The MTK NPT integrator carries a nhc_zero_eta_dot permanent-zero buffer that is passed to npt_barostat_half_step to mute the kernel's incorrect −eta_dot · h_dot drag. Canonical MTK has no thermostat coupling in the barostat half-step (cell damping comes from the separate b_scale operator), so the kernel API change in NVIDIA/nvalchemi-toolkit-ops#77 drops the eta_dots argument entirely.
Once nvalchemi-toolkit-ops > 0.3.1 ships:
- Drop the
nhc_zero_eta_dot buffer from _init_state and _make_new_state.
- Switch both
npt_barostat_half_step(...) calls to the eta_dots-free signature.
- Remove the
# Workaround for nvalchemi-toolkit-ops <= 0.3.1 comments and the npt_barostat_half_step op shim in nvalchemi/dynamics/_ops/npt_nph.py.
Introduced in #90.
The MTK NPT integrator carries a
nhc_zero_eta_dotpermanent-zero buffer that is passed tonpt_barostat_half_stepto mute the kernel's incorrect−eta_dot · h_dotdrag. Canonical MTK has no thermostat coupling in the barostat half-step (cell damping comes from the separateb_scaleoperator), so the kernel API change in NVIDIA/nvalchemi-toolkit-ops#77 drops theeta_dotsargument entirely.Once
nvalchemi-toolkit-ops > 0.3.1ships:nhc_zero_eta_dotbuffer from_init_stateand_make_new_state.npt_barostat_half_step(...)calls to the eta_dots-free signature.# Workaround for nvalchemi-toolkit-ops <= 0.3.1comments and thenpt_barostat_half_stepop shim innvalchemi/dynamics/_ops/npt_nph.py.Introduced in #90.