|
| 1 | +# Debugging |
| 2 | + |
| 3 | +Every (mortal) modeler writes models that contain mistakes or are susceptible to numerical errors in their hunt for the perfect model. |
| 4 | +Debugging such errors is part of the modeling process, and ModelingToolkit includes some functionality that helps with this. |
| 5 | + |
| 6 | +For example, consider an ODE model with "dangerous" functions (here `√`): |
| 7 | + |
| 8 | +```@example debug |
| 9 | +using ModelingToolkit, OrdinaryDiffEq |
| 10 | +using ModelingToolkit: t_nounits as t, D_nounits as D |
| 11 | +
|
| 12 | +@variables u1(t) u2(t) u3(t) |
| 13 | +eqs = [D(u1) ~ -√(u1), D(u2) ~ -√(u2), D(u3) ~ -√(u3)] |
| 14 | +defaults = [u1 => 1.0, u2 => 2.0, u3 => 3.0] |
| 15 | +@named sys = ODESystem(eqs, t; defaults) |
| 16 | +sys = structural_simplify(sys) |
| 17 | +``` |
| 18 | + |
| 19 | +This problem causes the ODE solver to crash: |
| 20 | + |
| 21 | +```@example debug |
| 22 | +prob = ODEProblem(sys, [], (0.0, 10.0), []) |
| 23 | +sol = solve(prob, Tsit5()) |
| 24 | +``` |
| 25 | + |
| 26 | +This suggests *that* something went wrong, but not exactly *what* went wrong and *where* it did. |
| 27 | +In such situations, the `debug_system` function is helpful: |
| 28 | + |
| 29 | +```@example debug |
| 30 | +try # workaround to show Documenter.jl error (https://github.com/JuliaDocs/Documenter.jl/issues/1420#issuecomment-770539595) # hide |
| 31 | +dsys = debug_system(sys; functions = [sqrt]) |
| 32 | +dprob = ODEProblem(dsys, [], (0.0, 10.0), []) |
| 33 | +dsol = solve(dprob, Tsit5()) |
| 34 | +catch err # hide |
| 35 | +showerror(stderr, err) # hide |
| 36 | +end # hide |
| 37 | +``` |
| 38 | + |
| 39 | +Now we see that it crashed because `u1` decreased so much that it became negative and outside the domain of the `√` function. |
| 40 | +We could have figured that out ourselves, but it is not always so obvious for more complex models. |
| 41 | + |
| 42 | +```@docs |
| 43 | +debug_system |
| 44 | +``` |
0 commit comments