Vous êtes sur la page 1sur 20

Journal of Molecular Structure (Theochem) 586 (2002) 91110

www.elsevier.com/locate/theochem

Conformations, vibrational frequencies and Raman intensities of short-chain fatty acid methyl esters using DFT with 6-31G(d) and Sadlej pVTZ basis sets
Roma E. Oakes a, J. Renwick Beattie a, Bruce W. Moss b, Steven E.J. Bell a,*
b a School of Chemistry, Queen's University, Belfast, Northern Ireland BT9 5AG, UK School of Agriculture and Food Science, Queen's University, Belfast, Northern Ireland BT9 5AG, UK

Received 26 December 2001; revised 14 February 2002; accepted 14 February 2002

Abstract Density functional calculations, using B3LPY/6-31G(d) methods, have been used to investigate the conformations and vibrational (Raman) spectra of three short-chain fatty acid methyl esters (FAMEs) with the formula CnH2nO2 n 35: In all three FAMEs, the lowest energy conformer has a simple `all-trans' structure but there are other conformers, with different torsions about the backbone, which lie reasonably close in energy to the global minimum. One result of this is that the solid samples we studied do not appear to consist entirely of the lowest energy conformer. Indeed, to account for the `extra' bands that were observed in the Raman data but were not predicted for the all-trans conformer, it was necessary to add-in contributions from other conformers before a complete set of vibrational assignments could be made. Provided this was done, the agreement between experimental Raman frequencies and 6-31G(d) values (after scaling) was excellent, RSD 12:6 cm21 : However, the agreement between predicted and observed intensities was much less satisfactory. To conrm the validity of the approach followed by the 6-31G(d) basis set, we used a larger basis set, Sadlej pVTZ, and found that these calculations gave accurate Raman intensities and simulated spectra (summed from two different conformers) that were in quantitative agreement with experiment. In addition, the unscaled Sadlej pVTZ, and the scaled 6-31G(d) calculations gave the same vibrational mode assignments for all bands in the experimental data. This work provides the foundation for calculations on longer-chain FAMEs (which are closer to those found as triglycerides in edible fats and oils) because it shows that scaled 6-31G(d) calculations give equally accurate frequency predictions, and the same vibrational mode assignments, as the much more CPU-expensive Sadlej pVTZ basis set calculations. q 2002 Elsevier Science B.V. All rights reserved.
Keywords: Density functional theory; Fatty acids; Fats; Lipids; Raman; Raman intensities; Sadlej

1. Introduction Saturated fatty acid methyl esters (FAMEs) have the general molecular formula CH3(CH2)nCOOCH3
* Corresponding author. Tel.: 144-28-9027-4470; fax: 144-289038-2117. E-mail address: s.bell@qub.ac.uk (S.E.J. Bell).

(see Fig. 1 for structure and numbering) and are commonly used as model compounds for the much more complex triglycerides that are found in edible fats and oils. We are interested in using vibrational spectroscopy for the non-destructive characterisation of foodstuffs, studies that are aimed, ultimately, at providing quality indicators for both raw and processed foodstuffs. To this end, we have embarked

0166-1280/02/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved. PII: S 0166-128 0(02)00064-7

92

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 1. Structure and numbering of FAMEs.

on detailed experimental and computational studies of extended series of model compounds, such as the fatty acid methyl esters discussed here, to determine the extent to which we can characterise complex foodstuffs on the basis of rich but highly complex and structured Raman spectra. Recently it has been shown [1,2] that a density functional theory (DFT) approach to the calculation of the vibrational frequencies using gradient corrected functionals can perform quite satisfactorily on molecules of moderate size. As the quality of the calculations has improved, the agreement between the experimental and the calculated spectra has also improved, to the point where calculations can provide a realistic basis for making denitive assignments of the bands in experimental spectra. This improvement is due, in part, to the fact that DFT methods have provided a way of including electron correlation in the study of moderately large molecules. Although the resulting vibrational frequencies tend to be higher than those observed experimentally, they are closer to the experimental frequencies than those calculated by semi-empirical and HartreeFock methods. The remaining overestimation of the vibrational frequencies is mainly due to neglect of anharmonicity, together with incomplete basis sets. However, this overestimation of vibrational frequencies is, generally, found to be uniform so that scaling factors can be used throughout a series of homologous molecules to give good agreement with the experimentally obtained vibrations. The determination of appropriate scaling factors for estimating experimental frequencies has received considerable attention in the literature [211]. In contrast to the, by now, fairly well established methods for calculating vibrational frequencies, the calculation of accurate Raman intensities has been more problematic because Raman intensities depend on the square of the polarizability derivative which requires the calculation of the third derivative of the

