Cu

Cu

Tip

Cu 4S

Cu 3dSide

View

A B

Figure 5.9: (A) Shown is a schematic model for tunneling into the CuO2 plane,as discussed in the text. (B) This model of tunneling results in the anisotropictunneling of electrons into the CuO2 plane. Shown here is a conductancecurve produced by the model of tunneling conductance discussed in the textoverlayed on an STM conductance spectrum taken on the CuO2 plane (Figureadapted from Ref. [17]).

try of the molecular orbitals involved in the tunneling process gives a fairly close match to

the measured low-energy conductance. The best fit with the experimental data is obtained

using ∆0 = 60mV , which is larger than the value expected for samples with a TC of 84K.

However, this value of the energy gap is close to that previously reported on underdoped,

but still superconducting, samples (δ = 0.1). [6] Furthermore, it has been suggested that

underdoping the sample will cause the broad hump seen at negative energies on BiO termi-

nated samples to shift to larger negative energies, providing a potential explanation for the

feature seen at higher negative energies on the surface CuO2 plane. [11] However, to obtain

the best match between the calculation and the data, we have had to use an unphysically

large amount of energy broadening (σ = 12mV ). In addition, the feature seen at higher

positive energies on the CuO2 plane has no analog in the BiO-terminated data. However,

both these can be rationalized as subtle differences caused by the removal of the other crys-

tal planes. More troubling is the fact that the same physical arguments used for invoking

this anistropic tunneling matrix element can be made for tunneling over BiO-terminated

107

Page 116: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

samples. The constrained c-axis tunneling that results from this model was, in fact, first

proposed as an explanation for the large anisotropy between the in-plane and out-of-plane

resistivity in the HTS. [31, 32] Although we cannot rule out the possibility that the surface

CuO2 plane is superconducting, it seems unlikely that anisotropic tunneling arising from the

symmetry of the atomic orbitals occurs only when the CuO2 plane is at the surface, and not

when it is in the bulk.

Another possibility is that the surface CuO2 plane is in the low-temperature pseudogap

regime that exists for dopings in between the superconductor and antiferromagnetic insulator.

This interpretation is tempting because the physics of the pseudogap regime is unsettled,

many competing proposals have been made (Chapter 4), and this would provide a new, direct

way of examining a CuO2 plane in the pseudogap regime at low temperatures. However,

in the cases where these theoretical proposals predict the density of states, they predict a

pseudogap- an energy gap with a finite number of states at the Fermi energy. In contrast, we

find that the tunneling conductance is strongly suppressed at the Fermi energy, and drops

below the noise floor for our measurement within 10mV of the Fermi energy. Moreover,

recent STM measurements on lightly doped Ca2−xNaxCu2O4Cl2 see a density of states with

remarkably different characteristics than the spectra measured by us on the surface CuO2

plane. Notably, the spectra measured on Ca2−xNaxCu2O4Cl2 have a ’real’ pseudogap, with

a measurable number of in-gap states (Fig. 5.10). [33] Nevertheless, the doped CuO2 plane

might possess an instability to different kinds of ordered states, depending on factors other

than doping and temperature. The electronic state of the surface CuO2 plane might thus

still be related to one of the various exotic states proposed for the pseudogap regime.

The final possibility is that the surface CuO2 plane is in the lightly-doped antiferromag-

netic insulator regime of the HTS phase diagram. At dopings below where superconductivity

occurs, resistivity data shows a crossover from metallic behavior at high temperatures to in-

sulating behavior at low temperatures. The authors of this work suggest that the reason

108

Page 117: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Figure 5.10: STM conductance spectra taken on the compoundCa2−xNaxCu2O4Cl2 for a doping of x = 0.06 (insulating state), x = 0.08(zero temperature pseudogap regime), and x = 0.1 (superconducting state) allshow an ill-defined energy gap reminiscent of a pseudogap. (Figure adaptedfrom Ref. [33]).

is that charge carriers exist at the chemical potential at these dopings, but are localized.

[34] They further note that the hole mobility scales with doping in a fashion similar to the

average spacing between holes (3.8A/√

δ). This could be consistent with our observation

of a spatially homogeneous energy gap if we treat the surface CuO2 plane as a collection

of localized holes. The energy gap in the spectra on this plane would then be the result

of Coulomb blockade associated with tunneling into an island the size of this localization

length scale. Moreover, an island of the same size, having the same energy gap, would sim-

ply follow the tip wherever we attempted to tunnel into the sample. We can attempt to

determine whether the size of the observed energy gap is consistent with this scenario. The

observed energy gap of Ec = 60mV indicates an island size of 100A from using the Coulomb

gap equation. The doping level of the sample could then be calculated using the average

separation between holes at a particular doping. Our energy gap translates into a doping

level of δ = .01, where the sample is expected to show this localization behavior. Although

we cannot make a conclusive claim that the surface CuO2 plane is in the doped insulating

regime, this explanation provides a self-consistent explanation of the data.

109

Page 118: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Clearly, more data is needed before we can distinguish which of the electronic phases

is present on the surface CuO2 plane. If step edges could be found cutting through this

plane along the 110 direction, we could test for the presence of Andreev bound states,

which should only be present in a d-wave superconductor (Chapter 1). If large data sets over

large regions of the sample had been taken, as in Chapter 3, we could examine the spatial

structure of the density of states at a given energy to look for the presence of local charge

ordering. Finally, if we had varied the tip-sample separation by a significant distance (≈ 1A),

we would have been able to test whether the observed energy gap came from a Coulomb

blockade effect arising from the localization of electrons in the doped insulator. Changing

the tip-sample separation would have changed the capacitance, and altered the size of the

Coulomb charge gap.

Irrespective of what the electronic phase of the surface CuO2 plane is, our demonstration

that it is possible to tunnel directly into a CuO2 plane opens a new way to probe the

HTS. Because its crystal structure appears unaltered by exposing it at the surface, STM

measurements made on the surface CuO2 plane should be representative of the behavior of

perovskite CuO2 planes in general. Furthermore, the primary effect of removing the other

crystal layers appears to be a change in the carrier concentration of the surface CuO2 plane.

As a result, it should be possible to measure properties of the HTS, which are thought

to be nothing more than doped CuO2 planes, by making STM measurements directly on

CuO2-terminated samples, with the added benefit that the process of measurement is not

complicated by the presence of the other crystallographic planes.

5.4 References

[1] P. Lindberg et al., Phys. Rev. B 39, 2890 (1989).

[2] J. Tarascon et al., Phys. Rev. B 37, 9382 (1988).

110

Page 119: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[3] H. Heinrich, G. Kostorz, B. Heeb, and L. Gauckler, Physica C 224, 133 (1994).

[4] C. Renner et al., Phys. Rev. Lett. 80, 149 (1998).

[5] D. Feng et al., Science 289, 277 (2000).

[6] J. Tallon and J. Loram, Physica C 349, 53 (2001).

[7] H. Ding et al., Phys. Rev. Lett. 87, 227001 (2001).

[8] M. Norman et al., Nature 392, 157 (1998).

[9] A. Loeser et al., Science 273, 325 (1996).

[10] H. Ding et al., Nature 382, 51 (1996).

[11] B. Hoogenboom et al., Phys. Rev. B 67, 224502 (2003).

[12] A. Damascelli, Z. Hussain, and Z.-X. Shen, Rev. Mod. Phys. 75, 473 (2003).

[13] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 59, 11191 (1999).

[14] A. Zakharov, I. Lindau, and R. Yoshizaki, Physica C 398, 49 (2003).

[15] H. Murakami and R. Aoki, J. Phys. Soc. Jpn. 64, 1287 (1995).

[16] S. Pan, E. Hudson, J. Ma, and J. Davis, Appl. Phys. Lett. 73, 58 (1998).

[17] S. Misra et al., Phys. Rev. Lett. 89, 87002 (2002).

[18] T. Hasegawa and K. Kitazawa, Jpn. J. Appl. Phys. 29, 434 (1990).

[19] K. Kitazawa, Science 271, 313 (1996).

[20] S. Sugota, T. Watanabe, and A. Matsuda, Phys. Rev. B 62, 8715 (2000).

[21] A. Yazdani et al., Phys. Rev. Lett. 83, 176 (1999).

111

Page 120: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[22] E. Hudson et al., Science 285, 88 (1999).

[23] S. Pan et al., Nature 403, 746 (2000).

[24] E. Hudson et al., Physica B 329, 1365 (2003).

[25] E. Hudson et al., Nature 411, 920 (2001).

[26] M. Asensio et al., Phy. Rev. B 67, 14519 (2003).

[27] J. Campuzano, M. Norman, and M. Randeria, in Nobel Lectures, Physics 1981-

1990, edited by K. Bennemann and J. Ketterson (Springer-Verlag, ADDRESS, 2004),

Chap. Photoemission in the High-TC Superconductors.

[28] D. Singh and W. Pickett, Phys. Rev. B 51, 3128 (1995).

[29] M. Kugler et al., Phys. Rev. Lett. 86, 4911 (2001).

[30] A. Hanna and M. Tinkham, Phys. Rev. B 44, 5919 (1991).

[31] S. Chakravarty, S. Sudbo, P. Anderson, and S. Strong, Science 261, 337 (1993).

[32] O. Anderson, A. Lichtenstein, O. Jepson, and F. Paulson, J. Phys. Chem. Solids 56,

1573 (1995).

[33] Y. Kohsaka et al., Phys. Rev. Lett. 93, 97004 (2004).

[34] Y. Ando et al., Phys. Rev. Lett. 87, 17001 (2001).

112

Page 121: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Chapter 6

Conclusion

We began this thesis motivated by the goal that scanning tunneling microscopy can teach

us new information about the high temperature superconductors. The characteristic length

scales of the HTS phase diagram are all on the atomic length scale- the coherence length for

superconductivity is ≈ 16A in-plane, and the various ordered states discussed in relation to

the HTS phase diagram typically have length scales on the order of tens of Angstroms. STM

is capable of measuring the density of states on atomic length scales, and correlating these

measurements with topographic data of the surface it takes the spectroscopic measurements

on. Through the course of this thesis, we have demonstrated how the nanoscale view of the

HTS provided by STM has lead to new discoveries throughout the phase diagram.

