Local-Global Solver API

[termi-official]

Feature List

• Local variable handler
• Custom quadrature rules via semidiscretize
• MultiLevelNewtonRaphson
• Stiff DAE solver baseline (via wrapper API)

TODOs

• Fix naming inconsistencies between local and internal
• Provide caches for the local solve
• Communication of convergence information between outer and inner solver
• Streamline multilevel setup Don't know how.

[termi-official]

Found some snippets. Probably part of this PR.

# Ferrite-specific
struct MaterialCache{M, DH, CV, FV}
    material::M
    dh::DH
    cv::CV
    fv::FV
end

# Standard assembly loop for global problem
function stress!(du, u, q, cache::MaterialCache, t)
    K = allocate_matrix(cache.dh)
    f = zeros(ndofs(cache.dh))
    # Allocate the element stiffness matrix
    n_basefuncs = getnbasefunctions(cache.cv)
    ke = zeros(n_basefuncs, n_basefuncs)
    fe = zeros(n_basefuncs)
    # Create an assembler
    assembler = start_assemble(K, f)
    # Loop over all cells
    for cell in CellIterator(cache.dh)
        # Update the shape function gradients based on the cell coordinates
        reinit!(cache.cv, cell)
        # Reset the element stiffness matrix
        fill!(ke, 0.0)
        fill!(fe, 0.0)
        # Compute element contribution
        assemble_cell!(ke, fe, u, q, cell, cache.cv, material)
        # Assemble ke into K
        assemble!(assembler, celldofs(cell), ke, fe)
    end
    for face in FacetIterator(cache.dh, getfacetset(cache.dh.grid,"right"))
        # Update the facetvalues to the correct facet number
        reinit!(cache.fv, face)
        # Reset the temporary array for the next facet
        fill!(fe, 0.0)
        # Access the cell's coordinates
        cell_coordinates = getcoordinates(face)
        for qp in 1:getnquadpoints(cache.fv)
            # Calculate the global coordinate of the quadrature point.
            x = spatial_coordinate(cache.fv, qp, cell_coordinates)
            tₚ = cache.material.traction(x, t)
            # Get the integration weight for the current quadrature point.
            dΓ = getdetJdV(cache.fv, qp)
            for i in 1:getnbasefunctions(cache.fv)
                Nᵢ = shape_value(cache.fv, qp, i)
                fe[i] += tₚ * Nᵢ * dΓ
            end
        end
        # Add the local contributions to the correct indices in the global external force vector
        assemble!(f, celldofs(face), fe)
    end
    du .= K*u .- f
end

function stress_Ju!(K, u, q, cache::MaterialCache, t)
    # Allocate the element stiffness matrix
    n_basefuncs = getnbasefunctions(cache.cv)
    ke = zeros(n_basefuncs, n_basefuncs)
    # Create an assembler
    assembler = start_assemble(K; fillzero=false)
    # Loop over all cells
    for cell in CellIterator(dh)
        # Update the shape function gradients based on the cell coordinates
        reinit!(cache.cv, cell)
        # Reset the element stiffness matrix
        fill!(ke, 0.0)
        # Compute element contribution
        assemble_cell_Ju!(ke, u, q, cell, cache.cv, cache.material)
        # Assemble ke into K
        assemble!(assembler, celldofs(cell), ke)
    end
end

function stress_Jq!(K, u, q, cache::MaterialCache, t)
    # Allocate the element stiffness matrix
    n_basefuncs = getnbasefunctions(cache.cv)
    ke = zeros(n_basefuncs, n_basefuncs)
    # Create an assembler
    assembler = start_assemble(K)
    # Loop over all cells
    for cell in CellIterator(dh)
        # Update the shape function gradients based on the cell coordinates
        reinit!(cache.cv, cell)
        # Reset the element stiffness matrix
        fill!(ke, 0.0)
        # Compute element contribution
        assemble_cell_Jq!(ke, u, q, cell, cache.cv, cache.material)
        # Assemble ke into K
        assemble!(assembler, celldofs(cell), ke)
    end
end

[Copilot]

Pull Request Overview

This PR introduces documentation and API reference updates for the new Local-Global Solver API along with dependency updates. Key changes include:

• Addition of a new section on Thunderbolt operators in the topics/operators.md file.
• Creation of detailed nonlinear solver documentation describing the Multi-Level Newton-Raphson method.
• Updating the API reference to include the MultiLevelNewtonRaphsonSolver.
• Numerous dependency version bumps and git-tree-sha1 updates in Manifest.toml files.

Reviewed Changes

Copilot reviewed 19 out of 34 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
docs/src/topics/operators.md	Added a new TLDT section with descriptive text on Thunderbolt operators.
docs/src/topics/nonlinear-solver.md	New comprehensive documentation for the Multi-Level Newton-Raphson method.
docs/src/api-reference/solver.md	Updated API reference to include MultiLevelNewtonRaphsonSolver.
docs/Manifest.toml	Dependency version and hash updates.
bak/examples/Manifest.toml	Similar dependency updates as in the main docs Manifest.

Files not reviewed (15)

• bak/examples/conduction-velocity-benchmark.jl: Language not supported
• benchmarks/benchmarks-multilevel.jl: Language not supported
• docs/make.jl: Language not supported
• docs/src/assets/references.bib: Language not supported
• docs/src/literate-tutorials/cm01_simple-active-stress.jl: Language not supported
• docs/src/literate-tutorials/cm03_3d0d-coupling.jl: Language not supported
• src/Thunderbolt.jl: Language not supported
• src/discretization/fem.jl: Language not supported
• src/ferrite-addons/InternalVariableHandler.jl: Language not supported
• src/ferrite-addons/transfer_operators.jl: Language not supported
• src/modeling/common.jl: Language not supported
• src/modeling/core/element_interface.jl: Language not supported
• src/modeling/coupler/tying.jl: Language not supported
• src/modeling/electrophysiology/ecg.jl: Language not supported
• src/modeling/functions.jl: Language not supported