Short tutorial:

Solving fractional differential equations

by Matlab codes
Roberto Garrappa
Department of Mathematics
University of Bari - Italy
[email protected]

June 30, 2014

1 Introduction

Fractional differential equations (FDEs) are becoming a very popular topic

and several reallife phenomena are described and modeled by means of
scalar or systems of FDEs.
Whilst most of the mathematical packages provide well designed and ro-
bust routines for solving ordinary differential equations, very few codes are
available for the numerical treatment of FDEs.
The aim of this short tutorial is to describe some Matlab codes recently
written for solving FDEs and provide some examples for their use.

2 The Matlab codes fde12.m and flmm.m

The Matlab code fde12.m and flmm.m are devised to numerically solve FDEs
and they are freely available on the file exchange service of Matlab central.
The code fde12.m implements the PredictorCorrector method proposed by
Diethelm and Freed in [1]. This methods is a combination of some product
integration rules, usually known as fractional AdamsBashforthmMoulton
methods and its stability properties were studied in [3]. The code is available
in the file exchange service of Matlab central at

1
 http://www.mathworks.com/matlabcentral/fileexchange/32918

The code flmm.m implements some fractional linear multistep methods (FLMMs)
introduced by Lubich in [5] of the second order. In particular the code im-
plements three different implicit methods: a generalization of the classical
Trapezoidal rule (sometimes referenced as the Tustin method), a generaliza-
tion of the Newton-Gregory formula and the generalization of a Backward
Differentiation Formula (BDF). We refer to [2] for a description of the way in
which the methods are implemented and for a discussion of the various cases
in which one method is preferable to the other. Also this code is available
in the file exchange service of Matlab central at
http://www.mathworks.com/matlabcentral/fileexchange/47081

All the methods are based on discrete convolution quadrature rules. To

keep at the lowest possible level the amount of computation required by the
solution of FDEs, the convolution quadrature rule are evaluated by means
of the FFT algorithm described in [4] allowing to reduce the computational
effort from N 2 to N (log N )2 , where N is the number of timepoints on the
interval of integration [t0 , T ].
We suggest to carefully read the explanations included in each code for a
complete description of the parameters and their use. The explanations can
be also read by means of the Matlab help command as help fde12 or help
flmm2.

2.1 Use of the code fde12.m

To solve a FDE by the fde12.m code the main command is

> [T,Y] = FDE12(ALPHA,FDEFUN,T0,TFINAL,Y0,h)
where ALPHA is the order of the fractional derivative, FDEFUN the vector field
of the FDE, T0 the starting point, TFINAL the ending point of the interval
of integration, Y0 the initial value and h the stepsize.
For a scalar T and a vector Y, FDEFUN(T,Y) must return a column vector
corresponding to the vector field f (t, y). It is possible to specify some pa-
rameters for FDEFUN by introducing the optional argument PARAM; in this
case the vector field is evaluated as FDEFUN(T,Y,PARAM) and fde12.m is
called as
> [T,Y] = FDE12(ALPHA,FDEFUN,T0,TFINAL,Y0,h,PARAM)
We refer to help fde12 for the meaning and the use of the other parameters.

2
2.2 Use of the code flmm.m
The code flmm.m is used in a very similar way to fde12.m but there are two
main differences:
since flmm.m implements three different methods, the parameter METHOD
is used to specify the selected method;

being the methods implemented by flmm.m of implicit type, it is nec-

essary to define the Jacobian of the vector field in order to solve at
each timestep the (usually nonlinear) system of algebraic equations.

The basic syntax for solving and FDE by the code flmm.m is therefore
> [T,Y] = FLMM2(ALPHA,FDEFUN,JFDEFUN,T0,TFINAL,Y0,h,PARAM,METHOD)
where JFDEFUN defines the Jacobian of the vector field FDEFUN and METHOD
the method selected for solving the FDE. The options for METHOD are: 1 for
the Trapezoidal method, 2 for the Newton-Gregory formula and 3 for the
BDF-2 (see [5, 2]). The parameter METHOD and is optional; when not defined
by default it is assumed METHOD=3.
We refer to help flmm2 for the meaning and the use of the other parameters.

