JOURNAL OF CHEMICAL PHYSICS

VOLUME 117, NUMBER 19

15 NOVEMBER 2002

Determination of a exible 12D water dimer potential via direct inversion of spectroscopic data
Claude Leforestiera) and Fabien Gatti
LSDSMS (UMR 5636), CC 014, Universite Montpellier II, 34095 Montpellier Cedex 05, France

Raymond S. Fellers and Richard J. Saykally

Department of Chemistry, University of California, Berkeley, California 94720-1460

Received 10 July 2002; accepted 27 August 2002 We report the determination of two dimer water potential energy surfaces via direct inversion of spectroscopic data. The rst surface, rigid, employs the MCY functional form originally tted by Clementi and co-workers from ab initio calculations, modied by adjunction of a fth, uncharged, site to improve the dispersion component. The vibration-rotation-tunneling energy levels were computed by means of the pseudospectral split Hamiltonian method that we developed previously. The tted surface shows considerable improvement as compared to the original one: transitions among the ground-state manifold are in error by at most 0.2 cm 1 , and excited state band origins up to 150 cm 1 ) are reproduced to within 0.5 to 3 cm 1 . For the second surface, exible, we used the same modied MCY functional form, considered now to depend on the 12 internal degrees of freedom, and augmented by the monomer potential energy terms. The water dimer is described in its full dimensionality by collision-type coordinates in order to access the whole conguration sampled by this oppy system. Internal motions of the monomers stretches and bends are explicitly considered by invoking an adiabatic separation between the slow intermonomeric and fast intramonomeric modes. This (6 6)d adiabatic formulation allows us to recast the calculations into an equivalent six-dimensional dynamics problem ( pseudorigid monomers on an effective potential energy surface. The resulting, tted, fully exible dimer potential leads to a much better agreement with experiment than does the rigid version, as examplied by the standard deviation on all observed frequencies being reduced by a factor of 3. It is shown that monomer exibility is essential in order to reproduce the experimental transitions. 2002 American Institute of Physics. DOI: 10.1063/1.1514977

I. INTRODUCTION

Literally hundreds of potential energy surfaces PES exist for use in computer simulations describing the behavior of liquid and solid water, and such activities, along with ab initio molecular dynamics simulations, constitute a major effort in contemporary science.1 The use of effective PES in this context usually engenders several explicit approximations, viz. 1 classical dynamics, 2 pairwise additive PES, and 3 frozen rigid monomers. Calculations with ab initio molecular dynamics MD methods transcend the latter two, but both approaches are also subject to the limited accuracy of the ab initio methods used to describe the intermolecular forces e.g., density functional theory fails to account for dispersion interactions .2 These approximations have been examined in numerous treatments,1 but never in a truly denitive manner e.g., either the PES or the dynamical method was approximate . Recently, several37 PES designed to accurately describe the water dimer have been developed by tting or tuning ab initio PES to the extensive data sets produced by terahertz vibration-rotation-tunneling VRT spectroscopy.8,9 These can be used as the basis for developing more accurate PES
a

Electronic mail: lefores@lsd.univ-montp2.fr 8710

for condensed water, and they can provide more denitive tests of the approximations described above. The spectroscopic PES published thus far labeled VRT ASP-W I, II, and III and SAPT-5St are based on a rigid monomer description and were determined via the associated 6D frozen monomer dynamics calculation. In this paper we describe the determination of two new experimental water dimer PES rigid VRT MCY-5r and exible VRT MCY-5f based on the ab initio MCY PES of Clementi and co-workers,10 and we explicitly examine the effects of including water monomer exibility. The development of an accurate water dimer potential is needed to compute the IR shifts of the monomers, and in the study of processes such as vibrational predissociation, or dimerization in the gas phase. Once such an accurate dimer potential is available, one also can address the role of exibility in the prediction of observed tunneling frequencies in the dimer, and its possible consequences for larger clusters. The exible calculations presented here can also serve as a benchmark for the nonrigid quantum Monte Carlo method,11,12 or approximate formulations such as the vibrational self-consistent eld1315 or the initial value representation16,17 approaches. The outline of this paper is as follows. In Sec. II, we rst describe the modication made to the original MCY PES,
2002 American Institute of Physics

0021-9606/2002/117(19)/8710/13/$19.00

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

12D calculations on the water dimer

8711

and present the rigid dimer potential VRT MCY-5r obtained from tting to (H2 O) 2 experimental transitions. In Sec. III, we introduce the exact quantum formulation describing the exible water dimer, and discuss the adiabatic approximation invoked to perform the calculations. This is then used in Sec. IV to derive a exible dimer potential VRT MCY-5f from the (H2 O) 2 experimental data. In this section, we fully address the role of exibility when comparing to experimental VRT transitions. Finally, Sec. V concludes and considers some directions for future work.
II. FITTING A RIGID DIMER POTENTIAL

basis B can be projected onto the different irreducible representations irrep of the G 16 molecular symmetry group governing the energy levels22 B

where each ; is a symmetry-adapted linear combination j A ,k A , A  j B ,k B , B of the angular functions  J,K,M .

2. Grid representation

In this section, we rst recall the original rigid monomer 6D formalism that we previously elaborated,18,19 and apply it to the MCY potential energy surface of Clementi and co-workers.10 A slight modication to this potential is then introduced. This modied potential is then tted to experimental results and shown to lead to a much better agreement.
A. Rigid monomer formalism

The most compact representation for the potential energy A B V(R, A , B ) is the six-dimensional grid g q s A B A B Rp , where . This grid is restricted to points where the potential energy is lower than some threshold V max , and only non-symmetry-equivalent points are stored, which typically corresponds to a dimension of around 106 .
3. Iterative calculation of energy levels

This formalism has already been described in two previous papers,18,19 the reader being referred to these for detailed information. We briey sketch it below in order to show how it can be extended to exible monomers, as in the next section. The description of nuclear motions in complexes comprising two polyatomic fragments A and B was rst studied by Brocks et al.20 In the case where these fragments can be considered as rigid, these authors gave an explicit expression for the quantum Hamiltonian see Eqs. 24 and 30 in their paper, and also Ref. 21 for many examples , which reads as
2

As very large bases have to be considered in order to converge energy levels of interest for the water dimer, we resort to an iterative method based on the Lanczos algorithm.23 It consists of repetitive actions of the Hamiltonian operator on a seed vector u 0
n 1

un

un

un

3 un .

in order to construct the requisite Krylov space K In this space, H is tridiagonal and symmetric
n n 1

u n H un , un
1

Hrigid

2 2

AB

2 1 A B R Hrot Hrot V R, R R2

H un .

