Commit Graph

287 Commits

Author SHA1 Message Date
Martin Diehl 871241e31b first round of loop removal 2020-03-24 12:43:41 +01:00
Martin Diehl b88ffb8d4f converged(g,i,e) matters only for FPI 2020-03-24 12:30:43 +01:00
Martin Diehl d16af3bfb3 one loop for Euler integrator 2020-03-24 12:19:36 +01:00
Martin Diehl d45f1f1c6f further integration into one loop 2020-03-24 11:37:00 +01:00
Martin Diehl 0e5f0a3068 no need for two loops 2020-03-24 11:27:53 +01:00
Martin Diehl 106cc1de92 complete iteration of each materialpoint step by step 2020-03-24 10:45:38 +01:00
Martin Diehl cde558e736 convergence check on the fly 2020-03-24 10:34:15 +01:00
Martin Diehl 4c485f1af0 state jump in same loop 2020-03-24 10:18:17 +01:00
Martin Diehl 3a6269e802 wrong omp pragmas 2020-03-24 10:06:08 +01:00
Martin Diehl 8d6dcd779c dotstate/residual/state calculation in the same loop 2020-03-24 06:34:42 +01:00
Martin Diehl 830ea61739 better readable 2020-03-24 06:27:43 +01:00
Martin Diehl 9c134e68fc in-line dot state 2020-03-23 08:15:33 +01:00
Martin Diehl e5743c7d27 same logic 2020-03-23 06:53:23 +01:00
Martin Diehl 027818c942 run in one loop 2020-03-23 00:32:41 +01:00
Martin Diehl 0b321bd9d4 bugfix for nonlocal
logic is hard to understand ...
2020-03-23 00:18:57 +01:00
Martin Diehl 86db8f8ca4 avoid flush 2020-03-23 00:15:00 +01:00
Martin Diehl e7d61e49fe can be done in the same loop 2020-03-22 23:16:00 +01:00
Martin Diehl 91e728d065 polishing 2020-03-19 11:30:36 +01:00
Martin Diehl 9ed48f7e5f getting rid of totalNslip in nonlocal 2020-03-16 10:09:58 +01:00
Martin Diehl 66302fa6da rotational part is always of 3x3 tensor 2020-03-15 14:21:11 +01:00
Martin Diehl ae49e6710d documenting for doxygen 2020-03-15 12:54:35 +01:00
Martin Diehl 8cc16da53f atol is part of the state structure, no need for suffix 2020-03-15 09:51:40 +01:00
Martin Diehl eb08f9f0b2 polishing 2020-03-14 19:40:05 +01:00
Martin Diehl 3713810592 simplified allocation for same shape 2020-02-29 13:53:04 +01:00
Martin Diehl bcccf06450 Merge remote-tracking branch 'origin/development' into less-public-variables 2020-02-29 13:18:02 +01:00
Martin Diehl 82dee9db0e matching names 2020-02-25 17:53:15 +01:00
Martin Diehl 48604292e2 Merge remote-tracking branch 'origin/development' into MiscImprovements 2020-02-25 17:33:39 +01:00
Martin Diehl 2c952c3410 these variables can be at leat write protected 2020-02-25 09:50:21 +01:00
Martin Diehl 7d7eff0d94 crystallite should be responsible of crystallite variables 2020-02-25 09:42:07 +01:00
Martin Diehl 2cf89b1fdd bugfix: output dataset was too large in case of Ngrains>1
material_phaseAt is defined per constituent/grain/component => no need
to multiply with homogenization_maxNgrains
2020-02-22 11:28:03 +01:00
Martin Diehl a2e710c89c alinged for better readability 2020-02-21 08:45:11 +01:00
Martin Diehl c9c78aa90d do not store invFp and invFi for all points
requires to explicitly calculate inverse of Fp and Fi for the tangent
calculation. Hence, classical tradeoff between memory consumption and
runtime.
2020-02-21 08:41:08 +01:00
Martin Diehl d108d76a61 was never read
probably it was optimized away by the compiler, still confusing
2020-02-20 15:08:32 +01:00
Martin Diehl a8e2ee0a86 [skip ci] is read only for other modules 2020-02-14 21:31:03 +01:00
Martin Diehl 6adb116712 [skip ci] whitespace adjustments 2020-02-14 06:24:17 +01:00
Martin Diehl 4f7bbb323e not needed 2020-02-13 18:43:20 +01:00
Martin Diehl 64e86666c6 also set initial det(Fp)=0
Marc element lib test failed otherwise for type 117
2020-02-13 17:10:27 +01:00
Martin Diehl 0f70a19266 Fp matters, not Fp^-1
mathematically absolutely equivalent, but numerically not. Sometikes makes a
huge difference in convergence behavior, even though abs(det(Fp)-1) is
in the order of 1e-15
2020-02-13 14:48:32 +01:00
Martin Diehl fc82ec9b1b tiny deviations from det(Fp) = 1 cause convergence problems 2020-02-13 09:18:49 +01:00
Martin Diehl b861ad11c1 tiny differences between orientation conversions result in more
iterations

Seen so far only for plasticityDetectChanges test (confirmed for dislotwin and
phenopowerlaw).
The max difference between the entries of the rotation matrix from
orientation0%asMatrix() and eu2om(Eulers)/eu2om(orientation0%asEulers) is
1e-15. This is the ratio of km/atom radius! Still, the number of
iterations is consistently higher.
Results are the same. I believe this is a strange coincidence where one
particular orientation causes problems. The current version recovers almost the 'good' behavior
of math_EulerToR(Eulers)
2020-02-12 20:49:02 +01:00
Martin Diehl 8770613e9c better readable 2020-02-12 06:26:22 +01:00
Martin Diehl ab475b7c6b need 'error return'
revert from change in 5b72110d
2020-02-12 06:02:37 +01:00
Martin Diehl 08174a119f consistent name 2020-02-11 17:50:07 +01:00
Martin Diehl 5b72110d0a simplified 2020-02-11 17:47:48 +01:00
Martin Diehl e212f91fac print statements prevents reading code
first re-structure and clean, than re-implement where useful
2020-02-11 17:41:30 +01:00
Martin Diehl 6463fcdabd consistent names 2020-02-11 17:36:43 +01:00
Martin Diehl c2c84d698f Merge remote-tracking branch 'origin/development' into MiscImprovements 2020-02-11 17:11:43 +01:00
Martin Diehl e932b386b9 Merge branch 'development' into MiscImprovements 2020-02-07 17:11:01 +01:00
Martin Diehl 5d4d1dcf9a all nonlocal parts are fully explicit
i.e. they are based on converged (partioned0) states
2020-02-07 12:41:01 +01:00
Martin Diehl f854dc27e9 explicit dotState for nonlocal
all flux related quantities are calculated based on the converged
quantities
2020-02-07 12:23:22 +01:00