Density Functional Modeling of the Local Structure of Kaolinite

Phone: +61 3 8344 8755. Fax: +61 3 ... Local structure-based experimental techniques are useful in understanding the atomic structure but do not provi...
2 downloads 0 Views 4MB Size
Download PDF
4988

J. Phys. Chem. A 2010, 114, 4988–4996

Density Functional Modeling of the Local Structure of Kaolinite Subjected to Thermal Dehydroxylation Claire E. White,† John L. Provis,*,† Thomas Proffen,‡ Daniel P. Riley,§ and Jannie S. J. van Deventer† Department of Chemical & Biomolecular Engineering, UniVersity of Melbourne, Victoria 3010, Australia, Lujan Neutron Scattering Center, Los Alamos National Laboratory, New Mexico 87545, and Department of Mechanical Engineering, UniVersity of Melbourne, Victoria 3010, Australia ReceiVed: NoVember 23, 2009; ReVised Manuscript ReceiVed: January 27, 2010

Understanding the atomic-level changes that occur as kaolinite is converted (thermally dehydroxylated) to metakaolin is critical to the optimization of this large-scale industrial process. Metakaolin is X-ray amorphous; therefore, conventional crystallographic techniques do not reveal the changes in local structure during its formation. Local structure-based experimental techniques are useful in understanding the atomic structure but do not provide the thermodynamic information which is necessary to ensure plausibility of refined structures. Here, kaolinite dehydroxylation is modeled using density functional theory, and a stepwise methodology, where several water molecules are removed from the structure, geometry optimization is carried out, and then the process is repeated. Hence, the structure remains in an energetically and thermodynamically feasible state while transitioning from kaolinite to metakaolin. The structures generated during the dehydroxylation process are validated by comparison with X-ray and neutron pair distribution function data. Thus, this study illustrates one possible route by which dehydroxylation of kaolinite can take place, revealing a chemically, energetically, and experimentally plausible structure of metakaolin. This methodology of density functional modeling of the stepwise changes in a material is not limited in application to kaolinite or other aluminosilicates and provides an accurate representation of the local structural changes occurring in materials used in industrially important processes. Introduction Metakaolin is an aluminosilicate mineral product which is produced in quantities of several million tonnes per year worldwide and used in applications including supplementary cementitious materials in concretes,1 an intermediate phase in ceramic processing,2 as a geopolymer precursor,3 and as a paint extender. Metakaolin is formed by the dehydroxylation of kaolinite, whereby the initial crystalline layered structure of kaolinite (Al2Si2O5(OH)4) is subjected to local buckling and strain due to the loss of chemically bound water upon heating.4 It is well understood that the structure loses its crystallographic order between 500 and 750 °C, observed experimentally in conventional diffraction studies via the lack of Bragg scattering from materials heated above these temperatures.5–8 However, a study of metakaolin using energy-filtered transmission electron microscopy found a periodicity of around 14 Å along the axis corresponding to the c-direction of the kaolinite structure,9 indicating that the layered structure of kaolinite is not completely destroyed during the formation of metakaolin. Given the importance of metakaolin in various industry sectors, it is interesting to note that not more has been done in a crystallographic sense to systematically investigate the atomic structural changes taking place during dehydroxylation. It should be noted that any transformation from crystalline to amorphous is not straightforward to investigate at the atomic level,due to * To whom correspondence should be addressed. E-mail: jprovis@ unimelb.edu.au. Phone: +61 3 8344 8755. Fax: +61 3 8344 4153. † Department of Chemical & Biomolecular Engineering, University of Melbourne. ‡ Los Alamos National Laboratory. § Department of Mechanical Engineering, University of Melbourne.

the inherent difficulties associated with the depiction and modeling of amorphous structures. Hence, there exists the need to use additional techniques such as quantum mechanics to probe the structural changes at the atomic level. Other experimental techniques that have been used to characterize metakaolin include nuclear magnetic resonance (NMR), thermogravimetry (TGA), differential thermal analysis (DTA), and infrared spectroscopy (FTIR), and there exists a wealth of information in the literature from such studies regarding the process of kaolinite dehydroxylation and formation of metakaolin.5–7,10 However, this study represents the first detailed local structure analysis of this process. The current state of the art in experimental atomic structure solution of disordered crystals, nanoparticles, liquids, glasses, and other amorphous materials is the total scattering technique.11 This technique utilizes both Bragg and diffuse scattering from X-ray and/or neutron diffraction data sets and hence provides information about the degree of disorder/amorphicity of a material. Studies can be performed in reciprocal space, whereby structure refinement is carried out against the total scattering pattern. Alternatively, taking the Fourier transform of this pattern provides atom-atom correlations in real space, in the form of a pair distribution function (PDF). Studies using total scattering have been carried out on aluminosilicates besides kaolinite, such as zeolites12 and geopolymers,13,14 revealing the local structure of these materials. Hence, here, PDF analysis has been used to validate the atomic structures generated in order to ensure experimental plausibility of the local structure. Investigations aimed at determining the atomic structure of metakaolin include the use of laboratory X-ray-based radial distribution function analysis,15 NMR spectroscopy,10 and

