Polarizability using automatic differentiation

Simple example for computing properties using (forward-mode) automatic differentiation. For a more classical approach and more details about computing polarizabilities, see Polarizability by linear response.

using DFTK
using LinearAlgebra
using ForwardDiff
using PseudoPotentialData

# Construct PlaneWaveBasis given a particular electric field strength
# Again we take the example of a Helium atom.
function make_basis(ε::T; a=10., Ecut=30) where {T}
    lattice = T(a) * I(3)  # lattice is a cube of ``a`` Bohrs
    # Helium at the center of the box
    pseudopotentials = PseudoFamily("cp2k.nc.sr.lda.v0_1.semicore.gth")
    atoms     = [ElementPsp(:He, pseudopotentials)]
    positions = [[1/2, 1/2, 1/2]]

    model = model_DFT(lattice, atoms, positions;
                      functionals=[:lda_x, :lda_c_vwn],
                      extra_terms=[ExternalFromReal(r -> -ε * (r[1] - a/2))],
                      symmetries=false)
    PlaneWaveBasis(model; Ecut, kgrid=[1, 1, 1])  # No k-point sampling on isolated system
end

# dipole moment of a given density (assuming the current geometry)
function dipole(basis, ρ)
    @assert isdiag(basis.model.lattice)
    a  = basis.model.lattice[1, 1]
    rr = [a * (r[1] - 1/2) for r in r_vectors(basis)]
    sum(rr .* ρ) * basis.dvol
end

# Function to compute the dipole for a given field strength
function compute_dipole(ε; tol=1e-8, kwargs...)
    scfres = self_consistent_field(make_basis(ε; kwargs...); tol)
    dipole(scfres.basis, scfres.ρ)
end;

With this in place we can compute the polarizability from finite differences (just like in the previous example):

polarizability_fd = let
    ε = 0.01
    (compute_dipole(ε) - compute_dipole(0.0)) / ε
end

1.7735581438063395

We do the same thing using automatic differentiation. Under the hood this uses custom rules to implicitly differentiate through the self-consistent field fixed-point problem. This leads to a density-functional perturbation theory problem, which is automatically set up and solved in the background.

polarizability = ForwardDiff.derivative(compute_dipole, 0.0)
println()
println("Polarizability via ForwardDiff:       $polarizability")
println("Polarizability via finite difference: $polarizability_fd")

n     Energy            log10(ΔE)   log10(Δρ)   Diag   Δtime
---   ---------------   ---------   ---------   ----   ------
  1   -2.770781111360                   -0.52    9.0    248ms
  2   -2.772061243184       -2.89       -1.32    1.0   98.6ms
  3   -2.772082952893       -4.66       -2.42    1.0    132ms
  4   -2.772083347686       -6.40       -3.14    1.0    102ms
  5   -2.772083417759       -7.15       -4.68    2.0    124ms
  6   -2.772083417780      -10.67       -4.76    2.0    126ms
  7   -2.772083417810      -10.52       -5.93    1.0    113ms
  8   -2.772083417811      -12.38       -6.95    2.0    129ms
  9   -2.772083417811      -13.88       -7.67    2.0    127ms
 10   -2.772083417811   +  -14.18       -8.03    1.0    113ms
Solving response problem
Iter  Restart  Krydim  log10(res)  avg(CG)  Δtime   Comment
----  -------  ------  ----------  -------  ------  ---------------
                                      13.0  60.4ms  Non-interacting
   1        0       1       -0.60     10.0   1.40s
   2        0       2       -2.42      8.0   110ms
   3        0       3       -3.55      5.0   120ms
   4        0       4       -5.33      4.0  94.7ms
   5        0       5       -7.76      1.0  83.4ms
   6        0       6      -10.09      1.0  90.1ms
   7        1       1       -7.37     13.0   213ms  Restart
   8        1       2       -8.94      1.0  82.6ms
                                      13.0   116ms  Final orbitals

Polarizability via ForwardDiff:       1.7725349859332262
Polarizability via finite difference: 1.7735581438063395