JOURNAL OF CHEMICAL PHYSICS VOLUME 112, NUMBER 13 1 APRIL 2000 On the importance of exchange effects in three-body interactions: The lowest quartet state of Na3 J. Higgins, T. Hollebeek, J. Reho, T.-S. Ho, K. K. Lehmann, H. Rabitz, and G. Scoles Department of Chemistry, Princeton University, Princeton, New Jersey 08544-1009 Maciej Gutowski Materials Resources, Pacific Northwest National Laboratory, Richland, Washington 99352 and Department of Chemistry, University of Gdańsk, 80-952 Gdańsk, Poland 共Received 8 November 1999; accepted 10 January 2000兲 Three-body interactions in a homonuclear van der Waals bound trimer 共the 1 4 A 2⬘ state of Na3) are studied spectroscopically for the first time using laser induced emission spectroscopy on a liquid helium nanodroplet coupled with ab initio calculations. The van der Waals bound, spin polarized sodium trimers are prepared via pickup by, and selective survival in, a beam of helium clusters. Laser excitation from the 1 4 A ⬘2 to the 2 4 E ⬘ state, followed by dispersion of the fluorescence emission, allows for the resolution of the structure due to the vibrational levels of the lower state and for the gathering of precise information on the three-body interatomic potential. From previous experiments on Na2 we know that the presence of the liquid helium perturbs the spectra by a very small amount 关see J. Higgins et al., J. Phys. Chem. 102, 4952 共1998兲兴. Ab initio potential energy calculations are carried out at 42 geometries of the lowest quartet state using the coupled cluster method at the single, double, and noniterative triple excitations level 关CCSD共T兲兴. The full potential energy surface is obtained from the ab initio points using an interpolation procedure based on a Reproducing Kernel Hilbert Space 共RKHS兲 methodology. This surface is compared to a second, constructed using an analytical model function for both the two-body interaction and the nonadditivity correction. The latter is calculated as the difference between the CCSD共T兲 points and the sum of the two-body interactions. The bound vibrational states are calculated using the two potential energy surfaces and are compared to the experimentally determined levels. The calculated bound levels are combined with an intensity calculation of the 2⬙ mode of E ⬘ symmetry derived from a Jahn–Teller analysis of the excited electronic state. The calculated frequencies of 1⬙ and 2⬙ are found to be 37.1 cm⫺1 and 44.7 cm⫺1, respectively, using the RKHS potential surface while values of 37.1 cm⫺1 and 40.8 cm⫺1 are obtained from the analytical potential. These values are found to be in good to fair agreement with those obtained from the emission spectrum and to be significantly different from any values calculated from additive potential energy surfaces. The 1 4 A ⬘2 Na3 potential energy surface is characterized by a D 3h symmetry minimum of ⫺850 cm⫺1 共relative to the three 3 2 S Na atom dissociation limit兲 with a bond distance of 4.406 Å. This bond distance differs by about 0.8 Å from the value of 5.2 Å found for the sodium triplet dimer. This means that approximately 80% of the binding energy at the potential minimum is due to three-body effects. This strong nonadditivity is overwhelmingly due to the deformability of the valence electron density of the Na atoms which leads to a significant decrease of the exchange overlap energy in the trimer. © 2000 American Institute of Physics. 关S0021-9606共00兲01813-4兴 and later by Wells and Wilson7 and by Chalasiński and collaborators8 and confirmed, again theoretically, by the accurate quantum Monte Carlo calculations of Mohan and Anderson on He3. 9 Recently, highly accurate calculations by Lotrich and Szalewicz of the binding energy of solid argon showed that, if agreement with experiment was to be reached, nonadditive exchange contributions had to be included.10,11 A recent, rather comprehensive, review on many-body effects in intermolecular forces has been published by Elrod and Saykally.12 Three-body interactions in the noble gases are not easily quantified because of two main reasons. The first is their small size which is due, in part, to cancellations arising from I. INTRODUCTION Since the pioneering work of John Barker,1 the independent knowledge of realistic two-body interactions in the rare gases has allowed an estimate of the contribution of the three-body forces to the properties of these substances in the condensed state. While in the earlier period it was believed that the triple dipole dispersion term 共the so-called Axilrod– Teller and Muto term兲 was sufficient,2 later high pressure research led to the realization that many-body effects could alter the exchange repulsion in an appreciable way3,4 in agreement with an early suggestion of Jansen.5 A convincing analysis of the importance of the three-body component of the exchange repulsion was provided theoretically by Bulski6 0021-9606/2000/112(13)/5751/11/$17.00 5751 © 2000 American Institute of Physics 5752 J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 the different signs of the main components. The total threebody effect summed over all terms is indeed reduced to be about 5% of the total binding energy, which is why cancellation of errors have marked the evolution of this field until not long ago.12 The second reason is of an experimental nature. It is in fact well known that the most precise information on the two-body intermolecular potential for argon has been obtained by measuring the v.u.v. spectrum of the argon dimer.13 However, the v.u.v. spectrum of an Ar trimer has not yet been obtained and, if obtained, would be quite difficult to analyze. Spectra of nonsymmetric trimers such as Ar2 – HF or Ar2 – Kr have instead been obtained in easier regions of the electromagnetic spectrum 共IR and microwave, respectively兲 and have been used to study three-body forces.14–16 However, the three-body terms in nonsymmetric trimers are much more numerous which makes their study substantially more complicated.17 Atoms with a low ionization potential such as those that fill the first two columns of the periodic table offer two main advantages for these types of studies. In the first place they are much more compressible than the noble gases, showing, therefore, much larger exchange effects. Furthermore, their spectra are located in a very convenient region of the electromagnetic spectrum. Not by chance the first theoretical evidence on the existence of three body exchange effects was gathered by Bulski on Be3. 