American Mineralogist, Volume 74, pages 160-176, 1989

Order-disorder transition-induced twin domains andmagnetic properties in ilmenite-hematite

GonooN L. Nonn, Jn., CH.lnr.ns A. LlwsoN959 National Center, U.S. Geological Survey, Reston, Virginia 22092,U.5.4.

Ansrnacr

The transition temperature (4) between disordered, R3c, and ordered, R3, ilmenite-hematite solid solutions in the ferrian ilmenite composition range between Ilmuo and Ilm'ohas been redetermined by observing the presence or absence of transition-induced cation-ordered domains and the behavior of pre-existing domains annealed below the transition.The transition was reversed for Ilmro and is bracketed between 1000 and 1050 "C. Thedomains were shown by dark-field transmission electron microscopy to be twin related bya 180" rotation about an axis parallel to a and to vary in size depending on Z" and thetemperature of quench.

A model of the twin-domain boundary indicates that such boundaries are disorderedand partially Fe-enriched. Magnetic and reiu observations on the same samples show thatthe room-temperature saturation magnetization is related to the surface area of the twinboundaries. Because the cation-ordered domains are ferrimagnetic with a strong magneticmoment and the disordered boundaries are probably antiferromagnetic with a weak mag-netic moment, the quenched samples are essentially mixtures of two magnetic phases. Themagrretization is therefore related to the proportion of each phase.

The twin boundaries act as the classic "x" phase, which allows ferrian ilmenite to acquirea self-reversed thermoremanent magnetization (rnvr). Our measurements indicate thatwhen the surface area of the twin boundaries in Ilmro exceeds approximately 25 x 106m2/m3, the quenched samples acquire reverse rnu during cooling in a 0.5-Oe field. Whenthe boundary surface area is less than this critical threshold, the quenched samples acquirea normal rnvr.

INrnonucrrox

Minerals of the ilmenite (FeTiOrFhematite (FerOr)solid-solution series have widespread importance in pe-trology and paleomagnetism. In petrology they provide,in conjunction with other coexisting Fe-bearing phases,information on oxygen fugacity, crystallization tempera-ture, and cooling rates. In paleomagnetism they can makea substantial contribution to the magnetic signal of rocks.In particular, ferrian ilmenites in the composition rangeIlmro to Ilmrr, which are common in silicic volcanic rocks,can have a magnetic intensity approaching one-third thatof magnetite.

The physical and chemical properties of natural andsynthetic ilmenite-hematite solid solutions result in partfrom coupled interactions between cation order-disorderof Fe and Ti, phase separation in a miscibility gap, andmagnetic order-disorder. These reaction boundaries areshown schematically in Figure l. The locations of thereaction boundaries are not well known in the ilmenite-hematite binary because of difficulty in synthesis andcharacteization ofthe pure phase as well as slow reactionrates at the temperatures at and below those of the mis-cibility gap. The first and only experimental determina-tion ofthe cation order-disorder transition curve was done

0003-004x/89/0 102-o l 60$02.00

by Ishikawa (1958) for compositions between Ilmou, andIlmur. He found that the quenched, high-temperature,disordered phase is only weakly ferrimagnetic, whereasthe quenched, low-temperature, ordered phase is stronglyferrimagnetic. In a series of annealing experiments, thepoint at which the room-temperature magnetization in-tensity changed was correlated by Ishikawa with the Fe-Ti cation order-disorder transition. Recent experimentaland theoretical work on the phase-boundary topologysuggests that the cation order-disorder transition is sec-ond-order in nature and intersects the top of the misci-bility gap at a tricritical point near Hemrollm'o and 700"C (Burton, 1984). As a consequence ofthe tricritical point,samples crystallizing on either side will experience verydiferent phase transitions upon cooling (Allen and Cahn,1976). Samples crystallizing in the R3c field with com-positions more ilmenite-rich than Ilmro will undergo cat-ion ordering prior to exsolution upon cooling, whereassamples crystallizing in the R3c field with compositionsless ilmenite-rich than Ilm'o will undergo clustering (exso-lution); only the ilmenite-rich precipitate will order. Themagnetic ordering transition (Fig. l) decreases with in-creasing dilution by Ti until it is below room temperatureat ilmenite-rich mole fractions.

160

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE 1 6 1

The magnetic behavior of ilmenite-hematite solid so-lutions is unusual because samples in two compositionalranges (Ilm,r_r, and Ilmr,_rr) are capable of acquiring areverse thermoremanent magnetization (reverse rnu). A"self-reversal" occurs when the remanent magnetization,which is acquired by a magnetic mineral when cooledthrough its Curie point in a magnetic field, is antiparallelto (or the reverse of) the applied field. In the first com-positional range (Ilm,r_rr) ilmenite-hematite solid solu-tions that have cooled very slowly in plutonic and meta-morphic rocks can acquire a reverse rnvr (Carmichael,1961; Merrill and Gromme,1967). The presence of thebroad, two-phase miscibility gap results in the formationof exsolution microstructures in the rhombohedral oxidesof these rocks. Slow cooling appears to be a necessarycondition because only the titano-hematite portion (Ilm,,to llmrr) of exsolved grains carries the reversed rema-nence. Study of the self-reversal mechanism in this com-positional range has been hampered by the nonreproduc-ible nature of the reverse rnv in laboratory experiments.The cause ofthis type ofreverse rnvr is unknown but isthought to involve a magnetic coupling between two in-timately intergrown phases.

The second compositional range is that of ferrian il-menite, Ilmr,_rr, where the limits were defined by West-cott-Lewis and Parry (l97lb). Under certain crystalliza-tion and cooling conditions, ferrian ilmenite can acquirea reverse rnu. This type ofself-reversal has been relateddirectly to ferrian ilmenites that have undergone Fe-Tiordering at high temperatures during cooling. Uyeda(1958) proposed a model for this self-reversal that re-quired the interaction of a second magnetic phase withthe host ferrian ilmenite. This second phase must (1) havea higher Curie temperature than the major phase (i.e., Fe-enriched), (2) have a disordered Fe-Ti distribution (i.e.,antiferromagnetic), (3) be metastable with respect to an-nealing, and (4) couple antiferromagnetically with the hostphase. This phase was called the "x" phase by Ishikawaand Syono (1963). The mechanism proceeds as follows.During cooling in a magnetic field, the high-Curie point"x" phase takes on the direction of 'che applied field. Uponfurther cooling, the lower-Curie point phase becomesmagnetized in a direction antiparallel to the "x" phase(and thus, the applied field) because of a negative ex-change interaction between it and the "x" phase. In na-ture, ilmenite-hematite minerals in this composition rangeare only found in acid extrusives and highly oxidized in-termediate and basic extrusives.

Lawson et al. (1981) and Lawson and Nord (1984),using transmission electron microscopy, showed that syn-thetic ferrian ilmenite (Ilmro and Ilmro) grown at 1300.Cin the high-temperature, R3c, disordered field (Fig. l)forms transition-induced domains and domain bound-aries upon cooling through the transition temperature intothe low-temperature, R3, ordered field. We also demon-strated that a high domain-boundary surface area wasnecessary for the operation ofthe magnetic self-reversalmechanism in synthetic Ilmro and inferred that the do-

T,oC

700

i ty Gop

Mognelic

Fe2O3 FeTiO3

Fig. 1. Schematic ilmenite-hematite binary phase diagramshowing the R3c to R3 cation order-disorder transition, the mis-cibility gap between disordered hematite and ordered ilmeniteand the magnetic order-disorder transition (Curie points). Thecation order-disorder transition and miscibility gap intersect ata tricritical point at approximately 700 'C and Ilmro (Burton,1984). The two compositional fields of samples that exhibit aself-reversing thermoremanent magnetization are shown as heavylines.

main boundary was the "x"-phase (Hoffman, 1975b, hadpreviously suggested that antiphase boundaries could actas the "x" phase). Similarly, studies of dacitic pumiceblocks from Mount Shasta. California. that exhibited re-verse rRM showed the presence of a high density of tran-sition-induced domain boundaries in ferrian ilmenite,Ilm,r_r, (Lawson et al., 1987). This suggests that the samemechanism responsible for self-reversal in ferrian ilmen-ite synthesized in laboratory experiments is also respon-sible for self-reversal in natural ferrian ilmenites in dacit-ic volcanic rocks.

