![]() |
(git:27c966b)
|
Dual integer-electron-endpoint charge mixing for constant-potential MD, built on top of CP2K's stock MIXED force_eval (METHOD MIXED / MIXING_TYPE LINEAR_COMBINATION). Energy, forces and virial are already interpolated by stock force_env_methods.F's mixed_energy_forces using the LAMBDA value read from the input structure at every call; this module (a) pushes the dynamically tracked lambda into that input structure each step, (b) mixes Fermi level / workfunction, which stock MIXED does not touch, and (c) rebuilds a sub_force_env at a new integer CHARGE whenever q(t) drifts outside the currently-built bracket ("boundary crossing"). More...
Functions/Subroutines | |
| subroutine, public | qs_electrons_charge_mixing_lambda (electrons, lambda, n_lo, n_hi) |
| Given the current electronic coordinate q and the integer span S, compute the bracketing integers N_lo/N_hi and the mixing weight lambda = (q - N_lo)/S. | |
| subroutine, public | push_lambda_to_mixed_env (force_env, lambda) |
| Push a dynamically computed lambda into the MIXED force_eval's input structure. force_env_methods.F::mixed_energy_forces reads MIXEDLINEARLAMBDA fresh from the input structure every call, so this alone is sufficient to make energy/force/virial mixing track our own extended-variable dynamics with zero changes to stock CP2K. | |
| subroutine, public | compute_mixed_workfunction (force_env, axis, wf_min, wf_max, q, span, fermi_level_ev, workfunction_ev, lambda_in) |
| Fermi level / workfunction mixing across the two integer-charge endpoints. Stock MIXED only combines energy/force/virial/dipole (mixed_environment_utils.F::mixed_map_forces), so these two quantities – needed every step by our own O-B-A-B-O potentiostat to decide how to move q(t) – are interpolated here, mirroring FFChargeMixTwoSockets._mix_results's fermi/workfunction lines. | |
| subroutine, public | charge_mixing_maybe_rebuild (force_env, electrons) |
| Check whether q(t) has drifted outside the currently-built [N_lo,N_hi] bracket and, if so, rebuild whichever endpoint(s) are now stale at the new integer electron count. Must be called every step, right after qs_electrons_drift and before push_lambda_to_mixed_env, so LAMBDA is always pushed against a bracket that actually matches the physically-built sub_force_env pair. | |
| subroutine, public | charge_mixing_sync_endpoint_state (force_env, electrons) |
| Record which integer electron counts the two FORCE_EVAL blocks are ACTUALLY built for, by interrogating them, so that the very first charge_mixing_maybe_rebuild can tell whether the input agrees with q. | |
Dual integer-electron-endpoint charge mixing for constant-potential MD, built on top of CP2K's stock MIXED force_eval (METHOD MIXED / MIXING_TYPE LINEAR_COMBINATION). Energy, forces and virial are already interpolated by stock force_env_methods.F's mixed_energy_forces using the LAMBDA value read from the input structure at every call; this module (a) pushes the dynamically tracked lambda into that input structure each step, (b) mixes Fermi level / workfunction, which stock MIXED does not touch, and (c) rebuilds a sub_force_env at a new integer CHARGE whenever q(t) drifts outside the currently-built bracket ("boundary crossing").
| subroutine, public qs_electrons_charge_mixing::qs_electrons_charge_mixing_lambda | ( | type(qs_electrons_type), pointer | electrons, |
| real(kind=dp), intent(out) | lambda, | ||
| integer, intent(out), optional | n_lo, | ||
| integer, intent(out), optional | n_hi ) |
Given the current electronic coordinate q and the integer span S, compute the bracketing integers N_lo/N_hi and the mixing weight lambda = (q - N_lo)/S.
| electrons | ... |
| lambda | ... |
| n_lo | ... |
| n_hi | ... |
Definition at line 78 of file qs_electrons_charge_mixing.F.
| subroutine, public qs_electrons_charge_mixing::push_lambda_to_mixed_env | ( | type(force_env_type), pointer | force_env, |
| real(kind=dp), intent(in) | lambda ) |
Push a dynamically computed lambda into the MIXED force_eval's input structure. force_env_methods.F::mixed_energy_forces reads MIXEDLINEARLAMBDA fresh from the input structure every call, so this alone is sufficient to make energy/force/virial mixing track our own extended-variable dynamics with zero changes to stock CP2K.
