DriverMLMOD

struct DriverMLMOD

The driver of the functionality for the shared library class. This specifies the prototype for the API interface with lammps.

Public Functions

DriverMLMOD(class FixMLMOD*, class LAMMPS*, int, char**)

Main constructor for mlmod package driver.

Parameters:

• FixMLMOD* – reference to the mlmod-lammps “fix” interface

• LAMMPS* – reference to the running lammps instance

• narg – number of command arguments

• args – list of strings for the arguments

DriverMLMOD()

Constructor for empy object which is used primarily for testing.

~DriverMLMOD()

The deconstructor.

int setmask()

Lammps interface code for signaling the type of simulation.

virtual void init()

Used for initializing the driver settings when created.

void setup(int vflag)

Used for setting up references in the driver after lammps is initialized.

virtual void initial_integrate(int)

Lammps calls this routine at the start of each time-step.

virtual void final_integrate()

Lammps calls this routine at the end of each time-step.

void reset_dt()

Lammps calls this routine when the time-step size changes.

void post_force(int vflag)

Lammps calls this routine after it finished computing forces.

void pre_exchange()

Lammps calls this routine before updating atom tables.

void end_of_step()

Lammps calls this at the end of a simulation step.

double compute_array(int i, int j)

Lammps calls this to extract an array of data from the fix.

void init_attributes()

This is called from mlmod routines in lammps to initializes the driver class attributes.

void init_from_fix()

This is called from mlmod routines in lammps when the “fix” is setup.

void parse_xml_params(char *filename)

Parses the parameters from the xml files associated with the model_type.

void params_parse(int narg, char **arg)

Parses the parameters from the xml files associated with the model_type.

void write_info()

Writes additional simulation data to disk each time-step. Writes a file giving information about the current simulation for post-processing scripts.

void write_final_info()

Writes a file for the final step of the simulation, see write_info().

void initial_integrate_dX_MF_Q1_ML1_Pair()

Dynamic integrator with ML model for the mobility tensor \(M(\mathbf{X})\) for a pair of particles with \( \left\{ \begin{array}{lll} {d\mathbf{X}}/{dt} & = & \mathbf{M}(\mathbf{X})\mathbf{F} + k_B{T}\nabla_X \cdot \mathbf{M}(\mathbf{X}) + \mathbf{g}, \\ \langle \mathbf{g}(s) \mathbf{g}(t)^T \rangle & = & 2 k_B{T} \mathbf{M}(\mathbf{X}) \delta(t - s). \\ \end{array} \right. \).

void initial_integrate_dX_MF_Q1_ML1_N2N()

Dynamic integrator with ML model for the mobility tensor \(M(\mathbf{X})\) for pairwise interactions for a collection of particles with \( \left\{ \begin{array}{lll} {d\mathbf{X}}/{dt} & = & \mathbf{M}(\mathbf{X})\mathbf{F} + k_B{T}\nabla_X \cdot \mathbf{M}(\mathbf{X}) + \mathbf{g}, \\ \langle \mathbf{g}(s) \mathbf{g}(t)^T \rangle & = & 2 k_B{T} \mathbf{M}(\mathbf{X}) \delta(t - s). \\ \end{array} \right. \).

void post_force_F_ML1()

Computes an ML model for a force determined and applied collectively for all the particles in the designated group \( \mathbf{X} = \{\mathbf{X}_i\}_{i \in \mathcal{G}} \), where \( F(\mathbf{X},\mathbf{V},\mathbf{F},\mathbf{I_T},t) \).

void post_force_F_X_ML1()

Computes an ML model for a force applied individually to each particle \( \mathbf{X}_i \), with \( F_i(\mathbf{X}_i,\mathbf{V}_i,\mathbf{F}_i,\mathbf{I_T}_i,t) \). To compute forces collectively over the particles, see post_force_F_ML1().

void post_force_F_Pair_ML1()

