hansjoerg seybold, 2009-07-22 04:05

~~DISCUSSION~~ hello, I was playing a bit around with the couette.cpp example using adaptive timestepping in a domain 15x2x15 with the following parameters

const Real dtmin = 0.001;
const Real dtmax = 0.02;
const Real CFLmin = 0.40;
const Real CFLmax = 0.55;

I tried several resolutions for the spectral calculations and when i increase the number of modes for Ny to values >=50 (49 works 50 not any more) i get a nan in the further calculations. The bug seems to be independant from the X-Z resolution.

I also tried fixed timestepping with very small dt (CLF<0.3) but the error appeared nevertheless

As integrators i tried SBDF3+SMRK2 and CNAB2 running with plain channelflow-1.4.1, g++ (GCC) 4.2.3 20071030 (prerelease) (SUSE Linux)

After the adaption step / first printout step for fixed dt i get the following output

TimeStep::adjust(CFL) {

 n : 95 -> 50
dt : 0.0105263 -> 0.02
CFL : nan -> nan
}
adjusting timestep
         t == 1
       CFL == nan
 L2Norm(u) == nan
divNorm(u) == nan
      dPdx == 0
     Ubulk == nan

I compiled channelflow in optimized mode as i do not have the debug fftw3 libraries on my system. Any hints where i could look for the bug?

hj