In the superconducting state, we investigated the spectacular changes that occur to

the parent d-wave superconducting state upon the introduction of certain types of defects.

Although the existence of the novel zero-energy state found at 110 step edges could be

predicted using the existing theory, we confirmed that it does form a one-dimensional state

at the interface. More importantly, the absence of this state at 110 twin boundaries,

where the established theory predicted it should also occur, opens up the issue of how

twin boundaries affect the superconducting state. As it is exceptionally difficult to grow

untwinned samples, and the effect of twin boundaries on other experiments is still debated,

113

Page 122: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

a definitive answer to this issue is needed. Although our data lend preliminary support to

the idea that 110 twin boundaries nucleate a subdominant s-wave superconducting order,

in itself an exciting possibility, more data will be needed to make a definitive statement. The

most promising line of inquiry would be making measurements on 100 oriented defects,

which are not expected to give rise to either of these novel effects.

Far away from these extended structural defects, we found that the in-gap electronic

states form modulated patterns in the density of states at T = 40K. Although the obser-

vation of modulated patterns over seemingly pristine HTS samples was not a new discovery

in itself, as they had been seen before at low temperatures, it represented a landmark step

forward for STM. This was the first time detailed spectral mapping had been accomplished

by STM on a correlated electronic system at elevated temperatures. That we were able

to make meaningful measurements, which took days to acquire, with the potentially fatal

complication that the tunnel junction becomes unstable due to thermal fluctuations, opens

up the possibility for exploring the temperature dependence of the density of states with

atomic precision in any number of materials systems for the first time. The analysis of the

modulations in terms of perturbative corrections to the density of states had also been in-

vestigated thoroughly in the literature. We demonstrated that the scattering interference

scenario qualitatively described the changes seen in these patterns between 4K and 40K,

including the disappearance of 5 of the 6 modes seen in the low temperature data due to

thermal fluctuations. Perhaps even more importantly, it lent credence to the idea that the

formalism developed in doing these calculations can be used to diagnose, if not describe, any

modulated pattern seen by STM.

In contrast to the superconducting state, we lack even a rough phenomenological under-

standing of the physics underlying the pseudogap regime. For this reason alone, our discovery

that the in-gap electronic states of the pseudogap regime form disperionless bond-oriented

modulations is perhaps the most significant work presented in this thesis. Having understood

114

Page 123: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

the presence of the patterns in the superconducting state to be a consequence of scattering

interference, which nominally requires the existence of quasiparticles, it came as a shock that

these patterns were present above TC , where quasiparticles have been thought to not exist

at all. The scattering interference formalism developed in attempting to understand the

superconducting state data proved to be invaluable in diagnosing the possible origin of these

modulations. We had hoped to be able to identify which of the candidate electronic Green

functions proposed for the pseudogap state were consistent with the observed modulations.

The failure of any Green function to describe the dispersionless modulations led us to the

conclusion that the problem lay in the scattering interference model itself. Moreover, the

reasons for this failure suggested that these patterns represent the signature of some locally

ordered state whose appearance correlates with the pseudogap energy scale.

Future works will have to determine whether the observed local order is the signature of a

thermodynamically significant phase. Our measurements lead to the tantalizing possibility

that this local order causes the pseudogap in the density of states (or vice versa), and,

coupled with similar results in other cases where superconductivity is suppressed, that this

order competes with superconductivity. However, we have merely determined that a single

point on the phase diagram shows this local order in one HTS compound. We still do

not know whether the appearance of the local order correlates with the appearance of the

pseudogap in the density of states throughout the phase diagram. Even if the local ordering

does not represent a thermodynamically significant phase, they could still be important to

the study of the HTS by explaining anomalies measured by other techniques. Ultimately,

the importance of this local ordering to the study of HTS lies in determining under what

conditions these patterns are seen, and what anomalies in other techniques their properties

are consistent with.

In the last experiment presented in this thesis, we measured, for the first time, the density

of states of a single CuO2 plane at the surface of a high temperature superconductor. The

115

Page 124: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

highly correlated electronic behavior of the cuprates is thought to originate on these planes,

which are common to all the HTS materials. Although density of states measurements are

typically made using surface sensitive techniques on samples terminated by planes other than

the CuO2 plane, measurements made on the BiO surface of Bi2Sr2CaCu2O8+δ do appear

to be representative of the subsurface CuO2 planes. We thus would expect to measure

the same density of states when tunneling directly into a CuO2 plane at the surface. We

found, however, that the tunneling conductance is remarkably different in this situation.

This indicates that the process of removing the other planes, which we had thought would

simplify the process of measuring the properties of the HTS using STM, instead alters their

electronic state. However, the surface CuO2 plane has an identical crystal structure to the

structure of the plane in the bulk of the material, suggesting that the process of exposing the

CuO2 plane does nothing more than change its doping level. Although more experiments are

required to determine where on the cuprate phase diagram the surface CuO2 plane is, this

geometry might still be a new way to probe the HTS, which are nothing more than doped

CuO2 planes, without having to deal with the added complication of tunneling through the

other crystallographic layers in these compounds.

Through the course of this thesis, we have illustrated various techniques that will prove

valuable to the study of strongly correlated electron systems with STM in the future. In

general, observing how a material responds to a perturbation by measuring the density of

states near specific impurities, and comparing these results, can reveal information about the

correlations present in a material (Chapter 3). Determining whether there is a modulated

spatial structure to the density of states at fixed energies, and seeing how these modula-

tions evolve as a function of energy can help determine the electronic state of the sample

(Chapters 3 and 4). Finally, the ability to carefully correlate density of states information

with topographic data can reveal experimental situations that are completely inaccessible to

other techniques (Chapter 5). Most importantly, we have demonstrated that STM can be

116

Page 125: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

used to map the temperature dependence of the density of states with unprecedented spatial

resolution. With this arsenal, STM appears poised to make a glut of important contributions

to the study of any number of correlated electron systems with short characteristic length

scales.

117

Page 126: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Appendix A

STM Methods

A.1 Construction

The basic requirements to make the tunneling measurements discussed in this thesis are that

the surface be atomically flat and clean, and that a tunnel junction can be established and

stably maintained for days at a time. Accordingly, the bulk of any STM system consists

of subsystems designed to prepare a sample surface for measurement, then bring a sharp

metallic wire (the tip) within a few Angstroms of the surface, and finally maintain the tip-

sample junction under unchanging conditions for several days. The two home built STMs

used to acquire the data presented in this thesis functionally have the same subsystems:

instruments for preparing clean metal surfaces, vacuum pumps to keep their surfaces clean,

a piezoelectric scan head that brings the tip close to the sample surface, cryostats to main-

tain a stable temperature, and spring-based vibration isolation to maintain the mechanical

stability of the tunnel junction. In this section, I will review how each of the two home-built

microscopes used in this thesis- one, a low temperature STM (LTSTM) and the other, a

variable temperature STM (VTSTM)- deals with each of these issues. The precise construc-

tion details of these instruments appear in the theses of Dan Hornbaker (LTSTM) [1] and

Michael Vershinin (VTSTM) [2], and will not be repeated here.

118

Page 127: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Both STMs share similar subsystems for sample preperation. Two types of samples are

routinely examined with these machines- noble metals (such as Au(111)), and metal oxides

(such as HTS). To clean the noble metal surfaces, the samples are first heated, for the

purpose of making the second step, sputtering, more effective. The LTSTM accomplishes

this through the resistive heating of a ceramic heating element located behind the sample

inside the sample holder. The VTSTM, instead, relies placing the sample inside an e-beam

heater, in which current is passed through a filament and then the hot electrons accelerated

at the back of the sample holder by means of an electric field. After heating, the sample

surface is ballistically sputtered clean. Ar gas is leaked into the chamber and ionized by

passing current through a filament. An applied voltage between the filament and the sample

surface accelerates the ionized Ar atoms at the sample. They collide with the atoms at

the surface, and ballistically knock them off, thus cleaning the surface. Finally, the now-

disordered surface is heated again to make the surface layer slightly molten. This causes

the atoms to rearrange themselves, and, upon quenching the sample to the temperature at

which measurements will be made, produces an atomically flat surface. Metal oxide surfaces

cannot be prepared by sputtering and annealing because most are doped with atoms, such

as Oxygen, that will leave the sample upon heating. Instead, a metal post is glued on to

the top of these samples and knocked off in vacuum, literally cleaving the samples in half.

This exposes a surface that is clean, having previously been confined to the interior of a

solid. In layered materials with weak interplane coupling, this also produces atomically flat

surfaces. Finally, both systems have turbomolecular pumps and ion pumps to maintain

background pressures of ≈ 5 × 10−10torr to prevent recontamination of these clean surfaces

while measurements are being made.

Once the sample surface has been prepared, the STM tip must be brought from an

essentially arbitrary macroscopic distance away from the sample to within a few Angstroms

of the surface in a controlled fashion. Both the microscopes employ what is referred to as

119

Page 128: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B

Figure A.1: (A) A Besocke-style STM heads was used in both the LTSTMand VTSTM designs. The three tubes at the perimeter are the so-called legpiezos, and the one in the center is the so-called scan piezo (from Ref. [2]).(B) The sample holder consists of three ramps which sit on top of each of thethree leg piezos, and a center part where the sample sits.

a Besocke-style scan head to accomplish this (Fig. A.1a). [3] The scan head consists of

four piezoelectric tubes, three placed at the vertices of an equilateral triangle, and one in

the center. All the tubes have been metalized on their inside, and in four quadrants on

their outside. Application of a differential voltage across two opposing quadrants produces

a deflection of the tube along that axis, and the application of a voltage between the outer

electrodes and the inner one produces an elongation of the tube. The central tube (scan

piezo) carries the STM tip, and can be used to laterally scan the tip, and change the height

of the tip in response to the feedback loop. The outer tubes (leg piezos), on the other

hand, both carry the sample holder, and, through the electrically isolated balls at their ends,

provide the sample bias. The sample holder consists of a disk whose outer part consists of

three sloped ramps, underneath which sit the leg piezos, and whose inner part contains the

sample (Fig. A.1b-d).

