EulerSolver(equation::MemoryEquation; t_max=10.0^2, Δt=10^-3, verbose=false)

Constructs a solver object that, when called solve upon will solve an MemoryEquation using a forward Euler method. It will discretise the integral using a Trapezoidal rule. Use this solver only for testing purposes. It is a wildy inefficient way to solve MCT-like equations.

Arguments:

• equation an instance of MemoryEquation
• t_max when this time value is reached, the integration returns
• Δt fixed time step
• verbose if true, information will be printed to STDOUT

ExponentiallyDecayingKernel{T<:Number} <: MemoryKernel

Scalar kernel with fields ν and τ which when called returns ν exp(-t/τ).

InterpolatingKernel(t, M; k=1)

Uses the package Dierckx to provide a kernel that interpolates the data M defined on grid t using spline interpolation of degree k.

MemoryEquation(α, β, γ, F₀::T, ∂ₜF₀::T, kernel::MemoryKernel) where T

Arguments:

• α: coefficient in front of the second derivative term. If α and F₀ are both vectors, α will automatically be converted to a diagonal matrix, to make them compatible.
• β: coefficient in front of the first derivative term. If β and F₀ are both vectors, β will automatically be converted to a diagonal matrix, to make them compatible.
• γ: coefficient in front of the second derivative term. If γ and F₀ are both vectors, γ will automatically be converted to a diagonal matrix, to make them compatible.
• δ: coefficient in front of the constant term. δ and F0 must by types that can be added together. If δ is a number and F0 is a vector, this conversion will be done automatically.
• F₀: initial condition of F(t)
• ∂ₜF₀ initial condition of the derivative of F(t)
• kernel instance of a MemoryKernel that when called on F₀ and t=0, evaluates to the initial condition of the memory kernel.

MemoryEquationCoefficients(a, b, c, d, F₀)

Constructor of the MemoryEquationCoefficients struct, which holds the values of the coefficients α, β, γ and δ. It will convert common cases automatically to make the types compatible. For example, if α and F0 are both given as vectors, α is converted to a Diagonal matrix.

MemoryEquationSolution

solution object that holds the solution of an AbstractMemoryEquation. It has 4 fields t: array of t values F: array of F for all t K: array of K for all T solver: solver object that holds the solver settings

an MCT object can be indexed such that sol=MemoryEquationSolution sol[2] gives the F[2] for all t.

SCGLEKernel(ϕ, k_array, S_array)

Constructor of a SCGLEKernel. It implements the kernel K(k,t) = λ(k)Δζ(t) where Δζ(t) = kBT / (144 π^2 ϕ) ∫dq⃗ M^2(k,q) Fs(q,t) F(q,t) in which k and q are vectors. For Fs and F being the self- and collective-intermediate scattering function.

Arguments:

• ϕ: volume fraction
• k_array: vector of wavenumbers at which the structure factor is known
• S_array: concatenated vector for ones(length(structure factor)) and structure factor

Returns:

an instance k of SCGLEKernel <: MemoryKernel, which can be called both in-place and out-of-place: evaluate_kernel!(out, kernel, F, t) out = evaluate_kernel(kernel, F, t)

SchematicDiagonalKernel{T<:Union{SVector, Vector}} <: MemoryKernel

Matrix kernel with field ν which when called returns Diagonal(ν .* F .^ 2), i.e., it implements a non-coupled system of SchematicF2Kernels.

SchematicF1Kernel{T<:Number} <: MemoryKernel

Scalar kernel with fields ν1, ν2, and ν3 which when called returns ν1 * F^1 + ν2 * F^2 + ν3 * F^3.

SchematicF1Kernel{T<:Number} <: MemoryKernel

Scalar kernel with field ν which when called returns ν F.

SchematicF2Kernel{T<:Number} <: MemoryKernel

Scalar kernel with field ν which when called returns ν F^2.

SchematicMatrixKernel{T<:Union{SVector, Vector}} <: MemoryKernel

Matrix kernel with field ν which when called returns ν * F * Fᵀ, i.e., it implements Kαβ = νFαFβ.

SjogrenKernel

Memory kernel that implements the kernel K[1] = ν1 F[1]^2, K[2] = ν2 F[1] F[2]. Consider using Static Vectors for performance.

SchematicF2Kernel{T<:Number} <: MemoryKernel

Scalar tagged particle kernel with field ν which when called returns ν F*Fs.

TimeDoublingSolver(N=32, Δt=10^-10, t_max=10.0^10, max_iterations=10^4, tolerance=10^-10, verbose=false, ismutable=true, init_with_memory=true)

Uses the algorithm devised by Fuchs et al.