10.1021/jp911108d  2010 American Chemical Society Published on Web 03/18/2010

Modeling Kaolinite Dehydroxylation “single-crystal” diffraction experiments on metakaolin synthesized from single crystals of kaolinite,16 all making advances toward understanding the structure. However, all apart from MacKenzie et al.10 presented the metakaolin structure based on the structural motif of a single kaolinite unit cell and thus were unable to replicate the amorphous nature of the local structure. MacKenzie et al.10 proposed a structure that was larger than the kaolinite unit cell, where the structure was determined from NMR results. These results, along with other known properties of metakaolin, were used to generate a representation of the metakaolin structure, although their investigation was limited by the fact that solid-state NMR provides only very limited information beyond the second coordination sphere of an atom. Our recent investigation into the atomic structure of metakaolin showed that an accurate representation of the structure can be achieved through an iterative density functional theory (DFT) and PDF refinement process,4 which is necessary when the structure is sufficiently complex that standard techniques are insufficient to accurately refine the local atomic structure. To further develop the understanding of how metakaolin is formed, here, we model the stepwise transition from kaolinite to metakaolin using DFT. In order to assess whether these models are experimentally plausible, pair distribution functions obtained from neutron and X-ray total scattering experiments are used for validation. Through the systematic comparison of simulated and experimental pair distribution functions, we present a route by which the dehydroxylation of kaolinite to metakaolin is able to proceed, maintaining agreement with thermodynamic and nanostructural data. Experimental Procedures Samples. High-purity kaolinite (KGa-1b, Source Clay Repository, Columbia MO) was calcined at temperatures of 450, 500, 550, 600, 650, and 750 °C. Powdered kaolinite was placed in an alumina tray in a small laboratory furnace being held at the desired temperature, calcined for 2 h in air, and then removed directly to cool in air under ambient conditions. Neutron PDF. High-resolution time-of-flight neutron powder total scattering was carried out on the NPDF beamline at the Lujan Neutron Scattering Center, Los Alamos National Laboratory.17 Samples were measured in standard vanadium cans in a Displex cryostat at 15 K. Standard data reduction was performed using PDFgetN,18 including background subtractions to remove incoherent scattering.19 For the most crystalline sample (450 °C), a Qmax of 35 Å-1 was used, whereas for the more disordered samples (500-750 °C), Qmax was set at 30 Å-1. A sine damping function was applied to each total scattering function at high Q, along with low-Q extrapolations.11 Neutron PDF analysis of unheated kaolinite is complicated by the high concentration of hydrogen, leading to a high incoherent scattering background. Work aimed at overcoming this difficulty is ongoing; neutron PDF data for unheated kaolinite are not presented here. X-ray PDF. High-resolution X-ray powder total scattering was performed using beamline 11-ID-B at the Advanced Photon Source, Argonne National Laboratory. Samples were measured in Kapton capillary tubes, and the measurements were performed under ambient conditions at a wavelength of 0.2127 Å. Standard data reduction was performed using Fit2D20,21 and PDFgetX2.22 Total scattering optimization was performed on each data set using the 1/E quadratic energy optimization over 15-20 Å-1.22 PDFs were produced using a Qmax of 19 Å-1. X-ray Diffraction. High-resolution powder diffraction patterns were obtained from the Australian Synchrotron, using the

J. Phys. Chem. A, Vol. 114, No. 14, 2010 4989 TABLE 1: Number of Water Molecules Removed Per Step before the Structure Is Geometry-Optimizeda structure step numberb

number of H2O molecules removed prior to optimization

1 2 3 4 5 6 7 8 9 10 11 12 13 (metakaolin)

1 1 1 1 2 2 2 3 4 3 4 3 1

a

Initially, the number removed is low to replicate a slow rise in the rate of dehydroxylation with increasing temperature, as observed in thermogravimetry (TGA).26,27 b Figure 1 displays the corresponding structures (geometry-optimized).