3 Examples
We provide here an example of application of the codes presented in this
tutorial. In particular we consider, as test problem, the fractional version of
the Brusselator model problem

2

t0 Dt y1 (t) = a ( + 1)y1 (t) + (y1 (t)) y2 (t)
2
t0 Dt y2 (t) = y1 (t) (y1 (t)) y1 (t)

, 0<<1 (1)


y1 (t0 ) = y1,0
y2 (t0 ) = y2,0

We assume that we need to integrate this FDE for the order = 0.8 in the
interval [0, 100] with a stepsize h = 26 . Moreover, we assume that the
main parameters has the values a = 1 and = 4.
The vector field of the test problem is
! !
f1 (y(t)) a ( + 1)y1 (t) + (y1 (t))2 y2 (t)
f (t, y(t)) f (y(t)) = =
f2 (y(t)) y1 (t) (y1 (t))2 y1 (t)

3
and it is easy to evaluate that the Jacobian


y f1 (y(t)) y2
f1 (y(t))
Jf (t, y(t)) = 1

f2 (y(t)) f2 (y(t))
y1 y2
of this function is given by
!
( + 1) + 2y1 (t)y2 (t) (y1 (t))2
Jf (t, y(t)) Jf (y(t)) =
2y1 (t) (y1 (t))2

There are two main ways to describe the above vector field and the corre-
sponding Jacobian. With recent versions of Matlab it is possible to define

a = 1 ; mu = 4 ;
fdefun = @(t,y) [ a-(mu+1)*y(1)+y(1)^2*y(2) ; mu*y(1)-y(1)^2*y(2) ] ;
Jfdefun = @(t,y) [ -(mu+1)+2*y(1)*y(2) , y(1)^2 ; mu-2*y(1)*y(2) , -y(1)^2 ] ;

Note that the Jacobian Jfdefun is used only by the flmm2.m code; in the
case of fde12.m it is sufficient to define only fdefun.
For the other parameters (order, interval of integration, initial value and
stepsize) we use the assignments

alpha = 0.8 ;
t0 = 0 ; tfinal = 100 ; y0 = [ 0.2 ; 0.03] ;
h = 2^(-6) ;

It is hence possible to solve the test problem by means of the following calls

[t, y_flmm2] = flmm2(alpha,fdefun,Jfdefun,t0,tfinal,y0,h) ;

[t, y_fde12] = fde12(alpha,fdefun,t0,tfinal,y0,h) ;

To plot the solution the following few Matlab commands can be used

figure(1)
plot(t,y_flmm2(1,:),t,y_flmm2(2,:)) ;
xlabel(t) ; ylabel(y(t)) ;
legend(y_1(t),y_2(t)) ;
title(FDE solved by the FLMM2.m code) ;

4
 FDE solved by the FLMM2.m code FDE solved by the FLMM2.m code
7 7
y1(t)
y2(t) 6
6

5 5

4 4
y(t)

y2
3 3

2 2

1 1

0 0
0 20 40 60 80 100 0 1 2 3 4 5 6
t y1

Figure 1: Solution of the Brusselator test problem in the plane (t, y) and in
the phase plane.

and similarly for plotting the results from the fde12.m code. The results
are shown in Figure 1.
An alternative approach (usually necessary in old versions of Matlab) to
define the vector field is by means of external functions of this kind

function y = f_Bruss(t,x,param)
a = param(1) ; mu = param(2) ;
y(1,1) = a -(mu+1)*x(1)+x(1)^2*x(2) ;
y(2,1) = mu*x(1)-x(1)^2*x(2) ;

which must be saved in a file named as f Bruss.m. similarly, for the corre-
sponding Jacobian matrix we define the external function

function J = Jf_Bruss(t,x,param)
a = param(1) ; mu = param(2) ;
J(1,1) = -(mu+1)+2*x(1)*x(2) ;
J(1,2) = x(1)^2 ;
J(2,1) = mu-2*x(1)*x(2) ;
J(2,2) = -x(1)^2 ;

which will be saved in a file named as Jf Bruss.m.

The way in which the FDE is solved by the codes flmm2.m and fde12.m
is slighly modified since the vector field and its Jacobian are given now as
the strings containing their names and the optional variable PARAM must be
used in order to communicate the values of the parameters a and

5
a = 1 ; mu = 4 ;
param = [a , mu] ;
[t, y_flmm2] = flmm2(alpha,f_Bruss,Jf_Bruss,t0,tfinal,y0,h,param) ;
[t, y_fde12] = fde12(alpha,f_Bruss,t0,tfinal,y0,h,param) ;

References
[1] Kai Diethelm and Alan D. Freed. The FracPECE subroutine for the nu-
merical solution of differential equations of fractional order. In S.Heinzel
and T.Plesser, editors, Forschung und wissenschaftliches Rechnen 1998,
pages 5771. 1999.

[2] R. Garrappa. Trapezoidal methods for fractional differential

equations: Theoretical and computational aspects. Mathe-
matics and Computers in Simulation, 2013. In press. DOI
http://dx.doi.org/10.1016/j.matcom.2013.09.012.

[3] Roberto Garrappa. On linear stability of predictorcorrector algorithms

for fractional differential equations. Int. J. Comput. Math., 87(10):2281
2290, 2010.

[4] E. Hairer, Ch. Lubich, and M. Schlichte. Fast numerical solution of

weakly singular Volterra integral equations. J. Comput. Appl. Math.,
23(1):8798, 1988.

[5] Christian Lubich. Discretized fractional calculus. SIAM J. Math. Anal.,

17(3):704719, 1986.