Computer Physics Communications 183 (2012) 16541657

Contents lists available at SciVerse ScienceDirect

Computer Physics Communications

www.elsevier.com/locate/cpc

Two interacting electrons conned in a 3D parabolic cylindrically symmetric

potential, in presence of axial magnetic eld: A nite element approach
Hanz Y. Ramrez, Alejandro Santana
Research Center in Basic and Applied Sciences, Universidad Antonio Nario, Cra. 3E N 47A-15 B4 P3, Bogota, Colombia

a r t i c l e i n f o a b s t r a c t

Article history: In this work, we explore the performance and accuracy of nite element numerical simulations for
Received 2 September 2011 the relevant and well-known problem of two interacting electrons conned in a parabolic cylindrically
Accepted 7 March 2012 symmetric potential, and under the inuence of axial magnetic eld. By using a full conguration
Available online 8 March 2012
interaction method as reference, we compare two different ways to obtain the eigenvalues of the
Keywords:
system by nite element simulations, one of which approximately separates the Coulomb interaction
Quantum dots and averages the longitudinal part. We nd that in the regime of low aspect ratios, the results from
Interacting electrons the approximate scheme with approximation are quite optimal (< 0.05% of difference respect to the
Finite elements reference), but once the dots turn more and more thicker, the results become just acceptable (> 0.5%
Conguration interaction of difference), due to underestimation of the Coulomb interaction. On the other hand, for the non-
approximated nite element approach the results are consistently reliable along the different eld and
aspect ratio regimes (< 0.02% of difference respect to the reference). This allows us to present a novel,
ecient and highly accurate method for obtaining electronic structure of interacting particles in 0D
nanostructures.
2012 Elsevier B.V. All rights reserved.

1. Introduction sition), which has been well understood and experimentally ob-
served for years [11,12]. Capitalizing this feature is obviously fa-
Given its potential for applications in spintronics and optoelec- vored by accurate and straightforward schemes in the calculation
tronic devices, the problem of two interacting electrons conned in of electron eigenenergies. Motivated on this premise, we explore
a 0D heterostructure under axial magnetic eld, is of scientic and reliability and precision of the nite element (FE) method to ob-
technological importance and has been recurrently studied [13]. tain electronic structure of cylindrically symmetric quantum dots
It is well known that the 3D cylindrically symmetric problem can- modeled by a 3D parabolic conning potential. The possibility of
not be exactly reduced to a 1D problem due to the mixing of calculating eciently electronic states in nanostructures through
radial and axial coordinates by the Coulomb interaction, and that a well-packed and kindly interfaced software, not only helps sci-
consequently no exact analytical solution has been found so far entists to improve their understanding of these relevant systems,
(except for fully 2D [4,5], and 3D spherically symmetric dots with but lays bridges for engineers to become more directly engaged in
no eld [6]). Approximate solutions using a effective radial poten- boosting nanotechnology.
tial have been proposed but they are expected to work well mostly
in the quasi-2D dot limit [5,7]. However, advances in growth tech- 2. Model Hamiltonian
niques nowadays allow to fabricate dots with on-demand aspect
ratio making this limit just a very particular case [8,9]. For the The studied system consists of two interacting electrons con-
fully 3D problem an approximate estimation of the interaction in ned by a fully 3D parabolic potential axially symmetric, in the so-
the z-direction was proposed by Nazmitdinov et al., by using semi- called Faraday conguration, i.e. under the inuence of a transver-
classical tools to separate the radial and axial variables [10].
sal magnetic eld B = (0, 0, B ). Within the effective-mass approx-
Between the characteristics of these two electron systems, per-
imation, the corresponding Hamiltonian in effective atomic units
haps the most promising for technological implementation is the
(h = e = me = (40 )1 = 1) is written
change of the total spin number of the ground state related to
interaction with the external magnetic eld (singlettriplet tran- i =2 
 