system's energy with respect to the nuclear coordinates and the electric eld. The quality of the basis set has been found to be the dominant factor in obtaining accurate Raman intensities [12] and Sadlej's polarised triple zeta basis set [13,14], which we have used here, has been shown to be good for Raman intensity calculations. We have started our theoretical studies of lipids and lipid-related compounds with short-chain FAMEs. Since these compounds have only short aliphatic chains, all of the possible conformers of their chains can be built and run as geometry optimisations using the 6-31G(d) basis set without undue computational cost. The purpose of these calculations is to allow us to establish the general rules which govern the relative energies of the conformers in short-chain FAMEs. It is hoped that we can then use these rules to limit the range of conformers which will have to be considered in later calculations on the much larger, long-chain systems which dominate plant- and animal-derived edible fats and oils. Furthermore, because the molecules are small, the computational times, which are short with the 6-31G(d) basis set, remain tractable even when larger basis sets are used. This means that the results of small calculations can be tested against both experimental data and much larger calculations, which are known to be more accurate but cannot be run on large FAMEs because of the vast amounts of CPU time they would require. Of course, one of the immediate objectives of the theoretical studies is to provide secure and unambiguous assignments of the observed Raman bands. The Raman spectra of these small molecules are not as simple as one might expect and we need to understand the molecular basis for this complexity before moving on to studies of the more common FAMEs, which have much longer (C6 C20) alkyl chains. We have found that the nomenclature of these FAMEs can become confusing, often different conformers are referred to as cis or trans and there are instances of these terms being used to label exactly opposite conformers when used by different researchers. The terms `syn and anti', `staggered and eclipsed', `s-trans and s-cis' and `E and Z' are also widely used in the literature, in many cases without any clear denition of what is meant [1521]. In this paper, we use a simple system of describing the conformations which uses the torsion angles of the

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

93

molecules' backbones, so that we dene the conformation of the FAME shown in Fig. 1 by three torsions, T0, T1 and T2. Of course, this is not a complete description because, in addition to the Tn torsions of the backbone, there can also be rotations of the methyl groups (i.e. CH torsions with respect to the backbone), but we have found that only a few calculations are needed to establish a pattern which allows the lowest energy methyl CH torsion angles to be chosen by inspection. 2. Methodology Most of the density functional calculations were performed in gaussian 98 using the hybrid B3-LYP functional and the 6-31G(d) basis set [22]. An `ultrane' grid (a pruned (99,590) grid, having 99 radial shells and 590 angular points per shell) was used because we found that, when we used Cs input geometries for conformers which might be expected to possess a mirror plane, the calculations would not converge with a default (ne) grid. Additionally, the modied GDIIS algorithm was used instead of the default Rational Function Optimisation (RFO) because the latter did not reach a stationary point under the `tight convergence' conditions used throughout this work. The energies for the potential energy surface of C5H10O2 were produced using the `Scan' keyword using 18 steps of 108 in T1 and T2. Data were transferred to Microsoft Excel for manipulation and display. Some calculations were carried out using Sadlej's polarised triple zeta basis set optimised for electric properties, rather than the standard 6-31G(d) basis set. This basis set, although somewhat larger than 631G(d), is reported to provide accurate calculated Raman intensities [12]. Although it was again necessary to use the GDIIS algorithm with this basis set, we were able to use the default (ne) grid rather than the ultrane grid required with 6-31G(d). Raman spectra were obtained using a conventional dispersive system that has been described previously [23]. Briey, the spectra were recorded using 785 nm excitation (Spectra-Physics Ti/sapphire laser pumped by a Spectra-Physics 2020 Ar 1 laser, typically 100 mW at sample) using a 1808 backscattering

geometry, a JobinYvon HR640 single stage spectrograph and a CCD detector. The FAMEs were obtained from SigmaAldrich Inc. and were used as received. Samples were held in 1.6 mm diameter holes within an aluminium block that was at ambient temperature for the liquids and cooled to liquid nitrogen temperature for recording the spectra of solids. 3. Results and discussion The Raman spectra of the solid and liquid FAMEs studied in this paper are shown in Fig. 2 and although the spectrum for methyl acetate (C3H6O2) is relatively straightforward, the Raman spectra of the other two FAMEs are surprisingly complex; in the spectrum of solid C4H8O2 two carbonyl stretching peaks are clearly visible ca. 1750 cm 21 and the spectrum of solid C5H10O2 has many more bands than that of solid C4H8O2, which is only slightly smaller. As a rst step towards resolving these apparent anomalies we have used DFT calculations to predict the structures and vibrational Raman spectra of these compounds. The shortest of the FAMEs we have studied is C3H6O2. Despite its apparent simplicity, this FAME has a considerable degree of conformational freedom. Even before carrying out detailed calculations, inspection of simple models shows that there are two obvious starting positions for the methyl group attached to the ester oxygen (T0 0; 1808). Furthermore, the hydrogens attached to this terminal methyl carbon may be rotated so that one hydrogen points up, eclipsing the carbonyl oxygen, or that two hydrogens point up, straddling the carbonyl oxygen. Finally, the hydrogens on the alkyl chain side of the molecule can also adopt two conformations which differ by a 608 rotation. This gives a set of eight starting structures (Fig. 3) which we used as the input for eight separate DFT calculations. Each of these calculations converged to stationary points (conformers) whose structures were similar to the input structures. The calculated heats of formation of these conformers are given in Table 1, together with the number of negative frequencies, if any, found in the calculations. Table 1 shows that only two of the conformers (d and h) correspond to minima, the rest being saddle

94

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 2. Experimental Raman spectra of: (a) liquid and (b) solid C3H6O2; (c) liquid and (d) solid C4H8O2; (e) liquid and (f) solid C5H10O2.