Arguments:

• equation: an instance of MemoryEquation
• N: The number of time points in the interval is equal to 4N
• t_max: when this time value is reached, the integration returns
• Δt: starting time step, this will be doubled repeatedly
• max_iterations: the maximal number of iterations before convergence is reached for each time doubling step
• tolerance: while the error is bigger than this value, convergence is not reached. The error by default is computed as the absolute sum of squares
• verbose: if true, information will be printed to STDOUT
• ismutable: if true and if the type of F is mutable, the solver will try to avoid allocating many temporaries
• init_with_memory if true, the solver will include the memory kernels for the short time expansion to bootstrap the algorithm. Sacrifices performance and memory for accuracy.

MSDModeCouplingKernel(ρ, kBT, m, k_array, Sₖ, sol, taggedsol; dims=3)

Constructor of a MSDModeCouplingKernel. It implements the kernel K(k,t) = ρ kBT / (6π^2m) ∫dq q^4 c(q)^2 F(q,t) Fs(q,t) where the integration runs from 0 to infinity. F and Fs are the coherent and incoherent intermediate scattering functions, and must be passed in as solutions of the corresponding equations.

Arguments:

• ρ: number density
• kBT: Thermal energy
• m : particle mass
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: vector with the elements of the structure factor
• sol: a solution object of an equation with a ModeCouplingKernel.
• taggedsol: a solution object of an equation with a TaggedModeCouplingKernel.

Returns:

an instance k of MSDModeCouplingKernel <: MemoryKernel, which can be evaluated like: k = evaluate_kernel(kernel, F, t)

MSDMultiComponentModeCouplingKernel(s, ρ, kBT, m, k_array, Sₖ, sol, taggedsol; dims=3)

Constructor of a MSDModeCouplingKernel. It implements the kernel Kₛ(k,t) = ρ kBT / (6π^2mₛ) Σαβ ∫dq q^4 csα(q)csβ(q) Fαβ(q,t) Fₛ(q,t) where the integration runs from 0 to infinity. F and Fs are the coherent and incoherent intermediate scattering functions, and must be passed in as solutions of the corresponding equations.

Arguments:

• s: the species for which to solve this equation
• ρ: number density
• kBT: Thermal energy
• m : particle mass of species s
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: vector with the elements of the structure factor
• sol: a solution object of an equation with a MultiComponentModeCouplingKernel.
• taggedsol: a solution object of an equation with a TaggedMultiComponentModeCouplingKernel.

Returns:

an instance k of MSDMultiComponentModeCouplingKernel <: MemoryKernel, which can be evaluated like: k = evaluate_kernel(kernel, F, t)

ModeCouplingKernel(ρ, kBT, m, k_array, Sₖ; dims=3)

Constructor of a ModeCouplingKernel. It implements the kernel K(k,t) = ρ kBT / (16π^3m) ∫dq V^2(k,q) F(q,t) F(k-q,t) in which k and q are vectors.

Arguments:

• ρ: number density
• kBT: Thermal energy
• m : particle mass
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: structure factor

Returns:

an instance k of ModeCouplingKernel <: MemoryKernel, which can be called both in-place and out-of-place: evaluate_kernel!(out, kernel, F, t) out = evaluate_kernel(kernel, F, t)

MultiComponentModeCouplingKernel(ρ, kBT, m, k_array, Sₖ; dims=3)

Constructor of a MultiComponentModeCouplingKernel. It implements the kernel Kαβ(k,t) = ρ / (2 xα xβ (2π)³) Σμνμ'ν' ∫dq Vμ'ν'(k,q) Fμμ'(q,t) Fνν'(k-q,t) Vμν(k,q) in which k and q are vectors and α and β species labels.

Arguments:

• ρ: vector of number densities for each species
• kBT: Thermal energy
• m : vector of particle masses for each species
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: a Vector of Nk SMatrixs containing the structure factor of each component at each wave number

Returns:

an instance k of ModeCouplingKernel <: MemoryKernel, which can be called both in-place and out-of-place: k = MultiComponentModeCouplingKernel(ρ, kBT, m, karray, Sₖ) `evaluatekernel!(out, kernel, F, t)out = evaluate_kernel(kernel, F, t)`

TaggedModeCouplingKernel(ρ, kBT, m, k_array, Sₖ, sol; dims=3)

Constructor of a Tagged ModeCouplingKernel. It implements the kernel K(k,t) = ρ kBT / (8π^3m) ∫dq V^2(k,q) F(q,t) Fs(k-q,t) in which k and q are vectors. Here V(k,q) = c(q) (k dot q)/k.