6 On the other hand, while Be3 and Mg3 have at least partially closed electronic shells, Li3 , Na3, etc. can be considered van der Waals systems only in their lowest quartet 共spin polarized兲 states which are normally next to impossible to prepare even in a molecular beam environment. However, quite recently, in extending the spectroscopic use of large helium clusters18,19 to alkali atoms, we have shown how this new kind of matrix spectroscopy introduces very small perturbations and can be used to prepare and characterize very unstable species such as alkali triplet dimers20 and quartet trimers.21 In the latter paper, upon excitation of the 2 4 E ⬘ electronic state of Na3, laser induced fluorescence excitation and emission spectra were obtained, confirming that, in this highly deformable, one valence electron atom, nonadditivity effects are very important. In the present paper, the data obtained in the experiment described in Ref. 21 are presented in a more extensive way and the potential used in their analysis is improved by increasing the number of points and geometries at which high level ab initio calculations are performed, and by using a general, multidimensional, reproducing kernel interpolation procedure. In this way very precise information on the threebody interactions is obtained, making this symmetric trimer a benchmark system in the study of nonadditivity effects. II. THE 2 4 E ⬘ \1 4 A 2⬘ EMISSION SPECTRUM OF Na3 The experimental apparatus used to obtain spectroscopic information on the lowest quartet state of Na3 has been described before.21,22 Here we shall only briefly sketch its main characteristics. Helium clusters ranging in average size up to 104 atoms per cluster are prepared in an expansion of helium gas through a 10 m nozzle at a temperature of 17.5 K and a stagnation pressure of 5.4 MPa. A short distance after the Higgins et al. FIG. 1. The solid line is the emission spectrum of the 2 4 E ⬘ →1 4 A 2⬘ transition of Na3 on a helium cluster surface. The dashed line is the excitation spectrum obtained by collecting the total fluorescence. The excited vibronic levels are marked with arrows. Experimental conditions were 53.4 bar He stagnation pressure and 17.5 K nozzle temperature 共11 000 atoms/cluster before pickup兲. beam is collimated by a 400 m skimmer the clusters cross a pickup cell that is connected to an alkali reservoir. A few centimeters downstream, the doped clusters cross a tunable continuous wave dye laser beam at the center of a two mirror laser induced fluorescence 共LIF兲 collector. Emission spectra were obtained using a monochromator coupled to a liquid nitrogen cooled charge coupled detector. To obtain the emission spectrum of the 2 4 E ⬘ ←1 4 A 2⬘ transition the dye laser frequency was positioned at 15 817.8 cm⫺1 on the strongest band of the excitation spectrum21 which is shown as a dashed line in Fig. 1. The resulting emission spectrum is shown as a continuous line in Fig. 1. The strongest fluorescence occurs at about 15 792 cm⫺1, which is 35 cm⫺1 to the red of the excitation. Since the laser is exciting the 0-0 band of a pure electronic transition of Na3, the red shift must result from the inelastic energy transfer to the elementary excitations of the helium cluster and to the solvent reorganization. The full width at half maximum 共FWHM兲 of the 0-0 emission band 共approximately 10 cm⫺1兲 is a consequence of the surface location of the trimer on the helium cluster as electronic chromophores excited inside the liquid droplets have typically much broader excitation spectra. Emission spectra were also acquired at excitation energies above the 0-0 band. Figure 2 displays the resonant emission measured upon excitation of the second, third, fourth, and sixth bands of the excitation spectrum. Jahn–Teller analysis of the excited state allows precise quantum numbers to be assigned to these transitions 共see Fig. 1兲. As can be seen from the spectra, the emission is in all cases dominated by transitions arising from the lowest vibronic level of the excited state which produces a spectrum similar to that obtained upon excitation to the 0-0 level. Indeed, the 0-0 band near 15 791 cm⫺1 is the most prominent feature in all the spectra. Some emission from the v ⫽1 – 3 vibronic levels in the excited state is observed at higher frequencies than the 0-0 transition, but this only constitutes a minor component of J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 Exchange effects in three-body interactions 5753 TABLE I. Peak positions of the resonant fluorescence channel of the 2 4 E ⬘ →1 4 A ⬘2 transition of Na3 excited at different ( v , j) vibronic levels. Peak number in spectrum 1 2 3 4 5 6 7 8 9 10 11 Position 共cm⫺1兲 Excitation at first band 共0,0兲 15 791.39 15 750.74 15 710.30 15 667.45 15 627.44 15 593.68 15 561.78 15 522.30 15 490.69 15 450.73 15 421.95 Difference 共cm⫺1兲 40.65 40.44 42.85 40.01 33.76 31.90 39.48 31.61 39.96 28.78 Excitation at second band 共1,1/2兲 共15 866.0 cm⫺1兲 1 15 868.84 50.83 2 15 818.01 23.97 3 15 794.05 40.66 4 15 753.39 41.33 5 15 712.06 41.99 6 15 670.07 FIG. 2. Resonant emission spectra obtained by exciting different vibronic levels of the 2 4 E ⬘ →1 4 A ⬘2 transition of Na3. The quantum numbers v , j of the excited 2 4 E ⬘ state are given in the upper right hand corner of each panel. The laser excitation frequency is marked by an asterisk. The labeling of the peaks in the spectra are given as the excited v of the 2 4 E ⬘ state and the vibrational levels of the lower 1 4 A 2⬘ state. the emission. These spectra reveal the presence of vibrational cooling in the excited quartet state similarly to what occurs in the spectroscopy of the alkali dimers on the surface of helium clusters.20 Vibrational cooling must occur on a relatively fast time scale since most of the population in the excited state