You are on page 1of 10

Mapped Fourier grid methods for ultracold molecules

Viatcheslav Kokoouline,

Department of Physics and JILA, University of Colorado, Boulder, Colorado


80309-0440, USA

Kai Willner, Olivier Dulieu, and Francoise Masnou-Seeuws

Laboratoire Aime Cotton, B^at. 505 Campus d'Orsay, 91405 Orsay Cedex, FRANCE.

I. INTRODUCTION
Although direct laser cooling of molecules is not an ecient process, the formation of ultracold molecules (T  100 K) in an ensemble of laser-cooled alkali
atoms has been observed by various groups [1{4]. Two steps are necessary, photoassociation and radiative stabilization. In the photoassociation reaction [5]
a pair of colliding cold atoms is absorbing a photon red-detuned from the resonance line, creating an excited molecule in a loosely-bound vibrational level.
This short-lived molecule usually decays back into a pair of free atoms, but stabilization schemes, either by spontaneous or by induced emission, into bound levels of the ground molecular state, have been found by the experimental groups.
Many other schemes should exist, and their investigation relies upon accurate
theoretical calculations of the various transition probabilities. The e orts must
be directed into three main directions:
 The precision required for the electronic potential curves is well beyond the
present achievements of ab initio calculations. Methods to t potentials
to accurate experimental data [6] have to be revisited [7].
 The dynamics of ultracold molecules is characterized by the large extension of the vibrational motion, up to distances of hundreds of atomic
units. Numerical methods should describe both the short-range region,
where potential curves can have depths of a few eV, and the long-range
region, where the potential is weaker than 1 cm 1. A time-independent
description, adapted to a perturbative treatment of radiative processes for
continuous wave (cw) lasers and weak intensities, is the subject of this article. Matrix elements of the dipole transition moment must be computed,
requiring an accurate knowledge of continuum and bound wavefunctions.

 email: francoise.masnou@lac.u-psud.fr

 For strong laser intensities, or when pulsed lasers are used, time-dependent
methods are also being developed [8{10].

II. THE MAPPED FOURIER GRID METHOD


A. One channel calculations
Let us consider the vibrational wavefunction for a bound level close to the
dissociation limit of a photoassociated molecule, such as Cs2 1g (6s + 6p 2 P3=2).
The potential has an asymptotic R 3 dipole-dipole behaviour. As an example,
for the v = 332 excited wavefunction in the 1g potential, the local de Broglie
wavelength (R), in atomic units with h = 1, at energy Ev ,
(R) = p 2
;
(1)
2[Ev V (R)]
is varying from a few tens of a0 in the outer region to a value less than 1a0 in
the inner region (see Fig. 1). The standard methods then have to be adapted.
(R)

0.1
0.0
0.1
0

100

200

300

400

R(a0)

FIG. 1. The wavefunction v = 332 in the potential 1 (6s + 6p 2 P3 2 of Cs2 ). The


local wavelength becomes very large in the asymptotic region.
g

The present paper deals with a mapping procedure developed to bring the
eciency of Fourier grid methods into the domain of cold molecules.
Grid numerical methods using fast Fourier transforms (FFTs) have proved to
be very ecient for quantum molecular dynamics [11,12]. Wavefunctions and
operators are represented on grids in both the momentumand coordinate spaces,
using a constant grid step s, which is linked to the minimum value of the local
de Broglie wavelength. For calculations of the vibrational levels of a diatomic
molecule, a Fourier grid method, originally proposed by Marston and BalintKurti [13] and by Colbert and Miller [14], was further developed by Monnerville
and Robbe [15] and Dulieu et al [16,17], and proved to be very successful. The
2

wavefunction may be represented by its values at a set of N equally spaced


radial distances provided we may de ne a set of approximate delta functions
at the grid points. This can be achieved by a discrete Fourier expansion on a
set of N ingoing and outgoing plane waves, with wavelength varying from 2s
to 2L, L being the length of the grid. It is then straightforward to represent
^ by N  N matrices either in
all operators, and therefore the Hamiltonian H,
coordinate or in momentum representation. The eigenvalues are obtained by
diagonalization, and the computation e ort involved scales as N 3 . The choice
of the constant step, and therefore of N, is linked to the minimum value of the
local de Broglie wavelength (with respect to a reference energy Eref , usually
chosen as the asymptotic value of the potential),

