OPTIMIZE - Open-Quantum-Platform/openqp GitHub Wiki

[OPTIMIZE] Defaults

Keyword	Default	Description
lib	scipy	Optimization engine.
optimizer	bfgs	Optimization algorithm.
step_size	0.1	Step size for optimization.
step_tol	1e-2	Tolerance for step size.
maxit	30	Maximum number of optimization iterations.
mep_maxit	10	Maximum iterations for MEP optimization.
rmsd_grad	1e-4	RMSD criteria for gradients.
rmsd_step	1e-3	RMSD criteria for step size.
max_grad	3e-4	Maximum gradient criteria.
max_step	2e-3	Maximum step size criteria.
istate	0	First state for optimization.
jstate	1	Second state for optimization.
energy_shift	1e-6	Energy shift criteria.
energy_gap	1e-5	Energy gap criteria.
meci_search	penalty	Method for MECI searching.
pen_sigma	1.0	Sigma value for penalty method.
pen_alpha	0.0	Alpha value for penalty method.
pen_incre	1.0	Increment for penalty method.
gap_weight	1.0	Weight for energy gap in optimization.
init_scf	False	Initial SCF option.

[OPTIMIZE] Details

The **optimize** section handles geometry optimization.

• lib: Choose the optimization library.
  • Options:
    • scipy: Use the scipy.optimize library. (Default)
    • dlfind: Use the DL-FIND library.
• optimizer: Choose the scipy optimizer.
  • Options:
    • bfgs: Use the BFGS method. (Default)
    • cg: Use Conjugate Gradient.
    • l-bfgs-b: Use L-BFGS-b method.
    • newton-cg: Use Newton Conjugate Gradient.
• step_size: Set the radius of the constraining hypersphere from the starting structure.
  • Default: 0.1 (the largest distance between the mass-weighted coordinates)
• step_tol: Set the threshold for the radius on a hypersphere from the starting structure.
  • Default: 1e-2 (the smallest distance between the mass-weighted coordinates)
• maxit: Set the maximum number of geometry optimization iterations.
  • Default: 30
• mep_maxit: Set the maximum number of MEP steps.
  • Default: 10
• rmsd_grad: Set the convergence threshold for RMSD of gradients.
  • Default: 1e-4
• rmsd_step: Set the convergence threshold for RMSD of structure changes.
  • Default: 1e-3
• max_grad: Set the convergence threshold for the maximum gradient.
  • Default: 3e-4
• max_step: Set the convergence threshold for the maximum structure changes.
  • Default: 2e-3
• istate: Choose the state for single-state optimization. This option is for either runtype=optimize, mep or meci. The latter requires setting the jstate too. In the case of HF/DFT calculation, istate=0. It should be emphasized that TDDFT can only calculate excited states. Thus, istate=1 for TDDFT means the first excited state (S1). However, both SF-TDDFT and MRSF-TDDFT can also calculate ground state. Therefore, istate=1 for these two theories correspond to the ground state (S0).
  • Note: Currently, time-dependent calculations ([input] method=tdhf) do not compute gradients for the reference state (grad=0).
  • Default: 1
• jstate: Choose the second state for conical intersection optimization.
  • Default: 2
• energy_shift: Set the convergence threshold for electronic energy changes.
  • Default: 1e-6
• energy_gap: Set the convergence threshold for energy gap changes.
  • Default: 1e-5
• meci-search: Choose the algorithm for conical intersection optimization.
  • Options:
    • penalty: Use the modified penalty method. (Default)
    • ubp: Use the Update Branching Plane method.

hybrid: Use the penalty function, then switch to ubp after the energy gap is below the threshold.

• pen_sigma: Set the sigma in the penalty function.
  • Default: 1.0
• pen_alpha: Set the alpha in the penalty function.
  • Default: 0.0
• pen_incre: Set the incremental factor in the penalty function.
  • Default: 1.0
• init_scf: Perform initial SCF iterations during geometry optimization.
  • Options:
    • True: Perform initial SCF iterations at every optimization step.
    • False: Do not perform initial SCF iterations after the first optimization step.

Back to Input File System with PyOQP