relaxes to the lowest vibronic state within the fluorescence lifetime of the excited quartet state which has been determined to be less than 1.5 ns using time resolved spectroscopy.23 The positions of the emission spectra bands are given in Table I. The 1 4 A ⬘2 state of Na3 has three fundamental vibrational modes, a symmetric stretch of A 1⬘ symmetry and a doubly degenerate mode of E ⬘ symmetry. Without vibronic coupling, fluorescence emission from levels in the 2 4 E ⬘ state could lead into any vibrational level of the symmetric stretch of the 1 4 A ⬘2 state, but only into even numbered quanta of the degenerate mode. Due to Jahn–Teller interaction in the 2 4 E ⬘ state, decay into all vibrational states becomes allowed. As mentioned above, the emission spectra are dominated by fluorescence originating from the v ⫽0 level of the excited state due to vibrational cooling. Hence, the analysis will focus only on the data obtained by exciting the v ⫽0 level of the 2 4 E ⬘ state. The line positions provide accurate spectroscopic data on the vibrational levels of the 1 4 A 2⬘ state of Na3 which will be used to test the accuracy of ab initio potential energy surfaces calculated for the lowest quartet Excitation at third band 共2,5/2兲 共15 932.6 cm⫺1兲 1 15 897.52 32.26 2 15 865.26 47.25 3 15 818.01 23.08 4 15 794.93 41.55 5 15 753.39 37.82 6 15 715.56 1 2 3 Excitation at fourth band 共2,1/2兲 15 955.2 cm⫺1 15 865.26 47.25 15 818.01 23.97 15 794.05 1 2 3 4 Excitation at sixth band 共3,1/2兲 16 038.0 cm⫺1 15 906.51 40.35 15 866.16 47.26 15 818.90 23.10 15 795.82 state of the sodium trimer assessing the importance of the nonadditive terms. The bound states of the potential energy surfaces will be calculated and compared to the vibrational levels obtained from the emission spectrum. III. THE POTENTIAL ENERGY SURFACE OF Na3 IN THE LOWEST QUARTET STATE „1 4 A 2⬘ … In this section, two global potential energy surfaces for the lowest quartet state 1 4 A 2⬘ of the Na3 trimer are introduced. One is based on a general multidimensional reproducing kernel 共RK兲 interpolation procedure,24–28 while the other is based on a fit to a Hatree–Fock plus Damped Dispersion 共HFD兲 analytic form for the two-body interaction and to a damped triple exponential for the nonadditive three-body interaction. Both employ a standard many-body expansion ansatz based on high level ab initio calculations at the coupled cluster level of theory with single, double, and noniterative triple excitations 关CCSD共T兲兴.29 While the three-body CCSD共T兲 interaction has been obtained as described in the 5754 Higgins et al. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 TABLE II. Calculated CCSD共T兲 potential of the 共a兲 1 3 ⌺ ⫹ u state of Na2. R 共Å兲 V(R) 共cm⫺1兲 R 共Å兲 V(R) 共cm⫺1兲 2.0 2.5 3.0 3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 16 304.56 7292.56 2978.40 1289.37 1019.36 793.37 604.68 447.59 317.40 210.11 122.37 51.22 ⫺6.0 ⫺51.57 ⫺87.52 ⫺115.51 ⫺136.89 4.8 4.9 5.0 5.0911 5.2 5.3 5.4 5.7 6.1 6.6 7.1 8.1 9.1 10.1 11.1 12.1 ⫺152.73 ⫺163.88 ⫺171.09 ⫺174.76 ⫺176.17 ⫺175.15 ⫺172.38 ⫺157.21 ⫺130.14 ⫺96.87 ⫺69.61 ⫺34.47 ⫺17.18 ⫺8.95 ⫺4.90 ⫺2.84 with this basis set differ by less than 0.03 Å and 3 cm⫺1, respectively, from our most accurate results reported for the dimer in Ref. 30. Our ab initio calculations for the trimer, shown in Table III, were performed at 42 carefully selected geometries which cover the weakly attractive region, the van der Waals minimum, and the repulsive region of the potential energy surface. We thoroughly covered the equilateral triangle geometry 共18 D 3h points兲, but we also included numerous bent structures 共16 C 2 v points兲, linear structures 共4 D ⬁h points兲, and low symmetry structures 共4 C s points兲. B. Reproducing Kernel Hilbert Space „RKHS… interpolation surface This interpolation procedure is based on a recently formulated one-dimensional 共1-D兲 distancelike reproducing kernel that possesses a correct asymptotic inverse power form: ⫺m⫺1 q n,m 共 x,x ⬘ 兲 ⫽n 2 B 共 m⫹1,n 兲 x ⬎ 2F 1 next section, the two-body lowest triplet Na–Na interaction, based on a larger basis set, has recently been published by one of us30 and is reported here in Table II. A. Ab initio calculation of the three-body interaction The full three-body interaction was calculated at the coupled cluster level of theory with single, double, and noniterative triple excitations 关CCSD共T兲兴.29,31 All electrons were correlated with the exception of those which occupy the atomic 1s levels. First, the total interaction energy was determined for the sodium trimer by subtracting atomic energies E at from the trimer energy E trim E int共 r 12 ,r 23 ,r 31兲 ⫽E trim共 r 12 ,r 23 ,r 31兲 兺 E iat共 r 12 ,r 23 ,r 31兲 , i⫽1 共4兲 where n and m are integers with n⭓1 and m⭓0, x ⬎ , x ⬍ are, respectively, the larger and smaller of x and x ⬘ , B is the Beta function, and 2 F 1 is the Gauss’ hypergeometric function. It is instructive to note that the reproducing kernel q n,m (x ⬍ ,x ⬎ ) ⫺(m⫹k⫹1) is a finite sum of inverse powers x ⬎ , k⫽0,...,(n ⫺1). This property enables one to correctly mimic the asymptotic forms of both the two- and three-body interactions as described below. The total potential energy surface V(r 12 ,r 23 ,r 31) of the lowest quartet state 1 4 A 2⬘ of Na3 can be written as a manybody expansion: 共1兲 where r i j is the interatomic distance between the ith and jth atoms and E at depends on r i j ’s through the function counterpoise procedure, which is applied to ensure basis set consistency between the trimer and monomer calculations.32–38 Next, the three-body term ⑀ was extracted from E int by subtracting the pairwise interactions ⑀ i j 3 ⑀ ⫽E int⫺ 兺 ⑀ i j , 共2兲 ij ⑀ i j ⫽E dim ⫺E iat⫺E atj 共3兲 i⬍ j 冊 x⬍ , x⬎ V 共 r 12 ,r 23 ,r 31兲 ⫽V 1 ⫹V 2 共 r 12兲 ⫹V 2 共 r 23兲 ⫹V 2 共 r 31兲 3 ⫺ 冉 ⫻ ⫺n⫹1,m⫹1;n⫹m⫹1; with every term on the right-hand side of Eq. 