[petsc-users] -snes_type test vs -snes_mf

[(Rebecca) Xuefei YUAN]

Dear Barry,

Thanks very much!

I am pretty sure that I was wrong with Jacobian now because I took  
Jacobian in multiprocessors as the same structure in single processor,  
which had made my indexing wrong in the MatSetValues().

I am working on that to fix it.

Thanks a lot!

Rebecca

Quoting Barry Smith <bsmith at mcs.anl.gov>:

>
>    I would focus on the Jacobian.
>
>
> On Mar 18, 2010, at 2:00 PM, (Rebecca) Xuefei YUAN wrote:
>
>> Dear all,
>>
>> I ran a code for np=1 and np=2.
>>
>> If hand coded Jacobian is provided, np=1 and np=2 gives me   
>> different convergence and solutions, so I want to know what is wrong.
>>
>> If option -snes_mf is used, np=1 and np=2 gives me the same   
>> convergence and close solutions. I thought there might be a bug in   
>> FormJacobian().
>>
>> If option -snes_type test is used, np=1 and np=2 gives different   
>> ratio and differences for finite difference Jacobian(fd) and hand   
>> coded Jacobian(hc): J_np1_hc = J_np2_hc, but J_np1_fd != J_np2_fd.   
>> so I thought there might be a bug in FormFunction().
>>
>> I am not sure which one should be my focus?
>>
>> Any suggestions?
>>
>> Thanks a lot!
>>
>> Rebecca
>>
>>
>>
>>
>> -- 
>> (Rebecca) Xuefei YUAN
>> Department of Applied Physics and Applied Mathematics
>> Columbia University
>> Tel:917-399-8032
>> www.columbia.edu/~xy2102
>>



-- 
(Rebecca) Xuefei YUAN
Department of Applied Physics and Applied Mathematics
Columbia University
Tel:917-399-8032
www.columbia.edu/~xy2102