Sei sulla pagina 1di 9

Evaluation of vibrational partition functions for polyatomic systems : quantum versus classical methods for H O and Ar CN 2

Antonio Riganelli, Frederico V. Prudente and Antonio J. C. Varandas* Departamento de Qu mica, Universidade de Coimbra, P-3049 Coimbra Codex, Portugal Received 3rd March 2000, Accepted 14th April 2000 Published on the Web 18th May 2000

The vibrational partition function of H O and Ar CN systems is calculated within the framework of 2 quantum and classical statistical mechanics. The phase space integral arising in the classical picture is solved adopting an efficient Monte Carlo technique. The temperature dependence of the partition function for the two molecules is exploited with a view to study the range of applicability of classical statistical mechanics. For the case of Ar CN van der Waals complex the role played by freezing the CN bond is also analyzed.

1 Introduction
The partition function (Q) is a fundamental concept in statistical mechanics and accurate values of Q are an essential ingredient of any statistical treatment of chemical processes. Values of partition functions for polyatomic species are also demanded by astrophysics for modeling cool stellar atmospheres,1 and are the starting point to derive several thermodynamic quantities. In principle one can obtain the partition function once the energy levels of the system are known. With the amazing growth of computational resources, much eort has been made2 in recent years to obtain Q theoretically using rst-principle methods to calculate the energy levels (i.e., the rotationalvibrational spectrum) of small molecules with an accuracy close to high resolution spectroscopic experiments. In spite of this, the calculations involved when dealing with real systems are formidable and the full description of the rotational motion is still out of reach for most polyatomic molecules. For this reason, historically, several strategies have been developed to explore alternative routes for the explicit calculation of the vibrationalrotational energy levels. These include classical methods with quantum corrections,3 semiclassical methods,4 path integral methods,5 and harmonic plus correction approximations.6 In a previous paper,7 hereafter referred to as paper I, we developed an alternative approach based on the solution of the phase space integral which arises in classical statistical mechanics. The method is based on the Monte Carlo technique and has been shown to yield accurate results for van der Waals systems over the whole range of temperatures studied when compared with the quantum calculations. It also turns out to be very simple to implement. However, it is well known that classical partition functions cannot give accurate results for strongly bound molecules at the low temperature regimes. Messina et al.5 have shown how to obtain accurate values of the partition function for stable molecules such as H O and SO by calculating classically the 2 2 partition function with the true potential function replaced by an eective one. It should be pointed out though that the potential function employed in their work is a force eld, and hence cannot describe correctly the dissociation limits of the molecule.
Presented at the Workshop on Photodynamics from Isolated Molecules to Condensed Phases, Havana, Cuba, February 1319, 2000.

For molecules described by realistic non-separable potentials, the solution of the phase space integral is far more complex. The traditional way to tackle the problem by transforming the phase space integral8h10 into a congurational integral (this approach is implicitly used by Messina et al.5) is not in this case a viable route. Indeed, for polyatomic molecules described by realistic potentials, it is not possible to carry out the integration over momentum space separately from the coordinates.11 Note that this route is usually adopted in conventional transition state theory12 and in conformational studies of macromolecules,13 with a harmonictype potential [lim V (q) \ O, where q is a generalized q?= coordinate vector] being often adopted. Vieth et al.14 proposed a method to estimate the partition function and equilibrium constant based on Monte Carlo simulations but here also the potential function diverges at innity as q increases. The aim of this work is to explore the limitations of the approach suggested in paper I, and further develop it to attain generality. Thus, the nal goal will be to obtain a general procedure to calculate the partition function of systems described by realistic potential functions but avoiding the cumbersome calculation of the energy levels. We limit our calculations to the vibrational partition function. This allows us to shed some light on the classical description limit when a quantity such as the partition function is needed. Although work is in progress to extend the method to include overall rotation of the system, this is not believed to alter the major conclusions of the present work. To test the methodology we consider the prototype systems H O and Ar CN (where dots emphasize the 2 weak nature of the van der Waals bond) which have been much studied both experimentally and theoretically. The rst is representative of molecules with a deep well, while the second is a oppy molecule of the van der Waals type. The paper is organized as follows. In Section 2 we review the theory, and in Section 3 the adopted Monte Carlo procedure. Section 4 presents the applications to the H O and 2 Ar CN molecules. The conclusions are in Section 5.

2 Theoretical background
2.1 Quantum statistical mechanics The partition function is usually expressed as Q\Q Q trans int Phys. Chem. Chem. Phys., 2000, 2, 41214129 (1) 4121

DOI : 10.1039/b001746i

This journal is ( The Owner Societies 2000

where Q is the external translational contribution comtrans monly obtained assuming perfect gas formalism and Q is the int internal (rotationalvibrational plus electronic) contribution given by Q \Q Q (2) int vr e where Q and Q are the rotationalvibrational and the elecvr e tronic partition functions, respectively. In this work, we assume that no electronic excited states are involved, i.e., Q \ 1. In turn, Q is obtained in quantum statistical mechae vr nics as an explicit summation of Boltzmann factors for all the energy levels of the system, namely (3) QQ \ ; g e~ei@kT vr i i where g is the degeneracy factor associated with level i. This i is given by (2J ] 1) times the nuclear spin factor g , where J n is the total angular momentum quantum number. We perform calculations only for J \ 0, and hence g \ 1. i A good example of the intricacies involved in the calculation of the rotationalvibrational partition function using eqn. (3) for bound molecules is given in ref. 15, where Tennyson and coworkers report a study for the H O molecule. The 2 calculation of the energy levels was performed using the PJT2 potential16 up to 30 000 cm~1, and the surface of Ho et al.17 for the calculations above this energy. It is important to point out that these calculations are extremely time consuming even when adopting modern parallel computer technologies, and hence approximate strategies are needed to extend the calculations to high J values. The summation in eqn. (3) is therefore commonly carried out only up to a cut-o energy value. In this work we assume as cut-o value the energy at which dissociation of the molecule occurs as will be discussed in the next section. The role of metastable states will not be addressed here (see ref. 18 and ref. 15 for a detailed discussion on this topic). 2.2 Classical statistical mechanics In the classical statistical mechanics approach, the summation of energy levels of eqn. (3) is replaced by the multidimensional phase space integral19 e~H(q, p)@kT dV (4) B with H being the classical Hamiltonian, n the number of degrees of freedom, and dV the volume element or extension in phase space dV \ dqn dpn (5) 1 QC \ vr hn

