[SciPy-User] Integrating large ODE problems with sundials.
Peter John Garrone
pgarrone at optusnet.com.au
Sat Jan 15 16:40:09 EST 2011
Hi,
Some time ago I wrote that I was having problems with ODE algorithm solving, where most of the time appeared to be lost in the algorithm.
In fact the problem appears to be within the PySUNDIALS implementation that I downloaded from sourceforge, pysundials-2.3.0-rc2. When I implemented my own simple interface, in fact only about five percent of the cpu time is spent within the algorithm. My approximate times for a 100000 state problem were:
dydt/right-hand side equation: 20 percent
Jacobian and sparse matrix creation: 35 percent
Preconditioner sparse solve: 40 percent
Remainder, including algorithm: 5 percent
With the pysundials implementation, the remainder portion would have been about 95 percent, with the other categories adding up to about 4 percent.
So I do conclude there is something poor about the pysundials implementation, possibly the N_Vector implementation, for big problems, using standard python.
Cheers,
More information about the SciPy-User
mailing list