Powder Diffraction beamline.23 Samples were measured in glass capillary tubes at 80 K with a wavelength of 0.82631 Å. DFT Modeling Methodology. Density functional modeling was carried out using the generalized gradient functional BLYP as implemented in the DMol3 v4.4 software using a Quad-core desktop workstation for smaller computations and eight CPUs (using 1.6 GHz Itanium2 processors) on an AC cluster (NCINF facility, hosted by Australian National University, Canberra) for larger supercells. The initial periodic 2 × 2 × 2 kaolinite supercell was created from the unit cell structure reported by White et al,24 with a ) 10.298, b ) 17.867, c ) 14.769, R ) 91.93, β ) 105.042, and γ ) 89.791. All modeling was carried out using periodic boundary conditions with the dimensions fixed to these values. The numerical basis set used was double numerical (two atomic orbitals for each occupied atomic orbital) plus a polarization p function on all hydrogen atoms (DNP)25 to account for hydrogen bonding. No pseudopotentials or effective core potentials were used. Convergence thresholds were set at 1 × 10-4 hartree for energy, 0.02 hartree/Å for maximum force, and 0.05 Å for maximum displacement. A SCF convergence of 10-4 was used, along with 1 × 1 × 1 k-point sampling and 0.005 hartree smearing to aid convergence when needed. The last structure (i.e., metakaolin) was subjected to a geometry optimization with higher accuracy convergence criteria (1 × 10-5 hartree for energy, 0.002 hartree/Å for maximum force, 0.005 Å for maximum displacement, SCF convergence of 10-6 hartree, 3 × 2 × 1 k-point sampling, and no smearing). Table 1 reports the number of water molecules which were removed from the structure at each step of the stepwise dehydroxylation. In total, there were 13 steps modeled to represent the stepwise transition, beginning with the removal of 1 water molecule per step and then increasing this number progressively to model the experimentally observed gradual increase in the rate of dehydroxylation.26,27 Also, the hydroxyl groups selected for removal were initially selected from around a single region of the supercell structure to represent the preferential loss of water from more strained parts of the structure. At the beginning, only inner-surface hydroxyls (i.e., those located “beneath” the alumina layer of the kaolinite structure)24 were selected to participate in the dehydroxylation process; then, as the extent of dehydroxylation increased, the inner hydroxyls became involved. The residual one in eight kaolinite hydroxyls which are present in metakaolin, as reported

4990

J. Phys. Chem. A, Vol. 114, No. 14, 2010

White et al.

Figure 1. Structural models used in the stepwise dehydroxylation process in DFT. The structures given here are for each step after geometry optimization. The enlarged atoms are those chosen to form water molecules upon dehydroxylation and are therefore those which are removed before the next optimization step is carried out.

by MacKenzie et al.,10 were located in the inner-surface hydroxyl positions in order to maintain the 1:1 layering of metakaolin without the silica and alumina layers collapsing into each other. Results and Discussion DFT Structures. The structures resulting from the systematic transition from kaolinite to metakaolin via stepwise dehydroxylation are displayed in Figure 1 (with cif files provided in Supporting Information). Hydroxyl groups selected to form water molecules in general obeyed two rules; only hydroxyl groups located close to each other will react with each other, and the process of removal of hydroxyl groups maintains the presence of IV-, V-, and VI-coordinated aluminum atoms, with a gradual decrease in coordination numbers upon heating.28 However, as will be discussed later, the aluminum coordination does not always remain in the IV-, V-, and VI-fold distribution when the structure is geometry-optimized using DFT (as was also shown in our previous work4). X-ray Diffraction. In order to confirm the nature and concentration of the crystalline components present in the PDF experiment samples, high-resolution synchrotron X-ray diffraction was performed. The diffraction patterns are shown in Figure 2, where the transition from kaolinite to metakaolin (crystalline to disordered) is clearly evident in the 450 and 500-550 °C samples (note that the temperatures given throughout the paper indicate ex situ calcination temperatures; all diffraction data collection was carried out at or below room temperature). The 500 °C sample still retains most of the kaolinite structure, albeit with lower peak intensity, indicating that a sizable percentage of the kaolinite has already transitioned to an amorphous

Figure 2. Synchrotron high-resolution powder diffraction of kaolinite and its calcined derivatives. Marked peaks are from anatase (A), rutile (R), and ice (I); all others are assigned to the kaolinite structure.

structure. This sample is therefore denoted as semicrystalline. The natural kaolinite sample contained a 1.66 wt% impurity of crystalline anatase,29 which is seen to undergo a phase transition to rutile in the 500 °C sample. This impurity was present at a concentration too low to detect in the PDFs from neutrons or X-rays; hence, only the aluminosilicate phases are modeled when comparing PDF simulations with experiment. Small amounts of ice accumulating on the outside of the capillaries from the cryostream blower are also visible in the 500 and 600-750 °C samples.

