American Mineralogist, Volume 77, pages 545-557, 1992

Conversionof to anataseand TiO2 (B): A TEM study and the use of fundamentalbuilding blocks for understandingrelationships among the TiO,

Jrr.r.r.lN F. BlNrrnr,Drt Dlvro R. VBnr,rN Department of Earth and Planetary Sciences,The Johns Hopkins University, Baltimore, Maryland 21218, U.S.A.

Ansrucr TiO, crystallizes in at least seven modifications encompassingseveral major structure types including those of , hollandite, and PbOr. We have adopted a fundamental building block (FBB) approach in order to understand better the relationships between these structures.The identification of common structural elementsprovides insights into possiblephase transformation mechanismsas well as the weatheringreactions that convert perovskite (CaTiOr) to TiO, anataseand TiO, (B). We have studied the weathering of perovskite from the Salitre II , Minas Gerais, BraziI, by transmission electron microscopy. Although the dissolution of perov- skite, a proposed component of the high-level nuclear waste form Synroc, has been mod- eled thermodynamically and studied experimentally, its direct natural replacementby Ti during weathering has not been previously described. Our results show that the perovskite is replaced topotactically by both anataseand TiO, (B). The reaction appears to involve an intermediate product that can be interpreted as a half-decalcified,collapsed derivative of perovskite. The second step, involving removal of the remaining Ca, can occur in one of two ways, leading either to or in rare casesto TiO, (B). The orientations of the interfaces between both perovskite and anataseand perovskite and TiO, (B) are consistent with collapse of the corner-linked Ti-O framework to form edge- sharing chains. Despite the orthorhombic structure of perovskite, the resultsshow that the anatase can develop with its c axis parallel to any of its three pseudocubic axes. The complete destruction of perovskite produces arrays primarily of anatasethat display a variety of textures, including porous aggregatesand recrystallized needles.REE elements releasedfrom perovskite are immobilized by precipitation of rhabdophane that is inti- mately intergrown with anatase.

INrnonucrroN is related to the rutile structure by a second-orderphase There are currently four naturally occurring TiO, poly- transformation (discussed by Hyde, 1987). The high- pressure polymorph, morphs [rutile, anatase,, and TiO, (B)] and at TiO, II, has the a-PbO, structure; least three (plus some very high-pressureforms) poly- the highly metastable TiO, (H) polymorph has the hol- morphs that have been produced synthetically. Details landite structure; and TiO, (B) has a structure that is for six of the distinctive polymorph structures are listed closely related to that of VO, (B). The similarities be- in Table I . It is generallyconsidered that at low pressures tween some of these polymorphs and the related struc- only rutile has a true field of stability; anataseand brook- tures of anataseand brookite will be discussedbelow. ite form metastably(Dachille et al., 1968; Post and Burn- All of the polymorphs contain edge-and corner-linked, ham, 1986).Post and Burnham (1986) used ionic mod- octahedrally coordinated Ti cations. The available struc- eling to show that the electrostatic energies of rutile, ture refinements(see asterisks in Table l) show relatively anatase,and brookite are not greatly different. Navrotsky constant Ti-O distancesand quite variable O-O distanc- and Kleppa (1967) measuredAH : -5.27 kJlmol for the es, particularly when the shared vs. unshared octahedral anataseto rutile reaction. edgesare compared.Pairs offace-sharing octahedra(con- The structureof the most commonly occurringand best taining trivalent Ti) are found only in crystallographic known TiO, phase,rutile, is sharedby a number of min- shear structuresin reduced rutile derivatives (TiOr_.). erals including pyrolusite, cassiterite,stishovite, and in a Previous discussionsofthe relationships between cer- distorted form, marcasite.The CaCl, polymorph of TiO, tain of these structures have focused on the presenceof sheetsof O atoms that approach a cubic closest-packing * Present address: Department of Geology and Geophysics, arrangement(such as in anatase),very distorted hexago- University of Wisconsin-Madison, 1215 West Dayton Street, nal closest packing that approachescubic (as in rutile; Madison,Wisconsin 53706, U.S.A. Hyde et al.,1974), or an approximately double hexagonal ooo3404x/ 9210506-0545$02.00 545 546 BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

TaBLE1. Data for some TiO2polymorphs and perovskite

Density Unit-celldata Structure (g/cm") (nm) Reference Rutile' P4"lmnm 4.13 a : 0.459,c: 0.296 Cromer and Herrington(1955) Anatase* Allamd 3.79 a : 0.379,c: 0.951 Cromer and Herrington(1955) Brookite' Pbca 3.99 a : 0.917,b : 0.546,c : 0.514 Baur(1961) Tio, (B) C2lm 3.64 a : 1.2'16,b : 0.374,c : 0.651,0 : 1O7.2V Marchand et al. (1980) Tio, il' Pbcn 4.33 a : 0.452.b : 0.550.c : 0.494 Simonsand Dachille(1967) Tio, (H) Alm 3.46 a : 1.018,c: 0.297 Latroche et al. (1989) Perovskite" Pcmn 4.03 a: 0.537.b: 0.764,c: 0.544 Kay and Bailey(1957)

