American Mineralogist, Volume 82, pages l10l-1 110, 1997

Transformation of titanomagnetite to titanomaghemite: A slow, two-step,oxidation-ordering process in MORB

WBrxrN Xurt Doxar,n R. PnAcoRrr WnyNB A. Dolr,lsEr2 RoB VIN DBn Voor'lNn Rrcr Bnaunounn3,t

rDepartment of Geological Sciences, University of Michigan, Ann Arbor, Michigan 48109-1063, U.S.A.rDepartment of Earth and Space Sciences, University of California Los Angeles, Los Angeles, California 90095-1567, U S.A.

rDepartment of Geosciences, University of Houston, Houston, Texas 77204-5503, U.S A.

Assrnncr

Magnetic iron oxides in a sequence of pillow basalts that were dredged from the AtlanticOcean floor have been studied to characteize titanomaghemite and to define the processesof maghemitization. Distances from the spreading ridge and ages (in parentheses) of thesamples are 0-10 (0-1), 160 (9), 450 (26), and 900 km (70 Ma).

Iron titanium oxides occur as I to 10 pm-sized dendritic and cruciform-shaped crystalswith identical appearances in all samples and with no signs of change or significant het-erogeneity in composition or structure as observed by TEM and AEM. Parameters changeprogressively from the youngest to-the oldest, e.g., Curie temperature : 180 to 360 'C;

lattice parameter : 8.466 to 8.361 A; number of octahedral cations per cell from Rietveldrefinement : 14.8 to l2.l; mean hyperfine (internal) fields at 300 K from Mdssbauer data: 37 to 45 T. The large Ti contents (Uvuo to Uvro) are nearly constant. SAED patternsshow superstructure reflections only for the oldest sample.

The youngest sample has parameters corresponding to nearly unoxidized titanomagne-tite, whereas the oldest is near-end-member titanomaghemite. Intermediate samples arepaftially altered but display no superstructure reflections, implying a lack of significantordering of vacancies. The data therefore show that the process of (titano)maghemitizationhas two distinctly different components: (l) oxidation and loss of Fe, with creation ofdisordered vacancies, and (2) ordering of vacancies. The data collectively imply a processo rminated by solid state diffusion of Fe from the crystals, oxidation of Fe, and creationo' vacancies wherein the O closest-packed framework is preserved, in sharp contrast to amodel of addition of O or to dissolution and neocrystallization.

INrnonucrroN

Magnetic minerals in ocean-floor basalts have beenmuch studied because magnetic anomalies of oceaniccrust have served as the primary data in our understand-ing of the evolution of oceanic lithosphere and of platetectonics. Titanomagnetite [FerTiOo (ulvcispinel).Fe.Oo(magnetite), or Fe,. *,Ti.Ool is the principal magnetic min-eral in unaltered ocean floor basalts, but it undergoes al-teration to (titano)maghemite (kving 1970, Ozima et al.1974; Petersen et al. 1979; Smith 1987; Pariso and John-son 1991). The transformation involves oxidation of fer-rous iron and production of (ordered) cation vacancres.O'Reilly and Banerjee (1967) defined the degree ofmaghemitization by the oxidation parameter Z in theexpression

* Present address: Department of Geology, Arizona State Uni-versity, Tempe, AZ 8528'7-1404, U.S.A.

f Present address: Exxon Production Research Co.. P.O. Box2189, Houston, TX 77252-2189. U.S.A.

Fe2* + Zro - ZFe3n I ( l - z)Fe2* +t ;o '

where Z is the fraction of ferrous iron that has been ox-idized to ferric iron, ranging from 0 to 1. The parameterZ canbe estimated using the measured Curie temperatureor cell parameter, provided the effects of solid solutionof components such as Ti are known (Readman andO'Reilly 1912: Ozima et al. 1974; Nishitani and Kono1983: Moskowilz 1987).

Maghemitization of primary titanomagnetite has beeninterpreted to be the main cause of decrease in NRMintensity with increasing age of MORB, concomitant withincreasing distance from spreading ridges (Irving 1970;Marshall and Cox 1972; Blell and Petersen 1983). Anunderstanding of such relations requires characterizationof the crystal chemisffy of titanomaghemite, but becauseof the difficulties in obtaining pure natural material, stud-ies since the 1960s have been largely restricted to syn-thetic titanomagnetite and titanomaghemite (e.g.,O'Reilly and Banerjee 1967; Ozima and Sakamoto l97l;

0003-004x/97l1 1 12-1 r01$05.00 1 1 0 1

rlo2 XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

Readman and O'Reilly 1972; Nishitani and Kono 1982,1983; Moskowitz 1987 Goss 1988). Natural materialshave been less extensively studied (e.g., Marshall andCox 19'72: Petersen et al. 1979; Beske-Diehl 1990; Furuta1993), in part because of their small grain size, which isgenerally less than approximately 5 pm. Consequently,the crystal-chemical properties of (titano)maghemite andintermediate products of maghemitization remain largelyuncharacterized. It is especially important to determinethe characteristics of intermediate products because theyare the key to an understanding of the process ofmaghemitization.

Although it is well known that the magnetic propertiesof oxidized titanomagnetite are directly related to themaghemitization process (Bleil and Petersen 1983;O'Reilly 1983; Banerjee 1991), the alteration productsand processes are incompletely defined. The transforma-tion is generally inferred to occur by one of two verydifferent processes: (1) loss of Fe relative to Ti, with theO closest-packed substructure being invariant (e.g., Pe-tersen et al. 1919), or (2) addition of O with constant Fe:Ti ratio (e.g., Readman and O'Reilly 1972). The generalconcensus favors transformation by diffusion of Fe outof crystals (e.g., Furuta 1993, and references therein).Both mechanisms appear to involve solid state diffusionat relatively low temperatures and in the presence of afluid in naturally occurring samples, as implied, for ex-ample, by retention of grain shapes (Petersen and Vali1987) and of Ti contents (Furuta 1993). Tiansformationin laboratory experiments usually has been promoted byheating at high temperatures in air to promote oxidation(e.g., Readman and O'Reilly 1972; Moskowitz 1987; Fu-kasawa et al. 1993), and infrequently by heating in fluid(Ryall and Hall 1979; Kelso er al. 1991). Although heat-ing in air presumably involves diffusion of Fe, it involvesconditions that are unlike those in nature, whereas heatingin HrO may involve dissolution and neocrystallizationrather than diffusion of Fe.