Modeling Kaolinite Dehydroxylation

J. Phys. Chem. A, Vol. 114, No. 14, 2010 4991

Figure 3. Neutron and X-ray PDFs of kaolinite and the products of its calcination at temperatures as marked. (a) Neutron and (b) X-ray PDFs, showing the short- and medium-range transition from kaolinite (X-ray) and 450 °C (neutron) to metakaolin (750 °C). Enlarged low-r regions of the (c) neutron and (d) X-ray PDF data, displaying the distinct change in the Al-O interaction distance with an increase in calcination temperature.

Experimental Pair Distribution Functions. Experimental PDFs are shown in Figure 3. Figure 3a and b shows the neutron and X-ray PDFs, respectively, for the crystalline (unheated for X-rays; 450 °C for neutrons), semicrystalline (500 °C), and amorphous (750 °C) samples. The crystalline sample is evident in both figures due to the medium-range structural peaks present in the range of 10-20 Å. As a percentage of kaolinite transitions to an amorphous structure (500 °C), the medium-range structural peaks become less intense yet remain observable. This result correlates with the high-resolution X-ray diffractogram (Figure 2) where kaolinite Bragg peaks were still visible, although lower in intensity, in this sample. The amorphous sample (750 °C) shows minimal medium-range structural order, with some noise present at high r in the neutron PDF data due to lower statistics in the corresponding total scattering function. First- and second-neighbor correlations are shown in Figure 3c and d, which displays the neutron and X-ray PDFs up to 4 Å. Immediately apparent in this figure is the significant change in the Al-O atomic interaction distance. The Al-O peak changes from well-defined (sharp) in kaolinite to being a shoulder on the Si-O peak in the amorphous samples. This result is in agreement with what is expected structurally upon removal of water molecules. A similar distinction between Al-O and Si-O bond lengths has been seen by Petkov et al.30 in calcium aluminosilicate glasses; however, the alumina in this instance is initially more ordered and dissimilar from the Si-O correlation as is seen by the distinct Al-O peak (rather than just a shoulder) in the crystalline structures in Figure 3.

As was discussed in detail by White et al.,4 the representative structure of metakaolin contains relatively rigid silica tetrahedra, while the once-octahedral alumina sites change to a mixture of III-, IV-, and V-coordinated as both silica and alumina layers buckle to accommodate the changes in the alumina layers. Hence, the distribution of Al-O interaction distances broadens, as seen in Figure 3c and d, and tends to move toward shorter distances, increasing the overlap with the Si-O peak. In contrast to the change in Al-O interaction distances, the Si-O peak remains distinct throughout the dehydroxylation process. The peak does broaden a little (Figure 3c) and shifts to a slightly larger interaction distance. Therefore, the silica tetrahedra present in kaolinite remain intact in metakaolin but do rearrange a little to accommodate the major changes in the alumina layers. The ability to generate bond length and angle distributions directly from structural information is one of the strengths of the DFT method (or any atomistic structure modeling technique). The correlations between experimental and modeling bond length distributions will therefore be explored in more detail below. DFT Structures and Experimental Validation. In order to provide experimental validation for the simulated stepwise dehydroxylation process, neutron and X-ray PDFs were generated from the DFT structures for each step and compared with the experimental PDFs. The same instrument and data processing parameters (Qmax, Qbroad, Qdamp) were used in the generation of these simulated PDFs as those for the experimental PDFs. Atomic displacement parameters were refined against the ex-

4992

J. Phys. Chem. A, Vol. 114, No. 14, 2010

White et al.

Figure 4. Comparison of the simulated (DFT model) and experiment PDFs of the kaolinite to metakaolin transition, illustrating that the transition can be modeled at the atomic level using a step-by-step method (based on DFT structure relaxations). The structure changes from crystalline to semicrystalline between 450 and 500 °C and then from semicrystalline to amorphous between 500 and 550 °C. There is relatively little change in the local structure between 550 and 750 °C (metakaolin).

perimental data and restricted to between 0.001 and 0.04 Å2 (all refined/set parameters are given in Supporting Information Table 1). The closest DFT match to each experimental PDF is shown in Figure 4. As is seen in the neutron and X-ray PDFs, DFT shows that the structure changes from the rigid alumina layers of kaolinite to ones that are buckled, disordered, and strained. This can be seen most clearly in the X-ray and neutron data in Figure 4 by monitoring the Al-O interaction distance, shown by the peak immediately below 2 Å. The DFT kaolinite structure and the experimental PDFs each contain a distinct Al-O bond peak, characteristic of the rigid alumina layer in this material. The model transitions to a semicrystalline structure by step 9 (which matches most closely to the 500 °C experimental data sets and corresponding TGA mass loss), where the Al-O peak is still visible but is greatly decreased in intensity. By step 12 (identified as being equivalent to a calcination temperature of 550 °C), the Al-O peak in the DFT model structure PDF has reduced to a shoulder, reiterating the fact that there is very significant disorder present in the alumina layers in metakaolin. It is also apparent from Figures 3 and 4 that the experimental PDFs for the samples obtained at 550 and 750 °C are very similar. In fact, if all of the experimental PDFs are overlaid on top of one another, 550 °C is only slightly different from 600 °C (not shown), and 600 to 750 °C are the same to within experimental uncertainty when comparing the region up to r ) 10 Å. This means that the local structure of metakaolin is in