[pi + A(ri )]2 m  2 2
  
H total = +  xi + y 2i 2 2
+
zi
2m 2
i =1
* Corresponding author.
1
+ g B (s1 + s2 )B,
E-mail addresses: hanzyrg@hotmail.com (H.Y. Ramrez),
+ (1)
alejandro.santana@uan.edu.co (A. Santana).
|r2 r1 |
0010-4655/$ see front matter 2012 Elsevier B.V. All rights reserved.
doi:10.1016/j.cpc.2012.03.002
 H.Y. Ramrez, A. Santana / Computer Physics Communications 183 (2012) 16541657 1655

out numerical diagonalization for nding eigenenergies of the sys-

tem. The matrix elements corresponding to the term H C are built
up using expressions (13) and (15) of [13].
We use in our calculations basis of up to 1225 two particle
states which includes s-, p-, d-, f-, and g-like single particle or-
bitals. This allows us to work with convergence levels of below
0.002% of relative error between subsequent runs as function of
the basis number, as shown in Fig. 1(b). Additionally, for checking
reliability of the code, it was tested to reproduce analytical results
for the full 2D case and for the spherical 3D case [4,6], and previ-
ous similar CI calculations [13].

4. Finite element calculations

To determinate the equations to be solved numerically by FE

Fig. 1. (Color online.) (a) Depiction of the 3D conning potential for different aspect
ratio values. (b) Convergence of the ground state eigenenergy as function of the
number of elements in the basis, obtained by the conguration interaction method.
(c) Convergence of the ground state eigenenergy as function of the number of ele-
ments in the calculation domain, obtained by the FE method for the 2D eigenvalue
problem. (d) as in (c) but obtained by the FE method for the 1D eigenvalue problem.
Results in (b), (c) and (d) correspond to an aspect ratio of a = 1.
methods, it is necessary to work on the Hamiltonian (1) by us-
ing the standard procedure of introducing centre-of-mass [R =
( X , Y , Z ) 12 (r1 + r2 )] and relative [r = (x, y , z) (r2 r1 )] co-
ordinates. Thus, ignoring the Zeeman term the Hamiltonian reads
H total = H cm + H r , (3)
where

[P + 2A(R)]2    
where m is the electron effective mass in units of electron mass H cm = + m 2 X 2 + Y 2 +
2 2
Z , (4)
me , is the dielectric constant, g is the effective g factor, s1 4m
(s2 ) is the total spin of electron 1 (2), and A(ri ) = B2 ( y i , xi , 0) and
is the vector potential in the symmetric gauge. We adopt a cylin-
drically symmetric parabolic conning potential with character- [p + 12 A(r)]2 m    
istic frequencies  and , for the in plane and growth di- H rel = + 2 x2 + y 2 +
2 2
z
m 4
rections, respectively (the corresponding characteristic lengths are 1
l = (1/2m  )1/2 and l = (1/2m )1/2 ). The choice of this unit + . (5)
system obeys the sake of generality, since energy results are pre- x2 + y 2 + z2

sented in units of effective Hartree Ha = m 2
Ha, making them This separability of the Hamiltonian is a peculiar property of
non-dependent on any particular material (thus, lengths are to be the parabolic connement, which allows us to decouple the sys-
taken in effective Bohr radius aB = m a B ). tem into the energy of the centre-of-mass and the energy of the
l
The aspect ratio a l
is dened to describe the geometric relative part [6,20]. The H cm part is a simple harmonic oscillator
character of the dot in such a way that a > 0 refers to prolate with different connement in the plane and in the z-direction with
ellipsoids (nanorod quantum dots), a < 0 to oblate ellipsoids (nan- uniform magnetic eld applied normally to the plane, and its so-
odisk quantum dots), and a = 1 to the particular spherical case lutions are the same as the ones used for the single particle basis
(see Fig. 1(a)). in the CI calculations, but for a particle with electric charge 2 and
effective mass 2m .
3. Full CI calculation On turn, the relative part represents a much more complicated
problem related to the mixing of variables due to the Coulomb in-
For monitoring the FE results, we perform full conguration in- teraction. Taking advantage of the axial symmetry of the problem,
teraction calculations for the studied dots. The implementation of the Hamiltonian (5) can be written in the form
the method is done following the work by Chen et al. in [13].
 