The reactant phase in marine basalts is dominantly ti-tanomagnetite with an ulvcispinel component of approxi-mately 0.62 (Bleil and Petersen 1977; Moskowitz and Ba-nerjee 1981; Pariso and Johnson 1991). Titanomaghemitewith ordered vacancies was found to retain primary Ticontents (Furuta 1993). However, little is known aboutthe relation between composition and cation oxidation insamples with intermediate values of ordering, becausesuch materials have not been well characterized. The abil-ity of the oxidized products to retain components such asTi is a key to the nature of the process, however, becauseTi should be retained during solid state diffusion and lostduring water-mediated dissolution and neocrystallization.

The specific vacancy distribution of natural titanom-aghemite is not known. The detailed structure of mag-hemite seems process dependent (Waychunas 1991, andreferences therein), because at least three symmetrieshave been proposed for maghemite, including one cor-responding to a face-centered cubic structure (Goss1988), one with a teffagonal superlattice having c : 3A

600

400

0' 0 2 0 4 0 6 0 8 0

Age (Ma)

Frcuno 1. Plot of saturation magnetization (M,) vs. age ofbulk rock samples, showing decrease of M" with increasing age.

(Boudeulle et al. 1983; Greaves 1983), and one with aprimitive cubic superlattice (Smith 1979; Goss 1988;Banfield et aI. 1994). A primitive cubic superlattice hasbeen reported for naturally occurring titanomaghemite(Collyer et al. 1988; Smith 1979; Price and Putnis 1979),but the oxidized equivalent of titanomagnetite occurringin MORB remains unstudied.

To investigate the relationship between MORB mag-netic properties and mineralogic properties, we studied aseries of four pillow basalts from the Atlantic Ocean col-lected at increasing distances, and therefore increasingages (<1 Ma to 70 Ma), from the mid-Atlantic Ridge(Beaubouef 1993; Xu et al. 1994). These samples are ofinterest because they display a systematic decrease inNRM with increasing age that was tentatively hypothe-sized to be related to the maghemitization process (Beau-bouef 1993). Figure 1, plot of saturation magnetizationvs. age, demonstrates decreasing magnetization with age.Xu et al. (1995) showed that the rock magnetic propertiesare not related entirely to titanomagnetite-titanomag-hemite with grain sizes on the order of a micrometer thatare easily observed and the source of past observations.Instead the magnetic properties are primarily related tochanges in submicrometer-sized titanomagnetite found ininterstitial glass. In this report, we characterize the com-mon kind of titanomagnetite-titanomaghemite that occursabundantly in MORB with grain sizes in the opticalrange. Because such material was incompletely charac-terized, in part because of its small grain size, this studyemphasizes TEM and AEM observations, but also in-cludes data from SEM, EMPA, XRD, Mdssbauer analy-sis, and magnetic studies. These data are then used toshow that the transformation of titanomagnetite to titan-omaghemite in oceanic basalt involves two processes thatmay occur at different times, oxidation-cation-diffusionand ordering of resultant cation vacancies.

S,q.Iupr,o DESCRrprroN AND EXPERTMENTALPROCEDURES

The samples were dredged from four sites at distancesof <10, 160, 450, and 900 km from the mid-Atlantic

800

cT)

N

E

E

{

n

trEI

ridge and are referred to with conesponding site numbersone through four. All hand samples are pillow basalts,which grade from black in the youngest samples, throughgrey or dark brown, to brown or reddish brown in theoldest sample, as consistent with increasing degree of ox-idation (see below). To ensure that the full range of tex-tures was observed, thin sections were prepared from pil-low rims as well as pil low interiors.

Thin sections were made with "sticky wax" as adhe-sive so that sections could first be studied by optical andSEM techniques, with selected areas subsequently re-moved and ion-milled for imaging by TEM. SEM obser-vations were made with a Hitachi 5-570 instrument fittedwith back-scattered electron detector and Kevex QuantumEDS system. TEM and AEM data were obtained using aPhilips CM-12 STEM fitted with a Kevex Quantum EDSsystem, with procedures as defined by Jiang et al. (1990).

To obtain concentrates of the magnetic phases for mea-surement of magnetic properties, Mossbauer spectra, andXRD patterns, one sample from each site was crushed inair to obtain grains <1 mm in size. These were thenground under acetone in a tungsten alloy mortar, and themagnetic fraction was separated using a REE magnet.The separates were subsequently found (see below) tocontain approximately 18 to 35Vo magnetic phases. Fur-ther separation was not possible owing to the fine-grainednature of the intergrowths. Curie temperatures (T.) andMcissbauer spectra were measured at the Institute forRock Magnetism, University of Minnesota. The T. valueswere obtained using an automated, vibrating sample mag-netometer, by heating and cooling samples sealed in silverfoil at a rate of 20 "C/min, in an applied field of 500 mT,both in helium and in air. These sTFe Mcissbauer data wereobtained as 1024-channel specffa, folded, and summedpairwise to yield 256-point spectra. The spectra were fit-ted with the program MOSFIT a program developed atUCLA. Spectra were fitted with a magnetically split, six-line component, a paramagnetic Fe3* doublet, and a par-amagnetic Fe'* doublet. Because of the variability or oth-er complexities of the magnetic phases (see below) in-dividual lines of the magnetic sextet were approximatedby one to three subcomponents, but no significance wasinferred for the individual subcomponents.

