Principal Component Analysis of X-ray Diffraction Patterns To Yield

Jan 9, 2007 - Principal Component Analysis of X-ray Diffraction Patterns To Yield ... for the Detection and Identification of Chemical Weapons and Exp...
0 downloads 0 Views 355KB Size
Download PDF
Anal. Chem. 2007, 79, 2091-2095

Principal Component Analysis of X-ray Diffraction Patterns To Yield Morphological Classification of Brucite Particles Charlene R. S. Matos, Maria Jose´ Xavier, Ledjane S. Barreto, Nivan B. Costa, Jr., and Iara F. Gimenez*

Departamento de Quı´mica, Universidade Federal de Sergipe (UFS), Av. Marechal Rondon s/n, Campus Universita´ rio Prof. Jose´ Aloı´sio de Campos, CEP 491000-000, Sa˜o Cristo´ va˜o SE, Brasil

Principal component analysis was applied to XRD data from a series of Mg(OH)2 samples prepared under different hydrothermal conditions from bischofite (MgCl2· 6H2O) and carnallite (KCl.MgCl2·6H2O), owing to differences in full width at half-maximum (fwhm) as well as in the intensity ratio I001/I101 of the respective diffraction peaks. According to the PCA results, the four principal components are able to explain 93% of the total variance and the samples can be classified into four main groups. For instance, the principal component PC1 can be interpreted as the crystallite size along the 101 direction since it is related to the fwhm of this peak. On the other hand, PC3 is related to orientation effects along 001 and 101 directions as it is dominated by the relative intensities of the two peaks. Finally, a comparison of the scanning electron microscopy images of the samples classified in each group revealed that in most of the cases a distinct morphology predominates within each group, which can be explained on the basis of the brucite growth mechanism. The increasing interest in the application of brucite (crystalline Mg(OH)2) as a flame retardant filler in polymer formulations is motivated by advantages of this material over halogenated flame retardants, which include environmental and health compatibility.1,2 The flame retardancy effect from brucite was found to be strongly dependent on the particle morphology,3 and in consequence, a great number of works described the study of morphological control of brucite particles.4-6 Alternatively, brucite is widely employed as a ceramic precursor and also for this application the particle morphology plays a key role in the solidstate process at high temperatures.7 Natural brucite can be found * To whom correspondence should be addressed. E-mail: [email protected]. Fax: (+55) 79 3212 6651. (1) Hornsby, P. R. Int. Mater. Rev. 2001, 46, 199-210. (2) Zhang, J. G.; Wilkie, C. A. ACS Symp. Ser. 2006, No. 92, 61-74. (3) Jiao, C. M.; Wang, Z. Z.; Ye, Z.; Hu, Y.; Fan, W. C. J. Fire Sci. 2006, 24, 47-64. (4) Hsu, J. P.; Nacu, A. Colloids Surf. A 2005, 262, 220-230. (5) Wu, H.; Shao, M.; Gu, J.; Wei, X. Mater. Lett. 2004, 58, 2166-2169. (6) Yan, C. L.; Xue, D. F.; Zou, L. J.; Yan, X. X.; Wang, W. J. Cryst. Growth 2005, 282, 448-454. (7) Mel’gunov, M. S.; Fenelonov, V. B.; Mel’gunova, E. A.; Bedilo, A. F.; Klabunde, K. J. J. Phys. Chem. B 2003, 107, 2427-2434. 10.1021/ac061991n CCC: $37.00 Published on Web 01/09/2007

