Enhanced catalytic cracking activity of kaolinites through ... - QUT ePrints

1 downloads 0 Views 454KB Size Report
Onuma Carmodya, Ray L. Frost 1a, János Kristóf b, Serge Kokota, J. Theo ... Technology, 2 George Street, GPO Box 2434, Brisbane Queensland. 4001 ...
COVER SHEET

This is the author-version of article published as:

Frost, Ray and Carmody, Onuma and Kloprogge, Theo and Mako, Eva and Kristof, Janos and Kokot, Serge (2006) Modification of kaolinite surfaces through mechanochemical activation with quartz a DRIFT and chemometrics study. Applied Spectroscopy 60(12):pp. 1414-1422. Please enter the information about this item. Fields mark Accessed from http://eprints.qut.edu.au © 2006 Society for Applied Spectroscopy

Modification of kaolinite surfaces through mechanochemical activation with quartz -a DRIFT and chemometrics study

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46

Onuma Carmody a, Ray L. Frost 1a, János Kristóf b, Serge Kokot a, J. Theo Kloprogge a, Éva Makóc a

Inorganic Materials Research Program, Queensland University of Technology, 2 George Street, GPO Box 2434, Brisbane Queensland 4001, Australia.

b

Department of Analytical Chemistry, University of Pannonia, Veszprem, Hungary

c

Department of Silicate and Materials Engineering, University of Pannonia, Veszprem, Hungary

Abstract Studies of kaolinite surfaces are of industrial importance. One useful method for studying the changes in kaolinite surface properties is to apply chemometric analyses to the kaolinite surface infrared spectra. A comparison is made between the mechanochemical activation of Kiralyhegy kaolinites with significant amounts of natural quartz and the mechanochemical activation of Zettlitz kaolinite with added quartz. DRIFT spectra were analysed using, Principal Component Analysis (PCA), and multi-criteria decision making (MCDM) methods, PROMETHEE and GAIA. The clear discrimination of the Kiralyhegy spectral objects on the two PC scores plots (400-800 and 800-2030 cm-1) indicated the dominance of quartz. Importantly, no ordering of any spectral objects appeared to be related to grinding time in the PC plots of these spectral regions. Thus, neither the kaolinite nor the quartz, are systematically responsive to grinding time according to the spectral criteria investigated. The third spectral region (2600-3800 cm-1 – OH vibrations), showed apparent systematic ordering of the Kiralyhegy and, to a lesser extent, Zettlitz spectral objects with grinding time. This was attributed to the effect of the natural quartz on the delamination of kaolinite and the accompanying phenomena (i.e. formation of kaolinite spheres and water). The mechanochemical activation of kaolinite and quartz, through dry grinding, results in changes to the surface structure. Different grinding times were adopted to study the rate of destruction of the kaolinite and quartz structures. This relationship (i.e. grinding time) was classified using PROMETHEE and GAIA methodology. INDEX HEADINGS: kaolinite surfaces, halloysite, chemometrics, PCA, PROMETHEE and GAIA, DRIFT

1

Corresponding author: Ray Frost Email address: [email protected] Address: 2 George Street, Brisbane Q 4001, Australia

1

47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95

1

Introduction

Interest in the mechanochemical treatment of kaolinite first originated as part of soil science 1-3. Takahashi (1959) undertook intensive research on the mechanochemical activation of kaolinite and related polytypes 4-8. It was found that there were two structural processes involved in the mechanochemical treatment (a) the reduction in particle size, and (b) a re-aggregation process. The first stage involves the production of a non-crystalline substance attended by the disordering of the crystals, and the other is the reaggregation process. The reduction step and production of the non-crystalline substance are connected to that of reaggregation. At a certain stage of dry grinding, the reaggregates are spherical particles, which have a zeolitic structure 7. Dry grinding causes the kaolinite layers to fragment, and results in the formation of spheroidal particles 9, 10. It has been proposed that dry grinding removes the hydroxyl units from the kaolinite, but ultimately facilitates the formation of new OH groups. Studies have shown that the mechanochemical treatment of kaolinite results in the loss of the d(001) peak intensity 11-14. This effectively makes the kaolinite into a non-diffracting material different to that of the original starting material with significantly larger surface areas of the particles 14, 15. Many techniques have been employed to study the structural processes during mechanochemical activation 16-18. Besides X-ray diffraction, thermal analysis methods have been used to show the conversion of hydroxyl groups to water, which is both weakly and strongly hydrogen bonded to the kaolinite surface 10, 12, 19. Infrared spectroscopy has also proven to be a most useful tool for exploring the changes in molecular structure of kaolinite during grinding 20-23. Recent studies have shown that the processes of mechanochemical treatment of kaolinite are as yet not well understood 24. Indeed, the effect of grinding with or without quartz appears to produce different results 14. Common avenues for data interpretation from the above studies have been effectively exhausted. However, in general, it is well known that multi-variate data analysis with the aid of chemometrics methods may often highlight patterns and relationships that cannot be observed by conventional means. A previous study 25 based on DRIFT spectroscopy of several different kaolinite samples, and using chemometrics for the interpretation of spectral data, indicated that the presence of the dispersed kaolinite sample in the presence of natural quartz lead to a different and mechanically much more efficient grinding process. It was observed that the presence of natural quartz is of considerable significance to the grinding process as quartz enhances the delamination process. This observation is in agreement with the work of Schrader 26, 27, which showed that when quartz is ground important changes occur to the nature of the crystals in the first few hours grinding. In addition, the chemometrics interpretation of the DRIFT spectra indicated that the kaolinite surface hydroxyl groups are being affected in parallel with the quartz during the grinding process. In this paper, the effects of mechanochemical activation of kaolinite by itself and in the presence of quartz are reported. The role of quartz as a grinding agent is investigated, particularly in altering the surface and structural modification of the

2

96 97 98 99 100 101 102 103 104 105 106 107 108

kaolinite. Three samples were investigated: natural sand (quartz); Zettlitz kaolinite with 75% quartz added; and Kiralyhedgy kaolinite with 70% natural quartz. The effects of grinding on the three different samples was followed by DRIFT spectroscopy, and spectral interpretation was facilitated by chemometrics. The study focuses on a search for patterns, trends and relationships and attempts to relate these directly to the spectra, and hence to the mechanochemical activation of the quartz and kaolinites. A well known chemometrics method, exploratory Principal Component Analysis (PCA)was applied, as well as the sparsely used multi-criteria decision making (MCDM) methodology, PROMETHEE (Preference Ranking Organisation METHod for Enrichment Evaluations) and GAIA (Geometrical Analysis for Interactive Assistance), are used to explore the relationships between spectroscopic data and other available information such as grinding time.