To bring the tip to a distance where electrons can tunnel between the tip and the sample,

120

Page 129: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

µN

kdd

Figure A.2: The procedue used to approach the sample operates like an inversepizza-toss. Shown are one of the three piezo legs and the sample holder/ rampassembly.

the legs are used to walk down the three ramps that sit on top of them (Fig. A.2). Walking

the sample towards the (comparably stationary) tip can be accomplished by suddenly con-

tracting the legs to disengage them from the surface of the ramp. Before the sample holder

falls back onto the legs, they are quickly deflected sideways. When the sample holder once

again makes contact with the legs, the contact points are located further ’down’ the ramps.

The voltage applied to the legs is then slowly relaxed to zero to restraighten the legs. In

this way, it is possible to make single ≈ 50A steps towards the sample. Finally, the feedback

loop is engaged, and the scanner extends the tip through a range of ≈ 200A. If, at any point

in this extention, tunneling current starts to flow, the feedback loop stabilizes the tip a few

Angstroms above the sample surface, and the approach is complete. If the tip extends to

the maximum of the scan piezo’s range, the feedback loop is opened, the tip retracted, and

another step is taken.

Having moved the sample to a position where we can find the surface by extending the

scan piezo, the STM must finally ensure that the tunnel junction remains stable. Slight

changes in position can have catastrophic effects, as the tip is merely 6A away from the sur-

face, and tunneling current depends exponentially on the tip-sample separation. This effect

is small for topographic measurements, as the feedback loop can compensate for changes in

121

Page 130: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

tip-sample separation, but can make spectroscopic measurements impossible, as the feed-

back loop must be opened to take spectra. First, the temperature of the microscope stage

and sample must be kept constant to within 10mK, as the assymetric thermal expansion

of the scan piezo/ tip assembly compared to the leg piezo/ sample assembly can lead to

significant changes in tip-sample separation. Second, the microscope stage cannot vibrate.

To aid in vibration isolation, the entire experiment is set on top of an optical table inside of

an acoustic room. The optical table prevents sudden shocks in the floor, such as dropping

a hammer, from propogating into the system, and crashing the tip into the sample. The

acoustic room stops ambient room noise from vibrating the tip-sample assembly (STMs with

poor acoustic shielding can make excellent low-bandwidth microphones). However, the bulk

of the low-frequency vibration isolation, and the thermal components, have been treated

differently in the two microscopes used here.

The way the LTSTM and VTSTM solve these problems have dramatically influenced

how they were constructed. In the LTSTM, the microscope assembly is at the end of a

large pendulum which sits in the sealed cavity of a liquid Helium dewar (Fig. A.3). [4]

An exchange gas intruduced into this cabity helps maintains the microscope at a constant

T ≈ 4.5K. Because of the large thermal mass that remains cold, the microscope assembly

doesn’t warm up even when refilling the dewar. This means that the LTSTM is capable of

examining a single sample for months at a time. The entire pendulum assembly is attached to

the bottom of the vacuum chamber under which it sits using vacuum bellows. This acts like a

spring, damping vibrations from traveling from the vacuum chamber into the pendulum. In

addition, the pendulum is suspended using a set of three low-vacuum bellows whose internal

volume is connected to a small gas (or vacuum) reservoir. The pressure of the reservoir

can be changed, raising and lowering the pendulum. In addition, these reservoirs act like

dashpots for the bellows, which act like springs. Combined, they provide excellent low-

frequency vibration isolation, with tunable dissipation, for the microscope stage. However,

122

Page 131: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Optical

Table

Helium

Dewar

Isolation

Bellows

Vacuum

Pumps

Pendulum

Microscope

Exchange

Gas

Load Lock/

Cleaving Chamber Surface Prep Chamber

Sample

Manipulation

A B

Figure A.3: (A) This schematic diagram of the LTSTM highlights the key partsdealing with surace preperation, vibration isolation, sample manipulation, andtemperature regulation (adapted from Ref. [1]) (B) The microscope stage ofthe LTSTM sits on a linear motion stage which lifts the ramp and the sampleholder, which plugs into the ramp (from Ref. [1]).

due to the physical size of this setup, large manipulators must be used for sample transfer,

resulting in relatively long turnaround times.

The VTSTM design is significantly more flexible (Fig. A.4a). [5] It employs a flow-

through cryostat, in which a small volume of liquid helium is cycled through the non-vacuum

side of a Copper cold finger. The thermal mass of the cold finger is relatively small, so its

temperature can be adjusted fairly quickly by either adjusting the Helium flow rate or by

flowing current through a heating element located on the vacuum side of the cold finger.

A feedback loop is employed to control the flow of power to the heating element, allowing

us to maintain a constant temperature on the cold finger. However, the temperature of

the microscope stage increases significantly on refils of the Helium dewar, meaning that the

time available to take measurements is limited by the amount of Helium the dewar can

123

Page 132: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

store. The microscope assembly itself consists of two sets of heat shields anchored to the

cold finger (Fig. A.4b). They shield the microscope stage, which sits inside, from radiative

heating. The cooling of the stage assembly occurs by turning a screw that pins the stage

against the heat shield. The screw will be turned in before a new sample is placed on

the stage to optimize cooling. Once the sample has thermalized, the screw is turned out,

and the microscope stage hangs from the cold finger by three springs at its perimeter. The

temperature of the microscope stage is now a balance between cooling through these springs,

and the heat leak of having electrical connections between the microscope and the outside

of the chamber (which are made with .001′′ diameter stainless steel wire). These springs

serve as the low frequency vibration isolation, with some additional magnetic damping.

This damping is provided by magnets that are attached to the microscope stage and induce

eddy currents in the Copper heat shields when the microscope stage moves. Because of the

compact construction, complete access to the entire chamber, including the mircoscope stage,

is provided by a single multimotion wobble stick. This setup allows for variable temperature

operation while maintaining good vibration isolation, and the compact construction results

in fast turnaround times.

A.2 Operation

A standard procedure was used to prepare both tips and samples in order to maximize the

probability that a good tunnel junction could be established. The goal of the tip preparation

procedure is to ensure that the STM tip is simultaneously sharp, stable, and clean before

moving on to measuring the samples of interest. This was accomplished by modifying the

tips on clean metal surfaces. In the case of the LTSTM, where the tip cannot be replaced

in situ, field emitting the Ir tip to clean Cu(111) and Ag(111) surfaces was necessary, as it

typically had significant debris at its apex from earlier experiments, or the hard oxide that

124

Page 133: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Optical

Table

Vacuum

Pumps

Load Lock/

Cleaving ChamberSurface Prep Tools

Flow-Through

Cryostat

Dewar

Heat Shields/Microscope

Sample

Manipulation

A B

Figure A.4: (A) This schematic diagram of the VTSTM highlights the keyparts dealing with surace preperation, vibration isolation, sample manipula-tion, and temperature regulation. (B) This is a schematic diagram of themicroscope stage, the thermal anchoring screw, and the inner heat shields,which are anchored to the cryostat. Not shown are the outer heat shields, andthe three springs used for vibration isolation(from Ref. [2]).

forms on exposure of Ir to air. A tip that has become dirty is first held close to the sample

surface with a large applied bias of +200V , which accelerates the electrons emitted by the

sample towards the apex of the tip. This causes large, sudden changes in the field emission

current (≈ 1 − 10µA), leading to a catastrophic event where the tip either crashes into the

sample (either the surface or tip melts and joins the other) or current drops to zero (the tip

melts and gets shorter, or drops off onto the surface). Afterwards, the microscope walks to

a different part of the sample, and pristine regions of the metal surface are imaged with the

new tip. Small modifications to the tip can be made by lightly poking it into the sample

surface. The procedure is repeated until a satisfactory tip results. The need for harshly

preparing tips in the LTSTM is alleviated due to its ability for insitu changing of the tip. In

order to avoid having to harshly prepare tips, commercially etched PtIr (90/10) tips were

used, as they are both sharp and do not form a hard oxide while being stored in ambient

conditions. Lightly poking the tip (by ≈ 10A) into prepared Au(111) surfaces is sufficient

to remove the water that naturally builds up on even inert metals in contact with air.

125

Page 134: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Each tip was then checked against a set of criteria which at first seem arbitrary, but have

led to the repeatable establishment of good tunnel junctions. First, the tunneling current is

confirmed to depend exponentially on tip-sample distance. This ensures that we really have

a vacuum tunnel junction. Second, the density of states of the metallic surface is verified to

be roughly flat, with a clearly visible surface state. This ensures that the tip itself is metallic,

and can reliably take density of states spectra. Finally, the tip is used to image step edges

running in four orthogonal directions. This ensures that the tip doesn’t have artifacts, for

example, from a double tip. The stability of the tip is tested as an accidental byproduct of

this procedure by requiring that no tip changes (and the corresponding sudden change in

current) occur while acquiring these images.

The Bi2Sr2CaCu2O8+δ samples examined in this thesis were all processed in a way to

yield reliably flat, clean cleaves. First, the samples themselves were chosen to be large

enough that the approach procedure could find the samples, but small enough that they

cleaved straight across the sample (typically 1− 2mm on a side, 0.5mm thick). Second, the

mechanical stress of cleaving a sample requires the sample to be the weakest joint in the

sample holder- sample- post sandwich. As a result, we must choose the strongest bonding

agents possible. We chose H20E from Epotek to join the sample to the sample holder, as it

is the strongest conducting epoxy commercially available. The cleaving post was attached to

the sample surface using TorrSeal, an insulating epoxy that is stronger than H20E. In extreme

cases, H74F from Epotek could be used for both purposes, with the electrical contacts to

the sample being painted on using H20E after mounting the sample. Both joints were cured

at 100C, which, although it might result in some oxygen loss, produces significantly better

cleaves than lower temperature cures. In addition, when mounting the sample, care must be

taken not to pot it with epoxy. Finally, because the best cleaves come from samples which

already have a flat surface to attach the cleaving post, samples must often be pre-cleaved a

few times on the bench. The sample is then inserted into the vacuum chamber and cleaved

126

Page 135: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

at a base pressure of 1e−9torr or better. This ensures nothing collects on the sample surface

when a new layer of the crystal is exposed in the cleaving process. Finally, the samples are

allowed to thermalize on the STM stage for a day. This is required because the sample is

cooled only by the legs of the microscope stage, which conducts heat very poorly. Even a

10mK change in sample temperature over the period of a day shows up as significant drift

of the tip-sample junction during open-feedback loop measurements, such as spectroscopic

measurements, and makes them difficult to take accurately.

