optimizer

Optimizers for refining TS-constrained conformers.

Constructor parameters:

  • conf_id_ref: int = -1 : reference conformer ID on the TS molecule
  • force_constant: float = 1e6 : force constant of distance constraints to TS reference points

MMFFOptimizer

Optimizes each conformer with MMFF94:

  1. Align to the TS reference conformer (on align_indices)
  2. Create RDKit force-field extra points at the TS coordinates for each align_indices atom and add distance constraints with force_constant
  3. Minimize and set the conformer energy property from ff.CalcEnergy() in kcal/mol
  4. Re-align to the TS reference

UFFOptimizer

Same flow as MMFF_optimizer using the RDKit implementation of UFF.

Note

  • MMFF: Good general-purpose performance, if MMFF parameters are available.
  • UFF: Use if no MMFF parameters available or as fallback.

ASEOptimizer (optional dependency)

ASEOptimizer uses the Atomic Simulation Environment (ASE):

  1. Align to the TS reference conformer on align_indices
  2. Convert each RDKit conformer to ASE Atoms
  3. Apply FixAtoms(indices=align_indices) to freeze these atoms
  4. Optimize with an ASE optimizer (BFGS by default) and the provided ASE calculator (calculator=..., instance or callable)
  5. Write positions back to RDKit and set conformer energy in kcal/mol (eV * 23.06054783061903)
  6. Re-align to the TS reference

Charge and multiplicity are the charge and multiplicity arguments, else the values stored on the molecule by generate_conformers, else the sum of the formal charges and the lowest multiplicity for the number of electrons. They reach the calculator in two forms: in atoms.info (charge, spin = multiplicity, as read by UMA/fairchem) and as initial charges and magnetic moments (whose sums tblite and xtb-python read). Calculators that take charge and multiplicity as their own arguments (e.g. ASE's ORCA(charge=..., mult=...), PySCF) need them passed there. A conformer whose calculation fails (e.g. SCF not converged) is left without an energy and is dropped when pruning; if all conformers fail, an error is raised.

Note

align_indices follow standard Python indexing (0-based), consistent with RDKit and ASE atom indices.

Install ASE support with:

pip install racerts[ase]

Example with a calculator instance:

from ase.calculators.lj import LennardJones
from racerts.optimizer import ASEOptimizer

optimizer = ASEOptimizer(
    calculator=LennardJones(),
    fmax=0.05,
    max_steps=100,
)

Example with a calculator callable:

from ase.calculators.lj import LennardJones
from racerts.optimizer import ASEOptimizer

optimizer = ASEOptimizer(
    calculator=LennardJones,
    num_workers=4,
)