Physics Letters A 304 (2002) 36–42

www.elsevier.com/locate/pla

Faddeev–Merkuriev equations for resonances in three-body

Coulombic systems
Z. Papp a,b,∗ , J. Darai c , A. Nishimura b , Z.T. Hlousek b , C.-Y. Hu b , S.L. Yakovlev d
a Institute of Nuclear Research of the Hungarian Academy of Sciences, PO Box 51, H-4001 Debrecen, Hungary
b Department of Physics, California State University, Long Beach, CA 90840, USA
c Department of Experimental Physics, University of Debrecen, Debrecen, Hungary
d Department of Mathematical and Computational Physics, St. Petersburg State University, St. Petersburg, Russia

Received 18 June 2002; received in revised form 24 September 2002; accepted 24 September 2002
Communicated by B. Fricke
This work is dedicated to the 70th birthday of Prof. Borbála Gyarmati

Abstract
We reconsider the homogeneous Faddeev–Merkuriev integral equations for three-body Coulombic systems with attractive
Coulomb interactions and point out that the resonant solutions are contaminated with spurious resonances. The spurious
solutions are related to the splitting of the attractive Coulomb potential into short- and long-range parts, which is inherent
in the approach, but arbitrary to some extent. By varying the parameters of the splitting the spurious solutions can easily be
ruled out. We solve the integral equations by using the Coulomb–Sturmian separable expansion approach. This solution method
provides an exact description of the threshold phenomena. We have found several new S-wave resonances in the e− e+ e−
system in the vicinity of thresholds.
 2002 Elsevier Science B.V. All rights reserved.

PACS: 03.65.Ge; 02.30.-f; 11.30.Qc

Keywords: Resonances; Three-body systems; Coulomb potential; Integral equations

1. Introduction Then, standard bound-state methods become applica-

Certainly, to calculate resonances in three body
atomic systems the methods based on complex rota-
tion [1] are the most popular. The complex rotation of
the coordinates turns the resonant behavior of the wave
function into a bound-state-like asymptotic behavior.
* Corresponding author.
E-mail address: zoltan.papp@moon.atomki.hu (Z. Papp).
0375-9601/02/$ – see front matter  2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 5 - 9 6 0 1 ( 0 2 ) 0 1 3 5 0 - 6
ble also for calculating resonances. The complex ro-
tation of the coordinates does not change the discrete
spectrum, the branch cut, corresponding to scattering
states, however, is rotated down onto the complex en-
ergy plane, and as a consequence, resonant states from
the unphysical sheet become accessible.
In practice bound state variational approaches are
used, that results in a discretization of the rotated con-
tinuum. By changing the rotation angle the points cor-
responding to the continuum move, while those cor-
 Z. Papp et al. / Physics Letters A 304 (2002) 36–42 37

responding to discrete states, like bound and resonant 2. Faddeev–Merkuriev integral equations
states, stay. This way one can determine resonance pa-
rameters. The Hamiltonian of a three-body Coulombic sys-
However, in practical calculations the points of the tem is given by
discretized continuum scatter around the rotated-down
H = H 0 + v1C + v2C + v3C , (1)
straight line. So, especially around thresholds it is not
easy to decide whether a point is a resonance point or where H0 is the three-body kinetic energy operator
it belongs to the rotated continuum. and vαC denotes the Coulomb potential in the subsys-
Recently, we have developed a method for calcu- tem α, with α = 1, 2, 3. We use throughout the usual
lating resonances in three-body Coulombic systems configuration-space Jacobi coordinates xα and yα .
by solving homogeneous Faddeev–Merkuriev integral Thus vαC , the potential between particles β and γ , de-
equations [2] using the Coulomb–Sturmian separable pends on xα .
expansion approach [3]. As a test case, we calculated The Hamiltonian (1) is defined in the three-body
the resonances of the negative positronium ion. This Hilbert space. The three-body kinetic energy, when the
system has been extensively studied in the past two center-of-mass motion is separated, is given by
decades (see, e.g., Ref. [4] and references therein) and
thus serves as test example for new methods. We found H 0 = h0xα + h0yα = h0xβ + h0yβ = h0xγ + h0yγ , (2)
all the 12 S-wave resonances presented in Ref. [5] and where h0 is the two-body kinetic energy. The two-
observed good agreements in all cases. body potential operators are formally embedded in the
With our method we succeeded in locating 10 more three-body Hilbert space v C = v C (x)1y , where 1y is a
resonances in the same energy region, all of them are unit operator in the two-body Hilbert space associated
very close to the thresholds. This is certainly due to the with the y coordinate.
fact that in Ref. [3] the threshold behaviors are exactly Merkuriev proposed [2] to split the Coulomb inter-
taken into account. Unexpectedly we also observed action in the three-body configuration space into short-
that in the case of attractive Coulomb interactions and long-range terms
the Faddeev–Merkuriev integral equations produce
numerous resonant solutions of dubious origin. We vαC = vα(s) + vα(l) , (3)
tend to regard them as spurious resonances which where the short- and long-range parts are defined via a
come about from the somewhat arbitrary splitting of splitting function:
the potential in the three-body configuration space into
short-range and long-range terms, that is an inherent vα(s) = vαC ζ(xα , yα ), (4)
attribute of the theory.

