Linearsolve3

Project.toml

@@ -1,10 +1,11 @@
 name = "VoronoiFVM"
 uuid = "82b139dc-5afc-11e9-35da-9b9bdfd336f3"
 authors = ["Jürgen Fuhrmann <[email protected]>", "Dilara Abdel", "Jan Weidner", "Alexander Seiler", "Patricio Farrell", "Matthias Liero"]
-version = "1.15.1"
+version = "1.16.0"
 
 [deps]
 BandedMatrices = "aae01518-5342-5314-be14-df237901396f"
+Colors = "5ae59095-9a9b-59fe-a467-6f913c188581"
 CommonSolve = "38540f10-b2f7-11e9-35d8-d573e4eb0ff2"
 DiffResults = "163ba53b-c6d8-5494-b064-1a9d43ac40c5"
 DocStringExtensions = "ffbed154-4ef7-542d-bbb7-c09d3a79fcae"
@@ -28,6 +29,7 @@ Symbolics = "0c5d862f-8b57-4792-8d23-62f2024744c7"
 
 [compat]
 BandedMatrices = "0.17,1"
+Colors = "0.12"
 CommonSolve = "0.2"
 DiffResults = "1"
 DocStringExtensions = "0.8,0.9"
@@ -40,7 +42,7 @@ LinearAlgebra = "1.6"
 LinearSolve = "2.2"
 Printf = "1.6"
 Random = "1.6"
-RecursiveArrayTools = "^2.14.2"
+RecursiveArrayTools = "^2.14.2,3"
 RecursiveFactorization = "0.2.18"
 SparseArrays = "1.6"
 SparseDiffTools = "^1.19, 2"

docs/src/changes.md

@@ -1,4 +1,9 @@
 # Changes
+## v1.16.0 Dec 15, 2023
+- Bugfix for assembly of outflow bc
+- Bugfix for matrixtype=:banded
+- Updated [`plothistory`](@ref): 
+
 ## v1.15.0 Dec 1, 2023
 - Adjusted time/embedding stepping scheme, added `num_final_steps` to  [`VoronoiFVM.SolverControl`](@ref). This may lead to
   sligthly different results when solving time dependent problems. Some unit test values have been adapted. Before,

examples/Example120_ThreeRegions1D.jl

@@ -7,6 +7,7 @@ using Printf
 using VoronoiFVM
 using ExtendableGrids
 using GridVisualize
+using LinearSolve
 
 function main(; n = 30, Plotter = nothing, plot_grid = false, verbose = false,
               unknown_storage = :sparse, tend = 10,
@@ -124,22 +125,24 @@ function main(; n = 30, Plotter = nothing, plot_grid = false, verbose = false,
                     clear = false, show = true)
     end
 
-    tsol = solve(sys; inival = 0, times = (0, tend), post = plot_timestep, verbose = verbose, Δu_opt = 1.0e-5)
+    tsol = solve(sys; inival = 0, times = (0, tend), post = plot_timestep, verbose = verbose, Δu_opt = 1.0e-5,
+                 method_linear=KLUFactorization())
 