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

95

Fig. 3. The eight conformers of C3H6O2 used as input geometries for the calculations.

points. Conformation h T0 1808 is the global minimum, while the local minimum (d) has T0 08: The published X-ray structure shows that the T0 torsion is 1808 in the crystal [24]. As expected, there is very good agreement between bond lengths and angles in the crystal structure and in the calculated (conformer h) geometry for example, the bond lengths between heavy atoms typically agree ^0.02 A. In addition to the geometry, the calculations give data on several physical properties of the compounds that have been determined experimentally. First, the difference in energy between the two minima (conformers d and h) is calculated to be 5.4 10 220 J (7.78 kcal mol 21) which is in very good agreement with the experimentally determined value of DH8; which is 8.5 kcal mol 21 [25]. Secondly, the calculations predict a barrier of 0.73 kcal mol 21 for rotation
Table 1 The energies of the various conformers of C3H6O2 shown in Fig. 3 Conformer a b c d e f g h E (Hartrees) 2 268.37090 2 268.37405 2 268.37409 2 268.37608 2 268.38699 2 268.38731 2 268.38814 2 268.38848 DE from lowest energy conformer (kcal mol 21) 11.032 9.055 9.030 7.781 0.935 0.734 0.213 0

of the methyl hydrogens (difference in energy between conformer h and conformer f, which is a saddle point corresponding to the transition state). The experimentally determined value for this barrier in the very similar methyl formate molecule is 1.1 kcal mol 21 [17]. Finally, the calculated dipole moment of conformer h is 1.78 Debye and the observed dipole moment is 1.72 Debye [26]. There is more than one stable conformer in even this most simple of model systems, and as the alkyl chain length increases, the potential for the existence of multiple conformers also grows. In this short-chain FAME, the Raman spectra of the solid and liquid forms are essentially identical and the crystal is known to adopt an all-trans (T0, T1 1808) conformation (c.f. h) so the obvious implication is that the compound is also predominantly all-trans in room temperature liquid. This would also accord with the calculated clear difference in energy between the most stable conformer (h) and the only other minimum (d), in which the OCH3 is rotated away from the carbonyl T0 1808: Since we are condent of the conformation of C3H6O2 we can use it to test the validity of using the Raman spectra to determine the conformations(s) adopted by the FAMEs both as solids and as room temperature liquids. In this case we need only to compare the calculated spectra of both the T0 08 and T0 1808 conformers with the experimental spectrum and to decide which of the calculated spectra best ts the experimental data. For consistency with the work on longer FAMEs shown below, the solid sample is used. Of course, this approach will only work if the differences between the spectra of the two conformers which are to be distinguished are larger than the inevitable

Negative frequencies 2 1 1 0 2 1 1 0

96

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 4. Raman spectra of C3H6O2: (a) experimental spectrum; (b), (c) calculated spectra of conformer h T0 1808 and d T0 08: Frequencies in the calculated spectra have been scaled by 0.95.

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

97

Table 2 Vibrational assignments for the three short-chain FAMEs considered in this paper. For C5H10O2, (a) is conformer q while (b) is conformer o. All calculated frequencies have been scaled as described in the text Vibration C3H6O2 Observed CH rock alkyls CH3 tw CH3 tw O CCO bend OCH3 rock Methyl rock COC bend CCC bend CCO bend COC deform CC str CH2 rock CH3 tw C Acyl str. CC str OC str, i ph CC str, o ph CH twist CC str, i ph CH twist (O) OCH bend CO str CH2 twist CH2 twist CH2 row CH2 rock Methyl umb. (2)CH2 sc Methyl umb(O) CH3 torsion (O) CH2 sc CH3 sc (O) CH3 torsion (C) CH3 sc C CyO str Calculated 49.0 127.7 279.4 178.1 C4H8O2 Observed Calculated 35.4 124.5 205.8 156.7 223.4 327.8 447 584 657 812 854 964 1027 1092 11291197 1264 1368 1377 14251461 1375.0 1444.3 1463.2 1475.2 1455.5 1452.1 1744.6 1390 14191476 440.9 572.0 650.6 809.3 856.3 981.0 1007.3 1083.3 1080.4 1147.1 1174.2 1196.4 1251.0 1354.3 1399.0 1444.5 1445.2 1463.4 1475.3 1475.6 1480.8 1736.8 C5H10O2 Observed Calculated (a) 30.1 79.8 127.6 146.5 175.5 243.3 303.5 334.9 429.0 575.8 696.5 746.5 884.4 888.9 907.8 999.0 1023.6 1097.8 1103.1 1146.1 1171.1 1188.0 1223.3 1293.9 1303.4 1377.6 1395.9 1442.5 1444.2 1463.1 1469.1 1475.4 1479.9 1487.3 1734.9 Calculated (b) 34.4 105.6 128.9 173.0 186.9 257.5 303.1 356.4 461.9 579.4 648.2 786.5 868.6 885.9 904.8 991.3 1036.7 1073.2 1100.4 1145.9 1173.3 1195.1 1222.5 1273.9 1348.8 1363.3 1393.8 1439.7 1444.0 1462.8 1464.6 1475.0 1479.5 1490.6 1731.7

436 614 642 849 985 1049 1039 1163 1193 1255

416.9 597.2 637.3 854.4 983.5 1051.1 1041.3 1146.0 1179.1 1247.4

319 344, 375 436, 469 591 705, 650 758, 786 846 887, 869 1001 1045 1116, 1082 1162 1184 1299, 1279

1422 , 1462 , 1462 , 1462 , 1462 1733

1740

1736, 1747

differences that will be found between the real and the calculated spectra. As shown in Fig. 4, the calculated spectra of the two conformers are rather similar in this case. This difculty is compounded by the fact that the frequencies of the calculated modes lie at higher cm 21 than what appear to be corresponding bands in the experimental spectrum. As discussed above, this overestimation is well-known (indeed it is expected) and in the calculated spectra in Fig. 4 all frequencies have been scaled by 0.95 to roughly bring the positions of the calculated and experimental bands together. The

scaling factor of 0.95 was chosen for this crude scaling procedure because it brings the carbonyl stretching vibrations in the calculated spectra of both conformers of C3H6O2 (which lie ca. 1750 cm 21) close to the position of the band in the experimental spectrum. Fig. 4 shows that even after only rough scaling there is reasonably good agreement between the experimental and calculated peak positions. The calculated spectrum of conformation h T0 1808 ts the experimental data better than does the T0 08

98

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 5. Raman spectra of C3H6O2: (a) experimental spectrum; (b), (c) calculated spectra with linewidths of 5 and 2 cm 21, respectively, used in the simulation.

