Variational quantum state module

class jVMC.vqs.NQS(net, logarithmic=True, batchSize=1000, seed=1234, orbit=None, avgFun=<function avgFun_Coefficients_Exp>)

Wrapper class providing basic functionality of variational states.

This class can operate in two modi:
  1. Single-network ansatz

    Quantum state of the form \(\psi_\theta(s)\equiv\exp(r_\theta(s))\), where the network \(r_\theta\) is a) holomorphic, i.e., parametrized by complex valued parameters \(\vartheta\). b) non-holomorphic, i.e., parametrized by real valued parameters \(\theta\).

  2. Two-network ansatz

    Quantum state of the form \(\psi_\theta(s)\equiv\exp(r_{\theta_r}(s)+i\varphi_{\theta_\phi}(s))\) with an amplitude network \(r_{\theta_{r}}\) and a phase network \(\varphi_{\theta_\phi}\) parametrized by real valued parameters \(\theta_r,\theta_\phi\).

Initializer arguments:
  • net: Variational network.

    A network has to be registered as pytree node and provide a __call__ function for evaluation. It is expected that the network is of type jVMC.nets.sym_wrapper.SymNet. If the network is composed of two networks, the correct wrapping structure is jVMC.nets.sym_wrapper.SymNet(jVMC.nets.two_nets_wrapper.TwoNets).

  • batchSize: Batch size for batched network evaluation. Choice of this parameter impacts performance: with too small values performance is limited by memory access overheads, too large values can lead to “out of memory” issues.

  • seed: Seed for the PRNG to initialize the network parameters.

__call__(s)

Evaluate variational wave function.

Compute the logarithmic wave function coefficients \(\ln\psi(s)\) for computational configurations \(s\).

Args:

  • s: Array of computational basis states.

Returns:

Logarithmic wave function coefficients \(\ln\psi(s)\).

get_parameters()

Get variational parameters.

Returns:

Array holding current values of all variational parameters.

get_sampler_net()

Get real part of NQS and current parameters

This function returns a function that evaluates the real part of the NQS, \(\text{Re}(\log\psi(s))\), and the current parameters.

Returns:

Real part of the NQS and current parameters

gradients(s)

Compute gradients of logarithmic wave function.

Compute gradient of the logarithmic wave function coefficients, \(\nabla\ln\psi(s)\), for computational configurations \(s\).

Args:

  • s: Array of computational basis states.

Returns:

A vector containing derivatives \(\partial_{\theta_k}\ln\psi(s)\) with respect to each variational parameter \(\theta_k\) for each input configuration \(s\).

gradients_dict(s)

Compute gradients of logarithmic wave function and return them as dictionary.

Compute gradient of the logarithmic wave function coefficients, \(\nabla\ln\psi(s)\), for computational configurations \(s\).

Args:

  • s: Array of computational basis states.

Returns:

A dictionary containing derivatives \(\partial_{\theta_k}\ln\psi(s)\) with respect to each variational parameter \(\theta_k\) for each input configuration \(s\).

set_parameters(P)

Set variational parameters.

Sets new values of all variational parameters.

Args:

  • P: New values of variational parameters.

update_parameters(deltaP)

Update variational parameters.

Sets new values of all variational parameters by adding given values. If parameters are not initialized, parameters are set to deltaP.

Args:

  • deltaP: Values to be added to variational parameters.