* Structures refined by X-ray ditfraction. closest-packingarrangement (as in brookite). These de- tal building blocks (FBB; Moore, 1986),emphasizes the scriptions emphasizedifferences among the structuresbut similarities betweenthe structures.The aim is to provide fail to highlight their close relationships. insights into possible direct mechanismsby which poly- Andersson (1969) showed that rutile could convert to morphic phasetransformations involving TiO, might oc- the TiO, II structure by cation displacement.Simons and cur. We discuss the application of this view to under- Dachille (1970) noted that similar structural elementsare standing the TiO, (B)-anatase transformation. In the presentin anataseand TiO, II and suggestedthat a trans- second part of this paper we use the FBB description as formation between cubic closest-packedand hexagonal well as experimental results to describe perovskite closest-packedarrangements involves shear between O (CaTiOr)-TiO, weathering reactions. Iayers. Bursill and Hyde (1972) and Hyde et al. (1974) de- AN FBB APPROACHTO THE STRUCTURAL scribed a relationship betweenhollandite and rutile struc- RELATIONSHIPS BETWEEN RUTILE, ANATASE' (B), (H) tures involving rotation of adjacent rutile blocks by BRooKrrE, TiO, aNo TiO' clockwise and counterclockwise rotation. Alternatively, There are many ways to describe structures and the Bursill (1979) demonstratedthat the hollandite structure relationships between them. For example, the utility of can be generatedfrom the rutile structure by intersecting the polysomatic-seriesapproach has been discussedin antiphase boundaries or crystallographicshear. Latroche detail by Veblen (1991). Although this structuralratio- et al. (1989) reportedthat the TiO, (H) polymorph con- nalization works well for many minerals, it cannot be verted to anataserather than rutile with increasingtem- simply applied to describethe TiO' polymorphs, primar- perature and noted that the structural relationship be- ily becauseof the absenceof slabs common to all poly- tween TiO, (H) and anatasewas not clear. morphs. As noted in the introduction, both crystallo- The dissolution of perovskite and its replacement by graphic shear models and descriptions based upon the TiO, has been studied experimentally under a range of geometry of the O framework have been used to describe hydrothermal conditions (e.g.,Myhra et al., 1984; Kas- and relate some of the TiO, structures.In this discussion, trissioset al., 1986,1987; Jostsons et al., 1990).The in- we take a rather different approach. Following Moore terest is partly due to the proposal that perovskite could (1986),we describethe TiO, structuresin terms of their be a major component of the synthetic assemblageof construction from a single structural unit, or FBB, by minerals known as Synroc, a potential nuclear either direct assembly or shear. This approach reveals waste form. Thermodynamic calculations have clearly both the close relationships (some groups of TiO, poly- shown that perovskite is unstable with respect to both morphs contain common structural layers) and the dif- and rutile at low temperaturesand in natural wa- ferencesbetween members of the group (e.g.,in the stack- ters (Nesbitt et al., l98l). Kastrissioset al. (1987) sug- ing ofthese layers). gestedthat perovskite dissolvedunder hydrothermal con- FBB ditions (between ll0 and 190 "C) and that euhedral Choice of an anatase and brookite grew epitaxially from an The attempt to use an approach basedon the presence amorphousTi-O layer.Jostsons et al. (1990)summarized of common structural elements in the TiO, polymorphs much of the experimental work on perovskite dissolution was hampered by two problems. First, the largest struc- including data ofPham et al. (unpublished data), which tural block common to all of them is merely a pair of suggestthat under epithermal conditions (<80 "C) an edge-sharingoctahedra. Second, the group of TiO, struc- amorphousTi-rich layer approximately l0 nm thick forms tures falls into two categories,namely, those that contain on the perovskite surface.This layer is believed to incor- chains ofedge-sharingoctahedra in one orientation [TiO, porate Ca and act as a protective barrier that inhibits (B), anatase]and those with chains in two orientations further dissolution. [rotation parallel to the chain relatesthe two orientations, The first part ofthis paper describesthe results ofour e.g.,in rutile, brookite, TiO, (H)1. attempts to find a unified way of describing the Ti All ofthese structures except rutile can be constructed structures.The approach, basedon the use offundamen- from a unit composed of four edge-sharingoctahedra. BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

Fig.2. FBBrepresentation ofanatase. The (l0l)- and(103)- boundedslabs are indicated. The FBB is shownby thesolid dark Fig. l. Representationsof the fundamentalbuilding block. line, and the octahedraincluded in the (103)-boundedslab are Iflight-stippledoctahedra have a coordinatez : O,l (z perpen- groupedby the dashedline. dicularto the page),then dark-stippled octahedra are located at z:0.5. sharing, so that light-stippled octahedraat y : 0,1 define the cornersof the as shown. Dashedparallel lines Becausethe basic rutile structural elementscan be created isolatea (101)-boundedslab ofanatasereferred to in the by stacking ofthis unit and the rutile structure itselfcan discussion of the brookite structure, and those labeled be generatedfrom this assembly by shear, we have se- (103) separategroups of octahedra (outlined) that form lected this four-octahedron unit to be the fundamental slabs common to the TiO, (B) structure. building block. The FBB is illustrated in four orientations Figure 3 shows a representationofa slice through the in Figure l. TiO, (B) structure. The FBB is indicated by the heavy Although we will not discussit further, this FBB also solid line, and the corner-linked FBB slab common to the occurs in many structures other than the TiO, poly- anatasestructure (Fig. 2) is shown by the dashedoutline. morphs.For example,it is the sameas a four-octahedron The common structure parallel to (103) anataserational- segmentof the Ml cation chain found in the , izes that TiO, (B) is converted to anataseby shear in- and it occurs as a fragment of dioctahedral and triocta- volving growth ofledgesalong the (103)planes ofanatase hedral sheetsin the layer silicates, oxides, and hydrox- (Banfieldet al., l99l). ides. The FBB representation of the brookite structure is shown in Figure 4a. This diagram illustrates that brookite Anatase, brookite, TiO, (B), and TiO, II and their relationships to rutile There are a number of reasonsto begin this discussion with the anatasestructure. Anatase is an important low- temperature alteration product of many Ti-bearing min- erals (biotite, , perovskite), and a number of thermally induced polymorphic transformations involv- ing this have been reported. The anatase-rutile reaction that occursduring low-grademetamorphism was studied experimentally by Shannon and Pask (1964) and subsequently(with conflicting results) by Kang and Bao (1986).The conversionof both TiO, (H) (Latrocheet al., 1989)and TiO, (B) (Brohan et al.,1982) polymorphs to anatasehas been reported. The structural rationalization for the TiO, (B) to anatasetransformation is well under- stood (Brohan et al., 1982; Tournoux et al., 1986; Ban- field et al., l99l). However, an explanationfor the TiO, (H) to anatasetransformation has not been provided. A representationof a slice of the anatasestructure con- structed from the FBB (in the orientation shown on the Fig. 3. FBB representationof TiO, (B). The FBB is shown far right in Fig. l) is given in Figure 2. Consecutivelayers by the dark solid line, and the (103)-bounded anatase slab is stack directly above the layer shown, linked by corner highlighted by the dashedline. 548 BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

A B A

A A B B B B B A B (a) Brookite (c) Anatase Fig. 6. The stackingofslabs shown in Figures5a and 5c (A and B) (a) to form brookite,(b) to form TiO, II. The same stackingof slabscontaining straight rather than kinked chains producesrutile (c) to form anatase.

shear,as indicated by the arrows. When viewed down Fig. 4. (a) FBB representationofthe brookitestructure. The by throughout the structure dark line showsthe FBB, and the dashedline highlightsthe c of brookite, shear of this type page). (101)-boundedanatase slab. Arrows indicate the brookiteaxes producesinfinite edge-linkedchains (out ofthe Al- (bold)and the anatase directions (italics) in theplane ofthe slabs, ternatively, conversion ofthese chains to infinite chains (b) the FBB in the secondorientation, (c) structuralelement could involve cation hopping (every second pair along relatedto the rutile structureby shear,as indicatedby the ar- the chains in Fig. 5 to form straight, edge-sharinghori- rows. zontal chains) rather than shear.Similar cation displace- ments relating rutile and TiO, II are discussedby Simons can be constructedfrom structural slabsfound in anatase and Dachille (1970)and Hyde and Andersson(1988). (the {101}-boundedslabs). In brookite, theseslabs occur Figure 5 showsthe arrangementof chains of FBBs (Fig. in two alternating orientations, shown in Figure 4a as 5a) that form the (100) layers of brookite, the (101)- stippled and unstippled. Figure 4b shows the orientation boundedslabs ofanatase, and the (100)layers ofTiO, II. of the FBB for the unstippled slab. The first anataseslab The stacking of layers as shown in Figure 5b (A) or iden- (stippled) is oriented with an anatasea axis coming out tical layers with an orientation as shown Figure 5c (B) ofthe page,the secondwith an anatasea axis vertical. generatesthe structural arrangementsillustrated in Fig- Figure 4c showsa strip ofbrookite structure that occurs ures 6a (brookite: AABBAABB . . .), 6b (TiO, II, and by along the plane where the stippled and unstippled ana- rearrangementas discussedfor Fig. 4c, rutile: ABAB . . .), tase-type slabs are joined. This strip can be converted and 6c (anatase:BBB . . .). Thus, ignoring the polyhedral into the basic unit of the rutile structure (l l0 rutile slab) distortions, anatase,brookite, and TiO, II can be consid- ered to be simply polytypes of TiO, basedon stacking of layers illustrated in Figure 5b. A 180'rotation ofoctahedrarelates the A- and B-type layerssuggesting, as noted by Simonsand Dachille (1970), that transformations betweenstructures involve shearbe- tween O planes. A similar mechanism involving shearof the O planes parallel to the planes ofoctahedra was pos- tulated for the to clinoenstatite transformation (Coe and Kirby, 1975).Although such mechanismsin- volving two sets of shear operations or cation hopping plus shear would convert anataseor brookite to rutile, we currently have no direct evidence to indicate the ac- tual mechanismsby which the transformationsoccur, and other mechanismsmay well operate. (b) Fig. 5. (a) Assembly of an FBB to form a chain, arrange- The TiO, (H) structure and its relationship to the ment ofchains to form sheetsfound in both anataseand brookite other polyrnorphs [the (100) plane ofbrookite], (c) an FBB that forms sheetsiden- tical to those in b but rotated 180o.The numbers indicate how An FBB representation of the TiO, (H) (hollandite) the sheetin c may be superimposedonto that in b. structure is given in Figure 7. Unlike the structures de- BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B) 549

(a) 11l

Fig. 7. FBB representationof the basicunit of the TiO, (H) structure.The structurecontains two orientationsof chainscon- structedfrom the FBB highlightedby the dark lines(a) andin a secondorientation (b). scribed above, in TiO, (H) the FBBs are stackedto form infinite chains ofedge-linked octahedralthese chains oc- cur in two orientations. The TiO, (H) polymorph seems Fig. 8. (a) Diagramillustrating the conversionoftwo chain : structurally more closely related to rutile than to anatase, orientationsin TiO, (H) to one by cationhopping: [l] TiO, : : and it has been shown that electron-beam damage can (H) structure,[2] cationdisplacement; [3] resultingstruc- (b) neededto transform to result in the conversion of synthetic hollandite com- ture. Arrows indicatethe shear [3] Tio, (B). pounds to the rutile structure (Bursill, 1979). Conse- quently, it is interesting that Latroche et al. (1989) report the conversion of TiO, (H) to anatase with increasing in the precedingsection thus imply that anatase,brookite, temperature. A possible mechanism for this reaction in- TiO, il, and in a more complex way, rutile are also re- volves displacement of cations from the chains in one lated to the perovskite structure. orientation into adjacent octahedral sites (Fig. 8a), fol- usED lowed by shear as indicated in Figure 8b. This produces ExpnnrprnNTAr, METHoDS AND SAMpLES the TiO, (B) structure, which can in turn be converted to We studied naturally weathered perovskite crystals from anataseby the previously reported shear mechanism. If the Salitre II in Minas Gerais, Brazil. The this type of structural transformation occurs, we predict samples,385-P4a,385-143, and 385-179,were collected that TiO, (B) developsas an intermediate phasebetween by Anthony Mariano. Samples 385-179 and 385-143 TiO, (H) and anatase. contain lamellar-twinned perovskite embayed and veined by anatase,whereas 385-P4a is composed primarily of The structural relationships betweenperovskite and anatase,exsolved magnetite,and only minor residual pe- TiO, structures rovskite. It has been previously noted that the TiO, (B) structure Samples were examined with a Philips 420 ST trans- can be described as a derivative of the ReO. structure mission electron microscope(TEM) operatedat 120 keV. (Tournoux et al., 1986). Likewise, the perovskite struc- Sampleswere preparedby dispersingcrushed grains on a ture is closely related to the ReO, structure (Hyde and holey carbon grid and by Ar- milling specimensre- O'Keefe, 1973). In perovskite, Ca sits within a corner- moved from petrographic thin sections. Minerals were linked Ti-O framework that is identical to that found in identified by their selected-areaelectron diffraction ReO, (i.e., perovskite is a stuffed derivative of ReOr). (SAED) patterns and by their compositions. Analytical Consequently,there are closesimilarities betweenthe pe- electron microscope (AEM) data were obtained from the rovskite and TiO, (B) structures.The relationships noted thin edgesof samples using an EDAX Lidrifted Si de- BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, G)