Arguments:

• ρ: number density
• kBT: Thermal energy
• m : particle mass
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: vector with the elements of the structure factor
• sol: a solution object of an equation with a ModeCouplingKernel.

Returns:

an instance k of TaggedModeCouplingKernel <: MemoryKernel, which can be called both in-place and out-of-place: evaluate_kernel!(out, kernel, Fs, t) out = evaluate_kernel(kernel, Fs, t)

TaggedMultiComponentModeCouplingKernel(s, ρ, kBT, m, k_array, Sₖ, sol; dims=3)

Constructor of a Tagged ModeCouplingKernel. It implements the kernel Kₛ(k,t) = ρ kBT / (8π^3 mₛ) Σαβ ∫dq V^2sαβ(k,q) Fαβ(q,t) Fₛ(k-q,t) in which k and q are vectors.

Arguments:

• s: the species for which to solve this equation
• ρ: number density
• kBT: Thermal energy
• m : particle mass
• k_array: vector of wavenumbers at which the structure factor is known
• Sₖ: vector with the elements of the structure factor
• sol: a solution object of an equation with a MultiComponentModeCouplingKernel.

Returns:

an instance k of TaggedMultiComponentModeCouplingKernel <: MemoryKernel, which can be called both in-place and out-of-place: evaluate_kernel!(out, kernel, Fs, t) out = evaluate_kernel(kernel, Fs, t)

allocate_results!(t_array, F_array, K_array, equation::AbstractMemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache; istart=2(solver.N)+1, iend=4(solver.N))

pushes the found solution, stored in temp_arrays with indices istart until istop to the output arrays.

allocate_temporary_arrays(equation::AbstractMemoryEquation, solver::TimeDoublingSolver)

Returns a SolverCache containing several arrays that are used for intermediate calculations.

computec3immutable(equation::MemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache, it::Int)

computes one of the matrices necessary to propagate F in time.

computec3mutable(equation::MemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache, it::Int)

computes one of the matrices necessary to propagate F in time, by mutating it inplace. The commented code is the corresponding scalar equivalent

`convert_multicomponent_structure_factor(Sk_in::Matrix{Vector{T}})``

This function takes a matrix of vectors Sk_in where each vector contains structure factor information for different species at different k-points and converts it into a vector of SMatrix.

Input:

• Sk_in::Matrix{Vector{T}} - A square matrix of vectors, where each vector contains structure factor information for a specific set of species at different k-points.

Output:

• Sk_out - A Vector of SMatrix where each SMatrix contains structure factor information for different species at a single k-point.

Example

Ns = 2;
Nk = 3;
S11 = [1,2,3]; S21 = [4,5,6]; S22 = [8,9,10];
S = [zeros(Nk) for i=1:2, j=1:2];
S[1,1] = S11; S[1,2] = S21; S[2,1] = S21; S[2,2] = S22;
convert_multicomponent_structure_factor(S)

    3-element Vector{StaticArraysCore.SMatrix{2, 2, Float64, 4}}:
    [1.0 4.0; 4.0 8.0]
    [2.0 5.0; 5.0 9.0]
    [3.0 6.0; 6.0 10.0]

do_time_steps!(equation::MemoryEquation, solver::TimeDoublingSolver, kernel::MemoryKernel, temp_arrays::SolverCache)

Solves the equation on the time points with index 2N+1 until 4N, for each point doing a recursive iteration to find the solution to the nonlinear equation C1 F = -C2 M(F) + C3.

evaluate_kernel(kernel::MemoryKernel, F, t)

Evaluates the memory kernel out-place. It may mutate the content of kernel.

Returns

• out the kernel evaluated at (F, t)

evaluate_kernel!(out, kernel::MemoryKernel, F, t)

Evaluates the memory kernel in-place, overwriting the elements of the out variable. It may mutate the content of kernel.

find_error(F_new::T, F_old::T) where T

Finds the error between a new and old iteration of F. The returned scalar will be compared to the tolerance to establish convergence.

find_relaxation_time(t, F; threshold=exp(-1), mode=:log)

Finds the time at which a signal first decreases below some threshold. Uses interpolation to find an accurate result. if it never decreases below the threshold, returns Inf.

Arguments:

• t: a vector of time values at which the signal F is known
• F: the signal
• threshold: see above
• mode: uses interpolation either in a logarithmic or linear space.

`get_F(sol::MemoryEquationSolution)`

obtains the solution F from a MemoryEquationSolution object. Equivalent to sol.F.