A question that arises naturally is which coordinate system is more suitable to express the Hamiltonian and, as a result, to solve efficiently the phase space integral. In paper I, we have chosen the set of Jacobi coordinates (R, r) to describe a van der Waals complex like Ar CN (BC is a bound diatomic molecule) where only one dissociation channel is possible. In our notation, r (modulus of r) represents the BC distance, and R (modulus of R) is the distance between Ar and the center of mass of CN. However, in the case of molecules such as H O, 2 the presence of two equivalent OH channels (we have two distinct fragments) makes the Jacobi coordinates unsuitable for our purpose. In such cases, the hyperspherical system of coordinates, where a unique radial coordinate allows a democratic description of all rearrangement or dissociation processes, suggests itself as the ideal choice. In the next section, we present the Monte Carlo technique used to evaluate the integral in eqn. (4) for the two dierent coordinate systems : massweighted Jacobi, and hyperspherical coordinates.

3 Monte Carlo method


If only the vibrational motion is considered, one has n \ 3 and the integral in eqn. (4) is six-dimensional. The integration can then be performed using a Monte Carlo type approach following the method described in paper I. In that work, it was shown that the problem of efficiency in the calculation is a crucial one. For this reason, we follow an adaptation7 of Barkers22 algorithm to calculate the density of states in the transition state theory context. We discuss here the basic principles of the method, the algorithm for its implementation, and the new features introduced. The idea is near in spirit to the well known importance sampling technique : the variables are not sampled independently of each other but instead some kind of dependence is introduced. This leads to a non-uniform distribution, and hence appropriate weighting factors need to be used. The goal is to choose the sampling volume as close as possible to the integrated one in order to maximize the efficiency of the Monte Carlo simulation. The main idea of this algorithm is to dene the hypervolume B of eqn. (4) by introducing simple bounds on each variable : B \ Mx o xmin O x O xmax #i \ 1, . . . , 2nN (7) i i i where x \ (q, p) \ (q , . . . , q , p , . . . , p ) is a vector with 2n 1 n 1 n components. The basic ingredients of the algorithm may be summarized in the following steps (for a formal demonstration, see ref. 22). 1. Find the equilibrium values of x that dene the global minimum of the Hamiltonian function H(x) : xeq \ (qeq, peq) where qeq \ (qeq, . . . , qeq) and peq \ (0, . . . , 0). In other words 1 n the molecule is set at its equilibrium geometry. 2. Find the integration domain (xmin, xmax) of x with all the 1 1 1 other variables at their equilibrium geometry : H(xmin, xeq, . . . , xeq) \ H(xmax, xeq, . . . , xeq) \ E (8) 1 2 2n 1 2 2n d 3. Sample randomly the x variable within this range 1 according to : xS \ xmin ] (xmax [ xmin)m (9) 1 1 1 1 where m is a random number in the range [0, 1]. 4. Find the integration domain (xmin, xmax) for x with the i i i rst (i [ 1) variables xed at the sampled value and the other (2n [ i) variables at the equilibrium value : H(xS, . . . xS , xmin, xeq , . . . , xeq) 1 i~1 i i`1 2n \ H(xS, . . . , xS , xmax, xeq , . . . , xeq) \ E (10) 1 i~1 i i`1 2n d 5. Sample randomly the x variable inside this range accordi ing to eqn. (9). 6. Repeat steps 4 and 5 for all the variables until the value (xS ) of the last variable is sampled. 2n

P P

where q stands for generalized coordinates and p for the associated conjugate momenta. Note that from the transformation theory of the Hamiltonian20,21 the volume element is an invariant for any canonical transformation of variables which does not explicitly involve the time, and hence dV in eqn. (4) always takes the form dV \ dqn dpn \ dq dq . . . dq dp dp . . . dp (6) 1 2 n 1 2 n This is independent of the coordinates adopted to express the Hamiltonian, a result which will be used in the following. The subscript B in eqn. (4) implies that the hypervolume of integration is restricted to the phase space region corresponding to a bound state situation : 0 O H(q, p) O E , where E is d d the dissociation energy of the molecule (we assume as reference energy the minimum of the potential well). In other words, we must select the region of phase space for which the internal energy is smaller than the dissociation energy. Thus, we will assume as reference atomdiatom dissociation energy the one associated to the channel of lowest energy. For the title systems the diatomic fragments will be CN and OH. 4122 Phys. Chem. Chem. Phys., 2000, 2, 41214129

7. Calculate the weight for each sampled point g, xS \ (xS, . . . , xS ), according to g 1 2n 2n w \ % (xmax [ xmin) (11) g i i i/1 which represents the phase space volume associated to xS . g 8. Calculate the integral in eqn. (4) as M S \ ; f w /M (12) g g g/1 where f \ exp[[H(xS)/kT ] is the function to be integrated g g and M is the total number of sampled points. The standard deviation associated with eqn. (12) is given by22 M p2 \ [M(M [ 1)]~1 ; ( f w [ S)2 (13) g g g/1 In the following subsections, we present the details of the sampling procedure in mass-weighted Jacobi coordinates and hyperspherical coordinates. 3.1 Sampling of phase space in Jacobi coordinates The classical Hamiltonian in mass-weighted Jacobi coordinates (R, r, r where r is the angle between the two Jacobi vectors) for J \ 0 assumes the form (see Appendix for details) H\ 1 1 1 ] P2 ] P2 ] P2 ] V (R, r, r) R r R2 r2 r 2k