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

99
21

conformer, both in the region .1000 cm where the number and displacement of experimental and conformer h bands is similar, and below 1000 cm 21, where there are two intense low cm 21 peaks around 600 and 800 cm 21. The calculated positions of these bands in conformer d are at lower cm 21 than the experimental values whilst calculated positions in h are at higher cm 21. This means that for all reasonable scaling factors (i.e. ,1.00), which is normally an essential requirement, the calculated bands for conformer d are moved further away from the experimental values. In these circumstances, it seems reasonable to conclude that we can use the experimentally determined Raman spectrum to show that the compound does indeed adopt the expected T0 1808 conformation. Despite the fact that the single scaling factor of 0.95 brought the calculated bands into the same spectral regions as the experimental frequencies, we have found that, in general, better agreement can be achieved using different scaling factors for three distinct areas within the spectra because these at least divide the vibrations between bands below 1000 cm 21 (which are predominantly bending and twisting vibrations), between 1000 and 1700 cm 21 (mainly skeletal stretching vibrations) and, nally, the carbonyl stretch between 17001800 cm 21. This scaling is a compromise between use of a single scaling factor, which is unnecessarily crude, and full SQM (scaled quantum-mechanical) treatment, which will probably give only slightly better frequencies at signicantly greater effort. For all our 6-31G(d) calculations we have used scaling factors of 0.988, 0.967 and 0.95, respectively, for the three spectral regions. Table 2 shows the assignments of the vibrations together with the observed frequencies and the scaled calculated frequencies, which we generally found, to be in good agreement with those proposed previously [19] on the basis of scaled 4-21G HF calculations. 1 Fig. 5 compares the calculated (scaled three factors) and experimental spectra of C3H6O2. Two calculated spectra, (b) and (c) are shown to emphasise the fact
We have found that all except two of our assignments agree with those due to Pyckhout. These are the bands at 1179 and 1455 cm 21, which were previously assigned as a CH3 rock and a CH3 stretch, respectively. Our assignments indicate that these bands arise from an OCH bend and CH3 torsion, respectively.
1

21

that using a reasonably broad (5 cm ) linewidth in the simulation (as in (b)) gives a trace which can be easily related to the experimental data but in which underlying structure is concealed. For example, the peaks at 638 and 854 cm 21 in (b) are each composed of only one vibrational band whereas the similarly broad peak centered around 1460 cm 21 is a combination of ve vibrations at 1444, 1452, 1456, 1463 and 1475 cm 21, a fact that is only apparent from inspection of the narrow width (2 cm 21) simulation. In the real data these bands overlap, so we prefer to compare the broader simulations and it is clear from the gure that the agreement between the calculated and observed spectra for this small molecule using the 6-31G(d) basis set is acceptable, provided that the frequencies are scaled and some error in the predicted intensities is allowed. The experimental spectra of C5H10O2 (Fig. 2) appear to be much more complex than those of C3H6O2 and C4H8O2, despite the fact that this molecule is only slightly larger. As illustrated in Fig. 1, three torsions dene the conformation of this FAME and, again, inspection of simple models shows that logical starting points would be conformers with T0 0; 1808 and both T1 and T2 0; ^60, ^120 and 1808, which gives two sets of 20 different possible starting geometries, one set for each of the two values of T0 (assuming mirror images count only once). We found that the energies of T0 08 conformers (OCH3 rotated away from carbonyl) were higher than their T0 1808 counterparts so only the lowest energy of the T0 08 conformers, a, is considered here. Additionally, within the set of 20 conformers with T0 1808 most of the conformers have only positive torsion angles (or are their mirror images, with negative torsions) but four of the 20 have a combination of positive and negative torsions (e.g. 160, 260). Finally, we were also forced in a single case to calculate the energies of two conformers with the same backbone torsions but different methyl hydrogen positions, e and e p. In the other conformers, we could use the results from our calculations on C3H6O2 to build starting structures with the lowest energy methyl CH torsion angles close to their optimum values but for this one structure it was not clear which of the starting geometries would converge to the lower energy structure. Overall this gave a set of 22 different starting structures (Fig. 6) which were used as the input for 22 separate DFT calculations.

100

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 6. The 22 conformations of C5H10O2 used as input geometries for the calculations. Conformations e p and e differ only in the rotation of the terminal methyl group (i.e. CH torsions).

In the set with positive torsions we found that there were eight stable conformations, three of which were saddle points (b, e and n) and ve of which were minima; a, g, i and o were local minima whilst q was the global minimum. The results of the energy calculations are shown in Table 3. Conformations c, d, f, h, j, k, l, m and p did not converge near their starting geometries but had nal geometries coincident with one or other of the minima g, i, o or q. The four conformers with mixed positive/negative

torsions also converged to one or other of these minima or their mirror images. The difference in energy between conformations a and q (T0 0 or 1808, all other torsions 1808) was 7.7 kcal mol 21, similar to the corresponding difference in C3H6O2, so there should be virtually undetectable amounts of even the lowest energy T0 08 conformer present in the solid sample. However, the energies of the other three minima are all very close to that of the lowest-lying, all-trans conformer q. The