`get_F(sol::MemoryEquationSolution, I...)`

obtains the solution F from a MemoryEquationSolution object and indexes into it. Enables convenient indexing into the multidimensional object. I is interpreted as a set of indices.

Examples:

If sol is the solution to a scalar equation get_F(sol, 2:4) gets the elements 2:4 If sol is the solution to a vector-valued equation get_F(sol, 5, 2:43) gets the solution at the 5th time point for vector indices 2:43. If sol is the solution to a vector-valued multicomponent equation get_F(sol, 5, 2:43, (1,2)) gets the solution at the 5th time point for vector indices 2:43, for species 1 and 2.

`get_K(sol::MemoryEquationSolution)`

obtains the kernel K from a MemoryEquationSolution object. Equivalent to sol.K.

`get_K(sol::MemoryEquationSolution, I...)`

obtains the solution K from a MemoryEquationSolution object and indexes into it. Enables convenient indexing into the multidimensional object. See get_F for examples.

`get_t(sol::MemoryEquationSolution)`

obtains the time grid t from a MemoryEquationSolution object. Equivalent to sol.t.

initialize_F_temp!(equation::MemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache)

Fills the first 2N entries of the temporary arrays of F using forward Euler without a memory kernel in order to kickstart the algorithm' scheme.

initialize_K_temp!(solver::TimeDoublingSolver, kernel::MemoryKernel, temp_arrays::SolverCache)

Evaluates the memory kernel at the first 2N time points.

initialize_integrals!(equation::AbstractMemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache)

Initializes the integrals on the first 2N time points as prescribed in the literature.

initialize_output_arrays(equation::AbstractMemoryEquation)

initializes arrays that the solver will push results into.

" mymul!(c,a,b,α,β)

prescribes how types used in this solver should be multiplied in place. In particular, it performs C.= βC .+ αa*b. defaults to mul!(c,a,b,α,β)

newtimemapping!(equation::AbstractMemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache)

Performs the time mapping central to Fuchs' algorithm with the conventions prescribed in "Flenner, Elijah, and Grzegorz Szamel. Physical Review E 72.3 (2005): 031508". Performs them inplace if solver.inplace = true in order to avoid unnecessary allocations.

solve(equation::AbstractMemoryEquation, solver::Solver)

Solves the AbstractMemoryEquation with the provided kernel using solver. Search for a specific solver or kernel object to find more specific information.

If no solver is provided, it uses the default TimeDoublingSolver.

Returns:

• t an array of time values
• F The solution in an array of which the last dimension corresponds to the time.
• K The memory kernel corresponding to each F

solve_steady_state(γ, F₀, kernel; tolerance=10^-8, max_iterations=10^6, verbose=false, inplace=true)

Finds the steady-state solution (non-ergodicity parameter) of the generalized Langevin equation by recursive iteration of F = (K + γ)⁻¹ * K(F) * F₀ This function assumes δ = 0, since that would lead to a divergence.

Arguments:

• γ: parameter in front of the linear term in F
• F₀: initial condition of F. This is also the initial condition of the rootfinding method.
• kernel: callable memory kernel
• max_iterations: the maximal number of iterations before convergence is reached
• tolerance: while the error is bigger than this value, convergence is not reached. The error by default is computed as the absolute sum of squares between successive iterations
• verbose: if true, information will be printed to STDOUT
• inplace: if true and if the type of F is mutable, the solver will try to avoid allocating many temporaries. default =true``

Returns:

• The steady state solution

Surface of a d-dimensional sphere

update_F!(solver::TimeDoublingSolver, temp_arrays::SolverCache, it::Int)

updates F using the formula c1F = -KC2 + C3.

update_Fuchs_parameters!(equation::MemoryEquation, solver::TimeDoublingSolver, temp_arrays::SolverCache, it::Int)

Updates the parameters c1, c2, c3, according to the appendix of "Flenner, Elijah, and Grzegorz Szamel. Physical Review E 72.3 (2005): 031508" using the naming conventions from that paper. If F is mutable (and therefore also c1,c2,c3), it will update the variables in place, otherwise it will create new copies. This is controlled by the solver.inplace setting.

update_K!(solver::TimeDoublingSolver, kernel::MemoryKernel, temp_arrays::SolverCache, it::Int)

evaluates the memory kernel, updating the value in solver.temparrays.Ktemp

update_integrals!(solver::TimeDoublingSolver, equation::AbstractMemoryEquation, temp_arrays::SolverCache, it::Int)

Update the discretisation of the integral of F and K, see the literature for details.

Volume of a d-dimensional sphere