fact obtained around 550 °C when preparing samples by the methodology used here. The implications of this result are significant in optimizing the industrial production of metakaolin, an aim which has been discussed in numerous investigations studying the correlation between calcination method and pozzolanic activity.1,6,31–33 Many such studies have indeed observed only limited differences between the reactivities of soak-calcined metakaolins produced at temperatures ranging from 550 to 750 °C, although with increasing calcination time and temperature, there can also be a decrease in particle surface area which affects reactivity and which is not captured by the local structure study presented here. Differences between the experimental and DFT-generated PDFs can be attributed to several points. First, the experimental samples of kaolinite and its various calcined forms contain stacking faults and preferred orientation. It was almost inevitable that during filling of sample capillaries for X-ray studies that the particles of kaolinite and calcined kaolinite powder would align along the c-axis, as can been seen in Figure 2, where the 001 (2θ ) 6.62°) and 002 (2θ ) 13.32) Bragg peaks are significantly more intense than would be expected by comparison with the intensities of the other peaks. This has been confirmed in the X-ray total scattering patterns (not shown). Preferred orientation was not detected in the neutron results, which was expected as sample cans are relatively large compared to capillaries and hence the particles tend not to align as strongly. There is currently not a widely accepted approach

Modeling Kaolinite Dehydroxylation

J. Phys. Chem. A, Vol. 114, No. 14, 2010 4993

Figure 6. Aluminum coordination of kaolinite and the dehydroxylated structures. Aluminum atoms transition from octahedral coordination (VI-fold coordinated) to increasing percentages of V-fold and IV-fold. The existence of small amounts of III-fold aluminum is indicative of the strained nature of the alumina layers at higher degrees of dehydroxylation, as seen previously.4 Figure 5. Neutron PDFs derived in (a) our previous study4 and (b) the current study of metakaolin, along with the 750 °C experimental PDF.

to account for preferred orientation during the generation and analysis of PDFs. However, given that the agreement between experiment and model in Figure 4 is the same for X-ray and neutron data, preferred orientation is deemed not to have a significantly adverse effect on the experimental PDFs. It is also well-known that this particular source of natural kaolinite clay (Georgia kaolin source mineral KGa-1b) does contain stacking faults,29 and this is evidenced by the “sharp rise-slow fall” peak shape in Figure 2 between 11 and 13° 2θ.34 Stacking faults can be modeled at the atomic level and subsequently in the model-generated PDF. Investigations by Masadeh et al.35 and Neder and Korsunskiy36 successfully modeled CdSe and CdS nanoparticles containing stacking faults. The modeling of stacking faults in kaolinite (and then successively in its calcined derivatives) is more complex due to the very large unit cells involved and is beyond the scope of this study. However, such an investigation would elucidate the exact effect of this phenomenon on the experimental PDFs reported in this study and will be a valuable area of future study. Other possible reasons for discrepancies between experimental and DFT-generated PDFs include limitations in the simulation methodology used to model the atomic structures. Calculations were performed on supercell structures that were constructed as 2 × 2 × 2 kaolinite unit cells (where a single kaolinite unit cell contained 34 atoms). Larger supercells proved to be extremely computationally intensive. Hence, there is a question regarding the supercell size and whether it restricts the development of an “amorphous-like” structure. It should be noted that a preliminary investigation was performed on kaolinite supercells, looking at determining the supercell size required to eradicate unphysical periodic interactions. For the convergence tolerance used in this study, the 2 × 2 × 2 supercell was found to be adequate. As can be seen in Figure 4, the supercell used does allow for development of a disordered structure over the range of comparison (0.0-8.0 Å). Finally, the level of precision used during the DFT simulations is a possible contributing factor in the discussion of differences between experimental and model PDFs. Simulations were performed using a relatively low precision convergence tolerance in order to study a large supercell due to memory limitations imposed by the computing architecture (i.e., 4GB per processor maximum). However, it has been shown that the computational parameter set (functional, basis set, and conver-