vα(l) = vαC 1 − ζ(xα , yα ) . (5)
Since the possible appearance of spurious states
in the formalism is new and surprising, for the sake The splitting function ζ is defined in such a way that
of the better understanding of its mechanism, in Sec- lim ζ (x, y)
tion 2 we outline briefly the Faddeev–Merkuriev in- x,y→∞
  
tegral equation formalism. In our particular case of 0 1/ν ,
= 1, if |x| < x 1 + |y|/y
0
the e− e+ e− system two particles are identical and (6)
0, otherwise,
the set of Faddeev–Merkuriev integral equations can
be reduced to an one-component equation. In Sec- where x 0 , y 0 > 0 and ν > 2. Usually the functional
tion 3 we discuss the spurious solutions. In Section 4 form
the Coulomb–Sturmian separable expansion method 2
ζ (x, y) =  , (7)
is applied to the one-component integral equation. In (x/x 0 )ν
1 + exp 1+y/y 0
Section 5 we present our new calculations for the
L = 0 resonances of the e− e+ e− system and compare is used. So, the separation into short- and long-range
them with the results of the complex scaling calcula- parts is made along a parabola-like curve over the
tions of Ref. [5] and with the more recent results of {x, y} plane. Typical shapes for v (s) and v (l) are shown
Ref. [4]. in Figs. 1 and 2, respectively.
38 Z. Papp et al. / Physics Letters A 304 (2002) 36–42

function into two components

|Ψ = |ψ1 + |ψ2 , (10)

with components defined by

|ψα = G(l)(z)vα(s) |Ψ , (11)

where α = 1, 2 and = (z −
G(l) (z) H (l))−1 ,
z is a
complex number.
In the case of bound and resonant states the wave-
function components satisfy the homogeneous two-
component Faddeev–Merkuriev integral equations
(l) (s)
|ψ1 = G1 (z)v1 |ψ2 , (12)
Fig. 1. Typical shape of v (s) for attractive Coulomb potentials.
(l) (s)
|ψ2 = G2 (z)v2 |ψ1 , (13)

at real and complex energies, respectively. Here G(l)

α is
the resolvent of the channel long-ranged Hamiltonian
Hα = H (l) + vα , Gα (z) = (z − Hα )−1 . Merkuriev
(l) (s) (l) (l)

has proved that Eqs. (12) and (13) possess compact

Fig. 2. Typical shape of v (l) for attractive Coulomb potentials.
In atomic three-particle systems the sign of the
charge of two particles are always identical. Let
us denote in our e− e+ e− system the two electrons
by 1 and 2, and the positron by 3. In this case
v3C , the interaction between the two electrons, is a
repulsive Coulomb potential which does not support
two-body bound states. Therefore, the entire v3C can be
considered as long-range potential. With splitting (3)
the Hamiltonian can formally be written in a form
which looks like an usual three-body Hamiltonian with
two short-range potentials
H = H (l) + v1(s) + v2(s) ,
where the long-range Hamiltonian is defined as
(l)
H (l) = H 0 + v1 + v2 + v3C .
(l)
kernels, and this property remains valid also for
complex energies z = E − iΓ /2, Γ > 0.
Further simplification can be achieved if we take
into account that the two electrons, particles 1 and 2,
are identical and indistinguishable. Then, the Faddeev
components |ψ1 and |ψ2 , in their own natural Jacobi
coordinates, have the same functional forms
x1 y1 |ψ1 = x2 y2 |ψ2 = xy|ψ . (14)
On the other hand
|ψ2 = pP|ψ1 , (15)
where P is the operator for the permutation of in-
dexes 1 and 2 and p = ±1 are the eigenvalues of P.
Therefore, we can determine |ψ from the first equa-
tion only
(l) (s)
|ψ = G1 v1 pP|ψ . (16)
It should be emphasized, that so far we did not make
any approximation, and although this integral equation
has only one component, yet it gives full account both
(8) of asymptotic and symmetry properties of the system.
(9) 3. Spurious resonance solutions