X-ray powder diffraction patterns were obtained usinga Philips XRG 3100 diffractometer fitted with a graphitemonochromator mounted near the receiving slit, CuKaradiation (35 kV 15 mA), and variable divergence slit.Scans were made from 15" to 75" 20 with a step intervalof 0.01' and count time of 6 s. Rietveld analysis wascarried out with the program DBW (Wiles and Young1981), with refinement of the usual pattern parameters(background, 28 zero-point, peak width, and Gaussian-Lorentzian mix). No regions of the diffraction patternswere excluded. The separates contained significantamounts of pyroxene and plagioclase. Positional param-eters of plagioclase (Toman and Frueh 1973), clinopyrox-ene (Peacor 1967), and orthopyroxene (Ghosc 1965) wereheld constant, but cell parameters and concentrations

1 103

Frcunn 2. SEM backscattered electron image showing thetypical texture of pillow basalt sample one with titanomagnetitecommonly concentrated in areas between dendritic pyroxene ag-gregations. Platy dark grey area (P1) : plagioclase lath; greyarea (Cpx) - dendritic pyroxene aggregations; bright spots (Mt): titanomagnetite; irregular dark grey area : interstitial glass

were refined. The magnetite-maghemite substructure pa-rameters that were refined included subcell parameter (A),oxygen coordinate (u), and occupancy of the octahedralsites. The tetrahedral site was assumed to be occupiedonly by Fe, and all Ti was assumed to be in the octahedralsite with the Fe/Ti ratio being held constant at a valueseparately determined for each sample by AEM. Theseassumptions regarding the site occupancies have previ-ously been verified by several refinements of titanomag-netite or titanomaghemite structures (e.g., Collyer et al.1988).

Dar.q A.Nn oBSERVATIoNS

SEM observations of thin sections

The texture shown in Figure 2 is representative of allfour samples, with the exception of the thin, glass-richpillow rinds. All samples consist primarily of microphen-ocrysts of plagioclase and olivine within a fine-grainedmatrix of plagioclase, pyroxene, olivine, titanomagnetite,and residual glass. The dominant opaque mineral is eu-hedral titanomagnetite or titanomaghemite, occurringcommonly with dendritic (Fig. 3a) and crucifotm (Fig.3b) shapes consistent with quenching from high temper-atures (Somboonsuk and Tiivedi 1985). Grain sizes thatcan be resolved by BSE imaging are dominantly in therange 1 to l0 pm. Qualitative EDS analyses are consis-tent with relatively high Ti concenffations that are ap-proximately equal in all samples. There is limited varia-tion in composition within a thin section, individualcrystals having compositions varying by up to Uv. (asshown by AEM data; see below) from crystal cores torims, but crystals are otherwise homogeneous.

STEM observations

Figure 4a is a low magnification, bright field image oftypical magnetic material from sample four. The magnetic

XU ET AL: TRANSFORMATION OF TITANOMAGNETITE

I 104 XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

Frcunn 3. (A) SEM secondary electron image showing dendritic titanomagnetite in sample four. (B) SEM backscattered electronimage showing crucifotm titanomagnetite in sample one. Pl : plagioclase; Cpx : clinopyroxene; Mt - titanomagnetite.

phase has shapes identical to those in the other three sam-ples and appears to be uniform in texture. There are nocracks that might arise because of decreases in latticeparameters as a result of maghemitization e.g., as ob-served by Petersen and Vali (1987). Figure 4b is a cor-responding [010] zone axis selected area electron diffrac-tion (SAED) pattern. It has strong substructure reflections(e.g.,202 and 400) characterisric of a phase with the spi-nel sffucture but also has weak superstructure reflections(e.g., 100) of a kind characteristic of maghemite. SAEDpatterns obtained from within a single micrometer-sizedcrystal were uniform in orientation and relatively sharp,implying that each grain is a well-formed, single crystal.

Superstructure reflections were observed in every crys-tal studied from sample four. They always were observedto have the same approximate intensities relative to thesubstructure reflections. By conffast, no superstructure re-flections were observed in magnetic phases for samplesone to three, even though SAED patterns were obtainedon tens of crystals from several ion-milled samples fromeach locality.

The space group of a natural sample of titanomag-hemite was determined by Collyer et al. (1988) to beP4,32 or P4,32, the cell having a -- 8.341 A, i.e., a cellwith the same dimensions as that of magnetite, but with

reflections in positions that violate F centering. Smith(1979) first suggested that maghemite satisfies those re-lations, and Banfield et al. (1994) made equivalent ob-servations. However, Banfield et al. also observed thatsome of their material gave extra reflections that requireddoubling of the magnetite cell parameter. Boudeulle et al.(1983) and Greaves (1983) reported that synthetic mag-hemite can be tetragonal with a superlattice with c :3A.The SAED patterns obtained in this study show super-structure reflections that only violate F centering andtherefore are consistent with a cell with a : A. However,reflections of the type ft00 with h odd are prominent andtherefore inconsistent with space grotp P4,32 or its en-antiomorph. The [100] SAED patterns of different crys-tals displayed fourfold symmetry in all cases, with re-flections of equivalent indices having equivalentintensities as visually estimated. Such relations are con-sistent with cubic symmetry or of unit-cell twinning of asymmetry with a noncubic space group. Our data do notpermit a direct identification of space group. However,imaging under various conditions, including attempts toimage possible antiphase domains by dark-field imagingwith supersffucture reflections, failed to detect furthertextural relations.

Some representative AEM analytical data are listed in

XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE I 105

FrcunB 4. (A) TEM bright field image of titanomaghemite (Mgh) from sample four, showing that it has the original cruciformtexture. (B) Electron diffraction pattem of the titanomaghemite of the same sample as in (A), showing that it has the diffractionpattern of a primitive cubic superlattice (100), which is distinct from that of the face-centered cubic lattice.