gence tolerances) used here is sufficient to obtain a representative structure of an amorphous material.4 A higher level of accuracy would be required when investigating other properties such as phonon vibrations and band structures, but the agreement between model and experiment is more than adequate for this level of structural investigation. Metakaolin Structure. The DFT-generated PDF of the metakaolin structure from our previous study4 is shown in Figure 5 compared against the experimental PDF, along with the current DFT-generated PDF. The previous metakaolin structure was obtained by removing seven in eight hydroxyls from a slightly larger (3 × 2 × 2) supercell of kaolinite and then allowing the structure to relax using DFT geometry optimization. The structure was then subjected to an iterative DFT-PDF refinement process using the 750 °C neutron PDF data set, so that the resulting structure was both energetically and experimentally plausible.4 As can be seen in Figure 5, the PDFs of the metakaolin structures obtained by the two techniques are not precisely the same; however, they both agree reasonably well with the experimental PDF. There does appear to be better agreement in the structure generated in the current study, especially in the first two peaks (∼1.7 and 2.6 Å), possibly due to the stricter convergence tolerances applied during the last DFT geometry optimization. Hence, it has been shown that a probable structure of metakaolin can be derived either via a DFT-PDF iterative process or via stepwise DFT modeling of the dehydroxylation process. There are scenarios in which either the former or the latter of these approaches would be strongly preferred, but we have proven the utility of each of the two approaches by investigation of the metakaolin structure in both ways. It should also be noted that the results from the current study depict only one of many possible transition paths from kaolinite to a metakaolin structure, that is, there are more than 3 × 1084 (64!/8!) different permutations, whereby 56 out of 64 hydroxyl groups may be removed sequentially from this kaolinite supercell. Further research would be required to investigate a representative sample of these many potential paths from kaolinite to metakaolin, which would ideally be looked at using a modeling technique (such as molecular dynamics) capable of searching configuration space for the many almost-equal-energy local minima which exist. To do so using DFT at the level of theory implemented here would be prohibitively expensive in terms of computational resources. However, the very close agreement between theory and experiment displayed in Figure 4 shows that the approach utilized here provides a very plausible

4994

J. Phys. Chem. A, Vol. 114, No. 14, 2010

White et al.

Figure 7. Bond length distributions of (a) kaolinite, (b) dehydroxylation step 4, (c) dehydroxylation step 9, and (d) dehydroxylation step 13 (metakaolin). As kaolinite transitions to metakaolin, the O-Si bond length distribution remains relatively unchanged, whereas the O-Al distribution undergoes significant change.

Figure 8. Bond angle distributions of (a) kaolinite, (b) dehydroxylation step 4, (c) dehydroxylation step 9, and (d) dehydroxylation step 13 (metakaolin). All bond angle distributions broaden significantly as kaolinite transitions to metakaolin.

mechanism for this process, which has never before been described in atomistic detail. DFT Structures s Chemical Feasibility. In order to assess the chemical feasibility of the dehydroxylated structures, coordination numbers, bond lengths, and bond angles were generated for each. For analysis of coordination numbers, the defined distances used to calculate the probability were rAl-O

) 2.2 Å and rSi-O ) 2.0 Å, reflecting the possibility of longer bonds being formed to aluminum atoms as they have more flexible coordination numbers than silicon atoms. These values of cutoff radii have been shown to be appropriate in our earlier study of the structure of metakaolin.4 The coordination of silicon atoms remains IV-fold throughout the dehydroxylation process, apart from in the case of one atom, where the collapse of the

Modeling Kaolinite Dehydroxylation interlayer spacing causes an oxygen atom from the neighboring alumina layer to come within 2.0 Å. This may be considered an artifact induced by the coordination number calculation method and choice of radius defined. The silica layers are thus seen to remain relatively rigid compared to the alumina layers. The change in aluminum coordination during dehydroxylation is shown in Figure 6, where the initially octahedral alumina layer of kaolinite undergoes significant disruption. As more and more water molecules are removed from the structure, the coordination number distribution shifts toward predominantly IV- and V-fold, with a small percentage of III-fold. This major change in aluminum coordination indicates that the alumina layers are losing their regularity and buckling due to the removal of water molecules. Hence, the existence of tricoordinated aluminum atoms appears plausible since the alumina layers are experiencing large changes. Previous investigations that have reported tricoordinated aluminum include our earlier work4 and a study by Benoit et al.,37 where ab initio molecular dynamics simulations were performed on a Ca-aluminosilicate melt. Bond length distributions, calculated from the bond lengths in the DFT structures (Figure 7) and plotted as histograms, can be compared with the experimental pair distribution functions in Figure 4 in order to assess chemical feasibility of the computed structure. The results show once again the slight change in the O-Si bond length distribution due to the fairly rigid behavior of the SiO2 layers, whereas there is significant change in O-Al distances. Hence, the energetically feasible atomic structures generated via DFT mimic what is measured experimentally. The bond angle distribution is not a direct output of the PDF method; however, it can be invoked via analysis of the atomic structure. These distributions, calculated from the DFT structure, are given in Figure 8, which indicates that it is not only the aluminum-related angles which undergo significant change. It is expected that due to the large changes of the structure in the alumina layers, the distributions of Al-O-Al and Si-O-Al bond angles would broaden. However, as shown in Figure 8, the Si-O-Si bond angle distribution also widens considerably. This indicates that the silica rings are rearranging slightly (via buckling) to accommodate the significant changes in the alumina layers without noticeable broadening of the O-Si bond length distribution, as is seen in Figures 1 and 7. Conclusion We have demonstrated that DFT can be used to model the structural changes of kaolinite as it undergoes dehydroxylation in a stepwise manner from crystalline to amorphous. By carrying out such a procedure, one out of the many possible dehydroxylation paths has been elucidated. Hence, we have shown that an amorphous structure can be generated without refinement of atomic positions against experimental data and, moreover, that the structure created is energetically feasible and therefore can be used to elucidate other (e.g., electronic) structural information. By ensuring that the simulated local structures match what is seen experimentally through pair distribution analysis, further investigations into the various dehydroxylated structures can be performed, such as phonon excitations, for which an energetically feasible structure is required. The local structural changes occurring in the DFT-generated kaolinite and dehydroxylation product structures are seen to mimic the changes seen experimentally by X-ray and neutron pair distribution analysis. As kaolinite transforms to metakaolin, the aluminum coordination changes from octahedral to a mixture of IV- and V-fold (with a small percentage of III-fold). On the