© 2007 American Chemical Society

as hexagonal plates8 but also may exhibit a fibrous habit,9 although the former is the preferred morphology since it is related to the crystal structure. Unusual morphologies such as needle-like,10 nanotubes,11 nanorods,12 and globular aggregates8 were also previously described by the use of hydrothermal and solvothermal methods. Several magnesium precursors such as Mg, MgSO4, Mg(NO3)2,13 and Mg10(OH)18Cl2·5H2O14 among others were used for the precipitation of brucite under basic conditions, but to our knowledge, no report of the use of carnallite (KCl‚MgCl2·6H2O) can be found in the literature. A possible advantage of this precursor is to increase the concentration of cations in the medium, which favors the morphological control at pH values above the isoelectronic point of brucite (pH 12).8 The main reason for this is that a relatively slow diffusion of crystal building blocks usually favors crystal growth over nucleation, resulting in wellformed crystals. In general, the particle morphology is analyzed by scanning and transmission electron microscopy techniques. Microstructural features such as shape, size, and agglomeration are very important due to their influence on several desired properties. In addition, the results obtained from other characterization techniques may also be strongly influenced by the morphology. For instance, in the case of materials with anisotropic crystal structures the powder X-ray diffraction (XRD) patterns can be very sensitive to particle morphology since it may reflect orientation effects of the crystallites.15 The relationship between changes in XRD patterns and particle morphology as observed from electron microscopy or other direct technique is a question to be addressed to each particular structure, and no general statements can be proposed to extend the conclusions to other materials. In this case, the application of (8) Henrist, C.; Matthieu, J.-P.; Vogels, C.; Rulmont, A.; Cloots, R. J. Cryst. Growth 2003, 249, 321-330. (9) Ren, Q. L.; Bin, L.; Chen, S. T. Rare Metal Mater. Eng. 2004, 33, 47-50. (10) Lv, J.; Qiu, L.; Qu, B. Nanotechnology 2004, 15, 1576-1581. (11) Fan, W.; Sun, S.; Song, X.; Zhang, W.; Yu, H.; Tan, X.; Cao, G. J. Solid State Chem. 2004, 177, 2329-2338. (12) Li, Y.; Sui, M.; Ding, Y.; Zhang, G.; Zhuang, J.; Wang, C. Adv. Mater. 2000, 12, 818-821. (13) Ding, Y.; Zhang, G.; Wu, H.; Hai, B.; Wang, L.; Qian, Y. Chem. Mater. 2001, 13, 435-440. (14) Yan, L.; Zhuang, J.; Sun, X.; Deng, Z.; Li, Y. Mater. Chem. Phys. 2002, 76, 119-122. (15) Varma, H. K.; Sureshbabu, S. Mater. Lett. 2001, 49, 83-85.

Analytical Chemistry, Vol. 79, No. 5, March 1, 2007 2091

a multivariate approach represents the best procedure for identifying trends and behaviors since a large number of data can be acquired with complex correlation patterns.16 The interest in establishing those relationships is not only to achieve a better understanding of the differences in XRD patterns but also to provide an indirect route to address the brucite morphological changes. Common reports on the application of principal component analysis (PCA) to the classification of powder XRD data are related to the identification of polymorphs in a series of related compounds17-19 and to the classification of archaeological samples with respect to the presence of oxides and clay minerals.20,21 On the other hand, recent reports on the application of combinatorial and high-throughput techniques to materials science using an XRD data set were based on the sensivity of the technique toward different crystal structures. This approach proved to be a powerful tool to correlate the occurrence of new zeolite and catalyst phases with synthesis conditions and also to construct QSPR models to predict structural and activity properties.22,23 In this work, PCA was used as an approach to classify a series of brucite samples (all representing a single polymorph) prepared from bischofite and carnallite under hydrothermal conditions on the basis of XRD data and then to verify if this classification reflects the particle morphology. EXPERIMENTAL SECTION Materials. In this work, reagent grade bischofite (MgCl2·6H2O; Merck), natural carnallite (KCl‚MgCl2·6H2O) from Sergipe State (Brazil), and “synthetic carnallite”sa mechanical mixture of 1:1 molar proportion of reagent grade KCl (Merck) and MgCl2·6H2Os were used as magnesium precursors. Natural carnallite was characterized by XRD, XRF, and atomic absorption spectroscopy prior to use (results not shown). Preparation. The Mg(OH)2 samples were prepared hydrothermally or solvothermally in Teflon-lined autoclaves of 120-mL capacity. Briefly, in each preparation, 0.01 mol of dry magnesium precursor was placed in the autoclave along with 90 mL of the solvent (0,1 mol L-1 aqueous NaOH (-), 0.1 mol L-1 aqueous NH4OH (+), or ethylenediamine) and heated at 110 (-) or 180 °C (+) for 8 (-) or 24 h (+), according to a factorial design (Table 1). After cooling to room temperature, the powdered brucite was collected by centrifugation, washed with distilled water and anhydrous ethyl alcohol, and dried under vacuum for 24,h. Characterization. The samples were characterized by X-ray diffractometry using Cu KR (λ ) 1.540 60 Å) on a Rigaku instrument, with 40 mA, 40 kV, and scanning rate of 3°/min. (16) Ferreira, M. M. C.; Antunes, A. M.; Melgo, M. S.; Volpe, P. L. O. Quim. Nova 1999, 22, 724-731. (17) Po ¨lla¨nen, K.; Ha¨kkinen, A.; Huhtanen, M.; Reinikainen, S.-P.; Karjalainen, M.; Rantanen, J.; Louhi-Kultanen, M.; Nystro¨m, L. Anal. Chim. Acta 2005, 544, 108-117. (18) Norrman, M.; Stahl, K.; Schluckebiera, G.; Al-Karadaghi, S. J. Appl. Crystallogr. 2006, 39, 391-400. (19) Barr, G.; Dong, W.; Gilmore, C. J. J. App. Crystallogr. 2004, 37, 243-252. (20) Gimenez, R. G.; Villa, R. V.; Dominguez, M. D. P.; Rucandio, M. I. Talanta 2006, 68, 1236-1246. (21) Villa, R. V.; Gimenez, R. G.; Dominguez, M. D. P.; Rucandio, M. I. Microchim. Acta 2003, 142, 115-122. (22) Corma, A.; Moliner, M.; Serra, J. M.; Serna, P.; Dı´az-Caban ˜as, M. J.; Baumes, L. A. Chem. Mater. 2006, 18, 3287-3296. (23) Corma, A.; Serra, J. M.; Moliner, M. J. Catal. 2005, 232, 335-341.