Then, the Faddeev method is applicable and, in this Let us suppose for a moment that the y 0 parameter
particular case, results in a splitting of the wave in ζ(x1 , y1 ) is infinite. Then ζ would not depend on y1 ,
 Z. Papp et al. / Physics Letters A 304 (2002) 36–42 39

the separation of the potential into short and long range way we can easily distinguish between physical and
(l) spurious resonance solutions.
parts would go along a straight line, and v1 (x1 , y1 )
would be like a valley in the y1 direction. The potential
(l)
v1 (x1 , y1 ), which is Coulomb-like along x1 , would
support infinitely many bound states and, as free 4. Coulomb–Sturmian potential separable
motion along the coordinate y1 would be possible, expansion approach
the Hamiltonian H 0 + v1(l) would have infinitely many
two-body channels. We solve Eq. (16) by using the Coulomb–Sturmian
If, however, y 0 is finite, the straight line along separable expansion approach [6]. The Coulomb–
the y1 direction becomes a parabola-like curve and the Sturmian (CS) functions are defined by
valley, as y1 goes to infinity, gets broader and broader,  1/2
n!
and shallower and shallower, and finally disappears r|nl =
(n + 2l + 1)!
(see Fig. 2). As the valley gets closed the continuum of
× (2br)l+1 exp(−br)L2l+1 (2br), (18)
H 0 + v1(l) associated with the y1 coordinate becomes n
(l)
discretized. So, if y 0 is finite, H 0 + v1 have infinitely with n and l being the radial and orbital angular
many bound states. Similar analysis is valid also for momentum quantum numbers, respectively, and b is
(l)
H 0 + v2 , and consequently, also H (l) has infinitely the size parameter of the basis. The CS functions {|nl }
many bound states. form a biorthonormal discrete basis in the radial two-
This, however, due to (11), can lead to spurious body Hilbert space; the biorthogonal partner is defined
solutions. If, at some energy, H (l) has bound state by r|nl = r|nl /r. Since the three-body Hilbert
G(l) has pole. Then, |ψα in Eq. (11), irrespective space is a direct product of two-body Hilbert spaces
of vα(s) , can be finite even if |Ψ is infinitesimal. These an appropriate basis can be defined as the angular
solutions are called spurious solutions: although the momentum coupled direct product of the two-body
bases
Faddeev components are not identically zero, but their
sum vanishes. |nνlλ 1 = |nl 1 ⊗ |νλ 1 (n, ν = 0, 1, 2, . . .), (19)
Let us examine now the spectral properties of the
Hamiltonian where |nl 1 and |νλ 1 are associated with the coordi-
nates x1 and y1 , respectively. With this basis the com-
(l) (s) (l)
pleteness relation takes the form (with angular mo-
H1 = H (l) + v1 = H 0 + v1C + v2 + v3C . (17) mentum summation implicitly included)
N
The three-body potential v2(l) is attractive and con-
(l) 1 = lim |nνlλ 11 nνlλ| = lim 1N
1 . (20)
structed so that v2 (x2 , y2 ) vanishes as y2 tends to in- N→∞ N→∞
n,ν=0
finity. Therefore, there are no two-body channels as-
sociated with fragmentation 2, and of course neither Note that similar bases can be constructed for frag-
mentations 2 and 3 as well.
with fragmentation 3, the Hamiltonian H1(l) has only
We make the following approximation on Eq. (16)
1-type two-body asymptotic channels.
What happens to the bound states associated with |ψ = G(l) N (s)
1 11 v1 pP11 |ψ ,
N
(21)
(l)
v2 ? They are embedded in the continuum of H 0 + (s)
i.e., the operator v1 pP in the three-body Hilbert
v1C , and become resonant states. So, by solving the
space is approximated by a separable form, viz.
Faddeev–Merkuriev integral equations, at some com-
plex energies, we can encounter spurious solutions. v1(s) pP = lim 1N (s) N
1 v1 pP11
These spurious solutions are not related to the Hamil- N→∞
(s)
tonian H , but rather only to some auxiliary potentials ≈ 1N N
1 v1 pP11
coming from the splitting procedure. Consequently, N
the spurious resonances should be sensitive to the pa- ≈ |nνlλ 1 v (s)