The final step in setting up an experiment is the microscope approach. This sensitively

depends the voltages supplied to the piezos, and the parameters for the feedback loop.

Experimentally, the voltages supplied to the leg piezos were the smallest ones that walked

the sample towards the tip without significantly skewing the ramp off-axis, which can cause

the tip to miss the sample and find the sample holder. Notably, the reliability of walking is

determined by the ability to unstick the balls at the end of the leg piezos from the sample

holder. The minimum vertical deflection d (Fig. A.2) is given by the force balance equation

kd > µmg, where k = Y A/h ≈ .05N/nm is the spring constant for the piezo, A = 7mm2

is the cross sectional area of the piezo, h = 12mm is the height of the piezo, m ≈ 20g is

the mass of the sample holder, µ ≈ 10 is the coefficient of sticking, and g = 9.8 ms2 is the

gravitational constant. In practice, the minimum deflection that will serve as the threshold

for reliable step motion is 400A, or about 80V for piezos with a sensitivity of 5A/V . Here

we have assumed that the slew rate of the power supply causing this motion, which will

be set by the output resistance of the supply coupled with the capacitance of the piezo, is

sufficiently fast for this force to be considered an impulse (10V/µs was seen to be adequate).

Once the vertical motion of the tube is set to be larger than the threshold, almost any

horizontal step can be taken. The angle of the ramp translates this horizontal motion into

a shortening of the tip-sample distance. Even when done carefully, this distance forms a

distribution whose width is of the order of the step size itself. Because the full extention of

127

Page 136: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

the scan piezo must be larger than this distance, lest the tip crash into the sample, relatively

small horizontal deflections were used (500 − 1000A, which translates to a reduction of

the tip-sample separation by 25 ± 25A to 50 ± 50A per step). Feedback loop parameters

for the approach were selected after attempting a large number of approaches on a clean

metal surface. Parameters that extend the tip too fast result in ringing when the sample is

found, which leads to tip instability. Parameters which extend the tip too slowly lead to an

overextention of the tip, and introduce the possibility for the tip touching the sample. Both

these extremes result in debris from the tip being deposited on the approach area, and can

thus be tested for. Fortunately, both the feedback loop parameters and the voltages supplied

to the piezos need only be determined exhaustively once.

In practice, the dominant mode of failure for these experiments is an unstable tunnel

junction. The two primary cultprits appear to be the tip and the samples. Even after

preparing the tip, changes can occur at its apex, for example, from the thermal shock of

putting a sample on the microscope. We believe, more often, tip-related instabilities result

from preparing a dull tip. If the tip is dull, it’s possible to be tunneling out of two different

parts depending on the slope of the sample surface. Hence you could end up preparing one

part of the tip, and tunneling into the experimental sample with a dirty part. This possibility,

which was only circumstantially confirmed, led us to use electrochemically etched tips that

were gently prepared, instead of tips that had become dull through field emission.

Failures related to the quality of the sample surface currently account for the majority of

rate of failure. The surface must be flat, with no irregularities resulting from cleaving from

two different parts of the crystal. To a certain extent, this can be detected by simply looking

at the sample after cleaving and putting it on the microscope stage. The surface itself must

also have a contact resistivity that is less than the typical resistivity of the tunnel junction.

This too can be checked by measuring the bulk resistivity of a sample and hoping that there

is no surface effect. However, currently, the overwhelming majority of our approaches do

128

Page 137: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

not work for a different reason. The failures derive from an interaction between the tip

and loose debris on the surface. The source of this debris does not appear to be related to

the background pressure of the chamber, as its concentration is insensitive to the pressure

under which the sample is cleaved. It also does not appear to come from outgassing as the

sample stage is warmed up when coming into contact with a room temperature sample- the

concentration is also insensitive to the temperature we take measurements at. We believe it

is an intrinsic property of the crystals themselves. Under the same conditions, some cleaves

produce cleaner surface. We are able to diagnose the quality of the samples to an extent using

high magnification phase cobntrast optical microscopy. Samples with any obvious surface

irregularities never work. However, of the ones which do not have any surface irregularities,

we have not yet been able to develop a procedure that determines which samples might

produce clean cleaves and which ones never will.

A.3 Limitations

Assuming we have established a stable tunnel junction, there are three limitations to our

ability to take good data. First and foremost are those that limit the signal to noise ratio

of the tunneling current, which can result in useless, noisy data. Then are those limiting

the resolution of the data, which can render a perfectly quiet data set useless. The final

limitation is time itself, which is currently the factor limiting us from taking even larger

data sets. Each limitation applies differently to topographic and spectroscopic data, and

will be treated separately.

The signal and noise in topographic measurements is the dc level of the tunneling current

and the peak-peak ac level on top of it, respectively. However, the feedback loop, whose

output we record as the height of the tip, has an extremely limited bandwidth (10Hz) and

thus the noise is unaffected by high frequency levels on the tunneling current. The relevant

129

Page 138: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

low-frequency ac noise comes from two sources: noise originating from the electronics, which

is independent of the setpoint of the feedback loop, and noise originating from vibrations,

which scales with the tunneling current.

Electronic noise in topographic measurements has two sources. High gain transimpedence

amplifiers (typically 1 × 1010V/A) have broadband noise whose amplitude increases with

the magnitude of the shunt capacitance on their input. The relevant capacitance is the

capacitance between the current line to ground and the bias lines to ground (≈ 100pF ).

Measured on the bench, with a 120pF shunt capacitance, the total noise at a bandwidth

of 1kHz is measured to be 2.5pArms for the DL Instruments Model 1211 on the 109V/A

setting (used in Chapters 3 and 5 of this thesis), 0.9pArms for the same amplifier on the

1010V/A setting (not used), and 1.1pArms for the Femto LCA-1K-5G (Chapter 4). The

Femto amplifier has the dual advantages of shorter cable length and fewer connections,

which reduces input capacitance from ≈ 120pF to ≈ 70pF , and results in better performance

(0.6pArms) than the DL Instruments amplifier. However, this noise has most of its weight at

frequencies outside of the bandwidth of the feedback loop, and can be ignored (only 0.1pArms

or less at the frequncies of interest). The limiting electronic noise actually arises from the

triboelectric coupling of vibrational noise into the bias and tunneling lines (≈ 1.5pArms).

The result is ≈ 4.2pA of peak-peak low frequency noise that does not depend on how much

dc current passes through the amplifier.

Although vibrations result in small changes of the tip-sample separation, the tunneling

current varies exponentially with tip-sample separation, making vibrations a major source

of topographic noise. The vibration isolation scheme has three primary components: an

acoustical room which attenuates air-borne sound above ≈ 100Hz, an optical table which

attenuates ground-borne vibration above ≈ 100Hz, and the springs suspending the micro-

scope stages which attenuate vibrations above ≈ 5Hz. The relatively well-isolated STMs

used here had vibrational noise at three primary frequencies: ≈ 5Hz and ≈ 42Hz arising

130

Page 139: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Frequency (Hz)

Curr

ent (r

ms, pA

)

Building Vibrations

Roughing Pump

Air handling system

0.0

0.2

0.4

0.6

0.8

1.0

0 20 40 60 80 100 120 140 160

Figure A.5: This shows the power spectrum of the tunneling current acquiredin open-feedback loop conditions with a I = 1nA, V = 100mV junction onAu111. The highest three noise peaks are labeled. The pumps are normallyturned off. The total vibrational noise was ≈ 10%.

from the vibrations of the building, and ≈ 2kHz arising from the resonance frequency of

the piezoelectric tubes (Fig. A.5). These accounted for an ac vibrational noise that was

typically 10% (peak-peak) of the dc tunneling current, with most of the power being at

low frequencies. Notably, additional changes in the tip-sample separation arising from the

voltage supplied to the piezos, which is heavily filtered (to 10Hz for the legs, and 1kHz for

the scanner), is not an issue for our experiments.

To maximize the signal to noise ratio on the tunneling current itself, the setpoint is set

such that the proportional noise, from vibrations (10%), is as large as fixed noise, from

triboelectric contributions (≈ 4.2pA). For a dc tunneling current of 50pA, we should have

6.5pA =√

(0.1 ∗ 50pA)2 + (4.2pA)2 of peak-peak ac current noise, yielding a maximum

height noise of 1A ∗ log(1 + 6.5pA/50pA) = 53mA in the topographs (typical atomic corru-

gations on Bi2Sr2CaCu2O8+δ are of the size 200mA). Increasing the tunneling current can

reduce the relative contribution of the triboelectric noise, and effectively reduce the height

131

Page 140: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

noise to 40mA. However, decreasing the tip-sample separation by increasing the tunneling

current often results in increased interaction of the tip with impurities on the sample sur-

face. These instabilities show up as spikes in the tunnel current, and, if there is a change in

the tip, a sudden change of height of the topograph. Altough these spikes are tolerable for

topographic data, they limit our ability to take spectroscopic measurements.

The role of noise in conductance measurements is much more complicated. In a typical

conductance measurement (Fig. A.6), an AC lockin provides a clean sin wave at a specific

frequency that is applied to the sample bias (EG&G models 124A and 5209 were used in

Chapters 3 and 5, and in Chapter 4, respectively). The AC response of the tunneling current

is fed to the transconductance amplifier, which goes to the lockin. The lockin measures the

amplitude of the wave at a specified phase relative to the reference wave using a demodulation

sin wave. The phase must be set to measure only the resistive response, and not the capacitive

response, of the tunnel junction. This can be accomplished by offseting the tip out of

tunneling range, such that the tip is fully extended and no tunneling current results, and

then opening the feedback loop. In this condition, the reponse is fully capacitive, and the

lockin phase can be set to measure the amplitude of the ac response of the tunnel junction out-

of-phase with the capacitance (i.e., the conductance). The approximate signal for tunneling

conductance measurements is then dIsignal(V ) = 2.8 ∗ dVrefI/V peak-peak, where dVref is

the rms amplitude of the bias modulation, I is the DC tunneling current, and V is the

bias. This signal, typically of order 2.8pA peak-peak, is more sensitive to noise than the

topograph, where the signal level was over an order of magnitude larger. On the other hand,

only the noise sources with a frequency close to that of the sin wave contribute. This noise

again has two primary sources: electronic, and vibrational.