2092 Analytical Chemistry, Vol. 79, No. 5, March 1, 2007

Table 1. Values Used for the Variables Studied in the Experimental Design sample 1 2 3 4 5 6 7 8 9 10 11 12

base

temperature

time

+ + + + ethylenediamine ethylenediamine ethylenediamine ethylenediamine

+ + + + + +

+ + + + + +

Scanning electron microscopy (JEOL JSM T-300) characterized the morphology of the samples. Principal Component Analysis. The PCA transforms the correlated data set to a smaller set of variables, the principal components (PCs), which are uncorrelated and contain nearly all of the original information. The first principal component is the linear combination with maximal variance; the second principal component is the linear combination essentially searching for a dimension along which the maximum separation of the observations takes place in a direction orthogonal to the first principal component, and so on. Loading plots allow identification of the more important variables, and from score plots, the sample cluster can be identified and it may reveal the presence of outliers in the data. In this study, all intensity data points from XRD in the 2θ range 15-75° (step size 0.2°) were used and the intensity was normalized such that the maximum in each XRD pattern is one. The PCA was carried out on the data set centered on the respective variable, but not scaled by unit variance since the variables are all in the same unit.24 Calculation were performed by a program wrote in Fortran and by the software Statistica 6.0.25 RESULTS AND DISCUSSION In this work, we aimed to obtain brucite (Mg(OH)2) with controlled morphological features from bischofite (MgCl2·6H2O) and carnalite (KCl‚MgCl2·6H2O) under hydrothermal conditions using NaOH and NH4OH as bases and also solvothermally using ethylenediamine. The efect of the experimental variables on the growth of brucite particles has been described in previous reports.8,13 It is generally assumed that formation of individual lamellae can take place either prior to stacking along the c-axis or simultaneously to this. As described by Jin and co-workers,26 the rate at which the octahedral Mg(OH)64- building blocks reach the crystal surface and the nucleation rate (which gives a measure of the formation of the building block itself) are key factors to the final morphology. From a theoretical study of formation of fibrous brucite in nature, it was concluded that the growth direction of the Mg(OH)2 units tend to favor one-dimensional direction, parallel to the basic brucite (24) Glen, W. G.; Dunn III, W. J.; Scott, D. R. Tetrahedron Comput. Methodol. 1989, 2, 349-376. (25) Statistica version 6.0; StatSoft, Inc.: Tulsa, OK, 2004. (26) Xiang, L.; Jin, Y. C.; Jin, Y. Chin. J. Inorg. Chem. 2003, 19, 837-842.