1
AB R 2

J2 j2 2j.J ,

where R is the distance between the centers-of-mass of the X two monomers A and B, and AB their reduced mass; Hrot X and j are, respectively, the rigid rotational Hamiltonian and total angular momentum of monomer X see Eq. 30 of Ref. 20 ; j jA jB is the coupled internal rotational angular momentum; J j L the total angular momentum L is the relative angular momentum between the monomer centers-ofmass ; and X ( X , X , X ) represents the Euler angles dening the orientation of monomer X in the dimer body-xed BF frame.
1. Spectral representation

The kinetic energy operator displays a simple expression in the overall basis set B n
 

j A ,k A ,

A A


B

j B ,k B , K ,

J,K,M

Diagonalization of H in the K space rst produces converged energy levels which are located in the sparse part of the spectrum, i.e., its low energy region. The different symmetries can be independently studied by dening a seed vector u 0 belonging to a given symmetry-adapted subspace n  ; . As the Hamiltonian operator belongs to the totally symmetrical irrep A 1 of G 16 , the above Lanczos scheme will conne the resulting Krylov space to the same irrep as u 0 . The action of H on a vector u n is performed by means of the pseudospectral split Hamiltonian PSSH method as briey described here. The effect of the kinetic energy operator can be calculated in the spectral representation, where it is analytical. To apply the potential term V, one switches to the grid representation where it is diagonal, i.e., the amplitude of the wave function at each point A B A , B ,Rp is simply multiplied by the value of g, q , s , the potential at that point. After multiplication, the wave function is then back-transformed to the spectral representation. Such a pseudospectral scheme is equivalent to a multiple quadrature of Gaussian accuracy if the grid points coincide with the zeros of analytical functions.24
B. The rigid MCY potential

where n is an appropriate basis for the interfragment distance R; j X ,k X , X is a Wigner basis set describing the rotation of monomer X; and J,K,M is the Wigner basis set associated with the overall rotation of the complex. The

Using the above scheme, energy levels have been computed up to 150 cm 1 of internal excitation for the MCY

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8712

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

Leforestier et al.

dimer potential.10 Convergence of transition frequencies between nondegenerate A and B symmetry levels to within 0.01 cm 1 was achieved by employing a Wigner basis set specied by j max 10 on each monomer. For this particular PES, degenerate E levels require a somewhat larger basis ( j max 11). Overall basis set sizes ranged from 73 000 for J 0 to 220 000 for J 1 calculations. The corresponding symmetry-reduced grid independent of total J) spans ca. 106 points. Due to the use of a Lanczos iterative scheme, the lowest four J 0 energy levels of each symmetry converge in about 15 min of CPU time on a Pentium IV Xeon PC. This time increases to 60 min for J 1 calculations. A comparison between experimental8,25,26 bold gures and calculated gures in parentheses transition frequencies appears in Fig. 1. Diagram a focuses on the splittings associated with the J 0,1 ground states, while b displays the band origins of low energy intermolecular vibrations. As noted before,18,19 this surface leads to interchange splittings (A i B i transitions that are much too small, and acceptor splittings such as the a 0 and a 1 values that are too large by almost a factor of 2.

C. Modifying the rigid MCY potential

The original MCY pair potential10 corresponds to the sum of the Coulomb interaction in the form of point charges ( q on the hydrogens and 2q located on the water C 2 axis near the oxygen , and a series of exponentials to model the short range exchange repulsion (A OO , A HH , and A OH terms and dispersion (A OH term at long range
molecules

V MCY

charged sites

q iq j Rij e e

A OOe

OOR OO

A HH A OH

HHR HH

A OH .

OHR OH

OHR OH

5
FIG. 1. Comparison between experimental Refs. 8, 25, 26 bold and calculated (H2 O) 2 transitions from the tted VRT MCY-5r dimer potential original MCY results; gures given in parentheses . a Splittings in the J 0,1 ground vibrational states; b band origins of the lowest excited states.

In the above equation, summations run on pairs of atoms belonging to different water molecules. This original MCY potential was modied by the adjunction of a fth sitea oating virtual uncharged site VS along the C 2 axis of each monomer as shown in Fig. 2. This virtual site is then used in place of the oxygen atom in the dispersion term ( A OH e OHR VS H) of the potential.

The six-dimensional energy level calculation was embedded in a standard nonlinear least-squares tting routine LevenbergMarquardt algorithm27 in order to minimize the chi-square normalized
2

D. Fitting the modied MCY rigid potential: VRTMCY-5r

This modied potential depends on 11 parameters corresponding to the 9 coefcients q 2 ,A OO , OO ,A HH , HH , A OH , OH ,A OH , OH appearing in Eq. 5 and the two distances d VC and d VS specifying the positions on the C 2 axis of the charge 2q and the virtual site, respectively. It should be noted that setting d VS to zero gives the original MCY potential.

1 Nt

Transitions t

E obs t
t

E calc t

where E obs represents a transition between two rovibrat tional states, and t corresponds to the associated uncertainty. In fact, the experimental uncertainty is negligible ( 1 MHz 8 10 5 cm 1 ) as compared to the agreement E obs E calc which can be expected from the t. In order t t to handle transitions which differ largely in magnitudes, t

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

12D calculations on the water dimer

8713

well depth D e of 4.95 kcal/mol as obtained in the largest ab initio calculations performed28 31 at xed equilibrium monomer geometry. The matrix of derivatives is determined via the HellmanFeynman theorem E pk V pk .