This kind of numerical diagonalization approach has been previ- 1 1
2 B2
Hr = + 2 + 2
ously used for studying dimensionality effects, and in explaining m z 16m
experimental results in quantum dots [7,1417].

m  2 2 2 2
 1 1 1
The four terms of the Hamiltonian in (1), can be labeled ac- +  + z +
cording to 4 2 + z2 m 2 2
B
H total = H 1 + H 2 + H C + H z , (2) i , (6)
2m i
where the H 1 (H 2 ) term corresponds to the kinetic and conne-
where clearly the angle dependence can be separated from ra-
ment energy for electron 1 (2), H C stands for the Coulomb inter-
dial and z-components, and replaced by a separation constant
action between electron 1 and 2, and H Z represents the Zeeman
l = 0, 1, +1, 2, +2, . . . associated to the quantum number of the
interaction between the total spin and the magnetic eld. From
z-component of angular momentum [6]. Then, Hamiltonian (6) be-
now on, for simplicity we do not further consider this Zeeman
comes
term since its energy contribution E Z = g B S B, is not relevant

 
for the comparison of methods in solving the coordinate depen- 1 1 2 m  2 2 2 2

dent terms. Hr = + + + z
m z2 4
The two single particle terms H 1 and H 2 are exactly diagonal-
izable in the well-known FockDarwin basis [18,19]. Thus, we use 1 l2 c l 1
+ + + , (7)
this basis to write the two particle Hamiltonian and then to carry m 2 2 2 + z2
1656 H.Y. Ramrez, A. Santana / Computer Physics Communications 183 (2012) 16541657


with c = B /m and = 2 + c2 /4, the cyclotron and effective
in-plane frequencies, respectively.
Aiming further simplication of the relative
 Hamiltonian, it is
possible to introduce the quantities l0 m , l 1 and
1
0

, in such a way that changing to dimensionless coordinates

/l0 and z/l0 , the time independent Schrdinger equation for
this term looks

 
1 2 1 
+ 2 + 2 + 2 2
4

l2
+ 2+ (, ) = l (, ), (8)
2 + 2
and the eigenenergy value corresponding to the relative part of the
Hamiltonian reads
c l Fig. 2. (Color online.) Eigenenergies as functions of the axial magnetic eld, obtained
E lr = l + . (9) from full CI calculations for three different aspect ratios (ground and rst excited
2 states). The dotted lines show the non-interacting two electron-energies for each
case.
Numerical treatment of the fully 3D problem requires to solve
the partial differential equation (8) for obtaining eigenenergies.
An simplied approach for this problem has been proposed by The corresponding eigenenergies for non-interacting particles are
Nazmitdinov et al. [10], in which implementation of the computa- also shown. Two important features are observed: First, the ne-
tional solution is favored by reduction of the dimensionality of the glecting of the Coulomb interaction between electrons introduce a
problem, turning the partial differential equation (8) in an approx- signicant error respect to the actual eigenenergies of the system,
imated ordinary differential eigenvalue problem (EVP) [20]. and then it should be strongly avoided (especially for nanorod-
Namely, the approximation drives to the modied time inde- like dots). Second, if Zeeman energy would be considered (which
pendent Schrdinger equation at principle does not depend on the geometry), the singlettriplet
ground state transition generally would require higher magnetic

 
1 2 l2 elds for smaller aspect ratios. This is contrary to the predicted
+ + behavior in [10], where the analysis is made assuming quasi-2D
4 2
 
dots, but makes physical sense due to the decreasing energy spac-
2  2 ing with the increase of the aspect ratio.
+ K ( ) = l ( ), (10)
2 +  2 +  2
2 In order to determinate accuracy of the FE results respect to the
CI ones, we compare for both FE approaches (2D and 1D EVPs), the
 2 relative differences for the obtained i-th eigenvalues as functions
where K ( 2 + 2 ) represents the rst kind-complete elliptical in-
 2 4j of the magnetic eld and aspect ratio by dening
tegral evaluated in 2 + 2 ,  2 , and j = n + 12 (with
 