In terms of these coordinates, the interatomic distances are given by r2 \ 1 d2 o2[1 ] sin(h/2)sin(/ ] m )] 12 2 3 3 r2 \ 1 d2 o2[1 ] sin(h/2)sin(/ [ m )] 13 2 2 2 r2 \ 1 d2o2[1 ] sin(h/2)sin /] (18) 23 2 1 where d2 \ (m /k)(1 [ m /M), k \ 1, 2, 3, are scaling paramk k k eters, with k and M being the reduced and total mass of the system [see Appendix, eqn. (25)], m \ 2 tan~1(m /k) and 3 2 m \ 2 tan~1(m /k) are the channel angles. 2 3 To clarify the advantage of using this coordinate system to treat molecules with two or more symmetric dissociation channels, we show in Fig. 1 the potential energy surface of H O at xed values of the hyperradius o as a function of the 2 other two hyperangles (h, /). Note that the H O potential 2 energy surface has a deep minimum and two equivalent H ] OH dissociation channels. As a result, three distinct situations corresponding to equilibrium, intermediate, and large o values are visible from Fig. 1. For the equilibrium o value (i.e., o \ 2.5813 a ), panel (a) shows that the minimum of the 0 potential energy surface occurs at h \ 0.6205 rad and / \ 3.2009 rad. At intermediate o values (i.e., o \ 4.0 a ) the 0 two equivalent H ] OH channels begin to appear. The asymptotic situation where the two OH channels are completely separated is illustrated for o \ 5.0 a in panel (c). Note 0

B D

(14)

where the volume element is given by dV \ dR dr dr dP dP R r dP . Note that the interatomic distances (r , r , r ) and r 12 13 23 internal mass-weighted coordinates are related by r \ dr 23

m dr 2 R2 2m 2 2 rR cos r ] [ r2 \ 13 m ]m d2 m ] m 2 3 2 3 m dr 2 R2 2m 3 3 rR cos r ] ] (15) r2 \ 12 m ]m d2 m ] m 2 3 2 3 where d is a mass scaling factor, and m is the mass of the ith i atom [see Appendix, eqn. (24)]. The approach followed to sample the coordinates (R, r, r) and associated momenta (P , P , P ) is that indicated in the R r r algorithm described above. When applying such a procedure, we should keep in mind that, for Ar CN, the R coordinate needs to be sampled before r. Instead, for H O, the order 2 of the sampling in R and r is irrelevant and can be done randomly. 3.2 Sampling of phase space in hyperspherical coordinates Except for an additive constant in the hyperangle /, we use a modied version of Johnsons hyperspherical coordinates23 suggested by Varandas and Yu.24 They consist of a hyperradius o, and two hyperangles (h, /) for the internal coordinates. Three external or rotational hyperangles (a, b, c) complete the set. As shown in the Appendix, the classical Hamiltonian for J \ 0 assumes in terms of these coordinates the form H\ 1 4P2 1 P2 ] 16P2 ] o o2 h sin2(h/2) 2k

A A

B B

BD

(16)

where the volume element is dV \ do dh d/ dP dP dP , o h and the coordinate ranges are 0 O o O O; 0 O h O n; 0 O / O 2n (17)

Fig. 1 Potential energy surface for the H O molecule as a function 2 of the hyperangles (h, /) for dierent values of o : (a) o \ 2.58 a ; (b) 0 o \ 4.00 a ; (c) o \ 5.00 a . 0 0

Phys. Chem. Chem. Phys., 2000, 2, 41214129

4123

4 Results and discussion


The strategies developed in the previous section are tested in this section for two representative systems, H O and 2 Ar CN. 4.1 H O 2 The H O molecule is characterized by a deep well and plays 2 the role of a benchmark system for both bound-state and reactive scattering calculations [O(1D) ] H reaction]. This 2 may be understood from the intrinsic interest of the title molecule due to its importance in atmospheric chemistry, and also to the complexity of its electronic structure, which makes the modeling of the potential energy surface (PES) particularly challenging (see ref. 25). In the present work we use the potential energy surface for the water molecule obtained by one of us26 using the energy switching (ES) method. Such a PES has spectroscopic accuracy, being also suitable for reactive dynamics calculations.27 We calculate the vibrational partition function (Q ) using v dierent methodologies : quantum statistical mechanics (QSM) and classical Monte Carlo (CMC) simulations, with the Hamiltonian being expressed both in hyperspherical (HCMC) and mass-weighted Jacobi (J-CMC) coordinates. The energy levels needed to get the QSM result according to eqn. (3) were calculated for each parity (even and odd) using the DVR3D program of Tennyson et al.28 For these calculations, we have employed Radau coordinates and the following optimum parameters for the DVR basis set : 50 DVR points for each radial coordinate, and 64 DVR points for the angular coordinate. The nal secular problem has a dimension 4000. A total of 460 and 373 vibrational states have been obtained for the even and odd parities (respectively) at energies below dissociation E \ 42 064.9 cm~1. The calculated zero point d energy is e \ 4710.329 cm~1. In the H-CMC calculations, we 0 have sampled 107 points to calculate the integral in eqn. (4) using the procedure described in Section 3.2 with omin \ 1.939 a and omax \ 9.0 a . In turn, for the J-CMC calculations, we 0 0 have adopted the procedure described in Section 3.1 to sample the same number of points using Rmin \ 1.262 a , 0 Rmax \ 8.210 a , rmin \ 1.231 a , and rmax \ 6.532 a (the 0 0 0 range of r is implicitly dened to be 0n). The H-CMC and J-CMC results, together with their respective errors calculated using eqn. (13), are reported in Table 1. In addition, we present the widely used results based on the harmonic oscillator (HO) approximation, which is (in conjunction with the rigid rotor assumption) routinely used in the calculation of most of the JANAF thermodynamic functions29 for polyatomic molecules. Note that the vibrational partition function is in this case given by20,30 3 exp([hcl /2kT ) i (19) QHO \ % v 1 [ exp([hcl /kT ) i i/1 where l denotes the three fundamental frequencies of H O : 29 i 2 l \ 3651.1 cm~1, l \ 1594.7 cm~1, l \ 3755.9 cm~1. 1 2 3 Table 1 shows that the classical description in H-CMC and J-CMC formulations gives values of Q of the same order of v magnitude only for temperatures above 1000 K, while approaching the quantum mechanical results as the temperature increases. For lower temperatures (T \ 1000 K), the classical values overestimate the exact ones by several orders of magnitude. It is important to point out that the classical curve always lies above the quantum results, as it should. A comparison between the results of the two dierent classical procedures also shows clearly that H-CMC produces better results than J-CMC. This validates the assertion that the hyperspherical coordinate system is more adequate when two or more equivalent dissociation channels are present. On the other hand, the HO approximation gives acceptable results

