qmctorch.sampler package

Module contents

class qmctorch.sampler.SamplerBase(nwalkers, nstep, step_size, ntherm, ndecor, nelec, ndim, init, cuda)[source]

Bases: object

Base class for the sampler

Parameters:
  • nwalkers (int) – number of walkers
  • nstep (int) – number of MC steps
  • step_size (float) – size of the steps in bohr
  • ntherm (int) – number of MC steps to thermalize
  • ndecor (int) – unmber of MC steps to decorellate
  • nelec (int) – number of electrons in the system
  • ndim (int) – number of cartesian dimension
  • init (dict) – method to initialize the walkers
  • cuda ([type]) – [description]
get_sampling_size()[source]

evaluate the number of sampling point we’ll have.

class qmctorch.sampler.Metropolis(nwalkers: int = 100, nstep: int = 1000, step_size: float = 0.2, ntherm: int = -1, ndecor: int = 1, nelec: int = 1, ndim: int = 3, init: Dict[KT, VT] = {'max': 5, 'min': -5}, move: Dict[KT, VT] = {'proba': 'normal', 'type': 'all-elec'}, cuda: bool = False)[source]

Bases: qmctorch.sampler.sampler_base.SamplerBase

Metropolis Hasting generator

Parameters:
  • nwalkers (int, optional) – Number of walkers. Defaults to 100.
  • nstep (int, optional) – Number of steps. Defaults to 1000.
  • step_size (int, optional) – length of the step. Defaults to 0.2.
  • nelec (int, optional) – total number of electrons. Defaults to 1.
  • ntherm (int, optional) – number of mc step to thermalize. Defaults to -1, i.e. keep only the last position
  • ndecor (int, optional) – number of mc step for decorelation. Defauts to 1.
  • ndim (int, optional) – total number of dimension. Defaults to 3.
  • init (dict, optional) – method to init the positions of the walkers. See Molecule.domain()
  • move (dict, optional) –

    method to move the electrons. default(‘all-elec’,’normal’)

    ’type’:
    ’one-elec’: move a single electron per iteration

    ’all-elec’: move all electrons at the same time

    ’all-elec-iter’: move all electrons by iterating through single elec moves

    ’proba’ :
    ’uniform’: uniform in a cube

    ’normal’: gussian in a sphere

  • cuda (bool, optional) – turn CUDA ON/OFF. Defaults to False.
Examples::
>>> mol = Molecule('h2.xyz')
>>> wf = SlaterJastrow(mol)
>>> sampler = Metropolis(nwalkers=100, nelec=wf.nelec)
>>> pos = sampler(wf.pdf)
log_data()[source]

log data about the sampler.

configure_move(move: Dict[KT, VT])[source]

Configure the electron moves

Parameters:move (dict, optional) –

method to move the electrons. default(‘all-elec’,’normal’)

’type’:
’one-elec’: move a single electron per iteration

’all-elec’: move all electrons at the same time

’all-elec-iter’: move all electrons by iterating through single elec moves

’proba’ :
’uniform’: uniform ina cube

’normal’: gussian in a sphere
Raises:ValueError – If moves are not recognized
move(pdf: Callable, id_elec: int) → <sphinx.ext.autodoc.importer._MockObject object at 0x7f59abd47410>[source]

Move electron one at a time in a vectorized way.

Parameters:
  • pdf (callable) – function to sample
  • id_elec (int) – index f the electron to move
Returns:

new positions of the walkers
Return type:

torch.tensor
class qmctorch.sampler.Hamiltonian(nwalkers: int = 100, nstep: int = 100, step_size: float = 0.2, L: int = 10, ntherm: int = -1, ndecor: int = 1, nelec: int = 1, ndim: int = 3, init: Dict[KT, VT] = {'max': 5, 'min': -5}, cuda: bool = False)[source]

Bases: qmctorch.sampler.sampler_base.SamplerBase

Hamiltonian Monte Carlo Sampler.

Parameters:
  • nwalkers (int, optional) – Number of walkers. Defaults to 100.
  • nstep (int, optional) – Number of steps. Defaults to 100.
  • step_size (int, optional) – length of the step. Defaults to 0.2.
  • L (int, optional) – length of the trajectory . Defaults to 10.
  • nelec (int, optional) – total number of electrons. Defaults to 1.
  • ntherm (int, optional) – number of mc step to thermalize. Defaults to -1, i.e. keep only last position
  • ndecor (int, optional) – number of mc step for decorrelation. Defaults to 1.
  • ndim (int, optional) – total number of dimension. Defaults to 3.
  • init (dict, optional) – method to init the positions of the walkers. See Molecule.domain()
  • cuda (bool, optional) – turn CUDA ON/OFF. Defaults to False.
static get_grad(func, inp)[source]

get the gradient of the pdf using autograd

Parameters:
  • func (callable) – function to compute the pdf
  • inp (torch.tensor) – input of the function
Returns:

gradients of the wavefunction
Return type:

torch.tensor
static log_func(func)[source]

Compute the negative log of a function

Parameters:func (callable) – input function
Returns:negative log of the function
Return type:callable
class qmctorch.sampler.GeneralizedMetropolis(nwalkers=100, nstep=1000, step_size=3, ntherm=-1, ndecor=1, nelec=1, ndim=1, init={'max': 5, 'min': -5, 'type': 'uniform'}, cuda=False)[source]

Bases: qmctorch.sampler.sampler_base.SamplerBase

Generalized Metropolis Hasting sampler

Parameters:
  • nwalkers (int, optional) – number of walkers. Defaults to 100.
  • nstep (int, optional) – number of steps. Defaults to 1000.
  • step_size (int, optional) – size of the steps. Defaults to 3.
  • ntherm (int, optional) – number of steps for thermalization. Defaults to -1.
  • ndecor (int, optional) – number of steps for decorelation. Defaults to 1.
  • nelec (int, optional) – number of electron. Defaults to 1.
  • ndim (int, optional) – number of dimensions. Defaults to 1.
  • init (dict, optional) – method to initialize the walkers. Defaults to {‘type’: ‘uniform’, ‘min’: -5, ‘max’: 5}.
  • cuda (bool, optional) – use cuda. Defaults to False.
move(drift)[source]

Move electron one at a time in a vectorized way.

Parameters:drift (torch.tensor) – drift velocity of the walkers
Returns:new positions of the walkers
Return type:torch.tensor
trans(xf, xi, drifti)[source]

transform the positions

Parameters:
  • xf ([type]) – [description]
  • xi ([type]) – [description]
  • drifti ([type]) – [description]
Returns:

[description]
Return type:

[type]
get_drift(pdf, x)[source]

Compute the drift velocity

Parameters:
  • pdf (callable) – function that returns the density
  • x (torch.tensor) – positions of the walkers
Returns:

drift velocity
Return type:

torch.tensor