Table 1. Although there are differences in compositionbetween materials from different localities, the range incomposition was found to be narrow for any one sample,with only relatively small concentration gradients fromcore to rim; e.g., Figure 5 shows a gradient correspondingto an increase in the TilFe ratio from 0.23 to 0.27 fromcore to rim, a relative change typical of all samples. Thecalculated average ulvcispinel components for each sam-

ple vary from x : 0.58 to x : 0.73, but there is no trendin composition with distance from the spreading ridge.Indeed, sample four has the highest ulvcispinel compo-nent, implying that there has been no significant loss ofTi. In addition, the small core-to-rim concentration gra-dients present in younger samples also occur in the oldestsamples.

A second mode of titanomasnetite-titanomashemite

TaeLe 1. Selected typical AEM analyses of titanomagnetite in studied samples of pillow basalts from mid-Atlantic Ocean*

Sample 1 Sample 2 Sample 3 Sample 4Oxide(vvt%)

Al,o3sio,CaOTio,MnOFerO"

Totalt

0 0 0 1 0 41 1 0 0 0 00.60 0 70

20.71 20.1 6o.73 2 02

76 38 76 08100 00 100 00

0 000 0.0420 039 0 0000021 0 0250.518 0 5070.021 0 0571 9 1 0 1 9 1 62 509 2545

0271 0 2650 640 0.666

1.90 1 630 6 8 0 7 90 5 3 0 6 9

2217 22941 4 6 1 1 5

7326 7276100 00 100.00

0 9 6 0 5 40.00 0 000.00 0 00

19 58 19.271.23 1 44

76 95 77 85100 00 100.00

0 063 0 0210 000 0 0000 000 0 000o 487 0 4810 034 0 0401 915 'l 9472537 2 535

0254 02470.608 0 594

2.86 1 960 00 0,950.93 0.51

2287 24740 96 072

7236 71 11100 00 100 00

0 1 1 0 0 0 7 50 000 0 0300 032 0 0180 560 0 6000 026 0 0201 771 1 7262 500 2 469

0 316 03470 681 0743

AISiCaTiMnFe

Total

Ti/Fe4

Number of cations calculated on the basis of four O atoms0 073 0 0630.022 0 0260 019 0 024o 544 0 5630 040 0.0321 799 1.7832 498 2.489

0 313 0 .316o 715 0.720

. Two columns for each sample represent two different analysest Oxide wt% normalized to 100%+ x represents the ulvospinel componenl in Fe, ,Fe,Tipo and is calculated from TilFe ratios x : 3Ti/(Ti + Fe).

A

I 106 XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

o.27

q) 0.26

': 0.25F

0.24

0.23

o.22

Frcunn 5. (A) Compositional zoning within dendritic titan-omagnetite, with numbers corresponding to AEM analyses forwhich areas are shown in Figure 58. (B) SEM secondary elec-tron image of sample one, showing a cross-section of one den-dritic titanomagnetite and areas of AEM analysis.

occulrence is as single domain-sized (approximately 200A) crystals within residual glass (Xu er al. 1995). Al-though important as a carier of the NRM, it accounts for<. l0%o of the magnetic material, and therefore it does

TeeLe 2. Rietveld-refined spinel structure parameters

not contribute significantly to the properties described inthis study.

X-ray diffraction data

No superstruchrre reflections were observed by powderXRD from any sample, including sample four for whichsuperstructure reflections were observed in SAED parterns (Fig. 4b). Powder X-ray diffraction patterns of syn-thetic maghemite (e.g., ̂ y-Fe,O. of Pfizer Chemical Co.),with completely ordered octahedrally coordinated cationsand vacancies, clearly show weak superstructure reflec-tions. The intensity of such reflections is maximized forboth maximum oxidation and cation-vacancy ordering.Because samples two through four are largely to com-pletely oxidized, these relationships imply that orderingmust be incomplete.

Individual Ru,"r, factors for the Rietveld refinement aregood especially for the magnetic phases, ranging from2.4to 4.3Vo. Structure parameters of the magnetic phases arelisted in Table 2, along with values obtained by othersfor refinements of well-characterized magnetic phaseswith the spinel sffucture. Extensive data on the cell sizeof the ulvcispinel-magnetite (0 < x < l; Z: 0) and mag-netite-maghemite (x : 0; 0 < Z < l) solid solutions (e.9.,Lindsley 1976; Fukasawa et al. 1993; Senderov et al.1993) show that A increases with increasing Ti contentand decreases with increasing proportion of vacancies.The variation of lattice parameter away from those joins(r : 0 and Z : 0) is more complex and is discussed indetail by Xu et al. (1996). Here we note that from sampleone to sample four the lattice parameters decrease froma : 8.466 (F-centered cell) to 8.361 A (cell with reflec-tions violating F centering). When corrected for the Ticontents measured by AEM and assuming linear relationsbetween the lattice parameters, Ti contents, and vacancycontents, those values correspond closely to values forend-member unoxidized titanomagnetite (8.475 A) andcompletely oxidized titanomaghemite (8.340 Al.

Proportions of components other than Fe and Ti wereshown by AEM to be small (<l wtVo). Using relationsfor lattice parameter variation as a function of concentra-tion of cations such as Mn and Al (Fukasawa el al. 1993),the small cation proportions were found, collectively, to

Sample A (A) References

No. 1N o 2N o 3N o 4Ti.Mh1-Fe,O"1-Fe.O.