Figure 1. Examples of powder X-ray diffraction patterns of the brucite samples prepared from natural carnallite (KCl‚MgCl2·6H2O) (a), from bischofite (MgCl2·6H2O) (b), and from synthetic carnallite (KCl + MgCl2·6H2O) (c).

layer, under natural hydrothermal conditions a thousand years ago.9 For a given experimental condition, the final particle morphology is determined by the relative stability of the crystal growth on the different directions of deposition of growth units on the crystal surface. According to previous reports, some significant variables during brucite growth under hydrothermal conditions are the solution pH, temperature, aging time, nature of the base, and concentration of ions in the medium.27,28 High pH values create a high-level supersaturation due to high OH- concentration resulting in a very fast nucleation of Mg(OH)64- building blockssprecursor to tiny and undefined nuclei. In order to decrease the surface energy, the small isotropic particles tend to aggregate, adopting an irregular shape in a process that is more pronounced at high temperatures. In addition, as the isoelectric point for brucite was determined by several authors as 12,29 above this value, the growing particles are negatively charged. In the case of samples prepared with NaOH as the base, the pH of the medium reached 13; so under this condition, the nature and concentration of cations present in the medium play a significant role. As cations are attracted to the negative surfaces, they hinder the income of the building blocks, thus decreasing the growth rate.8,13 Whether this cation adsorption occurs isotropically or not will influence directly on the final particle shape since it may favor specific growth directions. From powder XRD results, Figure 1, we observed that although all samples are undoubtedly identified as brucite, the full width at half-maximum (fwhm) and intensity ratio of some peaks in the XRD patterns varied significantly along the samples. The complete XRD results are presented in Figure S1 of Supporting Information. According to the JCPDS file for brucite,30 the intensity ratio between the first two peaks (I001/I101) is expected to be 0.53, but for some samples, it was found to be near 4. It is well described

that the intensity profile of XRD patterns can be modified by oriented crystal growth.31 This effect is highly pronounced for layered solids, and it is described that the high pressures found in hydrothermal treatment may restrict the growth direction.32,33 In general, the samples prepared here under hydrothermal conditions present high crystallinity with sharp and intense peaks in contrast to those prepared solvothermally, which present slightly broadened peaks indicative of reduced crystallite dimensions. Previous studies of layered double hydroxides in the literature proposed that examination of crystallite dimensions in the crystallographic directions [001] and [110] (Table 2, denoted as Dhkl) calculated with the Scherrer equation may give a measure of orientation effects.28 Note that the 110 peak is the relatively weak one observed around 2θ ) 58.7°. However, a better estimate of orientation effects can be given by the comparison of the number of crystal planes in each direction calculated by the ratio between the crystallite dimension and Bragg’s interplanar spacing. These values are presented in Table 2, where P relates to samples prepared from bischofite, M to samples prepared from synthetic carnallite, and C indicates that the Mg precursor was natural carnallite. From samples 1 to 8, the trends are very similar irrespective of the Mg precursor, which were prepared by a hydrothermal route. On the other hand, from samples 9 to 12, a specific behavior is observed for each precursor, which indicates a more pronouced influence of this variable in the solvothermal process. Clearly some experimental conditions favor the orientation effect in the 110 direction such as those used for samples 4 and 9-12. All these samples present particles with irregular morphologies, as will be discussed later in this work. Since the PCA is a method of reducing the number of variables in a correlated data set with minimal lost of variance, it was applied to our problem for an easier visualization of the data set and to clarify the connections among groups of samples. In comparison

(27) Kloprogge, J. T.; Hickey, L.; Frost, R. L. J. Solid State Chem. 2004, 177, 4047-4057. (28) Xu, Z. P.; Zeng, H. C. Chem. Mater. 2001, 13, 4555-4563. (29) Phillips, V. A.; Kolbe, J. L.; Opperhauser, H. J. Cryst. Growth 1977, 41, 228-234. (30) JCPDS (Joint Committee on Powder Diffraction Standards) standard 7-239.