FIG. 2. Modication of the MCY dimer potential by adjunction of a virtual uncharged site VS along the C 2 axis of the water molecules.

has been set to 10 1 E obs for all observed transitions. For t the t presented below, only transitions between nondegenerate (A/B) levels were considered. The reason is that E levels require signicantly larger angular basis sets, and thus would render the calculation much more costly. The experinJK n J K actually used in mental transitions the t are listed in Table I. In the above notation, n stands for the nth vibrational level starting from n 1) of symmetry associated with a total angular momentum projection number K. Some of the transitions appearing in this table are actually symmetry forbidden, but were t from experimental data and correspond to band origins. We also incorporated in the t a
TABLE I. Experimental (H2 O) 2 spectroscopic transitions Refs. 8, 25, 26 used in the t. nJK means the nth vibrational level of symmetry associated to a total angular momentum projection number K. Transition A2 B2 A1 A2 A1 A1 A1 A2 B2 A1 B1 A1 B1 A1 B1 A2 B2 A2 B2 A2 B2 A2 B2 100 110 100 111 110 100 100 100 100 110 110 100 100 100 100 100 100 100 100 100 100 100 100 A 2 B 2 A 1 B 2 B 1 B 1 B 1 B 2 B 2 A 1 B 1 A 1 B 1 B 1 A 1 B 2 A 2 A 2 B 2 B 2 A 2 A 2 B 2 110 100 110 111 111 111 100 100 111 211 211 211 211 200 200 200 200 200 200 300 300 100 100 Frequency (cm 1.057 0.245 1.163 0.541 13.666 15.535 0.752 0.652 0.290 86.62 88.48 88.89 87.03 109.78 106.08 98.04 97.38 114.03 103.97 143.71 141.18 54.94 51.74 1730.7
1

As the potential expression is particularly simple in this case, analytical derivatives V/ p k were used for the rst nine parameters, while the last two were numerically computed by nite differences. The pseudospectral scheme previously described for handling the potential term can be straightforwardly applied here by expressing the derivatives on the same six-dimensional grid. Figure 1 presents a comparison between observed bold gures and calculated transitions, obtained at the end of the tting procedure. The transitions obtained from the original MCY potential without the virtual site are recalled in parentheses in this gure. It can be seen that all transitions energies have been greatly improved, especially the donor acceptor splittings. The resulting standard deviation rms 1/N t E obs E calc 2 takes the value 3.62 (cm 1 ), t t down from 65.6 for the original MCY potential. Table II gives the new set of parameters for the tted potential VRT MCY-5r , the main changes in these parameters concerning the repulsion and dispersion linear terms. The equilibrium geometry and the rotational constants reported in this table were averaged over the ground A 1 wave function f
(A 1 ) 0

f R,

(A 1 ) 0

the above integral being efciently computed by the same pseudospectral scheme as already used for the potential term. We used the naming conventions given in Fig. 3 to describe this equilibrium geometry.

a a a a a a b b b b b b b b b b b b b b

III. FLEXIBLE MONOMER CALCULATIONS

In this section, we show how one can make use of an adiabatic decoupling between the intra- and intermolecular modes to make the calculations feasible. This allows us to recast the exible monomer formulation into that for the rigid case, provided one uses for the intermolecular modes an adiabatic potential generated by the intramolecular coordinates.
A. Flexible monomer formalism

Well depth D e
a b

Calculated from observed transitions. Band origins obtained by tting all the observed FIR data to the appropriate energy level expressions.

In order to describe the VRT dynamics of the water dimer in its full dimensionality, six intramolecular coordinates must be added to the intermolecular coordinates R, A , B considered in the previous section. The former can be chosen as the Jacobi coordinates qX R X ,r X , X describing the deformation of monomer X ( A,B) in its own body-xed frame, as shown in Fig. 4. The BF zX axis is dened along the RX Jacobi vector, and yX is in the molecular plane. The quantum exact Hamiltonian operator is then20,32

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8714

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002 TABLE II. Changes in the MCY potential parameters from original to tted rigid rigid to tted exible denition, and resulting equilibrium geometry. Parameters a.u. q2 A OO
OO

Leforestier et al.

Original MCY Ref. 10 0.514 783 1734.20 2.726 70 1.061 90 1.460 97 2.319 39 1.567 37 0.436 01 1.181 79 0.505 78 0 65.6

VRT MCY-5r 0.475 651 1998.13 2.523 48 0.884 429 1.522 05 1.942 69 1.520 02 0.547 74 1.226 86 0.462 31 0.369 80 3.62

VRT MCY-5f 0.583 599 2376.94 2.717 70 0.986 42 1.567 04 2.913 18 1.504 67 0.475 543 1.182 16 0.517 424 0.038 57 1.26 b

A HH
HH

A OH
OH

A OH
OH

d VC d VS rms (cm

Equilibrium geometry R OO () H O O
b

D A

2.87 2.91 4.17 142.2 2.99 3.33 2059.7 1351.2

2.99 3.10 9.63 107.3 1.75 2.98 1741.9 1043.5

3.04 3.10 9.97 118.0 2.14 3.11 1734.5 1231.6

(z A ,O A O D ) D D e (cm 1 ) D 0 (cm 1 ) Rot. constants (cm 1 ) A (B C)/2 B C

a

7.93 8.21 a 0.217 0.211 a 5.2( 4)(2.6( 2)) a

6.97 7.82 a 0.205 0.187 a 2.6( 4)(1.7( 2)) a

7.23 7.93 a 0.196 0.187 a 4.1( 5)(1.7( 2)) a

These quantities are averaged over the ground A 1 wave function Eq. 6 . Assuming gas phase equilibrium geometry for the monomers (d OH 0.9579 , HOH 104.50).

AB

2 1 A B R TVR TVR R R2 A

2 X TVR

2 1

2 1 2 R RX RX X

1 2 2 r 2m r X r X X

lX 2 2 2mr X
. 8

V qA ,qB ,R,

1 2
2 AB R

J2 j2 2j.J , 7

2 2 RX

jX

X 2

2jX . l

represents the vibration-rotation kinetic energy where operator KEO of monomer X

X TVR

2m H .m O /(2m H m O) and m In the above expressions, m H/2 are the reduced masses associated with the R and r variables, respectively, and l represents the angular momen-

FIG. 3. Equilibrium geometry of the water dimer and naming conventions used in the text: d means the donor molecule, and a the acceptor one; H f and Hb correspond, respectively, to the free and bound hydrogens in the donor molecule.

FIG. 4. Internal Jacobi coordinates used for describing the intramolecular vibrations on each monomer.

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

12D calculations on the water dimer

8715

tum associated with the r motion. jX , j jA jB , and J j L have the same meaning as in the rigid rotor formulation Sec. II A . An exact quantum calculation explicitly treating all 12 internal degrees of freedom is not possible for the time being, due to the computational demands. However, the above expression Eq. 7 can serve as a starting point for an approximate treatment as discussed below. The central concept is that for the intramolecular modes of the water dimer, a single quantum of excitation energy would break the hydrogen bond. This implies that only the ground vibrational state for these modes should be considered in a rst approximation as long as one is not interested in vibrational predissociation of the dimer . Such an adiabatic approximation was recently applied by Klopper, Quack, and Suhm33,34 to a (4 2)d treatment of the HF dimer, in which the eigenvalues of the two HF stretches dene effective potential surfaces for the intermolecular dynamics (4d). In this approach, different rotational constants B e were used for the monomer subunits depending on the number of excitation quanta. A similar adiabatic approximation, denoted (6 6)d, can be invoked here in order to separate out the internal degrees of freedom into two groups: i ii the fast intramolecular coordinates qA ,qB ; the slow intermolecular coordinates R, A , B , denoted collectively Q, which will govern the dynamics on the adiabatic potential dened with respect to the fast motion. Our approach will, however, differ in the way we treat the rotational constants B x , B y , and B z associated with each monomer, retaining their explicit dependence on the Euler angles A and B , as will be shown later on.

In order to apply this adiabatic decoupling scheme, one rst separates out the total Hamiltonian Eq. 8 into two parts H Hintra Q Tinter , 9

where Hintra(Q) represents the intramolecular vibrational Hamiltonian at xed intermolecular geometry Q Hintra Q
2 X TV A B TV TV V qA ,qB ;Q , 2 2 RX 2 2 2 rX

10

1
2 2 RX

2 1

2m
2

2 2mr X sin

sin
X X

X X

11

X X that is, TV corresponds to reducing TVR to the jX 0 case, and Tinter is to be specied later on. The adiabatic decoupling then consists of repetitively solving the six-dimensional vibrational equation for the the intramolecular modes

Hintra Q

nA nB

qA ,qB ;Q

EnA nB Q

nA nB

qA ,qB ;Q , 12

in order to dene the adiabatic potential EnA nB (Q), where nX stands for the three indices n s symmetric stretch , n b bending , and n a antisymmetric stretch labeling the vibrational modes of monomer X. Such a six-dimensional calculation has to be performed at every point of the six-dimensional intermolecular Q grid ( 106 points . In order to make this efcient, we used the following semiperturbative scheme at each geometry Q:

i ii

A B We rst optimize the intramolecular geometry (qopt ,qopt ;Q) by means of the Powell method,27 which does not require the energy derivatives; the intramolecular potential is then expanded around the minimum

V qA ,qB ;Q

A B V qopt ,qopt ;Q B V qA ,qopt ;Q A V qopt ,qB ;Q

V0 Q V0 Q V0 Q V A qA ;Q V B qB ;Q

iii

V qA ,qB ;Q V A qA ;Q V B qB ;Q V 0 Q V AB qA ,qB ;Q ; 13 Y the vibrational states of monomer X are exactly computed for monomer Y X at its minimum energy geometry qopt by means of a sequential truncation-reduction scheme35,36 X TV VX qX ;Q
nX

qX ;Q
AB A

X En Q
X

nX

qX ;Q ;
A nA B nB

14 Q V AB
A nA B nB

iv

the remaining term V EnA nB Q

(q ,q ;Q) can be taken into account by perturbation

A En Q
A

A B V qopt ,qopt ;Q

B En Q
B

Q .

15

In fact, we will show in Sec. III C that this correction, which is very small but very demanding on computation, as it results from a six-dimensional integration, can be omitted with negligible effects on the transition energies.
Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8716

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

Leforestier et al.

Let us now consider the denition of the Tinter term, which appears in Eq. 9 as the result of separating the total Hamiltonian into intra- and intermolecular contributions. The main consequence of having dened a Q-dependent vibrational basis set Eq. 12 is to make the Bs coefcients entering the rotational KEO depending explicitly on the Euler angles A and B , as well as on the R separation. It is thus necessary to reformulate these operators in a consistent way. Within an adiabatically constrained model for the KEO, Gatti et al.37 have shown that in such a case the correct expression for the rotational KEO associated with molecule X is given by BX x
X TR

0 BX y BX zy

0 BX yz
X Bz

X ,X ,z . jx jy jX

0 0

X jx X jy z jX

16

where B x , B y , B yz B zy and B z are the rotational constants associated with the instantaneous internal geometry specied by the (r,R, ) Jacobi coordinates Bx By
1 2

R 2 mr 2

17 18

FIG. 5. Changes in the intermolecular potential due to exibility: V rlx is the static decrease in energy upon relaxation of the monomer geometries; Z PE corresponds to the lowering of the zero-point energy of the monomers embedded in the dimer; E 0 corresponds to the sum of these contributions. These R-dependent quantities have been obtained from averaging over the Euler angles Eq. 32 .

1 , 2 R2 1 2 sin2 cot . 2 R2 cos R2

2

XM Bz 1 , mr 2 19

XM

XM XM

Hintra
N

N N

XM XM .

Tinter

B yz

20

In a near-adiabatic approximation, we do not allow for coupling between different adiabatic N states, which simplies the above expression as XM
N

The presence of the off-diagonal B yz term originates from the loss of the C 2 v symmetry for the water molecule in the bound dimer. Equation 16 can be recast into the more compact expression
X TR

H
MM

XM XM
N

EN

Tinter

XM .

One can then invoke a FranckCondon-type approximation j

X t

.B

(X)

.j .

21
N

Tinter

A B Tinter qopt ,qopt ,Q ,

23

In the following, we will neglect the effect of the B yz B zy contributions, as those terms are at least 2 orders of magnitude smaller due to being very close to /2 in the optimized geometry qopt dened earlier. This approximation typically changes the rotational energies by less than 10 4 in relative values. The nal expression retained for the intermolecular KEO, denoted Tinter , is
2 2

where the Tinter operator is evaluated for the optimized A B (qopt ,qopt) internal geometry corresponding to the instantaneous Q intermolecular conguration. This simplication is justied because each molecule is in its ground vibrational state, for which the harmonic approximation is almost exact. The B matrices appearing in Eq. 22 will thus be evaluated for the optimized internal geometry corresponding to the Q geometry
N

Tinter

2 2

AB

R2
2

jA t .B(A) .jA

jB t .B(B) .jB 22

BX qA ,qB ;Q

N qA qB

A B BX qopt ,qopt ;Q .

1
AB R

J2 j2 2j.J .

If we denote X M the basis set associated with the intermolecular coordinates, and N a shorthand notation for the A B nA nB (q ,q ;Q) adiabatic states dened in Eq. 12 , the matrix elements of the total Hamiltonian in the overall basis set will read as

A B Computation of the TR TR terms is about one order of magnitude more expensive than in the rigid case because one has to switch from the spectral to the grid representation several times in order to compute them. As this transformation is the most time-consuming part of the algorithm, this also slows down the whole calculation by one order of magnitude. We have thus investigated the approximation of averaging the BX matrices over the Euler angles

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

12D calculations on the water dimer

8717

TABLE III. Comparison of different denitions of the rotational constants B see the text for a selected set of transitions. B R, : explicit angular dependence of the B matrices Eq. 21 ; B(R) B: perturbative calculation of the angular dependence Eqs. 24 27 ; B(R): constants averaged over the angles Eq. 24 ; Bequ : constants from the averaged monomer equilibrium geometry in the dimer; B : asymptotical values dissociated dimer of the rotational constants; B(R),V AB : effect of the intramolecular six-dimensional V AB (qA ,qB ) term Eq. 15 . Transition A1 A1 B2 A2 A1 A1 B2 A1 A2 A1 A2 A2 100 100 100 111 110 100 100 100 100 100 100 100 B 1 A 1 B 2 A 2 A 1 A 1 B 2 A 1 B 2 B 1 A 2 B 2 D0 rms
2

Exp. (cm 100 110 111 111 111 111 100 211 200 200 200 300 0.752 1.163 0.290 0.541 13.666 15.535 51.74 88.89 98.04 109.78 114.03 143.71

B R, 0.596 0.971 0.280 0.460 13.748 15.312 50.70 88.96 98.17 106.49 114.69 143.11 1231.7 1.23 2.05

B(R) 0.596 0.972 0.279 0.460 13.754 15.318 50.70 88.97 98.17 106.50 114.69 143.11 1232.4 1.23 2.04

B(R) 0.602 0.978 0.338 0.465 13.698 15.278 51.03 89.22 97.81 106.14 114.44 143.29 1232.4 1.31 2.15

Bequ 0.593 0.969 0.362 0.458 13.645 15.205 51.02 88.98 97.73 105.95 114.27 141.64 1233.3 1.32 2.29

B 0.623 0.999 0.047 0.495 14.095 15.706 50.52 88.91 99.19 108.19 114.02 142.95 1227.8 0.89 2.89

B(R), V AB 0.597 0.972 0.376 0.461 13.684 15.253 51.27 89.34 97.85 106.14 114.54 143.43 1233.9 1.31 2.22

BX R

d e

BX R,

(V(R, A , B ) V min) ,

24

where is some constant ( 103 a.u.), while explicitly retaining the R dependence. This approximation allows us to recast the exible formulation into a rigid one Eq. 1 , exX cept for the rotational constants B X , B X , and B z , which x y depend now on the separation R Hrigid
n

potential is now evaluated for the instantaneous 12D geometry, and no longer for the water molecules in their equilibA B rium geometry (qeq ,qeq) as in Sec. II. It should be noted here that no new parameters have been introduced in the denition of the exible potential Eq. 28 , that is, it depends on the same 11 parameters see Sec. II D as in the rigid case. As discussed in Sec. III A, exibility will modify the intermolecular potential in two ways: i First, at every Q geometry the dimer energy is lowered due to relaxation of the monomer geometries A B A B 29 Vrlx Q V qopt ,qopt ,Q V qeq ,qeq ,Q ; the zero-point energies ZPE of the water molecules embedded in the dimer differ from their isolated values 30 ZPE Q Z PE Q 2Z PE H2 O .

R,

En

R,

25 ii

The actual angular dependence BX R,

A

BX R,

BX R ,

26

is retrieved by rst-order perturbation theory applied to the A , B) intermolecular eigenstates n (R, En

n

jA t . BA .jA

jB t . BB .jB

27

We will show in Sec. III C that this simplication essentially gives the exact transition energies. We will also investigate further approximations for the denition of the rotational constants BX .
B. Denition of the exible MCY potential

These two quantities represent the change from a rigid to a exible potential E0 Q
A B V MCY qeq ,qeq ,Q

V rlx Q

Z PE Q . 31

For denition of the exible potential, we employ the expansion V qA ,qB ,R,
A

In Fig. 5 we present the dependence of these two quantities, V rlx and Z PE, on the separation distance R, resulting from averaging over the Euler angles V R
0

V H2 O qA

V H2 O qB
A

V R,

32

V MCY qA ,qB ,R,

28

where V H2 O(q) is the JPT2 empirical potential determined by Tennyson and collaborators,38 and V MCY is the modied MCY potential dened in Sec. II C. This form guarantees that the dimer potential will exhibit the correct asymptotic behavior upon dissociation. In the above equation, the V MCY

where 0 is the ground (A 1 ) intermolecular wave function as determined from this potential. In order to t such a exible potential, the Hellman Feynman procedure referred to in Sec. II D will require the derivatives of the adiabatic potential EnA nB (Q) with respect to the parameters p k . From the denition 15 of this potential, its derivatives will be given as

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8718

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

Leforestier et al.

pk

EnA nB Q

pk

A B V qopt ,qopt ;Q

pk

A En Q
A

pk

B En Q ,
B

33

where we have ignored the VAB coupling term shown above X to be negligible. The derivatives of the E n contributions are X obtained from a second HellmanFeynman scheme applied to the intramolecular coordinates , that is
X En
X

pk

nX

VX pk

nX

34

Any approximation made in the evaluation of the energy derivatives E/ p k will not change the value of the minimum eventually obtained at the completion of the Levenberg Marquardt scheme, but only modify the path followed to reach it.

C. Test calculations

In Table III we present the results of calculations testing the different approximations made in the exible formulation. These results correspond to tests conducted with the potential energy surface obtained at the end of the tting procedure. In this table i ii B(R, ) means the actual angular dependence of the B matrices Eq. 21 ; B corresponds to retrieving the angular deB(R) pendence effect from perturbation theory Eqs. 24 27 ; (R) ignores this angular dependence, and only uses B the values averaged over the angles Eq. 24 ; Bequ denes the constants from the averaged monomer equilibrium geometry in the dimer; B uses the asymptotical values dissociated dimer of the rotational constants; and B(R),V AB , which should be compared to the B(R) results, tests the inuence of the intramolecular sixdimensional V AB (qA ,qB ) term Eq. 13 .

iii iv v vi

FIG. 6. Same as Fig. 1 for the tted VRT MCY-5f pair potential. The numbers in parentheses recall the values obtained from the tted rigid VRT MCY-5r potential.