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110 Table 3 The energies and geometries of the stable conformers of C5H10O2 shown in Fig. 6 Conformer b a n e0 e g i o q T0 180 0 180 180 180 179 179 178 180 T1 0 180 180 0 0 2 67 2 68 164 180 T2 0 180 0 180 180 2 63 2 179 2 69 180 DE from lowest energy conformer (kcal mol 21) 9.275 7.725 6.978 4.342 1.312 0.922 0.741 0.050 0 Negative frequencies 2 0 2 1 1 0 0 0 0

101

3-D representation of the potential energy surface of this molecule for all the T0 1808 conformations with positive torsions is shown in Fig. 7, along with the corresponding contour plot. These two representations show the atness of the potential energy surface around the area of the four minima, which is the reason why we had problems optimising to stationary points using the default (ne) grid and were obliged to use the ultrane grid. It is also clear from the gure and the table that although the energy barrier between conformations q and o is reasonably large (2.7 kcal mol 21), the difference in energy of these two conformers is very small (0.05 kcal mol 21) so thermal population of the second lowest lying conformer might be anticipated. In addition, the next two lowest-lying conformers are also ,1.0 kcal mol 21 above the minimum, so there are several conformers, lying well below the T0 08 form (which was the only other energy minimum in the C3H6O2 molecule), that involve torsions of the alkyl chain and may contribute to the Raman spectrum. The experimental Raman spectrum of C5H10O2 (Fig. 2) was discussed briey earlier, where it was pointed out that, even by simple inspection, there appeared to be more strong Raman bands than would be expected for such a simple molecule. This point becomes much more apparent when the calculated spectrum (Fig. 8(b)) of the lowest energy conformer (which was obtained and scaled using three factors in exactly the same way as for the shorter analogue discussed earlier) is compared to the experimental spectrum of the solid. Although the t between calculated and experimental data was acceptable for

the C3H6O2 FAME, in the longer molecule there are discrepancies that are too large to be due to error in the calculation. In particular, in the areas just above 400 and 900 cm 21 and between 10001400 cm 21, there appeared to be more peaks in the experimental spectrum than in the calculated spectrum. Of course, the obvious explanation is that the experimental spectrum contains contributions from the second lowest energy conformer as well as the global minimum, all-trans structure and when we combined the calculated spectra of conformers o and q we found that this gave a composite that more closely resembled the experimental spectrum (Fig. 8(c)). For example, in the experimental spectrum there are two sharp peaks in the region above 400 cm 21 while the calculated spectra of the two low energy conformers each have just one peak in this region (due to the same vibration). However, the sum of their spectra contains two peaks which match the pair in the experimental spectrum. Addition of bands due to small contributions from higher-lying conformers did not improve the t between calculated and observed spectra of the solid so these conformers were not considered further. The spectrum of the liquid sample shows evidence of many more conformers than that of the solid, the bands are generally broader and the ne structure (for example in the region ca. 1450 cm 21) is lost which makes denitive assignment of the bands to any single conformer almost impossible. For this reason, we have concentrated on assigning the bands in the spectra of the solids. The agreement between combined calculated (631G(d)) and experimental data is good but it is by no means perfect: in particular, the calculated, relative

102

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 7. Two representations of the potential energy surface, for twisting about T1 and T2 in C5H10O2: (a) the 2-D representation; white stars represent saddle points, black stars are minima. The four minima are labelled according to Fig. 6. The black arrows show the direction of movement from starting geometries to the fully optimised structures; (b) the 3-D representation of the same potential energy surface, the change in energy from the highest energy conformer (0, 08) to the lowest energy all-trans conformer (180, 1808) is 9.3 kcal mol 21.

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

103

Fig. 8. Comparison of Raman spectra of C5H10O2: (a) experimental spectrum; (b) the calculated spectrum of the lowest energy conformer q (T0, T1 and T2 1808) and (c) a simulated spectrum prepared by adding the spectrum calculated for o, the next lowest-lying conformer (T0, T1 1808 and T2 ^608), to that calculated for q.

104

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 9. The Raman spectra of C5H10O2: (a) experimental data; compared to: (b) the simulated spectrum created by adding spectra for the two lowest energy conformers, q and o, calculated with the Sadlej pVTZ basis set. The calculated spectra of the two individual conformers, q and o, are shown as (c) and (d), respectively. Vertical lines are guides to show how the bands due to each conformer contribute to the combined spectrum.

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

105

intensity of the carbonyl peak versus the broad peak at around 14001500 cm 21 is very different to that observed experimentally. The merely moderate agreement between calculated and observed band intensities is expected but is particularly signicant in these studies because it was necessary to include a second conformer to improve the t between calculation and experiment. The simulated data shown in Fig. 8(c) are therefore heavily manipulated, having been scaled in frequency and then summed and it is important to establish that the apparently acceptable correlation with experiment is not the result of over-tting. In particular, it is important to be condent that the appearance of bands in the experimental spectrum where the calculations predicted either very weak bands or none at all, is indeed due to a second conformer, rather than bands which were predicted to have low intensity appearing with higher than predicted intensity, i.e. good intensity predictions are needed. Previous studies have found that large polarised basis sets are required for the quantitative prediction of electric eld response properties [27,28]. However, it has been recently found that Sadlej's polarised triple zeta basis set, optimised for electric properties (the Sadlej pVTZ basis set) [14,29] can produce accurate Raman intensities when applied to small molecules [12]. This basis set is about the same size as the 631 1 G(2d, 2p) basis set but performs as well as the, very much larger, aug-cc-pVTZ basis set and so provides an accessible method for the calculation of quantitative Raman intensities. We have run geometry optimisation and Raman frequency/intensity calculations on both the low energy conformers of C5H10O2 using this basis set which, to our knowledge, is the rst time that it has been used in studies of molecular systems containing more than about 10 atoms. The calculated spectra of both of the low energy conformers are shown in Fig. 9(c) and (d), along with the simulated spectrum generated by adding conformers q and o in a 2:1 ratio (Fig. 9(b)) and the experimental data. It is clear that, as was the case for the 6-31G(d) calculations, it is necessary to include both conformers to get the best t between calculated and observed spectra. However, when both conformers are included the agreement is extremely good; there is a one-to-one correspondence between even the weaker bands in both spectra and the relative band intensities are now accurately reproduced, which