共3兲 being also calculated using the basis set of the trimer. We have used Sadlej’s medium-size polarized basis set, which was originally designed for calculations of molecular electric properties,39 modifying it to make it more suitable for calculations of this weak intermolecular potential. First, we uncontracted the more diffuse set of d functions. Second, we added one f and one g diffuse functions with exponents of 0.0641 and 0.079 36 a.u., respectively, which were obtained 30 by optimizing the dispersion interaction of the 3 ⌺ ⫹ u dimer. The CCSD共T兲 values of R e and D e for the dimer obtained ⫹V 3 共 r 12 ,r 23 ,r 31兲 , 共5兲 where V 1 , V 2 , and V 3 are, respectively, the one-, two-, and three-body terms. For simplicity, the one-body term, which may be assigned the value of the dissociation energy in the three atom limit Na⫹Na⫹Na is taken to be zero. The twobody terms, V 2 , where R⫽r 12 , r 23 , or r 31 , may be expressed as: N V 2共 R 兲 ⫽ 兺i ␣ 共i 2 兲q n,m共 i , 兲 共6兲 in terms of the 1-D distancelike reproducing kernel q n,m ( , ⬘ ). Here the reduced internuclear distance is defined as ⫽ 冉冊 R S 2 共7兲 with S being an arbitrary positive scaling factor chosen to keep the expansion coefficients ␣ i small. The index i runs over the ab initio data points. The two-body form has the following asymptotic form at long distances: n⫺1 lim V 2 共 R 兲 ⫽⫺ R→⬁ 兺 k⫽0 C 2 共 k⫹m⫹1 兲 , R 2 共 k⫹m⫹1 兲 共8兲 J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 Exchange effects in three-body interactions 5755 TABLE III. The CCSD共T兲 ab initio values of the three-body and total potential for the lowest quartet state 1 4 A ⬘2 of the trimer Na3. Geometry r 12 共Å兲 r 23 共Å兲 r 31 共Å兲 V tot 共cm⫺1兲 D 3h 2.0 2.5 3.0 3.6 3.9 4.2 4.3 4.4 4.5 4.6 4.8 5.0911 5.7 6.1 6.6 7.1 7.6 8.1 2.0 2.5 3.0 3.6 3.9 4.2 4.3 4.4 4.5 4.6 4.8 5.0911 5.7 6.1 6.6 7.1 7.6 8.1 2.0 2.5 3.0 3.6 3.9 4.2 4.3 4.4 4.5 4.6 4.8 5.0911 5.7 6.1 6.6 7.1 7.6 8.1 ⫺8876.46 ⫺6718.70 ⫺4510.49 ⫺2328.16 ⫺1545.93 ⫺972.61 ⫺824.24 ⫺694.64 ⫺582.46 ⫺485.86 ⫺333.47 ⫺186.20 ⫺47.59 ⫺16.69 ⫺2.99 0.50 1.12 0.92 41550.46 15675.49 4631.05 113.57 ⫺564.26 ⫺800.20 ⫺826.20 ⫺835.87 ⫺833.59 ⫺822.36 ⫺783.08 ⫺703.10 ⫺516.13 ⫺406.53 ⫺293.92 ⫺208.18 ⫺145.43 ⫺101.23 C 2v 2.4625 4.6281 6.5254 3.0098 5.6565 7.9755 3.4825 6.5450 9.2282 3.8990 7.3278 10.3319 4.1726 7.8420 4.51 8.4848 3.6 3.6 3.6 4.4 4.4 4.4 5.0911 5.0911 5.0911 5.7 5.7 5.7 6.1 6.1 6.6 6.6 3.6 3.6 3.6 4.4 4.4 4.4 5.0911 5.0911 5.0911 5.7 5.7 5.7 6.1 6.1 6.6 6.6 ⫺3481.83 ⫺1384.35 ⫺253.02 ⫺1330.77 ⫺300.46 ⫺67.29 ⫺464.16 ⫺60.38 ⫺21.51 ⫺155.99 ⫺10.46 ⫺7.45 ⫺69.44 ⫺1.91 ⫺21.50 0.95 6102.36 126.20 1272.74 1570.54 ⫺551.99 ⫺199.67 286.91 ⫺505.29 ⫺381.94 ⫺138.36 ⫺382.50 ⫺327.67 ⫺253.06 ⫺302.88 ⫺303.51 ⫺219.04 D ⬁h 7.2 8.8 10.1822 11.4 3.6 4.4 5.0911 5.7 3.6 4.4 5.0911 5.7 ⫺66.85 ⫺62.29 ⫺27.24 ⫺10.97 1493.05 ⫺178.78 ⫺380.32 ⫺327.40 4.558 784 4.538 180 4.677 665 4.818 342 4.323 550 4.402 865 4.411 442 4.425 702 ⫺691.04 ⫺691.04 ⫺680.59 ⫺663.29 ⫺816.67 ⫺815.42 ⫺753.73 ⫺648.48 Cs 4.322 079 4.263 256 4.128 087 3.994 653 M where the C coefficients take on the form: V 3 共 r兲 ⫽ n 2 B 共 m⫹1,n 兲共 ⫺n⫹1 兲 k 共 m⫹1 兲 k C 2 共 k⫹m⫹1 兲 ⫽⫺ 共 n⫹m⫹1 兲 k k! 2 共 m⫹1 兲 兺 ␣ 共i 2 兲R 2ki , i⫽1 兺 j⫽1 ␣ 共j 3 兲 再 1 P j Q n,m 共 r j ,r兲 3! 兵 123其 兵 123其 兺 冎 共10兲 in terms of a 3-D reproducing kernel defined as a tensor product: N ⫻S V 3 共cm⫺1兲 共9兲 Q n,m 共 r,r⬘ 兲 ⫽q n,m 共 x,x ⬘ 兲 q n,m 共 y,y ⬘ 兲 q n,m 共 z,z ⬘ 兲 共11兲 with reduced bond distances where (a) k is the Pochammer’s symbol. In the numerical implementation, we set n⫽2 and m⫽2 to yield a smooth interpolation through the ab initio data points and to correctly account for the two leading long-range dispersion interactions ⫺C 6 /R 6 and ⫺C 8 /R 8 . The three-body term V 3 (r), where r⫽ 兵 r 12 ,r 23 ,r 31其 , can be expressed as x⫽r 12 /S, y⫽r 23 /S, z⫽r 31 /S. 共12兲 The index j runs over the three-body ab initio data. P 兵j 123其 is the permutation operator with respect to an arbitrary permutation of the three integers 1, 2, and 3. The summation over the subscript 兵123其 is performed over all 3! symmetry permutations of the trimer bond lengths associated with the 5756 Higgins et al. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 FIG. 3. RKHS potential energy curve V 2 (R) of the Na2 共a兲 1 3 ⌺ ⫹ u state, obtained by interpolating, using the reproducing kernel expansion in Eq. 共6兲, high-level 共basis 1兲 CCSD共T兲 ab initio data 共see Table II兲. The potential can be characterized by: R e ⫽5.20 Å, D e ⫽176.17 cm⫺1, ⫽4.29 Å, C 6 ⫽7.6477⫻106 Å 6 , and C 8 ⫽1.8527⫻108 cm⫺1 Å 8 . jth data point of the potential surface. V 3 is then explicitly symmetric with respect to the exchange of any two sodium atoms. In the calculation, n is set to 2 for a smooth interpolation and m⫽2 is chosen such that the leading three-body term asymptotically behaves as: V 3 共 r 12 ,r 23 ,r 31兲 ⬇ 1 共13兲 3 3 3 r 12 r 23r 31 in line with the behavior of the lowest-order long-range triple and ␣ (3) dipole interaction. The expansion coefficients ␣ (2) i j can be obtained by solving the following linear algebraic equations: N V 2共 R k 兲 ⫽ 兺i ␣ 共i 2 兲q n,m共 i , k 兲 , M V 3 共 rk 兲 ⫽ 兺 ␣ 共j 3 兲 j⫽1 再 共14兲 冎 1 P j Q n,m 共 r j ,rk 兲 , 3! 兵 123其 兵 123其 兺 共15兲 where V 2 (R k ) and V 3 (rk ) are the two- and three-body CCSD共T兲 ab initio data tabulated in Tables II and III. The two-body interaction V 2 (R) of the Na2 a 3 ⌺ ⫹ u state obtained from interpolating the CCSD共T兲 ab initio calculations 共Table II兲 using Eq. 共6兲 is reported in Fig. 3. This new potential energy curve possesses a well depth D e ⫽176.17 cm⫺1 at the equilibrium position R e ⫽5.20 Å and crosses the zero value at the distance ⫽4.29 Å. Moreover, the calculated C 6 ⫽7.65⫻106 cm⫺1 Å 6 and C 8 ⫽1.85 ⫻108 cm⫺1 Å 8 关cf., Eq. 共9兲兴 agree well with the values derived in Ref. 40 fitting the outer part of the RKR results of 7.89⫻106 cm⫺1 Å 6 and 2.12⫻108 cm⫺1 Å 8 , respectively. The three-body interaction V 3 (r) for the lowest quartet state (1 4 A ⬘2 ) of Na3 was obtained from interpolating the FIG. 4. A view of the nonadditivity of the interaction energy in the lowest quartet state 1 4 A 2⬘ of Na3, as a function of bond length in the D 3h geometries. The solid curve is the nonadditivity of the RKHS potential while the dashed curve is calculated at the HF-SCF level. CCSD共T兲 ab initio calculations 共Table III兲 using Eq. 