mineralogical society of america - mineralogy, petrology and ......created date 11/11/2015 7:55:05...

11500003–004X/97/1112–1150$05.00

American Mineralogist, Volume 82, pages 1150–1175, 1997

Structural mechanism of Co21 oxidation by the phyllomanganate buserite

ALAIN MANCEAU,1,* VICTOR A. DRITS,2 EWEN SILVESTER,1,†CÉLINE BARTOLI,1 AND BRUNO LANSON1

1Environmental Geochemistry Group, LGIT-IRIGM, University of Grenoble and CNRS, 38041 Grenoble Cedex 9, France2Geological Institute of the Russian Academy of Sciences, 7 Pyzhevsky street, 109017 Moscow, Russia

ABSTRACTThe geochemistry of Co at the Earth’s surface is closely associated with that of man-

ganese oxides. This geochemical association results from the oxidation of highly solubleCo21 to weakly soluble Co31 species, coupled with the reduction of Mn41 or Mn31 ions,initially present in the manganese oxide sorbent, to soluble Mn21. The structural mech-anism of this Co immobilization-manganese oxide dissolution reaction was investigatedat the buserite surface. Co-sorbed samples were prepared at different surface coveragesby equilibrating a Na-exchanged buserite suspension in the presence of aqueous Co21 atpH 4. The structure of Co-sorbed birnessite obtained by drying buserite samples wasdetermined by X-ray diffraction (XRD) and powder and polarized EXAFS spectroscopy.For each sample we determined the proportion of interlayer cations and layer vacancysites, the Co21/(Co21 1 Co31) ratio, the nature of Co sorption crystallographic sites, andthe proportion of interlayer vs. layer Co. From this in-depth structural characterizationtwo distinct oxidation mechanisms were identified that occur concurrently with the trans-formation of low pH monoclinic buserite to hexagonal H-rich birnessite (Drits et al.1997; Silvester et al. 1997). The first mechanism is associated with the fast dispropor-tionation of layer Mn31 according to 2Mn → Mn 1 Mlayer 1 Mn , where M31 41 21layer layer solutiondenotes a vacant site. Divalent Co sorbs above or below a vacant site (M1) and is thenoxidized by the nearest Mn . The resulting Co31 species fills the M1 position while the31layerreduced Mn migrates to the interlayer or into solution creating a new vacant site (M2).This reaction can be written: Co 1 M1 1 Mn → Co 1 M1 1 Mn →21 31 21 31solution layer interlayer layerCo 1 M1 1 Mn → Co 1 M2 1 Mn . This mechanism may replicate along31 21 31 21interlayer layer layer sol/intera Mn31-rich row, and, because the density of vacancies remains constant, it can result inrelatively high Co concentrations, as well as domains rich in Co -Mn . During the31 41layer layerlow-pH buserite transformation, about one-half of the layer Mn31 that does not dispro-portionate migrates from the layer to the interlayer space creating new vacancies, withthe displaced Mn31 residing above or below these vacancies. The second oxidation mech-anism involves the replacement of Mn by Co ; the latter may eventually migrate31 31interlayer interlayerinto layer vacancies depending on the chemical composition of octahedra surroundingthe vacancy. The criterion for the migration of Co31 into layer vacancies is the need toavoid Mn -Co -Mn sequences. The suite of chemical reactions for this second31 31 31layer layer layermechanism can be schematically written: Co 1 Mn 1 M → Mn 121 31 21solution interlayer solutionCo 1 M → Mn 1 Co , the last step being conditional. In contrast to the first31 21 31interlayer solution layermechanism, this second mechanism decreases the density of vacant sites. At high surfacecoverage, Co-sorbed birnessite contains a substantial amount of unoxidized Co spe-21interlayercies despite some non-reduced Mn31 in the sorbent. This result can be explained by thesorption of Co onto vacant sites located in Co - and Mn -rich domains devoid21 31 41solution layer layerof Mn31. The number and size of these domains increase with the extent of oxidationand the total Co concentration in the solution, and this accounts for the decreasingcapacity of buserite to oxidize Co. The weight of structural evidence indicates that Cois oxidized by Mn31 rather than Mn41. Thermodynamic considerations indicate that underthe solution pH conditions employed in this study Mn31 is the more likely electron sinkfor the oxidation of Co21. This study also shows that the high affinity of Co for man-

* [email protected]† Present address: CSIRO, Division of Minerals, Box 312,

Clayton South Victoria, Australia, 3169.

1151MANCEAU ET AL.: Co OXIDATION BY BUSERITE

ganese oxides is not only due to its oxidation to weakly soluble Co31 species, but alsobecause of the reducted layer strains from the substitution of Co31 for Mn31.Results obtained for these model compounds were compared with those for natural

Co-containing asbolane and lithiophorite (Manceau et al. 1987). This comparison indi-cates that the different structural mechanisms explored in the laboratory can satisfactorilyaccount for the observations made on natural samples. Specifically, the present studyproves that Co substitutes for Mn in natural phyllomanganates and allows us to eliminatethe possibility of precipitation of discrete CoOOH particles.

INTRODUCTIONIt has long been recognized that manganese oxides ex-

hibit strong control on the environmental distribution ofCo. Early observations by Taylor and coworkers (Taylor1968; Taylor and McKenzie 1966; Taylor et al. 1964)showed that manganese minerals lithiophorite, hollandite,and birnessite (dehydrated form of buserite) present inAustralian soils contain relatively large amounts of Co.This geochemical Mn-Co association was also recognizedin deep-sea nodules and crusts (Burns and Burns 1977).In addition, monomineralic Co-containing Mn phases,such as asbolane and lithiophorite, have been identifiedin oceanic nodules, soils, and lateritic formations (Chu-khrov et al. 1982; Chukhrov et al. 1980; Llorca 1987;Manceau et al. 1987; Ostwald 1984). Many laboratoryexperiments have been performed to understand the rea-son for this unusually high geochemical selectivity ofmanganese oxides for Co (Balistrieri and Murray 1986;Gray and Malati 1979; Loganathan and Burau 1973;McKenzie 1967; Means et al. 1978; Murray 1975). Ingeneral, sorbed Co is strongly bound and sparingly ex-tractable. Its sorption is accompanied by a release ofMn21 and a darkening of the manganese oxide. Thesefacts were attributed to an electron transfer betweensorbed Co21 and manganese oxide (McKenzie 1970;McKenzie 1980; Traina and Doner 1985). Direct evi-dence for the oxidation of Co21 to Co31 at the surface ofmanganese oxides was first reported by Murray and Dil-lard (1979) using X-ray photoelectron spectroscopy(XPS), and this technique was then used to demonstratethe trivalent state of Co in oceanic nodules (Dillard et al.1982). More recently, Manceau et al. (1987) showed byextended X-ray absorption fine structure spectroscopy(EXAFS) that Co is trivalent and in a low spin electronicconfiguration (d6) in lithiophorite and asbolane from aNew Caledonian soil. Although the oxidation state of Coin Mn-rich natural samples is now well established, themechanism of its oxidation at the atomic scale and thestructural reasons for its immobilization remain poorlyunderstood. Dillard et al. (1982) showed that Co21 is ox-idized by Mn31 or Mn41 at the surface of the phylloman-ganate birnessite and that XPS spectra of Co-sorbed com-plexes were consistent with Co31 in an oxide or hydroxideenvironment such as CoOOH or Co(OH)3. EXAFSCo-(O,OH) and Co-(Co,Mn) interatomic distances werefound to be identical to those of Mn atoms in Co-con-taining lithiophorite and asbolane. This result indicates

that Co atoms are present in a layered, phyllomanganate-type structure. The structural environment of Co and Mnatoms strongly differed, however, by their short range or-dering: The amplitudes of Co-EXAFS spectra and radialstructure functions (Co-RSF) were systematically en-hanced compared to Mn-EXAFS and Mn-RSF. The originof this difference in local order is still a mystery. Theseexperiments allowed a random distribution of Co atomsin Mn layers to be ruled out, and Co was inferred to besegregated in Co-rich domains, either dispersed in phyl-lomanganate layers or forming individualized CoOOHlayers (Manceau et al. 1987).The work presented in this paper was undertaken to

gain further insight into the crystal chemistry of Co inphyllomanganate structures. The main challenge in study-ing this system originates from the fact that Mn41 andlow-spin Co31 ions have practically the same ionic radius(0.53 Å and 0.54 Å, respectively) and atomic number,preventing them from being formally distinguished byelectron and X-ray diffraction. To reach our objective,three factors proved to be critical. First, the phylloman-ganate buserite was used as the sorbent in aqueous so-lution. This mineral is a powerful oxidant and has oftenbeen used for studying surface redox reactions (Bidoglioet al. 1993; Fendorf and Zasoski 1992; Fendorf et al.1993; Manceau and Charlet 1992; Oscarson et al. 1983;Silvester et al. 1995; Stone and Morgan 1984; Stone andUlrich 1989). The poor knowledge of the defective natureof solids belonging to the buserite and birnessite group(Strobel et al. 1987) was the principal limitation to un-derstanding these reactions at the atomic level. This lim-itation has now been overcome with the recent structuralstudies by Drits et al. (1997) and Silvester et al. (1997),who have determined the spatial distribution of lower va-lence Mn (Mn31 and Mn21) and the density and distri-bution of layer cation vacancies. Second, Co-sorbed bir-nessite samples were prepared at different surfacecoverages so that birnessites with varying Co21/Co31 ra-tios were produced. As shown later, the simultaneouspresence of Co21 and Co31 in some samples has provedvery helpful in gaining an understanding of the earlysteps in the oxidation mechanism. Third, quantitativepowder X-ray diffraction (XRD) as well as powder andpolarized EXAFS methods were used as complementarytechniques. The unique combination of these three factorsproved to be critical in solving many aspects of the crys-tal chemical properties of the Co-phyllomanganatesystem.

1152 MANCEAU ET AL.: Co OXIDATION BY BUSERITE

FIGURE 1. Idealized structures of monoclinic Na-exchangedbuserite, hexagonal H-rich birnessite, and chalcophanite, afterDrits et al. (1997), Silvester et al. (1997), Wadsley (1955), andPost and Appleman (1988). Chemical compositions given forNaBu and HBi correspond to averaged values for microcrystal-lites type I and II.