The primary source of electronic noise can come from the process of making the lockin

measurement itself. The input to the lockin amplifier has a strong band-pass filter on it,

which filters the ac response of the tunnel junction to a 3dB bandwidth of the order of

132

Page 141: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

dV

Sample

Tip

+

V

A/V

Amplifier

I+dI

Bandpass

φ

Phase Shifter

X

Low-pass

Lockin Amplifier

Figure A.6: This schematic diagram highlights the five main functional partsof a lockin amplifier: the ac voltage source, the phase shifter, the bandpassinput stage, the multiplier, and the low-pass output stage.

10Hz. To a close approximation, the input can thus be treated as A sin(2πft). To measure

the amplitude of this wave, it is multiplied by a demodulating sin wave with a phase shift,

giving A sin(2πft)×2 sin(2πft+φ), which, after simplifying, is A× (cos(φ)+cos(4πft+φ)).

The first of these terms is a dc level corresponding to the amplitude of the input wave at a

phase φ, selected to measure the contribution from the conductance. The second is an equally

large 2f contamination of this measurement, consisting of a component A cos(φ) cos(4πft)

whose amplitude is also proportional to the conductance, and a component A sin(φ) sin(4πft)

whose amplitude is proportional to the capacitance. This capacitance arises from capacitive

coupling between the tip and bias lines, and is typically ≈ 1pF . At a frequency of 500Hz, it

has an amplitude of f ∗ 1pF = 0.5pArms per mV of dVref , which is equal to the magnitude

of both the dc lockin signal from the conductance and the 2f noise from the conductance

for an I/V = 0.5nS. The amplitude of the 2f noise cannot be decreased by increasing

dVref , as the amplitude of the dc conductive response, the 2f conductive response, and the

capacitive response are all proportional to dVref . It also cannot be decreased by changing

the dc current or bias of the tunnel junction itself, as the 2f conductance noise is equal in

magnitude to the lockin dc conductance term. These 2f components can only be filtered

using the output stage of the lockin, a 4th order low-pass filter with time-constant τ and a

133

Page 142: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

12dB/octave rolloff. This leads to an attenuation of the 2f contaminent by a factor of 10,

100, 300, and 5000 for time constants of 5mS, 10mS, 30mS, and 100mS, respectively. The

effectiveness of this filter cannot be increased by raising f , as this increases the capacitive

response. Although this analysis is specific to analog lockins, digital lockins suffer from

similar high-frequency contamination that depends on how they take Fourier Transforms.

Often the amplitude of the contamination is smaller, but it is distributed over a wider range

of frequencies, including low frequencies, making it harder to filter.

Assuming we have chosen a reasonable time constant, the primary source of electronic

noise in lockin measurements should come from noise on the input stage near the lockin

frequency, which will appear as low-frequency beats on its output. The frequency must thus

be chosen at a point where the background noise levels are featureless, and small. The lockin

frequency of 500Hz is far away from the charcteristic frequencies for triboelectric noise, and

thus represents a good choice. How much noise near the lockin frequency contributes to the

measurements depends sensitively on the bandwidth of the lockin. The limiting bandwidth

is determined by either the time constant of the output stage (a 4th order filter with a

bandwidth set by 1/τ), and or the band-pass filter on the input (a 2nd order filter with a

Q = 10, or a 4th order filter with a Q = 2 depending on the input mode used). The use of

filters with Q values much greater than 10 introduces phase shifts near the lockin frequency

that destabilize the lockin. Assuming noise near these frequencies is flat, these sharp filters

with long tails can be approximated by a hypothetical filter with a perfect cutoff, and an

effective bandwidth (Fig. A.7). For a lockin frequency of 500Hz, the calculated effective

bandwidth is 386Hz for the 4th order input filter, and 300Hz for the 2nd order input filter.

Both these are dwarfed by the effective bandwidths of the output stage- 300Hz for the

τ = 5mS setting, 156Hz for the τ = 10mS setting, 52Hz for the 30mS setting, and 16Hz

for the 100mS setting.

The sources of electronic noise near the lockin frequency are intrinsic to measuring tun-

134

Page 143: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Am

pli

tud

e

Frequency

BW

EBW

ω0

Figure A.7: The effective bandwidth of a bandpass filter tends to be quite abit larger than its 3dB bandwidth. One can be calculated from the other byintegrating the absolute value of A(iω) = ω0× iω/((iω)2 +BW ∗ (iω)+ω2

0)N

over all frequencies ω. In this expression, N refers to one half the order of thefilter, ω0 is the center frequency, and BW is the 3dB bandwidth (= ω0/Q,where Q is the quality factor of the filter).

neling currents. One comes from the noise intrinsic to tunneling discrete charges. So-called

shot noise has a contribution of 4fArms/√

Hz =√

2q ∗ 50pA for a 50pA dctunneling current.

The other cultprit is the amplifier. The measured power spectral density of noise near 500Hz

on the bench with the appropriate shunt capacitance are 3.2fArms/√

Hz for the Femto am-

plifier, 4.3fA/√

Hz for the DL Instruments on a gain of 1010V/A, and 10fA/√

Hz for the

DL Instruments on a gain of 109V/A. These two sources will add in quadrature to give about

5fArms/√

Hz of noise. For the τ = 5mS setting, this corresponds to 250fA (peak-peak) of

broadband low-frequency noise on the output of the lockin, 175fA for the τ = 10mS setting,

100fA for the τ = 30mS setting, and 56fA for the 100mS setting. This contribution is fixed

as a function of dVref and the dc tunnel junction parameters, and can thus be minimized by

changing either quantity. It should also be noted that some tunnel junctions become plagued

by blips in the tunneling current, usually coming from either changes in the tip configuration,

or interaction of the tip with impurities on the sample surface. The spectral content of a

delta function of amplitude a is distributed equally over all frequencies, including the lockin

135

Page 144: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

frequency. This is usually significantly larger than the lockin signal, and invalidates data

acquired within a time 100mS + 2τ of the blip. In principle, if they are infrequenct enough,

and small enough in amplitude, they can be accounted for simply measuring their spectral

density at the oscillator frequency (in fa/√

Hz) and multiplying by the square root of the

effective bandwidth to estimate their contribution to noise.

Vibration of the tunnel junction occur far away from the lockin frequency frequencies,

but still contributes noise because they change the conductance of the tunnel junction itself.

Their contribution to total noise is a flat 10% (peak-peak) of narrow-band low-frequency

noise. Like the 2f noise from the lockin, this also scales with the oscillator amplitude

and the absolute conductance, and can only be filtered out. However, it occurs at low

frequencies, and filtering is problematic becuase of the extremely long time constants that

must be employed to filter low-frequency noise.

The total effect of noise on spectroscopic measurements clearly depends strongly on the

time constant of the output circuit on the lockin. Adding the cumulative effects of noise

together, we get a peak-peak noise of

dInoise = [(7.8EBW × 2qI + SD2amp) +

(7.8dVref2

A2 × ( IV

)2 + (fC)2) +

(0.1dVref I

V)2]0.5, (A.1)

where I is the dc tunnel current, V is the dv bias voltage, dVref is the rms amplitude of the

lockin modulation at frequency f , EBW is the equivalent bandwidth of the lockin output

stage (Table A.1), A is the attenuation factor for 2f noise (Table A.1), q = 1.6e−19C, C is

the capacitance between the current and bias lines, and SDamp is the spectral noise density

in pArms/√

Hz of the amplifier near f (which should be replaced with the measured spectral

density of noise near f if the tunnel junction is unstable). For the τ = 5mS setting, the

dominant noise has a 2f component from the lockin itself and gives 40% (peak-peak) noise

136

Page 145: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Time Constant EBW(Hz) A

5mS 300 10

10mS 150 100

30mS 50 300

100mS 16 5000

Table A.1: The time constant on the output stage of the lockin affects thetotal noise of spectroscopic measurements by limiting the effective bandwidthof the measurement (EBW), and by attenuating 2f contamination by a factorA.

on the signal out of the lockin. Because it is invariant on the other parameters of the

measurement, this setting is almost useless. For the τ = 10mS setting, 2f noise (4% of total

signal) is dwarfed by an equally stubborn narrow-band low-frequency noise from vibrations

that gives 10% (peak-peak) noise on the lockin signal. There is an additional 175fA (peak-

peak) of broadband low-frequency noise from intrinsic sources. The latter becomes a serious

issue when total signal levels are small enough that 175fA > 10% × dIsignal, but can be

overcome by increasing dVref . For larger time constants, the influence of this noise decreases

(100fA for τ = 30mS, 56fA for τ = 100mS), and the measurements are often purely limited

by the vibration levels of the tunnel junction. These values agree within a few percent of

what was actually measured from tunnel junctions on metal surfaces on both microscopes

as measured using a HP3563A spectrum analyzer and a HP34401 voltmeter.

Beyond insuring that there is a clean enough signal to take meaningful data, we must also

make sure that we have can appropriately resolve the features we are searching for. In this

thesis, we have focused on taking spectroscopic measurements correlated with topographic

measurements, both of which can be limited by spatial resolution. Ultimately, spatial reso-

lution is limited by sources of positional error. The voltage noise on the piezoelectric scanner

is around ±2mV (peak-peak) of combined broadband noise from the power supply (filtered

137

Page 146: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

to 1kHz) and triboelectric noise on the voltage lines, and corresponds to a positional error of

about ±10mA. This is over an order of magnitude smaller than the typical distance between

atoms, > 1A, and is therefore never an issue. The voltage noise on the piezoelectric legs is

filtered to 10Hz and contributes even less to positional noise. Finally, horizontal vibrations,

which are much smaller than vertical vibrations (which themselved are only ≈ ±50mA),

also do not contribute to our spatial resolution. In fact, spatial resolution is limited purely

by user-selected pixelation (N × N pixels) and image size (xA × xA). The limitations on

resolution in real-space are then trivial- we can resolve features as small as x/N (Nyquit

limit), and as large as x (image size).