109

2

110

2.1

111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134

Experimental The sand and kaolinites

The (high-grade) Kiralyhegy kaolinite samples contain approximately 70% natural quartz. The low-defect Zettlitz kaolinite samples were made by homogenizing the kaolinite with 75 % sand. These samples were sourced from: • • •

natural sand (Hungary); Kiralyhegy kaolinite (Hungary); and Zettlitz (Sedlec) kaolinite (Slovakia).

Chemical composition of sand and kaolinites was determined by X-ray fluorescence, (Table I). 2.2

Milling procedure

A Fritsch pulverisette 5/2-type laboratory planetary mill was used for grinding (0, 1, 2, 3, 4, 6 and/or 10 hours). Each milling was carried out with a 10 g air-dried sample in an 80 cm3 capacity stainless steel (18 % Cr +8 % Ni) pot using 8 (31.6 g) stainless steel balls (10 mm diameter). The applied rotation speed was 374 r.p.m. All samples were ground under the same conditions.

135

2.3

136 137 138 139 140 141

The ground samples were analysed with the use of a Diffuse Reflectance Fourier Transform Infrared (DRIFT) accessory fitted to a Bio-Rad FTS 60A spectrometer. The scans (512) were obtained at a resolution of 2 cm-1 with a mirror velocity of 0.3 cm/s. Approximately 3 wt % of a sample was dispersed in 100 mg oven- dried spectroscopic grade KBr ( refractive index = 1.559; particle size = 5-20

DRIFT spectroscopy

3

142 143 144 145 146

μm), ground lightly in an agate mortar and pestle, and placed into the 6 mm diameter cup for spectral sampling. Background KBr spectra were obtained, and spectra ratioed to the background. The spectra were displayed in reflectance mode versus wavenumber.

147

2.4

148 149 150 151 152 153 154 155 156 157

DRIFT spectra were baseline corrected, zero-offset and subdivided into three regions, 400-830, 830-2030, and 2600-3800 cm-1 (GRAMS® 32). Thus, a data matrix for each spectral range consisted of 17 spectral objects and 250 wavenumber variables selected at equally spaced wavenumbers for each of the three spectral ranges so as to comply with the 256 column width of the Excel 5 spreadsheet. Each matrix was then transferred to an Excel spreadsheet (Excel 5, Microsoft) prior to submission to PCA (Sirius, Version 6.0, PRS, Bergen, Norway) and PROMETHEE and GAIA. (Decision Lab 2000, Version 2, Visual Decision, 2000)

158

2.5

159

2.5.1

160 161 162 163 164 165 166 167 168 169 170

The raw data matrix consisted of DRIFT spectra objects as rows with wavenumber variables as columns. To extract the data variance effectively, it is often necessary to rescale or pretreat the raw data matrix. In this work, the raw data matrix was double centred (i.e. y-mean scaled followed by x-mean scaling). This procedure removes the size component reflected by the first PC of the unpretreated data matrix leaving essentially the information responsible for the data variance and noise. The resulting matrix was then submitted to variance scaling or standardisation to bring each column variable to unit variance. The pretreated matrix was then submitted to PCA and PROMETHEE and GAIA.

171

2.5.2

172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187

DRIFT spectra processing

Chemometric analysis Data pretreatment

Principal Components Analysis (PCA)

PCA is a well known chemometrics method facilitating pattern recognition and display of data. PCA effects multi-variate data reduction by transforming the data into orthogonal components, which are linear combinations of the original variables. These new variables are often referred to as latent variables or principal components (PCs). The transformation is achieved without loss of information because each PC accounts for a certain amount of data variance and the PCs are extracted in order such that PC1 accounts for largest amount of data variance, PC2 for the next largest amount and so on. The data reduction process often encapsulates all the significant information or data variance in just a few PCs instead of hundreds or even thousands of original variables (i.e. wavenumbers in the case of DRIFT spectra). Each spectrum or object has a score value on each of the new PC variables. For example, a PC1 versus PC2 score plot may be displayed, which allows any patterns and trends to be studied. Commonly, such a score plot accounts for most of the data variance. In addition, each original variable (i.e. a wavenumber) is

4

188 189 190 191 192 193 194 195

characterised by a loadings value, which expresses the importance of a variable on a particular PC for the discrimination of the objects. Hence, plots of loadings values for a given PC will indicate the wavenumber ranges that are principally responsible for the separation or discrimination of the objects or spectra on a given PC. Thus, a combination of the PC score/score plots and their corresponding loadings displays may provide progressively more information which is often unavailable by conventional data analysis.