Fe"OoFe,TiO4

8.466(2)8.402(1)8 381 (3)8.361 (1)8 341 (1 )8 3s15(5)8.341I 33968.3970(1 )I 536(1 )

o 2625(2110.2583(20)0 2s34(55)0 2545(36)0.2554"0 2579-

0 2551 "0 25500 26s(10)

14 82(29)13.97(22)1 4 06(s3)12 .10 (31 )1 3 .1

(13 .3 )(13 .3 )(13 3 )(16 )(16 )

This paperThis paperThis paperThis paperCollyer et al (1988)Dollase (unpublished)Fukasawa et al (1993)Greaves (1 983)O'Nei l l & Dol lase (1994)Lindsley (1976)

Note: oclJcell: number of Fe-occupied octahedra per unit cell.'Refers to equivalent oxygen u-coordinate in space grcup Fd3m

XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

TeeLe 3. Estimated spinel composition based on Rietveld refinement of XRD data

I 107

Comoositions estimated from cell dimensions Compositions estimated from site occupancies

Sample Fe2* Fe3t Ti Vacancy A (A) Fe2' Vacancy O &,.nn (%)

N o 1N o 2N o 3N o 4

1 3970 658o 478o 211

0 9031 .351o.6421 716

0.6240 6580 5300 608

8.4664(1 1 )8.4028(1 4)8 3828(35)8 3607(1 3)

1 1670 7890 6840 201

1 0771 2511 4871 723

0.1 48o 2920.2860.469

0.076 40.333 40.350 40.465 4

0 6080 6680.5430 607

42392.434.313 0 0

have a negligible effect (<0.003 A) on lattice parameters.Therefore, assuming that the lattice parameters are relatedlinearly to the proportion of vacancies and Ti contents,compositions can be predicted as listed for each samplein Table 3. The compositions obtained from refined oc-cupancies, calculated assuming full occupancy of the tet-rahedral sites, are also given.

The proportions of vacancies determined by Rietveldrefinement (Table 3) increase in a regular way from sam-ple one to four. Likewise, the compositions calculated in-dependently and on the basis of relations between thecomposition and refined vacancies and of compositionand lattice parameters, are in good agreement.

Curie temperatures (T")

Table 4 shows that the measured Curie temperaturesfor the powdered, magnetic-concentrate samples increasefrom 180 to 350'C, with increasing distance of samplesfrom the spreading ridge. The Curie temperature of end-member magnetite (573 'C) has been shown to decreasewith increasing ulvcispinel content and increase with in-creasing oxidation parameter, Z (Readman and O'Reilly1972;Nishitani and Kono 1983;Moskowitz 1987). Usingthe AEM data of this study and the relationship betweenTi content and f of Readman and O'Reilly (1972) andNishitani and Kono (1983), Curie temperatures between0 atd220 "C are predicted for samples one through four,assuming no oxidation. Those values are significantlysmaller than the measured values, except for sample one.Other data imply little oxidation for sample one, but thedifferences in samples two through four can be attributedto maghemitization. Using the relationships developed byReadman and O'Reilly (1972), oxidation parameters wereestimated to vary from 0.1 to 0.9 for samples one throughfour, in good agreement with other estimates.

Teele 4. Measured and estimated 7.

Revers- MeasuredMeasuredx ib i l i ty . 4fC)

Estimated i" ("C)t

Sample N-K

Miissbauer data

The Mcissbauer spectra shown in Figure 6 are com-pared with spectra of pure maghemite and pure magnetite(Pankhurst and Pollard 1993). Parameters obtained fromthe spectra are listed in Table 5. All spectra are charac-teized by relatively broad peaks, but peaks become pro-gressively sharper from sample one through sample four.Broadening is, in general, caused by chemical heteroge-neity and small grain sizes. The titanomagnetite that oc-curs with grain sizes on the order of hundreds of ang-sffoms in interstitial glass is lost through alteration asdistance from the spreading ridge increases (Xu et al.1995). The increase in sharpness of peaks with increasingdistance therefore may be caused, at least in part, by lossof that fine-grained material.

The values of the isomer shift (Table 5) range from0.31 to 0.35 mm/s. They are close to the value of 0.3mm/s expected for octahedral Fe3* (Collyer et al. 1988)but are very different from the value of 0.67 mm/s forpure magnetite with 507o Fe'* in octahedral sites. Be-cause the peaks of spectra from samples close to the ridgeare extremely broad, the measured isomer shifts are in-accurate. Furthermore, in samples one and two, the iso-mer shifts listed in Table 5 refer only to the most prom-inent and readily identifiable component of the broadspectra. Because the specffa sharpen with degree of oxi-dation, focusing on that most prominent component em-phasizes only the most oxidized component. Thus sampleone, especially, may have a large Fe2* component. Isomershifts of the more sharpened spectra of samples three andfour, at least, imply that most of the Fe is present as Fe3*,however. The mean hyperfine (internal) fields at 300 Kincrease from sample one (37.0 T) to sample four (45.2T), consistent with increasing proportion of Fe3t.

DrscussroN

Characterization of magnetic minerals

All data from sample one through sample four are con-sistent with progressive maghemitization with increasingdistance from the spreading ridge, i.e., with increasingage. Sample one is unaltered or nearly unaltered titan-omagnetite as indicated by the following relations: (1)The lattice paramete! a, corresponds to a total of 2.92(octahedral plus tetrahedral) cations per four O atomswhen adjusted for the measured Ti content, in comparisonwith the ideal value of 3 for unoxidized titanomagnetite.(2) The composition calculated on the basis of refined

N o 1N o 2No. 3No. 4

0.64o.740 6 00 7 2

yes 180no 260no 300no 350

80-150 95-18020-80 0-95

110-200 120-22040-100 20-100

- Reversibility of the thermal curve measured in helium.t 4 estimated based on measured x values and Readman and

O'Reilly's (R-O) (1972) and Nishitani and Kono's (N-K) (1983) contourdiagrams, assuming Z: 0.

1 108 XU ET AL: TRANSFORMATION OF TITANOMAGNETITE

Flcunn 6. Mcissbauer spectra of: (A) magnetic separates from basalt sample one; (B) magnetic separates from basalt sampletwo; (C) magnetic separates from basalt sample three; (D) magnetic separates from basalt sample four; (E) pure maghemite; and(F) pure magnetite. Spectra A, B, C, D, and E were measured at room temperature, whereas spectrum F was measured at 200 K.Spectra E and F are used by permission of Plenum Publishing Corp., from Pankhurst and Pollard (1993).

