Skip to main content

Module ad

Module ad 

Source
Expand description

Rust-native automatic-differentiation primitives and optional rule adapters. Rust-native differentiation primitives for QMBED scientific operations.

The native layer owns the mathematical derivative semantics. Optional adapters such as the optional chainrules module only translate these tested primitives into another protocol; they do not reimplement the derivatives.

Structs§

ApplyCotangents
State and parameter cotangents returned by an operator pullback.
ApplyJvp
Primal value and forward tangent from parameterized operator application.
ApplyPullback
One-shot pullback for apply_vjp.
GradientDiagnostics
Backend-independent evidence returned with a native derivative.
GroundStateEnergyGradient
Ground-state energy, Hellmann–Feynman gradient, and solver evidence.
ParameterDirection
Ordered forward perturbation of ParameterValues.
ParameterGradient
Ordered reverse sensitivity of ParameterValues.
ParameterSchema
Stable names and differential domains for an ordered parameter vector.
ParameterValues
Ordered primal values for a parameterized operator.

Enums§

GradientStatus
Reliability classification attached to a scientific gradient.
ParameterDomain
Differential domain of one ordered operator parameter.

Functions§

apply_jvp
Evaluate a parameterized operator and its JVP without materializing A(θ).
apply_vjp
Evaluate a parameterized operator and prepare a one-shot reverse pullback.
ground_state_energy_gradient
Differentiate the isolated algebraic ground-state energy.