The goal of our research in the ilmenite-hematite bi-nary is to map out the reaction boundaries and under-stand the changes in magnetic properties with time andcooling history. In this paper we are concerned only withthe order-disorder transition in the composition rangeIlmuo to Ilm,oo and the magnetic properties of quenchedIlmro samples. The location of the transition was deter-mined by observing changes in domain microstructurethat arise when samples are quenched from different tem-peratures above and below the transition. The kinetics ofgrowth of the domains was monitored using samples thatwere heat-treated at different temperatures below thetransition for different lengths of time. The result is aninterpretation of the magnetic properties of ferrian il-menite that considers the mechanism of formation andcoarsening of the transition-induced microstructure, thekinetics of the high-temperature cation order-disordertransition, and the coupled interaction with the low-tem-perature magnetic ordering transition. It is the interactionbetween the microstructure and the magnetic transitionthat gives rise to the self-reversed rnvr. The reader shouldnote that the term "domain" in this paper will always

162 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE.HEMATITE

TraLe 1. Compositions and volumes of svnthetic ilmenite-hematite

900.c '1300 "c

l lmloollmeollm6llmTo llm?o llm8ollm@

Tio,FeOFerO.

Total

llm (mol%)No. analyses(1 o)

v.. (A)%", (A)

33.7630.3436.37

100.47

64.96'14(0.81)

1 0 3 . 1 7309.51

36.0132.413 1 . 1 199.53

69.841 3(1 .8)

103.34310.02

41.9937.7121.38

101 .08

79.721 4(1.02)

103.9431 1 .82

47.0342.2610.4699.75

90.00't0(0.46)

104.43313.29

cz.oc47.360.60

100.61

99 431 2(0.43)

10s.09315.27

3 0 927.944.29 9 0

o u o1 4(0.7)

36.332.629.8987

7 0 91 3(1 2)

t Oxides and totals are given after recalculation of ZAF-corrected microprobe data following Carmichael (1967).

refer to cation-ordered domains and not to masneticallvordered domains.

ExpnnrntnNTAl TEcHNTeuES

Compositions along the join FerOr-FeTiO3 were synthesizedat 1300 and 900'C. Different experimental techniques were usedfor each temperature. Synthesis at 1300'C was carried out bysuspending sintered pellets of a FerO, + TiO, mix in a verticaltube fumace through which flowed a mixture of COr-CO or COr-H, gas to control the oxygen fugacity. For a given composition,the oxygen fugacity in preliminary experiments was varied justuntil no magnetite-ulviispinel phase was detected by light opticaland seu techniques in the run product. All of the 1300 "C Ilmroruns were carried out in a vertical tube furnace fitted with aY-doped zirconia cell at a log/", of -5.68. After a synthesis timeof36 to 48 h, the pellets were dropped through the furnace intoHg or liquid Nr, resulting in a quench time to room temperatureof approximately 10 s. Grain sizes produced by the gas-flowmethod at 1300 "C ranged from 30 to 80 pm. Synthesis at 900oC was carried out using a stoichiometric mixture of TiOr, FerOr,and Fe metal wrapped in Ag foil, which was then sealed in anevacuated silica-glass tube and heated for approximately twoweeks. The silica-glass tube was dropped into water at the endof the experiment, resulting in a sample quench time to roomtemperature of approximately I min. Grain sizes in the 900 'C

silica-glass tube samples averaged 5 pm. The silica-glass synthe-ses produced <50/o magnetite-ulv<ispinel in addition to therhombohedral phase. A detailed discussion of the silica-glasstube synthesis can be found in Lawson (1978) and the gas-flowfurnace synthesis in Lawson (1981) and Huebner (1987). Ilmropellets, synthesized at 1300'C, were wrapped in Ag foil andannealed at 700-1000 'C in evacuated silica-glass tubes.

Run products were analyzed using reflected-light microscopy,electron-microprobe analysis, and X-ray powder diffraction (Ta-ble l). Electron-microprobe analyses were conducted at the Geo-physical Laboratory, Washington, D.C., using a rraec +oo' elec-tron microprobe operating at 15 keV. Matrix effects werecorrected with the ZAF scheme of uacrc rv. FerO. was deter-mined using the recalculation scheme of Carmichael (1967). Unit-cell volumes were calculated with a least-squares refinementprogram from Debye-Scherrer powder data using CaF, as aninternal standard and Fe-filtered CoKa, radiation.

' Use of brand names is for the purpose of identification onlyand does not constitute endorsements by the U.S. GeologicalSurvev.

Transmission electron microscopy was conducted on a JEoLzooe operating at 200 keV at the U.S. Geological Survey, Reston,Virginia, and a Philips 420T operating at I 20 keV at Johns Hop-kins University, Baltimore, Maryland. Samples were first pre-pared as ultra-thin thin sections mounted with an acetone-sol-uble thermoplastic cement (see Wandless and Nord, 1986, for adescription of this technique). Because of their fine-grained na-ture, the required final thicknesses achieved by hand polishingwere often less than 10-15 pm. Portions of these were attachedwith superglue to either Cu grids or washers depending on thesample size and coherency and thinned to electron transparencyby Ar-ion milling using 6-keV accelerating voltage and 18" tilt.A thin surface phase forms at 6-keV thinning voltages and isvisible only as weak diffuse reflections in selected-area diffrac-tion patterns. This phase was removed by final thinning at I keVfor 24 h.

All diffraction patterns and real space planes and directionsare indexed in the structural hexagonal unit cell by using theobverse setting of the hexagonal cell with respect to the rhom-bohedral cell. Although this statement may seem trivial, there ismuch confusion in the literature, especially in the indexing ofhexagonal forms and planes (see Snow and Heuer, 1973).

The average domain size was measured by determining theaverage intercept length (I) of a test line (I'), following thetechnique of Smith and Guttman (1953). In this case, a standardtest circle of approximately l0- or l5-cm circumference (l-) waslaid over the micrograph, and the number of circle-domainboundary intersections (A) counted. The circle was applied sev-eral times to each micrograph until greater than 50 intersectionswere counted. The average intercept length is then determinedfrom the following relationship, where I is the number of ap-plications of the test circle and M the magnification of the elec-tron micrograph. The average domain size L: Lr@)/N( O.

Smith and Guttman (1953) also pointed out that the averagenumber ofintercepts per unit length ofa random line (test circle)drawn through a three-dimensional structure is exactly halfthetrue ratio ofsurface to volume. Therefore, surface area per unitvolume (S") was determined by the relationship Sn : 2(N(MlL,(A).

Saturation magnetization, ,I,, was determined on an automat-ed Curie balance at the Rock Magnetics Laboratory, PrincetonUniversity. Measurements were made at room temperature infields up to a maximum of 7500 Oe. "/" was calculated by ex-trapolating the data to infinite field. The balance was calibratedwith standard Ni samples. To check whether the reverse Nnu ofthe samples is reproducible, we performed rru acquisition ex-periments on the samples. For these experiments, the samples

3c1:]

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE l o J

[-c tCO

totof-

c

<_qF+

Fig. 2. Structure of R3c ilmenite-hematite projected on to(1120). The a axis is normal to this view. Fe and Ti are disor-dered over the octahedral cation sites (small open circles) thatlie in the plane of the diagram. The large heary and light opencircles represent oxygen atoms lying in front ofand behind thisplane, respectively. Groups ofthree oxygens called triplets formthe common face of two face-sharing octahedra. In the R3cstructure, these cation sites are equivalent, as in the corundumand hematite structure and the disordered hematite-ilmenite sol-id solutions. A twofold rotation axis pierces the triplet and re-Iates the common cations. Other symmetry elements in the pro-jection are shown: centers (show.n as asterisks), twofold rotationand screw axes parallel to a, and c-glide and threefold rotationaxes parallel to c.

were heated in air at 620 oC in zero field for approximately 45min, after which they were allowed to cool to room tempgraturein an applied field of 50 pT (0.5 Oe). progressive AC demag-netization of the samples after acquisition of rnw was performedin AC f ields up ro 120 mT.