o ooo ?"n","." o o _@tt& otobr 4ln 0.19nm I rvD o oo.P,'. o2o.nOo.r9 nt t danatase

bperovskite a"n"r"a"f

a O2ooatl o &?.s.3 O. O. @^^rlO ro-+canatase o ct%?o ooo o o A danatasel ' Goo"n oro ro

OOOO@/OOOO+canatase 010p bperovskite ooo o

Fig. 9. SAED patterns (a) down [001] ofanatase (a) and [001] ofperovskite (p; pseudocubicunit cell); (b), (c) down [010] of anataseand [00 I ] of perovskite. In the SAED pattern in b the [00 I ] direction of anataseis perpendicularto and in c parallel to the doubled axis (2x) ofperovskite (D ofthe orthorhombic cell). (d) Indexing ofa, b, c using orthorhombic perovskite cell. tector and processedwith a Princeton Gamma-Tech Sys- ExprnrprnNTAl RESULTS tem IV X-ray analyzer. Anatase powder X-ray diffraction patterns were ob- SAED patterns indicate that perovskite is largely re- tained using a Scintag automated powder diffractometer placed by anataseand that the reciprocal lattices of the with CuKa, radiation. Peak positions were corrected for minerals are oriented preferentially with respectto each instrumental and physical aberrations through the use of other. When viewed with the beam parallel to [001] of internal standard lines from Si (Standard ReferenceMa- anatase,it is apparent that the a axes ofanatase are par- terial 640b). Cell dimensions were determined using Lat- allel and approximately equal in length to the pseudo- con, a Scintag least-squareslattice parameter refinement cubic a axes of perovskite (Fig. 9a). We make reference program. to the pseudocubicaxes ofperovskite becausethis setting BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B) 551

0.381

0.380

0 379

0 378

o 377

0 376 3

sample # Fig. 10. Comparisonbetween the lengthsof the a axes(in nm) of undopedanatase: I : Cromerand Herrington(1955); 2 : JCPDSvalue; 3, 4 : valuesfor undopedsynthetic anatase determinedusing the proceduresand equipmentdescribed here (Bischoftin preparation)and for anatasereplacing perovskite, 5. The 3oerror bars for the standarddeviations from the refined a-axisdimensions are shown. Fig. I l. Low-magnificationelectron micrograph down [001] ofboth anatase(A) and perovskite(P) showinglittle evidence permits an easiercomparison with anatase.The relation- for substantialvolume changeassociated with the perovskite- ship between the pseudocubicand orthorhombic cells is anatasereactron. given by Megaw(1973) and White et al. (1985).For clar- ification, we have shown the orientation of the pseudo- cubic axes on the SAED patternsin Figures9a,9b,9c undoped anatasemay indicate cationic substitutions in- and included the orthorhombic indices on the diagrams volving Fe3*,OH, and interstitial Ca within the anatase. in Figure 9d. Although electron diffraction patterns show At low magnification with the beam parallel to [001] doubling of one axis (b of the orthorhomic perovskite of anatase (Fig. I l), anatase and perovskite were inti- structure), the c axis of anatase shows no tendency to mately intergrown and the material exhibited very low preferential development parallel or perpendicularto this porosity. At higher magnification (Fig. l2), most inter- direction (Figs. 9b, 9c). Thus, as far as the weathering facesbetween perovskite and anatasewere approximately reactions are concerned,perovskite behavesas ifit were cubic or pseudocubic. The origin of broad streaks and extra reflections in SAED patterns and the associated moir6 fringes in images will be discussedbelow. Areas clearly showing the contact betweenanatase and perovskite were rare. In general,either unaltered perov- skite or porous aggregatesof anatasewere encountered. Where contacts were viewed with the beam parallel to [010] of anataseand to a pseudocubic a axis of perov- skite, strips of anataseonly a few tens of nanometerswide were preserved adjacent to the interface. No areas of amorphous Ti oxide were detectedalong the boundaries. AEM analysesofthe anatasetypically showedthe pres- ence of 2-5o/oFeO (occasionallyup to l0o/oFeO associ- ated with goethite-anataseintergrowths), l-2o/o CaO, and < loloAlrO, and MgO. No other impurities were detected (e.g., Nb, REE). Figure l0 illustrates that the c axes of perovskite larger val- anataseafter are demonstrably than Fig. 12. High-resolution image down [001] of both anatase ues for pure anatase.The value determined for the c axis and perovskite showing the orientation ofthe interface betrween was smaller at the lo level but not significantly different these minerals. Darker areaswithin the anatase(arrowed) may at the 2o level. The larger a dimension compared with be relict strips ofperovskite. 5s2 BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

