Professional Documents
Culture Documents
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 propose a Control Volume Function Approximation (CVFA) method to solve
Received 25 March 2014 equations involving the fractional Laplacian. The function approximation part is carried out
Received in revised form 22 May 2014 with Radial Basis Function (RBF) interpolation. The physical application of interest is the
Accepted 7 June 2014
eigenvalue problem for the time independent fractional Schrödinger equation. Fractional
Available online 3 July 2014
derivatives are considered in the Riesz potentials sense.
Ó 2014 Elsevier B.V. All rights reserved.
Keywords:
Fractional quantum mechanics
Fractional Laplacian
Control Volume method
Radial Basis Functions
1. Introduction
In recent years, fractional calculus has become a topic of great research activity. This is due to the fact that generalizations
of integer to fractional derivatives in connection with physical systems, have found applications in most fields. See for
instance the historical survey presented in Machado et al. [1].
The application of interest in this work is to fractional quantum mechanics. Here, the Schrödinger equation is generalized
by considering fractional powers of the Laplacian, the fractional Laplacian, in the Hamiltonian. A physically sound general-
ization is that of Laskin [2–4]. Therein a derivation of the fractional Schrödinger equation is developed based on Feynman
path integrals over Levy trajectories. Following these works, there has been an explosion on research regarding fractional
Schrödinger equations. For instance, in Wang and Xu [5] a fractional Schrödinger equation with space–time fractional deriv-
atives is considered. A solution is proposed by the method of integral transforms. Application to the free particle and square
well potential is presented. Of related interest is Tarasov [6], where a fractional generalization of Heisenberg equation is
developed, yielding a generalization to quantum Hamiltonian systems. An alternative approach, the D-deformed calculus,
to model fractional-dimension systems is introduced in Matos-Abigue [7]. The free particle and the harmonic oscillator
are considered as systems in the so called framework of D-deformed quantum mechanics, the process resembles to the clas-
sical case. Finally, on the theoretical side, Ionescu and Pusateri [8] carry out a well-posedness analysis for a nonlinear frac-
tional Schrödinger equation in one dimension.
This work is from a numerical perspective. Our purpose is to introduce a numerical method to solve the time independent
Schrödinger equation, namely,
q
Partially supported by CONACYT - Mexico.
⇑ Corresponding author.
E-mail addresses: moreles@cimat.mx, mamorele@ucalgary.ca (M.A. Moreles).
http://dx.doi.org/10.1016/j.cnsns.2014.06.013
1007-5704/Ó 2014 Elsevier B.V. All rights reserved.
A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613 605
a=2
2
Da
h D þ VðrÞ wðrÞ ¼ EwðrÞ; 1 < a 6 2:
Here Da is a generalized diffusion coefficient, and the fractional Laplacian is defined in terms of the Fourier Transform,
namely
a=2 Z p r
1
jpja wðpÞ:
2 3
h D wðrÞ ¼ 3
d p exp i
ð2pÞ h
Numerical methods to solve this eigenvalue problem are scarce at best. For the one dimensional case, a collocation method is
presented in Amore et al. [9]. Within the method, sampling functions, sampling points and interval of definition of the wave
functions, are to be determined according to the prescribed boundary conditions. There is no apparent extension to higher
dimensions. In Zoia et al. [10], a matrix representation of the fractional Laplacian is introduced. The eigenvalues and eigen-
functions of such a matrix, converge to the eigenvalues and eigenfunctions of the fractional Laplacian when the size of the
matrix tends to infinity. Apparently, the method requires large matrix sizes. For instance, in Kwaśnicki [11], the method is
used to approximate the eigenvalues associated to the infinite well potential. Therein, a matrix of size Oð103 Þ is required.
A difficulty on approximating the Fractional Laplacian, or any fractional derivative for that matter, is its non local property.
Consequently, one is led to full and large matrix systems. Moreover, on bounded domains boundary conditions need to be
dealt carefully. To address these issues, we propose a method based on the Control Volume Function Approximation method
(CVFA), see Li et al. [12]. After integrating on control volumes, function approximation is carried out on the boundaries by local
interpolation. Our proposal is to use Radial Basis Function (RBF) for global interpolation. It is well known that RBF approxi-
mations lead to ill-posed full matrix systems, but give satisfactory results on coarse discretizations. As we shall see, the latter,
and the global feature of RBF are well suited to approximate the fractional Laplacian in the eigenvalue problem. We provide
numerical results for the one dimensional cases in the works above. We remark that in all examples it suffices to solve Oð102 Þ
matrix problems. The matrix eigenvalue problem is solved simply by the inverse power method and deflation.