J. Phys. Chem. A, Vol. 114, No. 14, 2010 4995 other hand, silica coordination maintains a IV-fold nature as in kaolinite, which is usual in silicate materials. However, as more and more water molecules are removed from the structure, the alumina layers become substantially buckled, and the silica layers rearrange slightly to accommodate these changes. This progression from order to disorder is also visible from the bond length distributions of O-Si and O-Al, as well as the bond angle distributions. By using this methodology to model the local structural changes as materials transform from crystalline to amorphous, there exists the possibility to elucidate the local structures of amorphous materials that are responsible for performance of these materials at the macroscale. Acknowledgment. This work was funded in part by the Australian Research Council (ARC) (including some funding via the Particulate Fluids Processing Centre, a Special Research Centre of the ARC) and in part by a studentship paid to Claire White by the Centre for Sustainable Resource Processing via the Geopolymer Alliance. The authors would like to thank Dr. Hyunjeong Kim, Los Alamos National Laboratory, for assistance with the NPDF experiment. We also thank Dr. Katherine Page for collecting the X-ray PDF data from ID-11-B at the APS, Argonne National Laboratory, and Dr. Kia Wallwork for assistance with the experiment at the Australian Synchrotron. Use of the Advanced Photon Source at Argonne National Laboratory was supported by the U.S. Department of Energy, Office of Science, Office of Basic Energy Sciences, under Contract No. DE-AC02-06CH11357. The NPDF instrument is located at Los Alamos Neutron Science Center, funded by the DOE Office of Basic Energy Sciences. Los Alamos National Laboratory is operated by Los Alamos National Security LLC under DOE Contract DE-AC52-06NA25396. The upgrade of NPDF has been funded by the NSF through Grant DMR 0076488. X-ray diffraction data were collected on the Powder Diffraction beamline (10BM1) at the Australian Synchrotron, Victoria, Australia. The views expressed herein are those of the authors and are not necessarily those of the owner or operator of the Australian Synchrotron. The DFT modeling was supported by an award under the Merit Allocation Scheme on the NCI National Facility at the ANU. Supporting Information Available: Cif files containing coordinates of the structures shown in Figure 1 (steps 1-13) are given in the zip folder cifs.zip. These are the structures obtained using density functional modeling, prior to comparison with experimental PDFs. Parameters refined/set in PDFgui during comparison of DFT structures with experimental PDFs are provided in Table 1. This material is available free of charge via the Internet at http://pubs.acs.org. References and Notes (1) Sabir, B. B.; Wild, S.; Bai, J. Cem. Concr. Compos. 2001, 23, 441. (2) Brindley, G. W.; Nakahira, M. J. Am. Ceram. Soc. 1959, 42, 311. (3) Duxson, P.; Provis, J. L.; Lukey, G. C.; van Deventer, J. S. J. Cem. Concr. Res. 2007, 37, 1590. (4) White, C. E.; Provis, J. L.; Proffen, T.; Riley, D. P.; van Deventer, J. S. J. Phys. Chem. Chem. Phys. 2010, 12, 3239–3245. (5) Badogiannis, E.; Kakali, G.; Tsivilis, S. J. Therm. Anal. Calorim. 2005, 81, 457. (6) Shvarzman, A.; Kovler, K.; Grader, G. S.; Shter, G. E. Cem. Concr. Res. 2003, 33, 405. (7) Rahier, H.; Van Mele, B.; Biesemans, M.; Wastiels, J.; Wu, X. J. Mater. Sci. 1996, 31, 71. (8) He, H. P.; Guo, J. G.; Zhu, J. X.; Hu, C. Clay Miner. 2003, 38, 551. (9) Lee, S.; Kim, Y. J.; Moon, H. S. J. Am. Ceram. Soc. 2003, 86, 174.