Fig. 13. (a) High-resolution image down [010] of anatase(A) and [001] of perovskite (P) showingthe orientation of the interface. Strips ofa third phasewith a lattice spacinglarger than those found in anataseare indicated (I). (b) SAED pattern. (c) Enlargement of the image in a showing the perovskite margin.

parallel to the pseudocubic a axes of perovskite and a both ion-milled and crushedgrain-mount samples)as ho- axesof anatase.Lattice fringes are not presentin the im- mogeneousareas adjacent to perovskite surfaces(Fig. la) ageof anatasebecause the spacingsare so small that they and intergrown with anatase. were excludedby the objective aperture (spacingssmaller SAED patterns from the coexisting minerals indicated than 0.22 nm were excluded). Lattice fringes associated that the [0 I I ] TiO, (B) and I I I 2] perovskite zones(pseu- with the irregular strips within anataseare probably as- docubic axes)were subparallel,and (200) of TiO, (B) and sociatedwith relict perovskite. (l l0) of perovskite (pseudocubicaxes) were inclined to Figure 13 illustrates a contact between perovskite and each other by about 20". The interface is subparallel to anataseviewed down [010] of anatase.The numerous ( I I 0) of perovskite (pseudocubicaxes) and inclined to c* moir6 fringes inclined (two orientations) to the interface of TiO, (B) by about 45". are a sample preparation artefacl (discussedbelow) and When regionscomposed entirely of anatasewere viewed should be ignored. In addition to these, there is a set of down (100), the crystalswere found almost always to fringes with a spacingof approximately 0.67 nm parallel contain planar gapsparallel to (001). Thesegaps give rise to the interfaceand (001)ofanatase. The spacingofthese to streaking in some electron diffraction patterns (Fig. fringes cannot be readily explained as a moir6 effect from l 5b). overlapping perovskite and anatase,and they appear to The SAED patterns (Fig. 9) indicate that the c axis of be developed in areas where perovskite is clearly not anatasedevelops parallel to the three pseudocubicperov- present (Fig. l3c). These fringes were not observedat all skite axes.Consequently, the weatheredperovskite should contacts and are not present along interfacesother than contain crystals of anatase in three orientations. How- those subparallelto (001) ofanatase.The zone between ever, it is difficult to establish (statistically) whether c the perovskite and all crystalline products is very narrow, anatasedevelops parallel to each ofthese axeswith equal generally < I nm wide. No amorphous Ti oxide was de- probability. tected in this region. Figure 16 illustrates an area composed of intergrown In addition to anatase,a secondTiO, mineral was iden- anatasecrystals in both [001] and [00] orientations. The tified in thesesamples. SAED patterns indicated that this SAED pattern (Fig. l6b) indicates a very slight misori- is TiO, (B) (Marchandet al., 1980),which has been re- entation betweenthe (001) and (100) planesof the two ported by Banfieldet al. (1991).TiO, (B) was present(in sets of crystals. BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B) 553

Fig. 15. (a) High-resolutionimage down [010]of anatase showingregular gaps parallel to the (002)fringes. Moir6 fringes areinclined to (002)fringes. (b) SAEDpattern. X indicatesad- ditionalreflections referred to in the text.

of crystalsshow signsof recrystallizationand elimination of very fine-scaleporosity. The anataseis extensively intergrown with Ca-bearing light rare earth element (LREE) phosphateminerals (Fig. l8). Previouswork (Mariano. 1989)has shown that the

10 nm

Fig. 14. (a) Imageshowing an areaof TiO, (B) exhibiting 0.62-nm(001) fringes. TiO, (B) adjacentto perovskitehas been destroyedby ion milling. (b) SAEDpattern from TiO, (B). Ar- row indicatesthe c* direction.

Althougl some areasof anataseviewed with the beam Fig. 16. (a) Low-magnification image showing an inter- parallel to [001] show little porosity, some aggregatesof gro\4'thof anatasecrystals oriented with their [001] directions needle-shapedanatase crystals with a high intercrystal both perpendicular (as indexed) and parallel to the beam. (b) porosity were occasionallynoted (Fig. l7). The interiors SAED Datternlrom areashown in a. 554 BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

Fig. 18. (a) Iow-magnification image showing intimately in- pattern Fig. 17. Aggregateof anataseneedles with a highintercrystal tergrown anatase(A) and rhabdophane (R). (b) SAED porosrty. from rhabdophane.