was the objective of these calculations. There appears to be a systematic decrease in the intensities of the bands in the observed spectra versus the calculated values with increasing cm 21 but this disagreement is due to the fall off in the spectral response of the detector at higher cm 21 rather than a systematic error in the calculations. In addition to giving accurate relative band intensities, the calculations with the Sadlej pVTZ basis set also yield, without scaling, band positions that are marginally better than the scaled 6-31G(d) values obtained in the conventional way (RSD 9:1 vs 12.6 cm 21). 2 Table 4 details the positions of the vibrational bands calculated using both basis sets and compares them to their experimentally determined values. It is clear from Table 4 that the 6-31G(d) calculations give very similar information to the results obtained from the much larger Sadlej pVTZ calculations. This means that, provided one is willing to scale the frequencies and live with relative band intensities that can be factors of 3 or 5 in error, these smaller calculations can be used to make reliable assignments of FAME Raman bands. Of course, it is easier to use the larger basis set to give results that directly reproduce the experimental data but this option becomes increasingly costly (in term of CPU-time) as the size of the molecules of interest increases. The nal compound that we have investigated is C4H8O2. This FAME was left until the methodology was established because, in contrast to C3H6O2 and C5H10O2, the spectrum of the solid (Fig. 2) clearly has two peaks due to carbonyl stretches, implying that at least two conformers must be present in the sample. Since there were at least two conformers we found that it was essential to run calculations on a large number of starting geometries in much the same way as was done for C5H10O2, although in this case we had to consider only the T0 (0 and 1808) and T1 (0 ^60, ^120 and 1808) torsions, along with a variety of different orientations of the hydrogen atoms. In all, we performed geometry optimisations on 32 different starting conformations using the 6-31G(d)
2 In the spectra shown in Fig. 9, the frequency of the carbonyl stretching band was scaled by 0.964. No other bands were scaled at all and we believe that scaling the carbonyl band is justied because the scaling factor should be transferable to other esters.

106

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Table 4 Comparison of observed Raman frequencies for C5H10O2 with both the scaled 6-31G(d) and the Sadlej pVTZ basis sets for (a) the lowest energy all-trans conformer q and (b) the next lowest energy conformer o Vibration C5H10O2(observed) 6-31G(d) Calculated (a) CH rock alkyl s CH3 tw CH3 tw O CCO bend OCH3 rock Methyl rock COC bend CCC bend CCO bend COC deform CC str CH2 rock CH3 tw C Acyl str. CC str OC str, i ph CC str, o ph CH twist CC str, i ph CH twist (O) OCH bend CO str CH2 twist CH2 twist CH2 row CH2 rock Methyl umb. (2)CH2 sc Methyl umb(O) CH3 torsion (O) CH2 sc CH3 sc (O) CH3 torsion (C) CH3 sc C CyO str
a

Sadlej pVTZ Calculated (b) 34.4 105.6 128.9 173.0 186.9 257.5 303.1 356.4 461.9 579.4 648.2 786.5 868.6 885.9 904.8 991.3 1036.7 1073.2 1100.4 1145.9 1173.3 1195.1 1222.5 1273.9 1348.8 1363.3 1393.8 1439.7 1444.0 1462.8 1464.6 1475.0 1479.5 1490.7 1731.73 Calculated (a) 2 22.7 72.6 141.6 146.8 175.5 223.4 307.8 339.0 433.4 580.1 702.2 737.4 871.9 892.7 914.9 1027.4 1056.4 1112.1 1130.2 1156.6 1188.6 1207.1 1235.0 1312.8 1318.5 1397.2 1404.0 1442.5 1457.0 1458.2 1464.8 1477.6 1472.0 1484.4 1731.9 a Calculated (b) 31.8 99.7 135.5 168.2 190.6 253.3 308.3 365.5 463.7 588.1 657.5 784.6 876.7 891.0 905.4 1020.7 1066.1 1094.2 1122.5 1157.2 1190.3 1212.6 1248.9 1289.9 1362.7 1390.7 1407.2 1446.7 1456.8 1458.6 1466.2 1478.2 1476.4 1486.4 1730.3 a

319 344, 375 436, 469 591 705, 650 758, 786 846 887, 869 1001 1045 1116, 1082 1162 1184 1299, 1279

1422 , 1462 , 1462 , 1462 , 1462 1733

30.1 79.8 127.6 146.5 175.5 243.3 303.5 334.9 429.0 575.8 696.5 746.5 884.4 888.9 907.8 999.0 1023.6 1097.8 1103.1 1146.1 1171.1 1188.0 1223.3 1293.9 1303.4 1377.6 1395.9 1442.5 1444.2 1463.1 1469.1 1475.4 1479.9 1487.3 1734.9

These are the only frequencies, obtained using the Sadlej basis set, to have been scaled (0.964).