共10兲. In Fig. 4 the nonadditivity is displayed as a function of the bond length R at the D 3h geometry showing a very rapid drop in magnitude toward zero beyond R⫽7.0 Å. The CCSD共T兲 calculations indicate that there may be a very small barrier at R⫽7.6 Å 共see Table III兲 which is expected, as the longrange Axilrod–Teller–Muto 共ATM兲 triple dipole nonadditivity41,42 in the equilateral geometry is repulsive while the dominating term at shorter distance is strongly attractive. Figure 5 shows the contour plot of the total RKHS potential V(r 12 ,r 23 ,r 31) of the lowest quartet state 1 4 A ⬘2 of the Na3 trimer as function of two bond lengths r 12 and r 31 , with the angle between these two bond lengths fixed at 60 deg.43 The resulting Na3 potential energy surface has a global minimum value of ⫺849.37 cm⫺1 at the D 3h geometry with the bond distance R⫽4.406 Å, which is much smaller than the potential minimum position R e ⫽5.2 Å of the corresponding Na2 dimer potential curve 共see Fig. 3兲. A detailed comparison of the two- and three-body contributions to the total potential energy surface V(r), Eq. 共5兲, in the D 3h geometry is given in Fig. 6. It is found that about 80% of the total potential energy around the potential minimum (r 12⫽r 23 ⫽r 31⫽4.4 Å) comes from three-body contributions. In general, the two-body contribution dominates at geometries with small 共e.g., ⬍3.6 Å兲 and large 共e.g., ⬎4.8 Å兲 interatomic distances, whereas the three-body term is more important in the intermediate region that includes the potential minimum. C. The Hartree–Fock Damped Dispersion „HFD… plus CCSD„T… surface In our second potential energy surface the total interaction energy for the trimer was represented analytically as a sum of the two-body and three-body terms. The two-body J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 Exchange effects in three-body interactions ⌬E CORR共 R 兲 ⫽⫺ 冋兺 n 册 C ng n共 R 兲 f 共 R 兲, Rn 2 R 2 / 冑n 兲 n 兴 , f 共 R 兲 ⫽1⫺ R 1.68e ⫺0.78 R . FIG. 5. Contour plot of the total RKHS potential energy surface V(r) of the lowest quartet state 1 4 A 2⬘ of Na3, as a function of the two bond distances with included angle equal to 60 degrees. The corresponding HFD⫹CCSD共T兲 potential energy surface is not shown here because it only slightly differs from its RKHS counterpart. The total RKHS potential energy surface possesses a minimum value of ⫺849.37 cm⫺1 at r 12⫽r 23⫽r 31 ⫽4.406 Å. The contours start at the dotted line 共corresponding to ⫺100 cm⫺1兲 at the upper right corner, then decrease by 100 cm⫺1 per line until the innermost circle. Solid contour lines start at zero 共right next to the outer most dotted line兲 with an increment of 100 cm⫺1 per line. 共18兲 共19兲 n⫽3, 6, 8, 10, . . . , g n 共 R 兲 ⫽ 关 1⫺e 共 ⫺2.1 R/n 兲 ⫺ 共 0.109 5757 共20兲 共21兲 The universal damping functions f and g scale with , which was determined by using the ionization potential of Na. The long-range dispersion coefficients (C6, C8, C10) calculated by Marinescu et al.45 for Na(3 2 S)⫹Na(3 2 S) were used. The HF-SCF energy was calculated with the GAUSSIAN 46 92 suite of programs using the 6-311G** basis set. No further fitting was performed as in this way, the resulting potential energy curve is very similar to both the RKR curve and the ab initio CCSD共T兲 potential. The HFD two-body functional form was used to define the nonadditive potential surface V 3 via the relationship total V 3 共 r 12 ,r 23 ,r 31兲 ⫽V CCSD共T兲 共 r 12 ,r 23 ,r 31兲 ⫺ 兵 V 2 共 r 12兲 ⫹V 2 共 r 23兲 ⫹V 2 共 r 31兲 其 . 共22兲 The so-obtained ab initio three-body data were fit to an empirical formula47,48 that approaches zero at long range: V 3 共 r 12 ,r 23 ,r 31兲 ⫽A 3 f 共 r 12兲 e ⫺ ␥ r 12 f 共 r 23兲 e ⫺ ␥ r 23 f 共 r 31兲 e ⫺ ␥ r 31, 共23兲 where interaction energy between two sodium atoms (V 2 ) with parallel spins was represented by the Hartree–Fock Damped Dispersion 共HFD兲 model suggested by Douketis et al.44 According to this model, the total interaction energy between the two atoms (V 2 (r)) is split into a contribution calculated at the Hartree–Fock self-consistent field level (⌬E SCF , which ignores electron correlation兲 and a damped multipole expansion which accounts for dispersion interactions (⌬E CORR). V 2 共 R 兲 ⫽⌬E SCF共 R 兲 ⫹⌬E CORR共 R 兲 , ⌬E SCF共 R 兲 ⫽A 1 R ␣ 1 e ⫺  1 R ⫹A 2 R ␣2e ⫺2R, 共16兲 共17兲 FIG. 6. Comparison of the total RKHS potential energy surface of the 1 4 A ⬘2 state of Na3 with its two-body contribution in the D 3h geometry. Solid curve: the total potential surface, dotted curve: the two-body contribution, and dashed curve: the potential energy curve of the 共a兲 1 3 ⌺ ⫹ u state of Na2. f 共 r 兲 ⫽1⫺r e ⫺r 共24兲 is a damping function used to reproduce the correct form of the nonadditivity at short internuclear separations. A 3 , ␥, , and are adjustable parameters. The final HFD⫹CCSD共T兲 surface is obtained by summing the fitted two-body HFD terms 关cf., Eq. 共16兲兴 and the fitted three-body surface of Eq. 共23兲. It is found that two fitted HFD⫹CCSD共T兲 surfaces using 共a兲 only the D 3h data and 共b兲 using all of the available data differ only slightly from each other. The result based on the D 3h data is a potential whose binding energy, relative to three free ground state Na atoms, is ⫺835 cm⫺1 at an internuclear separation of 4.4 Å. The parameters of this potential surface of Na3 are listed in Table IV. The HFD⫹CCSD共T兲 potential surface along the symmetric stretch coordinate behaves very similarly to the RKHS surface which is displayed in Fig. 5. We note that the HFD⫹CCSD共T兲 surface does not contain a three-body long-range component of the ATM type. The fact that, as we will see later, this approximation leads to reasonably good agreement with experiment is due to the fact that the dominating three-body nonadditivity is already present at the self-consistent field 共SCF兲 level of theory where no electron correlation is included 共see Fig. 4兲. The nonadditivity of the potential energy is the result of a decrease in the overlap repulsion in the trimer as the third sodium atom 共with a parallel electron spin兲 approaches the triplet dimer. Since the valence electron clouds of the sodium atoms are highly deformable, the deformation induced by the presence of a third sodium atom is significant. This deformation allows the contraction of the bond distance in the trimer 5758 Higgins et al. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 TABLE IV. Parameters of the HFD potential energy surface of the 共a兲 4 1 3⌺ ⫹ u state of Na2 and of the HFD⫹CCSD共T兲 potential of the 1 A ⬘ 2 state of Na3. Values are given for r in Å and V in cm⫺1. Na2HFD Parameter Value A1 1 ␣1 A2 2 ␣2 C6 C8 C 10 De Re 1.010⫻106 5.7348 4.3736 9.6005⫻104 1.7342 1.7861 7.0789⫻106 1.509 85⫻108 4.181 65⫻109 0.5261 ⫺178.13 5.078 4.192 Na3 HFD⫹CCSD共T兲 ⫺1.3082⫻106 0.5869 25.575 9.8821 A3 ␥ and is accompanied by the consequent increase in binding energy. This increase is responsible for the relatively large difference between the additive two-body contribution to the binding energy of the trimer and the actual total binding energy including nonadditive effects. The RKHS method produces a potential surface that differs slightly from the HFD⫹CCSD共T兲 surface and can better describe the surface over a wider range of geometries. D. Bound-state calculations The vibrational levels of the 1 4 A ⬘2 state were calculated using the TRIATOM suite of programs.49 This program uses a variational approach utilizing a finite basis of Morse oscillators. The Hamiltonian for the triatomic system was set up using body-fixed coordinates (r 1 ,r 2 , ) where r 1 , r 2 are two bond lengths of the trimer and is the included angle. The angular basis functions are symmetrized rotation matrix elements: 兩 j,k,p 典 ⫽ 1 & 关 ⌰ jk 兩 JM K 典 ⫹ 共 ⫺1 兲 p ⌰ jk 兩 JM ⫺K 典 ], 共25兲 where ⌰ is an associated Legendre polynomial and p is the parity. Motion in the radial coordinate is carried out using a Morse oscillator basis set 兩 n 典 ⫽ 冑 N n ␣ L ␣n 共 y 兲 e 共 ⫺1/2兲 y y 共 ␣ ⫹1兲/2, 共26兲 where y⫽Ae ⫺  (r⫺r e ) , A⫽4D e /  ,  ⫽ e ( /2D e ) 1/2, and ␣ ⫽integer(A) are the Morse parameters and N n ␣ L n (y) is an associated LaGuerre polynomial. The total wave function for the lth energy level becomes: ⌿ l⫽ d kJljmn 兩 jk 典 兩 m 典 兩 n 典 . 兺k 兺 jmn 共27兲 FIG. 7. Calculated vibrational levels of the ab initio potential energy surfaces of the 1 4 A 2⬘ state of Na3. The top panel displays the experimental emission spectrum with the energy referenced to the ground vibrational level of the 1 4 A 2⬘ state. The middle panel is the vibrational levels calculated using the RKHS potential surface obtained by fitting the CCSD共T兲 ab initio data with the RKHS method. The lower panel are the levels calculated with the HFD⫹CCSD共T兲 potential surface constructed of the HFD two-body interaction and a damped triple exponential fit to the nonadditivity. A Morse oscillator basis set consisting of 3000 functions was used and the Hamiltonian was diagonalized to give the corresponding eigenvalues. The comparison of the energy levels calculated with the HFD⫹CCSD共T兲 three-body potential surface, the RKHS surface, and the experimental emission spectrum is given in Fig. 7. The first peak in the emission spectrum is the 0-0 band and is set as the zero of energy. The second prominent band is composed of two vibronic levels, 1 1 and 1 2 . In the absence of three-body forces, a symmetry analysis of the normal modes reveals that the asymmetric stretch 2 should be exactly 1/2 of the frequency of the symmetric stretch 1 . In the case of quartet Na3, the three-body forces constitute such a large part of the binding energy that the frequency of 2 actually becomes larger than that of 1 . The frequency of 2 therefore becomes a sensitive probe of the magnitude of the three-body forces present in the molecule. By calculating the energy of the two fundamental modes using a number of different potential energy surfaces, it is found that there is a direct correlation between the magnitude of the nonadditive component of the potential and the frequency of 2 . As the nonadditivity increases, the relative ratio of 2 to 1 increases. The calculated vibrational frequencies of the two normal modes are 1 ⫽37.1 cm⫺1 for the symmetric stretch and 2 ⫽40.8 cm⫺1 for the degenerate asymmetric stretch using the HFD⫹CCSD共T兲 surface, while the RKHS potential surface yields values of 1 ⫽37.1 cm⫺1 and 2 ⫽44.7 cm⫺1. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 Exchange effects in three-body interactions 5759 TABLE V. Calculated vibrational levels of the 1 4 A 2⬘ state of Na3. The two potential surfaces described in the text were utilized as input for the variational calculation of the vibrational levels. Vibrational Level HFD two-body ⫹CC three-body 共cm⫺1兲 Reproducing kernel fit to CC potential surface 共cm⫺1兲 共0,0兲 共1,0兲 共0,1兲 共2,0兲 共1,1兲 共0,2兲 共3,0兲 共2,1兲 共1,2兲 共0,3兲 共4,0兲 共3,1兲 共2,2兲 共1,3兲 共0,4兲 0.00 37.1 40.8 73.3 76.4 80.3 108.7 111.1 114.6 118.6 143.4 145.9 148.9 153.0 155.1 0.00 37.1 44.7 73.2 80.0 88.1 108.4 114.5 121.8 130.0 143.2 149.0 156.9 162.9 170.9 As the experimental line shape is uncertain we cannot carry out a quantitative comparison between theory and experiment, however, the vibrational frequencies calculated from both surfaces appear to be in good agreement with the measured emission spectrum. Table V lists the calculated vibrational levels of the 1 4 A 2⬘ state of Na3 using the two potential surfaces described previously. The notation ( 1 , 2 ) is used to denote the quantum numbers of the vibrational levels. All energies are referenced to the 共0, 0兲 level of the lowest quartet state. Incomplete convergence of the bound state calculation produces residual splitting of the two components of the asymmetric E ⬘ mode. The splitting does not allow accurate assignment of quantum numbers to calculated levels 200 cm⫺1 above the 0-0 level. The experiment also reveals congestion in the emission spectrum below 15 600 cm⫺1. Below this energy, an average was taken of the energies of the components of the E mode. Above the 共0, 4兲 level, all calculated vibrational levels are displayed in Fig. 7. Overall, the spectrum would allow, in principle, the determination the vibrational levels up to 700 cm⫺1 in energy which accounts for ⬃90% of the binding energy of the quartet trimer 关 D 0 ⬇775 cm⫺1 using the HFD⫹CCSD共T兲 surface, the RKHS surface gives D 0 ⬇783 cm⫺1]. However, to exploit this opportunity, a calculation of comparable quality of the excited state potential energy surface would be necessary, which would be a far from trivial