pardiso boundary condition

[array123]

hi,

i am using PARDISO to solve Ax=b type unsymmetric real sparse matrix. For the give problem if i set boundary conditions as pressure gradient = 0 then PARDISO gives result same as other solvers, but if i set BC's as p=0 then PARDISO gives different(wrong) result as compared to other solvers.

[Alexander_K_Intel2]

Hi,

What differential equation you are solve? Is "p = 0" Dirichle or Neumann boundary condition?

With best regards,

Alexander Kalinkin

[array123]

hi,

it is Dirichlet

[Alexander_K_Intel2]

Hi,

What about equation? Could determinant ofresult stiffness matrix is equal zero? If not, could you please attach testcase here?

With best regards,

Alexander Kalinkin

[array123]

hi,

no the determinant is not zero. The problem looks like (actual problem is large this is an example)

AP11	AE11	0	0	AN11	0	0	0	0	0	0	0		X1		B1
AW21	AP21	AE21	0	0	AN21	0	0	0	0	0	0		X2		B2
0	AW31	AP31	AE31	0	0	AN31	0	0	0	0	0		X3		B3
0	0	AW41	AP41	0	0	0	AN41	0	0	0	0		X4		B4
AS12	0	0	0	AP12	AE12	0	0	AN12	0	0	0		X5		B5
0	AS22	0	0	AW22	AP22	AE22	0	0	AN22	0	0	X	X6	=	B6
0	0	AS32	0	0	AW32	AP32	AE32	0	0	AN32	0		X7		B7
0	0	0	AS42	0	0	AW42	AP42	0	0	0	AN42		X8		B8
0	0	0	0	AS13	0	0	0	AP13	AE13	0	0		X9		B9
0	0	0	0	0	AS23	0	0	AW23	AP23	AE23	0		X10		B10
0	0	0	0	0	0	AS33	0	0	AW33	AP33	AE33		X11		B11
0	0	0	0	0	0	0	AS42	0	0	AW43	AP43		X12		B12

THIS IS COEFFICIENT MATRIX OF A 4x3 MATRIX.

in order to set right boundary as x=0 (Dirichlet) what i need to do is to set AE41,AE42,AE43 as zero. But when i do this and run PARDISO it gives wrong results (I solved it with gauss siedel and it gives correct result).

Matrix looks like

11	21	31	41
12	22	32	42
13	23	33	43

[Alexander_K_Intel2]

Hi,

Unfortunately, I couldn't find AE41, AE42 and AE43 in table of symbols you wrote. But in my opinion if one want to set Dirichlet boundary condition in some point he need to set zero in correspondent column/string of matrix and change correspondent diagonal element. Nevertheless, could you provide testcase that I could compile and execute for deeper investigation of problem?

With best regards,

Alexander Kalinkin