Check that your function evaluation is the same with 1 and 2 processors.
Check the ordering of the unknowns and how they are partitioned and that proper ghost point values are set. On Apr 21, 2009, at 2:05 PM, (Rebecca) Xuefei YUAN wrote: > Hi, > > I am testing my hand coded jacobian matrix in multi processors with > the option "-snes_type test -snes_test_display", and I find that > 1) my hand coded jacobian is different from the finite difference > jacobian running with 2 processors. > But > 2) my jacobian is the same as the finite difference jacobian matrix > running with 1 single processor. > Also > 3) my hand coded jacobian with 1 processor is the same as my hand > coded jacobian with 2 processors. > > Where could be wrong about multiprocessors' jacobian matrix? > > Thanks very much! > -- > (Rebecca) Xuefei YUAN > Department of Applied Physics and Applied Mathematics > Columbia University > Tel:917-399-8032 > www.columbia.edu/~xy2102 >