(31) Tian, Z. R.; Voigt, J. A.; Liu, J.; Mckenzie, B.; Mcdermott, M. J.; Rodriguez, M. A.; Konishi, H.; Xu, H. Nat. Mater. 2003, 2, 821-826. (32) Zhao, Y.; Li, F.; Zhang, R.; Evans, D. G.; Duan, X. Chem. Mater. 2002, 14, 4286-4291. (33) Tao, Q.; Zhang, Y.; Zhang, X.; Yuan, P.; He, H. J. Solid State Chem. 2006, 179, 708-715.

Analytical Chemistry, Vol. 79, No. 5, March 1, 2007

2093

Table 2. Crystallite Size (D) and the Ratio of the Number of Layers along the Directions 001/110 from XRD Dataa D001 (nm)

D110 (nm)

N layers (001/110)

sample

P

M

C

P

M

C

P

M

C

1 2 3 4 5 6 7 8 9 10 11 12

34.664 36.490 40.781 21.000 40.781 46.218 36.488 25.677 10.347 11.750 18.737 16.123

41.326 39.037 27.022 21.296 50.190 41.331 46.832 54.039 14.335 19.511 12.550 20.662

33.463 36.981 31.931 21.287 36.978 43.908 35.125 35.133 30.542 33.454 29.270 17.563

44.195 46.772 46.772 35.775 41.857 44.169 44.174 39.788 30.597 31.802 34.567 37.859

44.182 46.809 37.874 37.904 44.204 39.776 46.772 46.772 31.802 30.582 34.618 37.871

44.213 46.790 49.681 36.139 44.195 46.781 44.169 46.818 44.174 46.781 37.859 39.764

3.86 3.88 3.47 4.97 3.11 2.90 3.70 4.70 8.97 8.25 5.62 7.14

3.97 3.81 4.71 5.15 3.61 3.23 3.81 4.02 4.39 4.24 3.92 6.86

3.23 3.61 4.24 5.34 2.65 2.90 3.03 2.62 6.75 4.77 8.25 5.55

a In the columns, P refers to samples prepared from bischofite (MgCl ·6H O), M indicates samples prepared from synthetic carnallite (equimolar 2 2 mixtures of KCl + MgCl2·6H2O), and C indicates samples prepared from natural carnallite (KCl‚MgCl2·6H2O).

to the usual procedure, the chemometric analysis of XRD data can be more straightforward and practical due to the large number of original data. In the PCA of the samples, the first four principal components were able to describe nearly 93% of the total information in the set data. The loadings plots (Figure 2a) indicate that PC1 can be interpreted as a crystallite dimension of the samples since it reflects the width especially of the second peak. Thus, the broader the second peak, the more positive is the sample’s score on the first component. On the other hand, PC3 is dominated by the height of the first and second peaks. The higher the first peak of a given sample, the more positive should be its score on the PC3, while the higher the second peak, the more negative should be the value on the PC3. The score plot in Figure 2b shows the existence of four groups and P4 as an outlier. A possible explanation for the behavior of P4 can be proposed based on the XRD pattern obtained for this sample, in which the I001/I101 intensity ratio reached 4, while the value expected from the crystallographic file is only 0.53. The resulting groups were afterward compared on the basis of particle morphology from scanning electron microscopy (SEM) images to test for specific trends within the groups. A detailed examination of SEM images revealed that the classification from XRD data corresponds in 92% of the cases to the different morphological types. All SEM images can be observed in Figures S2-5 of Supporting Information. A clear distinction in the morphologies between different groups can be made. Group 1: Predominance of Particles with Morphology of Regular Hexagonal Plates (Figure 3a). This group contains samples prepared using NaOH as base and combinations of low temperature with long aging times (M2, C2) or high temperatures combined with short times (C3, P3). Also samples prepared using NH4OH are present, but in this case, almost all of them were prepared using synthetic carnallite as precursor (M6, M7, M8, P6), which contain K+ cations as determinant of low growth rates. In this group, the typical morphology is characterized by the presence of well-defined lamellae, which indicates a low growth rate in the initial stages followed by growth in the edge direction. The particles become more and more regular with increase in the aging time. Group 2: Predominance of Thickened Particles with Round Borders (Figure 3b). This group includes predominantly 2094 Analytical Chemistry, Vol. 79, No. 5, March 1, 2007