   
1 1 n ν l λ |, (22)
rameters of ζ , while the physical resonances not. This n,ν,n ,ν  =0
40 Z. Papp et al. / Physics Letters A 304 (2002) 36–42

where v (s) (s)     of G1 (z) = (z − H1 )−1 via solution of a Lippmann-

1 = 1 nνlλ|v1 pP|n ν l λ 1 . The compact-
ness of the equation and the completeness of the ba- Schwinger equation,
sis guarantee the convergence of the method. Utiliz-  (l) −1  −1
ing the properties of the exchange operator P these G1 = G1 − U 1, (26)
(s)
matrix elements can be written in the form v 1 = where

p × (−)l 1 nνlλ|v1(s) |n ν  l  λ 2 .
With this approximation, the solution of Eq. (16) G1 = 1 nνlλ|G1 |n
 ν  l  λ ,
1
 (l)    
turns into solution of matrix equations for the compo- U1 = 1 nνlλ| v2 + v3 |n ν l λ 1 .
C
nent vector ψ = 1 nνlλ|ψ
The most crucial point in this procedure is the cal-

(l) −1 (s)  culation of the matrix elements G1 , since the poten-
G1 (z) − v1 ψ = 0, (23)
(s)
tial matrix elements v 1 and U 1 can always be evalu-
where G1 = 1 nνlλ|G1 |n

(l) (l)
 ν  l  λ . A unique solu-
1 ated numerically by making use of the transformation
tion exists if and only if of Jacobi coordinates [8]. The Green’s operator G1

−1  is a resolvent of the sum of two commuting Hamil-
D(z) ≡ det G(l) 1 (z) − v (s)
1 = 0. (24) tonians, H1 = hx1 + hy1 , where hx1 = h0x1 + v1C (x1 )
(l)
The Green’s operator G1 is related to the Hamil- and hy1 = h0y1 , which act in different two-body Hilbert
tonian H1(l) , which is still a three-body Coulomb spaces. Thus, G1 can be given by a convolution inte-
Hamiltonian, seems to be as complicated as H itself. gral of two-body Green’s matrices, i.e.,

However, H1(l) has only 1-type asymptotic channels, 1
G1 (z) = dz g x1 (z − z )g y1 (z ), (27)
with asymptotic Hamiltonian 2πi
C
H1 = H 0 + v1C . (25)
where gx1 (z) = (z − hx1 )−1 and gy1 (z) = (z − hy1 )−1 .
Therefore, in the spirit of the three-potential formal- The contour C should be taken counterclockwise
ism [7], G(l)
1 can be linked to the matrix elements around the continuous spectrum of hy1 such a way

Table 1
e
S-wave 1S resonances of Ps− calculated by three different methods. In Ref. [5] the complex scaling method were used with Hylleraas-type
trial functions. In Ref. [4] the complex scaling method were combined with the stochastic variational method and correlated Gaussian were
used as trial functions. The energies and widths are expressed in rydbergs
Thresholds Ref. [5] Ref. [4] This work
−Er Γ −Er Γ −Er Γ
0.1520608 0.000086 0.1520594 0.00008472 0.1519 0.000085
0.12730 0.00002 0.127334 0.000180 0.1273 0.000017
0.1251 0.000002
N =2
0.070683 0.00015 0.0706658 0.000154 0.0707 0.00015
0.05969 0.00011 0.059690 0.000110 0.05968 0.00011
0.0564 0.00003
N =3
0.04045 0.00024 0.040428 0.00026
0.0350 0.0003 0.03502 0.00026
0.03463 0.00034 0.03462 0.00032
0.03263 0.0001
0.03158 0.00007
N =4
0.0258 0.00045 0.02606 0.00021
0.02343 0.00014 0.0234 0.0001
N =5
 Z. Papp et al. / Physics Letters A 304 (2002) 36–42 41

that gx1 is analytic on the domain encircled by C.
A contour, which appropriate also for resonances,
is given in Ref. [3]. The corresponding CS matrix
elements of the two-body Green’s operators in the
integrand are known analytically for all complex
energies (see [7] and references therein). It is also
evident that all the thresholds, corresponding to the
poles of g x1 , are at the right location and therefore
this method is especially suited to study near-threshold
resonances.
5. S-wave resonances in positronium ion
To calculate resonances we have to find the com-
plex zeros of the Fredholm determinant (24). To lo-
cate them we calculate D(z) along the real energy axis
with small step size. In the vicinity of zeros the D(z)
exhibits violent changes. Then we have a good starting
point for the zero search. This way we can easily find
resonances, at least the narrow ones. To find broad res-
onances one should proceed by calculating D(E − iε)
with finite ε > 0.
For the parameters of the splitting function we take
y 0 = 50a0 and ν = 2.1, a0 is the Bohr radius. To select
out the physical solutions we vary the short- and long-
range potentials by taking x 0 = 18a0, 25a0 and 30a0 ,
respectively. We can also vary the potentials by adding
and subtracting a short-range term
(s)
v1 = v1C ζ (x1 , y1 ) + v0 ζ (x1 , y1 ),


