## Linear equation sys with not positive definit left-hand side

Dee33
Registered Member
Posts
54
Karma
0
OS

### Re: Linear equation sys with not positive definit left-hand

Fri Jul 19, 2013 4:38 pm
Please, If you don"t mind, send us your matrix if we can reproduce the error.
In the meantime, could you backtrace with gdb to see where the assert is raised ?
Thanks
tonyhenz
Registered Member
Posts
7
Karma
0

### Re: Linear equation sys with not positive definit left-hand

Sat Jul 20, 2013 7:53 am
Dee33 wrote:Please, If you don"t mind, send us your matrix if we can reproduce the error.
In the meantime, could you backtrace with gdb to see where the assert is raised ?
Thanks

Hi Dee33,
Thanks for your reply. How shall i send you the matix, the subroutine or the pure data file. As mentioned before, the dimension of the matrix is 481^2 * 481^2, and the nonzero elements is 481^3, and all are complex<double>. I have exported nonzero elements of the sparse Matrtix A, with the format of i, j, A(i,j). The file is 4.4G. The matrix subroutine is a very small part of the large program. I have to think about how to make such matrix suroutine work independently.
Dee33
Registered Member
Posts
54
Karma
0
OS

### Re: Linear equation sys with not positive definit left-hand

Mon Jul 22, 2013 8:03 am
This is rather a large matrix. Are you sure you have enough memory for the factorization ?
tonyhenz
Registered Member
Posts
7
Karma
0

### Re: Linear equation sys with not positive definit left-hand

Mon Jul 22, 2013 9:26 am
Dee33 wrote:This is rather a large matrix. Are you sure you have enough memory for the factorization ?

Hi Dee33,
Thanks for your reply. The SparseLU still does not work, but i have solved this problem by using,
BiCGSTAB<SpMat, IncompleteLUT<complex> > solve,r;
solver.preconditioner().setDroptol(0.001);
VectorXcf x = solver.compute(A).solve(b);
Since at the very begining, i use the default jacobi precoditioner for BiCGSTAB, but it works quite quite slowly. Using IncompleteLUT preconditioner is much better for my problem.
However, for even large SpMat, there is also Std bad alloc issue for BiCGSTAB solver, i am still thinking about it. By the way, what is the maximum dimension or the memory required for the Spmat by using BiCGSTAB and SparseLU. By computing, it seems to me that SparseLU is much faster but requires more memory. Do you have a way to eatimate the memory requirement. And is there any limit of OS ( 64 bit) of declaring a sparse matrix ?

Best Regards,
Tony
Dee33
Registered Member
Posts
54
Karma
0
OS

### Re: Linear equation sys with not positive definit left-hand

Mon Jul 22, 2013 9:46 am
For the record : viewtopic.php?f=74&t=111967
mlimberger
Registered Member
Posts
9
Karma
0
OS

### Re: Linear equation sys with not positive definit left-hand

Tue Jul 23, 2013 11:19 am
Hi alltogether.

My matrix finally was really singular and therefore not solvable by SparseLU.
Now that I modified my algorithms it works fine with SparseLU.

Thanks for all the support and help!

## Who is online

Registered users: andreas_k, Artmessiah, Atalanttore, Baidu [Spider], Bing [Bot], catomic, davidemme, Exabot [Bot], fratti, Google [Bot], google01103, jmacleod, Majestic-12 [Bot], metzman, pepedopolous, Pepek, rblackwell, Steve Guilford, tearingproblems, vpinon, YaCy [Bot]