196

2.5.3

197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236

PROMETHEE and GAIA

PROMETHEE and GAIA MCDM methods have been described in detail elsewhere 28-30; in this paper, a summary of these methods is provided in sufficient detail to indicate the concepts and data treatment involved. PROMETHEE is a non-parametric method used to rank a number of actions (spectral objects) based on performance criteria (wavenumber variables) imposed on the data matrix. For each criterion, a specific preference function, threshold value and weighting condition must be defined. The preference function is a mathematical function, which is used to calculate the degree of preference associated with each action. To set up a model for PROMETHEE, a preference function must be selected. In commercially available software (e.g. Decision Lab (Visual Decision 2000)), six shapes of preference functions are defined. In this software, they are called Usual, Linear, Level, V-shaped, U-shaped, and Gaussian preference functions 30. For most functions, one or two classification thresholds must be provided by the user. These establish how preferences are to be attributed in accordance with the functions. Minimised and maximised conditions are allocated for each criterion to establish the preferred ranking sense. Therefore, the ranking can be undertaken bottom up (minimised) or top down (maximised) depending on the decision-maker’s preferences. Weights can also be allocated for each criterion to reflect the importance of one criterion over another. By default, a weighting of 1 is assigned for all criteria. However, weights can be altered by the decision-maker if alternative scenarios are required in the investigation. The model is now set up according to the user’s chosen scenario, and the raw data matrix may be submitted for calculation. 1. 2. 3. 4. 5.

A summary of the PROMETHEE procedure is given below: The raw data matrix is transformed to a difference d matrix. Such a matrix is constructed from pairwise subtraction of the entries in all possible combinations for each criterion. For each criterion, a preference function p(a, b) is applied to determine how much an outcome a is preferred to b. The overall outcome is a preference index matrix. A global (or overall) preference index is calculated for each object by summing all preference indices for each criterion. Positive (Φ +) and negative (Φ -) outranking flows are calculated by summing all the global indices. Φ + describes how the action outranks all others while Φ – describes how an action is outperformed by all others. The outranking flows are compared pairwise according to a set of rules 30. These are based on three possible outcomes, which lead to partial pre-order of the objects known as PROMETHEE I ranking: a) one action is preferred to another;

5

237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263

b) there is no difference between the two actions; and c) the two actions can not be compared. 6. The net outranking flows (Φ = Φ + - Φ -) are calculated which excludes rule 5 (c) above, and results in a unidimensional (PROMETHEE II) ranking. Although it may be more convenient to use PROMETHEE II net ranking, some information does get lost in the process. This information is retained in the PROMETHEE I partial ranking, where incomparable objects (or alternatives) are displayed. 7. Typical partial preorder rankings from PROMETHEE I and net outranking flows from PROMETHEE II are diagrammatically illustrated in Figures 1a and 1b.

264

3

265

3.1

266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284

GAIA is a display method and is linked to the PROMETHEE procedure. The GAIA matrix is constructed from a decomposition of the Φ net outranking flows 30. The data is then processed by a PCA algorithm and displayed on a GAIA biplot, which shows a distribution of objects, criteria vectors and a decision axis, π. Such displays are interpreted conventionally as normal PCA biplots, and the decision axis π, displays the degree of decision power and points to the approximate location of the preferred action. GAIA provides some guidance for criteria, which are important for net outranking, and which criteria influence the decision axis. An important difference between the application of the GAIA and a conventional PCA is its facility to model scenarios based on the choice of individual preference functions for each criterion, the choice of ranking sense and the criteria weights. The choice of such model specifications will be reflected in the distribution of the criteria vectors on the resulting biplot. This facilitates the testing of different experimental hypotheses, providing the user with options to test different scenarios. A typical simulated biplot is shown in Figure 1c. Results and discussion DRIFT Spectra

The infrared spectra for the three regions are shown in Figures 2a, 2b and 2c. In general, the first region represents the bands assigned to the OSiO and OAlO bending region. The second region reflects the absorptions due to the SiO and AlO stretching region and the third region is due to the OH stretching region. These three regions reflect the effects of mechanochemical activation of the kaolinite and quartz samples with increase grinding time as monitored by DRIFT spectroscopy. It has been previously reported 12, 24 that the mechanochemical activation grinding results in an increase in mean lattice strain culminating in the breakdown of the crystal structure of kaolinite by effecting the breakage of O-H, AlOH, Al-O-Si, and Si-O bonds. This is reflected in the spectral line broadening and a reduction of peak intensities. The key features in each region are discussed below. In the three wavenumber regions, spectra taken at time zero ie K-0, Q1-0 and Q.75-0, respectively, reflect the spectral responses of the materials in their initial crystalline states. All of the other spectra were taken after some grinding - normally up to 4 hrs, and for quartz itself (Q1 samples) up to 10 hrs. Thus, on a relative basis the general effects of grinding may be observed by comparison of these spectra with

6

285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334

those taken at time zero. To assist with spectral interpretation some well established quartz and kaolinite band frequencies were included. General similarities in the overall spectral shape might be expected between K-0 and Q.75-0 samples because they are both mixtures of kaolinite and quartz, but in K-0 sample, the quartz is present naturally, while in the Q.75-0 it has been added approximately in the same amount. Careful comparison of the time zero spectra for the three samples over the three regions indicates that they are remarkably similar. However, there are also clearly distinguishable differences such as the apparent shift of the 540 cm-1 band (Q.75-0) to ca.570 cm-1 (K-0) where it remains even after 4 hours of grinding. By comparison with the Q.75-0 sample, the 540 cm-1 band shifts to ca. 520 cm-1. The kaolinite band at 701 cm-1 (K-0) appears to be missing entirely in the Q.75-0 spectrum. Such significant band shifts are not readily apparent for the Q.75-0 and K-0 spectra over the other two wavenumber ranges, although other significant differences are readily apparent. In general, conventional examination of the 400-830 cm-1 region, indicates few significant changes to bands above 600 cm-1 for the three types of samples. The major quartz bands at ca. 697, 780 and 798 cm-1 remain apparently unchanged but the weaker kaolinite peaks at ca. 644 (Al-O-Si ) and 755 cm-1 (Al-O-Si ) here is a significant band overlap of kaolinite and quartz bands (Figure 2a). It is difficult to separate the bands in the 450-700 cm-1, however quartz bands at 780 and 798 cm-1 are clearly visible in all the three samples. For the 830-2030 cm-1 region, there are significant changes in the both the kaolinite and quartz samples due to grinding (Figure 2b). In the Kiralyhegy (K) samples, there is a reduction in the kaolinite peak intensities (920, 938 and 1115 cm-1) with increase grinding time. This implies that changes are occurring in the kaolinite surfaces under mechanochemical activation. Similarly, the natural quartz in the K samples is also undergoing changes with broadening of the quartz bands (1150 and 1172 cm-1) due to grinding. In the quartz (Q1) samples, the most noticeable difference is the lack of kaolinite bands in the 920-940 cm-1 region. This is expected since Q1 contains over 90% natural quartz with minor impurities and no kaolinite. The broadening of the quartz peaks at 1150 and 1172 cm-1 show that the quartz in these samples undergoes changes as the result of grinding. In the Zettlitz (Q.75) samples, with 75% added quartz, the kaolinite bands at 920 and 938 cm-1 disappear after two hours of grinding. This implies that the kaolinite crystal structure is destroyed after two hours mechanochemical activation. In addition, it appears that the (added) quartz has accelerated the mechanochemically induced amorphisation of the kaolinite structure. The role of quartz as a grinding agent will be discussed further in the chemometrics section. In the 2600-3800 cm-1 region, the most significant changes to kaolinite structure due to grinding is reflected in OH stretching range (i.e. losing hydroxyl units, formation of water and changes to the surface structure). According to Farmer 31, the hydroxyl stretching region of kaolinite displays five key features: a) in-phase inner surface hydroxyl stretching vibration at 3695 cm-1; b) two out-of-phase vibrations of the inner surface hydroxyl at 3668 and 3652 cm-1; c) hydroxyl stretching vibration of the inner hydroxyl at 3620 cm-1; d) transverse longitudinal optic vibration at 3684 cm-1; and