LREE pattern of the minerals is identical to that of the Drscussron perovskite, indicating that the REEs were releasedfrom Perovskite alteration the perovskite and reprecipitated without fractionation. Perovskite is a major igneous mineral in the Salitre II SAED patterns from several zone axeswere recorded from carbonatite. Some altered samplesare almost completely needlessuch as those shown in Figure 18. TheseSAED composedof TiO, minerals that pseudomorphthe perov- patterns were consistent with the identification of this skite. The coexistenceof anataseand TiO, (B) with rhab- mineral as rhabdophane.Sample 385-P4a also contains dophane,calcite, and goethite supports the currently held abundant, complexly exsolved magnetite that shows no view (Mariano, 1989) that the replacementof perovskite sign ofalteration. by Ti minerals occurred as the result of weathering. SAED patterns and images from intergrown perov- Sample preparation artefacts skite, anatase,and TiO, (B) (Figs.9, l3) clearlydemon- Most of the specimens examined in this study were strate that the TiO, minerals replacing perovskite are to- preparedfor transmission electron microscopy by Ar-ion potactically oriented. The anataseintergrowths composed milling without liquid N, cooling of the specimen. All of crystals in two or more orientations (Fig. 16) may re- diffraction patterns from the anatasesamples prepared in flect initiation ofthe anatase-formingreaction at separate this way showed extra reflections (particularly apparent sites in the perovskite. The direct overgrowth of TiO, in Figs. 9c, l3b, l5b) that could be indexed to a cubic minerals in specificorientations onto the perovskite sur- unit cell v/rth d2oo: 0.21 nm that developedin two ori- face is consistent with the possibility that the reactions entations with the ( 100) directions slightly rotated to ei- involve direct structural inheritance ofparts ofthe Ti-O ther side of the tetragonal anataseaxes. Moir6 fringes in framework of perovskite. high-resolutionimages (Figs. l3a, l5a) were pervasive, Alternative mechanisms, (l) dissolution followed by occurring in small patches in two symmetry-related ori- reprecipitation or (2) dissolution followed by formation entations over the entire sample. Dark-field images sug- of an amorphous Ti intermediate and of gestthat the phasecausing the moir6s was more abundant anatase,are not inconsistent with the topotactic relation- along the sample margins. Diffraction patterns and im- ship describedabove. However, given the extremely low agesfrom crushed grains of anatasesamples showed no solubility of Ti under weathering conditions, a mecha- evidence of extra spots or moir6 fringes. Furthermore, nism involving complete congruent dissolution of perov- specimens prepared by ion milling using a liquid-Nr- skite seemsunlikely. Furthermore, Jostsonset al. (1990) cooled stagedid not show the extra reflections or moir6 report no Ti in the solutions associatedwith hydrother- fringes. We conclude that a surface-coatingphase formed mal perovskite dissolution experiments.The perovskite- during non-cooledion milling. Basedon the sizeand shape TiO, mechanism suggestedby Kastrissios et al. (1986, of the unit cell, we tentatively suggestthat the material is 1987)and Myhra et al. (1984)emphasizes the importance a surfacecoating ofTiO. In any case,the extra spots and of Ca leachingand development of a disorderedor amor- moir6 fringes should be ignored since they are not related phous Ti-rich surfacelayer. Becausewe find no evidence to the natural sample microstructure. for an amorphous Ti phase along the interface between BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B) 555 anataseand perovskite (e.g.,Fig. l3), we suggestthat this Figure l9a shows an idealized (undistorted) represen- mechanism does not predominate during surficial weath- tation of the perovskite structure viewed down (100) enng. (pseudocubicaxes). Orthorhombic distortions from this High-resolution images indicate that the interface re- idealized cubic arrangementinvolve displacementof the gion between perovskite and anataseis very narrow (< I O atoms with minimal disruption of the cation array nm wide; Fig. l3c). If this remaining perovskite is to be (O'Keefeand Hyde, 1977;White et al., 1985). converted to anatasewithout large-scalesolution trans- The conversion of perovskite to anataseand TiO, (B) port of Ti, Ca leachingand structural reorganizationmust requires elimination of Ca, which is transported away occur along this planar region. In a zone as narrow as from the reaction site and reprecipitatedas calcite in oth- this, distinguishing among transport of components er areas. The reaction also requires conversion of some through a solution layer, surfacerestructuring, growth of corner-linked to edgeJinked pairs of octahedra. The re- an amorphous intermediate film, and direct structural in- moval of every secondplane of Ca atoms by leachingand heritance is extremely difrcult. the collapse of the Ti-O framework, as indicated by the The presenceof two metastable polymorphs in place arrows in Figure l9a, generatea hypothetical structure of the thermodynamically stable polymorph rutile sug- (Figs. l9b, l9c) with the composition Ca.rTiOrr. This geststhe importance of kinetic and structural factors in structure is closely related to shearedperovskite deriva- determining the reaction products.The formation of these tive structures in the (Na,K,Ca,Nb)TiO, system (Hyde polymorphs may be controlled by the specific structural and Andersson, 1988). The arrangementof more darkly inheritance involved in the reaction mechanisms or by stippled octahedrain Figures I 9b and I 9c is found in the the structure of the surface onto which epitactic growth TiO, (B) structure. occurs. The micrographsshown in Figures l3a and 13c illus- The growth of two TiO, minerals suggeststhat decal- trate a phase developed at the interface between anatase cification ofperovskite can occur by alternate routes with and perovskite that is characteized by a 0.67-nm spacing similar energeticrequirements. Although anataseis by far perpendicular to the interface, betweenthat of drooof an- the predominant Ti-bearing alteration product, the pres- atase(0.475 nm) and the 0.77-nm spacingequivalent to ence of TiO, (B) is of significance.TiO, (B) has been [020] perovskite (pseudocubicaxes). Because this mate- reported in low-grade metamorphic rocks (Banfield et al., rial has the characteristicspredicted for the intermediate, 1991), but it has not been recognizedpreviously as a partly decalcifredweathering product (Figs. l9b, 19c), it weathering product. may have the structure shown in Figure l9c. Images recorded with the beam parallel to [001] of an- The secondstep to form anatasefrom the intermediate atase (e.9., Fig. I l) srrggestedthat the areas occupied by phase may occur by removal of the remaining Ca and reactantand product structuresis approximatelythe same, collapse of the Ti-O framework, as indicated by the ar- as observed in this orientation. Contacts between perov- rows in Figure l9c. The experimentally observed orien- skite and anataseshowing the third dimension of the in- tations of the intergrown perovskite and anatase, the tergowths (beam parallel to ( 100) anatase)and preserv- presence of what is interpreted to be the predicted ing strips of anatasegreater than a few nanometerswide intermediate phase,the orientations of the interfacesbe- were extremely rare. In regions where perovskite had been tweenthe three phases,the seeminglysmall volume change completely destroyed, the anatase contained abundant associated with the intergrowths in two dimensions, gaps (Fig. l5), suggestingthat a substantial reduction in and the significant volume change associatedwith the volume occurs parallel to [001] of anatase. third dimension parallel to [001] of anataseare all con- Recrystallization of the anataseto form arrays of elon- sistent with the view that perovskite is converted to an- gate crystalsresults in the appearanceof very large holes. atase by leaching and structural collapse as illustrated The very significant volume change associatedwith the in Figure 19. Consequently, the reaction can be writ- perovskite-anatasereaction is particularly apparent where ten 2CaTiOr(pro")+ 2CaorTiOrnr,, + Cafrd,+ Oe) - this recrystallization has occurred. 2TiO2("n","".)+ 2Cai^[,+ 2Ol;or.The actual aqueous spe- cies are not known. The perovskite to anataseand TiO, (B) reactions The TiO, (B) structure can be generatedfrom the in- It is not possible to use textural arguments to prove termediate, partly decalcified structure (Fig. l9c) by an that transformations involved leaching of Ca and recon- alternative secondstep that involves Ca removal and col- struction of the existing Ti-O framework rather than oc- lapse of the Ti-O framework as indicated in Figures 19e curring through a transient amorphous intermediate. We and l9f. This reaction can be written 2CaTiO, (prov.)+ favor the former mechanism because(l) an amorphous 2CaorTiOrr,,",, + Cafg + Oi; - zTiO, @,+ 2Ct(^ + phase was not detected and (2) the direct mechanisms 2Offi. Consequently,we suggestthat the formation of TiO, described below, which are a consequenceof the close minerals from perovskite during weathering can be at- structural relationships between perovskite, anatase,and tributed to stepwise decalcification, with alternative di- TiO, (B), provide reasonablereaction pathways consis- rections of second-stagecollapse of the Ti-O framework tent with the orientation relationships between reactant resulting in either anataseof TiO, (B). The conversion of and products. perovskite to TiO, minerals by a mechanism involving 556 BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B)

(a) Perovskite: structural collapse eliminates the requirement for large- scaletransport of Ti in solution. For the reaction to involve inheritance of slabs of the perovskite Ti-O framework, it is necessaryto remove Ca. The presenceof abundant gaps perpendicular to the di- rections of volume reduction and the proposed structural collapsesuggest that relatively narrow slabsof perovskite react at one time. A possible mechanism involves leach- ing of Ca from the surface layer (as occurs to depths of at least 3 nm in labradorite dissolution; Hochella et al., 1988),with chargebalance maintained by bonding of two H* to two O atoms. This leaching opens pathways (via the vacant Ca sites)to the interior of the , through which Ca could be transported to the surface.Collapse as indicated by Figure l9a superimposesthe two OH groups, so that the excessstructural O can be eliminated as HrO. This could be written CaTiO. + 2H* + (s.z+ * HrTiOr; H,TiO3-TiOr+H,O. In this study we have not detectedamorphous Ti oxide or brookite in the alteration assemblage,which distin- guishesthe weatheringreaction from the reaction studied experimentally at higher temperatures(e.g., Kastrissios et al., 1987).The absenceofbrookite is not surprisingif, as we suggest,the natural weathering reactions involve a direct structural transformation. As shown in Figure 4, the brookite structure contains anatase-typeslabs in a secondorientation that cannot be generatedfrom the pe- rovskite or intermediate structure by simple collapse of the Ti-O framework.

CoNcr,usroNs We have presenteda way of viewing the TiO, struc- tures that highlights their similarities and provides clues to the mechanisms by which the polymorphic transfor- mations between them occur. Considering these struc- tures to be constructed from slabs or blocks related by shear or cation displacement simplifies the study of the perovskite to anataseand TiO, (B) weathering reactions (this study) and the TiO, (B) to anatase(Brohan et al., 1982; Banfieldet al., l99l) and rutile-TiO' II transfor- mation mechanisms(Hyde and Andersson,1988). The FBB approach to interpreting other transforma- tion mechanisms,such as brookite or anataseto rutile or TiO, (H) to anatase,cannot yet be assessed.However, we have made predictions about these reactions that can be testedwith future experimental work. If an FBB descrip- tion appropriate to all the observed transformations could be found, it would further enhanceunderstanding ofthe relationships between theseminerals. The results ofthe perovskite weathering study suggest that the reactions involve direct structural inheritance of planesofcorner-linked Ti octahedraand proceedby step-