The outline is as follows.
In Section 2 we introduce the classical definition of the fractional Laplacian in terms of Riesz potentials and the Fourier
Transform. The restriction to bounded domains is as in Musina and Nazarov [13]. In Section 3 we introduce a Control Volume
Radial Basis Function (CVRBF) Method for the eigenvalue problem, in any dimension. We use the simplest radial function,
/ðrÞ ¼ r, for approximation. The matrix problem for the one dimensional case is constructed in Section 4. Numerical results
are presented in Section 5. We solve the eigenvalue problem for the infinite well potential, the harmonic potential and the
quartic anharmonic potential. In Section 6 we draw conclusions and comment of future work.
The results in this section are well known, proofs are found for instance in Helgason [14].
We shall use the definition of Fourier Transform as follows
Z
bf ðkÞ F ðf ÞðkÞ ¼ exp ðik xÞf ðxÞdx
Rd
with inverse
Z
1
F 1 ðFÞðxÞ ¼ exp ðik xÞFðkÞdk:
ð2pÞd Rd
Let S and S the spaces of tempered functions and tempered distributions respectively. For the case s < 0, the Sobolev
0
By means of the Fourier Transform, the fractional Laplacian operator, ðDÞs is defined for f 2 Hs by
h i
2s
ðDÞs f ¼ F 1 jkj F ðf Þ :
with
Cðc=2Þ
Hd ðcÞ ¼ 2c pd=2 :
dc
C 2
I0 f ¼ lim Ia f ¼ f
a!0
and
Let X be a Lipschitz bounded domain in Rd . We shall consider ðDÞs f defined on the space
e s ðXÞ ¼ f 2 Hs : supp f X
H :
For 0 < s < 1 the spectrum of ðDÞs is discrete and contained in ð0; 1Þ. See Musina and Nazarov [13].
Hereafter we let s ¼ a=2; a > 0. In light of (1) we can write
c . We consider
We have that the wave function wðxÞ in (3) is zero for x 2 X
h i
ðDÞa=2 þ VðxÞ wðxÞ ¼ E wðxÞ; x 2 X;
ð4Þ
wðxÞ ¼ 0; x 2 @ X:
A partition T h of X consists of a set of (open) control volumes V i :
[
N
X¼ V i; V i \ V j ¼ £; i – j:
i¼1
XNi Z Z Z
r I2a w mik dr þ VðxÞwðxÞ dx ¼ E wðxÞ dx; ð5Þ
k¼1 eik Vi Vi
where Rik is the number of interpolation nodes xik;j for eik and these nodes can be located on or surrounding V i .
The coordinate functions uik;j are defined as follows:
(
1 at node xik;j
uik;j ðxÞ ¼
0 at other nodes
The other terms are also approximated in terms of the nodal values wik;j , leading to a matrix eigenvalue problem. We show
this in examples below.
Here /ðrÞ ¼ r; r ¼ jxj, is a conditionally positive definite radial function or order 1. Consequently, the system of Eq. (6) is
complemented with
X
N
aj ¼ 0:
j¼1
This guarantees existence and uniqueness of the coefficients in (7), see Wendland [15].
Our aim is to test the method with the simplest radial function. Extension to other choices of radial functions is straight-
forward. Noteworthy, the thin-plate spline, uðrÞ ¼ r2 log r is used in Li et al. [12].
with boundaries
n o
@V i ¼ xi1 ; xiþ1 :
2 2
608 A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613
n o n o
In each boundary xi1 ; xiþ1 , we shall consider all points in the partition as the neighboring points. Hence, the inter-
2 2
polant is of the form
X
N þ1
/h ðxÞ ¼ aj x xj þ b0
j¼0
The computation of Aa , the sub matrix associated to the Fractional Laplacian, is shown below.
In the examples that follow, we consider a regular grid.
We remark that to approximate the first smallest eigenvalues of (8), we simply use the inverse power method and
deflation.