basis set. The energies and geometries of the stable conformations obtained as a result of these calculations are shown in Table 5 and Fig. 10 and we found that there were three local minima and a global minimum, which was the expected T0 1808; T1 1808 (all-trans) conformer. Fig. 11 compares the calculated spectrum of the alltrans conformer (after scaling using the same factors as were used for the other FAMEs) with the experimental spectrum. In general, there is acceptable agree-

ment; the relative intensities are somewhat erratic and frequency scaling is necessary but the overall quality of the t is similar to that found for the other FAMEs using the 6-31G(d) basis set. The only indication of a second conformer is the appearance of a shoulder on the main carbonyl band in the experimental spectrum which is of course not predicted in the calculation of the all-trans conformer. Whatever the source of this second carbonyl band it must be very much a minor component of the sample because the shoulder has a

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110 Table 5 The energies and geometries of the different stable conformers of C4H8O2 shown in Fig. 10 Conformer a b i c d q r s t cc dd x ee ff T0 0 0 4 0 0 180 180 180 180 180 180 180 180 180 T1 180 180 95 180 180 0 0 180 180 0 0 60 180 180 DE from lowest energy conformer (kcal mol 21) 11.615 10.147 9.689 9.162 7.762 4.945 4.192 2.949 2.196 2.021 1.274 0.929 0.703 0 Negative frequencies 2 1 0 1 0 3 2 2 1 2 1 0 1 0

107

very small area compared to the main band. It is interesting to speculate on the possible origin of this minor component, bearing in mind that if the proportion of the sample in this second conformation is low then its contribution to the remainder of the spectrum may not be signicant. The difculty in distinguishing weak features due to a second minor component when we move away from the uncongested carbonyl area may explain why the doubling of bands around 800 cm 21 that was observed in the spectrum of C5H10O2 (characteristic of additional conformers) is not apparent here. This despite the fact that the calculations suggest that if there were two conformers in C4H8O2 (as there are in C5H10O2) then a pair of bands would appear in this spectrum in much the same way. Possible candidates for the second (minor) conformer that is present in the sample, along with the predominant all-trans form, are d or x, which have non-1808 torsions around either T0 or T1. The calculated position of the carbonyl stretching band in the T1 608 conformer (x) is only 2 cm 21 higher than that in the all-trans conformer. In contrast, the position of the same band in the T0 08 conformer (d) is calculated to lie 20 cm 21 higher, this is slightly larger than the ca. 10 cm 21 shift observed in the experiment but well within the errors of the calculation. This suggests that the C4H8O2 is predominantly in the all-trans form but that the solid samples we studied contained a small proportion of the higher energy conformer d. There may, in addition, be simi-

larly small amounts of x but since we would not expect x to present so clear a spectral signature it is more difcult to detect. This is a general problem, in C3H6O2 and C5H10O2, the carbonyl band is broad, so in those samples along with the predominant all-trans (or all-trans plus conformer o in the case of C5H10O2) a small proportion of the T0 08 conformer could well be present but it is not detected because of the width of the carbonyl band. 4. Conclusion Conventional B3LPY DFT methods with the 631G(d) basis set are well-established in predicting the geometries, physical properties and vibrational spectra of small organic molecules. In the compounds we have studied here, the DFT methods show that the lowest energy conformer has a simple (all-trans) structure but that there are other conformers, with different torsions about the backbone, which lie reasonably close in energy to the global minimum. The possibility of higher-lying conformers contributing to the Raman spectra would be expected to complicate assignment of observed Raman bands. To resolve this problem, we have used the DFT methods to predict the vibrational spectra of the various conformers as well as their energies. Although an excellent t between calculated and experimental frequencies RSD 12:6 cm21 can be obtained by

108

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

Fig. 10. The 14 conformations of C4H8O2 that correspond to stable geometries (Table 5). The sets of conformations {a, b, c and d} and {s, t, ee and ff} differ only in the rotation of the terminal methyl group (i.e. CH torsions).

scaling predictions from the economical 6-31G(d) basis set, the agreement between predicted and observed intensities is much less satisfactory. This would not be a problem normally because even if the predicted intensities are not accurate, they are good enough to relate predicted to observed bands. However, in some of these FAMEs it was necessary to include contributions from more than one conformer to account for `extra' bands observed in the experimental Raman spectra before a complete set

of vibrational assignments could be made. This leads to some degree of uncertainty because the relative contributions which each of the conformers make to the nal simulated spectrum is set by relative band intensities that are not well predicted in the calculations. The easiest way to resolve this uncertainty, and therefore to conrm the validity of the approach followed with the 6-31G(d) basis set, was to nd a basis set that gave accurate Raman intensities. The Sadlej pVTZ basis set proved to be ideal, giving inten-

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110

109

Fig. 11. The Raman spectra of C4H8O2: (a) experimental data; (b) calculated spectrum of the lowest energy all-trans conformer.

sities very close to their observed values and simulated spectra (summed from two different conformers) that were in quantitative agreement with experiment. Furthermore, because this basis set also gave acceptably accurate vibrational frequencies RSD 9:1 cm21 after scaling only the carbonyl frequency, the results could be used to test the validity of the scaling factors used for the 6-31G(d) calculations. We found that the unscaled Sadlej pVTZ, and the scaled 6-31G(d) calculations did indeed give the same vibrational mode assignments for all bands in the experimental data. This work provides the foundation for calculations on longer-chain FAMEs (which are closer to those

found as triglycerides in edible fats and oils) because it shows that scaled 6-31G(d) calculations give equally accurate frequency predictions, and the same vibrational mode assignments, as the very CPU-expensive Sadlej pVTZ basis set calculations which would be prohibitively time-consuming for FAMEs with chains greater than 14 carbon atoms. Acknowledgements R.E.O. would like to acknowledge the nancial support of the McClay Trust which enabled her to carry out this work. J.R.B. would like to acknowledge

