MultiScaleReactionDiffusion<T>

Models & Types · AiDotNet.PhysicsInformed.PDEs

A two-scale reaction-diffusion equation: the same physics at a coarse and a fine length scale, with a coupling term that penalises the two disagreeing where they overlap.

Inheritance: PDESpecificationBase<T> → MultiScaleReactionDiffusion<T>

Implements: IMultiScalePDE<T>

For Beginners

Some systems do the same thing at two very different sizes at once — heat spreading through a whole engine block while also spreading within each grain of the metal. Modelling only the coarse scale misses the fine detail; modelling only the fine one needs impossibly many points. A multi-scale model runs both and asks them to agree.

How It Works

This is the reference IMultiScalePDE, and until it existed there was none: MultiScalePINN takes one in its constructor and the library implemented the interface zero times, so the type could not be constructed at all (#2105).

The residual at each scale is the heat/diffusion form ∂u/∂t − k ∂²u/∂x², with a different diffusivity per scale — a large one for the coarse field and a small one for the fine.

Example

var pde = new MultiScaleReactionDiffusion<double>(
    coarseDiffusivity: 1.0,
    fineDiffusivity: 0.01);
            
var architecture = new NeuralNetworkArchitecture<double>(
    InputType.OneDimensional, NeuralNetworkTaskType.Regression, inputSize: 2, outputSize: 1);
            
var pinn = new MultiScalePINN<double>(
    architecture, pde, new IBoundaryCondition<double>[] { new DirichletBoundaryCondition<double>() });

Constructors

Constructor Summary
MultiScaleReactionDiffusion(double coarseDiffusivity, double fineDiffusivity, double coarseLength, double fineLength, double fineScaleWeight) Initializes a two-scale reaction-diffusion equation.

Properties

Property Summary
T[] ScaleCharacteristicLengths { get; }
int InputDimension { get; }
int NumberOfScales { get; }
int OutputDimension { get; }
string Name { get; }

Methods

Method Summary
T ComputeResidual(Vector<T> inputs, Vector<T> outputs, PDEDerivatives<T> derivatives)
T GetScaleLossWeight(int scaleIndex)
Tensor<T> ComputeScaleCoupling(int coarseIndex, int fineIndex, Tensor<T> inputs, Tensor<T> coarseOutputs, Tensor<T> fineOutputs, PDEDerivatives<T> coarseDerivatives, PDEDerivatives<T> fineDerivatives)
Tensor<T> ComputeScaleResidual(int scaleIndex, Tensor<T> inputs, Tensor<T> outputs, PDEDerivatives<T> derivatives)
Tensor<T> ComputeTapeResidual(Tensor<T> inputs, Tensor<T> outputs, PDEDerivatives<T> derivatives)
int GetScaleOutputDimension(int scaleIndex)