Figure 2. (a) Loading plot representing the contribution of the original variables for the principal components PC1 and PC3. As variables that contribute most are plotted around the borders of the plot, the inset shows a typical XRD pattern, where one can see that the values 37.56 and 38.28 (representing the broadness of peak 2) contribute to PC1 and the value 18,6 (representing the intensity of peak 1) contribute to PC3; (b) score plots showing the existence of four main groups with P4 as an outlier. In the plot, we see two largest components (PC1, horizontal, and PC3, vertical), and the scores t1 and t3 are new variables computed as linear combinations of all variables (intensities as a function of 2θ) and aim at describing as much of the original variation as possible without losing information.

Figure 3. SEM images illustrating the typical morphologies of brucite samples in different groups. (a) M2 (group1); (b) C7 (group 2); (c) P9 (group 3); (d) M3 (group 4).

samples prepared using NH4OH as base and natural carnallite and bischofite as Mg2+ precursor (C5, C5, C7, C8, P5, P7, P8).There are, however, a few samples prepared using NaOH, all of them with short aging times (C1, P1, M1). All these factors contribute to a high nucleation rate due to the reduced cation adsorption and provide the formation of tiny nuclei, which further aggregate into irregular particles. A large proportion of NaCl was detected in the natural carnallite used implying a reduced Mg2+ supply, which accounts for the reduced size of the aggregates. Small flat particles with round borders are evidence that the particle growth was interrupted in the initial stages. Group 3: Predominance of Large Aggregates of Tiny Crystallites (Figure 3c). In this group, there are only samples prepared using ethylenediamine as the solvent, and this group can be thought as a subgroup of group 4. As previously described, ethylenediamine acts as a coordinating solvent for Mg2+ ions, thus reducing the deposition rate of building blocks. In this study, the absence of water in the medium, except for the water molecules in the precursors MgCl2·6H2O or KCl‚ MgCl2·6H2O in the medium caused the formation of tiny brucite particles, which aggregated into large particles. Group 4: Predominance of Very Large Aggregates with Completely Irregular Shapes (Figure 3d). Note that we included P4 in this group only due to morphological similarity. This group includes predominantly samples prepared using combinations of high temperatures, long aging times, and NaOH as base (M3, C4, P4, M4) as well as samples prepared using ethylenediamine as solvent (M9, M10, C11, P11, C12, P12, M12). The ethylenediamine effect has been discussed in group 3. The effect of the combination of high temperatures, long times, and strong base probably involves indiscriminate growth in every direction along with aggregation, since the solubility of magnesium hydroxide decreases when increasing the temperature.

Considering that brucite is a candidate for substitution of halogenated flame retardants in the polymer industry, the possibility of use of a routine analysis such as XRD can be more suitable for the assessment of morphological aspects, which are fundamental for this application. To our knowledge, there is only a few investigations on the potential of XRD with a multivariate approach, and we recognize that brucite is a special case of layered solid for which XRD may provide indirect information on morphology. CONCLUSIONS From this study, we conclude that XRD patterns of brucite particles with different morphologies reflect these differences in a way that can be useful for classification in groups according to the predominant morphology. Clearly this observation cannot be extended directly to other solid phases because it depends on the specificities of each crystal structure. For brucite, PCA of XRD data evidenced that two principal components PC1 and PC3, i.e., the broadness of the 101 peak as well as the I001/I101 intensity ratio, are the parameters that suffer the most significant influence of the crystallite orientation, which in turn influences the final morphology. ACKNOWLEDGMENT The authors are grateful to Prof. A. E. Almeida Paixa˜o for the use of Statistica 6.0, to CNPq, to FAPITEC for financial support, and to LQES-IQ-Unicamp. SUPPORTING INFORMATION AVAILABLE Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review December 1, 2006.

October

23,

2006.

Accepted

AC061991N Analytical Chemistry, Vol. 79, No. 5, March 1, 2007

2095