s  (R2 e) = p
;
(2)
2[Eref V (Re)]
obtained at the equilibrium distance Re where the potential is minimum. For
the present problem, this can lead to a very large number of grid points. In the
example of the v = 332 wavefunction in the Cs2 1g potential presented above,
the calculations would require at least diagonalization of a  8200  8200 matrix.
However, the grid step is unnecessarily small in the long-range region. Fattal

et al [18] have proposed a mapping procedure in order to optimize the use of

phase space in grid methods, with application to Coulomb potentials. This


mapping procedure has been generalized to the present problem by Kokoouline
et al [19], who have introduced an adaptative coordinate x scaled on the local
de Broglie wavelength, such that a variable stepsize s(R) can be used:

p
s(R)  (R)
=
:
(3)
2
2[Eref V (R)]
The coordinate transformation involves a Jacobian J(x), which with our choice
for scaling reads
J(x) = dR
(4)
dx = s(R):
In the new coordinate x, the wavefunction displays regular oscillations (see Fig.
2).

(x)

0.1
0.0
0.1
450

500

550
x

FIG. 2. The wavefunction of Fig. 1 is shown here as a function of the coordinate x


near the outer turning point. The local wavelength in the classically allowed region is
nearly constant in x.

Therefore working grids with a constant grid step in x are ecient, corresponding to a variable step s(R) in R. The number of grid points is then
markedly reduced, and in the chosen example the same accuracy can be obtained
by diagonalization of a 564  564 matrix. The radial Schrodinger equation may
be rewritten in a symmetrical form by considering a new wavefunction (x),
(5)
(x) = J(x) 12 (x);
so that the Jacobian is eliminated from the scalar product. The Schrodinger
equation in the adaptative coordinate x then reads
h
i
T^ + V^ '(x) = E '(x):
(6)
The kinetic and potential energy operators T^ and V^ may be written as
1 J 21 (x) d J 1(x) d J 12 (x);
T^ = 2
(7)
dx
dx
V^ = V (x):
(8)
Both T^ and V^ are hermitian with respect to the scalar product. As discussed
in Ref. [19], T^ can be rewritten as


2
2
7 J 0 (x)2 + J 00(x) :
^T = 1 21 d 2 + d 2 21
(9)
4 J (x) dx dx J (x) 2 J(x)4 J(x)3
Expression (9) allows for a straightforward determination of the Hamiltonian
matrix, with no need for explicit summations. On the other hand, it requires
knowledge of the derivatives of the Jacobian J(x). If the mapping function
R = f(x) is de ned numerically, the determination of J 0 (x) and J 00(x) may
4

cause numerical problems. One may then resort to expression (7), which requires
knowledge of the Jacobian itself but not of its derivatives.
The grid point functions, obtained by discrete Fourier transform from the
plane waves,
 

(x
x
)
sin
N
1
n
L
 ;

un(x) = N
(10)
sin L (x xn )
are such that
un(xm ) = nm :
(11)
The matrix V^ of the potential energy is diagonal in this basis
hunjV^ jum i = V (xn ) nm ;
(12)
while the matrix of the kinetic energy may be derived from expression (9)


1
d2 ju i
^
hun jT jum i = 4 J(xn ) 2 + J(xm ) 2 hun j dx
(13)
m
2

7 J 0 (x )2J(x ) 4 + J 00(x )J(x ) 3  :
(14)
n
nm
n
n
nm
2 n
The matrix elements of the second derivative have been computed by Meyer [20]
8

2
< 1  2 (N 2 1)
2
n=m
d
3 L
(15)
hunj dx2 jum i = : 22 ( 1)n m cos( n Nm )
n 6= m :
L2
sin 2 ( n Nm )
Alternatively, the matrix T^ can be derived directly from expression (7):

hun jT^jum i =
where

N
X
k=1

d ju iJ(x ) 1 hu j d ju iJ(x )
J(xn) 21 hun j dx
k
k
k dx m
m

(
0
d
hun j dx jum i =  ( 1)nn mm
L sin( N )

n=m
n 6= m:

1
2

(16)

(17)

Eq. (16) does not require calculation of J 0 (x) and J 00(x). The drawback is
the computational e ort involved with the sum in (16): it can be overcome by
the numerical FFT algorithm [21]. Diagonalization of the Hamiltonian matrix
hun jH^ jum i yields very accurate eigenvalues and wavefunctions. In the working
5

grid the wavefunctions display regular oscillations. They have to be transformed


making use of Eqs. (4) and (5) to obtain results in the physical grid R. Extrapolation of the wavefunction between the grid points may be achieved using the
basis functions de ned in (10) according to
N
X

N
X

sin N L (x xn)
(x) = (xn ) un(x) = (xn) N1
;
sin L (x xn)
n=1
n=1
which is very convenient for quadrature procedures.

(18)

More generally, an adaptative coordinate may be de ned by replacing the


real potential V (R) in Eq. (3) by an enveloping potential Venv (R) located below
the physical potential for all internuclear distances, for instance the analytical
asymptotic potential C3=R3 or C6=R6 . Choosing Venv (R) = Cn=Rn with
n 6= 2, the change of variable in Eq. (4) becomes analytical, and for a grid
starting at distance Rin we may de ne an adaptative coordinate through formula

1
x(R) = 2(n2C2)n n1 2
(19)
n 2 :
2
R 2
Rin
The physical interpretation of this change of variable can be related to the
asymptotic calculations of Le Roy and Bernstein [6] where the Schrodinger equation is modi ed into a Bessel equation with analytical solutions.
It is often convenient to use a common potential Venv (R) for the de nition of
the same grid in the initial and nal states, in order to simplify the quadrature
involved in calculations of the Franck-Condon factors: this seems one of the
main advantages of a grid method.
The choice of the number of grid points is presently determined by optimization of the use of the classical phase space (Fattal et al, 1996, Kokoouline et al,
1999). As discussed by Kokoouline et al (1999), a better optimization is obtained
when the real potential is used in the de nition of the adaptative coordinate,
the analytical procedure being less ecient. Choosing Eq. (19), however, leads
to calculations with analytic formulas, which may save computing time. Fig. 3
shows the occupation of the phase space in calculations for a vibrational level
of a photoassociated molecule.

Analytic Mapping

40

4
0 Px

20
p (a.u.)

4
100 200 300 400

Venv(R)=C3/R

Numerical Mapping
4

20

0 Px
0

No mapping

40
0

100

200
R(a0)

300

100

200

4
300

Venv(R)=V(R)

FIG. 3. Solid lines: the classical energy shell in phase space with and without the coordinate transformation, for the vibrational level with binding energy
E = 0:295 cm 1 in the potential Cs2 1 (6s + 6p 2 P3 2 ). The area S = 1034 a.u.
de ned by contour is the same for all three cases, and is connected with the number
of phase cells N through S = 2  N . Broken lines: the rectangles de ning the phase
space used in the calculations. One can see that without mapping a much larger phase
volume is used, that requires proportionally more grid points, at least N =4573, to be
compared to N =172 in case of numerical mapping and N =272 in case of analytical
mapping.
u

B. Generalization
The method can be easily generalized to double-well potentials, where it is
more convenient than Numerov techniques and yields an accurate treatment of
tunneling e ects [22]. It has been also generalized to calculations with several
channels, including predissociation problems where coupling between bound and
continuum levels has to be considered [23{25]: the method is then well adapted
to the calculation involving optical potentials. Very accurate results have been
obtained: typically the method is now capable of computing the very last lev7

