Burgers Pseudospectral Methods Work-Precision Diagrams

HAO HAO
using ApproxFun, OrdinaryDiffEq, Sundials
using DiffEqDevTools
using LinearAlgebra
using Plots; gr()

Plots.GRBackend()

Here is the Burgers equation using Fourier spectral methods.

S = Fourier()
n = 512
x = points(S, n)
D2 = Derivative(S,2)[1:n,1:n]
D  = (Derivative(S) → S)[1:n,1:n]
T = ApproxFun.plan_transform(S, n)
Ti = ApproxFun.plan_itransform(S, n)

û₀ = T*cos.(cos.(x.-0.1))
A = 0.03*D2
tmp = similar(û₀)
p = (D,D2,T,Ti,tmp,similar(tmp))
function burgers_nl(dû,û,p,t)
    D,D2,T,Ti,u,tmp = p
    mul!(tmp, D, û)
    mul!(u, Ti, tmp)
    mul!(tmp, Ti, û)
    @. tmp = tmp*u
    mul!(u, T, tmp)
    @. dû = - u
end

burgers_nl (generic function with 1 method)

Reference solution using Rodas5 is below:

prob = SplitODEProblem(DiffEqArrayOperator(Diagonal(A)), burgers_nl, û₀, (0.0,5.0), p)
sol  = solve(prob, Rodas5(autodiff=false); reltol=1e-12,abstol=1e-12)
test_sol = TestSolution(sol)

tslices=[0.0 1.0 2.0 3.0 5.0]
ys=hcat((Ti*sol(t) for t in tslices)...)
labels=["t=$t" for t in tslices]
plot(x,ys,label=labels)

High tolerances

diag_linsolve=LinSolveFactorize(W->let tmp = tmp
    for i in 1:size(W, 1)
        tmp[i] = W[i, i]
    end
    Diagonal(tmp)
end)

DiffEqBase.LinSolveFactorize{Main.##WeaveSandBox#253.var"#5#6"}(Main.##Weav
eSandBox#253.var"#5#6"(), nothing)

In-family comparisons

1.IMEX methods (diagonal linear solver)

abstols = 0.1 .^ (5:8)
reltols = 0.1 .^ (1:4)
multipliers =  0.5 .^ (0:3)
setups = [Dict(:alg => IMEXEuler(linsolve=diag_linsolve), :dts => 1e-3 * multipliers),
          Dict(:alg => CNAB2(linsolve=diag_linsolve), :dts => 5e-3 * multipliers),
          Dict(:alg => CNLF2(linsolve=diag_linsolve), :dts => 5e-3 * multipliers),
          Dict(:alg => SBDF2(linsolve=diag_linsolve), :dts => 1e-3 * multipliers)]
labels = ["IMEXEuler" "CNAB2" "CNLF2" "SBDF2"]
@time wp1 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true,names=labels,
                            numruns=5,seconds=5,
                            save_everystop=false,appxsol=test_sol,maxiters=Int(1e5));

IMEXEuler

ERROR: LinearAlgebra.SingularException(504)

plot(wp1,label=labels,markershape=:auto,title="IMEX methods, diagonal linsolve, low order")

ERROR: UndefVarError: wp1 not defined

  1. ExpRK methods

abstols = 0.1 .^ (5:8) # all fixed dt methods so these don't matter much
reltols = 0.1 .^ (1:4)
multipliers = 0.5 .^ (0:3)
setups = [Dict(:alg => NorsettEuler(), :dts => 1e-3 * multipliers),
          Dict(:alg => ETDRK2(), :dts => 1e-2 * multipliers)]
labels = hcat("NorsettEuler",
              "ETDRK2 (caching)")
@time wp2 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true, names=labels,
                            numruns=5,
                            save_everystep=false, appxsol=test_sol, maxiters=Int(1e5));

NorsettEuler
ETDRK2 (caching)
164.590325 seconds (101.93 M allocations: 7.123 GiB, 1.34% gc time)

plot(wp2, label=labels, markershape=:auto, title="ExpRK methods, low order")

Between family comparisons

abstols = 0.1 .^ (5:8) # all fixed dt methods so these don't matter much
reltols = 0.1 .^ (1:4)
multipliers = 0.5 .^ (0:3)
setups = [Dict(:alg => CNAB2(linsolve=diag_linsolve), :dts => 5e-3 * multipliers),
          Dict(:alg => ETDRK2(), :dts => 1e-2 * multipliers)]
labels = ["CNAB2 (diagonal linsolve)" "ETDRK2"]
@time wp3 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true, names=labels,
                            numruns=5, error_estimate=:l2,
                            save_everystep=false, appxsol=test_sol, maxiters=Int(1e5));

CNAB2 (diagonal linsolve)

ERROR: LinearAlgebra.SingularException(264)

plot(wp3, label=labels, markershape=:auto, title="Between family, low orders")