Srnucrunn oF TLMENTTE-HEMATTTE

The solid solution between hematite, R3c, and ilmen-ite, R3, is characterized by a Fe-Ti order-disorder tran-sition. The R3c structure represented by hematite anddisordered ilmenite (Lindsley, 1976) consists of sheets ofoxygen anions in a nearly hexagonal close-packed arraylying parallel to (0001) (Fig. 2). Fe and Ti occupy thecation positions, as in the corundum structure. In purehematite, Fe3* is the only cation occupying these sites,whereas in high-temperature, disordered ilmenite-hema-tite solid solutions, Fs3*, Fe2*, and Ti4+ are randomly dis-tributed over the cation sites. Groups of three oxygenscalled triplets form the comrnon face of two adiacent oc-tahedra, and each ofthese oxygens in the triplei is linkedto the two cations in these octahedra (plus a third cation

+Q. .8+

Fig. 3. Structure of R3 ilmenite-hematite projected on to(1 120). This particular structure is of ordered ilmenite with anequal number ofFe and Ti cations. The cations are ordered ontoalternating planes; by convention Fe lies on A planes and Ti onB planes. The twofold rotation axes in the triplets are lost duringthe transition because the common pair of cations are no longerequivalent. The remaining symmetry elements are the centersand the threefold axes parallel to c.

in a lateral edge-shared octahedron). In the R3c structure,these two cation sites are equivalent. A twofold rotationaxis pierces the triplet and relates the common cations.

The R3 structure is represented by ordered ilmenite-hematite solid solutions (Fig. 3). The octahedral cationpositions are occupied as in the disordered form, but thecations are segregated onto alternating cation layers suchthat in pure ilmenite, layers contain only Fe2t or onlyTi4". With increasing hematite component, Fe3* substi-tutes for Fe2* and Ti4+ in both layers. By convention, theFe-rich layer is termed the A layer, and the Ti-rich layeris termed the B layer. As an additional consequence ofthe cation ordering, the oxygens move away from thelayers with the larger Fe cations and toward the layerswith the smaller Ti cations. Both the cation ordering andthe oxygen displacements destroy the c-glide and the two.fold rotation and screw axes of the high-temperature, R3c,disordered phase. Only the centers of symmetry andthreefold axes remain. This symmetry loss results in theappearance ofadditional reflections ofthe type hfr\[ l:2n * l, in the ordered phase. The unit-cell size in bothspace groups is similar.

The magnetic structure of the ilmenite and hematiteend-members and the intermediate solid solution is com-plex (Ishikawa,1962;' Lindsley, 1976). Above the mag-netic ordering transition in Figure l, all phases are para-magnetic. Below the magnetic ordering transition, thedisordered R3c hematite and hematite-rich phases up to

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE 165

compositions of approximately llmor_ro are antiferro-magnetic. In the disordered structure, Fe ions are ran-domly distributed on the different layers, so the total mo-ment of each individual layer is the same. Couplingbetween adjacent layers is antiferromagnetic. The cou-pling, however, is imperfect, resulting in a weak parasiticferromagnetic moment for the disordered phase of lessthan I emu/g. For samples with compositions betweenapproximately Ilmro and Ilm,oo, the phases are of the or-dered R3 structure, and between Ilmro and Ilm*., they areferrimagnetic at room temperature. Because the concen-tration of Fe on the A layers is greater than that on theB layers in ordered phases between Ilmro and Ilmro, thetotal magnetic moment due to A-layer contribution isgreater than that due to B-layer contribution. This givesrise to a ferrimagnetic structure with room-temperaturesaturation magnetizations that can be greater than 30emu/g. The transition between the weak parasitic ferro-magnetic ordering and the strong ferrimagnetic orderingis somewhere in the range of Ilm.r_r, and is highly depen-dent on the kinetics of the cation-ordering transition athigher temperatures as well as sample synthesis and ther-mal history. Compositions greater than Ilmro remain mag-netically disordered at room temperature and are thusparamagnetic. Short-range magnetic order and other com-plications do occur, but only below room temperature.

Tn,q,NslarssroN ELECTRoN MrcRoscopy ANDREFLECTED-LIGHT OBSERVATIONS

Smoothly curving domain boundaries were observedby transmission electron microscopy in all synthetic il-menites that were crystallized or annealed above the tran-sition curve (Figs. 4a-4d). No domain boundaries wereobserved in those samples crystallized at temperaturesbelow the transition. Coarsening of the domain micro-structures was observed in samples that had been syn-thesized above and quenched through the transition andthen subsequently annealed at temperature near but be-low the transition (Figs. 4g-4h). The domain boundariesare visible in dark-field images formed with the extrareflections that appear in the ordered phase, especially0003 consistent with antiphase domain boundaries.However, it is also possible to see the domains clearly by

a variation in background contrast between adjacent do-mains when using certain fundamental reflections (Fig.5). This contrast is consistent with twin-domain bound-aries. In order to determine the type of domain boundaryby transmission electron microscopy, a contrast analysisof two-beam conditions with low-order reflections needsto be performed. This analysis is discussed in the nextsection.

If large enough, the domains are also visible opticallywhen viewed in partially polarized reflected light in thepetrographic microscope. They can be seen by a slightcontrast difference (Fig. ah) when the analyzer of the pet-rographic microscope is rotated several degrees from thepolarizer. Under these conditions, the domains changecontrast as the stage is rotated, and the domain micro-structure is easily seen. This optical behavior is consistentwith the domains having a twinning relationship.

Ilmur, Ilmro, and Ilmro grown at 1300 oC and Ilmuo grownat 900 "C all have a domain microstructure (Fig. 4) in-dicating that they crystallized as the disordered phase andthat the transition must lie below their crystallizationtemperature. Ilmno and Ilm,oo grown at 1300'C and Ilmrogrown at 900 "C do not contain any domain boundaries,indicating that they crystallized as the ordered phase andthat the transition lies above their crystallization tem-perature.

The shape of the domains varies from slightly flattenedparallel to (0001) in Ilmuo and Ilmu, (these have the small-est domain size) to equidimensional in the more ilmen-ite-rich samples (these have progressively larger domainsizes). The flattening of the finest domains results in elon-gation ofthe l[0/ ordering reflections parallel to c (Fig.4e).The domain size of Ilmrr, Ilmro, and Ilmro quenched from1300 "C increases almost an order of magnitude for eachadditional 10 molo/o ilmenite component. An additionall0 molo/o ilmenite component increases the transitiontemperature by approximately 200 oC. Therefore, the do-main size increase simply reflects an increase in diffusionrates with increasing temperature.

The transition was reversed for the composition Ilmroby taking samples synthesized at one temperature andannealing them at another. Fine domains present in the1300 'C quenched sample coarsened rapidly at 1000 'C,

indicating that the transition is above 1000 'C for Ilmro.

Fig. 4. Dark-field transmission electron micrographs of do-main boundaries in quenched synthetic ilmenite-hematite solidsolutions. (a) Ilmuo quenched from 900'C showing very finedomains. (b) Ilmu, quenched from 1300 "C showing 300-A av-erage- domain size. (c) Ilmro quenched from 1300 .C showing700-A average domain size, and (d) Ilmro quenched from 1300'C showing domains greater than 1000 A. All micrographs weretaken at the same magnification with the ordering reflection g :0003 at 120 keV (a and b) and 200 keV (c and d). (e) and (f)Electron-diffraction patterns for the zones [1120] in Ilmuo andII00] in Ilmro, respectively. In the Il120] zone, the superlattice

reflections (hfr01, where I : 2n + l) are weaker than the fun-damental reflections and are slightly elongated parallel to c as aresult of the flattened shape of the fine domains in Figure 4a. Inthe zone [1 I00], only 0001, I : 2n f I are superlattice reflections.These reflections, 0003 and 0009, would appear by double dif-fraction in this zone in any case. (g) Dark-field electron micro-graph of coarsened domains in Ilmro (g : 0003, 200 keV). Thissample was crystallized at 1300 'C, quenched, and then annealedat 900'C for 100 h. (h) Partially polarized, reflectedJight mi-crograph of coarsened domains in Ilmro that was quenched from1300'C and annealed at 1000 .C for l0 hr.

t66 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

Fig. 5. Dark-field transmission electron micrographs of domain boundaries from the same area of Ilmro annealed at 900 'C forI h. The operating reflections, g, indicated on each micrograph correspond to the [1100] difraction pattem in Fig. 4f. An arrow oneach micrograph indicates the position of the boundary of a single closed domain. All micrographs at 200 keV.

A portion of the 1000'C sample with the coarsened do- mediately upon quenching. Samples crystallized at 1300mains was then annealed at 1050 'C for l0 h and "Candsubsequentlyloweredto 1200, l l00,and 1050'Cquenched. It was found to contain fine domains, indicat- and then quenched to room temperature contained pro-ing that it had disordered at 1050 'C and re-ordered im- gressively finer domains, indicating again that the tran-

etq

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE t67

sition lies below 1050 'C. The equililbrium transition,therefore, has been bracketed between 1000 and 1050'Cfor Ilmro.

