Verbosity Specification with SciMLLogging.jl

DifferentialEquations.jl uses SciMLLogging.jl to provide users with fine-grained control over logging and diagnostic output during ODE solving. The DEVerbosity struct allows you to customize which messages are displayed, from critical errors to detailed debugging information.

Basic Usage

Pass a DEVerbosity object to solve or init using the verbose keyword argument:

using OrdinaryDiffEq

# Define an ODE problem
function lorenz!(du, u, p, t)
    du[1] = 10.0 * (u[2] - u[1])
    du[2] = u[1] * (28.0 - u[3]) - u[2]
    du[3] = u[1] * u[2] - (8/3) * u[3]
end

u0 = [1.0, 0.0, 0.0]
tspan = (0.0, 100.0)
prob = ODEProblem(lorenz!, u0, tspan)

# Solve with detailed verbosity
verbose = DEVerbosity(SciMLLogging.Detailed())
sol = solve(prob, Tsit5(), verbose = verbose)

Example Use Cases

Debugging Algorithm Switching

When using auto-switching algorithms, see when and why switches occur:

using ODEProblemLibrary: prob_ode_vanderpol_stiff

verbose = DEVerbosity(alg_switch = SciMLLogging.InfoLevel())
sol = solve(prob_ode_vanderpol_stiff, AutoTsit5(Rodas5()), verbose = verbose)

Monitoring Stiff Solver Performance

Get detailed information about Jacobian updates, factorizations, and linear solver behavior:

function rober(du, u, p, t)
    y₁, y₂, y₃ = u
    k₁, k₂, k₃ = p
    du[1] = -k₁ * y₁ + k₃ * y₂ * y₃
    du[2] = k₁ * y₁ - k₃ * y₂ * y₃ - k₂ * y₂^2
    du[3] = y₁ + y₂ + y₃ - 1
end

u0 = [1.0, 0.0, 0.0]
tspan = (0.0, 0.1)
p = [0.04, 3e7, 1e4]
prob = ODEProblem(rober, u0, tspan, p)

# Monitor performance-related events and linear solver details
verbose = DEVerbosity(
    performance = SciMLLogging.InfoLevel(),
    linear_verbosity = SciMLLogging.Detailed()
)
sol = solve(prob, Rosenbrock23(), verbose = verbose)

Debugging Step Acceptance/Rejection

For adaptive methods, monitor which steps are accepted or rejected:

verbose = DEVerbosity(
    step_accepted = SciMLLogging.InfoLevel(),
    step_rejected = SciMLLogging.InfoLevel()
)
sol = solve(prob, Tsit5(), verbose = verbose)

Silent Operation

Completely disable all output:

verbose = DEVerbosity(SciMLLogging.None())
solve(prob, Tsit5(), verbose = verbose)

Using with init

The verbosity settings work the same way with init for manual stepping:

verbose = DEVerbosity(
    alg_switch = SciMLLogging.InfoLevel(),
    linear_verbosity = SciMLLogging.Detailed()
)
integrator = init(prob, Rosenbrock23(), verbose = verbose)

# Step through the solution
step!(integrator)

Controlling Subsolver Verbosity

When using implicit methods, linear_verbosity and nonlinear_verbosity are automatically passed to the linear and nonlinear solver caches, enabling diagnostics from LinearSolve.jl and NonlinearSolve.jl.

Using SciMLLogging Presets

The simplest approach is to pass a SciMLLogging preset, which is automatically converted to the appropriate verbosity type:

using OrdinaryDiffEqNonlinearSolve: NonlinearSolveAlg

# Enable detailed nonlinear solver diagnostics
verbose = DEVerbosity(nonlinear_verbosity = SciMLLogging.Detailed())
sol = solve(prob, ImplicitEuler(nlsolve = NonlinearSolveAlg()), verbose = verbose)

Using LinearVerbosity and NonlinearVerbosity Explicitly

You can also explicitly use LinearVerbosity and NonlinearVerbosity types from LinearSolve.jl and NonlinearSolve.jl:

using LinearSolve: LinearVerbosity
using NonlinearSolve: NonlinearVerbosity

# Create explicit verbosity objects
linear_verb = LinearVerbosity(SciMLLogging.Detailed())
nonlinear_verb = NonlinearVerbosity(SciMLLogging.Minimal())

# Pass them to DEVerbosity
verbose = DEVerbosity(
    linear_verbosity = linear_verb,
    nonlinear_verbosity = nonlinear_verb
)

# Use with a stiff solver
sol = solve(prob, Rosenbrock23(), verbose = verbose)

Combining Settings Flexibly

Individual field settings override group settings, allowing precise control:

# Set all error_control messages to WarnLevel,
# but make step_rejected silent and dt_NaN an error
verbose = DEVerbosity(
    error_control = SciMLLogging.WarnLevel(),
    step_rejected = SciMLLogging.Silent(),
    dt_NaN = SciMLLogging.ErrorLevel()
)
sol = solve(prob, Tsit5(), verbose = verbose)

API Reference