Determining the resolution limits on Fourier transforms of STM data, however, is a more

complicated affair, although it is limited by the same two principles. These limitations are

best illustrated by atttempting to establish error bounds on our estimation of the wavelength

from the location of a peak in a Fourier transform of the data (Fig. A.8). Assuming that

the peak width (δl, in pixels) is less than the peak location (l, in pixels), the wavelength

(s ± δs) is given by

s ± δs = x/(l ± δl)

= (x/l)/(1 ± δl

l)

=x

l× (1 ∓ δl

l). (A.2)

The wavelength and uncertainty are then simply

s = x/l (A.3)

δs =x

l(δl

l+ O(

δl

l)2) = s

δl

l. (A.4)

Notably, resolution can be improved by increasing the image size, which increases l, but not

by increasing N , which only determines the shortest resolvable wavelength (expands dynamic

range). Finally, it is valuable to know the error of the measured width of the peaks in FTS,

138

Page 147: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

which are used to estimate domain size. The full-width half-max 2δl in FTS corresponds to

a domain size of x2δl

, with an uncertainty in this quantity of x2δl±2

− x2δl

= ± x2(δl)2

.

After locating a feature in FTS, we often want to choose an optimal image size x and

pixelation N to learn about its dispersion. Assuming that a periodic feature in a spatial

density of states map has a wavelength s at one energy, and s + ds at another energy, we

want to choose x and N such that the first feature appears l pixels away from the center in

FTS and the second l + dl pixels away (Fig. A.8). This requires s = x/l and s + ds = xl+dl

.

Simplifying, the optimal choice for the image size

x =dl × s2

ds. (A.5)

The choice of N is limited by the Nyquist criterion, which requires 2x/N = s + ds or,

simplifying,

N =2s × dl

ds. (A.6)

A naive choice of dl is simply 1, the minimum to resolve the change in wavelength, but this

will lead to an uncertainty in the change in wavelength that is of the order of ds. Instead,

a better choice is one that’s large enough to observe the breadth of both wavelengths δs,

resulting from either energy broadening or domain size. Assuming we devote δl pixels to

resolving the width, we should pick

dl =ds × δl

δs. (A.7)

Here, the choice of δl should be determined by how well we need to determine the actual

width δs, and 1 is an appropriate choice. The bredth of a given peak in FTS is often the

result of energy broadening, which we discuss next.

For all spectroscopic measurements, we must also ensure that we can resolve the features

we are looking for as a function of energy. For conductance spectra taken at a fixed location

as a function of energy, thermal broadening and the lockin dVref amplitude limit our energy

139

Page 148: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

l

2δl

l+dl

Figure A.8: STM conductance maps with periodic modulations give FourierTransforms with distinct peaks. The peak location at one energy is l, with ahalf-width of δl. The peak location at a different energy is given by l + dl.

resolution. Recall from Chapter 2 that the tunneling conductance

dI(~r, V )

dV∝

∞∫

−∞

dEρS(~r, E)d

dV[f(E + V )], (A.8)

where ρS is the sample density of states at an energy E and position ~r, and f is a Fermi

function. Thermal fluctuations affect our ability to resolve the density of states through the

derrivative of the Fermi function. This can be approximated fairly well by a Gaussian of

full-width half-max 3.5kBT (1.4mV at 4.5K, 12mV at 40K, and 30mV at 100K). Added

in quadrature is the energy broadening from the amplitude of the bias oscillation that the

lockin uses to measure dIdV

. Assuming that we have added a modulation dVrms, the resultant

energy broadening is given by the peak-peak variation in the bias- 2dVrms

√2. Taking care

that it doesn’t significantly impact energy resolution, it is often desirable to increase dVrms

in order to increase the signal-noise ratio of conductance measurements limited by intrinsic

sources, such as shot noise and the noise of the amplifier. In cases where the signal-noise

ratio is limited either by 2f noise or vibrations, increasing dVrms has little impact outside of

degrading energy resolution. In sum, the total energy resolution

±δE =√

3(kBT )2 + 2(dV 2rms). (A.9)

140

Page 149: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

The energy resolution can also impact spatial resolution of spatial conductance maps.

These data are frequently taken at two (or more) fixed energies as a function of position

on the surface, and a Fourier Transform taken to determine the periodicity of modulations

often found on these maps. For example, modulations found in density of states maps often

arise from the elastic scattering of quasiparticles in a band, and thus the modulations have a

wavelength that changes as a function of energy (s(E)). The energy resolution ±δE will then

introduce a width to the peaks in the Fourier Transforms through this dispersion relation, in

effect reducing spatial resolution. For a given energy E (relative to EF ), an energy resolution

of δE will introduce an intrinsic breadth to the measured wavelength s of

δs =ds

dE× δE. (A.10)

Expressed in terms of the more-commonly known dispersion relation E(k),

δs =2πδE

k2

1

dE/dk. (A.11)

The ultimate limitation on the STM experiments presented here is simply time. After

checking a tip, and before cleaving a sample, the lHe dewar is refilled to its maximum

capacity. This defines the maximum amount of time (about 144 hours for the 60l dewars

we used) we have to take data from one area, as refilling the dewar requires us to walk

out of tunneling range, and then re-approach. After refilling, we cleave the sample, and let

it cool on the stage for a day (24 hours) to let it thermalize. The approach is relatively

slow, as we’re trying to move about 0.25mm in 50A steps, and can take another 24 hours.

Finally, even after we’re in tunneling range, we must typically wait another 6 hours for the

dielectric relaxation of the piezoelectric tubes to die down, as they will change the tip-sample

separation in the open-feedback loop conditions needed to take spectra. We have a seemingly

long 90 hours remaining in which to take measurements. This is more than enough for taking

topographs and spectra as a function of energy at a line of points, as done in Chapters 3

and 5 of this thesis.

141

Page 150: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Spatial maps of the conductance taken at several energies, on the other hand, are very

time-intensive due to the sheer volume of data involved, typically 500,000 data points or

more. Assuming we devote a time T to taking one set of data, we can take data at a number

of energies

ε = (T

N2− O)/t, (A.12)

where N is the number of pixels, O is the overhead time per pixel, and t is the time it takes

to make a single lockin measurement (Fig. A.9). The overhead time per pixel O should be

determined only by the inverse of the feedback loop bandwidth. Once we move our tip to a

new point, we must wait 100mS (the characteristic settling time for the feedback loop, which

has a bandwidth of 10Hz) to record the output of the feedback loop for the topographic

measurement. Because we must take spectra in open feedback loop conditions, we must wait

another 100mS after we take a spectra for the feedback loop to settle down again. There

is also the additional overhead of moving the tip to a new pixel (typically 5mS) and the

time to set the bias from the last point taken in the spectrum back to the scan condition

(typically 100mS). While this suggests O = 300mS, our software adds an additional 700mS

per point of overhead (only for spectral maps) to record the data to hard disk, creating a

total O = 1s. The time it takes to take a single lockin reading t should be determined by

how long it takes for the lockin to settle at a new reading when we set a new bias voltage.

For our lockin, this time is 100mS + 2τ , where τ is the time constant on the output stage of

the lockin. In addition, we must be careful to change the bias slowly enough that the lockin

doesn’t detect the change in voltage per unit time as a massive capacitance (which can cause

the lockin to rail), in effect doubling the settling time. Finally, we sample the lockin output

for the time τ . In sum, this makes the time to make a single lockin reading at one voltage

t = 200mS + 5τ . Unfortunately, our software adds an additional 50% to this time, making

t = 300mS + 7τ .

In sum, these limitations make taking data sets of conductance maps at several energies,

142

Page 151: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Time

Lock

in

Ou

tpu

t

Move tip

to new point

Wait for

Feedback

Loop, Record

Topograph

Ramp Bias

To Target

Wait for

Lockin reading

to Settle

Return

bias to

Scan

Condition

Re-engage

Feedback

Loop

Make Lockin

Measurement

1 2 3 4

5

6 7

Figure A.9: This is a rough, idealized diagram represnting the time it takesto take a single point in space of a spectral map- spectroscopy as a functionof position at one (or more) energies. The times 1,2,6 and 7 are part of theoverhead (O in the text)- the amount of time it takes to set up a spectroscopicmeasurement. The times 3, 4, and 5 are parts of the time it takes to take aspectroscopic measurement at one energy (t in the text).

such as presented in Chapter 4, very difficult. In Chapter 4, we attempted to track a

modulation in the density of states with a wavelength of s = 4.7a0 = 18A. Say that we want

to know whether its wavelength changes by ds = 0.2a0 = 0.76A. This forces an image size

of x = 182

0.76A = 432A. We still want to be able to see the atoms on this image, meaning

that the Nyquist limit dictates 2x/N = 3.8A or N > 228. The time constant we choose

is limited by what we want our signal to noise ratio to be, which would ideally be limited

only by vibrational noise, an almost unchangeable 10% noise level. The approximate signal

depends on the dc tunneling parameters I and V , which were chosen to be 40pA and 150mV

because they gave a reasonably small amount of noise arising from tip-impurity interactions.

The remainder of the noise arises from intrinsic sources, and has a total contribution of

100fA for τ = 30mS. In order to ensure that this contribution is smaller than that from

vibrations, we chose dIsignal = 28 ∗ 100fA = 2.8pA (vibrational noise will be 10% of this,

or 280fA). This requirement and the dc tunnel junction parameters force the choice for

143

Page 152: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

dVref = (2.8pA × 150mV )/(40pA × 2.8) ≈ 4mVrms. These choices should yield a noise level

of about 11% total, a t = 300mS + 7τ = 500mS, and an energy resolution that is almost

purely limited by kBT (±δE = ±16mV , where the thermal limit is ±15mV ). Assuming we

spend the entire T = 90 hours = 324000s to taking a single data set, this set can contain

just ε = (324000s2562 − 1s)/(0.5s) = 7.88 ≈ 7 − 8 energies. This kind of analysis was used to

determine our choice of parameters for the data set taken in the pseudogap state in Chapter

4. However, observed signal-noise ratio was worse than 11% due to interactions between

the tip and the sample, which created spikes in the tunneling current. We should have

attempted to estimate the spectral density of this noise, and set dVref to be higher, in effect

compromising a little bit of resolution for a better signal-noise ratio.

A.4 References

[1] D. Hornbaker, Ph.D. thesis, University of Illinois at Urbana-Champaign, 2003.

[2] M. Vershinin, Ph.D. thesis, University of Illinois at Urbana-Champaign, 2004.