Computes an ML model for a force applied on pairs of particles within the designated group, \( \mathbf{X}_{ij} = (\mathbf{X}_i,\mathbf{X}_j) \) for $i,j \in \mathcal{G}$, with \( F_{ij}(\mathbf{X}_{ij},\mathbf{V}_{ij},\mathbf{F}_{ij},\mathbf{I_T}_{ij},t) \). To compute forces collectively over the particles, see post_force_F_ML1().

void initial_integrate_QoI_ML1()

Computes an ML model for a quantity of interest (QoI), \( Q(\mathbf{X},\mathbf{V},\mathbf{F}) \).

void initial_integrate_Dyn_ML1()

Computes an ML model for the particle dynamics, \( \mathbf{X}^{n+1} = \Gamma(\mathbf{X}^n,\mathbf{V}^n,\mathbf{F}^n,t_n) \)

. Depending on the mask_fix settings this can be a single time-step or Verlet-style integrator. The initial step uses the machine learning model specified in dyn1_filenmae and the final step the model in dyn2_filename.

See the documentation pages and examples for more details.

void final_integrate_dX_MF_ML1()

Dynamic integrator with ML model for the mobility tensor \(M(\mathbf{X})\) for pairwise interactions for a collection of particles with \( {d\mathbf{X}}/{dt} = \mathbf{M}(\mathbf{X})\mathbf{F}. \).

void final_integrate_dX_MF()

Dynamic integrator for a test mobility tensor \(M(\mathbf{X})\) for pairwise interactions for a collection of particles with \( {d\mathbf{X}}/{dt} = \mathbf{M}(\mathbf{X})\mathbf{F}. \).

void final_integrate_dX_MF_Q1_ML1_Pair()

See initial_integrate_dX_MF_Q1_ML1_Pair().

void final_integrate_dX_MF_Q1_ML1_N2N()

See initial_integrate_dX_MF_Q1_ML1_N2N().

void final_integrate_QoI_ML1()

See initial_integrate_QoI_ML1().

void final_integrate_Dyn_ML1()

Dynamic integrator based on specified machine learning model. For a Verlet-style integrator this calls the second half of the integration step using the model specified by dyn2_filenmae. For more details, see initial_integrate_Dyn_ML1().

void parse_xml_params_dX_MF(XMLElement *model_data_element)

Parsers for the specific model cases.

double **create_db_vec(int num_rows, int num_cols, double **db_vec_block, int *db_vec_size)

Creates a vector array x[][] for storing double precision floating point values.

Parameters:

• num_cols – number of dimensions

• num_rows – number of indices

• db_vec_block – block of memory for vector (set to zero to allocate)

• db_vec_size – size of the block of memory allocated

Returns:

(double **) vector array with shape=[num_rows,num_cols].

void set_db_vec_to_tensor_full(double **z, at::Tensor output_m, int num_rows, int num_cols, int *rowII, int offset_tensor)

Sets a vector array x[][] from torch tensor data a[:][0].

This implements a copy of the form x[i][d] = a[I + offset][0], where I = i*num_cols + d x.shape=[n,num_cols], a.shape=[n*num_cols,1]

Parameters:

• double** – vector array, (typical shape=[n,num_cols])

• output_m – torch tensor

• num_cols – number of dimensions

• num_rows – number of indices

• rowII – subset of indices to set (rowII=NULL, use all)

• offset_tensor – for tensor array, offset for start of values

void add_tensor_full_to_db_vec(double **z, at::Tensor output_m, int num_rows, int num_cols, int *rowII, int offset_tensor)

Sets a vector array x[][] from torch tensor data a[:][0].

This implements a copy of the form x[i][d] += a[I + offset][0], where I = i*num_cols + d x.shape=[n,num_cols], a.shape=[n*num_cols,1]

Parameters:

• double** – vector array, (typical shape=[n,num_cols])

• output_m – torch tensor

• num_cols – number of dimensions

• num_rows – number of indices

• rowII – subset of indices to set (rowII=NULL, use all)

• offset_tensor – for tensor array, offset for start of values

void add_tensor_seg_to_db_vec(double **z, at::Tensor output_m, int num_rows, int num_cols, int *rowII, int offset_tensor)

Sets a vector array x[][] from torch tensor data a[:][0] which has the same overall size as x.