undertaking. The intensities of the transitions into the E ⬘ mode of the lowest 1 4 A 2⬘ quartet state were calculated through a Jahn– Teller analysis of the excited 2 4 E electronic state. The overlap integrals were calculated in the harmonic approximation and the resulting intensities are shown in Fig. 8, normalized to the 0-0 level. A complete modeling of the intensities of all the levels could not be accomplished since the frequency of 1⬘ could not be determined from the excitation spectrum. As stated above, only a full ab initio calculation of the excited state surface would provide such information. Good agreement is obtained between the experimentally determined and calculated emission intensity of the E ⬘ progression. The tran- FIG. 8. Comparison of the experimental emission spectrum and the calculated intensities of the transitions into the E ⬘ vibrational levels of the 1 4 A 2⬘ state of Na3. Panel 共A兲 displays the calculated intensities based on the level positions of the RKHS potential energy surface. Panel 共B兲 uses the level positions of the HFD⫹CCSD共T兲 surface. sitions group into bands for the first five prominent lines that appear in the emission spectrum due to the similar frequencies of the 1⬙ and 2⬙ modes. The first five bands of the emission spectrum beginning with the 0-0 band are composed of one, two, three, four, and five individual lines, respectively. The intensities could only be calculated for the transitions into the E ⬘ mode. For this reason, the intensity of only a single line of each clump involving an overtone level of the 2⬙ mode could be calculated. The intensities calculated using the line positions of the HFD⫹CCSD共T兲 surface appear to better match the experimental spectrum since the transition frequencies tend to clump together near the center of the bands as can be seen in Fig. 7, although better agreement with the absolute positions of the lines is obtained using the values calculated with the RKHS potential energy surface. This is demonstrated in panel A of Fig. 8 where it can be seen that the height of the line near 15 750 cm⫺1 matches the intensity of the shoulder seen in the experimental spectrum. IV. DISCUSSION Several comparisons were made in order to assess the strengths and weaknesses of the RKHS method when applied to the lowest quartet state of Na3. Two versions of the RKHS surface and the HFD⫹CCSD共T兲 surface were made. In each case, the first one 共not shown兲 was based only on the D 3h configurations, while the second one 共cf., Fig. 5兲 was based on all the available ab initio data. This allows effects due to symmetric and asymmetric configurations to be compared. In general, a better agreement is qualitatively observed between the experiment and the levels calculated with the 5760 Higgins et al. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 reproducing kernel interpolation of the CCSD共T兲 potential energy surface. When only D 3h ab initio data are used, the HFD⫹CCSD共T兲 and RKHS results are very similar. The HFD⫹CCSD共T兲 has the advantage of giving a simple intuitive form for the short-range interaction as well, while the RKHS form uses a more general but less physically motivated form which has the advantage of correctly reproducing all the ab initio calculations. When the non-D 3h configurations are added, the difference between the two is noticeably larger, and the RKHS method appears to give better qualitative agreement 共nearly quantitative, although comparison is difficult without information about peak shape or amplitudes兲 with the experimental spectrum. Since the HFD⫹CCSD共T兲 form does not include any three-body terms which attempt to model non-D 3h configurations, the HFD⫹CCSD共T兲 result only changes slightly when the new points are added to the fit. Better agreement would require the addition of new terms to give the function more flexibility. Doing this without sacrificing the intuitive nature of the HFD⫹CCSD共T兲 form would require careful insight into the nature of the interactions being modeled; simply adding ad hoc terms with less transparent physical justification to the HFD⫹CCSD共T兲 potential would sacrifice the intuitiveness that is the strength of the HFD⫹CCSD共T兲 method probably without achieving the accuracy of the RKHS surface. On the other hand, the RKHS method is able to incorporate the new data without any additional effort, after which the agreement with experiment is clearly improved. The reproducing kernel method of interpolation appears to be a good method to produce a complete potential surface from individual points since it possesses the correct asymptotic form at long range and has enough flexibility for incorporating data from non-D 3h geometries. This allows a larger portion of the surface to be calculated even though theoretical data may not be available for those particular geometries. V. CONCLUSION Using laser excitation and emission spectroscopy, the 1 4 A 2⬘ lowest quartet state of Na3 residing on the surface of a liquid He nanodroplet was identified for the first time and its vibrational levels were subsequently obtained. As the perturbations due to the liquid He droplet have been shown to be quite small20 we believe that the observed vibrational structure reflects quite closely the vibrational structure of those species in the gas phase. The ab initio potential energy surface of the 1 4 A 2⬘ state of Na3 displays strong nonadditivity effects. In particular, the experimentally determined asymmetric stretch frequency in the lowest quartet state provides a sensitive measure of the three-body contributions to the potential energy. An increased nonadditive component causes the frequency of 2 to increase from its expected value of 1/2⫻ 1 . With the aid of ab initio calculations, a potential surface has been obtained for the lowest quartet state of Na3. The validity of the surface was confirmed by comparing the calculated vibrational states of the theoretical potential energy surface with the experimental levels. The nonadditive contri- butions to the interaction energy of this homonuclear triatomic system have been found to be large. In fact, the equilibrium distance of the trimer is about 