[3] J. Frohn, J. Wolf, K. Besocke, and M. Teske, Ref. Sci. Inst. 60, 1200 (1989).

[4] D. Eigler and E. Schweizer, Nature 344, 524 (1990).

[5] B. Stipe, M. Rezaei, and W. Ho, Ref. Sci. Inst. 70, 137 (1999).

144

Page 153: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Appendix B

Elastic Scattering Interference

Calculations

Throughout this thesis, we have made use of the fact that the density of states in the immedi-

ate vicinity of a scattering site reveals information about the parent electronic state which is

inaccessible from examining the density of states over pristine regions. Of particular interest

is whether an or not a defect disturbs the parent electronic state in a perturbative fashion ex-

pected for elastic scattering. [1] To accomplish this, we have compared the measured density

of states to calculations based on Green functions proposed to describe the parent electronic

state of the system. In principle, such a comparison can reveal which of the proposed states

are consistent with the observed STM density of states data. They can also reveal if the

measured data is inconsistent with an elastic scattering description on general terms. In this

Appendix, we will describe the methods we used to perform these calculations.

B.1 Formulation

The elastic scattering of quasiparticles from an impurity can be treated using perturbation

theory. [2, 3, 4, 5, 6] The Green function near an impurity G(~r, ω) is then the sum of the

145

Page 154: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Green function over bare regions of the sample G0(~r, ω) (Fig. B.1a) plus corrections arising

from the elastic scattering. Mathematically, elastic scattering introduces a correction of the

form∫

d2~r′d2 ~r′′G0(~r − ~r′, ω)T (~r′, ~r′′, ω)G0( ~r′′ − ~r, ω) (B.1)

where the scattering matrix

T (~r′, ~r′′, ω) = V (~r′, ω) × δ(~r′ − ~r′′) +∫

d2 ~r′′′G0(~r′ − ~r′′′, ω)T ( ~r′′′, ~r′′, ω) (B.2)

and V is the bare scattering potential. In picture terms, this correction includes three kinds

of diagrams: weak scattering (Born approximation) from a single site (Fig. B.1b), strong

scattering from a single site (Fig. B.1c), and scattering from multiple sites (Fig. B.1d).

Assuming that scattering from spatially extended impurity potentials can be treated as a

spatial superposition of scattering patterns from single defects [5], the scattering matrix can

be simplified to

T (~r′, ~r′′, ω) = V (~r′, ω) × δ(~r′ − ~r′′) + G0(~r′ − ~r′′′, ω)δ(~r′ − ~r′′′)T ( ~r′′′, ~r′′, ω). (B.3)

This expression has the conventional T-matrix form, so we can rewrite the correction to the

Green function as∫

d2~r′G0(~r − ~r′, ω)T (~r′, ~r′, ω)G0(~r′ − ~r, ω) (B.4)

where

T (~r′, ~r′, ω) = V (~r′, ω) × [G0(0, ω)T (~r′, ~r′, ω)]

= (V (~r′, ω)−1 − G0(0, ω))−1. (B.5)

The second set of diagrams must be accounted for in the unitary limit (strong scattering).

However, this is important only when considering the possibility of forming bound states.

In cases where we are interested only in the spatial distribution of electronic states at a

given energy, such as in Chapters 3 and 4 of this thesis, we only need to retain the weak

146

Page 155: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

A B C

Figure B.1: (A) This diagram depicts the weak elastic scattering of a quasi-particle by a point defect. This diagram corresponds to making a Born ap-proximation. (B) This set of diagrams depicts higher-order scattering of thequasiparticle by the same defect. Including all of these corresponds to makingthe T-approximation for strong scattering. (C) These diagrams correspond tothe scattering of the quasiparticle off multiple defects. These cannot be ap-proximated by the linear superposition of multiple single-impurity scatteringevents, although their contribution is often negligible.

scattering term T (~r′, ω) = V (~r′, ω) (the Born approximation). The total Green function

near the impurity now has the simple form

G(~r, ω) = G0(~r, ω) +∫

d2~r′G0(~r − ~r′, ω)V (~r′, ω)G0(~r′ − ~r, ω), (B.6)

where V is the scattering potential.

One effect of introducing an elastically scattering defect into a material is to produce

modulated patterns in the density of states. In the language of Green functions, the den-

sity of states is n(~r, ω) = − 1πImG(~r, ω). Substituting our Born approximation into this

expression,

n(~r, ω) = − 1

πImG0(~r, ω) +

d2~r′G0(~r − ~r′, ω)V (~r′, ω)G0(~r′ − ~r, ω), (B.7)

where the first term will be spatially homogeneous because the electronic wavefunctions have

random phases, but the second will contain modulations because the defect pins the phase of

these wavefunctions. The modulated part of the density of states, which arises from elastic

147

Page 156: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

scattering, has the form

δn(~r, ω) = − 1

πIm

d2~r′G0(~r − ~r′, ω)V (~r′, ω)G0(~r′ − ~r, ω), (B.8)

in real space, and

δn(~q, ω) = − 1

πImV (~q, ω)

d2~kG0(~k, ω)G0(~k + ~q, ω) (B.9)

in Fourier Transform Space (FTS). The second of these forms is particularly useful because

the density of states modulations arising from elastic scattering now contains only two terms:

one term

V (~q, ω) =∫

d2~rei~q·~rV (~r, ω) (B.10)

that is the form factor for the scattering potential, and a term

Λ(~q, ω) =∫

d2~kG0(~k, ω)G0(~k + ~q, ω) (B.11)

which contains all of the wave-interference information.

B.2 Calculation

The challenge is then to calculate the wave-interference information Λ present in the density

of states modulations arising from elastic scattering. The structure factor V acts as a filter of

the wave-interference information, and can be ignored for the purposes of the calculation, as

it only serves to mask the set of information which could be present. The wave-interference

term is complicated to calculate in FTS, as it involves a convolution integral, but is simple

to calculate in real-space, where it reduces to a product Λ(~r, ω) = G0(~r, ω)2. Because most

Green functions are specified in FTS, the strategy will then be to calculate

δn(~q, ω) = − 1

πImFFT [IFFT [G0(~k, ω)]2], (B.12)

148

Page 157: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

where FFT denotes a forward fast Fourier transform, IFFT denotes an inverse fast Fourier

transform, G0 is the unperturbed Green function, and V has been taken to be an all-pass

filter.

The calculation of the density of states modulations arising from elastic scattering thus

boils down to a precise calculation of the Green function on a lattice and a few fast Fourier

transforms. However, care must be taken to perform the calculation in Equation B.12 on

a large lattice. Artifacts from the discrete sampling of the Green function, which contains

singularities in FTS, can contaminate the result (for example, Ref. [7]). The best approach

is to make the matrices in the calculation as large as possible. The calculation will use 3

equally sized matrices which will each occupy 1/4 of the RAM on your system, with the last

1/4 being devoted to the operating system. To determine the size of each of the matrices,

recall that double-precision complex numbers take 16 bytes of memory to store (usually

double in interpreted languages, such as MATLAB). The size of each dimension (n × n) of

each of the matrices should thus be n <√

RAM(bytes)/(16 × 4). On a computer with 1GB

of RAM, 2048×2048 matrices of double precision complex numbers should be used. We will

illustrate the procedure using MATLAB code, as it is fairly easy to understand and translate

into other languages.

The most time-intensive part of the calculation is simply to calculate the Green function

on the n × n lattice. Because we will want to repeat this calculation at several energies, it

pays to calculate quantities which will not change as a function of energy and keep them

stored in memory. This is the reason why we divide the available RAM into four equally

sized matrices. To determine which quantities should only be calculated once, consider the

fairly general form for the Green function G(~k, ω) = (ω − ε(~k) − Σ(~k, ω))−1, where ω is the

energy, ε(~k) is the band structure determined from ARPES, and Σ(~k, ω) is a self-energy that

encapsulates correlations in the system. Both ~k (of size 1 × n) and ε(~k) (size n × n) are

always energy-independent, and some parts of Σ(~k, ω) (size n × n) often are. Because ε(~k)

149

Page 158: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

often involves trigonometric functions of ~k, we start by storing the cosines of ~k instead of ~k

itself:

size=2048;

for n=1:size,

k=4*pi*(n-1)/(size)-2*pi;

cosk(n)=cos(k);

end;

Here, we have taken ~k values between (±2π,±2π) in order to make sure our calculation

spans enough of ~k-space to not ignore Umklapp processes. The two-dimensional matrix of

ε(~k) can now be calculated:

for n=1:size,

for m=1:size,

cos2kx=2*cosk(n)ˆ2-1;

cos2ky=2*cosk(m)ˆ2-1;

ek(n,m)=120.5-595.1/2*(cosk(n)+cosk(m))+ 163.6*(cosk(n)*cosk(m))

-51.9/2*(cos2kx+cos2ky) + 51*(cos2kx*cos2ky)

-111.7/2*(cos2kx*cosk(m)+cosk(n)*cos2ky);

...

Here, we have taken ε(~k) to have the Bi2Sr2CaCu2O8+δ band structure. [8] Finally, we

should also store whatever parts of Σ(~k, ω) are energy-independent. For a superconductor,

Norman et al. [9] have proposed a phenomenological Σ(~k, ω) = −iδ+∆(~k)2/(ω+ε(~k)+ i0+).

In this expression, ∆(~k) = ∆0/2(cos kx − cos ky) is energy independent, and should also be

calculated once and stored for future use:

...

150

Page 159: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Dk(n,m)=45/2*(cosk(n)-cosk(m));

end;

end;

We are now ready to calculate the density of states for a range of energies (Eq. B.12).

Begin by calculating Σ(~k, ω) (as a scalar), and from it, G0(~k, ω) (two-dimensional matrix):

delta=2.0

for en=-60:60,

for n=1:size,

for m=1:size,

sigma=-j*delta+Dk(n,m)ˆ2/(en+j*delta/10+ek(n,m));

G0(n,m)=1/(en-ek(n,m)-sigma);

end;

end;

...

Next, calculate the inverse Fourier transform of G0, and square the result for all points in

real-space before transforming back into FTS:

...

G0=ifft2(G0);

for n=1:size,

for m=1:size,

G0(n,m)=G0(n,m)ˆ2;

end;

end;

G0=fft2(G0);

...

