Physics Letters A 369 (2007) 128131

www.elsevier.com/locate/pla
Efcient numerical method for calculating exciton states
in quantum boxes
Yoji Kubota, Katsuyuki Nobusada

Department of Theoretical Studies, Institute for Molecular Science, Myodaiji, Okazaki 444-8585, Japan
Received 20 November 2006; received in revised form 17 April 2007; accepted 21 April 2007
Available online 25 April 2007
Communicated by R. Wu
Abstract
We have developed an efcient numerical method for exciton states conned in quantum boxes. The exciton wave function is expanded in terms
of discrete variable representation basis functions. Our numerical approach has proved to be computationally much less demanding in comparison
with the conventional conguration-interaction one.
2007 Elsevier B.V. All rights reserved.
PACS: 71.15.Dx; 71.35.-y; 73.21.La
Keywords: Excitons; Quantum box; Discrete variable representation
1. Introduction
Semiconductor nanostructures based on advanced fabrica-
tion technologies [13] have attracted considerable attentions
in recent years due to their wide range applications to func-
tional materials [46]. In the nanostructures, the quantum con-
nement has a great inuence on excitonic properties, such as
density of states, binding energy, and oscillator strength. These
properties are highly relevant to the functions inherent in the
semiconductor nanostructures. Since the connement effects
strongly depend on dimensions, sizes, and shapes of the nano-
structures, it is crucial to elucidate the exciton states in various
nanostructures.
Exciton states conned in nanostructures have often been in-
vestigated theoretically by using either a variational approach
[7,8] or a conguration-interaction (CI) approach [7,9]. In the
variational approach, the exciton wave function is represented
by a trial function. Validity of the approach depends on the
choice of the trial function and, in general, its application to ex-
cited exciton states is practically difcult. In the conventional
CI (cCI) approach, on the other hand, the wave function is ex-
*
Corresponding author.
E-mail address: nobusada@ims.ac.jp (K. Nobusada).
panded in terms of an appropriate basis set, and thus the ground
and excited states can be described accurately with a large num-
ber of bases. As will be shown later, the cCI approach requires
huge computational costs due to the multi-dimensional inte-
grals to calculate the Coulomb matrix elements. As a result, the
cCI approach is not suitable for calculating multi-exciton states,
dynamics of excitons, or excitons in complex nanostructures.
In the present study, we have developed an efcient numer-
ical method for calculating exciton states conned in nano-
structures. To overcome the drawback of the cCI approach
mentioned above, we adopt the discrete variable representa-
tion (DVR) for constructing a basis set [1114]. The DVR
method has been employed intensively in molecular science
and proved to be a powerful and efcient manner to calcu-
late molecular properties [15,16]. In the DVR-based CI (DVR-
CI) approach, the Coulomb matrix is reduced to a diagonal
matrix having single-point values with no integrals. This de-
sirable property is due to the fact that the DVR basis func-
tions are coordinate eigenfunctions localized on each grid point
associated with the Gaussian quadrature rule and satisfy the
Kronecker delta property at the grid points. To conrm the
numerical advantage of the DVR-CI approach over the cCI
one, we fully compare the results obtained by the two ap-
proaches.
0375-9601/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.physleta.2007.04.067
Y. Kubota, K. Nobusada / Physics Letters A 369 (2007) 128131 129
Fig. 1. Schematic diagram of a two-dimensional quantum box with width L.
2. Method of calculation
We consider exciton states conned in nanostructures. The
Hamiltonian operator is given by
(1) H=

i=e,h
_

h
2
2m
i

2
i
V
i
( r
i
)
_

e
2
r
eh
,
where m
e
and m
h
are the electron and hole effective masses.
The last term in Eq. (1) is the electronhole Coulomb interac-
tion potential with e and being the elementary charge and
the dielectric constant, respectively. V
e
(V
h
) is the conne-
ment potential for the electron (hole), which is determined by
the conned structure. Although the present method is gener-
ally applicable to various types of nanostructures, we choose a
two-dimensional square box with width L as an example (see
Fig. 1). The connement potential is given in the form of
(2) V
i
( r
i
) =
_
0 for 0 < x
i
< L and 0 < y
i
< L,
otherwise,
where r
i
=(x
i
, y
i
).
In the cCI approach, the exciton wave function can be ex-
panded in terms of products of the electron and hole single-
particle eigenfunctions [7,9]

jlmn
=
2
L
sin(k
j
x
e
) sin(k
l
y
e
)
2
L
sin(k
m
x
h
) sin(k
n
y
h
),
(3) j, l =1, . . . , N
e
, m, n =1, . . . , N
h
,
where k
j
(= j/L) refers to the wave number, and N
e
and
N
h
are the number of basis functions for the electron and hole
states, respectively. Then, the Hamiltonian matrix is written by
(4)
H
jlmn
j
/
l
/
m
/
n
/
=E
(0)
jlmn