11 .999

11 .82 t

- c u

Velocity (mm/s)

site occupancies has a total of 2.85 cations per four Oatoms, consistent with a small amount of oxidation. (3)The O coordinate, u : 0.2625, is compatible with a non-oxidized phase with significant Ti but is considerablylarger than that (0.255) predicted for ritanomaghemite(with the speciflc Ti content measured by AEM). (4) TheCurie temperature (180 'C) compares well with the pre-dicted range of values (80-180 "C) for unoxidized titan-omagnetite with the composition measured by AEM. (5)The thermal curve in He (magnetization vs. temperature)is reversible (Ozima and Ozima 1971; O'Reilly 1983).

By contrast with sample one, sample four is dominated

Tlele 5. Mean hyperfine (internal) fields at 300 K

Samp le 4 (T ) l .S - "kFet Misfit (%)+

-37 -0.35 78 0 26-44 :0.35 84 0 31

443 0.34 54 0 5645.2 0 31 76 0 25

* LS. : lsomer (chemical) shift in mm/s relative to Fe metalt %Fe : Percent of total Fe in magnetic phase(s)f Misfit : Percent of spectrum remaining un{itted

0

Velocity (mm/s)

by titanomaghemite that is nearly fully oxidized, asshown by: (1) The lattice paramete\ A, has a value con-sistent with 0.21 Fe2' and a total of 2.54 (tetrahedral andoctahedral) cations per four O atoms. (2)The compositiondetermined from the site occupancy data has 0.20 Fe,*.There is a total of 2.53 cations per four O atoms. (3) TheO coordinate , u : 0.2545, is compatible with nearly com-plete oxidation. (4) The Mdssbauer pattern is compatiblewith complete or nearly complete oxidation of Fe. (5) TheCurie temperature (350 "C) compares well with that pre-dicted (350-450 'C) for oxidation parameter Z : 0.9(Readman and O'Reilley 1972; Nishitani and Kono1983). (6) The thermal curves in He are irreversible (Ozi-ma and Ozima l97l; O'Reilly 1983). (7) The SAED parterns of all material that was observed showed superstruc-ture reflections consistent with ordered vacancies.However, superstructure reflections were not observed inXRD patterns, whereas many authors have shown thatXRD patterns of end-member, fully ordered maghemiteshow well-defined superstructure reflections. The inten-sities of such reflections are a function of both the pro-portions of vacancies and the degree of vacancy-cation

N o 1N o 2N o 3N o 4

order. The relative weakness of such reflections maytherefore be due to incomplete ordering, given the near-end-member state of oxidation and therefore the high pro-portion of vacancies, even taking into consideration thesmaller proportion of vacancies relative to Ti-freemaghemite.

Intermediate samples two and three have lattice param-eters, O coordinates, cation site occupancies, and param-eters of Mcissbauer patterns that are intermediate to thoseof samples one and four. For example, compositions de-termined from lattice parameters or octahedral site oc-cupancies have octahedral plus tetrahedral cation totals inthe range 2.65 to 2.7I, in comparison with the ideal valueof 3 for unoxidized material. Those parameters thereforedemonsffate that samples two and three have intermediatestates of maghemitization. However, the spinel phasesgive SAED patterns that, despite overexposure of pho-tographs, fail to show any superstructure reflections, im-plying that the vacancies are not ordered or at most areonly partially ordered.

Transformation mechanism

It is generally believed that the transformation of titan-omagnetite to titanomaghemite occurs by oxidation of Fe,with solid-state diffusion of Fe to crystal boundaries andthence to pore fluids rather than by addition of O. Perhapsthe most compelling evidence for a process involving lossof Fe is the electron microprobe data of Furuta (1993)who showed that the average Ti/O ratio of titanomag-hemite is unchanged relative to titanomagnemite, whereasthe TilFe ratio is diminished. Data of this study are con-sistent with such a process, including retention of crystalmorphologies and the three-dimensional translational pe-riodicity of the O atom array of the primary, igneous,single crystals, as indicated by SAED patterns. The dif-ferences in TilFe ratios vary randomly over the range ofsamples studied as a function of oxidation state, appar-ently in part as a function of primary igneous composi-tions. They are therefore not definitive in supporting aprocess of loss of Fe; nevertheless, the sample with thelargest Ti/Fe ratio and one that is significantly larger thanthat of unaltered MORB (0.25; Furuta 1993) is the mostoxidized (sample four). The retention of subtle zoningfrom cores to rims in oxidized crystals implies minimaldisruption in structure and composition, however. A pro-cess of diffusion and loss of small Fe cations is one thatmaintains the integrity of the O closest-packed array, pre-sumably by a process of cation-vacancy hopping. Al-though the effect of O diffusion into the sffucture cannotbe directly assessed, it must involve at least local disrup-tion in the anion array, and the occurrence of such a pro-cess is unlikely given the preservation of subtle featuressuch as core-to-rim zoning. We therefore conclude thatthe data of this and other studies are conclusive with re-spect to a process of Fe diffusion from crystals concom-itant with oxidation rather than with addition of O. Inaddition, the data are entirely inconsistent with a processof dissolution and neocrystallization of the micrometer-

I 109

sized oxides; such a process generally leads to more sta-ble materials, but Ti-rich titanomagnetite and vacancy-rich spinels such as maghemite are inherently metastable,implying the same for titanomaghemite.