STRUCTURAL CHEMISTRY OF BUSERITE ANDBIRNESSITE

The structural study of buserite and birnessite mineralshas been a topic of sustained interest for the last ten years(Chukhrov et al. 1989; Chukhrov et al. 1985; Drits et al.1997; Kuma et al. 1994; Manceau et al. 1992; Post andVeblen 1990; Silvester et al. 1997). In this section themain structural and chemical features of these phyllo-

manganates, as determined by Drits et al. (1997) and Sil-vester et al. (1997), are summarized.Buserite and birnessite possess a layer structure formed

of edge-sharing Mn octahedra (Fig. 1). Na-exchangedbuserite, hereafter referred to as NaBu, contains two H2Olayers giving a characteristic 10 Å periodicity along thec axis. Its layer has an orthogonal unit cell with a 5 5.23and b 5 2.85 Å, and is free of vacancies (Fig. 1). Theaverage composition of NaBu is Na0.30(Mn Mn )O;2.IV III0.69 0.31Partial dehydration of 10 Å NaBu leads to the formationof the one-layer monoclinic structure of Na-exchangedbirnessite (hereafter referred to as NaBi), with sub-cellparameters a 5 5.173, b 5 2.850, c 5 7.342 Å, and b 5103.28. The average chemical composition of NaBi isNa0.30Mn (Mn Mn M0.05)O;2, where M denotes va-21 41 310.05 0.74 0.21cant structural sites. NaBi consists in fact of two types ofmicrocrystals, each having different structural formulaeand layer cell parameters. NaBi type I is characterized bythe structural formula Nax(Mn Mn )O;2 with 0.16 # x41 3112x x# 0.25, and the parameters A 5 3a 5 15.52 Å, b 5 2.85Å, and g 5 908. NaBi type II has a better constrainedstructural formula Na0.333Mn (Mn Mn M0.055)O;221 41 310.055 0.722 0.222with A 5 3a 5 15.52 Å, B 5 3b 5 8.55 Å, and g 5 908.The supercell A 5 3a arises from the ordered distributionof Mn31-rich rows parallel to [010] and separated fromeach other along [100] by two Mn41 rows. Mn31-rich rowsof NaBi type I and type II contain some Mn41 distributedregularly with a periodicity of 6b. As microcrystals oftype II are the dominant form in the NaBi samples, theaverage chemical composition of the powder is very closeto the structural formula for NaBi type II. However itshould be kept in mind that the structural formula forNaBi type I is inaccurate, and therefore the relative pro-portions of each NaBi type cannot be determined pre-cisely. As a consequence, in the rest of the paper theresults are discussed using either the average compositionof NaBi or the structural formula of NaBi type II. It isassumed that the structural formula for NaBu that servesas a matrix for the formation of NaBi type II isNa0.33(Mn Mn )O2. Again, this formula is very close41 310.67 0.33to the average composition of NaBu.H-rich birnessite (HBi), the low-pH form of birnessite,

is obtained by equilibrating an NaBu suspension underacidic conditions. Species synthesized at pH 5 4 have theunit-cell parameters a 5 2.848, c 5 7.19 Å, g 5 1208and compositionMn Mn (Mn Mn M0.167)O1.70(OH)0.30. In contrast21 31 41 310.05 0.116 0.74 0.093to NaBu and NaBi, layers of HBi have an hexagonal sym-metry and contain a considerable amount of octahedralvacancies. Interlayer cations are located either above orbelow vacant sites.The transformation of NaBu to HBi in aqueous solu-

tion is more complex and is not completely understood(Fig. 2). It is characterized by an initial rapid process inwhich part of the interlayer Na is exchanged with H1from solution and 0.1 layer Mn31 per octahedron dispro-portionate to Mn41 and Mn21 according to: Mn31 1 Mn31→ Mn41 1 Mn21 (Fig. 2), with Mn21 migrating into so-


FIGURE 2. Schematic structural representation of the low pHtransformation from monoclinic NaBu type II to hexagonal HBitype II. Half super-cell projected in the a-b plane. Interlayer oc-tahedra are drawn in black. The oxidation state of Mn is indicatedby arabic numbers.

lution. Following this initial transformation, the structuralformula of crystallites of type II can be written:Na (Mn Mn M0.055)O2.020.112x(OH)0.111x (Fig. 2b).1 41 310.3332x 0.722 0.223A slower exchange process then ensues, in which the re-mainder of the interlayer Na desorbs and, depending uponthe solution pH, Mn21 re-adsorbs above or below vacan-cies: At pH 5 5 all desorbed Mn21 is re-adsorbed, where-as at pH 5 2 no re-adsorption occurs. Associated withthis slower exchange process is the migration of aboutone-half of the remaining layer Mn31 atoms to the inter-layer space located above or below newly formed vacan-cies. At equilibrium, the disproportionation and migrationreactions result in a 50% occupancy of layer cation po-sitions along the initial Mn31-rich rows. Because theserows alternate with two successive Mn41 rows, the pro-portion of vacant sites in the structure is 1⁄6 5 0.167 andthe structural formula of HBi type II at pH 5 is Mn -210.055Mn (Mn Mn M0.167)O1.67(OH)0.33 (Fig. 2c).31 41 310.112 0.722 0.111

MATERIALS AND METHODSSample preparationNaBu was prepared following the procedure of Gio-

vanoli et al. (1970). The solid phase was washed by cen-trifugation (more than six times) until the supernatant pHwas approximately 9–10. Co sorption was achieved byadding a Co(NO3)2 solution to an NaBu suspension at pH4. All samples were prepared in an argon-gas-saturatedand constant ionic strength (0.1 mol/dm3 NaNO3) aqueousmedium at 25 8C. The NaBu was not pre-equilibrated atpH 4 and, consequently, the low pH transformation ofNaBu to HBi (Silvester et al. 1997) and the sorption ofCo occurred simultaneously. The Co concentration in so-lution was adjusted to obtain different Co/Mn molar ra-tios of 0.06 (S1), 0.14 (S2), and 0.24 (S3) on solids. Afterallowing several hours for equilibration, S1 and S3 sam-ples were filtered, rinsed, and dried, yielding Co-contain-ing birnessite (CoBi) solids. S2 was prepared as a self-supported oriented film (Silvester et al. 1997). Total Coconcentrations were determined by inductively coupledplasma (ICP) spectroscopy after dissolution of the driedsolids in hydroxylamine hydrochloride (NH3OH·Cl).

X-ray diffractionPowder X-ray diffraction patterns were recorded on a

Siemens D5000 diffractometer equipped with a Si(Li)solid-state detector. CuKa radiation was used with acounting time of 20 or 30 s per 0.048 2u step. The ab-solute precision of Bragg angles was determined with aquartz standard and was found to be better than 0.018 2uover the full angular range.Because of the high density of defects the Rietveld

approach cannot be applied, and one of the most effectiveways of revealing the fine structural and chemical fea-tures of disordered layer minerals is by simulating dif-fraction patterns from realistic structural models (Dritsand Tchoubar 1990). The program used for simulatingXRD patterns was written by A. Plançon at the Universityof Orléans. It is based on the mathematical formalism

described by Plançon (1981), Sakharov et al. (1982a,1982b), and Drits and Tchoubar (1990). Calculationswere performed for the 10l reflections, which are the mostsensitive to site occupancy. Coherent-scattering domains(CSD) in the layer plane have a disk-like shape, the meanradius of which was determined by fitting the 100 reflec-tion. The CSD size distribution along the c axis was quan-tified by two parameters, the average (NMean) and maxi-mum (NMax) number of layers (Reynolds 1989). Where


necessary, random stacking faults were introduced andtheir amount adjusted with the WR probability parameter.According to the structural model (space group P3)

described by Chukhrov et al. (1985) the unit cell of theone-layer birnessite contains one cation and two O siteswithin the layer and two cations and two H2O sites withinthe interlayer. The x and y coordinates of these sites werefixed to values corresponding to the hexagonal symmetry.Starting values for z coordinates were set to those of Chu-khrov et al. (1985) and were further refined during thefitting process.

Powder and polarized EXAFSCo and Mn K-edge EXAFS spectra were recorded at

the LURE synchrotron radiation laboratory (Orsay,France) on the EXAFS 1 station. The positron energy ofthe DCI storage ring was 1.85 GeV and the current be-tween 280 and 320 mA. The incident X-ray beam wasmonochromatized with a channel cut Si(331) crystal.X-ray absorption data for Mn and Co were recorded overthe energy ranges 6400–7400 eV and 7600–8400 eV, re-spectively. Over these energy ranges the Bragg angle var-ied from 518 to 368. Within this angular range the polar-ization rate of X-rays is almost 100% (Hazemann et al.1992). Measurements were performed in transmissionmode with the beam intensity measured by gas ionization.Ionization chambers were filled with an air-helium mix-ture, proportioned to attenuate the beam intensity by;20% before and ;50% after the samples. Special carewas taken to avoid sample thickness and heterogeneityeffects in the collection of EXAFS spectra (Manceau andGates 1997; Stern and Kim 1981). Polarized EXAFSspectra were recorded on a flat self-supported CoBi film.Angular measurements were made by turning film layersaround a rotation axis normal to both the beam directionand the electric field vector e. The angle (a) between eand the layer plane was varied from 08 to 608. The func-tion x(a 5 908) was calculated by linear regression usingthe formula x(a) 5 [x(08) 2 x(908)]cos2a 1 x(908)(Brouder 1990; Heald and Stern 1977; Manceau et al.1988). Powder EXAFS spectra were recorded at a 5 358to eliminate any texture effects originating from possiblepreferential orientation of birnessite platelets (Manceau etal. 1990).X-ray absorption spectra were normalized and the EX-

AFS spectra then Fourier-filtered according to standardprocedures (Teo 1986). A Kaiser-function window wasused in Fourier transforms to minimize the intensity ofside lobes resulting from truncation effects (Manceau andCombes 1988). Co-O, Mn-O, and Mn-Mn distances andthe number of atoms in nearest O (CNO) and Mn (CNMn)coordination shells were determined by using experimen-tal phase shift and amplitude functions derived fromCoOOH and a stoichiometric l-MnO2 reference. InCoOOH, Co31 is surrounded by six O atoms at 1.90 Åand six nearest Co atoms at 2.85 Å (Delaplane et al.1969). In l-MnO2, Mn41 is surrounded by six nearest Oatoms at 1.91 Å and six nearest Mn atoms at 2.84 Å

(Thackeray et al. 1993). In the absence of a suitable ref-erence for Co-Mn pairs, McKale’s phase shift functionswere used for calculating Co-Mn distances. The accuracyof these functions was tested by comparing EXAFS de-rived atomic distances with the crystallographic valuesfor Co-Co and Mn-Mn distances in CoOOH andl-MnO2. For these two compounds the EXAFS distancesdiffered by less than 0.01 Å from the crystallographicvalues. The absolute accuracies of interatomic distancesand number of atomic neighbors determined in this studyare estimated to 60.02 Å and 610% for the oxygen shell,60.02 Å and 20% for the nearest cation shell, and 60.04Å and 30% for the next-nearest cation shell. The relativeaccuracy between samples is better.The amplitude of polarized EXAFS spectra, and ac-

cordingly of RSF peaks, depends on the angle that theparticular atomic pair makes with the electric field vectore. This angular dependence can be written:

Nj2 j jx(k, u) 5 3 cos (u )x (k) (1)O O i iso

j i51

where j is the number of the neighboring atomic shell, iruns over all the CNj atoms of the jth shell, u is the anglejjbetween the polarization vector e and the vector r thatjjbinds the absorbing atom to the ith atom of the jth shell,and xiso is the isotropic (i.e., powder) contribution of thejth shell. For a completely random powder, there is noangular variation, and x(k,u) is reduced to:

jx(k) 5 x(k, u) 5 x (k).O isoj

It is a direct result from Equation 1 that P-EXAFS ex-periments can provide unique angular structural infor-mation and can be used to probe the in-plane andout-of-plane structure of phyllomanganates. The quanti-tative interpretation of P-EXAFS spectra is examined inthe case of an idealized phyllomanganate layer. In thelayer plane Mn-Mn1 pairs across edges are rotated by 608,so for the in-plane orientation of the electric vector e (a5 08),

6a508 2CN 5 3 cos (u ) 5 9OMn-Mn i1

i51

instead of six for a powder spectrum (Fig. 3a). Regardlessof the orientation of e within the layer plane, the apparentnumber (CNapp) of nearest Mn atoms is 1.5 times greaterthan the crystallographic number (CNcryst). It can be easilydemonstrated that this factor also holds whenever CNcryst, 6, that is in the presence of layer vacancies, providedthat there is a threefold, or higher, symmetry axis per-pendicular to the layer plane. For the out-of-plane ori-entation (e \ c*, a 5 908), the contribution of Mn-Mn1pairs is zero because u 5 908. Mn-Mn2 pairs across cor-ners have a different angular dependence (Fig. 3b). Thesix vectors connecting an Mn atom located above or be-low a vacancy to the six layer Mn2 make an identical bangle with the c* direction. Thus, for the out-of-planeorientation of the electric vector,


FIGURE 3. Possible relative orientations of a phyllomanga-nate layer to the electric field vector e. (a) projection in the a-bplane; (b) projection in the b-c plane. a is the angle between thelayer and e, b is the angle between the vector connecting theatomic pair of interest and the perpendicular to the layer, and uis the angle between the vector connecting the atomic pair ofinterest and e.