Fig. 2 Values of the equilibrium hyperangles as a function of o for the H O potential energy surface : (a) /eq ; (b) heq. 2

that in both situations (intermediate and large o values) the two H ] OH channels are described symmetrically. The sampling of phase space in these coordinates presents some new features. The major points are : the equilibrium hyperangle values (heq, /eq) are not constant for dierent o values, and hence for molecules such as H O there will not be 2 a unique pair of equilibrium hyperangle values (heq, /eq). This can be seen in Figs. 2 and 3. In particular, Fig. 2 shows the values of the equilibrium hyperangles for the H O potential 2 energy surface plotted as function of the hyperradius o. Clearly, panel (b) of Fig. 2 shows that there is only one heq for each o value. Conversely, panel (a) shows that there are two equivalent values of o starting at o \ 3.1 a , which are associ0 ated to the two possible OH ] H dissociation channels. Of course, the plot of /eq for the Ar CN molecule in Fig. 3 shows only one dissociation channel. For the sampling procedure in hyperspherical coordinates we start for convenience with the variable o. For this, the omin and omax values were obtained according to step 2 [eqn. (8)] of our algorithm. The equilibrium values of the two other internal coordinates (h, /) were then obtained (see Figs. 2 and 3) by a cubic spline interpolation of a uniform o grid. To warrant a uniform sampling of phase space, for each sampled o, the two channels in H O were selected randomly. 2

Fig. 3 Values of /eq as a function of o for the Ar CN potential energy surface.

4124

Phys. Chem. Chem. Phys., 2000, 2, 41214129

Table 1 Vibrational partition function for the H O moleculea 2 H-CMCc T /K 100 500 1000 2000 3000 4000 5000 6000 7000 8000 9000 10 000 11 000 12 000 QSMb 3.69([30)g 1.31([6) 1.28([3) 5.77([2) 2.91([1) 8.21([1) 1.77 3.28 5.47 8.44 12.22 16.84 22.26 28.42 QC v 3.78([17) 4.48([5) 4.61([3) 9.38([2) 3.80([1) 9.70([1) 1.98 3.55 5.79 8.79 12.61 17.24 22.70 28.86 ed 1.90([17) 1.50([5) 7.49([4) 8.59([3) 2.24([2) 3.9([2) 5.75([2) 7.74([2) 9.84([2) 1.21([1) 1.44([1) 1.69([1) 1.94([1) 2.20([1) J-CMCe QC v 4.06([11) 1.70([3) 1.96([2) 1.28([1) 4.16([1) 1.01 2.04 3.63 5.93 9.03 12.00 17.87 23.61 30.20 ed 2.87([11) 1.05([3) 9.71([3) 3.29([2) 5.67([2) 8.33([2) 1.13([1) 1.45([1) 1.80([1) 2.17([1) 2.558([1) 2.94([1) 3.35([1) 3.76([1) HOf 5.99([35) 1.44([7) 4.25([4) 3.29([2) 1.96([1) 5.89([1) 1.31 2.44 4.09 6.33 9.27 13.00 17.60 23.18

a The calculations are performed with the zero of energy at the minimum of the potential energy surface. b Quantum statistical mechanical result, this work. c Classical Monte Carlo simulation in hyperspherical coordinates, this work. d Error calculated with eqn. (13). e Classical Monte Carlo simulation in Jacobi coordinates, this work. f Harmonic oscillator approximation, this work. g The number in parentheses indicates the power of 10.

for this range of temperatures but worse results than the classical simulation for higher temperatures (T [ 1000 K), as expected for molecules characterized by deep potential wells. 4.2 Ar CN The Ar CN van der Waals molecule is a weakly bound and very oppy system. The potential used to describe this molecule is a pairwise potential where the two-body terms have been represented by using the realistic extended HartreeFock approximate correlation energy (EHFACE)31,32 model. The parameters of the potential for the Ar C and Ar N fragments can be found in ref. 33, and ref. 34 for the CN molecule. We calculate the vibrational partition function by adopting two dierent methodologies : QSM and CMC. Usually, the study of van der Waals molecules (such as Ar CN) is carried out by considering the chemically stable diatomic as a rigid rotor35 (in the present case, the CN molecule). To tackle this system, we have performed the classical calculation with the Hamiltonian (J \ 0) expressed in three dierent ways : hyperspherical coordinates (H-CMC), full (3D) mass-weighted Jacobi coordinates (J-CMC-3D), and 2D mass-weighted Jacobi coordinates (J-CMC-2D) where the coordinate r is xed at its equilibrium value. To get the QSM results according to eqn. (3) we have used the energy levels calculated in paper I using the BOUND computer code.36,37 The number of energy levels obtained below the dissociation energy (E \ d
Table 2 Vibrational partition function for the Ar CN moleculea