n = 0, 1, 2, . . .) stands as a quantum number for excitations in the  E CI E 2DFE 
 E CI ,2D 
100 i i , (12)
approximately separated coordinate . Thus, the eigenenergy cor- i
E CI 
i
responding to the relative part of the Hamiltonian is
  and
w c l  
E lr = l + j + . (11)  E CI E 1DFE 
2  E CI ,1D
100 i i , (13)
i  E CI 
i
To solve (10) numerically, is obviously expected to be compu-
tationally less expensive than to solve (8). Nevertheless, how less respectively; where E CI i
represents the i-th eigenvalue obtained
expensive and how large is the error introduced by the proposed from the CI method while E 2DFE i
(E 1DFE
i
) the i-th eigenvalue ob-
approximation, to the best of our knowledge, remained so far as tained from the 2D (1D) FE calculation.
subject of speculation. Then, for solving the eigenvalue problems Fig. 3 allows to compare the mismatch between each of the
(8) and (10) we perform nite element calculations at different FE schemes and the CI results. Clearly the approximated charac-
magnetic eld and aspect ratio regimes. ter of the 1D EVP approach is reected in these plots, and its
For the FE calculations, the well-reputed nite element tool increased inaccuracy for higher aspect ratios becomes evident. It
COMSOL was used [21]. Convergence tests for the obtained eigen- is interesting to note that for strong elds the introduced error is
values of the systems were carried out for both cases, 1D and 2D slightly erased due to the increase of the ratio between conne-
EVPs. Convergence levels below the 0.002% for subsequent runs as ment strength and Coulomb interaction magnitude, on which the
functions of the number of elements in the domains were achieved approximation is exerted.
(see Figs. 1(c) and 1(d)). While the 2D EVP scheme ts very well to the reference values
in all the different studied regimes, the reliability of the 1D scheme
5. Results starts failing notably with the elongation of the dot. The approach
performance is enhanced for the rst excited state as expected,
We x the in plane conning frequency in a reference value of considering the reduction of the Coulomb interaction associated to
18.33 Ha (which in the case of InAs dots would be equivalent to the charge delocalization of excited states.
50 meV; i.e. a characteristic length of 5.75 nm). The longitudinal Regarding calculation times. In Table 1, the elapsed time in
frequency values are thus dened by the studied aspect ratios. seconds for the three different calculations (carried out in the con-
Fig. 2 shows ground and rst excited eigenenergies for three ditions in which convergences reported in Fig. 1 are obtained),
different aspect ratios, as functions of the external magnetic eld. are shown. The calculations were made using a commercial core
 H.Y. Ramrez, A. Santana / Computer Physics Communications 183 (2012) 16541657
Fig. 3. (Color online.) Upper panel: Relative difference between the CI obtained
ground state eigenvalue and the FE obtained value from the 1D EVP (left), and the
FE obtained value from the 2D EVP (right). Lower panel: as in the upper panel but
for the rst excited state.
Table 1
Elapsed calculation times.
CI method FE 2D EVP FE 1D EVP
470 s 8s 2s
i5 processor of 3.6 GHz. Evidently the time consumptions for the
FE schemes, beyond being shorter than that for the CI approach,
conrm the eciency of the method. Albeit longer than the time
required for the 1D EVP calculation, the time for the 2D EVP seems
very reasonable and not much more computationally expensive.
Two electron wave function calculations can be easily imple-
mented in each of these three numerical approaches and they will
be subject of further studies. However it is worth noting that the