Regardless of whether or not the process involves dif-fusion of Fe out of crystals or of O in, the data of thisstudy show that it is one in which oxidation of Fe withcreation of vacancies through loss of Fe is only a firststep, before long-range ordering of vacancies. The ab-sence of superstructure refections in the partially oxi-dized, intermediate samples of this study demonstratesthat ordering occurs after oxidation and hence is at leastin pan a separate process. Likewise, the nearly completestate of oxidation of sample four, in conjunction with su-perstructure reflections that are weak in comparison withthose of well-ordered structures, also implies that oxida-tion and creation of vacancies occurs before ordering. Wetherefore propose that the transformation of titanomag-netite to titanomaghemite may involve two processes,with oxidation and diffusion occurring first, at least inpart independent of vacancy ordering. Ordering of vacan-cies and cations on octahedral sites occurs subsequently.The slow rates of solid state diffusion favor such a pro-gressive sequence, in contrast to the more step-like func-tions that reflect dissolution and neocrystallization.

As noted above, maghemite and processes of mag-hemitization have been studied extensively in laboratorysyntheses. Most experiments involve heating and oxida-tion in air, a process that presumably is based on theaddition of O. Some experiments involve heating in fluidsat high temperatures, conditions that favor dissolution andcrystallization. Although the end-member phases may bereadily synthesized by such methods, the data of this pa-per imply that the processes by which phases form insuch laboratory experiments and in nature may be verydifferent. Not only the composition, but the state of orderand degree of homogeneity of synthetic materials mustbe identical to natural phases if the properties of syntheticmaterials are to be used as analogues of natural materials.Caution should therefore be used when synthetic phasesare treated as analogous to naturally occurring magneticoxides.

Acxnowr,nncMENTS

We thank B. Moskowitz and B R Frost who provided helpful commentson a previous draft of this paper and P Solheid who collected Mdssbauerdata The manuscript was also substantially improved by the suggestionsof two anonymous reviewers Research was supported by National ScienceFoundation (grant no EAR 93-15913) and a visiting scholarship from theInstitute for Rock Magnetism (IRM). The IRM is funded by grants fromthe Keck and National Science Foundation and the University ofMinnesota

RnnnnnNcps crrEDBanerjee, S.K. (1991) Magnetic properties of Fe-Ti oxides In Mineral-

ogical Society of America Reviews in Mineralogy, 25, 1Ol 128Banfield, J.E, Wasilewski, PJ., and Veblen, D R (1994) TEM study of

relationships between the microstructures and magnetic properties ofstrongly magnetized magnetite and maghemite American Mineralogist,79.654 661

XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

1 l r 0 XU ET AL.: TRANSFORMATION OF TITANOMAGNETITE

Beaubouef, RT Jr. (1993) Three case studies in the application ofpaleo-magnetic and rock magnetic techniques to geologic problems: 1. 15'20' fracture zone, Atlantic Ocean, 2 Tauride Mountains, southern Tur-key, and 3 Port au Port Peninsula, western Newfoundland,28T p Ph Dthesis, University of Houston, Houston, Texas

Beske-Diehl, S J (1990) Magnetization during low-temperature oxidationof seafloor basalts: no large scale chemical remagnetization Journal ofGeophysical Research, 95, 21,413-21,432

Bleil, U and Petersen, N. (1977) Magnetic properties of basement rocks,Leg 37, Site 332 In Initial Reports DSDR 37, Washington, 446,456.

-(1983) Variation in magnetization intensity and low-temperaturetitanomagnetite oxidation of ocean floor basalts. Nature, 301, 384 388

Boudeulle, M, Batis-Landoulsi, H, Leclercq, C -H, and Vergnon, P(1983) Structure of 1-FerO. microcrystals: vacancy distribution andstrxcture Journal of Solid State Chemisrry, 48,21-32

Collyer, S , Grimes, N W, Vaughan, D J, and Longworth, G. (1988) Stud-ies of the crystal structure and crystal chemistry of titanomaghemiteAmerican Mineralogist, 73, 153-160

Fukasawa, T, Iwatsuki, M, and Furukawa, M (1993) State analysis andrelationship between lattice constants and compositions including minorelements of synthetic magnetite and maghemite Analyica ChimicaActa,28l , 413-419.

Furuta, T (1993) Magnetic propefties and ferromagnetic mineralogy ofoceanic basalts Geophysical Journal International, 113, 95-114

Ghose, S (1965) Ifilg:.-psz- order in an orthopyroxene, Mguo.Fe,orSi.OuZeitschrift Fiir Kristallographie, Kristallgeometrie, Kristallphysik, Kris-tallchemie, 122,81-99

Goss, C J (1988) Saturation magnetization, coercivity and lattice param-eter changes in the system Fe.Oo-1FerO., and their relationship to struc-ture. Physics and Chemistry of Minerals, 16, 1,64-111

Greaves, C (1983) A powder neutron diffraction investigation of vacancyordering and covalence in ^y-Fe,O.. Journal of Solid State Chemistry,49.325-333.

Irving, E (1970) The mid-Adantic ridge at 45" N XIV Oxidation andmagnetic properties of basalt: Review and discussion Canadian Journalof Earth Sciences. 7. 1528 153

Jiang, W-T., Peacor, D.R., Merriman, RJ, and Roberts, B (1990) Trans-mission and analytical electron microscopic study of mixed-layer illite/smectite formed as an apparent replacement product of diagenetic illiteClays and Clay Minerals, 38, 449 468

Lindsley, D.H. (1976) The crystal chemistry and structure of oxide min-erals as exemplified by the Fe-Ti oxides In Mineralogical Society ofAmerica Reviews in Mineralogy, 3, L1-L88.

Kelso, PR., Banerjee, S K, and Worm, H-U (1991) The effect of low-temperature hydrothemal alteration on the remanent magnetization ofsynthetic titanomagnetites: A case for acquisition of chemical remanentmagnetization Journal of Geophysical Research, 96, 19,545-19,553.

Manhall, M and Cox, A, (1972) Magnetrc changes in pillow basalt due tosea floor weathering Joumal of Geophysical Research, 77,6459-6469.

Moskowitz, B M (1987) Towards resolving the inconsistencies in char-acteristic physical properties of synthetic titanomaghemites. Physics ofthe Earth and Planetary Interiors, 46,173-183.