els in a long range R 3 potentials, when the binding energy is only  10 10


cm 1, the motion extending well beyond 2000 a0 , that is a few tens of m. In
two-channel calculations, we have been able to extract the short-range reaction
matrix Y from generalized Lu-Fano plots [26,27] of the computed vibrational energies, and found an excellent agreement, over an energy domain of 4000 cm 1,
with independent MQDT calculations using the method developed by Mies and
Raoult [28].
One advantage of the present method is its non-local character. In Ref. [24],
a new global deperturbation approach, involving the Fourier Grid Hamiltonian
method for energy level calculation, was implemented for determination of the
potential curves and interactions of two coupled electronic states of the potassium dimer by tting to experimental data. A standard deviation of 1.2 is
obtained, corresponding to a variance of 7:5  10 3 cm 1, representing a signi cant improvement compared to the standard deviation of 4 yielded by the
traditional local deperturbation approach.
However one remaining problem is the occurrence of some spurious unphysical
levels. Those ghost levels are easily identi ed by irregularities in the calculated
spectrum of H^ and in the rotational constants. Their wavefunctions are characterized by their non WKB-like behaviour (rapid oscillations and non vanishing
amplitude in the classically forbidden region). We have not found a mathematical explanation but we believe that the problem is linked to the periodic
boundary conditions. It can be corrected by use of a periodic condition in the
de nition of the adaptative coordinate, or more elegantly by choosing alternative
basis sets; recent work has shown the sine basis set is very promising [29].
Future work shoud generalize the mapping procedure to multidimension problems for calculations of loosely bound levels in triatomic molecules.
Stimulating collaboration with Ronnie Koslo is gratefully acknowledged.

[1] A. Fioretti, D. Comparat, A. Crubellier, O. Dulieu, F. Masnou-Seeuws, and P. Pillet, \Formation of cold Cs2 molecules through photoassociation", Phys. Rev. Lett.
80, 4402 (1998).
[2] T. Takekoshi, B. M. Patterson, and R. J. Knize, \Observation of optically trapped
cold cesium molecules", Phys. Rev. Lett. 81, 5105 (1999).
[3] N. Nikolov, E. E. Eyler, X. Wang, J. Li, H. Wang, W. C. Stwalley, and P. Gould,
\Observation of translationally ultracold ground state potassium molecules",
Phys. Rev. Lett. 82, 703 (1999).
[4] C. Gabbanini, A. Fioretti, A. Lucchesini, S. Gozzini, and M. Mazzoni, \Obser-

vation of translationally cold ground state rubidium molecules", Phys. Rev. Lett.
84, 2814 (2000).
[5] H. R. Thorsheim, J. Weiner, and P. S. Julienne, \Laser-induced photoassociation
of ultracold sodium atoms", Phys. Rev. Lett. 58, 2420 (1987).
[6] R. J. Le Roy and R. B. Bernstein, \Dissociation energy and long-range potential
of diatomic molecules from vibrational spacings of higher levels", J. Chem. Phys.
52, 3869 (1970).
[7] C. Amiot, O. Dulieu, R. Gutteres, and F. Masnou-Seeuws, \Determination of
the Cs2 0 (p3 2 ) state and of the Cs 6P1 2 3 2 atomic radiative lifetimes from
photoassociation spectroscopy", Phys. Rev. A 66, in press (2002).
[8] J. Vala, O. Dulieu, F. Masnou-Seeuws, P. Pillet, and R. Koslo , \Coherent control
of cold molecule formation through photoassociation using chirped pulsed laser
eld", Phys. Rev. A 63, 013412 (2001).
[9] M. Vatasescu, O. Dulieu, R. Koslo , and F. Masnou-Seeuws, \Toward optimal
control of photoassociation of cold atoms and photodissociation of long range
molecules: characteristic times for wavepacket propagation", Phys. Rev. A 63
(033407 2001).
[10] M. Vatasescu and F. Masnou-Seeuws, \Time-dependent analysis of tunneling effect in the formation of ultracold molecules via photoassociation of laser-cooled
atoms", Phys. Rev. A 21, in press (2002).
[11] R. Koslo , \Time Dependent Methods in Molecular Dynamics", J. Phys. Chem.
92, 2087 (1988).
[12] R. Koslo , \Quantum molecular Dynamics on Grids", In R. H. Wyatt and J. Z.
H. Zhang, ed., Dynamics of Molecules and Chemical Reactions , (Marcel Dekker,
New York, 1996), p. 185.
[13] C. C. Marston and G. Balint-Kurti, \The Fourier Grid Hamiltonian method for
bound state eigenvalues and eigenfunctions", J. Chem. Phys. 91, 3571 (1989).
[14] D. T. Colbert and W. H. Miller, \A novel discrete variable representation for
quantum mechanical reactive scattering via the S -matrix Kohn method", J. Chem.
Phys. 96, 1982 (1992).
[15] M. Monnerville and J. M. Robbe, \Optical potential coupled to discrete variable representation for calculations of quasibound states: application to the
CO(B 1 + D0 1 + ) predissociating interaction", J. Chem. Phys. 101-112,
7580 (1994).
[16] O. Dulieu and P. S. Julienne, \Coupled channel bound states calculations for alkali
dimers using the Fourier grid method", J. Chem. Phys. 103, 60 (1995).
[17] O. Dulieu, R. Koslo , F. Masnou-Seeuws, and G. Pichler, \Intermediate long
range molecules: bound and quasibound states for alkali dimers", J. Chem. Phys.
107, 10633 (1997).
[18] E. Fattal, R. Baer, and R. Koslo , \Phase space approach for optimizing grid
representations: The mapped Fourier method", Phys. Rev. E 53, 1217 (1996).
[19] V. Kokoouline, O. Dulieu, R. Koslo , and F. Masnou-Seeuws, \Mapped Fourier
methods for long range molecules: Application to perturbations in the Rb2 (0+ )
spectrum", J. Chem. Phys. 110, 9865 (1999).
[20] R. Meyer, \Trigonometric interpolation method for one-dimensional quantum meg