     return testval
 end
 
 using Test
+
 function runtests()
     testval = 0.359448515181824
-    @test main(; unknown_storage = :sparse, rely_on_corrections = false, assembly = :edgewise) ≈ testval &&
-          main(; unknown_storage = :dense, rely_on_corrections = false, assembly = :edgewise) ≈ testval &&
-          main(; unknown_storage = :sparse, rely_on_corrections = true, assembly = :edgewise) ≈ testval &&
-          main(; unknown_storage = :dense, rely_on_corrections = true, assembly = :edgewise) ≈ testval &&
-          main(; unknown_storage = :sparse, rely_on_corrections = false, assembly = :cellwise) ≈ testval &&
-          main(; unknown_storage = :dense, rely_on_corrections = false, assembly = :cellwise) ≈ testval &&
-          main(; unknown_storage = :sparse, rely_on_corrections = true, assembly = :cellwise) ≈ testval &&
-          main(; unknown_storage = :dense, rely_on_corrections = true, assembly = :cellwise) ≈ testval
+    @test main(; unknown_storage = :sparse, rely_on_corrections = false, assembly = :edgewise) ≈ testval
+    @test main(; unknown_storage = :dense, rely_on_corrections = false, assembly = :edgewise) ≈ testval
+    @test main(; unknown_storage = :sparse, rely_on_corrections = true, assembly = :edgewise) ≈ testval
+    @test main(; unknown_storage = :dense, rely_on_corrections = true, assembly = :edgewise) ≈ testval
+    @test main(; unknown_storage = :sparse, rely_on_corrections = false, assembly = :cellwise) ≈ testval
+    @test main(; unknown_storage = :dense, rely_on_corrections = false, assembly = :cellwise) ≈ testval
+    @test main(; unknown_storage = :sparse, rely_on_corrections = true, assembly = :cellwise) ≈ testval
+    @test main(; unknown_storage = :dense, rely_on_corrections = true, assembly = :cellwise) ≈ testval
 end
 
 end

src/.JuliaFormatter.toml

@@ -1,4 +1,7 @@
 style = "sciml"
+pipe_to_function_call=false
 always_for_in = false
 separate_kwargs_with_semicolon = true
 margin = 132
+yas_style_nesting=true
+ignore = "examples/PlutoGridFactory.jl"

src/gridvisualize.jl

@@ -1,5 +1,6 @@
 # Extend GridVisualize methods
 import GridVisualize
+import Colors
 
 """
     $(TYPEDSIGNATURES)
@@ -79,14 +80,18 @@ end
 
 """
     plothistory(sys, tsol; 
-                plots=[:timesteps,:newtonsteps,:updates], 
-                size=(700,400), 
+                plots=[:timestepsizes,
+                       :timestepupdates,
+                       :newtonsteps,
+                       :newtonupdates], 
+                size=(700,600), 
                 Plotter=GridVisualize.default_plotter(),
                 kwargs...)
 
 Plot solution history stored in `tsol`. The `plots` argument allows to choose which plots are shown.
 """
-function plothistory(sys, tsol; plots=[:timesteps,:newtonsteps,:updates], size=(700,400), Plotter=GridVisualize.default_plotter(), kwargs...)
+function plothistory(tsol::TransientSolution; plots=[:timestepsizes,:timestepupdates,:newtonsteps,:newtonupdates],
+                     size=(700,150*length(plots)), Plotter=GridVisualize.default_plotter(), kwargs...)
     hist=history(tsol)
     t=hist.times
     if length(t)==0
@@ -96,17 +101,37 @@ function plothistory(sys, tsol; plots=[:timesteps,:newtonsteps,:updates], size=(
     vis=GridVisualize.GridVisualizer(;layout=(nplots,1), size, Plotter, kwargs...)
 
     for iplot in eachindex(plots)
-        if plots[iplot]==:timesteps
+        if plots[iplot]==:timestepsizes
             GridVisualize.scalarplot!(vis[iplot,1],t[1:(end - 1)], t[2:end] - t[1:(end - 1)];
                                       title = "Time step sizes",  xlabel = "t/s", ylabel = "Δt/s", color=:red, kwargs...)
+        elseif plots[iplot]==:timestepupdates
+            GridVisualize.scalarplot!(vis[iplot,1],t, hist.updates,xlabel = "t/s", ylabel = "du",
+                                      title="Time step updates", color=:red, kwargs...)
         elseif plots[iplot]==:newtonsteps
             newtons=map(h->length(h.updatenorm),hist.histories)
             GridVisualize.scalarplot!(vis[iplot,1],t, newtons,xlabel = "t/s", ylabel = "n",
-                        title="Newton iterations", color=:red, limits=(0,maximum(newtons)+1), kwargs...)
-        elseif plots[iplot]==:updates
-            GridVisualize.scalarplot!(vis[iplot,1],t, hist.updates,xlabel = "t/s", ylabel = "du",
-                        title="Time step updates", color=:red, kwargs...)
+                        title="Newton steps", color=:red, limits=(0,maximum(newtons)+1), kwargs...)
+        elseif plots[iplot]==:newtonupdates
+	    nhist=length(hist)
+	    dc=1/nhist
+	    r=0.0
+	    b=1.0
+	    g=0.0
+	    for h in hist
+   	        GridVisualize.scalarplot!(vis[iplot,1],
+                                          1:length(h),h,
+                                          xlabel="step",
+                                          ylabel="du",
+                                          clear=false,
+                                          yscale=:log,
+                                          color=Colors.RGB(r,g,b),
+                                          linewidth=0.5, title="Newton updates")
+		r+=dc
+		b-=dc
+	    end
         end
     end
     GridVisualize.reveal(vis)
 end
+
+plothistory(sys,tsol;kwargs...)=plothistory(tsol,kwargs...)