7

335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358

e) water hydroxyl stretching vibration of weakly hydrogen bonded interstitial water at 3595 cm-1. The DRIFT spectra of Kiralyhegy (K) and Zettlitz (Q.75) samples exhibit these key features (Figure 2c). Mechanochemical treatment results in dehydroxylation of the kaolinite. This occurs in two processes: a) delamination of the kaolinite structure which exposes the inner surface hydroxyl groups to form water b) disintegration of the kaolinite crystal structure and the resulting reaggregation of the crystallites to form a new amorphous material 9, 32. With reference to Figure 2c, the two most significant spectral features are the reduction in peak intensities in the 3600 – 3700 cm-1 range with increase grinding time and the formation of the broad band in the 3000 – 3500 cm-1 range which is attributed to adsorbed and coordinated water 24. In the Kiralyhegy (K) samples, the inner surface and inner hydroxyls groups are being affected as a result of grinding (i.e. progressive reduction in the peak intensities of the spectra). In addition, a formation of the broad water band is visible after three hours grinding. In the quartz (Q1) samples, there are no kaolinite bands in the DRIFT spectra. Therefore, changes in the hydroxyl region are not reflected in the quartz samples. In contrast, the Zettlitz (Q.75) samples undergo significant changes as the result of grinding. The added quartz (75% quartz content) appears to accelerate the dehydroxylation process. After two hours grinding there is a significant reduction of the inner hydroxyls peaks. This reduction is more pronounced than those experienced for the Kiralyhegy samples where natural quartz (70%) is present.

359

3.2

360

3.2.1

361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382

Chemometrics Principal Components Analysis

PC1 versus PC2 scores plots for the three spectral regions are shown on Figures 3, 4 and 5, and each pair of PCs accounts for over 85% of the data variance. In fact, most of the differences in the spectra are accounted by just two to four PCs. In the 400-800 cm-1 range (Figure 3), three clusters of objects are apparent – the Zettlitz (Q.75) objects (positive scores on PC1 and PC2) form a fairly tight group; a more spread out Kiralyhegy (K) cluster is separated on PC2 (negative scores), and the narrow Q1 quartz group with objects spread out roughly parallel with PC1 (mostly –negative PC1 and low positive PC2 scores). Thus, the three groups of spectral objects are discriminated. Considering the projection of the spectral objects on PC1, each group shows roughly the same rank order as a function of grinding time. Thus, the unground spectral object of each series of samples has the highest PC1 score and, on the whole, the scores decrease with grinding time. However, in each set of objects, the spectrum corresponding to four hours grinding is offset and is found to be close to the unground and ground (one hour) objects. The objects are discriminated by high value loadings in the spectral ranges which contain infrared bands attributed to kaolinite: 410-440 cm-1 (SiO), 570-610 cm-1 (SiO), 648-690 cm-1 (SiO, and Si-O-Al) and 700-770 cm-1 (OH translation).