= ; =

chanical problems", J. Chem. Phys. 52, 2053 (1970).


[21] A. G. Borisov, \Solution of the radial Schrodinger equation in cylindrical and
spherical coordinates by mapped Fourier transform algorithms", J. Chem. Phys.
114, 7770 (2001).
[22] M. Vatasescu, O. Dulieu, C. Amiot, D. Comparat, C. Drag, V. Kokoouline,
F. Masnou-Seeuws, and P. Pillet. \Multichannel tunneling in the Cs2 0 photoassociation spectrum", Phys. Rev. A 61, 044701 (2000).
[23] V. Kokoouline, O. Dulieu, R. Koslo , and F. Masnou-Seeuws, \Theoretical treatment of channel mixing in excited Rb2 and Cs2 ultra-cold molecules: determination of predissociation lifetimes with coordinate mapping", Phys. Rev. A 62,
032716 (2000).
[24] C. Lisdat, O. Dulieu, H.Knockel, and E. Tiemann, \Inversion analysis of K2 coupled electronic states with the Fourier grid method", Eur. Phys. J. D 17, 319
(2001).
[25] P. Pellegrini, O. Dulieu, and F. Masnou-Seeuws, \Formation of Cs2 molecules via
Feshbach resonances stabilized by spontaneous emission: theoretical treatment
with the Fourier grid method", Eur. Phys. J. D 20, 77 (2002).
[26] V. Kokoouline, O. Dulieu, and F. Masnou-Seeuws, \Theoretical treatment of channel mixing in excited Rb2 and Cs2 ultra-cold molecules. Perturbations in 0+ photoassociation and uorescence spectra", Phys. Rev. A 62, 022504 (2000).
[27] V. N. Ostrovsky, V. Kokoouline, E. Luc-Koenig, and F. Masnou-Seeuws, \Lu-Fano
plots for potentials with non-Coulomb tail: application to vibrational spectra of
long-range diatomic molecules", J. Phys. B 34, L27 (2001).
[28] F. H. Mies and M. Raoult, \Analysis of threshold e ects in ultracold atomic
collisions", Phys. Rev. A 62, 012708 (2000).
[29] K. Willner, O. Dulieu, and F. Masnou-Seeuws, \A mapped sine grid method for
long range molecules and cold collisions", J. Chem. Phys., submitted (2002).
g

10

You might also like