add a ShiftedProximableQuadraticNLPModel by MaxenceGollier · Pull Request #113 · JuliaSmoothOptimizers/RegularizedProblems.jl

MaxenceGollier · 2026-03-25T19:41:56Z

closes #112.

What is implemented

I implement a new type ShiftedProximableQuadraticNLPModel.
The documentation I add here is self-explanatory:

    subproblem = ShiftedProximableQuadraticNLPModel(reg_nlp, x; kwargs...)

Given a regularized NLP model `reg_nlp` representing the problem

    minimize f(x) + h(x),
  
construct a shifted quadratic model around `x`:

    minimize  φ(s; x) + ½ σ ‖s‖² + ψ(s; x),

where φ(s ; x) = f(x) + ∇f(x)ᵀs + ½ sᵀBs is a quadratic approximation of f about x,
ψ(s; x) is either h(x + s) or an approximation of h(x + s), ‖⋅‖ is the ℓ₂ norm and σ > 0 is the regularization parameter.

How do I implement it

I add an abstract type

abstract type AbstractShiftedProximableNLPModel{T, V} <: AbstractRegularizedNLPModel{T, V} end

as well as a mutable struct

mutable struct ShiftedProximableQuadraticNLPModel{T, V, M <: AbstractNLPModel{T, V}, H <: ShiftedProximalOperators.ShiftedProximableFunction, I, P <: AbstractRegularizedNLPModel{T, V}} <:
       AbstractShiftedProximableNLPModel{T, V}
  model::M
  h::H
  selected::I
  parent::P
end

where model represents φ(s; x) + ½ σ ‖s‖², h represents ψ and parent, the original problem.

Why do I implement it

In RegularizedOptimization.jl, I want to construct the subproblem with a sparse Hessian to use here in my exact penalty solver. This allows me to make a different constructor, by overwriting ShiftedProximableQuadraticNLPModel where the Hessian is sparse. In the shift! function, I will be able to update the sparse Hessian as well by overwriting.
In RegularizedOptimization.jl there is a ton of redundant code between each solver. For example, R2N defines

  φ1 = let ∇fk = ∇fk
    d -> dot(∇fk, d)
  end

  mk1 = let ψ = ψ
    d -> φ1(d) + ψ(d)::T
  end

and so does TR, TRDH, R2, R2DH, etc. When a step is accepted in each of those, there is a lot of noisy code as well. With this feature, we will now be able to simply call shift!(subproblem, x) when the step x is accepted.

How do I test it

I added unit tests here. I plan to make a draft PR in RegularizedOptimization.jl where I use this code.
Link to draft PR: JuliaSmoothOptimizers/RegularizedOptimization.jl#316.

Copilot

Pull request overview

This PR adds a new shifted subproblem model type to support regularized optimization workflows that build and update quadratic + prox subproblems around an iterate.

Changes:

Introduces AbstractShiftedProximableNLPModel and ShiftedProximableQuadraticNLPModel with shift!, obj, and σ helpers.
Wires the new implementation behind @require ShiftedProximalOperators and @require QuadraticModels.
Adds unit tests for construction, sigma handling, shifting behavior, and allocations.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 8 comments.

File	Description
`src/shiftedProximableNLPModel.jl`	Implements the new shifted quadratic + proximable subproblem model, including `shift!`, `obj`, and forwarding methods.
`src/RegularizedProblems.jl`	Conditionally includes the new file when `QuadraticModels` and `ShiftedProximalOperators` are available.
`test/rmodel_tests.jl`	Adds tests validating objective evaluation, sigma toggling, shifting, bounds shifting, and allocations.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

test/rmodel_tests.jl

src/shiftedProximableNLPModel.jl

Copilot · 2026-03-25T19:47:10Z

src/shiftedProximableNLPModel.jl

+  σ_c = skip_sigma ? zero(σ_temp) : σ_temp
+  update_sigma!(reg_nlp, σ_c)
+
+  φs = cauchy ? dot(φ.data.c, s) + σ_c * dot(s, s)/2 : obj(φ, s)
+  ψs = ψ(s)
+
+  update_sigma!(reg_nlp, σ_temp) # restore original σ
+


obj temporarily mutates the model’s σ via update_sigma! and then restores it. If obj(φ, s) or ψ(s) throws, σ will not be restored, leaving the model in a corrupted state (and it’s also not thread-safe). Consider avoiding mutation (e.g., compute with the current σ and subtract the sigma term when skip_sigma=true), or wrap the restore in a try/finally to guarantee restoration.