Extending by zero outside the interval ða; bÞ we write the Riesz potential for a function in the form
where
Z x
1
I2a
f ðxÞ ¼ j x yj1a f ðyÞ dy ð9Þ
H1 ð2 aÞ a
and
Z b
1
I2a
þ f ðxÞ ¼ j x yj1a f ðyÞ dy: ð10Þ
H1 ð2 aÞ x
d 2a 1 h i
I x xj ¼ ðxj aÞðx aÞ1a þ 2ðx xj Þ2a ðx aÞ2a ; x > xj
dx 2a
and
d 2a
I 1 ¼ ðx aÞ1a :
dx
A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613 609
Similarly
d 2a 1 h 2a
i
1a
I x xj ¼ 2ðxj xÞ2a þ ðb xÞ ðb xj Þðb xÞ ; x < xj ;
dx þ 2a
Remark. In the terminology of Musina and Nazarov [13], we are approximating the eigenvalue problem for the Dirichlet
fractional Laplacian. Therein it is also considered the eigenvalue problem for the Navier fractional Laplacian. Let us denote by
ED N
n , and En the corresponding eigenvalues.
The Navier case is solved in Laskin [3]. The normalized wave functions are
n pxo
ev en
w2mþ1 ðxÞ ¼ cos ð2m þ 1Þ ; m ¼ 0; 1; . . . ð16Þ
2
and
n p xo
wodd
2m ðxÞ ¼ sin ð2mÞ ; m ¼ 1; 2; . . . : ð17Þ
2
Table 1
Infinite potential well.
a E1 E2 E3 E4 E5 E6 E7 E8 E9 E10
1.5 1.5998 5.0669 9.6075 15.039 21.217 28.071 35.532 43.559 52.111 61.159
0.0017 0.0016 0.0016 0.0016 0.0015 0.0015 0.0015 0.0014 0.0014 0.0013
1.8 (1e3) 2.0501 7.5079 15.809 26.739 40.135 55.891 73.918 94.147 116.515 140.974
0.9695 0.9538 0.9216 0.8823 0.8341 0.7744 0.7033 0.6214 0.5266 0.4206
1.9 (1e3) 2.2448 8.5984 18.72 32.47 49.73 70.42 94.48 121.86 152.51 186.38
0.7162 0.6736 0.6335 0.5816 0.5207 0.4451 0.3552 0.2501 0.1301 0.0048
1.99 (1e3) 2.4438 9.7332 21.828 38.707 60.355 86.760 117.91 153.79 194.40 239.71
0.4311 0.3973 0.3521 0.2889 0.2100 0.1133 0.0011 0.3281 0.2834 0.4507
610 A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613
Fig. 1. First wave functions, wi ðxÞ; i ¼ 1; 2; 3; 4, for the infinite well potential. a ¼ 1:5 solid line. a ¼ 2 dashed line.
See Corollary 4 in Musina and Nazarov [13]. It is readily seen that our estimates in Table 1 satisfy this inequality.
VðxÞ ¼ x2 ; x 2 R:
We consider the fractional quantum harmonic oscillator
2 !a=2 3
2
4 d þx 25
wðxÞ ¼ E wðxÞ; x 2 R:
2
dx
Let DL ¼ ðL; LÞ. For approximation we seek w 2 H b a=2 ðDL Þ. We solve a 200
200 system for a ¼ 3=2 and L ¼ 5:5. Wave
functions are plotted in Fig. 2. In Table 2 we show numerical results in comparison to the method of Amore et al. [9], and
the WKB estimates in Laskin [4], namely
2a
p 1 2þa
EWKB
n ¼ n ; n ¼ 1; 2; . . . : ð20Þ
Bð1=2; 1=a þ 1Þ 2
Fig. 2. First wave functions, wi ðxÞ; i ¼ 1; 2; 3; 4, for the harmonic potential. a ¼ 1:5 solid line. a ¼ 2 dashed line.
Table 2
Harmonic oscillator.
a E1 E2 E3
1.5 1.0011 2.7076 4.1762
Rel. error 0.0000756 0.0004866 0.0007545
Rel. error WKB 0.0496 0.0024 0.0021
Table 3
Anharmonic oscillator.
a E1 E2 E3 E4 E5 E6 E7 E8 E9 E10
4/3 0.9947 2.8204 4.7065 6.6022 8.4794 10.369 12.249 14.136 16.016 17.902
0.0550 0.0028 0.0016 0.0004 0.0007 0.0001 0.0006 0.0004 0.0007 0.0006
Remark. Nontrivial issues of the collocation method of Amore et al. [9], are the choice of L and the number of sampling
points, N. It is noticed that the number of sampling points increases significantly with L. First they start with a number N of
collocation functions, then a number of sampling points and an optimal choice of L are determined. Refinement may be
required. In contrast, our method only requires a good estimate for the decay of the wave function.