FIGURE 4. Agreement between experimental (dots) and cal-culated (solid line) XRD profiles. The comparison is for 10l re-flections (l # 3). Simulation parameters are listed in Table 3. Theinfluence of the interlayer cation site occupancy on the distri-bution of intensity for calculated XRD profiles is illustrated forsample S2. Site occupancy was varied from 0.050 to 0.065,whereas the optimum value is 0.058. Arrows indicate discrep-ancies between experimental and calculated intensity distribu-tions. Intensities are normalized to 100 for the 100 reflection.CuKa X-ray radiation.

6a5908 2 2CN 5 3 cos (u ) 5 18 cos b.OMn-Mn i2

i51

It is a particularly important result that if b 5 54.78,CN 5 6, as in the powder. The calculation ofMn-Mn2CN is less straightforward since each of the sixa 5 08Mn-Mn2atomic pair vectors makes a different u angle for a givenorientation of e in the layer plane. Manceau et al. (1990)showed that when the atomic pair under consideration(e.g., Mn-Mn2) possesses at least a threefold axis perpen-dicular to the layer plane, x(k,u) is independent of the

position of e within the layer plane. In this case, by av-eraging over the in-plane angles, it is possible to elimi-nate u, yielding the following formula:

^cos2ui& 5 cos2b sin2a 1 (sin2b cos2a)/2 (2)

The relationship between CNcryst and CNapp can then bewritten (Manceau et al. 1990):

CNapp 5 3 CNcryst [cos2b sin2a 1 (sin2b cos2a)/2] (3)

These equations are valid provided that the atomic pairunder study makes at least a threefold axis with the layerplane. In self-supporting films, the rotation of platelets inthe a-b plane results in an ` axis parallel to c* and, con-


TABLE 1. Unit-cell parameters determined for variousCo-sorbed birnessite samples

S1 S2 S3 HBi

[Co]solution(Co/Mn)solida (Å)c (Å)

10230.062.8387.180

5 3 10230.142.8407.212

2 3 10220.242.8427.215

——

2.8487.189

Note: H-rich birnessite data are from Drits et al. (1997).

TABLE 2. X-ray powder data for Co-sorbed birnessite

hkl

S1

dhkl(calc) dhkl(exp)

S2


S3


001Nsutite002100003/Nsutite101Nsutite102004103NsutiteNsutite104110111112105200

7.180

3.5902.4582.3932.325

2.0281.7951.715

1.4501.4191.3921.3201.2401.229

7.193

3.5852.4592.388 w2.328

2.0311.7951.717

1.4511.4191.3941.3201.240 vb1.229

7.212

3.6062.4602.4042.328

2.0321.8031.719

1.4541.4201.3931.3211.2441.230

7.2223.9613.6062.4602.405 w2.3302.1312.0331.8041.7221.6381.6131.4541.4201.3951.3251.246 b1.230

7.215

3.6082.461

2.329

2.0331.8041.720

1.4551.4211.3941.3221.2451.231

7.212

3.6072.4602.386 w2.330

2.0321.8051.723

1.459 vb1.4211.3981.326vb1.230

Notes: Spacings in Å. Theoretical values are calculated for the varioussets of unit-cell parameters given in Table 1.

sequently, this condition is fulfilled. It can be easily ver-ified that Equations 1 and 2 give the same result, butEquation 2 is much easier to apply as it directly connectsexperimental a values to crystallographic b angles. Ac-cordingly, from the knowledge of b it is possible to cal-culate CNapp for every a angle. A particularly importantresult is that for a 5 35.38, ^cos2ui& 5 1⁄3 and

jx(k) 5 x(k, u) 5 x (k),O isoj

regardless of the b angle. Thus the powder EXAFS spec-trum of a solid may be readily recorded from a self-sup-ported film oriented at a 5 35.38. Similarly, if an atomicpair makes a b angle of 54.78, then this is geometricallyequivalent to an a angle of 35.38, in which case CNapp 5CNcryst. The value b 5 54.78 is a magic angle for whichthe EXAFS amplitude of the considered pair is indepen-dent of the orientation of the sample in the X-ray beam.For b , 54.78, x(k,a) increases with a, whereas for b .54.78, x(k,a) decreases with increasing a. This contrastingvariation of x(k,a) with b was recently demonstrated ex-perimentally for the smectite nontronite (Manceau et al.,in preparation).

RESULTSXRDQualitative analysis. Powder XRD patterns for CoBi

samples are shown in Figure 4. These patterns are quitesimilar to those reported by Chukhrov et al. (1985) andDrits et al. (1997) for natural and synthetic one-layer hex-agonal birnessite. Note that S2 contains some nsutite (g-MnO2) impurity. The unit-cell parameters increase sys-tematically with increasing Co content (Table 1). Exper-imental and calculated dhkl values are given in Table 2.The XRD patterns contain diagnostic 102 and 103 peaks(2.03 and 1.72 Å, respectively; Fig. 4). The similar in-tensities of these peaks indicate the presence of largeamounts of interlayer cations lying above or below vacantsites (Chukhrov et al. 1985). The relative intensity andprofile shape of the hkl reflections vary with Co loading.In particular, 10l reflections of sample S3 are broader andweaker than S1 and S2 indicating a decrease in the three-dimensional (3D) ordering due to the presence of stackingfaults.Comparison of experimental and calculated pat-

terns. Experimental and calculated patterns are shown inFigure 4. The corresponding structural parameters arelisted in Table 3. All experimental XRD patterns (10l re-

flections, l # 3) are correctly reproduced by the simula-tion. The extreme sensitivity of calculated XRD profilesto site occupancy is illustrated in Figure 4. This figureillustrates the dramatic effect of a small variation of theinterlayer cation occupancy (from an optimum value of0.058, to 0.050 and 0.065, respectively) on the intensitydistribution. With these results, the estimated error in cat-ion site occupancy is less than 0.01. The fitted structuralparameters reported in Table 3 confirm that a high pro-portion of vacancies is overlaid by interlayer cations.Variations in the amount of interlayer cations, vacancies,and stacking faults as a function of Co content may bedescribed as follows.(1) The cation site occupancy within layers increases

strongly from 0.83 (Co-free HBi) to 0.89 (S1) and 0.90(S2 and S3). Associated with this increase in layer oc-cupancy is an initial decrease in the amount of interlayercations from 0.167 (Co-free HBi) to 0.11 (S1). At higherCo levels the amount of interlayer cations again increasesto 0.13 (S3). For the lower Co content (S1) the amountof interlayer cations exactly matches the amount of va-cancies. For higher Co contents, the amount of interlayercations is higher than the amount of vacancies, and it isnecessary to locate interlayer cations above and belowsome layer vacancies.(2) With increasing Co, the a parameter first decreases

from 2.848 Å (Co-free HBi) to 2.838 Å (S1). The a pa-rameter then increases with increasing Co content. Thisbehavior may be related to the evolution of the interlayeroccupancy that, as noted above, strongly decreases at firstand then increases again. Another important parameterthat may be related to the increase of the a parameterwith increasing Co content is the ratio of interlayer cat-ions located above and below vacancies relative to inter-layer cations located above or below vacancies. This ratioincreases from 0 (S1) to 0.50 (S2) and to 0.86 (S3).(3) The average cation-oxygen distances within layers


TABLE 3. Structural parameters used for the simulation of X-ray diffraction profiles

S1 S2 S3

Mean NMax NRadiusWR

8302200.23

8301700.22

9301550.40

O (position d) xyzOccupancy

0.6670.3331.021.00

0.6670.3331.021.00

0.6670.3331.021.00

Mn or Co within layer (position a) xyzOccupancy

0000.89

0000.90

0000.90

Mn or Co interlayer (position c) xyzOccupancy

002.050.06

002.050.058

002.050.065

H2O (position d) xyzOccupancy

20.66720.3333.470.30

20.66720.3333.470.25

20.66720.3333.470.27

Bond length (Å) Mn-OO-Mnint

1.931.94

1.931.94

1.931.94

Mnint-H2OH2O-Onext layer(Mn,Co)layer-(Mn,Co)interlayer

2.172.693.50

2.172.723.50

2.172.733.50

Notes: All parameters were adjusted to fit the experimental profiles (Fig. 4). The x and y coordinates are expressed in fraction of a and b parameters,respectively; z and bond lengths are expressed in Å, to emphasize the thickness of layer and interlayer octahedra. The radius of the CSDs in thea-b plane are also expressed in Å, whereas CSDs along c* (N) are given in layers. WR is the proportion of random stacking faults. Positions andoccupancies are given for the P3 space group.

are about the same for all samples (;1.93 Å). The aver-rage distances from interlayer cation to O atom (1.94 Å),on the one hand, and to H2O molecule (2.17 Å), on theother hand, are quite similar for all samples, as the av-erage thickness of interlayer octahedra is constant. Theaverage distance between layer and interlayer cations is3.50 Å (Table 3). This value is indicative of a tridentatecorner linkage, as in chalcophanite where d(Zn-Mn) 53.55 Å (Fig. 1) (Wadsley 1955).(4) The amount of stacking faults generally increases

from S1 (WR 5 0.23) and S2 (WR 5 0.22) to S3 (WR 50.40). No regular trend could be observed between theCo concentration and the amount of stacking faults.