separation between centre-of-mass and relative part of the Hamil-
tonian leads to an interesting conditional nature of the relative part
of the wave function which can be thoroughly investigated by us-
ing the described FE approaches.
6. Conclusions
In summary, we have shown two different ways to use the
nite element method for calculating two interacting electron
eigenenergies in 0D nanostructures.
The so-called approach 1D EVP, where average of the interac-
tion in z is applied, presents eciency and fairly accurate results
1657
for quantum disk-like dots, however its reliability showed fails for
increased aspect ratios. This effect is clearly related to underesti-
mation of the Coulomb interaction by the average in the approxi-
mation.
On the other hand, the results from the non-approximated FE
approach exhibited highly accuracy when compared to the refer-
ence CI calculations. This optimal performance is consistent along
the different eld and aspect ratio regimes explored in this study.
One of the main advantages of the presented FE approaches
is their versatility to describe systems of various geometries and
materials. Thus, these methods allow us to study a wide range
of cases. Likewise, these exact results can be used as reference
data for more sophisticated methods as Density functional theory,
which recently has begun to be used in describing nanostructures.
As a main conclusion we propose the presented FE based
schemes (especially the so-called 1D EVP); as novel, ecient; and
reliable techniques to calculate electron structures in quantum
dots.
Acknowledgements
The authors would like to thank to the Faculty of Sciences of
Universidad Antonio Nario for nancial support. Also to Prof. An-
gela Camacho for fruitful discussions and generous cooperation.
References
[1] M. Reimann, M. Manninen, Rev. Mod. Phys. 74 (2002) 1283.
[2] M. Rontani, C. Cavazzoni, D. Bellucci, G. Goldoni, J. Chem. Phys. 124 (2006)
124102.
[3] A. Wensauer, M. Korkusinski, P. Hawrylak, Solid State Commun. 130 (2004)
115.
[4] M. Taut, Phys. Rev. A 48 (1993) 3561.
[5] M. Dineykhan, R.G. Nazmitdinov, Phys. Rev. B 55 (1997) 13707.
[6] M. Taut, J. Phys. A: Math. Gen. 27 (1994) 1045.
[7] M. Rontani, F. Rossi, F. Manghi, E. Molinari, Phys. Rev. B. 59 (1999) 10165.
[8] S. Kalliakos, M. Rontani, V. Pellegrini, C.P. Garca, A. Pinczuk, G. Goldoni, E.
Molinari, L. Pfeiffer, K. West, Nature Phys. 4 (2008) 467.
[9] L.P. Kouwenhoven, D.G. Austing, S. Tarucha, Rep. Prog. Phys. 64 (2001) 701.
[10] M. Dineykhan, R.G. Nazmitdinov, J. Phys.: Cond. Matter 11 (1999) L83.
[11] M. Wagner, U. Merkt, A.V. Chaplik, Phys. Rev. B 45 (1992) 1951.
[12] R.C. Ashoori, H.L. Stormer, J.S. Weiner, L.N. Pfeiffer, K.W. Baldwin, K.W. West,
Phys. Rev. Lett. 71 (1993) 613.
[13] Y.T. Chen, S.J. Cheng, C.S. Tang, Phys. Rev. B 81 (2010) 245311.
[14] M. Rontani, F. Rossi, F. Manghi, E. Molinari, Appl. Phys. Lett. 72 (1998) 957.
[15] S. Tarucha, D.G. Austing, T. Honda, R.J. van der Hage, L.P. Kouwenhoven, Phys.
Rev. Lett. 77 (1996) 3613.
[16] C. Steinebach, C. Schller, G. Biese, D. Heitmann, K. Eberl, Phys. Rev. Lett. 57
(1998) 1703.
[17] C. Ellenberger, T. Ihn, C. Yannouleas, U. Landman, K. Ensslin, D. Driscoll, A.C.
Gossard, Phys. Rev. Lett. 96 (2006) 126806.
[18] R. de Sousa, S. Das Sarma, Phys. Rev. B. 67 (2003) 033301.
[19] O. Makarovsky, O. Thomas, A.G. Balanov, L. Eaves, A. Patan, R.P. Campion, C.T.
Foxon, E.E. Vdovin, D.K. Maude, G. Kiesslich, R.J. Airey, Phys. Rev. Lett. 101
(2008) 226807.
[20] T. Ihn, Semiconductor Nanostructures, Oxford University Press, New York, 2010,
p. 366.
[21] COMSOL Multiphysics Version 4.0 released, Sealing Technology 2010 (2010)
1350.