Why the sparse pattern of a sparse matrix changed

I'm building a sparse matrix through the mex function. Both the Ir and Jc properties are appointed in the mex function A. Later, this sparse matrix is returned to matlab, and is further transferred to another mex function B together with other parameters to assemble a large matrix.

For one test case, part of the Jc property of the created sparse matrix in the mex function A is as follows:

368

392

416

440

464

488

512

But when this sparse matrix is further transferred to mex function B, this part is altered and is as follows:

368

392

416

440

463

487

511

Actually, after this sparse matrix is established, I did no alternation about it. I want to know what may case this.

[Matt J]

I did no alternation about it

But you did (no one else could have - I certainly didn't). The question is where.

[Rylan]

What if some data in the sparse matrix is zero, whther or not the structure of the sparse matrix would be altered when the matrix returned to matlab

[Bruno Luong]

You MUST NOT store 0 elements internally. Use spok I indicate bellow to check if your mex-build result is valid.

[Rylan]

(9,22) -0.0000

(10,22) 0

(11,22) 0.0000

(12,22) -0.0000

(13,22) -0.2500

(14,22) 0.5000

It seems problem was caused by (10,22), and use of spok indicate that "0 jumbled row indices, 3 explicit zeros". To remove this problem, may be a vary small number, for example, 1e-16, should be added to all the data member.

Thanks for your suggestion!

Answer 1

[Bruno Luong]

The mex probably creates a non-valid sparse matrix, use spok to check validity

[James Tursa]

Note that Tim Davis's spok function has been removed from the FEX. You can use my replacement function sarek instead:

https://www.mathworks.com/matlabcentral/fileexchange/158161-sarek-sparse-analyzer-real-et-komplex?s_tid=srchtitle