Slender beams go wrong with Pastix

[victorkemp]

This works fine with SOLVER=SPOOLES but displacement is all wrong with the default PASTIX. All 3 beams should be bending the same as each other as cantilevers. Refining the mesh so each element is stockier leads to a more realistic-looking but still completely wrong solution.

*NODE
1,0,1,0
2,3,0,0
3,3,-1,0
4,3,1,0
5,0,0,0
6,0,-1,0
7,1.5,-1,0
8,1.5,0,0
9,1.5,1,0
*ELEMENT,TYPE=B32R,ELSET=EAll
1,6,7,3
2,5,8,2
3,1,9,4
*MATERIAL,NAME=Material
*ELASTIC,TYPE=ISOTROPIC
200000000000,0
*BEAM SECTION,ELSET=EAll,MATERIAL=Material,SECTION=RECT
0.005,0.005
0,1,0
*BOUNDARY
1,1,6,0
5,1,6,0
6,1,6,0
*STEP
*STATIC
*CLOAD
2,3,1.23
3,3,1.23
4,3,1.23
*NODE FILE
U
*END STEP

[Dhondtguido]

Hi Victor, the way PastiX is adressed in CalculiX it solves directly in single precision and iterates in double precision. Solving directly in double precision should solve the issue. Guido victorkemp ***@***.***> schrieb am Do., 3. Aug. 2023, 04:03:

…

This works fine with SOLVER=SPOOLES but displacement is all wrong with the default PASTIX. All 3 beams should be bending the same as each other as cantilevers. Refining the mesh so each element is stockier leads to a more realistic-looking but still completely wrong solution. [image: b32r_pastix] <https://user-images.githubusercontent.com/118495694/257980505-30139b82-7097-4275-bf53-615adceef7e0.png> *NODE 1,0,1,0 2,3,0,0 3,3,-1,0 4,3,1,0 5,0,0,0 6,0,-1,0 7,1.5,-1,0 8,1.5,0,0 9,1.5,1,0 *ELEMENT,TYPE=B32R,ELSET=EAll 1,6,7,3 2,5,8,2 3,1,9,4 *MATERIAL,NAME=Material *ELASTIC,TYPE=ISOTROPIC 200000000000,0 *BEAM SECTION,ELSET=EAll,MATERIAL=Material,SECTION=RECT 0.005,0.005 0,1,0 *BOUNDARY 1,1,6,0 5,1,6,0 6,1,6,0 *STEP *STATIC *CLOAD 2,3,1.23 3,3,1.23 4,3,1.23 *NODE FILE U *END STEP — Reply to this email directly, view it on GitHub <#51>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQMNEXAJK54Y7KR7SVHJ4WTXTMBGVANCNFSM6AAAAAA3CCGPOY> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[victorkemp]

I wonder if this also shows a problem with identifying convergence failure from Pastix? I'd expect the double precision iterations should fail to converge rather than ending up wrong. The output shows error being pretty big and barely reducing before it finishes "successfully". Normal good solves have tiny error like 1e-13.

 - iteration 1 :         error                                  3.4318
 - iteration 2 :         error                                  3.4317
...
- iteration 49 :         error                                  3.3754
- iteration 50 :         error                                  3.3721

[victorkemp]

Confirmed that the problem goes away by setting the environment variable PASTIX_MIXED_PRECISION = 0 which seems to force double-precision mode.