Powder EXAFSQualitative analysis. EXAFS spectra and radial struc-

ture functions (RSFs) for the different CoBi samples areshown in Figures 5 and 6 and compared to those of Co-containing asbolane and lithiophorite (Manceau et al.1987). Mn and Co K-edge EXAFS and RSFs for S1 arevery similar to those observed for natural samples. Spe-cifically, the characteristic shape of the Co K-edge signalin the range 4–5.5 Å21 (arrows) observed for asbolaneand lithiophorite (Fig. 3 in Manceau et al. 1987) is re-produced at low Co sorption. This salient result indicatesthat the peculiar crystal chemical behavior of Co atomsfound in natural manganese oxides was reproduced in thelaboratory. The amplitude of the Co K-EXAFS signal andRSF peaks for S1 are markedly enhanced compared to

Mn K-edge results. With increasing Co concentration, twospectral modifications are noticed. First, the relative am-plitudes of Co and Mn waves are reversed (Fig. 5). Thisevolution can be clearly observed in the respective RSFswhere the intensities of the first two Co peaks drop whilethose for the Mn peaks increase (Figs. 6, 7). Second, theshoulder at 6.3 Å21 becomes progressively more intenseand appears as a resolved oscillation maximum on the S3spectrum (arrows, Fig. 5). It will be shown later that thisspectral evolution accounts for the increase of 3rd peakin the Co RSFs at increasing Co loading. The Mn RSFsfor the low Co sample (S1) and HBi compare closely withone another. At higher Co contents the amplitude of thesecond peak in the Mn RSFs increases. This trend is re-ally noteworthy as it indicates that the sorption of Comodifies the Mn-Mn contributions across edges. Statedanother way, the results show that Co readily interactswith birnessite layers, being either surface adsorbed ortrapped within Mn41 layers, and does not precipitate ascobalt (oxyhydr)oxide.Quantitative analysis: Nearest O shell. RSF peaks

were singled out and Fourier back-transformed from dis-tance space (R) to the wavevector space (k). This math-ematical operation yields a partial EXAFS spectrum,that is, the contribution to the whole EXAFS spectrumof the selected atomic shells. Co-(O,OH,OH2) andMn-(O,OH,OH2) (hereafter noted Co-O and Mn-O) con-tributions to EXAFS spectra were filtered by selectingthe first peaks in the Co and Mn RSFs, respectively.


FIGURE 5. Various k3-weighted Co and Mn K-edge EXAFSspectra for CoOOH, asbolane, lithiophorite, and Co-sorbed bir-nessite. Solid line 5 Co K edge; dotted line 5 Mn K edge exceptin the case of CoOOH (Co K edge). Asbolane and lithiophoritespectra are from Manceau et al. (1987).←

Figure 8a shows that Co-O pairs for the Co-sorbed sam-ples do not have the same wave amplitude and frequen-cy as the reference CoOOH. Increasing the Co loadingresults in a beat node pattern at 9–10 Å21, which indi-cates the presence of multiple wave frequencies and,therefore, of several O shells around Co. Consequently,the spectral fitting of S2 and S3 requires a minimum oftwo shells. The spectra were sucessfully fitted (Q 5 0.01–0.015, Table 4) by assuming a short Co-O distance of1.91 Å typical of low-spin Co31-O bonds, and a larger of2.09 Å typical of the high spin Co21-O pair. The per-centage of Co31 in S2 and S3 was evaluated from thenumber of O atoms in each sub-shell and was found tobe 80 6 10% and 70 6 10%, respectively. As demon-strated by the strong wave beating of Figure 8a, in thisparticular system the sensitivity of EXAFS to the Co31/Cotot ratio is high as a result of the strong phase contrastbetween Co31-O and Co21-O pairs, which are separatedby 0.18 Å. In contrast to S2 and S3, S1 does not displaya marked beat pattern, and its electronic wave is almostin phase with CoOOH. S1 essentially differs from Co-OOH by a lower wave amplitude, which indicates thatCo-O distances show a greater spread in S1 than in thereference. These observations led us to suspect the pres-ence of some Co21-O pairs in this sample, but in anamount that is too low to give rise to wave beating. Con-sequently, S1 was fitted by assuming a single O shell.The spectral simulation yielded 5.5 O at 1.92 Å and aDebye-Waller factor (s) logically slightly greater than forthe other samples (Ds 5 0.03 Å vs. 0.00–0.01 Å). In thissample, the maximum amount of Co21 was estimated byindependently varying CN and s during the fit, and wasevaluated to be 10%. In conclusion, the spectral analysisof the first O shell provides compelling evidence for theoxidation of divalent Co at the birnessite surface. Ex-amination of Table 4 shows that the percentage of Co31decreases with increasing Co loading, which points to adecreasing capacity of birnessite to oxidize Co with in-creasing surface coverage.Hexagonal and monoclinic birnessite possess substan-

tial amounts of lower valency Mn (Mn21 and Mn31) cat-ions (Drits et al. 1997; Silvester et al. 1997). The multi-plicity of Mn structural environments is a source ofdisorder that tends to lower the amplitude of the wavebackscattered by nearest O atoms as previously observedat the Co K edge for S1. This disorder effect results inboth an increase in the Debye-Waller factor (s) and anapparent reduction of coordination numbers (CN), arisingfrom the high correlation between CN and s (Teo 1986).These considerations are illustrated in Table 4, which


FIGURE 6. Radial structure functions (RSF) from EXAFSmeasurements for asbolane, lithiophorite, H-rich birnessite, andCo-sorbed birnessite. Solid line 5 Co K edge; dotted and dashedline 5 Mn K edge except in the case of CoOOH (Co K edge).The K edge is indicated in parenthesis. First peaks correspond tothe O shell, second peaks to the Co,Mn edge-sharing contribu-tion, and the weak third peak near 3 Å to the Co,Mn corner-sharing contribution. RSFs were not corrected for phase shift;accordingly peaks are shifted toward shorter distances by DR ¯0.3–0.4 Å with respect to crystallographic values. Real distancesobtained from least-squares fits are reported in Tables 4 and 5.←

shows that for the Co-free hexagonal birnessite sample,CNMn-O 5 4.7, instead of the crystallographic value of six,and Ds 5 0.01 Å, instead of 0.0 in the non-disorderedl-MnO2 reference. The increase of CNMn-O from 4.2 (S1)to 5.3 (S3) with increasing Co concentration (Table 4)accounts for the enhancement of Mn-O RSF peaks notedin Figure 7a. This trend stands in strong contrast withthat observed at the Co K edge since the average struc-tural disorder about Co atoms because of their mixed va-lency (i.e., Co21 plus Co31) increases with the Co con-centration (Fig. 8b). In summary, the higher the Cocontent, the greater the valency mixture of Co atoms andthe lower the valency mixture of Mn atoms. As discussedbelow, this interpretation is in good agreement with theevolution of structural formulae at increasing Coconcentration.The mean Mn-O distance of 1.91 Å is indicative of an

overwhelming presence of tetravalent and trivalent Mnatoms. For example, in the tetravalent manganese oxidespyrolusite and ramsdellite, ^d(Mn41-O)& 5 1.89 Å, where-as in the trivalent manganese oxide groutite (a-MnOOH)the four nearest Oequatorial are located at ^1.93 Å& and thetwo distant Oapical at 2.18 Å and 2.34 Å (Glasser and In-gram 1968). In the case of mixed-valency minerals, thecontribution of the two distant Mn31-Oapical pairs is partic-ularly weak and remains undetected. Mn21-O distancesare typically 2.21 Å (Christensen 1965), but the numberof divalent Mn atoms in birnessite samples is very lowand only marginally influences mean Mn-O distances.Quantitative analysis: Nearest (Co,Mn) shells. Fou-

rier-filtered nearest Co-(Co,Mn) contributions for CoBisamples, obtained by selecting the second peak in the CoRSFs, are shown in Figure 8b. These curves, which cor-respond to edge-sharing octahedra, differ in their enve-lope amplitude but possess precisely the same phase, in-dicating that Co-(Co,Mn) distances are identical for thethree samples. Because Co21 and Co31 octahedra havedifferent sizes, the constancy in edge-sharing distanceleads us to conclude that either Co21 or Co31 ions shareedges with neighboring octahedra, but not both. Other-wise the simultaneous presence of Co21-(Co,Mn) andCo31-(Co,Mn) pairs would have shifted the wave fre-quency at increasing loading, similar to that observed forthe Co-O contribution (Fig. 8a). Continuing this line ofreasoning, given that Co in S1 is almost exclusively Co31


TABLE 4. EXAFS parameters for Co-O and Mn-O pairs

Sample K edge R (Å) CNO Ds (Å) Co31 Q

HBiS1

S2

S3

MnCoMnCo

MnCo

Mn

1.911.921.901.912.091.911.912.091.91

4.75.54.24.01.05.14.11.75.3

0.010.030.000.010.000.000.010.000.00

90–100%

80 6 10%

70 6 10%

0.010.010.030.02

0.020.02

0.03

Notes: CN is the coordination number; Ds is the difference of Debye-Waller factor between the sample and the reference. Q is the figure ofmerit for the spectral fitting: Q 5 S(k3xexp 2 k3xth)2/S(k3xexp)2. Co31 5N /(N 1 N ).31 31 21Co Co Co

TABLE 5. EXAFS parameters for Co-(Mn,Co) and Mn-(Mn,Co) pairs and comparison with ^CN& values obtained for a randomdistribution of layer and interlayer cations

Sample Atomic pair

Me-MeedgeR (Å) ^CN&edge Ds (Å)

Me-MecornerR (Å) ^CN&edge Ds (Å)

^CN&

^CN&edge ^CN&cornerHBiS1

S2

S3

Mn-MnCo-MnMn-MnCo-MnMn-MnCo-MnMn-Mn

2.862.842.862.832.862.832.86

4.14.64.53.35.22.75.2

0.000.000.010.000.010.000.01

3.493.473.483.503.483.513.48

2.02.01.72.51.33.21.0

0.030.020.020.020.020.020.01

4.44.73.35.03.05.1

1.51.22.51.12.81.0

Note: For all fits Q # 0.02.

coordination numbers (CN ) from ;4.5 to ;5.2 be-Mn-Me1cause of the filling of vacancies by Co31. The relativevariation of CN is 15%, which is significant. At the sametime, the number of Co-Me pairs (CN ) was found toCo-Me1decrease from ;4.6 to ;2.7 with increasing Co contentdespite the presence of 5–6 nearest Mn atoms aroundeach vacancy (Table 5). These two contrasting trends ap-pear to conflict and thus warrant some explanation. Theapparent loss of Co neighbors across edges results fromthe multiplicity of Co sites and the property of EXAFSto average the different local environments. In the presentsituation, Co atoms are located either at the layer surface(corner links) or within Mn layers (edge links). EXAFScoordination numbers are a weighted average of the dif-ferent crystallographic environments, so that a modifica-tion of the partitioning of Co atoms between these twosites results in an apparent loss or gain of coordinationnumber. For example, Figure 7 and Table 5 show that therelative decrease in the number of edge-sharing Co-Mnpairs (^CN& ) at increasing loading is accompanied byCoedgean increase in the number of corner-linkages (^CN& ).CocornerConsequently, the opposite trends that are observed forthe amplitudes of the edge-sharing peaks in the Mn andCo RSFs with increasing Co concentration simply indi-cate that ^CN& increases faster than ^CN& . Thus,Co Cocorner edgeboth Co and Mn K-edge measurements are consistentwith an increase in the number of layer edge-sharing pairsat increasing Co concentration.Quantitative analysis: Next-nearest (Co,Mn) shells.