58.0 cm~1) is 13, and the calculated zero point energy is e \ 0 16.893 cm~1. Note that this zero point energy is associated only with the bending and the R-stretching vibrational modes of the pseudo-diatomic Ar(CN). For the H-CMC calculation, 107 points have been sampled to obtain the integral in eqn. (4) using the procedure described in Section 3.2 with omin \ 8.9 and omax \ 55.0 while, for the J-CMC-3D calculation, the procedure described in Section 3.1 has been adopted by sampling the same number of points using Rmin \ 8.72 a , 0 Rmax \ 100.00 a , rmin \ 1.755 a and rmax \ 1.791 a . Note 0 0 0 that the J-CMC-2D calculations dier from the J-CMC-3D ones only because the coordinate r has been kept xed at its equilibrium value (req \ 1.772 a ). The H-CMC, J-CMC-3D, 0 J-CMC-2D, and the QSM results are presented in Table 2. In turn, Fig. 4 shows the logarithm of the dierent QC values as a v function of temperature. Clearly, the H-CMC and J-CMC-3D results dier by orders of magnitude from the QSM results, and only the J-CMC-2D calculations give good agreement with the QSM ones at all temperatures. However, it is also clear from Fig. 4 that the H-CMC and J-CMC-3D results dier from the J-CMC-2D ones approximately by a constant factor. This suggests that the phase space volume associated to the zero point energy has not been properly treated in the classical calculations (for a discussion on this topic in relation to classical reaction dynamics, see ref. 38 and references therein). In order to rationalize this observation, one may estimate how many states of the Ar CN complex have been

J-CMC-2Dd T /K 100 200 300 400 500 600 700 800 900 1000 H-CMCb 4.48([3)g 5.98([3) 6.60([3) 6.94([3) 7.14([3) 7.28([3) 7.39([3) 7.47([3) 7.53([3) 7.58([3) J-CMC-3Dc 4.90([2) 6.74([2) 7.51([2) 7.92([2) 8.13([2) 8.36([2) 8.49([2) 8.59([2) 8.67([2) 8.73([2) QC v 7.28 9.86 10.94 11.52 11.89 12.14 12.32 12.46 12.57 12.66 ee 2.94([2) 3.75([2) 4.09([2) 4.27([2) 4.38([2) 4.46([2) 4.52([2) 4.56([2) 4.59([2) 4.62([2) QSMf 7.07 9.55 10.57 11.13 11.48 11.72 11.89 12.03 12.13 12.21

a The calculations are performed with the zero of energy at the minimum of the potential energy surface. b Classical Monte Carlo simulation in hyperspherical coordinates, this work. c Classical Monte Carlo simulation (3D) in Jacobi coordinates, this work. d Classical Monte Carlo simulation (2D) in Jacobi coordinates, this work. e Error calculated with eqn. (13). f Quantum statistical mechanics, this work. g The number in parentheses indicates the power of 10.

Phys. Chem. Chem. Phys., 2000, 2, 41214129

4125

Table 3 Values of QC/QQ as a function of reduced temperature T QC/QQ T 0.01 0.05 0.1 0.5 1.0 5.0 10.0 50.0 100.0 500.0 Ar CN 2.41(]39) 903 376.52 167.68 1.22 1.04 1.03 1.03 1.04 1.04 1.04 H O 2 1.16(]20) 553.61 9.42 1.25 1.06 1.02 1.03 1.05 1.05 1.05

Fig. 4 Values of logarithm of Q as a function of temperature for the v Ar CN van der Waals molecule. J-CMC-2D (full line), J-CMC-3D (dashed line), H-CMC (dotted line).

counted below the dissociation energy. Classically, such a number can be gauged by using the approximation X\ B

dP dr (20) r B which can be justied from the fact that in van der Waals systems the internal degree of freedom corresponding to the stable (or chemically bound) fragments may be treated separately (see ref. 39) from the others. By evaluating the second term of eqn. (20) with a standard MC technique, we have obtained (1/h) // dP dr \ 0.03, i.e., the dissociation energy is B r much smaller than the quantum zero point energy of the diatom CN. Indeed, this term should be equal to one, if the energy were sufficient to populate the ground vibrational state of the molecule. Note that for the Ar CN case we do not even report the HO approximation. Indeed, we have shown in paper I that such an approximation is completely unrealistic for van der Waals molecules. 4.3 General discussion Until now, we have discussed separately the results obtained for each system under consideration (H O and Ar CN). 2 However, a major goal of this work is to study the range of applicability of the classical Monte Carlo calculations (and, in general, the classical statistical mechanics approach). The central aspect will be to dene for any molecule a limiting temperature value above which the classical Monte Carlo calculations give good agreement with the QSM results. Tables 1 and 2 show that, for H O, this temperature is around 6000 K, 2 while for Ar CN it is below 100 K. On the other hand, the zero point energy and the relative vibrational energy values of the two title systems dier greatly : D103104 cm~1 for H O, 2 and D101 cm~1 for Ar CN. Then, to exploit the temperature dependence of the classical partition function, we dene a reduced temperature T as kT T \ 0.695 (21) e e 0 0 where T is given in K, e is the zero point energy of the 0 system in cm~1, and k \ 0.695 cm~1 K~1. In Table 3, we report the values of the ratio classical/quantum partition function against T for the title systems with the classical calculations performed utilizing the H-CMC method for H O and 2 the J-CMC-2D method for Ar CN. Note that T \ 1 correT\ 4126 Phys. Chem. Chem. Phys., 2000, 2, 41214129