ERROR: UndefVarError: wp3 not defined

Low tolerances

In-family comparisons

1.IMEX methods (band linear solver)

abstols = 0.1 .^ (7:13)
reltols = 0.1 .^ (4:10)
setups = [Dict(:alg => ARKODE(Sundials.Implicit(), order=3, linear_solver=:Band, jac_upper=1, jac_lower=1)),
          Dict(:alg => ARKODE(Sundials.Implicit(), order=4, linear_solver=:Band, jac_upper=1, jac_lower=1)),
          Dict(:alg => ARKODE(Sundials.Implicit(), order=5, linear_solver=:Band, jac_upper=1, jac_lower=1))]
labels = hcat("ARKODE3", "ARKODE4", "ARKODE5")
@time wp4 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true, names=labels,
                            numruns=5, error_estimate=:l2,
                            save_everystep=false, appxsol=test_sol, maxiters=Int(1e5));

ARKODE3
ARKODE4
ARKODE5
329.881717 seconds (115.35 M allocations: 22.295 GiB, 1.38% gc time)

plot(wp4, label=labels, markershape=:auto, title="IMEX methods, band linsolve, medium order")

2.ExpRK methods

abstols = 0.1 .^ (7:11) # all fixed dt methods so these don't matter much
reltols = 0.1 .^ (4:8)
multipliers = 0.5 .^ (0:4)
setups = [Dict(:alg => ETDRK3(), :dts => 1e-2 * multipliers),
          Dict(:alg => ETDRK4(), :dts => 1e-2 * multipliers),
          Dict(:alg => HochOst4(), :dts => 1e-2 * multipliers)]
labels = hcat("ETDRK3 (caching)", "ETDRK4 (caching)",
              "HochOst4 (caching)")
@time wp5 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true, names=labels,
                            numruns=5, error_estimate=:l2,
                            save_everystep=false, appxsol=test_sol, maxiters=Int(1e5));

ETDRK3 (caching)
ETDRK4 (caching)
HochOst4 (caching)
347.360721 seconds (105.81 M allocations: 8.976 GiB, 0.95% gc time)

plot(wp5, label=labels, markershape=:auto, title="ExpRK methods, medium order")

Between family comparisons

abstols = 0.1 .^ (7:11)
reltols = 0.1 .^ (4:8)
multipliers = 0.5 .^ (0:4)
setups = [Dict(:alg => ARKODE(Sundials.Implicit(), order=5, linear_solver=:Band, jac_upper=1, jac_lower=1)),
          Dict(:alg => ETDRK3(), :dts => 1e-2 * multipliers),
          Dict(:alg => ETDRK4(), :dts => 1e-2 * multipliers)]
labels = hcat("ARKODE (nondiagonal linsolve)", "ETDRK3 ()", "ETDRK4 ()")
                        #"ARKODE (Krylov linsolve)")
@time wp6 = WorkPrecisionSet(prob,abstols,reltols,setups;
                            print_names=true, names=labels,
                            numruns=5, error_estimate=:l2,
                            save_everystep=false, appxsol=test_sol, maxiters=Int(1e5));

ARKODE (nondiagonal linsolve)
ETDRK3 ()
ETDRK4 ()
241.722719 seconds (42.80 M allocations: 4.273 GiB, 0.75% gc time)

plot(wp6, label=labels, markershape=:auto, title="Between family, medium order")

using SciMLBenchmarks
SciMLBenchmarks.bench_footer(WEAVE_ARGS[:folder],WEAVE_ARGS[:file])

Appendix

These benchmarks are a part of the SciMLBenchmarks.jl repository, found at: https://github.com/SciML/SciMLBenchmarks.jl. For more information on high-performance scientific machine learning, check out the SciML Open Source Software Organization https://sciml.ai.

To locally run this benchmark, do the following commands:

using SciMLBenchmarks
SciMLBenchmarks.weave_file("MOLPDE","burgers_spectral_wpd.jmd")

Computer Information:

Julia Version 1.4.2
Commit 44fa15b150* (2020-05-23 18:35 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i7-9700K CPU @ 3.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-8.0.1 (ORCJIT, skylake)
Environment:
  JULIA_LOAD_PATH = /builds/JuliaGPU/DiffEqBenchmarks.jl:
  JULIA_DEPOT_PATH = /builds/JuliaGPU/DiffEqBenchmarks.jl/.julia
  JULIA_CUDA_MEMORY_LIMIT = 2147483648
  JULIA_NUM_THREADS = 8

Package Information:

Status: `/builds/JuliaGPU/DiffEqBenchmarks.jl/benchmarks/MOLPDE/Project.toml`
[28f2ccd6-bb30-5033-b560-165f7b14dc2f] ApproxFun 0.11.14
[f3b72e0c-5b89-59e1-b016-84e28bfd966d] DiffEqDevTools 2.24.0
[7f56f5a3-f504-529b-bc02-0b1fe5e64312] LSODA 0.6.1
[09606e27-ecf5-54fc-bb29-004bd9f985bf] ODEInterfaceDiffEq 3.7.0
[1dea7af3-3e70-54e6-95c3-0bf5283fa5ed] OrdinaryDiffEq 5.41.0
[91a5bcdd-55d7-5caf-9e0b-520d859cae80] Plots 1.5.6
[c3572dad-4567-51f8-b174-8c6c989267f4] Sundials 4.2.5
[37e2e46d-f89d-539d-b4ee-838fcccc9c8e] LinearAlgebra

Published from burgers_spectral_wpd.jmd using Weave.jl v0.10.2 on 2020-07-25.