Co and Mn RSFs all display a third peak at ;3.1 Å (Fig.6). This peak was repeatedly observed in various of man-ganese compounds (Friedl et al. 1997; Manceau andCharlet 1992; Manceau and Combes 1988; Silvester et al.1997) and can either correspond to a side-lobe peak, orig-inating from the limited reciprocal space integrated in theFourier transform (i.e., a truncation effect), or it may bestructural in origin. The first hypothesis can be discardedfor several reasons. First, Manceau and Combes (1988)and Manceau (1995) showed that side lobes are best min-imized by using a Kaiser window function, the side-lobeintensity being about 5% of the intensity of main struc-tural peaks. Second, as shown by Silvester et al. (1997),this peak is absent in monoclinic birnessite, which has nocorner linkages. If this peak at ;3 Å was a side lobe, itwould be observed at the bottom of the huge edge-sharingsecond RSF peak even in the absence of corner links.Third, for a given Fourier-integration range, the ratio ofintensities of the main and side-lobe peaks is constant.Thus, when main peaks in the RSF drop, their side lobesshould drop in unison. Examination of Figure 6 showsthat the intensities of the second and third RSF peaks areindependent. A fourth argument can also be presented,which is based on the comparison with inverse Fouriertransforms of the second and third RSF peaks, windowedtogether (Fig. 8c). If the third peak was a side lobe of thesecond one, the frequency of the resulting wave wouldbe unique and close to that observed in Figure 8b wherethe second RSF peak was isolated. Comparison of Figures8b and 8c demonstrates that the addition of the third peakof the RSFs in the Fourier transform produces a wavebeating between 5 and 7 Å21. This beating is apparenteven for S1, which has the weakest 3rd RSF peak, man-ifest in the form of similar amplitudes of the second andthird oscillations. Last, according to XRD data (Chukhrovet al. 1985; Drits et al. 1997), hexagonal birnessite hasinterlayer cations above or below vacant sites located at;3.4–3.5 Å from nearest layer Mn atoms. This crystal-lographic distance should yield an RSF peak at the phase-shift-uncorrected distance of ;3.1 Å. In conclusion, allthese arguments solidly support a structural origin for thethird peak in the RSFs often observed in phyllomangan-ate minerals (Manceau and Charlet 1992; Manceau andCombes 1988). In Co-sorbed samples, this peak can be


FIGURE 9. The k3-weighted Co and Mn K-edge polarizedEXAFS spectra for S2 at a angles of 08, 208, 358, 508, and 608.The 908 spectra were calculated using the method described inthe text. Note the presence of isosbestic points where, for a cer-tain value of k, all x(k,a) functions are equal.

attributed to corner-linkages between Mn-Co octahedra aswell as Mn-Mn octahedra.The chemical nature of the corner linkages may be in-

ferred from examination of Figure 8c. In fact, the effecton the Co-Me2 wave frequency of introducing the thirdpeak of the RSFs in the back Fourier transforms has aclose parallel with results obtained for the first O shellanalysis (compare Figs. 8a and 8c). In Figure 8c the in-tensity of the wave beating near 5–7 Å21 increases withCo loading and, consequently, with increasing Co21/Co31ratio. Because the increasing amount of Co21 modifiesonly the contribution of corner links (Fig. 8c), and doesnot effect the edge links (Fig. 8b), it can be deduced thatCo21 octahedra bond birnessite layers by sharing corners.From S1 to S3, the number of Co-Me2 pairs increasesfrom 2.0 to 3.2 while the interatomic distance increasesfrom 3.47 to 3.51 Å (Table 5). It will be shown later thatthis change of distance is due to a variation of the Co31/Co21 ratio in the interlayer.In conclusion, the analysis of powder EXAFS spectra

shows that Co21 bonds the layer surface exclusively bysharing corners with Mn octahedra and that Co31 is pres-ent within the phyllomanganate layer. On the basis of theevidence presented thus far it is not possible to eliminatethe possibility that Co31 also occupies corner-sharing po-sitions, in addition to edge-sharing positions in the layers.This point is important for understanding the oxidationmechanism of Co and is addressed next.

Polarized EXAFSCo and Mn P-EXAFS spectra are displayed in Figure

9. Note the extremely high signal-to-noise ratio even athigh k values. The presence of isosbestic points where,for a certain value of k, x(k,a) is independent of the ex-perimental a angle provides a stringent test for the reli-ability of the measurements. All spectra must cross pre-cisely at these specific k values, and a close examinationof Figure 9 shows that this is the case up to at least 11Å21. The k- and k3-weighted RSFs derived from the x(k,a)spectra in Figure 9 are shown in Figure 10. This figureshows that all peaks in the RSF drop in amplitude whena is increased, which indicates that b . 54.78 for allatomic pairs. This result is obvious for Mn-Mn1 pairs (b5 908) but it provides critical information about the oxi-dation state of sorbed Co atoms. It is reasonable to as-sume that b depends on the size of sorbed cations becausethe larger the ionic radii, the greater the Mn-Me distanceand, consequently, the lower the b value. b should in-crease in the following order: Mn21 (0.83 Å) , Zn21 (0.74Å) ¯ Co21 (0.745 Å) , Mn31 (0.645 Å) , Co31 (0.545Å) , Mn41 (0.53 Å). Because b 5 53.58 ¯ 54.78 in chal-cophanite (Wadsley 1955; Post and Appleman 1988), itis likely that cations smaller than Zn would have a largerb value. Thus, a b value larger than 54.78 for the CoBisample S2 points to the presence of Co31 in the interlayerbecause for Co21 the b angle is expected to be ;54.78,as in chalcophanite. However, the presence of some Co21was undoubtedly identified from the analysis of powderspectra with increasing surface loading, which indicatesthe presence of both divalent and trivalent Co above layervacancies. Additional support for a mixture of Co oxi-dation states comes from the quantitative analysis ofP-EXAFS spectra.Normalized numbers of nearest

a a508[CN /CN ](Mn,Co)-Me (Mn,Co)-Me1 1and next-nearest

a a508[CN /CN ](Mn,Co)-Me (Mn,Co)-Me2 2Mn,Co neighbors determined from least-squares fits arereported in Figure 11. Note that CN and CN area5908 a 5 908Mn-Me Co-Me1 1not zero because of the partial disorientation of particlesin the film plane. This disorientation can be quantifiedfrom the reduction of CNapp in going from a 5 08 to a 5908, amounting to ;60% at the Co K edge and ;55% atthe Mn K edge. These two values are in reasonably goodagreement and are reinforced by the parallel nature of the(CN /CN ) 5 f(a) and (CN /CN ) 5 f(a)a a508 a a508Mn-Me Mn-Me Co-Me Co-Me1 i 1 1


FIGURE 10. Co and Mn K-edge polarized RSFs for S2 at a angles of 08, 208, 358, 508, and 608 derived from the EXAFS spectrashown in Figure 12. The 908 spectra were calculated.

lines shown in Figure 11. The small difference can beattributed to some unidentified systematic errors. Themagnitude of reduction of the third peak in the RSFs(CN /CN and CN /CN ) against a de-a a508 a a508Mn-Me Mn-Me Co-Me Co-Me2 2 2 2pends on two factors: the disorientation of particles andthe b value, because b ± 908. The difference in slopebetween the CN /CN 5 f(a) and CN /a a508 aMn-Me Mn-Me Co-Me2 2 2CN 5 f(a) lines reflects the difference in b anglea508Co-Me2between the two pairs and hence provides informationabout the Mn31 and Co atoms located at the layer surface.Since the radius of Mn31 is larger than Co31, but the an-gular dependence of CN /CN less than that fora a508Co-Me Co-Me2 2CN /CN (Fig. 11), it must be concluded that botha a508Mn-Me Mn-Me2 2Co21 and Co31 are located above or below layer vacancysites. Put another way, if only Co31 were located at va-cancy sites, the angular dependence of CN /CNa a508Co-Me Co-Me2 2would be greater than CN /CN . Therefore wea a508Mn-Me Mn-Me2 2conclude that a mixture of Co21 and Co31 is adsorbed atlayer vacancies.It is concluded from this polarized EXAFS analysis

that vacant sites are exclusively filled by trivalent Co,whereas external surface sites contain both Co21 andCo31. A closer look at the EXAFS results shows that therelative amount of interlayer divalent and trivalent Covaries with the total Co concentration. Indeed, theCo-Me2 distance was shown to increase from ;3.47 to;3.51 Å with increasing Co concentration (Table 5). Thisvariation is noticeable in the Co RSFs because the thirdpeaks for S2 and S3 are shifted to longer distances rela-

tive to S1 (Fig. 7b). Associated with this peak shift is aprogressive enhancement and broadening. The first effectis simply due to the increasing number of TC-sharingpairs, whereas the latter stems from the increasing con-tribution of Co21-Me2 to the Co-Me2 peak at increased Coloading. Thus, one may deduce that, unlike S1 which hasno or very little Co21, the interlayers of S2 and S3 containboth Co21 and Co31 cations.We have seen in Figure 5 that with increased Co load-

ing, the shoulder at 6.3 Å21 is reinforced at the expenseof the main oscillation at 6.8 Å21 (see arrows). The samespectral evolution is observed for S2 with increasing aangle (Fig. 9, arrows) due to the increased weighting ofthe corner-sharing contribution relative to edge sharing.This structural interpretation is further reinforced by the^CN& /^CN& values, which increase from 0.3 for S1Co Cocorner edgeto 1.2 for S3 (Table 5). Thus, we may conclude that thefrequency that peaks at 6.3 Å21 is characteristic of corner-sharing octahedra, and the frequency that peaks at 6.8 Å21is characteristic of edge-sharing octahedra.

