 
  
  
   
The grid spacing and time step were chosen to reliably create spiral waves
with appropriately smooth profiles and propagation speeds. A criterion of
numerical accuracy was that small changes in the grid or time step should
not give rise to large differences in the solution. With numerical
parameters chosen as in Sect.2.4 neither reducing the steps   or
 
or   by
  by   nor reducing the time step by
  nor reducing the time step by   changed
the wave period by more than
  changed
the wave period by more than   . Nevertheless, even in the absence of any
forcing, a radial drift of spiral waves away from the centre was observed.
The radial drift velocity
 . Nevertheless, even in the absence of any
forcing, a radial drift of spiral waves away from the centre was observed.
The radial drift velocity   su/tu was less than 0.2% of
the plane wave velocity. Numerical experiments evidenced that the mentioned
radial drift velocity decreased as the
  su/tu was less than 0.2% of
the plane wave velocity. Numerical experiments evidenced that the mentioned
radial drift velocity decreased as the   became smaller,
  became smaller,   su/tu for
  su/tu for   ;
 ;   su/tu for
  su/tu for 
  ;
 ;   su/tu for
  su/tu for   ; and
 ; and   su/tu for
  su/tu for   . In these test computations the time
step
 . In these test computations the time
step   was used to satisfy the stability criterion for the
case of the smallest
  was used to satisfy the stability criterion for the
case of the smallest   . The very slow radial drift should
probably be attributed to the grid anisotropy produced by the summation of
local errors of approximation in integrating the discrete analog of
equations (1)-(5) in polar coordinates. In numerical experiments
described below, this numerical artefact was always much smaller than the
resonant drift.
 . The very slow radial drift should
probably be attributed to the grid anisotropy produced by the summation of
local errors of approximation in integrating the discrete analog of
equations (1)-(5) in polar coordinates. In numerical experiments
described below, this numerical artefact was always much smaller than the
resonant drift.