IMPORTANT sign convention: CP2K's own mixed_energy_forces computes E = LAMBDA*E(force_eval#1) + (1-LAMBDA)*E(force_eval#2). In our input, force_eval#1 is the N_lo endpoint and force_eval#2 is the N_hi endpoint, but our own lambda argument here uses the opposite (standard interpolation) convention, lambda = (q-N_lo)/S, i.e. lambda=0 at N_lo and lambda=1 at N_hi. So CP2K's LAMBDA must be set to (1-lambda), NOT lambda directly, or the mixing comes out backwards (mostly N_hi when q is near N_lo).
| force_env | the top-level MIXED force_env (force_envin_use must be use_mixed_force; force_envforce_env_section must contain a MIXED subsection with MIXING_TYPE LINEAR_COMBINATION) |
| lambda | our convention: 0 at N_lo, 1 at N_hi |
Definition at line 123 of file qs_electrons_charge_mixing.F.
| subroutine, public qs_electrons_charge_mixing::compute_mixed_workfunction | ( | type(force_env_type), pointer | force_env, |
| integer, intent(in) | axis, | ||
| real(kind=dp), intent(in) | wf_min, | ||
| real(kind=dp), intent(in) | wf_max, | ||
| real(kind=dp), intent(in) | q, | ||
| integer, intent(in) | span, | ||
| real(kind=dp), intent(out) | fermi_level_ev, | ||
| real(kind=dp), intent(out) | workfunction_ev, | ||
| real(kind=dp), intent(in), optional | lambda_in ) |
Fermi level / workfunction mixing across the two integer-charge endpoints. Stock MIXED only combines energy/force/virial/dipole (mixed_environment_utils.F::mixed_map_forces), so these two quantities – needed every step by our own O-B-A-B-O potentiostat to decide how to move q(t) – are interpolated here, mirroring FFChargeMixTwoSockets._mix_results's fermi/workfunction lines.
| force_env | the top-level MIXED force_env; force_envsub_force_env(1) and (2) must each be an ordinary QS force_env |
| axis | ... |
| wf_min | ... |
| wf_max | ... |
| q | current electronic coordinate (total electron count) |
| span | charge_mixing_span (S) |
| fermi_level_eV | ... |
| workfunction_eV | ... |
| lambda_in | optional explicit mixing weight; callers pass the value from qs_electrons_charge_mixing_lambda so a FIXED lambda is honoured here too |
Definition at line 155 of file qs_electrons_charge_mixing.F.
| subroutine, public qs_electrons_charge_mixing::charge_mixing_maybe_rebuild | ( | type(force_env_type), pointer | force_env, |
| type(qs_electrons_type), pointer | electrons ) |
Check whether q(t) has drifted outside the currently-built [N_lo,N_hi] bracket and, if so, rebuild whichever endpoint(s) are now stale at the new integer electron count. Must be called every step, right after qs_electrons_drift and before push_lambda_to_mixed_env, so LAMBDA is always pushed against a bracket that actually matches the physically-built sub_force_env pair.
| force_env | the top-level MIXED force_env |
| electrons | ... |
Definition at line 243 of file qs_electrons_charge_mixing.F.
| subroutine, public qs_electrons_charge_mixing::charge_mixing_sync_endpoint_state | ( | type(force_env_type), pointer | force_env, |
| type(qs_electrons_type), pointer | electrons ) |
Record which integer electron counts the two FORCE_EVAL blocks are ACTUALLY built for, by interrogating them, so that the very first charge_mixing_maybe_rebuild can tell whether the input agrees with q.
This closes the "startup endpoint mismatch" hole. The port assumes throughout that sub_force_env(1) carries N_lo and sub_force_env(2) carries N_hi – push_lambda_to_mixed_env relies on it, since CP2K evaluates E = LAMBDA*E(1) + (1-LAMBDA)*E(2) and we therefore push LAMBDA = 1 - lambda. Nothing in stock CP2K enforces that ordering: an input whose two blocks are written high-charge-first would mix BACKWARDS, silently, for the whole run, and the only symptom would be a potentiostat that pushes q the wrong way.
Reading the counts back here and letting maybe_rebuild correct them removes the dependence on the order of the input blocks.
Reading rather than blindly forcing a rebuild matters: seeding the tracking with -HUGE(0) would also fix the ordering, but it would force two full qs_init rebuilds at step 0 of every run, including the vast majority whose input is already correct.
| force_env | the top-level MIXED force_env |
| electrons | ... |
Definition at line 311 of file qs_electrons_charge_mixing.F.