HessianFunction.FromProduct Method

public static Func<Vector<double>, LinearOperator<double>?, LinearOperator<double>> FromProduct(
	Action<Vector<double>, Vector<double>, Vector<double>> multiply
)

Public Shared Function FromProduct ( 
	multiply As Action(Of Vector(Of Double), Vector(Of Double), Vector(Of Double))
) As Func(Of Vector(Of Double), LinearOperator(Of Double), LinearOperator(Of Double))

public:
static Func<Vector<double>^, LinearOperator<double>^, LinearOperator<double>^>^ FromProduct(
	Action<Vector<double>^, Vector<double>^, Vector<double>^>^ multiply
)

static member FromProduct : 
        multiply : Action<Vector<float>, Vector<float>, Vector<float>> -> Func<Vector<float>, LinearOperator<float>, LinearOperator<float>> 

Parameters

multiply  Action<Vector<Double>, Vector<Double>, Vector<Double>>

A callback receiving the point, direction and destination. Write every component of the objective Hessian times the direction into the destination.

Return Value

Each evaluation copies the point into a protected snapshot and returns a new operator. The previous-operator argument is ignored. An earlier operator remains valid after later evaluations or changes to the original point.

The callback must represent a real symmetric Hessian. Transpose and conjugate transpose products use the same callback; symmetry is a caller obligation and is not checked. Use the original objective sign, including when maximizing.

The point and direction are read-only inputs. The destination belongs to the operator's caller; fully overwrite it and do not retain it. A destination is allocated when none is supplied. The adapter separates a direction and destination that are the same vector instance. Callback exceptions propagate. Concurrent products require a thread-safe callback.

The operator provides multiplication only. It supplies no explicit matrix entries, direct solves or sparse preconditioner. Use a matrix Hessian when these are needed.

Reference