DISCUSSIONChemical composition for Co-sorbed birnessiteSample S1. The average chemical composition of S1

can be determined from the combination of chemicalanalysis, EXAFS, and XRD data. According to EX-AFS, almost all Co in S1 is trivalent and located pre-dominantly, if not entirely, within layers. Indirect evi-


FIGURE 11. Number of nearest and next-nearest Mn,Coneighbors as a function of cos2a for S2.

dence for the existence of Co31 atoms in vacant sitescomes from the simulation of XRD patterns, whichshowed that the density of vacant sites, CNM, decreasedfrom 0.17 in HBi (Silvester et al. 1997) to 0.11 in S1.The fraction of interlayer cations NIC was found to de-crease by the same amount, which indicates that thefilling of vacant sites occurs at the expense TC-sharingcations. Interestingly, this variation of the layer andinterlayer site occupancies matches precisely the pro-portion of Co atoms in the solid as determined bychemical analysis (Co/Mn 5 0.05). Thus, the combi-nation of chemical analysis and XRD simulations pro-vides additional support for the migration of Co31 tovacant sites. If one assumes that all divalent Co is ox-idized by trivalent Mn (this hypothesis is discussed lat-er), then the total number of trivalent cations remainsthe same as in HBi (i.e., 0.21 cations per octahedron),and the chemical composition for S1 can be written:Mn Mn Co (Mn Mn Co M0.11 )O1.71 (OH)0.29 .21 31 31 41 31 310.06 0.05 0.01 0.74 0.10 0.05Examination of this chemical formula leads to the fol-lowing remarks. (1) S1 and HBi have the same anioncomposition. Thus, interlayer Na cations of the initialNaBu are completely replaced by protons, the same asCo-free HBi (Drits et al. 1997; Silvester et al. 1997).(2) The amount of Co31 in layers (5%) corresponds tothe amount of vacant sites initially formed during thelow-pH equilibration resulting from the rapid dispro-

portionation of layer Mn31 according to 2Mn31 → Mn411 Mn21. In addition, the presence of 0.05 Co31 in lay-ers is accompanied by a decrease of the same amountof interlayer Mn31.Assuming a random mixing of Co and Mn atoms

within the layers and the interlayers it is possible tocalculate ^CN& and ^CN& from the chemicalMn,Co Mn,Coedge cornercomposition of S1. These values can be compared withthose determined from the quantitative EXAFS analy-sis. The number of atomic neighbors determined byEXAFS is a weighted average of the different environ-ments: ^CN& 5 SiWiCNi, where i refers to the differentsites of a given element, Wi is the site occupancy ofthe element, and CNi is its number of neighbors at sitei. From the chemical composition of S1 we calculate

0.05Co^CN& 5 (0.89 3 6) 5 4.4,edge 0.06

0.05 0.01Co^CN& 5 (0.12 3 6) 1 5.34 5 1.5,corner 0.06 0.06

0.84Mn^CN& 5 5.34 5 4.7, andedge 0.95

0.84 0.11Mn^CN& 5 0.72 1 5.34 5 1.2.corner 0.95 0.95

These values compare well with those determined fromthe quantitative EXAFS analysis (Table 5). It should beemphasized that this agreement simply confirms the con-sistency of this chemical formula with EXAFS data anddoes not exclude the possible existence of a particular Coand Mn ordering, which could also give a good agree-ment. The question of the distribution of Mn and Co cat-ions within birnessite layers is addressed later.Samples S2 and S3. CNIC, CNM, Co31/(Co31 1 Co21),

and Co/Mn are known from XRD, EXAFS, and chemicalanalysis. Thus, if we again assume that the total numberof trivalent cations is the same as in HBi, then their av-erage cation composition can be written in the generalform:

21 21 31 31(Mn , Co ) (Mn , Co ) -20.05CN 2CN CNIC M10.05 M41 31 31[Mn (Mn , Co ) M ] (4)0.74 0.26-CN CNM M

The value 0.05 arises from the initial disproportionationof 0.1 Mn31 per octahedron. For S2, CNIC 5 0.12, CNM5 0.10, Co31/(Co31 1 Co21) 5 0.8, and Co/Mn 5 0.14,and formula 4 can be written as (Mn21,Co21)0.07(Mn31,Co31)0.05[Mn (Mn31,Co31)0.16M0.10]O1.73(OH)0.27 or Mn -41 210.74 0.072xCo Mn Co (Mn Mn Co M0.10)O1.73(OH)0.27 be-21 31 31 41 31 31x 0.052y y 0.74 0.162z zcause (y 1 z)/(x 1 y 1 z) 5 0.8 and (x 1 y 1 z) 5 0.14(1.02 2 x 2 y 2 z) we have x 5 0.25 (y 1 z); x 1 y 1z 5 0.125; (y 1 z) 5 0.100; z 5 0.100 2 y andMn Co Mn Co (Mn Mn Co M0.10)O1.73-21 21 31 31 41 31 310.04 0.03 0.052y y 0.74 0.061y 0.102y(OH)0.27 where y has been optimized to obtain a goodagreement with the number of Mn-Mn and Mn-Co edge-and corner-sharing octahedra determined by EXAFS (Ta-ble 5). The average chemical composition for S2 can be


finally expressed as: Mn Co Mn Co (Mn Mn21 21 31 31 41 310.04 0.03 0.03 0.02 0.74 0.08Co M0.10)O1.73(OH)0.27.310.08For S3, CNIC 5 0.13, CNM 5 0.10, Co31/(Co31 1 Co21)

5 0.7, and Co/Mn 5 0.24, and the set of possible chem-ical formulae is: Mn Co Mn Co (Mn Mn -21 21 31 31 41 310.02 0.06 0.052y y 0.74 0.021yCo M0.10)O1.75(OH)0.25. y 5 0.03 and y 5 0.04 values310.142ygive very similar ^CN& and ^CN& values, whichMn,Co Mn,Coedge cornerare both in fair agreement with EXAFS results consid-ering a precision of ;10%. The average chemical com-position for S3 can be written: Mn Co Mn Co -21 21 31 310.02 0.06 0.02 0.03(Mn Mn Co M0.10)O1.75(OH)0.25.41 31 310.74 0.05 0.11Comparison of the chemical formulae for HBi, S1, S2,

and S3 leads to the following remarks.(1) In spite of the small variation of CNIC (0.12–0.13),

the chemical composition of the interlayer changes mark-edly with increasing Co sorption, with Co tending31interlayerto substitute for Mn and Co for Mn . The31 21 21interlayer interlayer interlayersame trend is observed for trivalent layer cations.(2) The amount of interlayer protons decreases with

increasing Co concentration. In S1 the layer charge isbalanced equally by interlayer cations (0.30) and protons(0.29), as in the case of HBi, whereas in S3 the layercharge is mainly compensated by interlayer cations (0.31vs. 0.25). The decrease in the amount of protons is dueto the formation of pairs of TC octahedra above and be-low vacant sites, because for S2 and S3 the number ofinterlayer cations is greater than the number of vacancies(0.12–0.13 vs. 0.10). The bonding of interlayer metal ionsto undersaturated surface O atoms at vacant sites logicallyleads to the departure of protons, for charge-balancereasons.(3) The a parameter is reduced significantly in S1

(2.838 Å) compared with HBi (2.848 Å), but with in-creasing Co content it increases slightly to 2.840 Å (S2)and 2.842 Å (S3). These changes can be understood byconsidering the modifications that occur to the layer andinterlayer chemical compositions. Three scenarios can beenvisaged. First, the decrease in the density of vacanciesfrom 0.11 (S1) to 0.10 (S2, S3) increases the electrostaticrepulsion between layer cations, which could increase thea parameter for S2 and S3 compared with S1. This hy-pothesis can be rejected because the opposite trend is ob-served between HBi and S1, where a decrease in the den-sity of vacancies from 0.17 to 0.11 leads to a decrease inthe a parameter by 0.01 Å. The second alternative is thatthe replacement of Mn by the smaller Co cation31 31layer layercontributes to a decrease in a, but this interpretation isnot consistent with the increase of a in S3 compared withS1 because the former contains more Co . The third31layerexplanation is related to the amount, and nature, of inter-layer cations and their distribution around vacant sites. Itis reasonable to suppose that the presence of cationsabove and below vacant octahedra increases the lateralsize of these sites: The more interlayer cations the higherthe a parameter. For example, the strong decrease in theamount of interlayer cations in S1 comparison with HBileads to a decrease in the a parameter (2.848 vs. 2.838Å). The dependence of a on the interlayer occupancy is

more pronounced when cations are located above and be-low vacant sites. In going from S1 to S3, the percentageof vacancies that have pairs of TC-sharing cations in-creases from 10% (S1), to 20% (S2), and to 30% (S3),and this evolution is accompanied by an increase in thea parameter (2.838–2.842 Å). Additional support for thisinterpretation is provided by the structure of chalcopha-nite. The layers of chalcophanite contain only Mn41. Onein every seven layer cation sites is vacant, with pairs ofTC-sharing Zn21 octahedra located above and below (Fig.1). As a consequence, chalcophanite has a relatively largea parameter of 2.850 Å (based on an hexagonal layersymmetry) in comparison with CoBi.(4) With increased Co loading the proportion of Co21

(of total Co) increases while the proportion of Mn31 (oftotal Mn) decreases. As a consequence, at high Co load-ing the valency mixture of Co is greater, while that forMn is lower. This accounts for the observed lower inten-sity of the Co-O peak in Co RSFs and the correspondinghigher intensity of the Mn-O peak in Mn RSFs with in-creased Co concentration (Fig. 5). This trend is confirmedby the quantitative treatment of EXAFS data, with anincrease of ^CNMn-O& from 4.2 (S1) to 5.3 (S3) and thesplitting of the Co-O shell into one Co31-O and oneCo21-O sub-shell (Table 4). This agreement between EX-AFS and chemical compositions supports the hypothesisof oxidation of Co21 by Mn31.

Structural transformation of Na-exchanged buserite toCo-sorbed birnessiteStructural models for the oxidation of Co. In the

absence of Co, the low-pH weathering of NaBu stronglyalters its structure (Fig. 2). Two major transformationshave been identified (Drits et al. 1997; Silvester et al.1997). Na is exchanged by H1 and 0.1 Mn31 per octa-hedron disproportionate to form vacancies and solutionMn21, the latter of which then re-adsorb above and belowvacancies. This transformation is associated with a mi-gration of ;50% of the remaining Mn to the interlayer31layerspace, accounting for the monoclinic-to-hexagonal bir-nessite conversion. This sequence of structural transfor-mations is shown schematically in Figure 2 for micro-crystallites of type II, which possess a constant andwell-defined stoichiometry.In the present study, Co and NaBu were added simul-

taneously and then equilibrated at pH 4. Consequently,the weathering of NaBu and the sorption of Co occurredconcurrently. Analysis of the chemical formulae of theCoBi prepared in this study leads to the following re-marks. First, Co-sorbed samples and HBi have a similaranion and proton composition (1.70 , O , 1.75; 0.25 ,OH , 0.30). The H1 content is nearly equal to the amountof interlayer Na originally present in NaBu (0.30), whichmeans that the mechanism of Na1 desorption and H1 ad-sorption was not perturbed by the presence of Co, at leastat low Co concentration. Stated another way, the adsorp-tion and oxidation of Co21 at the buserite surface inter-fered with, but did not prevent, the exchange of interlayer


FIGURE 12. First mechanism (M1) for the oxidation of Co21 to Co31 at the HBi-water interface. For each step, structural unitsare projected in the a-b and b-c plane. This latter projection is cut along an Mn31-rich row. Arabic numbers correspond to theoxidation state of Mn. In step 2, an electron is transferred from Co21 to Mn31, and the reduced Mn21 leaves the layer in step 3.Steps 5 and 6 are replicates of steps 2 to 4.