F Fig. 19. Diagram illustrating the stepwiseconversion ofpe- rovskite (viewed down one pseudocubicaxis) to anatase(a), (b), (c), (d); and Tio, (B) (a), (b), (e), (f). BANFIELD AND VEBLEN: PEROVSKITE TO ANATASE AND TiO, (B) 55',7 wise removal of Ca and structural collapseto form edge- fect structuresin crystallinesolids. Annual Review of Material Science, sharing octahedralpairs. Although some redistribution of 4,43-92. M.G., Hart, trLP., Lumpkin, G.R., Ti by dissolution reprecipitation (e.9., Jostsons,A., Smith, K.L., Blackford, and may occur McGlinn, P., Myhra, S., Netting, A., Pham, D.K., Smart, R.SI.C., and during recrystallization), the mechanism described here Turner, P.S. (1990) Description of Synroc durability: Kinetics and eliminates the need for large-scalesolution transport of mechanismsof reaction. NERDDP Report no. 1319,261 p. ANSTO this very insoluble element.The formation of anataseand Lucas Heights Research l-aboratories, New South Wales, Australia. TiO, (B),rather than other TiO, polymorphs,emphasizes Kang,Z., and Bao, Q.X. (1986) A study of the interaction of VrO'/TiO, (anatase)by high-resolution electron-microscopy (HREM). Applied the importance of the structure of the precursor mineral Catalysis,26, 25l-263. in determining the weathering product formed. Kastrissios,T.K., Stephenson,M., Tumer, P.S., and White, T.J. (1986) Epitaxial growth ofTiO, polymorphs on perovskite.Proceedings ofthe llth International Congresson Electron Microscopy, Kyoto, 1697- AcxNowr,BocMENTS l 698. Kastrissios,T.K., Stephenson,M., and Turner, P.S.(1987) Hydrothermal We thank Anthony Mariano for providing the weatheredSalitre II car- dissolution of perovskite:Implications for Synrocformulation. Journal bonatite samplesand initial information about their .Timothy of the American Society,70, Cl44-C146. White, George Guthrie, Jeffrey Post, and Dimitri Sverjensky provided Kay, H.F., and Bailey, P.C. (1957) Structure and properties of CaTiOr. helpful comments that geatly improved this manuscript. This work was Acta Crystallographica,10, 219-226. supporredby NSF grant EAR-8903630. Latroche, M., Brohan, L., Marchand, R., and Toumoux, M. (1989) New hollandite oxides;TiO, (H) and IQouTiO'. Joumal of Solid StateChem- RnrnnrNcns crrno istry, 81, 78-82. Marchand, R., Brohan, L., and Tournoux, M. (1980) TiO, (B), a new Andersson,S. (1969) Mechanism of transformation of rutile - a-PbO:. form of dioxide and the potassiumoctatitanate K,TisOr?. Ma- A simple cation rearrang,ement.Bulletin de la Societ6 Chimique de teriafs ResearchBulletin, 15, 1129-1133. France.4.1065-1066. Mariano, A.N. (1989) Economic geologyof rare earth elements.In Min- Banfield,J.F., Veblen, D.R., and Smith, D.J. (1991)The identificationof eralogicalSociety of America Reviews in Mineralogy, 21, 309-337. naturally occurring TiO, (B) by structure determination using high- Megaw, H.D. (1973) Crystal structures:A working approach,563 p. W.B. resolution electron microscopy, image simulation, and distance-least- Saunders,Philadelphia. squaresrefinement. American Mineralogist, 76, 343-353. Moore, P.B. (1986) Quartz: Variations on a theme. American Mineralo- Baur, W.H. (1961) Atomabst?indeund bindungswinkelim brookit, TiOr. gist, 7 1, 540-546. Acta Crystallographica,14, 214-216. Myhra, S., Savage,D., Atkinson, A., and Riviere, J.C. (1984) Surface Brohan, L., Verbaere, A., and Tournoux, M. (1982) La transformation modification of some titanate minerals subjected to hydrothermal TiO, (B) - anatase.Materials ResearchBulletin, 17, 355-361. chemical attack. American Mineralogist, 69, 9O2-9O9. Bursill, L.A. (1979) Structural relationshipsbetween p-gallia, rutile, hol- Navrotsky, A., and Kleppa, O.J. (1967) Enthalpy of the anatase-rutile landite, psilomelane, ramsdellite and gallium titanite type structures. transformation. Journal ofthe American Ceramic Society, 50, 626. Acta Crystallographica,B35, 530-538. Nesbitt, H.W., Bancroft, G.M., Fyfe, W.S., Karkhanis, S.N., and Nishi- Bursill, L.A., and Hyde, B.G. (1972) Rotation faults in crystals.Nature- jima, A. (1981) Thermodynamic stability and kinetics of perovskite Physical Science, 240, 122. dissofution. Nature. 289. 358-362. Coe, R.S., and Kirby, S.H. (1975) The orthoenstatite to clinoenstatite O'Keefe, M., and Hyde, B.G. (1977) Somestructures topologically related transformation by shearing and reversal by annealing: Mechanism and to cubic perovskite (E2,), ReO3(DOJ and Cudu (LlJ. Acta Crystal- potential applications.Contributions to Mineralogy and Petrology,52, lographica.B33, 3802-38I 3. 29-55. Post, J.E., and Burnham, C.w. (1986) Ionic modeling of mineral struc- Cromer, D.T., and Herrington, K. (1955) The structuresof anataseand tures and energiesin the electrongas approximation: TiOr polymorphs, rutife. Journal ofthe American Chemical Society,77, 4708-4709. quartz, forsterite, and diopside. American Mineralogist, 7 1, 142-1 50. Dachille, F., Simons,P.Y., and Roy, R. (1968)Pressure-temperature stud- Shannon, R.D., and Pask, J.A. (1964) Topotaxy in the anatase-rutile ies of anatase,brookite, rutile, and TiO, (ID. American Mineralogist, transformation. American Mineralogist, 49, 1707 - 1717 . 53, 1929-1939. Simons, P.Y., and Dachille, F. (1967) The structure of TiO'-II, a high Hochella, M.F., Ponader,H.B., Turner, A.M., and Harris, D.W. (1988) pressurephase of TiOr. Acta Crystallographica,23, 334-335 The complexity of mineral dissolution as viewed by high resolution - (1970) Possibletopotaxy in the TiO, system.American Mineral- scanning Auger microscopy: Labradorite under hydrothermal condi- ogist,55, 403-415. tions. Geochimica et CosmochirnicaActa, 52,385-f94. Tournoux, M., Marchand, R., and Brohan, L. (1986) I:yered K,TioOt Hyde, B.G. (1987) The effectofnon-bonded, anion-anion interactionson and the open metastable TiO, (B) structure. Progress in Solid State the Caclrlrutile transformation and on the bond lengths in the rutile Chemistry,17,33-52. type. Zeitschrift fiir Kristallographie, 179, 205-213. Veblen, D.R. (1991) Polysomatismand polysomatic series:A review and Hyde, B.G., and Andersson,S. (1988) Inorganic crystal structures,430 p. applications.American Mineralogist, 76, 80l-826. Wiley Interscience,New York. White, T.J., Seeall,R.L., Barry, J.C., and Hutchison, J.L. (1985) Twin Hyde, B.G., and O'Keefe, M. (1973) Relations between the DO, (ReO) boundariesin perovskite. Acta Crystallographica,84l, 93-98. 'bronze' 'tunne[' structuretype and some and structures.Acta Crystal- lographica, 429, 243-248. M.c.NuscRrrr RrcErvEDApnrl 22, l99l Hyde, B.G., Bagshaw,A.N., O'Keefe, M., and Andersson,S. (1974) De- MeNuscnrrr AccEprEDDrcBMger 18, l99l