jj
/
ll
/
mm
/
nn
/

_
d r
e
_
d r
h

jlmn
e
2
r
eh

j
/
l
/
m
/
n
/ ,
(5) E
(0)
jlmn
=

h
2
2m
e
_
k
2
j
k
2
l
_

h
2
2m
h
_
k
2
m
k
2
n
_
.
This expression clearly shows that the computation of the ma-
trix H requires the order of N
2
four-dimensional numerical
integrals, where N (=N
2
e
N
2
h
) is the dimension of the matrix H.
In the case of a three-dimensional square box, N
3
six-dimen-
sional numerical integrals are required. These multi-dimen-
sional integrals are apparently computationally demanding and
become a practical difculty in calculating exciton states [9,10].
To reduce the computational costs of the cCI approach, we
alternatively use DVR to construct the basis set. In the DVR-CI
formalism, the single-particle basis functions are transformed
into the grid-point ones. The transformation matrix is deter-
mined by the grid points and the weights associated with an ap-
propriate Gaussian quadrature rule [11]. Since the second term
in Eq. (4) can be evaluated by the GaussChebyshev quadra-
ture of the second kind [17], we use the GaussChebyshev DVR
[18]. The grid points are given by
(6) x
e,s
=
L
N
e
1
s, s =1, . . . , N
e
,
and the weights are constants, w
xe,s
= L/(N
e
1). For the
other coordinates y
e
, x
h
, and y
h
, the grid points and the weights
are given in a similar way. Then, the transformation matrix is
derived in the form of
U st uv
jlmn
=

w
xe,s
w
ye,t
w
xh,u
w
yh,v

jlmn
( r
e,st
, r
h,uv
)
=
2
N
e
1
sin(k
j
x
e,s
) sin(k
l
y
e,t
)
(7)
2
N
h
1
sin(k
m
x
h,u
) sin(k
n
y
h,v
).
We nally obtain the matrix H in DVR as follows:
H
(DVR)
st uv
s
/
t
/
u
/
v
/
=
_
UHU
1
_
st uv
s
/
t
/
u
/
v
/
=T
(e)
ss
/

t t
/
uu
/
vv
/ T
(e)
t t
/

ss
/
uu
/
vv
/
T
(h)
uu
/

ss
/
t t
/
vv
/ T
(h)
vv
/

ss
/
t t
/
uu
/
(8)
e
2
[ r
e,st
r
h,uv
[

ss
/
t t
/
uu
/
vv
/ ,
where T denotes the kinetic energy term given by [19]
T
(i)
ss
/
=

h
2
2m
i

2
L
2
(1)
ss
/
2
__
sin
2

2(N
i
1)
(s s
/
)
_
1

_
sin
2

2(N
i
1)
(s s
/
)
_
1
_
for s ,=s
/
,
=

h
2
2m
i

2
L
2
1
2
_
1
3
_
2(N
i
1)
2
1
_
(9)
_
sin
2

N
i
1
s
_
1
_
for s =s
/
.
Owing to the great advantage of DVR, the matrix of the multi-
dimensional Coulomb integrals [Eq. (4)] is reduced to the di-
agonal matrix with single-point values [12]. Consequently, the
computational time for the matrix elements decreases sharply.
In addition, the matrix H becomes very sparse, i.e., the number
of nonzero elements is only N (2N
e
2N
h
). Therefore, we
can deal with a large size of Hamiltonian matrix by combining
the Lanczos diagonalization. These great advantages allow us
to calculate even the excited exciton states very efciently. The
eigenvalues of H correspond to the energy of each exciton state
E
ex
, and the exciton wave function is expressed by
( r
e
, r
h
) =

st uv
a
st uv

(DVR)
st uv
( r
e
, r
h
)
(10) =

st uv
jlmn
a
st uv
U st uv
jlmn

jlmn
( r
e
, r
h
),
130 Y. Kubota, K. Nobusada / Physics Letters A 369 (2007) 128131
where a
st uv
is the expansion coefcient to be determined for
each exciton state. Here, we should note one of the advan-
tages of the DVR-CI approach. The characteristic properties
of the exciton states can be obtained by a simple sum since
the DVR basis functions satisfy the following Kronecker delta
property [18]
(11)
(DVR)
st uv
( r
e,s
/
t
/ , r
h,u
/
v
/ ) =
N
e
1
L
N
h
1
L