Moskowitz, B M. and Banerjee, S K (1981) A comparison of the mag-netic propenies of synthetic titanomaghemites and some oceanic ba-salts. Journal of Geophysical Research, 86, 11,869-11,882.

Nishitani, T. and Kono, M (1982) Grain size effects on the low-temper-ature oxidation of titanomagnetite. Journal of Geophysics, 50, l3'7-142

-(1983) Curie temperature and lattice constant of oxidized titan-omagnetite Geophysical Journal of the Royal Astronomical Society, I 4,585-600

O'Neill, HSIC and Dollase, VIA (1994) Crystal structures and cation dis-tributions in simple spinels from powder XRD structural refinements;MgCrrOo, ZCrrOo, Fe.O,, and the temperature dependence of the cationdistribution in ZnAl,Oo Physics and Chemtsrry of Minerals, 20, 541-555

O'Reilly, W. (1983) The identification of titanomagnetites: Model mech-anisms for the maghemitization and inversion process and their mag-netic consequences Physics of the Earth and Planetary Interiors, 31,65 76

O'Reilly, W. and Banerjee, S K. (1967) The mechanism of oxidation in trtan-omagnetites: A magnetic study Mineralogical Magazine, 36,29-31

Ozima, M and Ozima, M (1971) Characteristic thermomagnetic curve insubmarine basalts. Journal of Geophysical Research, 16,2051 2056

Ozima, M and Sakamoto, N (1971) Magnetic properties of synthesizedtitanomaghemite Journal of Geophysical Research, 16,1035 1046

Ozima, M , Joshima, M., and Kinoshita,H (1974) Magnetic properties ofsubmarine basalts and the implications on the structure of the oceaniccrust Journal of Geomagnetism and Geoelectricity,26, 335-354

Pankhurst, Q A and Pollard, R J. (1993) Applied field Mcissbauer spec-troscopy of magnetic powders In G.J Long and EG GrandJean, Eds,Mcissbauer spectroscopy applied to magnetism and material science,vol 1, p 77-113 Plenum Press, New York

Pariso, JE and Johnson, HP (1991) Alteration processes at deep seadrilling project/ocean drilling program hole 5048 at the Costa Rica rift:Implications for magnetization of oceanic crust Journal of GeophysicalResearch, 968, 11,7 03- | 1,7 22

Peacor, DR (1967) Refinement of the crystal structure of pyroxene offormula M,M,, (Sir.A105)06 American Mineralogist, 138,331 342

Petersen, N. and Vali, H (1987) Observation of shnnkage cracks in oceanfloor titanomagnetrte. Physics of Eafth and Planetary lntenors, ,16, 197-205.

Petersen, N., Eisenach, P, and Bleil, U (1979) Inw temperahre altemtion ofthe magnetic minerals in ocean floor basalts In C G A Harrison, M. Thl-wani, and D Hayes, Eds., Deep drillhg results in the Atlantic ocean: Oceancrust, 2, p 169-209 American Geophysical Union, Washington, DC

Price, G D and Putnis, A,. (19'79) Oxidation phenomena in pleonaste bear-ing titanomagnetites Contributions to Mineralogy and Petrology, 69,355 159

Readman, PW and O'Reilly, W. (1972) Magnetic properties of oxidized(cation-deficient) titanomagnetites (Fe,Ti,n).O. Journal of Geomag-netism and Geoelectricity, 69-90.

Ryall, R J C and Hall, J M (1979) Laboratory alteration of titanomag-neites in submarine pillow lavas Canadian Journal of Earth Science,16. 496-504

Senderov, E., Dogan, A U , and Navrotsky, E. (1993) Nonstoichiometryof magnetite-ulvdspinel solid solutions quenched from 1300'C Amer-ican Mineralogist, 7 8, 565-51 3

Smith, B M (1987) Consequences ofthe maghemitization on the magneticproperties of submarine basalts: Synthesis of previous works and resultsconcerning basement rocks from mainly D S D P Legs 5 I and 52, Phys-ics of the Eanh and Planetary Interiors, 46,206-226.

Smith, PPK (1979) The observation of enantiomorphous domains in anatural maghemite. Contributions to Mineralogy and Petrology, 69,249-254

Somboonsuk, K and Trivedi, R. (1985) Dynamical studies of dendriticgrowth Acta Metallurgica, 33, 1051-1060

Toman, K and Frueh, A J (1973) On the centrosymmetry of intermediateplagioclases Zeitschrift Fuer Kristallographie, Kristallgeometrie, Kris-tallphysik, Kristallchemie, 138, 331 -342

Waychunas, G A (1991) Crystal chemistry of oxides and oxyhydroxidesIn Mineralogical Society of America Reviews in Mineralogy, 25, 11-68

Wiles, D B. and Young, R (1981) A new computer program for Rietveldanalysis of X-Ray powder diffraction patterns Journal of Applied Crys-tallography, 14, 149-151

Xu, W., Van der Voo, R, Peacor, D R, and Beaubouef, R (1994) Magh-emitization of titanomagnetite in dredged pillow basalts fiom the At-lantic ocean (abs ) Eos, 75, 5126

Xu, W., Peacor, D.R., Van der Voo, R, Dollase, WA, and Beaubouef, R(1995) Alteration of fine-grained titanomagnetite in the interstitial glassof pillow basalts from the Atlantic ocean: A component of the decreasein NRM intensity with aging (abs ) Eos, 76, S99

- (1996) Modified laftice paameter/Curie temperanue diagrams for ti-tanomagnetite/titanomaghemite within the quadrilateral Fe.O. Fe,TiOn-Fe,O.-FeJiO, Geophysical Research lrtters, 23, 2811-2814.

M,qNuscprm RECETvED DeceMeEn 9. 1996MlNuscruvr ACCEpTED Jur-v 30. 1997