Merge pull request #86 from harmoniqs/fix/diverging-iterates-tol-default

solver(ipopt): bump diverging_iterates_tol default 1e8 → 1e20

Author: jack-champagne

Project.toml

@@ -1,6 +1,6 @@
 name = "DirectTrajOpt"
 uuid = "c823fa1f-8872-4af5-b810-2b9b72bbbf56"
-version = "0.9.3"
+version = "0.9.4"
 authors = ["Aaron Trowbridge <aaron.j.trowbridge@gmail.com> and contributors"]
 
 [deps]

src/solvers/ipopt_solver/options.jl

@@ -44,7 +44,11 @@ Base.@kwdef mutable struct IpoptOptions <: Solvers.AbstractSolverOptions
     acceptable_constr_viol_tol::Float64 = 1.0e-2
     acceptable_compl_inf_tol::Float64 = 1.0e-2
     acceptable_obj_change_tol::Float64 = 1.0e-5
-    diverging_iterates_tol::Float64 = 1.0e8
+    # Ipopt's default is 1e20; the previous 1e8 fired false-positive "Iterates
+    # diverging" exits on smooth-pulse NLPs whose unscaled second derivatives
+    # legitimately reach O(1e8) (e.g. ddu ≈ u_max / Δt² with Δt in ns).
+    # Until DTO-level variable scaling lands, defer to Ipopt's default.
+    diverging_iterates_tol::Float64 = 1.0e20
     eval_hessian = true
     hessian_approximation = eval_hessian ? "exact" : "limited-memory"
     hsllib = nothing
@@ -70,3 +74,10 @@ Base.@kwdef mutable struct IpoptOptions <: Solvers.AbstractSolverOptions
     watchdog_shortened_iter_trigger = 0
     watchdog_trial_iter_max = 3
 end
+
+@testitem "IpoptOptions: diverging_iterates_tol default matches Ipopt" begin
+    using DirectTrajOpt: IpoptSolverExt
+    # Locks in the bump from 1e8 (false-positive on smooth-pulse NLPs) to Ipopt's
+    # native default of 1e20. See hopper/dto-variable-scaling for context.
+    @test IpoptSolverExt.IpoptOptions().diverging_iterates_tol == 1.0e20
+end