Na for protons at low pH. Second, all CoBi samples con-tain large amounts of vacant sites in the layers. For S1the sum of layer Co and vacancies equals the amount ofvacancies in HBi (0.167), and for S2 and S3 this sum iseven slightly higher (0.18–0.21). This important resultsuggests that the filling of layer vacancies by Co31 doesnot modify the vacancy formation mechanism associatedwith the monoclinic buserite to the hexagonal birnessiteconversion (i.e., disproportionation and migration of layerMn31).From these general observations, two structural mech-

anisms can be envisaged for the oxidation of Co21 (Figs.12 and 13). In the first mechanism (M1), Co sorbs21solutiononto vacant sites created as a result of the fast Mn31 dis-proportionation (steps 1–2, Fig. 12). The sorbed Co21 isthen oxidized by the nearest layer Mn31, which is reducedto Mn21 (step 3). The newly formed interlayer Co31 mi-grates into the underlying vacant site while the adjacentMn21 moves out of the layer, into either solution or aninterlayer position (step 4). This Co is then surrounded31layerby five Mn41 and one vacancy. The oxidation processmay end at this point, but the sorption of a new Co21solutionand its further oxidation by the next Mn31 along theMn31-rich row seems feasible (steps 4–5). Thus, thismechanism may be replicated along a Mn31-rich row aslong as a sorbed Co21 has an adjacent Mn31. Ultimatelythis mechanism results in the formation of short

Mn41Co31Co31 . . . M chains, in place of the former Mn31rows.This oxidation process has several important conse-

quences that deserve to be outlined. First, since Co31layercations are only surrounded by Mn41 and Co31, the fiveto six Co-(Mn,Co) distances should be identical. Thishigh structural ordering should lower the EXAFS Debye-Waller factor and, accordingly, enhance the electronicwave amplitude associated with this nearest cationic shellcontribution. Second, because Co atoms contain moreelectrons than Mn atoms, the possible existence of Co-Copairs may contribute to a further enhancement of thewave amplitude. This mechanism therefore provides anexplanation of the notable increase in amplitude of thesecond peak in the Co RSFs relative to the equivalentpeak in the Mn RSFs that is consistently observed innatural and synthetic Co-containing phyllomanganates(Fig. 6). Third, mechanism 1 also accounts for the reduc-tion of the Co -(Mn41,Co31)layer distance in comparison31layerwith Mnlayer-(Mn31/41,Co31)layer (2.83–2.84 Å vs. 2.86 Å,Table 5) because Co is surrounded on average by near-31layerest cations smaller than Mnlayer.The second mechanism (M2), schematically drawn in

Figure 13, considers the possibility that the sorption ofsolution Co21 occurs after the structural transformation ofNaBu to HBi is complete (step 1). In this mechanism,solution Co21 is oxidized by Mn , and the resulting31interlayer


FIGURE 13. Second mechanism (M2) for the oxidation ofCo21 to Co31 at the HBi-water interface. Projection in the b-cplane at the level of an Mn31-rich row. Co21 is oxidized andreplaces an interlayer Mn31 (step 2). Its further migration to themanganate layer depends on the local environment. It is favor-able whenever Co31 is surrounded by a layer vacancy or Mn41and forbidden when it is surrounded by two Mn31.

Mn21 either enters into the solution or sorbs onto vacantsites (step 2). Three different cation ordering sequencesalong b can be created depending on the local chemicalsurroundings of Co . They are: Mn -Co -31 41 31interlayer layer interlayerMn , M-Co -Mn , or Mn -Co -Mn . As31 31 31 31 31 31layer interlayer layer layer interlayer layerfor M1, it is envisaged that Co migrates into the31interlayerunderlying vacant site in the two first situations (step 3).In the third case, however, Co31 migration is unlikely be-cause this results in the formation of Mn -Co -31 31layer layerMn sequences that are sterically unfavorable. This31layerproposition is supported by comparison with the Co-freebuserite and birnessite structures in which Mn41 and Mn31atoms do not mix at random but form Mn41-rich andMn31-rich rows along b (Fig. 1). Although some Mn41appears in the Mn31-rich rows it is not distributed ran-domly, but is instead always adjacent to a vacant site(Mn -Mn -M-Mn sequences). Because Co31 and31 41 31layer layer layerMn41 ions have identical ionic radii and a similar elec-tronic configuration (3d6 vs. 3d3, empty eg orbitals) it islikely that they behave similarly and, in particular, avoidmixing with Mn31 at the atomic scale. If Co in a31interlayerMn -M-Mn sequence were to migrate into the un-31 31layer layerderlying vacant site, it would be surrounded by four Mn41and two Mn31 atoms. This structural environment is steri-cally unfavorable due to both the different sizes of Mn41and Mn31 ions and the Jahn-Teller distortion of trivalentMn octahedra. This structural environment is contrary tothe observed enhancement of the edge-sharing Co-MnEXAFS contribution in Co21-free samples (i.e., S1, as-bolane, lithiophorite).The juxtaposition of these two mechanisms, M1 and

M2, with the chemical formulae of CoBi shows that nei-ther can individually account for the oxidation of divalentCo. Each has its own merit and limitations. For instance,the second mechanism can account for the decrease inthe amount of vacancies with increasing Co loading andfor the presence of Co and Co species, but the31 31interlayer layernumber of edge-sharing Co allowed by this process is31layerlow due to the high probability of having Mn -31layerCo -Mn sequences. This mechanism could never31 31interlayer layerallow for twice as many Co as Mn , as is observed31 31layer layerfor S3 (Co /Mn 5 0.11/0.05). On the other hand, the31 31layer layerfirst mechanism allows the accumulation of large amountsof Co but provides no explanation for the existence of31layerinterlayer Co31. In addition, this mechanism maintains aconstant density of vacant sites with increased Coloading.Structural models and oxidation mechanism of Co

on sorption samples. The two mechanistic models M1and M2, presented in the previous section, do allow anunderstanding of the mechanism of Co oxidation on thedifferent CoBi samples and enables the formation of re-alistic structural models that account for the whole set ofstructural and chemical data. To achieve these goals,models M1 and M2 must be applied to stoichiometricallywell-defined initial solids. For this reason, the discussionuses type II microcrystallites (Fig. 2). Due to the mixingof crystallites I and II in bulk samples, only average

chemical formulae are experimentally accessible, so thatthe cationic composition of Co-sorbed crystallites of typeII cannot be known precisely. Fortunately, owing to theoverwhelming predominance of type II microcrystallitesin the bulk sample, their chemical composition is closeto the bulk values. In any case, their structural formulaecan be reasonably inferred from some simple considera-tions. First, NaBu crystallites II [Na0.333(Mn Mn )O;2]41 310.667 0.333contain 2% more Mn31 than the sample average[Na0.30(Mn Mn )O;2]. Consequently, type II particles41 310.69 0.31should be more reactive and should be enriched inMn , Co , and Mn in comparison with type I31 31 21layer layer interlayerparticles and the average chemical composition. Second,because Co21 is oxidized by Mn31, the total amount oftrivalent cations in type II CoBi samples should be thesame as in type II HBi crystallites (0.22, Fig. 2). On thisbasis, the chemical formulae of the three type II CoBiare:

21 31 31S1 : Mn Mn Co -II 0.07 0.04 0.0141 31 31(Mn Mn Co M )O (OH)0.72 0.11 0.06 0.11 1.68 0.3221 21 31 31S2 : Mn Co Mn Co -II 0.05 0.03 0.02 0.0241 31 31(Mn Mn Co M )O (OH)0.72 0.09 0.09 0.10 1.70 0.3021 21 31 31S3 : Mn Co Mn Co -II 0.03 0.06 0.01 0.0341 31 31(Mn Mn Co M )O (OH)0.72 0.06 0.12 0.10 1.72 0.28

These chemical formulae are close to those calculatedon an average basis, keeping in mind that the precisionon the cation site occupancy is estimated to be 0.01. Thisinherent uncertainty in the exact chemical composition oftype II Co-containing crystallites is acceptable for de-scribing the general oxidation pathway of Co as well as


FIGURE 14. Idealized structure of type II microcrystallites forS1, S2, and S3. Structural formulae are written on a per-octa-hedron basis and for the half super-cell. Projection in the a-bplane. Interlayer octahedra are solid, and arabic numbers corre-spond to the oxidation state of Mn atoms.

FIGURE 15. Schematic structural representation in the a-bplane of the oxidation of Co21 to Co31 on type II microcrystallitesfor S1. Seq. 1 5 Sorption of Co ; Seq. 2 5 Oxidation and21solutionmigration of Co according to mechanism M1; Seq. 3 521interlayermigration of Mn to interlayer site. Seq. 4 5 sorption of31layerMn and Co and oxidation of Co by Mn ac-21 21 21 31solution solution solution interlayercording to mechanism M2.

allowing sufficiently accurate structural models of the dif-ferent CoBi end-products.S1II. These particles contain little Co and the31interlayer

amount of Co per octahedron (6%) is equal, within31layerthe precision of the site occupancy evaluation (;0.01),to the amount of vacant sites formed during the initialdisproportionation step in the conversion of NaBu to

HBi at low pH (5.5%). Consequently, Co21 that is con-verted to Co is in all likelihood oxidized by Mn31 31layer layer(first mechanism), and Co21 that becomes Co is ox-31interlayeridized by Mn (second mechanism). A structural31interlayermodel for the half super-cell of S1II CoBi, which satisfiesall crystal chemical requirements stated previously, isdrawn in Figure 14. A structural description of the typeII NaBu to S1II transformation is shown in Figure 15.The suite of chemical reactions corresponding to thisstructural transformation are listed below. Comments


FIGURE 16. Idealized structural representation of (Mn,Co)31-rich rows in type II microcrystallites for S1 (b-c projection).

written above the arrows correspond to step numbers inFigure 12 for M1 and Figure 13 for M2, while commentsbelow the arrows refer to the sequential steps shown inFigure 15. The reactions are:

41 31 1 21Na (Mn Mn )O 1 0.33H 1 0.065Co0.333 0.667 0.333 2M1-step2

21 41 31→ Co (Mn Mn M )O (OH)0.055 0.722 0.223 0.055 1.67 0.33seq1

21 211 0.055Mn 1 0.01Co 1 0.333NaM1-step3-4

21 41 31 31→ Mn (Mn Mn Co M )O (OH)0.055 0.722 0.168 0.055 0.055 1.67 0.33seq2

21 211 0.055Mn 1 0.01Co 1 0.333NaM2-step1

21 31 41 31 31→ Mn Mn (Mn Mn Co M )O -0.055 0.055 0.722 0.113 0.055 0.11 1.67seq3

21 21(OH) 10.055Mn 10.01Co 10.333Na0.33M2-step2

21 31 31 41 31 31→ Mn Mn Co (Mn Mn Co M )-0.065 0.045 0.01 0.722 0.113 0.055 0.11seq4

21O (OH) 1 0.045Mn 1 0.333Na1.67 0.33For this structural model, ^CN& 5 4.2, ^CN& 5 2.0,Co Coedge corner^CN& 5 4.3, and ^CN& 5 1.7. These values compareMn Mnedge edgewell with those determined from the quantitative EXAFSanalysis (Table 5). However, this agreement is not verydiscriminating because similar values were obtained byassuming a random mixing of Mn and Co atoms in mi-crocrystallites I and II (^CN& 5 4.4, ^CN& 5 1.4).Co Coedge cornerThis is a consequence of the similarity of local environ-ments for Mn and Co atoms in the two models and in thetwo types of crystals. Consequently, EXAFS spectros-copy provides direct evidence for the oxidation and mi-

gration of Co into buserite layers, but has little sensitivityto the relative distribution of Mn and Co atoms in thelayers.The possibility of mechanism 1 replicating along the