v1(l) = v1C 1 − ζ(x1 , y1 ) − v0 ζ(x1 , y1 ), (29)
respectively, while keeping a fixed x 0 = 18a0 value.
This new kind of splitting goes beyond the Merkuriev’s
original suggestion, but since the character of v (s) and
v (l) remains the same all the nice properties of the
original Faddeev–Merkuriev equations are retained. In
these calculations we used v0 = 0, 0.01, 0.05, 0.1 and
0.5 values.
We found that some solutions, especially their
widths, are very sensitive to the change of either x 0 or
v0 parameters. We regard them spurious solutions. The
solutions given in Tables 1 and 2 were stable against
all changes of parameters and we can consider them as
physical resonances. We recovered all the resonances
presented in Ref. [5] with very good agreements,
but besides that we found 10 more resonances, all
of them are in the vicinity of some thresholds. In
Tables 1 and 2 we also show the results of a recent
calculation using complex scaling with stochastic
variational method [4]. In Ref. [5] Hylleraas-type
trial functions were used while in Ref. [4] correlated
Gaussian functions were adopted. We can see that
these approaches, especially the method of Ref. [4],
have serious deficiencies in locating resonances at
higher energies or in the vicinity of thresholds.
6. Conclusions
In this article we pointed out that in the Faddeev–
(28) Merkuriev integral equation approach of the three-

Table 2
e
The same as Table 1 but for 3S states
Thresholds Ref. [5] Ref. [4] This work
−Er Γ −Er Γ −Er Γ
0.12706 0.00001 0.127074713 0.000000016 0.127 0.000000006
0.12501 0.00015 0.1251 0.000000001
N =2
0.05873 0.00002 0.05870352 0.00004 0.05874 0.00000037
0.0561 0.0000002
N =3
0.03415 0.00002 0.0553 0.0001
0.03420 0.000001
0.03237 0.0000011
0.03172 0.0000002
N =4
0.031035 0.0000938
N =5
42 Z. Papp et al. / Physics Letters A 304 (2002) 36–42
body Coulomb problem the complex-energy spectrum
is contaminated with spurious solutions. The spurious
solutions, however, are sensitive to the splitting of the
potential into short- and long-range terms. This offers
an easy way to select the physical solutions.
We solved the integral equations by using the
Coulomb–Sturmian separable expansion approach.
This method gives an exact description of the thresh-
old phenomena, thus the method is applicable also for
close-to-threshold and high-lying resonances. In the
e− e+ e− system we located several new resonances,
all of them are in the close vicinity of some threshold.
Acknowledgements
This work has been supported by the NSF Grant
No. Phy-0088936 and OTKA Grants under Contracts
No. T026233 and No. T029003.
References
[1] J. Aguilar, J.M. Combes, Commun. Math. Phys. 22 (1971) 269;
E. Balslev, J.M. Combes, Commun. Math. Phys. 22 (1971) 280;
B. Simon, Commun. Math. Phys. 27 (1972) 1.
[2] S.P. Merkuriev, Ann. Phys. (N.Y.) 130 (1980) 395;
L.D. Faddeev, S.P. Merkuriev, Quantum Scattering Theory for
Several Particle Systems, Kluwer, Dordrech, 1993.
[3] Z. Papp, J. Darai, C.-Y. Hu, Z.T. Hlousek, B. Kónya,
S.L. Yakovlev, Phys. Rev. A 65 (2002) 032725.
[4] J. Usukura, Y. Suzuki, Phys. Rev. A 66 (2002) 010502(R).
[5] Y.K. Ho, Phys. Lett. A 102 (1984) 348.
[6] Z. Papp, W. Plessas, Phys. Rev. C 54 (1996) 50.
[7] Z. Papp, C.-Y. Hu, Z.T. Hlousek, B. Kónya, S.L. Yakovlev, Phys.
Rev. A 63 (2001) 062721.
[8] R. Balian, E. Brézin, Nuovo Cimento B 2 (1969) 403.