110

R.E. Oakes et al. / Journal of Molecular Structure (Theochem) 586 (2002) 91110 [16] J. Blomqvist, L. Ahjopalo, B. Mannfors, L. Pietila, J. Mol. Struct. (Theochem) 488 (1999) 247. [17] R.K. Bohn, K.B. Wiberg, Theo. Chem. Acc. 102 (1999) 272. [18] M. Oki, H. Nakanishi, Bull. Chem. Soc. Jpn 43 (1970) 2558. [19] W. Pyckhout, C. Vanalsenoy, H.J. Geise, J. Mol. Struct. 144 (1986) 265. [20] M.W. Wong, Chem. Phys. Lett. 256 (1996) 391. [21] S. Xu, C. Wang, G. Sha, J. Xie, Z. Yang, J. Mol. Struct. (Theochem) 467 (1999) 85. [22] M.J. Frisch, G.W. Trucks, H.B. Schlegel, G.E. Scuseria, M.A. Robb, J.R. Cheeseman, V.G. Zakrzewski, J.A. Montgomery, R.E. Stratmann, J.C. Burant, S. Dapprich, J.M. Millam, A.D. Daniels, K.N. Kudin, M.C. Strain, O. Farkas, J. Tomasi, V. Barone, M. Cossi, R. Cammi, B. Mennucci, C. Pomelli, C. Adamo, S. Clifford, J. Ochterski, G.A. Petersson, P.Y. Ayala, Q. Cui, K. Morokuma, D.K. Malick, A.D. Rabuck, K. Raghavachari, L.B. Foresman, J. Cioslowski, J.V. Ortiz, B.B. Stefanov, G. Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. Gomperts, R.L. Martin, D.J. Fox, T. Keith, M.A. Al-Laham, C.Y. Peng, A. Nanayakkara, C. Gonzalez, M. Challacombe, P.M.W. Gill, B.G. Johnson, W. Chen, M.W. Wong, J.L. Andres, M. Head-Gordon, E.S. Replogle, J.A. Pople, Gaussian, Inc., Pittsburgh, PA, 1998. [23] D.S. Jones, A.F. Brown, A.D. Woolfson, A.C. Dennis, L.J. Matchett, S.E.J. Bell, J. Pharm. Sci. 89 (2000) 563. [24] M.J. Barrow, S. Cradock, E.A.V. Edsworth, D.W.H. Rankin, J. Chem. Soc. Dalton (1981) 1988. [25] C.E. Blom, H.S. Gunthard, Chem. Phys. Lett. 84 (1981) 267. [26] J. McMurray, Organic Chemistry, Brooks/Cole, Pacic Grove, CA 1996. [27] C.E. Dykstra, Ab Initio Calculation of the Structure and Properties of Molecules, Elsevier, The Netherlands, 1988. [28] Y. Yamaguchi, M. Frisch, J. Gaw, H.F. Schaefer, J.S. Binkley, J. Chem. Phys. 84 (1986) 2262. [29] A.J. Sadlej, Theor. Chim. Acta 81 (1992) 339.

the award of a studentship by the Department of Agriculture and Rural Development, N. Ireland. The authors thank Dr A. Feron and Mr A. Beattie for supplying samples. References
[1] A.P. Scott, L. Radom, J. Phys. Chem. 100 (1996) 16502. [2] G. Rauhut, P. Pulay, J. Phys. Chem. 99 (1995) 3093. [3] C.W. Bauschlicher, H. Partridge, J. Chem. Phys. 103 (1995) 1788. [4] D.J. DeFrees, A.D. McLean, J. Chem. Phys. 82 (1985) 333. [5] J.W. Finley, P.J. Stephens, J. Mol. Struct. (Theochem) 357 (1995) 225. [6] N.J. Harris, J. Phys. Chem. 99 (1995) 14689. [7] R.F. Hout, B.A. Levi, W.J. Hehre, J. Comp. Chem. 3 (1982) 234. [8] M.A. Palafox, Int. J. Quan. Chem. 77 (2000) 661. [9] P.E. Peterson, M. Abuomar, T.W. Johnson, R. Parham, D. Goldin, C. Henry, A. Cook, K.M. Dunn, J. Phys. Chem. 99 (1995) 5927. [10] J.A. Pople, H.B. Schlegel, R. Krishnan, D.J. DeFrees, J.S. Binkley, M.J. Frish, R.A. Whiteside, R.F. Hout, W.J. Hehre, Int. J. Quan. Chem. 15 (1981) 269. [11] J.A. Pople, A.P. Scott, M.W. Wong, L. Radom, Is. J. Chem. 33 (1993) 345. [12] M.D. Halls, H.B. Schlegel, J. Chem. Phys. 111 (1999) 8819. [13] Sadlej's polarised electric property basis set was obtained from the Extensible Computational Chemistry Environment Basis Set Database, Version 1.0 Pacic Northwest Laboratory, P.O. Box 999, Richland, Washington 99352 (http:// www.emsl.pnl.gov:2080/forms/basisform.html). [14] A.J. Sadlej, Collect. Czech. Chem. Commun. 53 (1988) 1995. [15] E. Bicknell-Brown, K.G. Brown, W.B. Person, J. Am. Chem. Soc. 102 (1980) 5486.

Vous aimerez peut-être aussi