P P A P P A PP B
1 h3 1 h2 ] 1 h

dP dP dP dR dr dr R r r B

dP dP dP dR dr dr R r r B

sponds to the temperature associated with the zero point energy, which is T \ 6771.1 K and T \ 24.3 K for H O and 2 Ar CN respectively. Clearly, Table 3 shows a similar type of deviation behavior for the two title systems. For example, for T \ 0.5 the deviation of the classical result from the quantum ones is 22% for Ar CN (T \ 12.2 K) and 25% for H O (T \ 3388.5 K), 2 while for T \ 1.0 the deviation is 4% for Ar CN (T \ 24.3 K) and 6% for H O (T \ 6777.1 K). Thus, the present results 2 indicate that the classical Monte Carlo method can be utilized to calculate the internal partition function of the title systems for temperatures above T \ e /k. This is a useful empirical 0 relation which may be used for polyatomic systems described by realistic potentials.

5 Conclusions
The vibrational partition function has been obtained by solving the classical phase space integral for the two title systems described by realistic non-separable potentials. An efficient Monte Carlo approach has been implemented for that purpose using hyperspherical and Jacobi coordinate systems. For this, we have provided in the Appendix the expression of the Hamiltonian, being the vibrational, rotational, and coupling parts fully separated. This should be of help for several applications. The numerical algorithm adopted looks stable and its extension to larger systems straightforward. For the H O molecule, the classical values show signicant 2 deviations from the quantum results, especially at low temperatures, as expected for strongly bound molecules. The classical results converge to the QSM at the high temperature limit. On the other hand, for the oppy Ar CN molecule, the classical Monte Carlo simulation gives accurate results for the temperatures studied only when the diatom CN is maintained rigid at its equilibrium geometry. This has been attributed to the phase space associated to the zero point energy of the CN fragment. Another aspect that has been pointed out refers to the absolute temperature at which the classical description begins to give accurate results for the two systems : D6000 K for H O, D20 K for Ar CN. To estimate 2 this temperature we have proposed a useful empirical relation in the form of the ratio of the absolute temperature and the zero point energy of the system. Work is currently in progress to introduce quantum corrections into the present methodology. Our purpose is to obtain a general procedure based on the phase space integral to calculate the internal partition function and therein thermodynamic functions for molecules described by realistic potential energy surfaces, including the low temperature regime.

Acknowledgements
This work has the support of Fundacao para a Ciencia e Tec 8 nologia, Portugal, under programme PRAXIS XXI. It has

also beneted from an EC grant under contract CHRX-CT 94-0436.

transpose matrix. In matrix notation, it assumes the form

AB AB

r5 Q 0 \M \ Mq 5 R 0 x

(30)

Appendix
This Appendix provides the classical Hamiltonians utilized in this work. First, following Johnson,40 we delineate the procedure used for their determination, and then we present the classical Hamiltonians both for mass-weighted Jacobi coordinates and for a modied version24 of Johnsons hyperspherical coordinates23,40,41 (both in body-xed reference axes). Consider a system of three particles, with m representing i the mass of the ith particle, and X its position vector with i respect to space-xed axes X@Y @Z@. Let the mass-weighted Jacobi vectors be dened as42 r \ d~1(X [ X ) 3 2 m X ]m X 3 3 R\d X [ 2 2 1 m ]m 2 3 where d is the mass scaling or normalizing factor,

where M depends on the (q , q , q ) internal coordinates. 1 2 3 Substituting eqn. (30) in eqn. (26), the kinetic energy formula can be rewritten in compact matrix notation as T\ k k qTMTMq \ qTgq 5 5 5 5 2 2 (31)

where the symmetric matrix g \ MTM is named the metric tensor. It is useful to partition this metric tensor in the same manner as the generalized velocity vector. One has g\

(22) (23)

G C CT K

(32)

d\ and

CA BA
m 1 k

m 1[ 1 M

BD

1@2

where G, C, and K are (3 ] 3) square matrices which depend on the particular choice of congurational coordinates. The components of the generalized momentum vector can now be obtained by dierentiating the kinetic energy T with respect to the generalized velocities, p\ dT \ kgq 5 dq 5 (33)

(24)

k\

m m m 1@2 1 2 3 M

(25)

where we have used eqn. (31). Using eqn. (33), we can then express the kinetic energy in terms of the momentum vector, T\ 1 pTg~1p 2k (34)

is the three body reduced mass ; M \ m ] m ] m is the 1 2 3 total mass of the system. In this coordinate system, the expression for the kinetic energy of relative motion assumes the form T\ k 0 0 0 (r5 2 ] r5 2 ] r5 2 ] R2 ] R2 ] R2) y z x y z 2 x (26)

Both the momentum vector (p) and the inverse of the metric tensor (g~1) can also be written in partitioned form. In particular, p can be partitioned as pT \ (PT, JT) where P is the momentum conjugate to Q, PT \ (P , P , P ) 1 2 3 and J is the angular momentum, JT \ (J , J , J ) x y z The inverse of g is given by43 [see eqn. (32)] g~1 \ where U \ (K [ CTG~1C)~1 (39) (36) (35)

where (r , r , r ) and (R , R , R ) are the Cartesian comx y z x y z ponents of the r and R column vectors in the space-xed axes, respectively, and the dot over a symbol stands for its time derivative. We now introduce the body-xed reference frame axes XY Z (rotating coordinates) which are oriented parallel to the principal axes of inertia of the three-particle system. In particular, we assume that the positive Z axis points in the direction of the vector r ] R. Thus, the mass-weighted Jacobi vectors can be written in the space-xed reference frame as r \ R~1(a, b, c)r* R \ R~1(a, b, c)R* (27)