CoNrnLsr ANALySIS oF cATIoN-ORDERED DoMAINSAND DOMAIN BOUNDARIES

The crystallographic relationship between adjacent do-mains can be determined from an analysis of the contrastof the domains and the domain boundaries under variousdiffraction conditions in the transmission electron micro-scope (Amelinckx and Van Landuyt, 1976, provide anexcellent review). In general, two tlpes of interfaces aredeveloped as a result of an order-disorder transition: atranslation interface and a twin interface. For orderingtransitions that undergo a change in the size of the unitcell (formation of a superlattice), the domains are relatedby a translation vector that is a lattice vector ofthe dis-ordered phase. For ordering transitions that result in theloss of a symmetry operation, the domains are related bya symmetry operation of the disordered phase. Ame-linckx and Van Landuyt (1976) described the image con-trast, extinction criterium, and effects on the diffractionpattern that are indicative ofthese diferent planar inter-faces. The most pertinent for our quenched products isthe fact that domains related by only a translation nevershow any contrast difference between domains whereasdomains related by a twinning operation commonly showcontrast differences between adjacent domains for manyfundamental reflections. The domains in ordered hema-tite-ilmenite solid solutions show contrast differences andtherefore must be related by a twin operation, and thedomain boundaries must be twin interfaces. This is alsoconsistent with the reflected-light observations.

In order to adequately describe the diffraction contrastof the twin-domain microstructure, we must construct amodel that will account for all observations. The diffrac-tion contrast of the domains and the domain boundariesis a function of the amplitude and phase of the operatingreflections diffracted by each domain. Table 2 lists thecalculated structure factors lFoo,,l and phase angles (a)for given reflections hkil of the host and of the twin. Inorder to calculate the values for the twin, the atomic co-ordinates of the host were transformed by the relation-ship X, : Y", Y, : Xn, and Zt : 0.5 - ZH. This trans-formation relates the atomic coordinates of the host tothe twin by a 180'rotation about an axis parallel to thea axis (two-fold axis in R3c). Because the disordered (R3c)and ordered (R3) phases are centrosymmetric, the twinoperation is a rotation and not an inversion.

The amplitude of the transmitted or diffracted electronbeam exiting the specimen that forms an image by dif-fraction contrast is mainly a function of the deviation (s)from the exact Bragg angle and the extinction distance({). The extinction distance is the critical distance in aperfect crystal over which the intensity of the diffractedbeam varies from zero to some intensity and back to zero.The extinction distance depends on the operative reflec-

Treue 2. Extinction distances in ordered ilmenite

Struc-ture a (

h k i t tactor f) (A) t",lt""

HostTwin

HostTwin

HostTwin

HostTwin

HostTwin

HostTwin

HostTwin

HostTwin

HostTwin

0 00 0

0 00 0

0 00 0

0 00 0

1 21 2

1 21 2

1 21 2

1 21 2

1 21 2

5 5 35.53

29 4529 45

10.3810.38

24.7024.70

24.4024.40

59.9649.04

44.3058.61

31 9221.93

4.6217.95

71477147

13421342

38093809

1 6001 600

16201 620

806

892674

1 2381 802

85582202

0.82

0.75

0.69

00

00

00

00'I

,l

1

1

1

1

I

1

1'I

o

o

6

I

I

1 21 2

002

o

o

9a

1 2t z

1 . 0

1 . 0

1 . 0

1 . 0

1 0

1800

180180

0180

00

00

00

00

180180

180180

Note:200 kev, s- 0.

tion and may be calculated from the relationship (Hirschet a l . , 1965)

f : (a'Z.cos O)/(>\Fhk,),

where Z. is the unit-cell volume, d is the Bragg angle, andtr is the electron wavelength calculated at the operatingvoltage of the rnv.

Extinction distances in Table 2 were determined forpure ordered ilmenite with electron scattering factors fromSmith and Burge (1962) and atomic positions from Ray-mond and Wenk (1971). These distances are applicableonly for the exact Bragg position where s : 0 and are onlyaccurate to within 100/0, primarily because of uncertaintyin the electron scattering factors.

The data in Table 2 can be compared to dark-field two-beam observations in the transmission electron micro-scope for the zone [ 100] (Fie. 5).

For g : 0003, we flnd that ,F"{0003} : F,{0003} andthat the diference in the phase angles is 180'. Thereforeboth domains should show the same intensity, and thedomain boundaries should exhibit the fringe contrast ofa zr fault (Fig. 5a). Because the extinction distance for{0003} is large, almost 0.75 pm, only the central darkfringe is visible in micrographs taken in thin areas. Figure49 of the 900 "C, 100-h anneal, however, was taken in athicker area at low magnification and a central black fringebordered by white fringes can be seen. This is typical ofthe dark-field symmetry of a' fringes. Similarly, for g :0009, symmetrical fringes are visible at the boundary,and the domains have identical diffraction contrast.

For g: 0006 we find that FH{0006} : F,{0006} andthat the difference in the phase angles is zero. Both do-

1 6 8 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

Fig. 6. Transition-induced twin-boundary model. Projectionof the crystal structure of Ilmro on to (l 120). For this composi-tion, approximately 7.7 of the cations are Fe and 4.3 are Ti perunit cell. During cation ordering, the right side of the modelordered on an A-B scheme while the left side ordered on a B-Ascheme, where A layers contain only Fe and B layers contain Feand Ti. Upon growth and impingement of the two ordered do-mains, a twin boundary is developed (stippled band). The do-mains are related by a twofold axis at the boundary representedby the filled oxygens. A 1 80' rotation about these axes will bringthe A and B layers as well as the oxygen lattice back into register.

mains should show the same intensity, and the boundaryshould not be visible. This is in fact observed. Similarly,for g : 0.0.0.12 and 1 120 (Fig. 5b), no boundary contrastis visible, and the domains have identical contrast.

Fo rg : l l 23 ,we f i nd tha t F " t l l 23 f + F r l l l 23 t , bu tthat the difference in the phase angles is zero. In thissituation, adjacent domains should show an intensity dif-ference for a crystal ofconstant thickness, and the bound-aries should not show fringe patterns. Images using g :I 123 (Fig. 5c) show thickness contours that are offset whenthey cross the domain boundaries, indicating a change inextinction distance between the two simultaneously op-erating reflections from the host and twin. The boundaryis barely visible with weak fringes in thicker areas. Thereflections 1126, 1129, and 1.1.2.12 also give unequalstructure factors between the host and twin. The ratio ofthe extinction distance between that ofthe host and thatof the twin decreases as / increases, and we would there-fore expect that the domains will show increasing diffrac-tion-contrast differences as / increases. This is observedin the images (Figs. 5d-50. In fact, the extinction dis-tances for g: 1.1.2.12 are so different as to permit ablack-white intensity difference between the domains. Inthis set of images, the domain boundary shows a weakfringe contrast. Mclaren and Phakey (1969) showed forDauphin6 twins in quartz (also 180'rotation twins) thatthe weak fringes in the boundary do not arise from any

phase diference or misorientation between the twins butfrom the fact that the extinction distances in the two twinsare widely different.

TwrN-noulrN BoUNDARY MoDEL

The twin-domain boundaries form during coolingthrough the order-disorder transition. At high tempera-ture, Fe and Ti are disordered over the cation sites. Asthe transition temperature is approached, small orderedareas began to form with Fe concentrating on one layerand Ti on the adjacent layers. These areas can order in-or out-of-phase with adjacent ordering areas. Ifadjacentareas order in-phase, then upon growth and impinge-ment, a perfect match is achieved and no boundary re-sults. Ifthe adjacent areas order out-of-phase, then upongrowth and impingement, Fe-rich layers abut Ti-rich lay-ers and vice versa. Figure 6 illustrates such a case forferrian ilmenite with a bulk composition of Ilmro. The Aand B cation layers of one ordered domain are out-of-phase with the A and B layers of the adjacent ordereddomain. The stippled boundary between the domains hasits own local symmetry. Twofold rotation axes that werepresent in the disordered phase are retained in the bound-ary. Rotation of 180" about any one of the filled oxygenswill bring the A and B cation layers of the adjacent do-mains into register. Thus, the model boundary is a twinboundary as required by the diffraction-contrast experi-ments.