From this series of tests, it rst appears that a perturbative treatment of the angular dependent term B(R, ) essentially gives the exact transition values, the difference being at most 0.006 cm 1 for all the transitions considered in the t. The next column of Table III shows that ignoring this B term constitutes a very good approximation, which reproduces the exact values to better than 1%, except for the B 2 100 B 2 111 . Using the values corresponding to the dimer equilibrium geometry (Bequ) basically reproduces the B(R) results, but one can note larger discrepancies for the higher transitions. Interestingly, using the monomer equilibrium geometry (B ) leads to the smallest standard deviation, but to the largest 2 value. Apart, from the B case, the 2 values show that the results stay essentially similar as long as a reasonable denition of the rotational constants is used.

The last column (B(R),V AB ), which should be com pared to the B(R) one, displays the results of explicitly considering the V AB coupling term in the evaluation of the sixdimensional ZPE of the intramolecular modes Eq. 15 . It can be seen that neglect of this term produces marginal changes, being at most 0.04 cm 1 for the ground transitions, and 0.2 cm 1 for the band origins.
IV. FULL DIMENSIONALITY RESULTS

We show in this section that considering a exible potential with the same number of parameters 11 as for the rigid case allows us to achieve a much better t of the VRT

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002 TABLE IV. Equilibrium geometry of the relaxed dimer for the VRT MCY5f dimer potential; the notations are dened in Fig. 3. Distances are in , angles in degrees. Relaxed geometry R OO H O O
b

12D calculations on the water dimer

8719

D A

3.034 9.67 117.7 0.965 0.968 0.959 102.8 102.4 1799.0

(z A ,O A O D ) R OA H R OD Hb R OD H f HO H
b D f

HOA H D e (cm

transitions. We further assess the effect of exibility by performing calculations on this tted exible potential but rigidied, that is, reduced to the rigid case 6D , and show that the results are essentially identical to those obtained from the rigid SAPT5S dimer potential untuned of Szalewicz et al.4 This exible potential is then used to compute the infrared spectral shifts, and we compare them to experimental values. Finally, we use the exible potential, tted exclusively with the (H2 O) 2 transitions, to predict those of the (D2 O) 2 species, and compare against experimental results.
A. Fitted exible potential VRTMCY-5f

Using the formalism given in Sec. III, we t a exible pair potential VRT MCY-5f without considering the transitions involving degenerate E states. An attempt to include those transitions in the t yielded only marginal changes in the results. E-state data are not independent of those already t, as they involve transitions between E states associated with the A/B states explicitly considered in the analysis. Very small bifurcation splittings do depend explicitly on the E states, but these were not included in the present analysis. Figure 6 presents a comparison between observed and calculated transitions obtained from the nal t. The transitions obtained from the previously tted rigid potential VRT MCY-5r are recalled in parentheses in this gure. Comparing with these latter values, it can be seen that inclusion of exibility systematically improves the agreement with the observed transitions it is reiterated that only the same set of 11 parameters was used in this t, the PJT2 monomer potential being left unchanged during the tting process , yielding a standard deviation of 1.23 cm 1 , down from 3.62 cm 1 for the tted rigid potential. The new parameters are displayed in Table II, together with the equilibrium geometry and rotational constants averaged over the ground A 1 wave function; see Eq. 6 . The virtual site introduced in our modied MCY potential Sec. II C is now very close to the oxygen atom (d VS 0.04 ). Although this value is very small, an attempt to t the experimental data while xing this site on the oxygen atom signicantly degraded the results, yielding a standard deviation of 2.5 cm 1 , and a 2 value of 6.04. Table IV gives the 12D equilibrium geometry of the relaxed dimer. Flexibility increases the D e value by only 65 cm 1 0.18 kcal/mol , but results in an increase much larger in D 0 ( 190 cm 1 ). The reason stems from the

FIG. 7. Global picture of all nondegenerate levels up to 200 cm 1 of excitation. Dotted lines indicate the width of the donoracceptor interchange tunneling splitting.

change in the value of the intramolecular ZPE, which is lowered in the dimer due to hydrogen bonding see Fig. 5 . Of particular interest in the ground-state splitting diagram Fig. 6 a is the sum a 0 a 1 of the (H2 O) 2 acceptor tunneling splittings, which has been measured experimentally to be 13.92 cm 1 . This quantity was not used in the tting process as it involves E states, and can be dened as a0 a1 E E E E 100 E E 100 E E 211 35

111 ,

using our conventions see Table I . The value computed from the tted surface is 13.87 cm 1 , in nearly perfect coincidence with the experiment. This result supports our preceding remark concerning the nonindependence of transitions involving E states with respect to those involving A/B states. Rigid calculations performed on the tted potential VRT MCY-5r lead to a value of 14.08 cm 1 , also in close agreement. Figure 7 provides a global picture of all VRT levels of

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8720

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

Leforestier et al. TABLE V. Band origins for the IR excitations: b bending, OHb D donor oxygenbound hydrogen, OH f D donor oxygenfree hydrogen, ss A acceptor symmetric stretch, as A acceptor antisymmetric stretch. Values in parentheses correspond to frequencies computed at the equilibrium geometry of the dimer. H2 O IR excitation b H2O ss H2O as H2O Expt. 1594.7 3657.1 3756.0 (H2 O) 2 Expt. b OHb OH f b ss as
a b

Calc. 1595.4 3659.3 3758.8

Calc. 1630 3573 3727 1624 3592 3703 1637 3561 3729 1624 3581 3683

D D D A A A

1615 3530,b 3601b 3730,b 3735c 1601a 3600,b 3660c 3745b

Reference 40. Reference 39. c Reference 41.

tions V H2 O(q), these latter terms must be responsible for the improvement. They enter the denition of the adiabatic potential through the change in the ZPE that they contribute to, and the intramolecular geometry relaxation effect they generate see Fig. 5 . In order to assess the role of these exible monomer terms, we have run rigid 6D calculations using the VRT MCY-5f potential restricted to the monomer equilibrium geometry, that is the rigid version of our tted exible pair potential
f V rigid R, A

