Use L-BFGS if the Hessian of the Lagrangian is not available

.github/workflows/Breakage.yml

@@ -159,8 +159,14 @@ jobs:
             fi
           done >> summary.md
 
+      - name: Display summary in CI logs
+        run: |
+          echo "### Breakage Summary" >> $GITHUB_STEP_SUMMARY
+          cat breakage/summary.md >> $GITHUB_STEP_SUMMARY
+
       - name: PR comment with file
-        uses: actions/github-script@v6
+        if: github.event.pull_request.head.repo.fork == false
+        uses: actions/github-script@main
         with:
           github-token: ${{ secrets.GITHUB_TOKEN }}
           script: |

.github/workflows/ci.yml

@@ -13,26 +13,26 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        version: ['1.6', '1']
-        os: [ubuntu-latest, macOS-latest, windows-latest]
+        version: ['lts', '1']
+        os: [ubuntu-latest, macos-latest, windows-latest]
         arch: [x64]
         allow_failure: [false]
         include:
-          - version: 'nightly'
+          - version: 'pre'
             os: ubuntu-latest
             arch: x64
             allow_failure: true
-          - version: 'nightly'
-            os: macOS-latest
+          - version: 'pre'
+            os: macos-latest
             arch: x64
             allow_failure: true
-          - version: 'nightly'
+          - version: 'pre'
             os: windows-latest
             arch: x64
             allow_failure: true
     steps:
       - uses: actions/checkout@v2
-      - uses: julia-actions/setup-julia@v1
+      - uses: julia-actions/setup-julia@v2
         with:
           version: ${{ matrix.version }}
           arch: ${{ matrix.arch }}

Project.toml

@@ -9,11 +9,11 @@ NLPModelsModifiers = "e01155f1-5c6f-4375-a9d8-616dd036575f"
 SolverCore = "ff4d7338-4cf1-434d-91df-b86cb86fb843"
 
 [compat]
-Ipopt = "1"
-NLPModels = "0.19, 0.20, 0.21"
+Ipopt = "1.11"
+NLPModels = "0.21.6"
 NLPModelsModifiers = "0.7"
 SolverCore = "0.3"
-julia = "^1.6"
+julia = "1.10"
 
 [extras]
 ADNLPModels = "54578032-b7ea-4c30-94aa-7cbd1cce6c9a"

docs/Project.toml

@@ -9,6 +9,6 @@ NLPModels = "a4795742-8479-5a88-8948-cc11e1c8c1a6"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 
 [compat]
-ADNLPModels = "0.7"
+ADNLPModels = "0.8"
 Documenter = "1.0"
-NLPModels = "0.20"
+NLPModels = "0.21.6"

docs/src/tutorial.md

@@ -6,13 +6,15 @@ You can call Ipopt with the L-BFGS Hessian approximation by passing the followin
 stats_ipopt = ipopt(nlp,
   hessian_approximation="limited-memory",
   limited_memory_update_type="bfgs",
-  limited_memory_max_history=10)
+  limited_memory_max_history=6)
 ```
 
-This will use the L-BFGS method for Hessian approximation with a history size of 10.
+This will use the L-BFGS method for Hessian approximation with a history size of 6.
+Note that these options are used by default in `NLPModelsIpopt.jl` if `nlp.meta.hess_available` is `false`.
 
 Reference:
 - [Ipopt.jl Manual: Hessian approximation](https://coin-or.github.io/Ipopt/OPTIONS.html#OPT_Hessian_Approximation)
+
 # Tutorial
 
 NLPModelsIpopt is a thin IPOPT wrapper for NLPModels. In this tutorial we show examples of problems created with NLPModels and solved with Ipopt.

src/NLPModelsIpopt.jl

@@ -61,6 +61,7 @@ mutable struct IpoptSolver <: AbstractOptimizationSolver
 end
 
 function IpoptSolver(nlp::AbstractNLPModel)
+  @assert nlp.meta.grad_available && (nlp.meta.ncon == 0 || nlp.meta.jac_available)
   eval_f, eval_g, eval_grad_f, eval_jac_g, eval_h = set_callbacks(nlp)
 
   problem = CreateIpoptProblem(
@@ -248,6 +249,13 @@ function SolverCore.solve!(
   SolverCore.reset!(stats)
   kwargs = Dict(kwargs)
 
+  # Use L-BFGS if the sparse hessian of the Lagrangian is not available
+  if !nlp.meta.hess_available
+    AddIpoptStrOption(problem, "hessian_approximation", "limited-memory")
+    AddIpoptStrOption(problem, "limited_memory_update_type", "bfgs")
+    AddIpoptIntOption(problem, "limited_memory_max_history", 6)
+  end
+
   # see if user wants to warm start from an initial primal-dual guess
   if all(k ∈ keys(kwargs) for k ∈ [:x0, :y0, :zL0, :zU0])
     AddIpoptStrOption(problem, "warm_start_init_point", "yes")