151

Page 160: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Finally, we can calulcate the power spectrum of δn(~q, ω), which we will do in the variable

’G0’ to save RAM:

...

for n=1:size,

for m=1:size,

G0(n,m)=-imag(G0(n,m))ˆ2;

end;

end;

We have, again, assumed the structure factor V to be an all-pass filter, although if we had

not, it should be applied inside the nested loops before calculating ’G0’ in the above code.

Finally, write the resulting elastic scattering interference image δn(~q, ω) to disk and move

on to the next energy:

...

mi=min(G0(:));

ma=max(G0(:));

G0=(G0-mi)/(ma-mi);

imwrite(G0,strcat(’c:\calc\’,int2str(en),’.tif’),’tif’);

end;

The procedure outlined here calculates the scattering interference contribution to the

density of states efficiently enough to be useful in most practical situations. With an Athlon

XP 2000MHz CPU and 1GB of RAM, and using MATLAB 6.1 in Linux, the calculation of

the power spectrum of δn(~q, ω) on a 2048×2048 grid takes about 2 minutes per energy, with

a total overhead of 2 minutes. Speed can be improved by a factor of about 4 by vectorizing

the calculations. The speed can be further improved by about a factor of 2 using a compiled

152

Page 161: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

(0,0) (0,2π)

(2π,2π)

Figure B.2: This is the imaginary part of the Green function described in thetext (ω = −20mV ). It was calculated between kx = −2π to 2π and ky = −2πto 2π. However, we need only calculate the part in the box, and then usesymmetry transformations to copy (instead of calculate) the rest of the Greenfunction.

language such as Fortran. Another small increase in speed, and a hefty decrease in memory

footprint, can be attained by taking advantage of the fact that Green function obeys various

symmetries (e.g. four-fold symmetry) in most cases (Fig. B.2). This important improvement

was not included in the above code because it makes the code significantly harder to read.

In sum, the calculations presented in this thesis were all performed on 4096× 4096 grids for

81 energies (−80mV to 80mV ) in a total time of less than 90 minutes for a single Green

function. The complete Fortran code used to generate the plots in Chapter 3 is included

here (compiled using the Intel Fortran Compiler and the Intel Math Kernel Library with

ifl -G7 -O3 -w95 gf.f mkl c.lib). In particular, notice that we have made use of symmetry

to decrease the memory footprint, and that we have vectorized the calculation to improve

speed.

Contents of File: gf.f

153

Page 162: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

subroutine csymm(orig,mat,n)

c This routine duplicates -2pi to -pi part of k-space, stored in

c orig, to occupy the space -2pi to 2pi in mat.

complex*16 orig(n/4,n/4)

complex*16 mat(n,n)

integer i,j

do i=1, n/4

do j=1,n/4

mat(i,j)=orig(i,j)

enddo

enddo

do i=n/4+1, n/2

do j=1, n/4

mat(i,j)=mat(n/2+1-i,j)

enddo

enddo

do i=1, n/2

do j=n/4+1, n/2

mat(i,j)=mat(i,n/2+1-j)

enddo

enddo

do i=1, n/2

do j=n/2+1,n

mat(i,j)=mat(i,j-n/2)

154

Page 163: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

enddo

enddo

do i=n/2+1, n

do j=1,n

mat(i,j)=mat(i-n/2,j)

enddo

enddo

return

end

subroutine matrix2pgm(mat, x, n)

integer x

double precision mat(x,x)

integer n

integer i,j,k

character(len=20) name

double precision max, min

integer val

write (name, 20) n

20 format(’test’,i3.3,’.pgm’)

open (UNIT=1, file=name)

write (1,’(A2) ’) ”P2”

write (1,’(i6,i6) ’) x, x

155

Page 164: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

write (1,’(i6) ’) 255

max=mat(1,1)

min=mat(1,1)

do i=1,x

do j=1,x

if (max .lT. mat(i,j)) then

max=mat(i,j)

endif

if (min .gT. mat(i,j)) then

min=mat(i,j)

endif

enddo

enddo

do i=1,x

do j=1,x

val=int((mat(i,j)-min)/(max-min)*255.0+0.5)

write (1,*) val

enddo

enddo

close(1)

return

end

program dN

156

Page 165: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

c Constants

integer energy

parameter (energy=121)

integer k pts

parameter (k pts=4096)

double precision mw,pw

parameter(mw = -60.0d0,pw=60.0d0)

double precision PI

c Band Structure

double precision t 0, t 1, t 2, t 3, t 4, t 5

parameter(t 0=120.5d0, t 1=-595.1d0, t 2=163.6d0)

parameter(t 3=-51.9d0, t 4=-111.7d0, t 5=51.0d0)

double precision D 0, delta

parameter(D 0=45.0d0, delta=.6d0)

c variables

double precision outp(k pts/2,k pts/2)

integer n,m,e, nn, mm, status, nn2

double precision ek(k pts/4, k pts/4), Dk(k pts/4,k pts/4)

double precision cosk(k pts)

double complex sigma1, sigma2

complex*16 G0(k pts/4,k pts/4), F0(k pts/4,k pts/4)

complex*16 temmy(k pts,k pts)

double precision cos2kx, cos2ky, w

157

Page 166: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

PI=4.0d0*datan(1.0d0)

nn=k pts

mm=k pts

nn2=k pts/2

c Set up cosines (calculationally expensive, so do once)

do n=1, k pts

cosk(n)=dcos(4.0d0*PI*dble(n-1)/dble(k pts)+2.0d0*PI)

enddo

c Set up stuff that’s energy independent, like band structure and gap

c Note in particular that we only need to calculate this between

c -2pi and -pi, not for the whole region

do n=1,k pts/4

do m=1, k pts/4

cos2kx=2.0d0*cosk(n)*cosk(n)-1.0d0;

cos2ky=2.0d0*cosk(m)*cosk(m)-1.0d0;

ek(n,m)=t 0+t 1/2.0d0*(cosk(n)+cosk(m))

c + t 2*(cosk(n)*cosk(m))

c + t 3/2.0d0*(cos2kx+cos2ky) + t 5*(cos2kx*cos2ky)

c + t 4/2.0d0*(cos2kx*cosk(m)+cosk(n)*cos2ky)

Dk(n,m)=D 0/2.0d0*(cosk(n)-cosk(m))

enddo

enddo

c Main Loop

do e=1, energy

w=(pw-mw)*dble(e-1)/dble(energy-1)+mw

158

Page 167: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

print *,e,w

c calculate G0 on a 1/4 of the BZ

G0=(dcmplx(w-ek,delta)*dcmplx(w+ek,delta/10))-Dk*Dk

F0=dcmplx(Dk,0.0d0)/G0

G0=dcmplx(w+ek,delta/10)/G0

c Now use symmetry transforms to get G0 for the whole BZ

c And then calculate dN

call csymm(G0,temmy,nn)

status=1

call zfft2d(temmy,nn,mm,status)

temmy=temmy*temmy

status=-1

call zfft2d(temmy,nn,mm,status)

do n=1,k pts/2

do m=1,k pts/2

outp(n,m)=dimag(temmy(n+k pts/4,m+k pts/4))

enddo

enddo

c Repeat for F0

call csymm(F0,temmy,nn)

status=1

call zfft2d(temmy,nn,mm,status)

temmy=temmy*temmy

status=-1

call zfft2d(temmy,nn,mm,status)

do n=1,k pts/2

159

Page 168: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

do m=1,k pts/2

outp(n,m)=outp(n,m)+ dimag(temmy(n+k pts/4,m+k pts/4))

enddo

enddo

c write dN to file

outp=outp*outp

call matrix2pgm(outp,nn2,100+e)

enddo

stop

end

B.3 References

[1] S. Misra, M. Vershinin, P. Phillips, and A. Yazdani, Phys. Rev. B in press (2004).

[2] Q.-H. Wang and D.-H. Lee, Phys. Rev. B 67, 20511 (2003).

[3] L. Capriotti, D. Scalapino, and R. Sedgewick, Phys. Rev. B 68, 14508 (2003).

[4] M. Salkola, A. Balatsky, and D. Scalapino, Phys. Rev. Lett. 77, 1841 (1996).

[5] L. Zhu, W. Atkinson, and P. Hirschfeld, Phys. Rev. B 69, 60503 (2004).

[6] S. Misra et al., Phys. Rev. B 66, 100510 (2002).

[7] C. Bena, S. Chakravarty, J. Hu, and C. Nayak, cond-mat/0405468 (unpublished).

[8] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 52, 615 (1995).

160

Page 169: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

[9] M. Norman, M. Randeria, H. Ding, and J. Campuzano, Phys. Rev. B 57, 11093 (1998).

161

Page 170: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

Curriculum Vitae

Degrees

1998 B.S. AMEP & Mathematics, University of Wisconsin- Madison

2004 Ph.D. Physics, University of Illinois at Urbana- Champaign

Honors

Radke Award Winner (Oustanding Physics Undergraduate at Madison) 1998

Hilldale Award Winner (Undergraduate Thesis at Madison) 1998

UIUC University Fellowship (Fellowship for 1st year of Graduate School) 1998

GAANN Fellowship (Fellowship for 2nd year of Graduate School) 1999

Bardeen Award Winner (departmental award for outstanding research by a graduate student

in condensed matter physics)- 2003

PCCM Fellow (Postdoctoral appointment at Princeton University)- 2005

Papers and Invited Talks

1. S. Misra, et al., PRB 58, 8905 (1998).

2. D.J. Hornbaker, et al., Science 295, 828 (2002).

3. S. Misra, et al., PRL 89, 87002 (2002).

4. S. Misra, et al., PRB 66, 100510 (2002).

5. S. Misra, March Meeting of the APS (2003).

162

Page 171: c Copyright by Shashank Misra, 2004ev1l.freeshell.org/SMThesis.pdf · SHASHANK MISRA B.S., University of Wisconsin - Madison, 1998 DISSERTATION Submitted in partial ful llment of

6. M. Vershinin, S. Misra, et al., Science 303, 1995 (2004).

7. S. Misra, et al., Phys. Rev. B, in press, cond-mat/0405204 (2004).

8. S. Misra, March Meeting of the APS (2005).

163


Recommended