molecular_dynamics

Full Documentation for hippynn.molecular_dynamics package. Click here for a summary page.

Molecular dynamics driver with great flexibility and customizability regarding which quantities which are evolved and what algorithms are used to evolve them. Calls a hippynn Predictor on current state during each MD step.

This module is only available if the ase package is installed.

class LangevinDynamics(force_db_name: str, temperature: float, frix: float, units_force: float = None, units_acc: float = None, seed: int | None = None)

Bases: VariableUpdater

Implements the Langevin algorithm

post_step(dt: float, model_outputs: dict)

Updates to variables performed during each step of MD simulation after HIPNN model evaluation

Parameters:

• dt – timestep

• model_outputs – dictionary of HIPNN model outputs

pre_step(dt: float)

Updates to variables performed during each step of MD simulation before HIPNN model evaluation

Parameters:

dt – timestep

required_variable_data = ['position', 'velocity', 'mass']

class MolecularDynamics(variables: list[Variable], model: Predictor, device: device | None = None, dtype: dtype | None = None)

Bases: object

Driver for MD run

get_data()

Returns a dictionary of the recorded data

reset_data()

Clear all recorded data

run(dt: float, n_steps: int, record_every: int | None = None)

Run n_steps of MD algorithm.

Parameters:

• dt – timestep

• n_steps – number of steps to execute

• record_every – frequency at which to store the data at a step in memory, record_every = 1 means every step will be stored, defaults to None

to(arg)

to(arg: device)

to(arg: dtype)

property device

property dtype

property model

property variables

class NullUpdater

Bases: VariableUpdater

Makes no change to the variable data at each step of MD.

post_step(dt, model_outputs)

Updates to variables performed during each step of MD simulation after HIPNN model evaluation

Parameters:

• dt – timestep

• model_outputs – dictionary of HIPNN model outputs

pre_step(dt)

Updates to variables performed during each step of MD simulation before HIPNN model evaluation

Parameters:

dt – timestep

class Variable(name: str, data: dict[str, Tensor], model_input_map: dict[str, str] = {}, updater: VariableUpdater = None, device: device | None = None, dtype: dtype | None = None)

Bases: object

Tracks the state of a quantity (eg. position, cell, species, volume) on each particle or each system in an MD simulation. Can also hold additional data associated to that quantity (such as its velocity, acceleration, etc…)

to(arg)

to(arg: device)

to(arg: dtype)

property data

property device

property dtype

property model_input_map

property updater

class VelocityVerlet(force_db_name: str, units_force: float = None, units_acc: float = None)

Bases: VariableUpdater

Implements the Velocity Verlet algorithm

post_step(dt: float, model_outputs: dict)

Updates to variables performed during each step of MD simulation after HIPNN model evaluation

Parameters:

• dt – timestep

• model_outputs – dictionary of HIPNN model outputs

pre_step(dt: float)

Updates to variables performed during each step of MD simulation before HIPNN model evaluation

Parameters:

dt – timestep

required_variable_data = ['position', 'velocity', 'acceleration', 'mass']

Submodules

• md module