B A

A B VRT MCY 5 f qeq ,qeq ,R,

The results, depicted in Fig. 8, show that the acceptor tunneling splitting sum a 0 a 1 increases to a value of 20.3 cm 1 in the rigid case. This behavior reects the increase of both individual splittings a 0 and a 1
FIG. 8. Same as Fig. 1 for the rigidied VRT MCY-5f pair potential. The numbers in parentheses recall the values obtained from the rigid SAPT5S Ref. 4 ab initio pair potential.

a 0 : 11.6616.33, a 1 : 2.213.97. In the same gure we compare with the results obtained from the original untuned ab initio SAPT5S rigid pair potential of Szalewicz et al.4 given in parentheses. The striking point is that both sets of transitions coincide within a few hundredths of a wave number for all transitions involving the J 0 and J 1 ground vibrational states Fig. 8 a . Notice, however, that some of these transitions are in error by several wave numbers with respect to the experimental values. In order to rationalize the coincidence of the two sets of results, let us consider as a working hypothesis that our tted exible potential is essentially exact at least for the low energy range considered in this work . Similarly, consider the SAPT5S potential as a nearly exact representation of the true pair potential when the monomers are xed at their equi-

(H2 O) 2 up to 200 cm 1 of excitation. Each fork represents a pair of levels split by acceptor tunneling, the dotted lines recalling the further splitting by donoracceptor interchange tunneling. For sake of clarity, only nondegenerate levels are depicted in this gure.
B. Assessing the role of exibility

The most important result deduced from this study concerns the much better agreement achieved when using a exible potential as compared to a rigid one. As the functional form with 11 free parameters retained in the two cases was identical, except for the presence of the monomer contribu-

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

12D calculations on the water dimer

8721

each intermolecular geometry Q, one identies the acceptor (a) and donor (d) molecules from the shortest HO distance between unbound atoms see Fig. 3 ; namely, the hydrogen denes the donor and the oxygen the acceptor. One can then attribute the bound (OHb ) and free (OH f ) hydrogenoxygen frequencies of the donor molecule to the lowest and highest stretch frequencies, respectively. Similarly, the symmetric and antisymmetric stretches of the acceptor molecule are assigned according to increasing frequencies. This procedure allows us to dene six excited adiabatic potentials
(X) En Q , X a or d,

1,0,0 , 0,1,0

or 0,0,1 ,

each one corresponding to a single excitation quantum in one of the six intramolecular modes. Band origins for the IR transitions are computed as the energy difference between the lowest tunneling (A 1 ) level associated with such an ex(X) cited adiabatic potential En and the lowest tunneling (A 1 ) level of the lowest adiabatic potential (E0 )
(X) n (X) E n A 1 100

E 0 A 1 100

The computed frequencies are compared to the experimental values39 41 in Table V. The rst three rows of Table V present a comparison between experimental monomer transition frequencies and those obtained from the method used to calculate the dimer intramolecular ZPE Sec. III A . It constitutes a check that the parameters used in the calculations provide a sufcient accuracy. The following rows report the calculated band origins, as well as their experimental values.39 41 The values in parentheses correspond to frequencies calculated at the equilibrium geometry, i.e., without averaging over the intramolecular wave function. On the whole, averaging tends to increase these values, except for the bending frequencies. One can note that the calculated transitions involving the donor molecule (d) are in better agreement with the experimental values, particularly the oxygen-free hydrogen (OH f ) frequency. For the acceptor transitions (a), the calculated shifts in energy appear exaggerated as compared to the experimental values : measured acceptor transitions are almost identical to those of the free monomer.
FIG. 9. Comparison between experimental Ref. 9 and calculated (D2 O) 2 transitions using the VRT MCY-5f potential tted on (H2 O) 2 transitions.

librium geometry. This hypothesis is substantiated by the high level of ab initio calculations used for its denition. Our rigidied exible potential should then coincide with this latter surface, as demonstrated by the above results. It thus appears that the tuning of the SAPT5S potential,5 in order to reproduce the experimental VRT data, constitutes an articial improvement of the surface: it compensates for the missing key ingredient of the actual pair potential, viz., exibility of the monomers.
C. Infrared shifts

D. D2 O 2 results

The exible dimer potential allows us to compute the IR shifts, within the adiabatic approximation invoked earlier. At

In order to assess the accuracy of the tted exible surface, we used it to predict VRT transitions of the (D2 O) 2 isotopomer. No change was made in the code except for the mass of the hydrogen atom. The results are compared in Fig. 9 to the experimental values of Braly et al.9 It can be seen rst that the donoracceptor interchange splittings are quite well described, even though they are reduced by almost a factor of 20 as compared to (H2 O) 2 . One does, however, observe a systematic error of 0.015 cm 1 ( 40%) for these values. The relative errors are much smaller ( 6%) concerning the acceptor tunneling splittings. Finally, errors for the excited vibrational states range from 1 to 7 cm 1 .

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp

8722

J. Chem. Phys., Vol. 117, No. 19, 15 November 2002

Leforestier et al.

V. DISCUSSION

In this work, we showed that the wealth of experimental results accumulated for the water dimer (H2 O) 2 allowed us to determine, or more precisely, to rene an existing potential energy surface via direct inversion of spectroscopic data. The functional form initially retained in this study was the rigid MCY potential of Clementi and co-workers, originally tted from ab initio calculations. This PES displays the Coulomb interaction, short-range repulsion, and dispersion terms, but no polarization. A t of a polarizable rigid MCY PES to the same set of experimental data is under progress. The determination of a PES by tting to experimental data is particularly efcient for the exible case considered here. As the potential now depends on 12 degrees of freedom, tting from ab initio calculations would require sampling a tremendous number of geometries. Furthermore, one knows a priori a zero-order description of this PES by means of the one-body terms Eq. 28 . Spectroscopic accuracy has already been achieved for these terms from previous tting to experimental results. The t of the exible surface thus deals only with the correction with respect to this zero-order description. Flexibility was included in our calculations by dening Jacobi-type intramolecular coordinates, which minimize the coupling to intermolecular ones. This 12-dimensional system was handled by means of a (6 6)d adiabatic separation between intra- and intermolecular coordinates. This formulation allowed us to recast the exible calculation into a rigid one using the adiabatic energy as an effective potential. Our calculations have shown that exibility plays a crucial role in the description of the acceptor tunneling splitting: its value is modied by the change in the adiabatic zero-point energy with respect to the intermolecular geometry. This was demonstrated by doing test calculations using a rigidied version of the tted exible dimer potential: it produced transitions nearly identical to those originating from the high quality rigid SAPT5S untuned potential of Szalewicz and coworkers. The exible dimer potential obtained in this study reaches a near-spectroscopic accuracy for all transitions reported in the microwave and far-infrared regions. However, its predictions of intramolecular IR shifts are less satisfactory. This can be attributed to the very simple functional form used: the expression we retained, identical to the rigid case, has to account for the changes in intramolecular frequencies of the molecules imbedded in the dimer. A better expression for the potential is under study, which will allow us to include the IR shifts in the experimental tting data, as well as the transitions measured for the (D2 O) 2 isotopomer.
ACKNOWLEDGMENTS

keley effort is supported by Experimental Physical Chemistry Program of the National Science Foundation.