4996

J. Phys. Chem. A, Vol. 114, No. 14, 2010

(10) MacKenzie, K. J. D.; Brown, I. W. M.; Meinhold, R. H.; Bowden, M. E. J. Am. Ceram. Soc. 1985, 68, 293. (11) Egami, T.; Billinge, S. J. L. Underneath the Bragg Peaks: Structural Analysis of Complex Materials; Pergamon: Elmsford, NY, 2003. (12) Martínez-Iñesta, M. M.; Peral, I.; Proffen, T.; Lobo, R. F. Microporous Mesoporous Mater. 2005, 77, 55. (13) Bell, J. L.; Sarin, P.; Driemeyer, P. E.; Haggerty, R. P.; Chupas, P. J.; Kriven, W. M. J. Mater. Chem. 2008, 18, 5974. (14) Bell, J. L.; Sarin, P.; Provis, J. L.; Haggerty, R. P.; Driemeyer, P. E.; Chupas, P. J.; van Deventer, J. S. J.; Kriven, W. M. Chem. Mater. 2008, 20, 4768. (15) Gualtieri, A.; Bellotto, M. Phys. Chem. Miner. 1998, 25, 442. (16) Brindley, G. W.; Nakahira, M. J. Am. Ceram. Soc. 1959, 42, 314. (17) Proffen, T.; Egami, T.; Billinge, S. J. L.; Cheetham, A. K.; Louca, D.; Parise, J. B. Appl. Phys. A 2002, 74, S163. (18) Peterson, P. F.; Gutmann, M.; Proffen, T.; Billinge, S. J. L. J. Appl. Crystallogr. 2000, 33, 1192. (19) Page, K.; White, C. E.; Estell, E.; Llobet, A.; Proffen, T. In preparation. (20) Hammersley, A. P. FIT2D V9.129 Reference Manual V3.1; European Synchrotron Radiation Facility: Grenoble, France, 1998. (21) Hammersley, A. P.; Svensson, S. O.; Hanfland, M.; Fitch, A. N.; Ha¨usermann, D. High Pressure Res. 1996, 14, 235. (22) Qiu, X.; Thompson, J. W.; Billinge, S. J. L. J. Appl. Crystallogr. 2004, 37, 678.

White et al. (23) Wallwork, K. S.; Kennedy, B. J.; Wang, D. AIP Conf. Proc. 2007, 879. (24) White, C. E.; Provis, J. L.; Riley, D. P.; Kearley, G. J.; van Deventer, J. S. J. J. Phys. Chem. B 2009, 113, 6756. (25) Delley, B. J. Chem. Phys. 2000, 113, 7756. (26) Guggenheim, S.; van Groos, K. A. F. Clays Clay Miner. 2001, 49, 433. (27) Lecomte, G. L.; Bonnet, J. P.; Blanchart, P. J. Mater. Sci. 2007, 42, 8745. (28) Rocha, J. J. Phys. Chem. B 1999, 103, 9801. (29) Pruett, R. J.; Webb, H. L. Clays Clay Miner. 1993, 41, 514. (30) Petkov, V.; Billinge, S. J. L.; Shastri, S. D.; Himmel, B. Phys. ReV. Lett. 2000, 85, 3436. (31) Mejı´a de Gutierrez, R.; Torres, J.; Guerrero, C. E. Mater. Constr. 2004, 54, 65. (32) Bich, C.; Ambroise, J.; Pera, J. Appl. Clay Sci. 2009, 44, 194. (33) Salvador, S. Cem. Concr. Res. 1995, 25, 102. (34) Young, R. A.; Hewat, A. W. Clays Clay Miner. 1988, 36, 225. (35) Masadeh, A. S.; Bozin, E. S.; Farrow, C. L.; Paglia, G.; Juhas, P.; Billinge, S. J. L.; Karkamkar, A.; Kanatzidis, M. G. Phys. ReV. B 2007, 76, 11. (36) Neder, R. B.; Korsunskiy, V. I. J. Phys.: Condens. Matter 2005, 17, S125. (37) Benoit, M.; Ispas, S.; Tuckerman, M. E. Phys. ReV. B 2001, 64, 224205.

JP911108D