15% smaller than the bond length of the dimer. This large three-body contribution is caused by a decreased overlap repulsion of the electrons in the trimer which is due to the highly deformable valence electron shells of the interacting sodium atoms. The lowest quartet state of alkali trimers represents somewhat of an extreme case of atomic deformability in which three-body effects are large and almost exclusively due to exchange overlap effects. This contrasts the case of the noble gases in which the relatively small ATM dispersion contributions are instead larger than those due to electron overlap. For instance, the equilibrium distance calculated for Ar3 is the same as the bond length in the dimer.50 As the rest of the atoms in the periodic table have ionization potentials, and therefore valence electron distribution deformability, which are intermediate between the two groups mentioned above, we now have a more precise feeling for what to expect when three-body effects in homonuclear systems are considered. ACKNOWLEDGMENTS J.H., J.R., K.K.L, and G.S. were supported by AFOSR 共HEDM program兲. T.H.T.-S.H., and H.R. were supported by the Department of Energy. M. G. acknowledges support of this work by the Division of Chemical Sciences and the Division of Geosciences and Engineering both of the Office of Science of the US Department of Energy. His work was performed in part under auspices of the U.S. Department of Energy, under Contract No. DE-AC06-76RLO 1830, with Battelle Memorial Institute, which operates the Pacific Northwest National Laboratory. 1 J. A. Barker. In M. L. Klein and J. A. Venables, Eds., Rare Gas Solids 共Academic, New York, 1976兲, Chap. 4. 2 J. A. Barker, R. A. Fisher, and R. O. Watts, Mol. Phys. 21, 657 共1971兲. 3 M. Ross, H. K. Mao, P. M. Bell, and J. A. Xu, J. Chem. Phys. 85, 1028 共1986兲. 4 J. A. Barker, J. Chem. Phys. 86, 1509 共1987兲. 5 L. Jansen, Adv. Quantum Chem. 2, 119 共1965兲. 6 M. Bulski, Mol. Phys. 29, 1171 共1975兲. 7 B. H. Wells and S. Wilson, Mol. Phys. 55, 199 共1985兲. 8 G. Chalasiński, M. M. Szalewicz, and S. M. Cybulski, J. Chem. Phys. 92, 2481 共1990兲. 9 V. Mohan and J. B. Anderson, J. Chem. Phys. 92, 6971 共1990兲. 10 V. F. Lotrich and K. Szalewicz, Phys. Rev. Lett. 79, 1301 共1997兲. 11 E. M. Mas, V. F. Lotrich, and K. Szalewicz, J. Chem. Phys. 110, 6694 共1999兲. 12 M. J. Elrod and R. J. Saykally, Chem. Rev. 94, 1975 共1997兲. 13 P. R. Herman, P. E. LaRosque, and B. P. Stoicheff, J. Chem. Phys. 89, 4535 共1988兲. 14 A. Ernesti and J. M. Hutson, J. Chem. Phys. 106, 6288 共1997兲. 15 Y. Xu, W. Jaeger, and M. C. L. Gerry, J. Chem. Phys. 100, 4171 共1994兲. 16 Y. Xu and W. Jaeger, J. Chem. Phys. 107, 4788 共1997兲. 17 R. Moszynski, P. E. S. Wormer, T. G. A. Heijmen, and A. van der Avoird, J. Chem. Phys. 108, 579 共1998兲. 18 S. Goyal, D. L. Schutt, and G. Scoles, Phys. Rev. Lett. 69, 443 共1992兲. 19 M. Hartmann, R. E. Miller, J. P. Toennies, and A. Vilesov, Phys. Rev. Lett. 76, 4560 共1996兲. 20 J. Higgins et al., J. Phys. Chem. 102, 4952 共1998兲. 21 J. Higgins et al., Phys. Rev. Lett. 77, 4532 共1996兲. 22 F. Stienkemeier, J. Higgins, W. E. Ernst, and G. Scoles, Phys. Rev. Lett. 74, 3592 共1995兲. J. Chem. Phys., Vol. 112, No. 13, 1 April 2000 23 J. Reho, J. Higgins, M. Radcliff, K. K. Lehmann, and G. Scoles, to be published. 24 T.-S. Ho and H. Rabitz, J. Chem. Phys. 104, 2584 共1996兲. 25 T.-S. Ho, T. Hollebeek, H. Rabitz, L. B. Harding, and G. C. Schatz, J. Chem. Phys. 105, 10472 共1996兲. 26 T. Hollebeek, T.-S. Ho, and H. Rabitz, J. Chem. Phys. 106, 7223 共1997兲. 27 G. C. Schatz, A. Papaioannou, L. A. Pederson, L. R. Harding, T. Hollebeek, T.-S. Ho, and H. Rabitz, J. Chem. Phys. 107, 2340 共1997兲. 28 T. Hollebeek, T.-S. Ho, and H. Rabitz, Annu. Rev. Phys. Chem. 50, 537 共1999兲. 29 R. J. Bartlett and J. F. Stanton, in Reviews in Computational Chemistry, Vol. V, edited by K. B. Lipkowitz and D. B. Boyd 共VCH Publishing, 1994兲. 30 M. Gutowski, J. Chem. Phys. 110, 4695 共1999兲. 31 P. R. Taylor, in Lecture Notes in Quantum Chemistry II, edited by B. O. Roos 共Springer Verlag, New York, 1994兲. 32 H. B. Jansen and P. Ros, Chem. Phys. Lett. 3, 140 共1969兲. 33 S. F. Boys and F. Bernardi, Mol. Phys. 19, 553 共1970兲. 34 M. Gutowski, J. van Duijneveldt-van de Rijdt, J. H. van Lenthe, and F. B. van Duijneveldt, J. Chem. Phys. 98, 4728 共1993兲. 35 M. Gutowski and G. Chałasiński, J. Chem. Phys. 98, 5540 共1993兲. 36 M. Gutowski, M. M. Szczȩśniak, and G. Chałasiñski, Chem. Phys. Lett. 241, 140 共1995兲. 37 F. B. van Duijneveldt, J. G. C. M. van Duijneveldt-van de Rijdt, and J. H. van Lenthe, Chem. Rev. 94, 1873 共1994兲. 38 F. B. van Duijneveldt, in Molecular Interactions: From van der Waals to Exchange effects in three-body interactions 5761 Strongly-Bound Complexes, edited by S. Scheiner 共Wiley, New York, 1997兲. 39 A. J. Sadlej and M. Urban, J. Mol. Struct.: THEOCHEM 234, 147 共1991兲. 40 L. Li, S. F. Rice, and R. W. Field, J. Chem. Phys. 82, 1178 共1985兲. 41 B. M. Axilrod and E. Teller, J. Chem. Phys. 11, 299 共1943兲. 42 Y. Muto, Proc. Phys.-Math. Soc. Jpn. 17, 692 共1943兲. 43 A copy of Fortran program for generating the RKHS potential energy surface of the lowest quartet state 1 4 A ⬘2 of the Na3 trimer can be obtained by forwarding a request to: taksan@chemvax.princeton.edu. 44 C. Douketis, G. Scoles, S. Marchetti, M. Zen, and A. J. Thakkar, J. Chem. Phys. 76, 3057 共1982兲. 45 M. Marinescu, H. R. Sadeghpour, and A. Dalgarno, Phys. Rev. A 49, 982 共1994兲. 46 GAUSSIAN 92, Revision A, for sample input profiles. M. J. Frisch, G. W. Trucks, M. Head-Gordon, P. M. W. Gill, M. W. Wong, J. B. Foresman, B. G. Johnson, H. B. Schlegel, M. A. Robb, E. S. Replogle, R. Gomperts, J. L. Andres, K. Raghavachari, J. S. Binkley, C. Gonzalez, R. L. Martin, D. J. Fox, D. J. Defrees, J. Baker, J. J. P. Stewart, and J. A. Pople 共Gaussian, Inc., Pittsburgh, PA, 1992兲. 47 J. N. Murrell, S. Carter, S. C. Farantos, P. Huxley, and A. J. C. Varandas, Molecular Potential Energy Functions 共Wiley, New York, 1984兲. 48 T. P. Lybrand and P. A. Kollman, J. Chem. Phys. 83, 2923 共1985兲. 49 J. Tennyson, S. Miller, and C. R. Le Sueur, Comput. Phys. Commun. 75, 339 共1993兲. 50 V. F. Lotrich and K. Szalewicz, J. Chem. Phys. 106, 9688 共1997兲.