(37)

G~1 ] G~1CUCTG~1 [UCTG~1

[G~1CU U

(38)

where R~1(a, b, c) is the inverse of the rotation matrix and r* and R* are the Jacobi vectors in the body-xed axes XY Z. The vectors r* and R* are dened by three internal or conguration coordinates (q , q , q ) which determine the size 1 2 3 and shape of the triangle formed by the three particles. The Euler angles (a, b, c), external or orientation coordinates, specify the absolute orientation of this triangle. Let x be the instantaneous angular velocity vector of the rotating axes XY Z with respect to the stationary space-xed axes X@Y @Z@. The velocity vectors r5 and R0 , relative to X@Y @Z@, are given by21 r5 \ r5 * ] x ] r* and R0 \ R0 * ] x ] R* (29) (28)

Substituting eqns. (35) and (38) into eqn. (34), and using the denition j \ CTG~1P we obtain a simpler expression for the kinetic energy T\ 1 [PTG~1P ] (J [ j)TU(J [ j)] 2k (41) (40)

These two relations indicate that the Cartesian coordinates of r and R in the space-xed axes can be written in terms of the internal and rotation components of velocities qT \ (QT, xT) 5 0 where QT \ (q5 , q5 , q5 ), xT \ (u , u , u ), and T indicates the 0 1 2 3 x y z

Note that in this way one has to obtain only the inverse of matrices of 3 ] 3 dimension. For a convenient choice of internal coordinates (q , q , q ), eqn. (41) can then be easily evalu1 2 3 ated. The next step consists of obtaining the classical Hamiltonian for two dierent choices of the (q , q , q ) set. The rst is 1 2 3 the hyperspherical coordinate system. The internal coordinates are dened by q \ o, q \ h and q \ /, being related 1 2 3 to the SmithWhitten (o, H, U) ones44 by the relations h \ n [ 4H and / \ n/2 [ 2U. Moreover, they can be dened Phys. Chem. Chem. Phys., 2000, 2, 41214129 4127

implicitly by r* \ o cos H cos U x r* \ [o sin H sin U y r* \ 0 z R* \ o cos H sin U x R* \ o sin H cos U y R* \ 0 (42) z 5 5 By dening the generalized velocity vector as qT \ (o, h5, /, w , 5 x w , w ) we can now obtain matrix M from eqn. (30), and hence y z the blocks of the metric tensor g, namely

obtained for the metric tensor g : 1 G\ 0 0

1 2
0 0 1 0 0 r2 [r2 cos r sin r r2 sin2 r 0 0 0 0 0 0 [r2

r2 cos2 r ] R2 K \ [r2 cos r sin r 0

0 C\ 0 0

0 0 R2 ] r2

(51)

(52)

(53)

G\

a b
1 0 0 o2 16 0 0 0 0 o2 4 0 0 I z 0 0 0

43) (

After the necessary algebraic calculations, and adding the potential energy term to eqn. (41), we nally obtain the classical Hamiltonian in the mass-weighted Jacobi coordinates as H\ 1 1 1 ] P2 ] P2 ] P2 R r R2 r2 r k ] 2 cos r 1 J x] J J k R2 R2 sin r x y r2 cos2r ] R2 J2 J2 ] z R2r2 sin2 r y R2

B D B

1 Ix 0 K\ k 0 Iy 0 0

0 C \ 1 o2 cos (h/2) 0 2 0 where

1 2 1 2
0 0 1

(44) ]

(45)

P J ] r z ] V (R, r, r) kR2

(54)

k I \ o2[1 [ sin(h/2)] x 2 k I \ o2[1 ] sin(h/2)] y 2

(46) (47)

I \ ko2 (48) z are the principal moments of inertia. Doing the necessary algebraic calculations and adding the potential energy term to eqn. (41), we obtain the classical Hamiltonian in hyperspherical coordinates 16 4 1 P2 ] P2 ] P2 H\ o o2 h o2 sin2(h/2) 2k ]

Here, the rst term in eqn. (54) represents the congurational or vibrational kinetic energy, the second the rotational kinetic energy, and the third term the coupling or Coriolis energy. In order to use the classical Hamiltonian [eqn. (49) and eqn. (54)] to calculate the internal (rotationalvibrational) partition function of triatomic molecules, the angular momentum components (J , J , J ) must be explicitly expressed in terms of x y z the Euler angles (a, b, c) and their conjugate momenta (P , P , a b P ). The complete relations are given by Johnson40 in eqn. c (A16) (Appendix A) of his paper. Since in the present work we use only J \ 0, we do not present here the complete expression of the classical Hamiltonians as they are fairly long.