8

383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432

These observations indicate that grinding causes changes in the spectral regions reflecting bending modes of vibrations of kaolinites i.e. the lattice distortion of kaolinite in the Zettlitz (Q.75) samples is occurring in preference to any changes in the quartz. This may well be expected as quartz in this type of sample (Q.75) is added as a grinding agent. However, the situation is rather different when the Kiralyhegy (K) spectral objects are considered. In these samples (K), quartz is a natural component of the kaolinite deposit. The spectral object from the unground sample is similarly separated as the Zettlitz group (similar positive scores on PC1) but the ground (one hour sample) has similar quite low scores on this PC as the first two spectral objects of the quartz (Q1) series. These samples (Q1) do not contain any kaolinite. Thus, it is likely that the grinding process in the Kiralyhegy samples is affecting quartz in preference to the kaolinite mineral even at low grinding times. This proposition is supported when loadings negative on PC1 are considered. Clearly, the spectral objects of quartz (Q1) ground for long periods of time dominate the negative PC1 side. Reference to the PC1 loadings plot shows that the strong negative loadings responsible for the discrimination of the respective spectral objects are associated with changes to the infrared bands corresponding to the characteristic SiO bending modes of quartz. Apart from most of the quartz objects, two Kiralyhegy samples are also affected, and they continue more or less the trend which commences from the object representing the unground Kiralyhegy sample. This means that the nature of the quartz and/or its action (i.e. grinding mechanism) are different in the two kaolinite containing samples (Q.75 and K) as reflected by the DRIFTS criterion. This conclusion is further supported by the comparison of the relevant spectra in the 8302000 cm-1 region that will be discussed below but before this is considered, the information reflected on PC2 of the 400-830 cm-1 is investigated. Members of the quartz (Q1) series cluster very closely together on PC2 with quite low positive score values. The Zettlitz (Q.75) series is also positive on this PC and the lower members of this series have scores similar to those of the Q1 set. However, this latter series is roughly rank ordered along this PC with increase in grinding time. This indicates that the PC2 loadings plot with positive values should reflect changes both in kaolinite and quartz spectral regions. Reference to the spectra (Figure 2a) indicates that there are intense bands in the 400-550 cm-1 region reflecting molecular bending vibrations from the two different materials. There are spectral changes in the spectra of both the kaolinite and quartz in this region. Thus, the kaolinite SiO band at 540cm-1 almost disappears after about two to three hours grinding and leaving just the quartz band at 512 cm-1, and the band present at about 490 cm-1 in the Q1-0 sample of quartz changes on grinding with the development of a new strong band at about 470 cm-1. Since the quartz added to the Zettlitz kaolinite as a grinding agent is the same as that studied in the Q1 quartz series, the ordering of the spectral objects on PC2 cannot be attributed to changes in quartz as it is unlikely to behave differently from the quartz of the Q1 series. Thus, the rank ordering of these objects must be attributed to changes in kaolinite i.e. the lattice distortion reflected by the changes in the bending modes of vibration of the kaolinite bands, which culminate in the disappearance of this band (presumably the transformation of the crystalline kaolinite into an amorphous form). However, original discrimination of the quartz spectral

9

433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482

objects as well as the first few members of the Zettlitz group from the Kiralyhegy ones has to be attributed to the differences in quartz in these two samples. Apparently PC2 does not reflect the substantial changes in the quartz spectra at 470 cm-1otherwise the Q1 spectral objects will be rank ordered as well. It is interesting to note that the Kiralyhegy objects (positive on PC2) are rank ordered on PC2. The loadings plot clearly shows that the ordering is due to changes in both quartz and kaolinite with the loadings at approximately 540-570 cm-1 reflecting the kaolinite SiO modes of vibration, whereas the loadings at approximately 700 cm-1 as well as 773 and 788 cm-1 clearly correspond to the OH translation and the characteristic SiO stretching bands of quartz, respectively. Thus, in the Kiralyhegy sample, the grinding appears to affect both components. This would suggest that both the quartz and the kaolinite may be different in this sample as compared to the Zettlitz one. (Ray/Serge: may need to discuss order/disorder, Hinckley indices and crystallinity of kaolinites – what do you think?) In the 830-2030 cm-1 region, the spectra of the Zettlitz (Q.75) samples is roughly rank ordered according to grinding times on PC1 (positive scores) with the unground sample having the highest score (Figure 4). However, the positive PC1 loading are not informative about the specific vibrational bands that may be responsible for the discrimination of this series from the others. On the other hand, the negative PC1 loadings indicate that the roughly ordered members of the quartz (Q1) series are discriminated on the basis of high positive loadings in the 950- 1240 cm-1. These loadings correspond to the SiO stretch bands of quartz which undergo changes with increase in grinding time. These changes are particularly manifested in the disappearance of the band at 1170 cm-1 and the development of a band at 1100 cm1 . The 1170 cm-1 band is regarded as a characteristic of the SiO stretch in the three dimensional structure of quartz whereas the 1082 cm-1 band in quartz and the 1100 cm-1SiO stretch vibration in kaolinite are attributed to the Si-O-Si modes of vibration perpendicular to that at 1170 cm-1 . Thus, it would seem that with grinding, the distortion of the quartz lattice changes from a three dimensional structure to an increasingly two dimensional one. On PC2, the Zettlitz (Q.75) series has scores of almost zero and therefore does not contribute to this PC, but the quartz has low but significant positive score values with all spectral object forming more or less a point cluster. This behaviour is similar to that observed in the 400-830 cm-1 range discussed above except that in the latter case there was some interference from the Zettlitz spectral objects. Here the positive values of the PC2 loadings plot are quite definitive of the spectral features that discriminated the quartz (Q1) series from the Kiralyhegy (K) series. There is no contribution from the Zettlitz spectral objects. The particular features are the loadings at 1080, 1170 and 1240cm-1, all of which have been discussed above with respect to the grinding of quartz in the Q1 series. Here however, the distinction made is between the Q1 and K series quartz. In addition, the Kiralyhegy set of spectral objects is discriminated from the quartz on the basis of positive loadings on PC2 in the region of 900-1050 cm-1. The loadings maxima correspond to the Al-O-H bands of kaolinite as well as 1008 and 1032 cm-1 attributed to the Si-O-Al and Si-O-Si stretch bands in the same mineral. This point to severe lattice distortion that takes place during grinding involving not only the SiO modes vibration but also the AlOH sites and the bonds between SiO and Al. Comparison of the spectra of the Zettlitz and