The twin-boundary model in Figure 6 provides someinsight into the properties of the boundary. (l) Theboundary represents a ribbon of disordered R3c phasebecause of the presence of the R3c symmetry elements.This implies that the boundary region should consist ofa weak, parasitic, ferromagnetic moment similar to a dis-ordered hematite-rich phase, in contrast to the ferrimag-netic cation-ordered domains. (2) The boundary is locallyenriched in Fe or Ti. If the adjacent layers are both Alayers, the boundary is enriched in Fe, which results in aribbon of magnetic phase with a higher Curie tempera-ture than the ordered domains. There is some indepen-dent evidence for Fe enrichment of the boundaries. Hoflman (1975a) found that the blocking temperature of the"x" phase in Ilmuo was approximately 340'C, which cor-responds to approximately Ilmoo-'. (3) The boundary hasa surface energy because oflattice strain. Ordering ofFeand Ti on adjacent layers causes the nearly close-packedoxygen layers to move toward the Ti-rich layer and awayfrom the Fe-rich layer because of the difference in atomicradius of the two cations. This adjustment of the dis-tances between layers must reverse itself upon crossing atwin-domain boundary. The strain between the two do-mains will be maximum for boundary orientations par-allel to c and a minimum for boundary orientations par-allel to the basal plane. Thus the domains are expectedto be flattened parallel to the basal plane as seen in Figure4. This flattening is confirmed by the elongated orderingreflections in selected-area electron diffraction patterns offine-grained twin domains.

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE r69

1050 0c

Fig. 7. Dark-field transmission electron micrographs of domain microstructures in Ilmro quenched from temperaiures progres-

sively closer to the transition temperature at approximately 1025 oC. Each sample was crystallized at 1300'C for 48 h, and then

the furnace temperature was reduced to the annealing temperature indicated on the micrographs; the gas-mixing ratio was held

constant. The samples were annealed for 10 h and then quenched into liquid Nr. The resulting domain size is reduced by a factor

oftwo as I. is approached.

V.lnrarroNr oF DoMAIN sIZE wITH eUENCHTEMPERATURE

The size of the twin domains that arise from coolingthrough the order-disorder transition progressively de-creases as the quench temperature approaches the tran-sition temperature. Figure 7 shows electron micrographsof twin domains in Ilmro samples that were quenchedfrom 1300, 1200, 1100, and 1050'C; mean domain di-ameters are 706, 633, 520, and 370 A, respectively. Eachsample was synthesized at 1300 oC, and then the furnacetemperature was lowered while maintaining the same COrlH, ratio of 99. The samples were then held for l0 h toequilibrate at the lower temperature before quenching toroom temperature. At temperatures below 1200 oC, thebuffering gases trended offthe equilibrium curve and be-came slightly more reducing than the calculated equilib-rium value. At 1050 "C, the oxygen fugacity was approx-

imately 0.5 log unit more reducing than the equilibriumvalue. However, examination of the quenched samplesindicated that no spinel phase was formed.

There are a number of possible explanations as to whythe domains formed from higher quench temperaturesare larger. (1) The high-temperature quenches are at sub-I" temperatures for a longer period of time, allowingcoarsening offiner domains. (2) Greater vacancy concen-trations occur at the higher temperatures and thus in-crease diffusivities at 4. (3) Short-range order (SRO) just

above the transition temperature provides abundant clus-ters of the ordered phase, providing a higher density ofordering sites and consequently forming finer domains.The further the quench temperature is above 7", lheweaker the SRO clustering. (4) Diflusivity is a functionof oxygen fugacity, and the lower-temperature quenchesare more reducing. Any of these explanations or a com-bination ofthem are possible causes for the decrease in

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

- \"*

Fig. 8. Dark-field transmission electron micrographs of Ilm,o annealed at 800 .C for (a) 0. t h, (b) 1.0 h, (c) 10.0 h, and (d) 100.0h. g : 0003 in all micrographs.

t70

domain size with quench temperature; the fact that thedecrease exists is important when understanding themagnetic properties and extrapolating the properties ofthe synthetics to natural ferrian ilmenite occurrences.

Coa.nsrNrNc KINETTCs AND MoRpHolocy oFTHE TWIN DOMAINS

The surface energy of the twin-domain boundaries isdue to lattice strain at the boundary. This surface energyis reduced by reducing boundary area or aligning theboundary along orientations of minimal strain. Boundaryarea is reduced by enlarging or coarsening the domains.The kinetics of domain coarsening were studied by an-nealing quenched samples, with a fine initial domain mi-crostructure, in the single-phase field below the order-disorder transition. Nord and Lawson (1988) found thatthe domains coarsen according to the relationship D, :k/ where n : 2.44, D is the domain diameter, ft is a rateconstant, and / is time. The activation energy for coar-sening was found to be 77 kcaV(deg.mol) [32] kJ/(deg.mol)1. The fact that the coarsening rate does not follow a

parabolic law (n : 2) is attributed to nonstoichiometryand possibly to concentration ofvacancies on the bound-ary.

The increases in domain size during the coarsening ex-periments are given in Table 3 where the domain diam-eters were determined by the previously described inter-cept method. The domains coarsen by straightening outcurved boundaries and forming small totally enclosed do-mains that shrink and disappear. These small domainscan be seen in Figure 5 where they form football shapeswith the long dimension parallel to (0001). Figure 5a alsoshows a good example of a domain in the process ofpinching offfrom a larger domain. This domain will shrinkand disappear with further annealing below 2". Duringthe coarsening process, the domain boundaries also ap-pear to orient themselves along rhombohedral planes.Figure 8 illustrates this in a sequence of annealing runsat 800 "C. Unfortunately, the orientation is not developedenough to determine it exactly. The strong curvature andsteplike surface of the domain as it swings through planesparallel to c certainly indicates a high boundary energy in

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

TABLE 3, Microstructure and magnetic intensity of llm?o

r71

Thermal historyAverage domain size' Boundary area

( x 10u mrlm.)

Saturationmagnetization

(emu/g)

70,02,0070,20,0070,21,0070,23,0070,29,0070,228,OO

70,148,0170,048,0170,078,01170,028,0170,328,0170 ,158 ,0170,058,0170,088,0170, '128,0170,308,0170,068,0170,098,0170,338,0170 ,118 ,0170 ,168 ,01

1300 (37 h) , O1300 (67 h) to 1200 (10 h) , O1300 (41 h) to 1100 (10 h) , O1 300 (46 h) to 1 050 (1 0 h), O1300 (41 h) to 1000 (10 h) , o1300 (41 h) to 1000 (10 h) , O

to 1050 (10 h) , O

Runs annealed in silica-gtass tubes for the duration indicated+900 (0 t h), Q 5074 + 214 3.9 + 0.2

706 + 54oJu a cu520 + 50370 + 502 000 000

n.o.

28.3 + 2.231.6 + 2 .53 8 3 + 2 . 5c4.c ! z.c

0.02 est.n d .

30.5 + 0.629.3 + 0.326.8 + 0.526.1 + 0.430.9 + 0.3

n.d.

34.2 + 0.8J 5 . 4 a U . O

28.9 + 0.836.0 + 0.332.4 + 0.731.7 + .t.2

32.7 + 1.434.6 + 0.335.7 + 0.230.8 + 0.531.8 + 0 .233.4 + 0.132.8 + 0.934.1 + 0.6J C , C

900 (1 h), o900 (1 0 h), o900 (1 00 h), o800 (0.1 h), o800 (0.1 h), o800 (1 h), o800 (1 0 h), o800 (100 h), o700 (0.1 h), o700 (1 h), o700 (1 0 h), o700 (10 h), o700 (100 h), o700 (1000 h), o

11992 + 400411749 + 2248350 000 est.

n.o.1296 + 983752 + 2928293 + 713

25008 + 6272n.d.

876 + 662084 + 132

n.o5086 + 754

11944

1 .8 + 0 .61 .8 + 0 .40.06 est.

n.o.15 4 + ' t .2

5.3 + 0.32.4 + O.20.8 + 0.2

n.d22 .9 + 1 .7

9.2 + 0.6n.o.

3.5 + 0.41 7

Notei n.d : no data. Q : ouench.. Plus or minus 1o standard deviation.t Magnetite precipitated in this run during annealing.+ The starting materials were synthesized at 1300'C in the gas-mixing furnace

those orientations. The steps are rhombohedral planesparallel to { I 104} and { I 102), and their presence mini-mizes the boundary surface energy.

