Custom solvers

In this example, we show how to define custom solvers. Our system will again be silicon, because we are not very imaginative

using DFTK, LinearAlgebra

a = 10.26
lattice = a / 2 * [[0 1 1.];
                   [1 0 1.];
                   [1 1 0.]]
Si = ElementPsp(:Si, psp=load_psp("hgh/lda/Si-q4"))
atoms = [Si => [ones(3)/8, -ones(3)/8]]

# We take very (very) crude parameters
model = model_LDA(lattice, atoms)
kgrid = [1, 1, 1]
Ecut = 5
basis = PlaneWaveBasis(model, Ecut; kgrid=kgrid);

We define our custom fix-point solver: simply a damped fixed-point

function my_fp_solver(f, x0, max_iter; tol)
    mixing_factor = .7
    x = x0
    fx = f(x)
    for n = 1:max_iter
        inc = fx - x
        if norm(inc) < tol
            break
        end
        x = x + mixing_factor * inc
        fx = f(x)
    end
    (fixpoint=x, converged=norm(fx-x) < tol)
end;

Our eigenvalue solver just forms the dense matrix and diagonalizes it explicitly (this only works for very small systems)

function my_eig_solver(A, X0; maxiter, tol, kwargs...)
    n = size(X0, 2)
    A = Array(A)
    E = eigen(A)
    λ = E.values[1:n]
    X = E.vectors[:, 1:n]
    (λ=λ, X=X, residual_norms=[], iterations=0, converged=true, n_matvec=0)
end;

Finally we also define our custom mixing scheme. It will be a mixture of simple mixing (for the first 2 steps) and than default to Kerker mixing. In the mixing interface δF is $(ρ_\text{out} - ρ_\text{in})$, i.e. the difference in density between two subsequent SCF steps and the mix function returns $δρ$, which is added to $ρ_\text{in}$ to yield $ρ_\text{next}$, the density for the next SCF step.

struct MyMixing
    α  # Damping parameter
end
MyMixing() = MyMixing(0.7)

function DFTK.mix(mixing::MyMixing, basis, δF::RealFourierArray, δF_spin=nothing; n_iter, kwargs...)
    if n_iter <= 2
        # Just do simple mixing on total density and spin density (if it exists)
        (mixing.α * δF, isnothing(δF_spin) ? nothing : mixing.α * δF_spin)
    else
        # Use the KerkerMixing from DFTK
        DFTK.mix(KerkerMixing(α=mixing.α), basis, δF, δF_spin; kwargs...)
    end
end

That's it! Now we just run the SCF with these solvers

scfres = self_consistent_field(basis;
                               tol=1e-8,
                               solver=my_fp_solver,
                               eigensolver=my_eig_solver,
                               mixing=MyMixing());

n     Energy            Eₙ-Eₙ₋₁     ρout-ρin   Diag
---   ---------------   ---------   --------   ----
  1   -7.166864152879         NaN   3.62e-01    0.0
  2   -7.238616681087   -7.18e-02   1.83e-01    0.0
  3   -7.247822849851   -9.21e-03   7.74e-02    0.0
  4   -7.248791085965   -9.68e-04   4.22e-02    0.0
  5   -7.249087294750   -2.96e-04   2.33e-02    0.0
  6   -7.249176872284   -8.96e-05   1.31e-02    0.0
  7   -7.249204219070   -2.73e-05   7.46e-03    0.0
  8   -7.249212750601   -8.53e-06   4.32e-03    0.0
  9   -7.249215491475   -2.74e-06   2.54e-03    0.0
 10   -7.249216402278   -9.11e-07   1.52e-03    0.0
 11   -7.249216715866   -3.14e-07   9.19e-04    0.0
 12   -7.249216827647   -1.12e-07   5.62e-04    0.0
 13   -7.249216868788   -4.11e-08   3.47e-04    0.0
 14   -7.249216884363   -1.56e-08   2.16e-04    0.0
 15   -7.249216890400   -6.04e-09   1.36e-04    0.0

Note that the default convergence criterion is on the difference of energy from one step to the other; when this gets below tol, the "driver" self_consistent_field artificially makes the fixpoint solver think it's converged by forcing f(x) = x. You can customize this with the is_converged keyword argument to self_consistent_field.