10

483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532

the Kiralyhegy series shows quite clearly that the bands in the 900-1050 cm-1 region for the Zettlitz samples effectively disappear after four hours of grinding but they are still present for the Kiralyhegy samples. This supports our contention raised earlier that both the nature of the kaolinite and quartz are important during the grinding process. The PC1 versus PC2 scores plot (Figure 5) for the 2600-3800 cm-1 range shows the best correlation with respect to grinding time. Both the Zettlitz and Kiralyhegy samples are highly ordered on the PC1 and PC2 axes on the basis of grinding time. The objects positive on PC1 are discriminated by a broad band of positive loadings in the 3000-3570cm-1 range. Spectra in Figure 2c show growing broad bands in this region which are commonly attributed to the OH stretch of hydrogen-bonded water. The expanded version of this frequency range (Figure 6) presents the spectra for the samples ground for four hours, and shows that in all three spectra there is a distinct maximum at 3383cm-1 commonly attributed to the OH stretching vibrations of strongly H-bonded water. Also, in the two quartz and Zettlitz spectra there is a band maximum at 3550cm-1 commonly associated with OH stretch bands corresponding to weakly H-bonded water. Thus, it is reasonable to suggest that the Kiralyhegy spectral objects are rank ordered and discriminated from the quartz and Zettlitz samples on the basis just described. The spectral objects with negative PC1 scores are discriminated on the basis of a broad loadings band in the 26003000cm-1. This is difficult to assign specifically. However, it should be noted that as may be anticipated, the quartz spectral objects form a small cluster of samples that are not ranked according to the grinding time, but on the other hand the Zettlitz spectra are well ranked on the basis of the negative PC1 loadings. Even though it is difficult to assign the underlying spectral reasons for this case, it is clear that the phenomenon reflected in this part of the spectrum must be quite systematic and varies with grinding time. Otherwise, the Zettlitz objects could not be ranked. On PC2 the spectral objects with positive scores are discriminated by a broad band of loadings with a maximum at 3030 cm-1. It is difficult to find an exact spectral association with the frequency at which this loadings maximum occurs because there is no obvious infrared band apparent in the spectra. However, it is noted that this maximum corresponds to a frequency that is well noted in the broad bands associated with the OH stretch. Also, interestingly it must be derived principally from the kaolinite samples because the quartz series spectra form almost a point cluster and are not rank ordered. On the whole the kaolinite samples involved correspond to those that have been well ground and in the case of the Zettlitz samples almost completely changed into the amorphous form. This implies a significant progressive increase of surface with grinding and the possibility of binding more adsorbed water. It is proposed that there is a hidden infrared band that describes a state of water that is specifically adsorbed on the fine, amorphous particles of kaolinite during grinding. The spectral objects negative on PC2 principally refer to the spectra from Kiralyhegy unground and the two samples ground for short times. These three samples are separated on the basis of changes in the hydroxyl stretch region and clearly account for considerable variance on this PC. Since the spectral intensities of the Kiralyhegy and the Zettlitz samples quickly decrease and indeed in the case of the latter series almost disappear, it suggests that the inner surface and inner hydroxyls are significantly changed or destroyed.

11

533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575

3.2.2

PROMETHEE and GAIA

In the context of this work, PROMETHEE and GAIA were applied to provide additional information not readily available using the PCA technique alone. PROMETHEE allows the criteria to be modelled independently (e.g. grinding time) and does not require full comparability between the criteria. The mechnoactivation of kaolinite and quartz, through dry grinding, results in changes to the surface structure. Different grinding times were adopted to study the rate of destruction of the kaolinite and quartz structures. This relationship (i.e. grinding time) was classified using PROMETHEE and GAIA. As previously indicated with PROMETHEE, it is necessary to model each criterion independently. In this case, PCs were used as criteria to represent spectra in a PROMETHEE matrix with scores as the new data entries. The application of this approach is illustrated in the presented scenario below. One of the advantages of PROMETHEE is that it is a non-parametric method, which means that, in principle, it is possible to compare as few as two objects. Therefore, the 17 x 3 matrix consisted of seventeen objects (kaolinite and quartz samples) and three variables (PC1 and PC2 scores, and grinding time). The Linear Vshaped preference function was applied for all three criteria, and the PC variables were set to maximise and grinding time variable set to minimise. This setting indicates that high values for the PCs are preferred while for grinding time low values dominate. The weights for each of the three variables were set to 1. From the model described above, the spectral objects should align with the grinding time vector in the GAIA biplot, if there is a correlation between the spectral objects and the grinding time loadings variable. However, the samples used are inherently different, therefore other physico-chemical properties (as reflected by the spectral data) may interfere and shift the objects roughly parallel with the grinding time vector. It should be noted that the grinding time vector would still dominate in this scenario. A matrix of spectral objects (K, Q1, and Q.75), spectral PCs and grinding time loadings variables were submitted to PROMETHEE for analysis (Table II) and the results presented in Figure 7. The GAIA biplot (Figure 7, 84% variance described) displays the relationship between the spectral objects, spectral PCs and the grinding time vector. From Figure 7, Kiralyhegy (K) and Zettlitz (Q.75) series are roughly parallel with the grinding time vector. This supports the initial PCA observation (Figure 5) regarding an apparent trend with respect to grinding time and the K and Q.75 spectral objects. In addition, the GAIA also shows that the Q1 spectral objects are roughly aligned with the grinding time vector in this region. This trend was not observed in the scores plot (Figure 5).

12

576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624

4

Conclusions

In general, this study demonstrated that DRIFT spectra of two different kaolinites mechanochemically activated with quartz can be successfully compared and interpreted in respect to their grinding by the common chemometrics display method, PCA. In addition, it was shown how the kaolinites may be globally compared on the basis of their chemical composition and physical properties, and how their spectral characteristics may be included in such comparisons with the aid of the MCDM methods, PROMETHEE and GAIA. The band assignments of the DRIFT spectra of kaolinite samples mechanochemically ground with quartz originating were generally consistent with those found in the literature. The three spectral regions chosen for interpretation viz,. 400-800, 800-2030, 2600-3800 cm-1, were then analysed by the two different chemometrics approaches. In regard to PCA, for each of the spectral regions, a PC scores plot was analysed in some detail with the aid of loadings plots. The scores plots produced concise displays of all the spectral data and allowed a comparative examination of the spectra. In general, the clustering could be interpreted in terms of the known properties of kaolinite samples as well as the presence of natural quartz. The dominance of this component was a notable feature in the two PC scores plots for the first two spectral regions by the distinctive discrimination of the Kiralyhegy spectral objects. Importantly, while some ordering of the objects in some PC clusters was apparent, it was not according to the grinding time sequence. Thus, it would appear that neither the kaolinite nor the quartz, are systematically responsive to grinding time according to the spectral criteria. The third spectral region, attributed to the OH vibrations, was the only one to show apparent systematic ordering of the Kiralyhegy and, to a lesser extent, Zettlitz spectral objects with grinding time. These two series of objects displayed approximate linear trends in the PC spectral plane. This was attributed to the effect of the natural quartz on the delamination of kaolinite and in the case of the Kiralyhegy kaolinites and added quartz in the case of the Zettlitz kaolinte and the accompanying phenomena such as the formation of kaolinite spheres and water. The key observation was that the Kiralyhegy and Zettlitz objects formed roughly linear trends in parallel with the grinding time criterion vector.