TwrN-souNDARy SURFACE AREA, ANNEALING TIME,AND SATURATION MAGNETIZATION

Ishikawa (1958) showed that the room-temperaturesaturation magnetization, "I", of quenched ilmenite-he-matite solid solutions is dependent on the degree of cat-ion order. The disordered R3c phase has a weak magneticmoment, whereas the ordered R3 phase has a strong mag-netic moment. In addition, "I" decreases when the quenchtemperature approaches the order-disorder transition.This is because the degree oflong-range order in the cat-ion-ordered domains decreases as Z" is approached. TheIlmro microstructure shown in the previous figures con-tains both disordered domain boundaries and ordereddomains. Therefore, the room-temperature saturationmagnetization (,I") of quenched Ilmro will depend on thevolume percent of disordered twin-domain boundary andordered domains as well as the degree of long-range orderin the domains themselves. Because we cannot measurethe thickness of the domain boundary, we cannot deter-mine the volume ratio; however, the twin-boundary sur-face area per unit volume (S") can be measured. Sn and

"I" are tabulated in Table 3 for all samples. Figure 9a showsthat S, decreases with increasing annealing time, whichis expected because of domain coarsening. It also showsthat the reduction in S, is much faster at higher temper-atures where diffusion rates are higher. Figure 9b showsthat as Sn decreases, ,I" increases. This trend is also ex-

pected because the rEM study indicates that the twin-do-main boundaries are ribbons ofthe disordered phase andtherefore will have a weaker "I" than the ordered phase.As the volume of the disordered boundary phase de-

o.t l.o loo loo.o looooLog Anneoling Time (hours)

20 25 30 35Iniensity of Mognetizotion ot RT (emu/gm)

Fig. 9. (a) Twin-boundary surface area versus log annealingtime. (b) Twin-boundary surface area versus room-temperaturesaturation magnetization. Symbols are as in (a); the 1000 "Canneal is an inverted triangle, and the four samples quenchedfrom above 7" are hexagons.

40

7060

40?n

20t oo

6-E

'E 30@o5 z ooE=g ro:b9o t0a

4 r o

Seof a ^

(, rw

3 4 0oE 3 ca

E2cs t a

, t ' "

t72 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

40

Eq

Eo- - ? ^

F " "(ro

o

N 4 V

o

oo

o l oi

oc

transition-temperatrue quenches were able to order sig-nificantly during quench (Fig. 7), whereas the lower-tran-sition-temperature quenches (Fig. 4a) were not. There-fore, the Ilmr,,r' curve follows the expected Bragg-Williams approximation, whereas disorder in the moreilmenite-rich compositions is more difficult to quench.

Particularly interesting is the area above the transitionwhere the magnetization increases with increasing quenchtemperature for Ilmro and Ilrnr,,ur,. Ishikawa (1958) sug-gested that samples were able to partially reorder, thusresulting in an increase in the magrretization.In fact, ourseries ofexperiments quenched from 1300, 1200, 1100,and 1050 oC shows that the disordered phase is notquenchable. The selected-area diffraction patterns all givestrong and sharp ordering reflections from the orderedR3 phase. What is apparent is that the amount of disor-dered domain boundary is higher in the samples quenchedfrom temperatures just above Z. (Figs. 7, 9b, and Table3). The samples quenched closest to f, therefore, wouldbe expected to have lower magnetization values. Themagnetization trend above I. will also depend on theability of the ordered domains to approach the fully or-dered configuration. The kinetics of ordering will be af-fected by any variation in stoichiometry and any differ-ence in the rate of heat loss during quench. Our sampleswere not encapsulated and were quenched directly intoliquid Nr, whereas Ishikawa's samples were contained ina quartz tube and were quenched into brine. The twocurves are very similar in form, the only difference beingthe magnitude of the intensity. Ishikawa (1958) made hismeasurements at 8000 Oe, whereas our "I" values werecalculated by extrapolating to infinite field. These differ-ences in experimental techniques can account for the dif-ference between the two sets ofcurves. The 700 and 800'C anneals show decreasing magnetization as the anneal-ing temperature is decreased, whereas the samples of Ishi-kawa show a constant intensity at the lower temperatures.This is the direct result of the thermal history of oursamples. The lower-temperature anneals have a largerdomain surface area (Table 3) and thus lower magnetiza-tion intensity. We cannot evaluate Ishikawa's samplesannealed below f because the total thermal histories arenot known.

The sample annealed at 1000 oC has a twin domainsize of over 100 pm, large enough to be visible in reflectedlight. We would expect the "I" of this sample to be thehighest of all the ordered samples because of the very lowboundary area. However, it has a "I. of 3l emu/g, whereasthe samples annealed at lower temperatures with muchhigher boundary surface areas have higher saturationmagnetizations. We believe we have quenched in somelong-range cation disorder that was present in the samplewhile it was annealed just below f . This cation disordermanifests itself by reducing the room temperature "I".

Dprpnvrrlq.arroN oF rsn R3c ro R3 TRANSTTToN

The first determination of the order-disorder transitionin ilmenite-hematite solid solutions was made by Ishi-kawa (1958). He measured the room temperature mag-

o' | | | | ^ {-r ,-0 200 400 600 800 roo0 1200 1400

Quenching Temperolure "C

Fig. 10. Intensity of magnetization measured at room tem-perature versus quenching temperature.

creases, the volume of the ordered, higher-,I" phase in-creases. Increases in the degree of order within the do-mains themselves with longer annealing times will alsocontribute to the increase in "I".

The horizontal line at a twin-boundary surface area ofapproximately 25 x 106 m2lm3 in Figures 9a and 9b sep-arates samples that exhibit a reverse rnu and samplesthat exhibit a normal rnrra. All of the Ilmro samplesquenched through the transition have a reverse rRM.However, short annealing times, even at 700 .C, are suf-ficient to reduce the boundary surface area and conse-quently destroy the self-reversing capability. The 0.1-hanneal at 700'C was found to self-reverse, but no bound-ary areas have been measured. After l-h annealing at 700'C, the samples become normal.

It is not possible to separate the contribution to thesaturation magnetization ("O due to the disordered do-main boundaries and that due to the long-range cationorder in the domains. As the cation-ordered transitiontemperature is approached from below, the degree oflong-range cation order decreases slowly and then rapidly tozero at the transition temperature, 2.. Figure l0 shows agraph of room-temperature magnetization versus thequench temperature for our experiments with Ilmro andIshikawa's experiments with Ilmu,,u, and Ilmr,,rr,. (Weactually believe his compositions are closer to Ilmu, andIlmrr, as explained in the next section; thus the values inparentheses.) For both studies, the samples plotted in Fig-ure l0 were annealeld at temperature for some period oftime and quenched. The curves for the Ilmro and Ilmu,,u,data sets show a steep drop in magnetization at the tran-sition, whereas the Ilm,,,ruy cuw€ shows a gradual dropto very low values for the high-temperature quenches.The lower-temperature quenches reflect equilibration inthe cation-ordered R3 field, and the high-temperaturequenches reflect equilibration in the cation-disorderedR3cfield. The difference between the shape of the curves isdue to the 200 "C difference in transition temperature;1000-1050'C for Ilmro and 800 'C for Ilmr,,rr,. The higher-

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE 173

TABLE 4, Exoerimental data for F3c to F3 transition

l lm l lm(mol%) 4 CC) (mol%) 7" ('C)

A

o lshikowo, 1958E Hoffmon, 1975

J Bur fon ,1982

^ This study: obove Tca This study: below Tc o*o

lsh ikawa (1958) '46.5 -650148 710s1 80056 8806 1 1 0 1 068 1150

Hoftman (1975a)

Redetermined usingLindsley's (1965) volume

vs. composition curve-650

710800880

1 0 1 01 150

o+o

,L

53

58bz.3b / . c74

R3c

\ Weslcott-Lewis

...- J{;o"t''tt'

50 60 70 80 FeTiO3

M o l % I l m e n i t e

Fig. 11. FerOr-FeTiO, phase diagram showing the fields ofR3c and R3. The filled triangles indicate the composition andcrystallization temperature of synthetic samples that containeddomain boundaries after quenching. The open triangles indicatesamples that did not contain domain boundaries after quenchingor showed coarsening of pre-existing domains. The results fromearlier studies are discussed in the text. The order-disorder tran-sition is drawn as a single line.

netization of synthetic samples that had been annealed ata constant temperature and then quenched into brine (Fig.l0). Ishikawa equated the decrease in magnetization withthe long-range order parameter, S, and placed the tran-sition temperature, 2., at the change in slope or wherethe long-range order parameter reached a minimum. Fromhis plots of magnetization versus quenching temperature,we have plotted Z. for six compositions on Figure I l.

