You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
lynnmunday opened this issue
Oct 23, 2024
· 0 comments
· May be fixed by #28923
Assignees
Labels
P: normalA defect affecting operation with a low possibility of significantly affects.T: defectAn anomaly, which is anything that deviates from expectations.
@lindsayad noticed that OptimizeSolve gives TAO PETSC ownership of a solution vector pointer that is then also being modified by OptimizeSolve. So both the TAO optimization algorithms and OptimizeSolve are changing values in this vectors. This behavior is wrong and causes the TAO optimization algorithms to be incorrectly applied. This is also not allowed in a PETSC 3.22 update that uses vecloc on solution vectors.
Steps to Reproduce
This test now converges in one fewer iteration after fixing this:
modules/optimization/executioners/constrained/shape_optimization.equality_constraint
You will also note that the gradient is reduced at a much faster rate.
Impact
This was causing linesearch to fail and resulted in more tests using a trust-region solver or a unit line-search. Fixing htis will enable us to use line-search algorithms again. Modifying the tao parameter vector probably has other unintended consequences I am unaware of.
[Optional] Diagnostics
No response
The text was updated successfully, but these errors were encountered:
lynnmunday
added
P: normal
A defect affecting operation with a low possibility of significantly affects.
T: defect
An anomaly, which is anything that deviates from expectations.
labels
Oct 23, 2024
P: normalA defect affecting operation with a low possibility of significantly affects.T: defectAn anomaly, which is anything that deviates from expectations.
Bug Description
@lindsayad noticed that OptimizeSolve gives TAO PETSC ownership of a solution vector pointer that is then also being modified by OptimizeSolve. So both the TAO optimization algorithms and OptimizeSolve are changing values in this vectors. This behavior is wrong and causes the TAO optimization algorithms to be incorrectly applied. This is also not allowed in a PETSC 3.22 update that uses vecloc on solution vectors.
Steps to Reproduce
This test now converges in one fewer iteration after fixing this:
modules/optimization/executioners/constrained/shape_optimization.equality_constraint
You will also note that the gradient is reduced at a much faster rate.
Impact
This was causing linesearch to fail and resulted in more tests using a trust-region solver or a unit line-search. Fixing htis will enable us to use line-search algorithms again. Modifying the tao parameter vector probably has other unintended consequences I am unaware of.
[Optional] Diagnostics
No response
The text was updated successfully, but these errors were encountered: