What problem does it solve? Computing gradients of parameterized quantum circuits is essential for variational algorithms, but analytic gradient methods only work with specific gate sets. This Skill numerically approximates gradients for any differentiable quantum circuit using finite difference methods, providing a universal fallback when parameter-shift or LCU methods are not applicable. ## Core Features & Use Cases - Estimator and Sampler Gradients: Supports both FiniteDiffEstimatorGradient for expectation value gradients and FiniteDiffSamplerGradient for probability distribution gradients. - Three Difference Schemes: Central (second-order accurate, 2n evaluations), forward, and backward (first-order, n+1 evaluations) schemes with configurable step size epsilon. - Batched Execution: All perturbed circuit evaluations are submitted as a single primitive job to minimize overhead. - Use Case: When training a variational quantum eigensolver whose ansatz contains gates outside the parameter-shift rule's supported set, use central finite difference with epsilon=1e-2 to compute energy gradients for the optimizer. ## Quick Start Ask the AI to compute the gradient of a parameterized Qiskit circuit with respect to an observable using the finite difference method with central differencing.