Hoffman (1975a) determined a T. for Ilmuo at 800 'C

and Ilm' near 650 oC. Even though he determined fusing the same magnetic measurements, his values areseveral hundred degrees below those oflshikawa. Burton(1982) determined 4 for Ilm,. to lie between 700 and750 "C by observing the change in intensity ofan orderingreflection during in situ heating on a single-crystal dif-fractometer. This value is consistent with Hoffman's de-terminations but again is several hundred degrees belowthat of Ishikawa's. Both Hoffrnan and Burton's determi-nations for 7. fit on the transition bracketed by the be-havior of the transition-induced domains (Fig. I I andTable 4).

The key to the disrepancy between Ishikawa's datapoints and those of Burton, Hoffman, and this study liesin understanding how the sample compositions were de-termined. Nagata and Akimoto (1956) determined thatthe FeTiO. content of the hematite-ilmenite solid solu-

- 4 taken as change in slope on diagrams of intensity of magnetization

vs. temperature in lshikawa (1958).t Appears to be a smeared transition.+ "X: 0.51 . . . near 650'C," p. 76.$ "4 : 800 'C" and Fig. 5, P.72.

tion is related in a linear fashion to the volume of theunit cell (Fig. l2). The compositions of their syntheticsolid solutions of ilmenite-hematite were determined bychemical analysis, but the analytical totals ranged from890/o to 1000/0. Ishikawa and Akimoto (1957, p. 1085),assuming that the empirical linear composition-volumerelationship was correct, determined the chemical com-position of specimens used for the order-disorder studyonly by X-ray diffraction. Subsequently, Ishikawa (1958)also used the unit-cell volumes to determine the molefraction of FeTiOr. However, it was shown by Lindsley(1965) that the cell volumes show a negative deviationfrom the linear relationship (Fig. l2). Subsequent workby Hoffman (1975a) and Lawson (1981), where cell vol-umes were determined on samples that were also ana-lyzed by the electron microprobe, agreed with the volumeversus composition curve of Lindsley. The deviation ofthe volume from ideality is not trivial. For instance, thecorrection for Ilmru determined from the linear relation-ship to a composition determined from Lindsley's curvewould be approximately 7 rnolo/o or a corected value ofIlmu.. In Figure I I and Table 4, we have corrected Ishi-kawa's data points by determining the composition as itwould fall on Lindsley's curve. The transition curve drawnis now a reasonable fit for all the data points and bracketsfrom the four studies of the transition critical tempera-ture.

The Curie points determined by Ishikawa and Aki-moto (1957) can be treated the same way. The Curie pointtrends for both Ishikawa and Akimoto (1957) and laterdeterminations by Westcott-kwis and Parry Q97la)along with points for Ilmuo by Hoffman (1975a) and Ilm'ofrom this study are plotted in Figure I l Again, the dis-

Q)

= 5160

-650+800$

Burton (1982)54.6 700 < T" < 750

This study60 <90065 <13007 0 1 0 0 0 < 4 < 1 0 5 080 < t30090 >1300

100 >1300

>k

Hof fmon, 1975This StudyLindsley, 1965

ro l

!o

c

E

t

oEf

Fe2O" lO 20 30

174 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

Mol 7o Ilrnenile

Frg. 12. Volume of the rhombohedral unit cell of ilmenite-hematite solid solutions. This figure compares the linear volumeversus composition relationship suggested by Nagata and Aki-moto (1956) with thar determined by Lindsley (1965). Syntheticsfrom this study and those of Hoffman (1975a) lie on Lindsley'svolume-composition curve.

crepancy between the two sets ofdata can be reduced bycorrecting Ishikawa's sample compositions.

Appr,rc.l,rroNs To NATURAL FERRIAN TLMENTTESNatural ferrian ilmenites exhibiting self-reversed TRM

have been found mainly in dacitic tufts. The composi-tions range from Ilmr. to Ilmu, and have crystallizationtemperatures between 850 and 950 "C (Lawson et al.,1987). This temp€rature-composition range lies above theorder-disorder curve in Figure I l. Within this range liesthe Ilmuo synthetic sample that was crystallized at 900 "C.This sample was found to be self-reversing (Lawson, l98l)and to contain a 100-A twin domain microstructure (Fig.4a). Two natural self-reversing ferrian ilmenites have beenexamined by us in the reu, and both were found to con-tain a similar fine twin-domain microstructure. One wasan Ilm58-5e from a dacitic pumice block at Mount Shasta,California (Fig. 9, in Lawson et al., 1987), and the otherwas an Ilmrr_rn from a dacite at White River, Alaska (Ler-bekmo et al., 1975).

The similarity between the natural and synthetic fer-rian ilmenites in composition, temperature of formation,cooling history, transition-induced cation-ordered do-main microstructure, and magnetic characteristics lendssupport to the placement of the transition curve in FigureI l. The simlarity also supports the hypothesis put forthby Lawson et al. (1981) that the "x" phase is the twinboundary and is the second magnetic phase necessary for

the self-reversing mechanism. Therefore, self-reversal willbe confined only to those ferrian ilmenites that crystallizein the R3c field and cool fast enough to supress coarsen-ing of the twin domains in order to retain a high twin-boundary surface area.

The temperature of the transition and the kinetics oftwin-domain gro*.th and coarsening in natural ferrian il-menites will depend on the composition and defect pop-ulation. Solid solutions with AlrOr, CrrOr, VrOr, MgTiOr,and MnTiO, as well as deviations from cation-anion stoi-chiometry will raise or reduce the transition tempera-ture, but more important \Mill change the kinetics of do-main growth or coarsening. The presence of increaseddefect population may increase Fe-Ti diffusion rates thusincreasing twin-domain growth, but will also decreasecoarsening rates by solute-drag on the domain bound-aries. Therefore, impurities and defect densities will havecompeting roles on natural ferrian ilmenites.

There are a number of potential uses for the twin-do-main microstructure that arises from the order-disordertransition. The presence of the twin domains indicatesthat the ilmenite passed through the transition, and there-fore a minimum temperature of crystallization is knownif I is known. The size and morphology of the twin do-mains will be a function of cooling rate, provided suitablecorrections can be made for the effect of kinetics by im-purities. The absence of twin domains in suitable com-positions would indicate a maximum temperature ofcrystallization.

Concr-usroNs

l. The transition between disordered R3c and orderedR3 ilmenite-hematite has been redetermined. The tran-sition was reversed for Ilmro and lies between 1000 and1050 ,C.

2. Samples cooled through the transition contain a do-main microstructure. The domains were shown by dark-field transmission electron microscopy to be twin-relatedby a 180" rotation about an axis parallel to a.

3. A structural model of the domain boundary suggeststhat the boundaries are disordered and thus antiferro-magnetic and Fe enriched.

4. The size of the domains increases with increasingilmenite component; this is directly related to the in-crease in 4 with increasing ilmenite component. The sizeof the domains also increases with increasing quenchtemperature above Z"

5. The room-temperature saturation-magnetization in-tensity is directly related to the surface area per unit vol-ume ofthe dornain boundary and the long-range order ofthe cation-ordered domains. This is because the domainsare ferrimagnetic with a strong magnetic moment where-as the boundary is parasitically ferromagnetic with a weakmagnetic moment. The quenched products are mixturesof these two magnetic phases.

6. When the twin-boundary surface area exceeds thecritical threshold of approximately 25 x 106 m2lm3, the

domain boundaries are able to couple antiferromagneti-cally with the cation-ordered domains, causing the sam-ple to acquire a reverse thermoremanent magnetizationduring cooling from 600'C in an applied field of 0.5 Oe.Therefore, the boundaries are the "x" phase oflshikawaand Syono (1963).