Mn31-rich rows, as described in Figure 12, cannot beexcluded a priori and needs to be addressed. An ide-alized representation of the structure of (Mn,Co)31-richrows in HBi and CoBi is drawn in Figure 16. This fig-ure shows that in the case where replication of mech-anism 1 occurs, the result is both a clustering ofCo and an increase of the vacancy density. In S1II,31layerthe total number of Co plus vacant sites equals that31layerin HBi (0.17), and consideration of Figure 16 showsthat this condition is fulfilled where Co atoms are31layerisolated. We may conclude that replication of mecha-nism 1 does not occur in S1II and that Co cations are31layerdiluted in the layers. Additional support for this struc-tural interpretation is provided by EXAFS and XRD.First, if Co atoms were clustered in domains, their localstructure would be similar to that of CoOOH, which,as discussed previously, is not supported by the EX-AFS analysis. Second, the clustering of Co atoms in afew domains separated by large Co-free HBi areas isincompatible with the strong reduction of the layer pa-rameter from 2.848 to 2.838 Å.S2II. The presence of Co and Co in S2II in-31 31layer interlayer

dicates that both mechanisms 1 and 2 occurred. In con-trast with S1II, the amount of Co is greater than the31layerinitial amount of vacancies created by the rapid dispro-portionation of Mn (5.5%). This additional amount31layerof Co is either a result of the replication capacity of31layerM1 (Fig. 12) or from the migration of Co origi-31interlayernating from M2 into vacant sites (Fig. 13). In fact, itproved impossible to build acceptable structural for-


mulae for S2II or S3II based on M1 alone. Indeed, evenin the most realistic models, one is left with Co -21interlayerMn pairs or Mn -Co -Mn sequences, which31 31 31 41layer layer interlayer layerare not likely for thermodynamic and crystal chemicalreasons, respectively. In the former case interlayer Co21should be oxidized by the adjacent Mn , and in the31layerlatter it should migrate to the underlying vacant site.In addition, the assumption that M1 dominates isequivalent to assuming that the formation ofMn41Co31Co31 . . . M sequences is faster than theMn → Mn → Co → Co reaction (M2).31 31 31 31layer interlayer interlayer layerThis is hardly tenable because the result would be thereplacement of the former Mn31-rich rows by Co31-richrows, without substantial formation of Co . Con-31interlayersequently, to explain the observed cation distributionsin S2II and S3II it is necessary to conclude that mech-anisms 1 and 2 occur in parallel. A structural modelfor the half super-cell of S2II is drawn in Figure 14 anda schematic description of the type II NaBu-to-S2IItransformation presented in Figure 17. The negativelayer charge throughout this transformation is compen-sated by Co21, Mn21 and H1 in the interlayer space.The sorption of Co21 as Co is shown explicitly in21interlayerthe reaction scheme because it is the further oxidationof this species, by means of M1 or M2, which is ofprincipal interest. Mn21 and H1 are essentially ‘‘spec-tator’’ species and may be sorbed at any moment. Forthe sake of convenience, the sorption of Mn is21interlayersupposed to occur during the last step of the transfor-mation sequence. Accordingly, intermediate solidphases are artificially enriched in OH groups. The re-action scheme is:

41 31 1 21Na (Mn Mn )O 1 0.44H 1 0.14Co0.333 0.667 0.333 2

M1-step221 41 31→ Co (Mn Mn M )O (OH)0.055 0.722 0.223 0.055 1.67 0.33

seq1

21 2110.055Mn 1 0.085Co 1 0.333Na

11 0.11H

M1-step3-441 31 31→ (Mn Mn Co M )O (OH)0.722 0.168 0.055 0.055 1.56 0.44

seq2

21 211 0.11Mn 1 0.085Co 1 0.333Na

M2-step131 41 31 31→ Mn (Mn Mn Co M )O (OH)0.055 0.722 0.113 0.055 0.11 1.56 0.44

seq3

21 211 0.11Mn 1 0.085Co 1 0.333NaM1-step21M2-step2

21 31 31 41 31 31→Co Mn Co (Mn Mn Co M )-0.017 0.015 0.04 0.722 0.113 0.055 0.11seq4

21 21O (OH) 1 0.15Mn 1 0.028Co1.59 0.4111 0.333Na 1 0.03H

M1-step3-4 1M2-step331 31 41 31 31→ Mn Co (Mn Mn Co M )-0.015 0.023 0.722 0.096 0.089 0.098

seq5

21 21O (OH) 1 0.15Mn 1 0.028Co1.59 0.4111 0.333Na 1 0.03H

M2-step131 31 41 31 31→ Mn Co (Mn Mn Co M )O -0.02 0.023 0.722 0.091 0.089 0.098 1.59

seq6

21 21(OH) 1 0.15Mn 1 0.028Co0.4111 0.333Na 1 0.03H

21 21 31 31→ Mn Co Mn Co -0.05 0.028 0.02 0.023seq7

41 31 31(Mn Mn Co M )O (OH)0.722 0.091 0.089 0.098 1.69 0.3121 11 0.1Mn 1 0.333Na 1 0.13H

A detailed analysis of the structural sequence drawn inFigure 17 shows that the degree of freedom of the natureand order of the different chemical reactions, leading tothe formation of S2II, is relatively limited. This is due tothe existence of several strong constraints, which restrictthe possibilities at each step of the NaBu-to-CoBi trans-formation. Among them are (1) the structural formula ofinitial NaBu crystallites II and the distribution of Mn41layerand Mn within these crystallites, (2) the fact that the31layerstructural formula for type II CoBi crystallites obtainedat the end of the reaction must fit the average chemicalcomposition determined for the whole sample, (3) theavoidance of M-M pairs at any step of the chemical se-quence, and (4) the simultaneous presence of Co and21interlayerMn31 in the solid. Indeed, the coexistence of Co21 andMn31 in S2II may appear surprising because normally theoxidation process continues as long as Mn31 is present.Thus we conclude that Co does not sorb near a layer21interlayeror interlayer Mn31, otherwise it would be readily oxidizedaccording to mechanism 1 or 2. The only structural so-lution is to consider that Co is surrounded by21interlayer(Mn41,Co31)layer or (Mn21,Co31)interlayer. Examination of Fig-ure 17 shows that this stringent requirement is actuallyfulfilled. Consequently, one merit of this structural modelis showing that unoxidized Co21 species must be sorbedlater in the structural transformation, when the amount ofMn31 in the solid is sufficiently lowered to allow the for-mation of Co -Co -Co or Co -Co -(Co31,31 21 31 31 21layer interlayer layer layer interlayerMn21, Co21)interlayer sequences along b. The probability ofa sorbed Co21 species encountering Co -M-Co or31 31layer layerCo -M-(Co31, Mn21, Co21)interlayer sequences increases31layerwith Co loading. This avoidance of Co21-Mn31 pairs isfully consistent with the observed increase of Co21interlayerfrom ;0 (S1II) to 0.06 (S3II) with increasing Coconcentration.The fact that the number of interlayer cations is greater

than the number of vacancies points to the existence ofcation pairs across some vacant sites. In birnessite layers,vacant sites result from either the disproportionation re-action (sites 1) or from the migration of Mn31 from layer


FIGURE 17. Schematic structural representation in the b-cplane of the Co21 oxidation to Co31 on type II microcrystallitesfor S2. Seq. 1–3 5 same as in Figure 15; Seq. 4 5 sorption ofCo and partial oxidation by Mn according to mecha-21 31solution interlayernism M2. Seq. 5 5 partial migration of Cointerlayer species to layersites according to mechanisms M1 and M2. Seq. 6 5 new mi-gration of Mn to interlayer site. Seq. 7 5 sorption of divalent31layerspecies. This suite of chemical reactions obeys the followingcrystal chemical conditions: (1) The oxidation of Co and theacidic weathering of NaBu occur simultaneously; (2) SorbedCo21 is oxidized by nearest Mn31 (i.e., absence of Co -21interlayerMn pairs); and (3) avoidance of Mn -Co -Mn and M-M31 31 31 31layer layer layer layersequences. The sorption of Mn is not temporally constrained,21solutionwith the entire amount of Mn arbitrarily sorbed at the end21interlayerof the oxidation sequence.

→

to interlayer positions (site 2). Interlayer cations compen-sate the layer charge deficiency created by the absence ofMn . For electrostatic reasons it is unlikely that41layerMn sorbs on site 2 (opposite an Mn31 or Co31) be-21solutioncause this would form (Mn,Co) -M-Mn pairs31 21interlayer interlayeracross vacancies, and therefore cause an excess of charge.For this reason, it is most likely that Mn sorbs on21solutionsite 1, as would Co (Figs. 14 and 17). In Figure 14,21interlayerthese sites correspond to positions with an occupancygreater than one.S3II. S3II microcrystallites contain large amounts of

Co (0.12) and Co (0.06) as well as some31 21layer interlayerCo (0.03) and Mn (0.01). The scarcity of31 31interlayer interlayerMn in S3II compared with HBi (0.11) can be ex-31interlayerplained by its reduction to Mn21 species because of thelarge concentration of Co (2 3 1022 M, Table 1).21solutionThis sample has more interlayer cations (0.13) thanlayer vacancies (0.10), and this excess of interlayer cat-ions points to the presence of pairs of TC-sharing cat-ions across vacancies [(Mn,Co) -M-(Mn,Co) ]21 21interlayer interlayeras for S2II. Interlayer cations, together with OH groups,compensate the deficiency of layer cationic chargeoriginating from heteroionic substitutions and latticevacancies. The increased amount of interlayer speciesrelative to vacant sites is fully consistent with the low-ering of OH groups from 0.32 (S1II) to 0.28 (S3II).A realistic structural model of the layer structure for

microcrystallites type II is drawn in Figure 14, and aschematic description of the type II NaBu to S3II trans-formation is shown in Figure 18. The structural formulaobtained for these crystallites II is in good agreementwith the average chemical composition of the bulksample.

The structures of Co-rich asbolane and Co-richlithiophorite revisitedThe results obtained for the synthetic CoBi samples

shed light on the immobilization mechanism of Co innatural asbolane and lithiophorite. Figure 6 shows thatRSFs for natural samples and S1 and S2 bear a strongresemblance. Consequently, the increase in intensity of


FIGURE 18. Schematic structural representation in the b-cplane of the Co21 oxidation to Co31 on type II microcrystallitesfor S3. Seq. 1–5: same as in Figure 18; Seq. 6 5 new sorptionand oxidation of Co by Mn according to mechanism21 31solution interlayerM2. Seq. 7 5 migration of Co and sorption of divalent31interlayerspecies.←

the Co-O and Co-Mn1 RSF peaks relative to Mn-O andMn-Mn1 in natural samples is fully accounted for by thestructural Co21-to-Co31 oxidation mechanism proposedfor birnessite. This startling enhancement of Co-RSFpeaks was noted by Manceau et al. (1987), but they failedat that time to determine whether Co was precipitated asCoOOH or segregated or substituted for Mn in the phyl-lomanganate layers. The evidence supporting the latticemigration of Co31 (Fig. 14) answers this important geo-chemical question and provides an atomic-scale expla-nation of the weak extractability of Co once s

mineralogical society of america - mineralogy, petrology and ......created date 11/11/2015 7:55:05...

Documents