[SciPy-dev] Test case for integrate/ode.py: banded systems

[Jesper Friis]

I hope this is the correct place to post messages like this. Last Friday 
I submitted a small patch for integrate/ode.py making it working for 
banded systems. I have now successfully applied the solver to the banded 
example problem included in the cvode package. The attached script might 
both work as a test and as an example on how to solve banded systems.

I think, especially the following comment considering the Jacobian might 
be of general interest:
     # The Jacobian. 

     # For banded systems this function returns a matrix pd of
     # size ml+mu*2+1 by neq containing the partial derivatives 

     # df[k]/du[u]. Here f is the right hand side function in 

     # udot=f(t,u), ml and mu are the lower and upper half bandwidths 

     # and neq the number of equations.  The derivatives df[k]/du[l] 

     # are loaded into pd[mu+k-l,k], i.e. the diagonals are loaded into 

     # the rows of pd from top down (fortran indexing). 

     # 

     # Confusingly, the number of rows VODE expect is not ml+mu+1, but 

     # given by a parameter nrowpd, which unfortunately is left out in 

     # the python interface. However, it seems that VODE expect that 

     # nrowpd = ml+2*mu+1. E.g. for our system with ml=mu=5 VODE expect 

     # 16 rows. Fortunately the f2py interface prints out an error if 

     # the number of rows is wrong, so as long ml and mu are known in 

     # beforehand one can always determine nrowpd by trial and error... 




Regards
/Jesper
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test_ode_banded.py
Type: text/x-python
Size: 5695 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/scipy-dev/attachments/20060306/250169a4/attachment.py>