Suggested change

σ_c = skip_sigma ? zero(σ_temp) : σ_temp

update_sigma!(reg_nlp, σ_c)

φs = cauchy ? dot(φ.data.c, s) + σ_c * dot(s, s)/2 : obj(φ, s)

ψs = ψ(s)

update_sigma!(reg_nlp, σ_temp) # restore original σ

if cauchy

# In the Cauchy case, construct the quadratic model explicitly using either

# the original σ or zero, depending on skip_sigma, without mutating σ.

σ_used = skip_sigma ? zero(σ_temp) : σ_temp

φs = dot(φ.data.c, s) + σ_used * dot(s, s) / 2

else

# Use the current σ in the model for φ, and, if requested, subtract the

# σ contribution from the objective afterwards instead of mutating σ.

φs = obj(φ, s)

if skip_sigma

φs -= σ_temp * dot(s, s) / 2

end

end

ψs = ψ(s)

why not but I think @dpo will not agree because we might compute $$\lVert \cdot \rVert$$ twice.

src/shiftedProximableNLPModel.jl

github-actions · 2026-03-25T19:55:04Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-25T20:24:31Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-25T20:38:25Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-25T20:54:13Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-25T20:57:46Z

Package name	latest	stable
RegularizedOptimization

codecov · 2026-03-25T21:05:41Z

Codecov Report

❌ Patch coverage is 98.27586% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 79.27%. Comparing base (85de184) to head (3fc6090).
⚠️ Report is 35 commits behind head on main.

Files with missing lines	Patch %	Lines
src/shiftedProximableNLPModel.jl	98.27%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main     #113       +/-   ##
===========================================
- Coverage   99.18%   79.27%   -19.92%     
===========================================
  Files          11       20        +9     
  Lines         369      521      +152     
===========================================
+ Hits          366      413       +47     
- Misses          3      108      +105

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

github-actions · 2026-03-30T18:00:19Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-31T14:54:47Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-31T21:42:04Z

Package name	latest	stable
RegularizedOptimization

github-actions · 2026-03-31T21:46:17Z

Package name	latest	stable
RegularizedOptimization

MaxenceGollier added 2 commits March 25, 2026 15:24

add a ShiftedProximableQuadraticNLPModel

12d5ce1

make the struct mutable

0c3c60e

MaxenceGollier requested review from Copilot and dpo March 25, 2026 19:41

Copilot started reviewing on behalf of MaxenceGollier March 25, 2026 19:42 View session

Copilot AI reviewed Mar 25, 2026

View reviewed changes

apply copilot suggestions

c6628c3

compute hess_op on shifts

b6095e7

MaxenceGollier mentioned this pull request Mar 25, 2026

Improve subproblem definition JuliaSmoothOptimizers/RegularizedOptimization.jl#316

Draft

MaxenceGollier added 2 commits March 25, 2026 16:40

fix tests

93bdf71

update doc

c886f42

MaxenceGollier mentioned this pull request Mar 26, 2026

Allow for sparse Hessians in solvers JuliaSmoothOptimizers/RegularizedOptimization.jl#317

Open

start bounds

d028a70

MaxenceGollier added 2 commits March 31, 2026 10:28

add bounds logic

a559eeb

improve readibility: add IndicatorFunction types logic

7a6a816

MaxenceGollier marked this pull request as draft March 31, 2026 15:00

MaxenceGollier added 2 commits March 31, 2026 17:28

fix tests

8295b47

refactor to set_sigma!

3fc6090

MaxenceGollier marked this pull request as ready for review March 31, 2026 21:34

-  σ_c = skip_sigma ? zero(σ_temp) : σ_temp
-  update_sigma!(reg_nlp, σ_c)
-  φs = cauchy ? dot(φ.data.c, s) + σ_c * dot(s, s)/2 : obj(φ, s)
-  ψs = ψ(s)
-  update_sigma!(reg_nlp, σ_temp) # restore original σ
+  if cauchy
+    # In the Cauchy case, construct the quadratic model explicitly using either
+    # the original σ or zero, depending on skip_sigma, without mutating σ.
+    σ_used = skip_sigma ? zero(σ_temp) : σ_temp
+    φs = dot(φ.data.c, s) + σ_used * dot(s, s) / 2
+  else
+    # Use the current σ in the model for φ, and, if requested, subtract the
+    # σ contribution from the objective afterwards instead of mutating σ.
+    φs = obj(φ, s)
+    if skip_sigma
+      φs -= σ_temp * dot(s, s) / 2
+    end
+  end
+  ψs = ψ(s)

Conversation

MaxenceGollier commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What is implemented

How do I implement it

Why do I implement it

How do I test it

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI Mar 25, 2026

Choose a reason for hiding this comment

Uh oh!

MaxenceGollier Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

github-actions bot commented Mar 25, 2026

Uh oh!

github-actions bot commented Mar 25, 2026

Uh oh!

github-actions bot commented Mar 25, 2026

Uh oh!

github-actions bot commented Mar 25, 2026

Uh oh!

github-actions bot commented Mar 25, 2026

Uh oh!

codecov bot commented Mar 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

github-actions bot commented Mar 30, 2026

Uh oh!

github-actions bot commented Mar 31, 2026

Uh oh!

github-actions bot commented Mar 31, 2026

Uh oh!

github-actions bot commented Mar 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

MaxenceGollier commented Mar 25, 2026 •

edited

Loading

MaxenceGollier Mar 25, 2026 •

edited

Loading

codecov bot commented Mar 25, 2026 •

edited

Loading