Acknowledgements The support of the Hungarian Research Fund under Grant OTKA K62175 is gratefully acknowledged. The financial and infra-structure support of the Queensland University of Technology Inorganic Materials Research Program is gratefully acknowledged. The Australian Research Council (ARC) is thanked for funding the instrumentation.

13

625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674

References 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

R. D. Dragsdorf, H. E. Kissinger and A. T. Perkins, Soil Sci. 71, 439 (1951) J. B. Holt, I. B. Cutler and M. E. Wadsworth, Clays Clay Minerals, Proc. Natl. Conf. Clays Clay Minerals, 12th 55 (1964) M. L. Jackson and E. Truog, Soil Sci. Soc. Am. Proc. 4, 136 (1939) H. Takahashi, Bull. Chem. Soc. Japan 32, 381 (1959) H. Takahashi, Bull. Chem. Soc. Japan 32, 252 (1959) H. Takahashi, Bull. Chem. Soc. Japan 32, 245 (1959) H. Takahashi, Bull. Chem. Soc. Japan 32, 235 (1959) H. Takahashi, Clays, Clay Minerals. Proc. Natl. Conf. Clays, Clay Minerals, 6th, Berkeley 279 (1959) F. Gonzalez Garcia, M. T. Ruiz Abrio and M. Gonzalez Rodriguez, Clay Miner. 26, 549 (1991) F. Gonzalez Garcia, M. Gonzalez Rodriguez, C. Gonzalez Vilchez and M. Raigon Pichardo, Bol. Soc. Esp. Ceram. Vidrio 31, 297 (1992) K. Tsunematsu, H. Tateyama and K. Kimura, Shigen to Sozai 116, 19 (2000) R. L. Frost, E. Mako, J. Kristof, E. Horvath and J. T. Kloprogge, Langmuir 17, 4731 (2001) R. L. Frost, J. Kristof, J. T. Kloprogge and E. Horvath, Langmuir 17, 4067 (2001) E. Mako, R. L. Frost, J. Kristof and E. Horvath, Journal of Colloid and Interface Science 244, 359 (2001) I. S. Ismael, M. K. Abd El-Rahman and M. S. Hassan, Int. J. Soc. Mater. Eng. Resour. 7, 339 (1999) R. M. T. Sanchez, E. I. Basaldella and J. F. Marco, J. Colloid Interface Sci. 215, 339 (1999) I. D. R. Mackinnon, P. J. R. Uwins, A. J. E. Yago and J. G. Thompson, Clays Controlling Environ., Proc. Int. Clay Conf., 10th 196 (1995) A. K. Bandopadhyay, D. G. Bharathi, S. Maitra, S. H. Ansari, S. Mitra and R. Sen, Fuel Sci. Technol. 16, 115 (1997) J. Kristof, R. L. Frost, J. T. Kloprogge, E. Horvath and E. Mako, Journal of Thermal Analysis and Calorimetry 69, 77 (2002) S. Yariv, Powder Technol. 12, 131 (1975) S. Yariv, A. Nasser, K. H. Michaelian, I. Lapides, Y. Deutsch and N. Lahav, Thermochim. Acta 234, 275 (1994) S. Yariv and S. Shoval, Clays Clay Miner. 24, 253 (1976) S. Yariv, J. Chem. Soc., Faraday Trans. 1 71, 674 (1975) R. L. Frost, J. Kristof, E. Mako and W. N. Martens, Langmuir 18, 6491 (2002) O. Carmody, J. Kristof, L. Frost Ray, E. Mako, J. T. Kloprogge and S. Kokot, Journal of colloid and interface science 287, 43 (2005) R. Schrader, and Dusdrof, W., Kristall und Technik 1, 59 (1966) R. Schrader, Silikattechnik 21, 196 (1970) J. P. Brans, B. Mareschal and P. Vincke, European Journal Operational Research 24, 228 (1986) J. P. Brans and P. Vincke, Management Science 31, 647 (1985) H. R. Keller, D. L. Massart and J. P. Brans, Chemometrics and Intelligent Laboratory Systems 11, 175 (1991) V. C. Farmer, Clay Miner. 33, 601 (1998)

14

675 676

32.

Z. Juhasz, Acta Mineral.-Petrogr. 24, 121 (1980)

15

677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716

List of Tables Table I. Chemical composition and physical parameters of kaolinite and quartz samples used in the study. Table II. Matrix submitted to the PROMETHEE analysis to determine a relationship between grinding time and changes in kaolinite structure (2600-3800 cm-1 region).

List of Figures FIG 1. (a) PROMETHEE 1 partial ranking. A3 is the most preferred object followed by A2 and A1 (which are alternatives and can not be compared), and A4 is least preferred. (b) PROMETHEE II net ranking. The incomparable option 5c) has been removed and the objects are net ranked. (c) The GAIA plot – π points in the direction of the preferred object (A3) and is strongly correlated to vector C1 and, to a lesser extent, vector C2. The other vectors point to objects which perform best on the respective criteria. FIG 2. Infrared spectra of kaolinite and quartz samples in (a) 400-830 cm-1 region (b) 830-2030 cm-1 region (c) and 2600-3700 cm-1 region. Note: Kiralyhegy (K), >90% Quartz (Q1) and 25% Zettlitz kaolinite and 75% Quartz (Q.75) and number beside label denotes grinding time (h). FIG 3. (a) Scores plot, (b) PC1 loadings plot and (c) PC2 loadings plot for 400-830 cm-1 region. FIG 4. (a) Scores plot, (b) PC1 loadings plot and (c) PC2 loadings plot for 830-2030 cm-1 region. FIG 5. (a) Scores plot, (b) PC1 loadings plot and (c) PC2 loadings plot for 2600-3800 cm-1 region FIG 6. Infrared spectra of kaolinite in the OH stretching region. FIG 7. GAIA biplot of kaolinite and quartz samples, grinding time (GT) and PC1 and PC2 scores.

