Back to Eqiora docs Skip to main content

Module differentiation

Module differentiation 

Source
Description

Implicit forward and adjoint differentiation over lowered relations.

Structs§

AcceptedLinearization
A linearization whose primal residual has been independently accepted.
AcceptedOutputLinearization
One accepted relation and output projection bound for a derivative action.
AdjointGradient
Accepted adjoint and total objective gradient.
DiscreteAdjointCheckpoint
One explicit accepted-state boundary inside a discrete adjoint trajectory.
DiscreteTrajectoryAdjoint
Reverse accumulation result over one validated discrete trajectory.
ForwardOutputSensitivity
Accepted implicit state sensitivity and its selected output projection.
ParameterJacobian
Matrix-free parameter Jacobian action R_p from the same relation.
StateJacobian
Matrix-free state Jacobian action R_w derived from one relation JVP/VJP.

Functions§

adjoint_gradient
Solve the transposed state system and form the total parameter gradient.
adjoint_objective_gradient
Compose one accepted relation and one objective intended for its accepted point.
adjoint_output_gradient
Pull one complete output cotangent back through an accepted implicit relation.
discrete_trajectory_adjoint
Compose reverse-mode cotangents over accepted discrete step relations.
forward_output_sensitivity
Solve the implicit forward system and project it into complete outputs.
forward_sensitivity
Solve R_w dw = -R_p dp at one accepted linearization.