Merge pull request #11 from JuliaComputing/myb/newtonad

Scalar nonlinear solve AD

Author: YingboMa

Project.toml

@@ -20,6 +20,7 @@ Reexport = "0.2"
 Setfield = "0.7"
 StaticArrays = "0.11, 0.12"
 UnPack = "0.1, 1.0"
+julia = "1"
 
 [extras]
 BenchmarkTools = "6e4b80f9-dd63-53aa-95a3-0cdb28fa8baf"

src/NonlinearSolve.jl

@@ -3,6 +3,7 @@ module NonlinearSolve
   using Reexport
   using UnPack: @unpack
   using FiniteDiff, ForwardDiff
+  using ForwardDiff: Dual
   using Setfield
   using StaticArrays
   using RecursiveArrayTools

src/scalar.jl

@@ -19,6 +19,51 @@ function solve(prob::NonlinearProblem{<:Number}, ::NewtonRaphson, args...; xatol
   return NewtonSolution(x, MAXITERS_EXCEED)
 end
 
+function scalar_nlsolve_ad(prob, alg, args...; kwargs...)
+  f = prob.f
+  p = value(prob.p)
+  u0 = value(prob.u0)
+
+  newprob = NonlinearProblem(f, u0, p; prob.kwargs...)
+  sol = solve(newprob, alg, args...; kwargs...)
+
+  uu = getsolution(sol)
+  if p isa Number
+    f_p = ForwardDiff.derivative(Base.Fix1(f, uu), p)
+  else
+    f_p = ForwardDiff.gradient(Base.Fix1(f, uu), p)
+  end
+
+  f_x = ForwardDiff.derivative(Base.Fix2(f, p), uu)
+  pp = prob.p
+  sumfun = let f_x′ = -f_x
+    ((fp, p),) -> (fp / f_x′) * ForwardDiff.partials(p)
+  end
+  partials = sum(sumfun, zip(f_p, pp))
+  return sol, partials
+end
+
+function solve(prob::NonlinearProblem{<:Number, iip, <:Dual{T,V,P}}, alg::NewtonRaphson, args...; kwargs...) where {iip, T, V, P}
+  sol, partials = scalar_nlsolve_ad(prob, alg, args...; kwargs...)
+  return NewtonSolution(Dual{T,V,P}(sol.u, partials), sol.retcode)
+end
+function solve(prob::NonlinearProblem{<:Number, iip, <:AbstractArray{<:Dual{T,V,P}}}, alg::NewtonRaphson, args...; kwargs...) where {iip, T, V, P}
+  sol, partials = scalar_nlsolve_ad(prob, alg, args...; kwargs...)
+  return NewtonSolution(Dual{T,V,P}(sol.u, partials), sol.retcode)
+end
+
+# avoid ambiguities
+for Alg in [Bisection, Falsi]
+  @eval function solve(prob::NonlinearProblem{uType, iip, <:Dual{T,V,P}}, alg::$Alg, args...; kwargs...) where {uType, iip, T, V, P}
+    sol, partials = scalar_nlsolve_ad(prob, alg, args...; kwargs...)
+    return BracketingSolution(Dual{T,V,P}(sol.left, partials), Dual{T,V,P}(sol.right, partials), sol.retcode)
+  end
+  @eval function solve(prob::NonlinearProblem{uType, iip, <:AbstractArray{<:Dual{T,V,P}}}, alg::$Alg, args...; kwargs...) where {uType, iip, T, V, P}
+    sol, partials = scalar_nlsolve_ad(prob, alg, args...; kwargs...)
+    return BracketingSolution(Dual{T,V,P}(sol.left, partials), Dual{T,V,P}(sol.right, partials), sol.retcode)
+  end
+end
+
 function solve(prob::NonlinearProblem, ::Bisection, args...; maxiters = 1000, kwargs...)
   f = Base.Fix2(prob.f, prob.p)
   left, right = prob.u0

src/types.jl

@@ -78,3 +78,6 @@ function sync_residuals!(solver::BracketingImmutableSolver)
     @set! solver.fr = solver.f(solver.right, solver.p)
     solver
 end
+
+getsolution(sol::NewtonSolution) = sol.u
+getsolution(sol::BracketingSolution) = sol.left

src/utils.jl

@@ -216,7 +216,7 @@ Move `x` one floating point towards x0.
 function prevfloat_tdir(x::T, x0::T, x1::T)::T where {T}
   x1 > x0 ? prevfloat(x) : nextfloat(x)
 end
-  
+
 function nextfloat_tdir(x::T, x0::T, x1::T)::T where {T}
   x1 > x0 ? nextfloat(x) : prevfloat(x)
 end
@@ -234,3 +234,7 @@ function value_derivative(f::F, x::R) where {F,R}
     out = f(ForwardDiff.Dual{T}(x, one(x)))
     ForwardDiff.value(out), ForwardDiff.extract_derivative(T, out)
 end
+
+value(x) = x
+value(x::Dual) = ForwardDiff.value(x)
+value(x::AbstractArray{<:Dual}) = map(ForwardDiff.value, x)

test/runtests.jl

@@ -57,18 +57,41 @@ end
 f, u0 = (u, p) -> u * u - p, 1.0
 
 g = function (p)
-    probN = NonlinearProblem{false}(f, u0, p)
+    probN = NonlinearProblem{false}(f, oftype(p, u0), p)
     sol = solve(probN, NewtonRaphson())
     return sol.u
 end
 
-@test_broken ForwardDiff.derivative(g, 1.0) ≈ 0.5
+@test ForwardDiff.derivative(g, 1.0) ≈ 0.5
 
 for p in 1.1:0.1:100.0
     @test g(p) ≈ sqrt(p)
     @test ForwardDiff.derivative(g, p) ≈ 1/(2*sqrt(p))
 end
 
+f, u0 = (u, p) -> p[1] * u * u - p[2], (1.0, 100.0)
+t = (p) -> [sqrt(p[2] / p[1])]
+p = [0.9, 50.0]
+for alg in [Bisection(), Falsi()]
+    global g, p
+    g = function (p)
+        probN = NonlinearProblem{false}(f, u0, p)
+        sol = solve(probN, Bisection())
+        return [sol.left]
+    end
+
+    @test g(p) ≈ [sqrt(p[2] / p[1])]
+    @test ForwardDiff.jacobian(g, p) ≈ ForwardDiff.jacobian(t, p)
+end
+
+gnewton = function (p)
+    probN = NonlinearProblem{false}(f, 0.5, p)
+    sol = solve(probN, NewtonRaphson())
+    return [sol.u]
+end
+@test gnewton(p) ≈ [sqrt(p[2] / p[1])]
+@test ForwardDiff.jacobian(gnewton, p) ≈ ForwardDiff.jacobian(t, p)
+
 # Error Checks
 
 f, u0 = (u, p) -> u .* u .- 2.0, @SVector[1.0, 1.0]