Dr. Linda Braly, Nir Goldman, and Serena Anderson are gratefully acknowledged for very helpful discussions. This work was partially funded by a CNRS-NSF grant. The Ber-

B. Guillot, J. Mol. Liq. 101, 1 2002 . P. L. Sivestrelli and M. P. Parrinello, J. Chem. Phys. 111, 3572 1999 . 3 R. S. Fellers, L. B. Braly, M. G. Brown, C. Leforestier, and R. J. Saykally, Science 284, 945 1999 . 4 E. M. Mas, R. Bukowski, K. Szalewicz, G. C. Gronenboom, P. E. S. Wormer, and A. van der Avoird, J. Chem. Phys. 113, 6687 2000 . 5 G. C. Gronenboom, P. E. S. Wormer, A. van der Avoird, E. M. Mas, R. Bukowski, and K. Szalewicz, J. Chem. Phys. 113, 6702 2000 . 6 G. C. Gronenboom, E. M. Mas, R. Bukowski, K. Szalewicz, P. E. S. Wormer, and A. van der Avoird, Phys. Rev. Lett. 84, 4072 2000 . 7 N. Goldman, R. S. Fellers, M. G. Brown, L. B. Braly, C. J. Keoshian, C. Leforestier, and R. J. Saykally, J. Chem. Phys. 116, 10148 2002 . 8 L. B. Braly, K. Liu, M. G. Brown, F. N. Keutsch, R. S. Fellers, and R. J. Saykally, J. Chem. Phys. 112, 10314 2000 . 9 L. B. Braly, J. D. Cruzan, K. Liu, R. S. Fellers, and R. J. Saykally, J. Chem. Phys. 112, 10293 2000 . 10 O. Matsuoka, E. Clementi, and M. Yoshimine, J. Chem. Phys. 64, 1351 1976 . 11 M. Quack and M. A. Suhm, J. Chem. Phys. 95, 28 1991 . 12 V. Buch, J. Chem. Phys. 97, 726 1992 . 13 J. M. Bowman, Acc. Chem. Res. 19, 202 1986 . 14 J. O. Jung and R. B. Gerber, J. Chem. Phys. 105, 10332 1996 . 15 G. M. Chaban, J. O. Jung, and R. B. Gerber, J. Chem. Phys. 111, 1823 1999 . 16 X. Sun and W. H. Miller, J. Chem. Phys. 108, 8870 1998 . 17 W. H. Miller, J. Phys. Chem. A 105, 2942 2001 . 18 C. Leforestier, L. B. Braly, K. Liu, M. J. Elrod, and R. J. Saykally, J. Chem. Phys. 106, 8527 1997 . 19 R. S. Fellers, L. B. Braly, R. J. Saykally, and C. Leforestier, J. Chem. Phys. 110, 6306 1999 . 20 G. Brocks, A. van der Avoird, B. T. Sutcliffe, and J. Tennyson, Mol. Phys. 50, 1025 1983 . 21 A. van der Avoird, P. E. S. Wormer, and R. Moszynski, Chem. Rev. 94, 1931 1994 . 22 T. R. Dyke, J. Chem. Phys. 66, 492 1977 . 23 C. Lanczos, J. Res. Natl. Bur. Stand. 45, 255 1950 . 24 C. Schwartz, J. Math. Phys. 26, 411 1985 . 25 G. T. Fraser, R. D. Suenram, and L. H. Coudert, J. Chem. Phys. 90, 6077 1989 . 26 E. Zwart, J. J. ter Muelen, W. L. Meerts, and L. H. Coudert, J. Mol. Spectrosc. 147, 27 1991 . 27 W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T. Vetterling, Numerical Recipes Cambridge University Press, Cambridge, 1986 . 28 S. S. Xantheas, J. Chem. Phys. 104, 8821 1996 . 29 M. Schutz, S. Brdarski, P.-O. Widmark, R. Lindh, and G. Karlstrom, J. Chem. Phys. 107, 4597 1997 . 30 J. G. C. M. van Duijeneveldt van de Ridjt and F. B. van Duijeneveldt, J. Chem. Phys. 111, 3812 1999 . 31 W. Klopper and H. P. Luthi, Mol. Phys. 96, 559 1999 . 32 F. Gatti, J. Chem. Phys. 111, 7225 1999 . 33 W. Klopper, M. Quack, and M. A. Suhm, Chem. Phys. Lett. 261, 35 1996 . 34 W. Klopper, M. Quack, and M. A. Suhm, J. Chem. Phys. 108, 10096 1998 . 35 Z. Bacic and J. C. Light, J. Chem. Phys. 85, 4594 1986 . 36 Z. Bacic and J. C. Light, J. Chem. Phys. 87, 4008 1987 . 37 F. Gatti, Y. Justum, M. Menou, A. Nauts, and X. Chapuisat, J. Mol. Spectrosc. 181, 403 1997 . 38 O. L. Polyansky, P. Jensen, and J. Tennyson, J. Chem. Phys. 105, 6490 1996 . 39 Z. S. Huang and R. E. Miller, J. Chem. Phys. 91, 6613 1989 . 40 J. B. Paul, R. A. Provencal, C. Chapo, K. Roth, R. Casaes, and R. J. Saykally, J. Phys. Chem. A 103, 2973 1999 . 41 U. Buck and F. Huisken, Chem. Rev. 100, 3863 2000 .
2

Downloaded 02 Dec 2002 to 128.32.220.140. Redistribution subject to AIP license or copyright, see http://ojps.aip.org/jcpo/jcpcr.jsp