This implements a copy of the form x[iii][d] += a[I + offset][0], where I = iii*num_cols + d, iii = rowII[i], x.shape=[num_rows,num_cols], a.shape=[num_rows*num_cols,1].

Parameters:

• double** – vector array, (typical shape=[n,num_cols])

• output_m – torch tensor

• num_cols – number of dimensions

• num_rows – number of indices

• rowII – subset of indices to set (rowII=NULL, use all)

• offset_tensor – for tensor array, offset for start of values

void set_to_zero_db_vec(double **z, int num_rows, int num_cols, int *rowII)

Sets a vector array x[][] to zero.

This implements a copy of the form x[i][d] = 0.0 x.shape=[n,num_cols], a.shape=[n*num_cols,1]

Parameters:

• double** – vector array, (typical shape=[n,num_cols])

• num_cols – number of dimensions

• num_rows – number of indices

• rowII – subset of indices to set (rowII=NULL, use all)

int get_tensor_input_size(int num_II, int *II, int num_dim, int mask_input, int mask_list_n, int *mask_list, int groupbit, int *mask_atom)

Get tensor size given the masks and groups from lammps.

void build_tensor_from_masks(at::TensorAccessor<float, 2, at::DefaultPtrTraits, long> input_m_a, int num_indices, int *indices, int mask_input, int mask_list_n, int *mask_list, int groupbit, int *mask_atom, double **x, double **v, double **f, int *type, double time)

Creates tensors for positions, velocity, and forces from lammps.

void get_indices_for_group(int groupbit, int *mask_atom, int nlocal, int *II_size_ptr, int **II_ptr, int *num_II_ptr)

Get indices for group from lammps.

void print_tensor_2d(const char *name, at::Tensor t)

Display information.

void write_array_txt_1D(const char *filename, int n, double *v)

Writing tensor arrays to disk for processing and debugging.

void test1()

Test codes helpful in debugging.

Public Members

LAMMPS *lammps

lammps instance

FixMLMOD *fixMLMOD

our lammps “fix” interface instance

int mlmod_seed

seed for random number generator

string params_filename

level of output to print

RanMars *random

random number generator

int model_type

integer flag for the model type to use

string model_type_str

name of the model type for setting up integer flags and displaying information (note may be current length limit)

void *model_data

general model specific data

int MODEL_TYPE_NULL

Model types (constants)

Friends

friend class FixMLMOD

struct ModelData_dX_MF_ML1_Type

Data for a basic ML mobility-based simulation

Public Members

string *M_ii_filename

filename for the ML model for the self-mobility tensor components

string *M_ij_filename

filename for the ML model for pair-mobility tensor components

torch::jit::Module *M_ii_model

torch ML model for self-mobility

torch::jit::Module *M_ij_model

torch ML model for pair-mobility

int flag_init

internal variables

struct ModelData_dX_MF_Q1_ML1_N2N_Type

State information for specified simulation modality.

Public Members

int flag_init

internal variables

struct ModelData_dX_MF_Q1_ML1_Pair_Type

Data for an overdamped stochastic ML mobility-based simulation, see initial_integrate_dX_MF_Q1_ML1_Pair() and final_integrate_dX_MF_Q1_ML1_Pair().

Public Members

int flag_init

internal variables

struct ModelData_dX_MF_Type

Data for hydrodynamic model case (used primarily for testing)

Public Members

int flag_init

internal variables

struct ModelData_Dyn_ML1_Type

State information for specified simulation modality.

struct ModelData_F_ML1_Type

State information for specified simulation modality.

struct ModelData_F_Pair_ML1_Type

State information for specified simulation modality.

struct ModelData_F_X_ML1_Type

State information for specified simulation modality.

struct ModelData_QoI_ML1_Type

State information for specified simulation modality.

[mlmod github codes]