src/vfvm_assembly.jl

@@ -273,7 +273,7 @@ function eval_and_assemble(
                             system.matrix,
                             idofK,
                             jdofL,
-                            -jac_outflow[ispec, jspec+nspecies],
+                            jac_outflow[ispec, jspec+nspecies],
                             edge.fac
                         )
                     end
@@ -283,7 +283,7 @@ function eval_and_assemble(
                             system.matrix,
                             idofL,
                             jdofK,
-                            +jac_outflow[ispec, jspec],
+                            -jac_outflow[ispec, jspec],
                             edge.fac
                         )
                         _addnz(

src/vfvm_linsolve.jl

@@ -228,11 +228,14 @@ function LinearAlgebra.ldiv!(u, cache::LinearSolve.LinearCache, b)
     copyto!(u, sol.u)
 end
 
+canonical_matrix(A)=A
+canonical_matrix(A::ExtendableSparseMatrix)=SparseMatrixCSC(A)
+
 function _solve_linear!(u, system, nlhistory, control, method_linear, A, b)
     if isnothing(system.linear_cache)
-        Pl = control.precon_linear(SparseMatrixCSC(A))
+        Pl = control.precon_linear(canonical_matrix(A))
         nlhistory.nlu += 1
-        p = LinearProblem(SparseMatrixCSC(A), b)
+        p = LinearProblem(canonical_matrix(A), b)
         system.linear_cache = init(
             p,
             method_linear;
@@ -242,11 +245,11 @@ function _solve_linear!(u, system, nlhistory, control, method_linear, A, b)
             Pl,
         )
     else
-        system.linear_cache.A=SparseMatrixCSC(A)
+        system.linear_cache.A=canonical_matrix(A)
         system.linear_cache.b=b
         if control.keepcurrent_linear
             nlhistory.nlu += 1
-            system.linear_cache.Pl=control.precon_linear(SparseMatrixCSC(A))
+            system.linear_cache.Pl=control.precon_linear(canonical_matrix(A))
         end
     end
 

src/vfvm_solver.jl

@@ -29,18 +29,16 @@ function _solve_timestep!(solution::AbstractMatrix{Tv}, # old time step solution
         update = system.update
         _initialize!(solution, system; time, λ = embedparam, params)
 
-        method_linear = control.method_linear
-        if isnothing(method_linear)
+        method_linear = system.matrixtype == :sparse ? control.method_linear : nothing;
+        if isnothing(method_linear) &&  system.matrixtype == :sparse
             if Tv != Float64
                 method_linear = SparspakFactorization()
+            elseif dim_space(system.grid)==1
+                method_linear = KLUFactorization()
+            elseif dim_space(system.grid)==2
+                method_linear = SparspakFactorization()
             else
-                if dim_grid(system.grid) == 1
-                    method_linear = KLUFactorization()
-                elseif dim_grid(system.grid) == 2
-                    method_linear = SparspakFactorization()
-                else
-                    method_linear = UMFPACKFactorization()
-                end
+                method_linear = UMFPACKFactorization() # seems to do the best pivoting
             end
         end