16

717 718

719 720 721 722

Table I. Sample

Natural Sand (Q1) wt%

SiO2 Al2O3 Fe2O3 CaO MgO K2O Na2O TiO2 Loss on Ignition (LOI) Kaolinite Quartz Illite Feldspar Corundum Hinckley Index (HI)

91.97 5.65 0.14 0.2 1.11 0.12 0.02 0.35 92 6 4 ???

Kiralyhedgy kaolinite (K) wt% 81.59 12.1 0.07 0.35 0.51 0.05 0.03 5.1 33 67 1.4

Table II. Sample Q.75-0 Q.75-.5 Q.75-1 Q.75-2 Q.75-3 Q.75-4 Q.1-0 Q1-1 Q1-2 Q1-4 Q1-6 Q1-10 K-0 K-1 K-2 K-3 K-4

PC1 Scores 0.14 5.59 2.53 8.88 10.54 14.42 3.85 0.00 3.34 4.56 1.76 2.30 14.94 19.44 24.37 39.14 42.43

PC2 Scores 24.80 22.79 28.52 32.14 33.95 35.66 32.24 31.36 31.96 32.22 31.67 31.54 0.00 15.41 25.83 30.82 37.03

723

17

Grinding Time 0 .5 1 2 3 4 0 1 2 4 6 10 0 1 2 3 4

Zettlitz kaolinite (Q.75) wt% 46.97 36.32 0.37 0.54 0.26 1.21 0.05 12.9 92 4 (+75% quartz) 4 0.7

724 725 726 727

18

728 729 730 731 732 733 734 735 736 737 738

Figure 1a, b, and c

Least preferred

Most preferred

A3 Φ

A2 Φ

A1 Φ

A4 Φ

Most preferred

Least preferred

A3 Φ

A3

A2 Φ

π

A4 Φ

A1 Φ

A2

C1 C3 C2

C4 A1

A4

C5

19

739 740

Figure 2a

741 742

20

K1033/1073 SiO stretch

K1117 SiO stretch Q1150/1172 SiOSi stretch

Figure 2b)

K920 OH deformation K938 inner surface OH bending

743 744

K-4

Reflectance

K-3 K-2 K-1 K-0

Q1-10 Q1-6 Q1-4 Q1-2 Q1-1 Q1-0

Q.75-4 Q.75-3 Q.75-2 Q.75-1 Q.75-.5 Q.75-0

830

1030

1230

1430

1630

Wavenumbers / cm-1 745 746

21

1830

2030

K3652/3669/3697 inner surface OH

K3620 inner OH

Figure 2c

K-4 K-3 K-2 K-1

Reflectance

747 748 749

K-0

Q1-10 Q1-6 Q1-4 Q1-2 Q1-1 Q1-0

Q.75-4 Q.75-3 Q.75-2 Q.75-1 Q.75-.5 Q.75-0

2600

2800

3000

3200

3400

Wavenumbers / cm-1

22

3600

3800

Figure 3 (a)

PC2 (32%)

15

10 Q1-10

Q1-2

5 Q1-1

Q1-4

Q1-6

Q.75-3

Q1-0

Q.75-.5

Q.75-2 Q.75-4 Q.75-1 Q.75-0

0 -25

-20

-15

-10

-5

0

5

10

15

20

PC1 (54%)

-5 K-4 K-2 -10 K-3 -15 K-1

K-0

-20

(b) PC1 K410

K570 Q470

K610

K648

Q490 Q512

K710 Q700

K750 Q785

Q800

(c) PC2 K440

K474

K625 Q540 Q570

K725 Q700

Q770

Q790

Loadings

750 751 752 753

400

500

600

Wavenumbers / cm-1 23

700

800

Figure 4

754 755 756

(a)

PC2 (17%)

10 Q1-6

Q1-0 Q1-2

Q1-10

Q1-1

5

Q1-4

Q.75-3 0 -30

-20

-10

Q.75-4 Q.75-2

Q.75-.5 10

0

Q.75-1 Q.75-0 20

30

PC1 (78%) K-2

-5 K-4

K-3

K-1 -10 K-0 -15

(b) PC1 Q1080 Q1150

Q1184

(c) PC2 Q1080 K938

Q1240

Loadings

K914

830

1030

1230

1430

Wavenumbers / cm-1 24

1630

1830

2030

757 758 759 760

Figure 5 (a) 15

PC2 (32%) Q.75 trend

K trend

10 Q.75-3 Q1-0 5 Q1-6 Q1-4 Q1-1 Q.75-2 Q1-10Q1-2 Q.75-1 0 -15 -10 -5 0 Q.75-0 -5 Q.75-.5

K-4

Q.75-4 K-3 5

10

15

20

25

30

K-2

35

PC1 (68%)

-10 K-1 -15 -20 -25 -30

(b) PC1

K-0

3383 strong OH bonding

3550 weak OH bonding

(c) PC2

K3693

Loadings

K3619

2600

2800

3000

3200 25 Wavenumbers / cm

3400 -1

3600

3800

761 762

Figure 6

3383

Kaolinite OHstretching absorption bands

3550

3619

Reflectance

3653 3693

K-4 Q1-4

Q.75-4

2600

2800

3000

3200

3400

Wavenumbers / cm-1 763 764

26

3600

3800

765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814

Figure 7.

Q1-10 PC2 Q1-6 Q1-4 Q.75-4 Q.75-3 Q1-2 Q1-1 Q.75-2 Q.75-1 Q1-0 Q.75-0 Q.75-.5 K2 K1

K4 K3

π PC1

GT

K0

Δ = 84%

27