7. It is only in quenched samples within the composi-tion range Ilmr,_r. that the boundary-surface area is ableto exceed the critical threshold value. The dependence ofcritical twin-boundary surface area on the compositionof the twins is unknown. Compositions more hematite-rich than Ilm,o (to the left of the tricritical point in Fig.l) will not form ordered domains because of the misci-bility gap; instead clustering will predominate rather thanordering. For compositions more ilmenite-rich than Ilmr.,the critical surface area may not be obtained because ofthe increased domain growth rate at the higher transitiontemperatures. Quenching Ilmro into liquid N, from 1300'C produces a domain microstructure an order of mag-nitude larger than the identical quench for Ilmro. Some-where between these two compositions, the domain-boundary surface area falls below the critical value ofliquid N, quench rates. We also must keep in mind thatthe Curie temperature goes below room temperaturb forIlmrr_ro. This will complicate interpretation of the criticalsurface area. If we were able to measure rnu below roomtemperature, perhaps a different value ofthe critical sur-face area for reversal would hold.

AcxNowr,BncMENTS

We would especially like to acknowledge the contribution of GingerWandless to this study She patiently and carefully prepared the fnablesynthetic samples for ion milling and rsu analysis. Many thanks toPrinceton University and Robert Hargraves for access to the Rock Mag-netics Laboratory and to Johns Hopkins University, David Veblen, andKen Livi for the use of their reu facilities. Howard Evans, Jr., USGS,kept us on the appropriate crystallographic path, and Bruce Moskowitz,University of California at Davis, kepr us straight magnetically HowardEvans, Jr , and Malcolm Ross, USGS, reviewed an early version of themanuscript.

RrrnnnNcns crrnoAllen, S.M., and Cahn, J.W. (1976) Mechanism of phase transformations

within the miscibility gap of Fe-rich Fe-Al alloys Acta Metallurgrca,24. 425437.

Amelinckx, S, and Van Landu)4, J. (1976) Contrast efects at planarinterfaces In H.-R. Wenk, Ed, Electron microscopy in mineralogy, p.68-l l2 Springer-Verlag, Amsterdam.

Burton, B P. (1982) Thermodynamic analysis of the systems CaCOr-MgCO., a-Fe,Or, and FerO,-FeTiOr Ph.D ihesis, State University ofNew York, Stony Brook, New York

Burton, B (1984) Thermodynamic analysis of the system FerOr-FeTiO..Physics and Chemistry ofMinerals, ll,132-139.

Carmichael, C.M. (1961) The magnetic properties of ilmenite-hemaritecrystals. Proceedings ofthe Royal Society ofLondon, A263, 508-530

Carmichael, I.S E (l 967) The iron-titanium oxides of salic volcanic rocksand their associated ferromagnesian silicates Contributions to Miner-alogy and Petrology, 14,36-64.

Hirsch, P.B., Howie, A., Nicholson, R.B., Pashley, D.W., and Whelan,M.J (1965) Electron microscopy of thin crystals, 549 p. Butterworths,London.

Hoffman, K.A. (1975a) Cation diffusion Drocesses and self-reversal of

175

thermoremanent magnetization in the ilmenite-hematite solid solution

series Geophysical Journal ofthe Royal Astronomical Society, 41, 65-

80.-(1975b) On the kinetics of cation ordering in hemoilmenites and

a reexamination ofthe self-reversal process. EOS, 56, 975.Huebner, J S (1 987) Use of gas mixtures at low pressure to specify oxygen

and other fugacities of fumace atmospheres In G C Ulmer and H.L.Barnes, Eds., Hydrothermal experimental techniques, p. 20-60. wiley,New York

Ishikawa, Y (1958) An order-disorder transformation phenomena in theFeTiOrFerOr solid solution series. Journal of the Physical Society ofJapan, 13, 838-837.

-(1962) Magnetic properties of ilmenite-hematite system at lowtemperature Journal ofthe Physical Society ofJapan, 17, 1835-1844.

Ishikawa, Y., and Akimoto, S. (1957) Magnetic properties of the FeTiOr-FerO, solid solution series. Journal of the Physical Society of Japan,1 2 . 1 0 8 3 - 1 0 9 8

Ishikawa, Y, and Syono, Y. (1963) Order-disorder transformation andreverse thermoremanent magnetism in the FeTiO,-FerO, system. Jour-nal of Physics and Chemistry of Solids, 24, 517-528.

Lawson, C.A (l 978) Problems in the use of evacuated, sealed silica tubesfor synthesis of single-phase compositions in the system FeO-Fe'Or-TiO, Camegie Institution of Washington Year Book 77 , 917 .

-(1981) Magnetic and microstructural properties of minerals of theilmenite-hematite solid solution series with special reference to the phe-nomenon of reverse thermoremanent magnetism, Ph.D. thesis, Prince-ton University, Princeton, New Jersey

Lawson, C.A., and Nord, G L., Jr. (1984) Remanent magnetization of a"paramagnetic" composition in the ilmenite-hematite series. Geophys-ical Research Letters, I l,197-200

Lawson, CA., Nord, G.L, Jr , Dowty, E. , and Hargraves, RB. (1981)Antiphase domains and reverse thermoremanent magnetism in ilmen-ite-hematite minerals. Science, 2 I 3, 137 2-137 4.

Lawson, C A., Nord, G.L., Jr., and Champion, D.E. (1987) Fe-Ti oxidemineralogy and the origin of normal and reverse remanenl magneti-zation in dacitic pumice blocks from Mt. Shasta, California Physics ofEarth and Planetary Interiors, 46,270-288

Lerbekmo, J F, Westgate, J.A., Smith, D.G W, and Denton, G.H. (1975)New data on the character and history of the White River volcaniceruption, Alaska. In R P. Suggate and M.M. Cresswell, Eds, Quater-nary studies. Royal Society ofNew Zealand Bullerin, 13,203-209.

Lindsley, DH. (1965) Iron-titanium oxides. Carnegie Institution ofWashington Year Book 64,144-148.

-(1976) The crystal chemistry and structure of oxide minerals asexemplified by the Fe-Ti oxides: Experimental studies of oxide min-erals. Mineralogical Society of America, Reviews in Mineralogy, 3, Ll-L88

Mclaren, A.C., and Phakey, P P. (1969) Diffraction contrast from Dau-phine twin boundaries in quartz Physica Status Solidi, 31, 723-737.

Merrill, R.T., and Gromme, C.S. (1967) Non-reproducitrle self-reversalofmagnetization in diorite Joumal ofGeophysical Research, 74,2014-2024

Nagata, T., and Akimoto, S. (1956) Magnelic properties of ferromagneticilmenites Geofisica Pura E Applicata, 34, 36-50.

Nord, G L, Jr, and Lawson, C.A. (1988) The order-disorder transitionin'FerO,-FeTiO,: Structure and migration kinetics of transformation-induced twin domain boundaries. In Phase transformations, 87, in press.Institute of Metals. London

Raymond, K.N., and Wenk, H R. (1971) Lunar ilmenite (refinement ofthe crystal structure). Contributions to Mineralogy and Petrology, 30,I 35-140

Smith, G.H., and Burge, RE. (1962) The analytical represenlation ofatomic scattering amplitudes for electrons. Acta Crystallographica, 15,I 82- l 86.

Smith, C.S., and Guttman, L (1953) Measurement of intemal boundariesin three-dimensional structures by random sectioning TransactionsAIME, Joumal of Metals, 197, 8l-87

Snow, J.D., and Heuer, A.H (1973) Slip systems in Al,O.. Journal of theAmerican Ceramic Society, 56, 153-157

Uyeda, S. (1958) Thermo-remanent magnetization as a medium ofpaleo-

NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

176 NORD AND LAWSON: ORDERING AND MAGNETIC PROPERTIES IN ILMENITE-HEMATITE

magnetism, with special reference to reverse thermo-remanent mag- -(l97lb) Thermoremanence in synthetic rhombohedral iron-tita-netism.JapaneseJournalofGeophysics,2, l-123. nrumoxrdes.AustralianJournalofPhysics,24,'135-742,

Wandless, M.-V, and Nord, G.L Jr., (1986) Sample preparation tech-niques for transmission electron microscopy of geologic materials. U.S.Geological Survey Open-File Report 86-255, 20 p.

Westcott.Lewis, M.F., and Parry, L.G. (1971a) Magnetism in rhombo-hedral iron-titanium oxides Australian Joumal of Physics, 24, 719- Mnxr.rscnrgr REcEn'ED Jury 25, 1988734. Mer.ruscnrpr AccEprED SEsrsvssr 3.0, 1988