uniqc.pytorch.gradient module

Parameter-shift rule gradient computation for quantum circuits.

The parameter-shift rule allows exact gradient computation for parametric quantum gates without finite differences. For gates with generator G:

∂⟨G(θ)⟩/∂θ = (⟨G(θ + π/2)⟩ - ⟨G(θ - π/2)⟩) / 2

This module provides functions to compute gradients for all parameters in a parametric circuit.

uniqc.pytorch.gradient.compute_all_gradients(circuit, expectation_fn, shift=1.5707963267948966)

Compute gradients for all parameters in a circuit.

Uses the parameter-shift rule for each parameter independently.

Parameters:

• circuit (Circuit) – Parametric circuit

• expectation_fn (Callable[[Circuit], float]) – Function that computes expectation value from a circuit

• shift (float) – Shift value for parameter-shift rule

Returns:

Dictionary mapping parameter names to gradient values

Return type:

dict[str, float]

Example

>>> grads = compute_all_gradients(circuit, expectation)
>>> print(grads)  # {'theta': 0.5, 'phi': -0.3}

uniqc.pytorch.gradient.parameter_shift_gradient(circuit, param_name, expectation_fn, shift=1.5707963267948966)

Compute gradient using the parameter-shift rule.

For a parametric gate G(θ), the gradient of an expectation value is:

∂⟨G(θ)⟩/∂θ = 0.5 * (⟨G(θ + π/2)⟩ - ⟨G(θ - π/2)⟩)

Parameters:

• circuit (Circuit) – Parametric circuit with the parameter

• param_name (str) – Name of the parameter to differentiate

• expectation_fn (Callable[[Circuit], float]) – Function that computes expectation value from a circuit

• shift (float) – Shift value (default π/2 for standard Pauli rotation gates)

Returns:

Gradient value

Return type:

float

Example

>>> def expectation(c):
...     # Simulate and compute <Z0>
...     sim.simulate(c.originir)
...     return sim.expectation([("Z", [0])])
>>> grad = parameter_shift_gradient(circuit, "theta", expectation)