Our last example is the anharmonic oscillator associated to the potential VðxÞ ¼ x4 .
For a ¼ 4=3 it is known that a good approximation for the eigenvalues is provided by the asymptotic WKB estimates. The
predicted spectrum is evenly spaced and is given by
2p 1
EWKB
n ¼ nþ : ð21Þ
Cð1=4ÞCð7=4Þ 2
We apply the Control Volume method with L ¼ 3 and solve a 300
300 system. Results are listed in Table 3, a ¼ 4=3
En ; n ¼ 1; 2; 3; . . . 10 and relative errors with respect to WKB. The first wave functions are plotted in Fig. 3.
Results in Table 3 show that the eigenvalues of the anharmonic oscillator follow the WKB line (21), see Fig. 4.
612 A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613
Fig. 3. First wave functions, wi ðxÞ; i ¼ 1; 2; 3; 4, for the anharmonic potential. a ¼ 4=3 solid line. a ¼ 2 dashed line.
Remark. Eqs. (20) and (21), are special cases of the WKB general equation derived in the pioneering work of Laskin [4].
Accuracy of the WKB estimates, even in small eigenvalues, is remarkable.
We have introduced a Control Volume Radial Basis Function method to solve the eigenvalue problem associated to the
time independent Schrödinger equation. The method has been tested on one dimensional problems with satisfactory results.
In order to test the applicability of the method, we have used for interpolation the simplest radial function. In line with
RBF approximation theory, it is straightforward to test other radial functions, as well as node location strategies.
Currently, we are working on applications of the method to higher dimensions and irregular domains. The latter is pos-
sible thanks to the control volume approach. Hopefully, RBF interpolation will make the method practical in higher
dimensions.
A. Guerrero, M.A. Moreles / Commun Nonlinear Sci Numer Simulat 20 (2015) 604–613 613
It is apparent that the proposed method is practical, and its implementation is straightforward. The approximation theory
supporting the method is left for future investigations.
Acknowledgments
Part of this work was carried out while M.A. Moreles was visiting the Department of Chemical and Petroleum Engineering,
University of Calgary. M.A. Moreles is thankful for their hospitality. The authors are also grateful to the anonymous referees
for their helpful comments.
References
[1] Machado JT, Kiryakova V, Mainardi F. Recent history of fractional calculus. Commun Nonlinear Sci Numer Simul 2011;16:1140–53.
[2] Laskin N. Fractional quantum mechanics and Levy path integrals. Phys Rev A 2000;268:298–305.
[3] Laskin N. Fractals and quantum mechanics. Chaos Interdisciplin J Nonlinear Sci 2000;10.4:780–90.
[4] Laskin N. Fractional Schrödinger equation. Phys Rev E 2002;66:056108.
[5] Wang A, Xu M. Generalized fractional Schrödinger equation with space-time fractional derivatives. J Math Phys 2007;48:043502.
[6] Tarasov VE. Fractional Heisenberg equation. Phys Lett A 2008;372:2984–8.
[7] Matos-Abiague A. Deformation of quantum mechanics in fractional-dimensional space. J Phys A Math Gen 2001;34:11059–68.
[8] Ionescu AD, Pusateri F. Nonlinear fractional Schrödinger equations in one dimension. J Funct Anal 2014;266:139–76.
[9] Amore P, Fernández FM, Hofmann CP, Sáenz RA. Collocation method for fractional quantum mechanics. J Math Phys 2010;51(12):122101.
[10] Zoia A, Rosso A, Kardar M. Fractional Laplacian in bounded domains. Phys Rev E 2007;76(2):061121.
[11] Kwaśnicki M. Eigenvalues of the fractional Laplace operator in the interval. J Funct Anal 2012;262(5):2379–402.
[12] Li B, Chen Z, Huan G. Control volume function approximation methods and their applications to modeling porous media flow. Adv Water Resour
2003;26:435–44.
[13] Musina R, Nazarov AI. On fractional Laplacians. Commun Partial Differ Equ 2013. http://dx.doi.org/10.1080/03605302.2013.864304.
[14] Helgason S. The radon transform. 2nd ed. Boston: Birkhäuser; 1999.
[15] Wendland H. Scattered data approximation. Cambridge: Cambridge University Press; 2005.