References
1 A. W. Irwin, Astron. Astrophys. Suppl., 1988, 74, 145. 2 O. L. Polyansky, N. F. Zobov, J. Tennyson, S. Viti, P. F. Bernath and L. Wallace, Science, 1997, 277, 346. 3 K. S. Pitzer and W. D. Gwinn, J. Chem. Phys., 1942, 10, 428. 4 R. M. Stratt and W. H. Miller, J. Chem. Phys., 1977, 67, 5894. 5 M. Messina, G. K. Schenter and C. Garrett, J. Chem. Phys., 1993, 98, 4120. 6 D. G. Truhlar, J. Comput. Chem., 1990, 12, 266. 7 A. Riganelli, W. Wang and A. J. C. Varandas, J. Phys. Chem. A, 1999, 103, 8303. 8 D. R. Herschbach, J. Chem. Phys., 1959, 31, 1652. 9 N. Davidson, Statistical Mechanics, McGraw-Hill, New York, 1962. 10 J. E. Mayer and M. G. Mayer, Statistical Mechanics, Wiley, New York, 1963. 11 V. D. Knyazev and W. Tsing, J. Phys. Chem. A, 1998, 102, 9167. 12 K. Pitzer, J. Chem. Phys., 1957, 79, 1804. 13 M. Karplus and J. N. Kushick, Macromolecules, 1981, 14, 325. 14 M. Vieth, A. Kolinsky and J. Skolnick, J. Chem. Phys., 1995, 102, 6189. 15 G. J. Harris, S. Viti, H. Y. Mussa and J. Tennyson, J. Chem. Phys., 1998, 109, 7197. 16 O. L. Polyansky, P. Jensen and J. Tennyson, J. Chem. Phys., 1996, 105, 6490. 17 T. S. Ho, T. Hollebeck, H. Rabitz, L. B. Harding and G. Schatz, J. Chem. Phys., 1996, 105, 10472. 18 A. W. Irwin, Astron. Astrophys., 1987, 182, 348. 19 L. Landau and E. Lifshitz, Statistical Physics, Pergamon Press, New York, 1969. 20 A. H. Wilson, T hermodynamics and Statistical Mechanics, Cambridge University Press, Cambridge, 1957.

J2 1 J2 J2 x] y] z I sin2(h/2)I 2 I x y z 2 cos(h/2)J P z ] V (o, h, /) [ (49) sin2(h/2)I z The rst term in this expression represents the congurational (or vibrational) kinetic energy, the second the rotational kinetic energy, and the third term the coupling or Coriolis energy. The second choice for internal coordinates is mass-weighted Jacobi coordinates : q \ o R o \ R, q \ o r o \ r, and q \ 1 2 3 arccos (R r/Rr) \ r. They are implicitly dened by r* \ r sin r ; r* \ [r cos r ; r* \ 0 x y z R* \ 0 ; R* \ R ; R* \ 0 (50) x y z A similar denition of these internal coordinates has been used by Tennyson and Sutclie45 to obtain the quantum Hamiltonian and by Acioli et al.46,47 to derive the rotating wavefunctions in correlation function quantum Monte Carlo calculations. In this case, the generalized velocity vector is 5 q5 T \ (R, r5 , r, w , w , w ), with the following blocks being 0 x y z 4128 Phys. Chem. Chem. Phys., 2000, 2, 41214129

21 22 23 24 25 26 27 28 29

30 31 32

H. Goldstein, Classical Mechanics, Addison-Wesley, Reading, MA, 2nd edn., 1980. J. R. Barker, J. Phys. Chem., 1987, 91, 3849. B. R. Johnson, J. Chem. Phys., 1980, 73, 5051. A. J. C. Varandas and H. G. Yu, J. Chem. Soc., Faraday T rans., 1997, 93, 819. A. J. C. Varandas, A. I. Voronin and P. J. S. B. Caridade, J. Chem. Phys., 1998, 108, 7623. A. J. C. Varandas, J. Chem. Phys., 1996, 105, 3524. A. J. C. Varandas, A. I. Voronin, A. Riganelli and P. J. S. B. Caridade, Chem. Phys. L ett., 1997, 278, 325. J. Tennyson, J. R. Henderson and N. G. Fulton, Comput. Phys. Commun., 1995, 86, 175. M. W. Chase, Jr, C. A. Davies, J. R. Downey, Jr, D. J. Frurip, R. A. McDonald and A. N. Syveraud, JANAF T hermodynamic T ables, American Chemical Society and American Institute of Physics for the National Bureau of Standards, New York, 3rd edn., 1985. D. A. McQuarrie, Statistical Mechanics, Harper and Row, New York, 1976. A. J. C. Varandas and J. D. Silva, J. Chem. Soc., Faraday T rans. 2, 1992, 88, 941. A. J. C. Varandas and J. D. Silva, J. Chem. Soc., Faraday T rans. 2, 1986, 82, 593.

33 34 35 36 37 38 39 40 41 42 43 44 45 46 47

S. P. J. Rodrigues and A. J. C. Varandas, J. Phys. Chem. A, 1998, 102, 6266. A. J. C. Varandas and S. P. J. Rodrigues, J. Chem. Phys., 1997, 106, 9647. S. P. J. Rodrigues and A. J. C. Varandas, J. Phys. Chem., 1999, 103, 6366. J. M. Hutson, BOUND computer code, version 5, distributed by Collaborative Computational Project No. 6 of the Science and Engineering Research Council, UK, 1993. J. M. Hutson, Comput. Phys. Commun., 1994, 84, 1. A. J. C. Varandas, Chem. Phys. L ett., 1994, 225, 18. A. van der Avoird, P. E. S. Wormer and R. Moszynski, Chem. Rev., 1994, 94, 1931. B. R. Johnson, J. Chem. Phys., 1983, 79, 1906. B. R. Johnson, J. Chem. Phys., 1983, 79, 1916. F. T. Smith, Phys. Rev., 1960, 120, 1058. G. G. Hall, Matrices and T ensors, Pergamon Press, Oxford, 1963. R. C. Whitten and F. T. Smith, J. Math. Phys., 1968, 9, 1103. J. Tennyson and B. T. Sutclie, J. Chem. Phys., 1982, 77, 4061. P. H. Acioli, L. S. Costa and F. V. Prudente, J. Chem. Phys., 1999, 111, 6311. L. S. Costa, F. V. Prudente and P. H. Acioli, Phys. Rev. A, 2000, 61, 012506.

Phys. Chem. Chem. Phys., 2000, 2, 41214129

4129

Potrebbero piacerti anche