ss
/
t t
/
uu
/
vv
/ .
For example, the expectation value of the electronhole separa-
tion r
eh
is given by
(12) [r
2
eh
[)
1
2
=
_

st uv
a
2
st uv
[ r
e,st
r
h,uv
[
2
_1
2
.
3. Numerical results
We demonstrate the numerical performance of the present
DVR-CI approach by comparing the result with the cCI one. As
an interesting example, we use a set of parameters for the typi-
cal GaAs system: =13.1, m
e
=0.067m
0
, and m
h
=0.377m
0
(the heavy-hole effective mass) with m
0
being the electron mass
in the vacuum. The width of the connement box is set to
be L = 10 nm. In the DVR-CI approach, to avoid the numer-
ical divergence of the Coulomb interaction potential, we set
N
h
=N
e
1 such that (N
e
1) and (N
h
1) satisfy a coprime
integer pair. Fig. 2(a) shows the convergence behavior of the ex-
citon binding energy E
b
(=E
(0)
grnd
E
ex
) for the exciton ground
state as a function of N, where E
(0)
grnd
denotes the lowest-energy
of the noninteracting single electron and hole system. The nu-
merical accuracy is rapidly improved in the small N region and
the calculated data converges to a certain value with increasing
N. In Fig. 2(b), we plot the computational time (closed circles)
required to obtain each data point in Fig. 2(a). The numerical
calculations were carried out by using a 3.8 GHz Pentium 4
PC. We note that the horizontal axis in Fig. 2(b) is equal to the
vertical axis in Fig. 2(a). The inset is the magnication ranging
from E
b
= 55 to 58.5 meV in a semi-log plot. We also show
the computational time for the cCI approach (open circles) re-
quired to obtain E
b
at the same level of accuracy corresponding
to those for the DVR-CI one. As is clearly seen from the inset in
Fig. 2(b), the DVR-CI approach is almost ten times faster than
the cCI one.
The DVR-CI computational advantage over the cCI ap-
proach becomes remarkable when calculating the exciton states
in larger L or the excited exciton states. This is simply be-
cause such exciton wave function should be expanded in terms
of a larger number of single-particle eigenfunctions [20] and
thus the multi-dimensional integral in the cCI approach requires
much more computational costs. On the other hand, the DVR
basis functions are the coordinate eigenfunctions specied at
each grid point, so that the number of DVR basis functions
used in the expansion is not directly dependent on the charac-
teristics of the exciton wave functions. The DVR-CI approach
does not actually require much additional computational time
even when calculating the excitons in the quantum boxes with
up to L =30 nm or for up to the third excited exciton state. In
Fig. 2. (a) Convergence behavior of the exciton binding energy E
b
for the
ground state as a function of the total number of the grid points based on the
DVR-CI approach. (b) Computational time (") required to obtain each data
point in Fig. 2(a) in comparison with that of the cCI approach (!) required to
obtain the binding energies corresponding to those of the DVR-CI one. The
inset is the magnication in a semi-log plot.
other words, the DVR-CI approach is not necessarily highly op-
timized for calculating the exciton ground state in a small quan-
tum box in which the exciton wave function can be expanded
in terms of a small number of single-particle eigenfunctions,
but a versatile numerical method for calculating the excitons
in any size of the boxes and in highly excited states. As de-
scribed above in the case of the quantum box with L =10 nm,
the DVR-CI approach is almost ten times faster than the cCI
approach. Since it is computationally demanding to carry out
the cCI calculations of the excitons in larger boxes (> 10 nm)
or for excited states, we cannot directly demonstrate the high
performance of the present DVR-CI approach. However, it is
reasonably expected that the DVR-CI approach has a greater
advantage over the cCI one when calculating the excitons in
such systems. Our preliminary calculations of the third excited
state for L = 20 nm showed that the DVR-CI approach was
about 300 times faster than the cCI one.
Before concluding this Letter, we briey mention the char-
acteristic properties of the ground and excited exciton states
calculated by the present efcient numerical method. The de-
tailed discussion of these properties will be reported in the
forthcoming paper. Fig. 3 shows the L-dependence of the ex-
