Dear Sella developers and maintainers:
I'm currently performing geometry relaxations for a periodic organic molecule system. The relaxation runs of desire should allow for changes in not only atomic positions but also cell dimensions and shape (i.e., lattice vector optimization), so as to release any significant internal stresses and strains introduced in the initial structure.
Through testing, I noticed that when using the Sella alone for relaxation, the dimensions of the simulation cell do not change, and it seems that sella currently doesn't support the optimization of lattice vectors internally. Alternatively, I tried to combine Sella with external cell filters like the 'FrechetCellFilter' (an ase cell filter) to include cell vectors as optimization variables, but it appears incompatible.
My questions are:
Are there any settings in sella that I missed that actually permit cell shape and vector adjustments?
If not, is sella compatible with any external cell filters that allow for lattice vector changes?
Given that I get no better solutions now, I plan to adopt a two-stage relaxation strategy to overcome the limitation:
Stage 1: Use L-BFGS (an ASE optimizer or another compatible optimizer) in combination with 'FrechetCellFilter' to perform an initial relaxation, releasing the bulk of internal stresses and strains and optimizing both atomic positions, cell shape, and cell vectors.
Stage 2: Follow up with the 'Sella' optimizer to perform a more detailed refinement of the atomic positions on the fixed cell determined in Stage 1 (atomic positions relaxation only).
We would highly appreciate any input from the maintainers regarding this limitation and our proposed two-stage strategy. We are also interested to know if there are any plans in the future to introduce support for lattice vector optimization into the Sella optimizer.
Best regards,
Jo
Dear Sella developers and maintainers:
I'm currently performing geometry relaxations for a periodic organic molecule system. The relaxation runs of desire should allow for changes in not only atomic positions but also cell dimensions and shape (i.e., lattice vector optimization), so as to release any significant internal stresses and strains introduced in the initial structure.
Through testing, I noticed that when using the Sella alone for relaxation, the dimensions of the simulation cell do not change, and it seems that sella currently doesn't support the optimization of lattice vectors internally. Alternatively, I tried to combine Sella with external cell filters like the 'FrechetCellFilter' (an ase cell filter) to include cell vectors as optimization variables, but it appears incompatible.
My questions are:
Are there any settings in sella that I missed that actually permit cell shape and vector adjustments?
If not, is sella compatible with any external cell filters that allow for lattice vector changes?
Given that I get no better solutions now, I plan to adopt a two-stage relaxation strategy to overcome the limitation:
Stage 1: Use L-BFGS (an ASE optimizer or another compatible optimizer) in combination with 'FrechetCellFilter' to perform an initial relaxation, releasing the bulk of internal stresses and strains and optimizing both atomic positions, cell shape, and cell vectors.
Stage 2: Follow up with the 'Sella' optimizer to perform a more detailed refinement of the atomic positions on the fixed cell determined in Stage 1 (atomic positions relaxation only).
We would highly appreciate any input from the maintainers regarding this limitation and our proposed two-stage strategy. We are also interested to know if there are any plans in the future to introduce support for lattice vector optimization into the Sella optimizer.
Best regards,
Jo