quantax.optimizer.MinSR#

class quantax.optimizer.MinSR#

Bases: TDVP

MinSR optimization, specifically designed for Sequential networks. The optimization utilizes gradient checkpointing method and structured derivatives to reduce the memory cost. See MinSR paper for details.

__init__(state: Variational, hamiltonian: Operator, solver: Callable | None = None)#
Parameters:

  • state – Variational state to be optimized.

  • hamiltonian – The Hamiltonian for the evolution.

  • solver – The numerical solver for the matrix inverse, default to minsr_pinv_eig.

…warning:

The model must be `~quantax.nn.Sequential`, otherwise one should use
`~quantax.optimizer.TDVP`.

The vs_type of the variational state should be ``real_or_holomorphic`` or
``real_to_complex``. In the latter case, the complex neurons are only allowed
in the last few unparametrized layers.
get_Tmat(samples: Samples) → Array#

Compute the \(T\) matrix in MinSR

Ohvp(samples: Samples, vec: Array) → Array#

Compute \(\bar O^† v\). vec @ jac is used instead of vjp for better precision.

solve(samples: Samples, Tmat: Array, Ebar: Array) → Array#

Solve the equation \(\bar O \dot \theta = \bar \epsilon\) for given \(\bar O\) and \(\bar \epsilon\).

get_step(samples: Samples) → Array#

Obtain the MinSR step from given samples

property VarE: float | None#

Energy variance \(\left< (H - E)^2 \right>\) of the current step.

property energy: float | None#

Energy of the current step.

get_Ebar(samples: Samples) → Array#

Compute \(\bar \epsilon\) for given samples. The local energy is \(E_{loc, s} = \sum_{s'} \frac{\psi_{s'}}{\psi_s} \left< s|H|s' \right>\), and \(\bar \epsilon\) is defined as \(\bar \epsilon = \frac{1}{\sqrt{N_s}} (E_{loc, s} - \left<E_{loc, s}\right>)\).

get_Obar(samples: Samples) → Array#

Calculate \(\bar O = \frac{1}{\sqrt{N_s}}(\frac{1}{\psi} \frac{\partial \psi}{\partial \theta} - \left< \frac{1}{\psi} \frac{\partial \psi}{\partial \theta} \right>)\) for given samples.

property hamiltonian: Operator#

The Hamiltonian for the evolution.

property holomorphic: bool#

Whether the state is holomorphic.

property imag_time: bool#

Whether to use imaginary-time evolution.

property kazcmarz_mu: float#

Whether to use the kazcmarz scheme.

property state: Variational#

Variational state to be optimized.

property vs_type: int#

The vs_type of the state.