citon energy E
ex
measured from E
(0)
grnd
. It is clearly seen from
this gure that the energies of the excited exciton states for
L 15 nm are partly larger than E
(0)
grnd
. In contrast to bulk sys-
tems, the single-electron states in the quantum box form the
Y. Kubota, K. Nobusada / Physics Letters A 369 (2007) 128131 131
Fig. 3. L-dependence of the exciton energy E
ex
for the ground (dashed curve),
rst (dot-dashed curve), second (solid curve), and third (dotted curve) excited
exciton states. The dashed horizontal line denotes the reference energy E
(0)
grnd
.
Fig. 4. L-dependence of the electronhole separation r
eh
for the ground
(dashed curve), rst (dot-dashed curve), second (solid curve), and third (dot-
ted curve) excited exciton states.
discrete energy levels. Some of the excited exciton states are en-
ergetically stabilized due to the electronhole interaction below
these energy levels. Therefore, such excited exciton states are
experimentally detectable. Fig. 4 displays the L-dependence of
the electronhole separation r
eh
. It is demonstrated that due to
the quantum connement effect, the crossover of r
eh
appears
at L =22 nm between the second and third excited states.
4. Summary
We have developed the efcient numerical method for cal-
culating the exciton states conned in quantum boxes. This
method was found to be much more efcient and promising
for calculating the exciton states than the cCI approach. The
computational time for the Hamiltonian matrix elements de-
creased sharply and the resultant matrix is sparse, so that we
can deal with a large number of basis functions. Although we
presented the results in the two-dimensional quantum boxes
in this study, the DVR-CI approach is straightforwardly ap-
plied to three-dimensional quantumboxes as is easily seen from
Eqs. (7) and (8). Since the DVR-CI formalism is based on the
Gaussian quadrature rule only, it is also applicable to the ex-
citon states in complex nanostructures or to the multi-exciton
states. We believe that our numerical method has high poten-
tialities for studying properties of bi- and triexcitons in various
nanostructures.
Acknowledgements
We would like to thank G.W. Bryant and H. Gotoh for valu-
able comments on the CI approach. This work was supported
by Grant-in-Aid (Nos. 17550024, 18066019) and by the Next
Generation Super Computing Project, Nanoscience Program,
MEXT, Japan. We employed the Lanczos algorithm in TIT-
PACK Version 2 developed by H. Nishimori.
References
[1] P.M. Petroff, J. Cibert, A.C. Gossard, G.J. Dolan, C.W. Tu, J. Vac. Sci.
Technol. B 5 (1987) 1204.
[2] Y. Miyamoto, M. Cao, Y. Shingai, K. Furuya, Y. Suematsu, K.G. Raviku-
mar, S. Arai, Jpn. J. Appl. Phys. 26 (1987) L225.
[3] D.K. Ferry, S.M. Goodnick, Transport in Nanostructures, Cambridge
Univ. Press, Cambridge, 1997.
[4] K. Imamura, Y. Sugiyama, Y. Nakata, S. Muto, N. Yokoyama, Jpn. J. Appl.
Phys. 34 (1995) L1445.
[5] N. Kirstaedter, O.G. Schmidt, N.N. Ledentsov, D. Bimberg, V.M. Usti-
nov, A.Y. Egorov, A.E. Zhukov, M.V. Maximov, P.S. Kopev, Z.I. Alferov,
Appl. Phys. Lett. 69 (1996) 1226.
[6] X. Jiang, S.S. Li, M.Z. Tidrow, Physica E 5 (1999) 27.
[7] G.W. Bryant, Phys. Rev. B 37 (1988) 8763.
[8] P. Harrison, Quantum Wells, Wires and Dots, second ed., Wiley, London,
2005.
[9] H. Gotoh, H. Ando, J. Appl. Phys. 82 (1997) 1667.
[10] In the case of the quantum box, Bryant [7,21] showed that the multi-
dimensional integrals of the Coulomb potential can be reduced to single
integrals of a combination of complex error functions. This is an unusual
case due to the extremely simple boundary condition of the quantum box.
[11] J.V. Lill, G.A. Perker, J.C. Light, Chem. Phys. Lett. 89 (1982) 483.
[12] R.W. Heather, J.C. Light, J. Chem. Phys. 79 (1983) 147.
[13] J.C. Light, I.P. Hamilton, J.V. Lill, J. Chem. Phys. 82 (1985) 1400.
[14] S.E. Choi, J.C. Light, J. Chem. Phys. 90 (1989) 2593.
[15] K. Nobusada, O.I. Tolstikhin, H. Nakamura, J. Chem. Phys. 108 (1998)
8922.
[16] K. Nobusada, J. Phys. B 35 (2002) 3055.
[17] M. Abramowitz, I.A. Stegun (Eds.), Handbook of Mathematical Func-
tions, Dover, New York, 1972.
[18] J.T. Muckerman, Chem. Phys. Lett. 173 (1990) 200.
[19] D.T. Colbert, W.H. Miller, J. Chem. Phys. 96 (1992) 1982.
[20] G.W. Bryant, D.B. Murray, A.H. MacDonald, Superlatt. Microstruct. 3
(1987) 211.
[21] G.W. Bryant, Phys. Rev. B 31 (1985) 7812.