GitHub

Fronts.ParamEstim module: parameter estimation support

The ParamEstim submodule provides support for optimization-based parameter estimation runs using Fronts.

Fronts.ParamEstim.ScaledSolution — Type
ScaledSolution

Wrapper for a solution scaled in o as if affecting the diffusivity by a constant factor D0.

Extra fields

  • original: original unscaled solution.
  • D0: scaling factor.
source
Fronts.ParamEstim.AbstractCostFunction — Type
abstract type AbstractCostFunction{fit_D0} end

Abstract cost function for parameter estimation.

Type parameters

  • fit_D0::Bool: whether to fit an additional constant factor D0 that affects the diffusivity. Values

of D0 can be found with relative efficiency without additional solver calls; so if any such constant factors affecting the diffusivity are unknown, it is recommended not to fit those factors directly but set fit_D0 to true instead. Values of D0 are found internally by local optimization. If true, the candidate function will return a ScaledSolution that includes the found value of D0.

source
Fronts.ParamEstim.RSSCostFunction — Type
RSSCostFunction{fit_D0}(func, prob::InverseProblem[; D0tol, oi_hint]) <: AbstractCostFunction

Residual sum of squares cost function for parameter estimation.

Type parameters

  • fit_D0::Bool: whether to fit an additional constant factor D0 that affects the diffusivity. Values

of D0 can be found with relative efficiency without additional solver calls; so if any such constant factors affecting the diffusivity are unknown, it is recommended not to fit those factors directly but set fit_D0 to true instead. Values of D0 are found internally by local optimization. If true, the candidate function will return a ScaledSolution that includes the found value of D0.

Arguments

  • func: function that takes a vector of parameter values and returns either a Fronts.Solution or a

Fronts.AbstractSemiinfiniteProblem. If func returns an AbstractSemiinfiniteProblem, it is solved with solve. Solutions with successful ReturnCode are passed to the cost function; otherwise, the cost is set to Inf.

Keyword arguments

  • D0tol=1e-3: if fit_D0 is true, a tolerance for D0.
  • oi_hint=nothing: if fit_D0 is true, an optional hint as to the point in o where the initial

condition begins. The hint will be used as an aid in finding the optimal value for D0. Otherwise, the fitting process will start by attempting to match sorptivities.

References

GERLERO, G. S.; BERLI, C. L. A.; KLER, P. A. Open-source high-performance software packages for direct and inverse solving of horizontal capillary flow. Capillarity, 2023, vol. 6, no. 2, p. 31-40.

See also: candidate, ScaledSolution, Fronts.Solution, Fronts.AbstractSemiinfiniteProblem

(::RSSCostFunction)(p::AbstractVector)

Return the cost of the solution obtained with parameter values p.

The RSSCostFunction object is meant to be passed to your optimizer of choice for minimization as the objective function.

If you need to know more than just the cost, call the candidate function instead.

source
Fronts.ParamEstim.candidate — Function
candidate(cf::AbstractCostFunction, ::AbstractVector)
candidate(cf::AbstractCostFunction, ::Fronts.AbstractSemiinfiniteProblem)
candidate(cf::AbstractCostFunction, ::Fronts.Solution)

Return the candidate solution for a given cost function and parameter values, problem, or solution.

source