Fix Ruiz equilibration skip heuristic to also check column imbalance by mlubin · Pull Request #1601 · NVIDIA/cuopt

mlubin · 2026-07-21T21:33:07Z

The skip heuristic in scaling() only measured row-norm imbalance, so a QP/SOCP with balanced rows but severely imbalanced columns (e.g. QPLIB_8515, where free variables appear only with ~1e-8 coefficients against O(1) rows) received no scaling at all. The resulting augmented KKT system was so ill-conditioned that barrier terminated at a suboptimal point while reporting convergence (320.13 vs the true optimum 320.00 on QPLIB_8515, non-deterministically across runs). With the column check, Ruiz runs and the instance solves to 319.999989 in 14 iterations instead of 338.

Adds QPLIB_8515 as a downloaded test dataset (new
download_qplib_test_dataset.sh, wired into ci/test_cpp.sh and CONTRIBUTING.md) and a full-solve regression test that asserts the optimal objective matches the QPLIB reference value.

A summary of impacts on public benchmark instances that are affected:

Improvements

Instance	Source	Iterations	Solve time (s)	Rel. obj. error	Notes
QPLIB_8515	QPLIB	177 → 14	2.62 → 0.36	3.2e-04 → 9.4e-10	7.3× faster; now matches reference objective
PRIMALC1	Maros–Mészáros	23 → 16	0.13 → 0.13	0 → 4.7e-08
PRIMALC2	Maros–Mészáros	25 → 16	0.13 → 0.12	0 → 2.8e-08
PRIMALC5	Maros–Mészáros	21 → 15	0.13 → 0.12	0 → 1.9e-08
PRIMALC8	Maros–Mészáros	26 → 15	0.14 → 0.12	5.5e-09 → 5.5e-09
QPCBLEND	Maros–Mészáros	17 → 17	0.13 → 0.14	5.6e-08 → 7.4e-09	accuracy only; time unchanged

Neutral

Instance	Source	Iterations	Solve time (s)	Rel. obj. error	Notes
QPLIB_8938	QPLIB	18 → 17	0.43 → 0.40	2.1e-07 → 3.9e-07
DTOC3	Maros–Mészáros	12 → 12	0.83 → 0.83	0 → 0
STADAT2	Maros–Mészáros	20 → 17	0.86 → 0.81	0 → 3.6e-07
STADAT3	Maros–Mészáros	20 → 18	0.86 → 0.85	0 → 2.8e-07
dsNRL	CBLIB	41 → 41	67.8 → 69.1	2.4e-07 → 2.4e-07
wbNRL	CBLIB	28 → 29	25.9 → 28.4	5.3e-06 → 4.8e-06
firL1	CBLIB	24 → 24	28.2 → 30.0	3.5e-06 → 1.1e-05	both errors well inside 1e-4 tolerance
firLinf	CBLIB	23 → 24	84.7 → 91.3	1.3e-06 → 4.0e-07	+8% time, better accuracy
firL1Linfalph	CBLIB	36 → 37	71.4 → 76.4	1.3e-07 → 7.3e-08	+7% time, better accuracy

Regressions

Instance	Source	Iterations	Solve time (s)	Rel. obj. error	Notes
db-plate-yield-line	CBLIB	42 → 46	6.4 → 8.7	9.9e-07 → 1.4e-06	reproducible: main stable at 42 it / ~6 s; branch 47–56 it / 8.5–9.5 s across reruns
STADAT1	Maros–Mészáros	64 → 381	0.93 → 1.46	0 → 0	mixed: main flipped optimal/suboptimal across reruns; branch always optimal (193–381 it)
HUES-MOD	Maros–Mészáros	10 → 26	0.12 → 0.14	7.0e-05 → 7.0e-05	iteration count only; sub-second solve
HUESTIS	Maros–Mészáros	15 → 32	0.12 → 0.15	7.0e-05 → 7.0e-05	iteration count only; sub-second solve
QPLIB_9008	QPLIB	—	54.1 → 169.0	—	both variants fail numerically ("search direction computation failed"); branch takes 3.1× longer to fail

Adds a --qcqp_hyper_ruiz_equilibration flag to override the behavior.

copy-pr-bot · 2026-07-21T21:33:11Z

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

mlubin · 2026-07-21T21:46:15Z

/ok to test ed675a1

github-actions · 2026-07-21T22:12:54Z

CI Test Summary

✅ All 31 test job(s) passed.

yuwenchen95 · 2026-07-22T08:55:53Z

+# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+INSTANCES=(


Are we only downloading this specific example?

Yes, it's the first instance from qplib that we're using in tests.

chris-maes · 2026-07-22T13:58:15Z

+ }
+ f_t col_norm_ratio = (min_col_norm > 0) ? max_col_norm / min_col_norm : 1.0;
+
+ if (row_norm_ratio < 100.0 && col_norm_ratio < 100.0) {


We might want to add an option --row-scaling that by default is -1 and we use the heuristic below, but if the user sets --row-scaling=0 they can turn it off, and if they set --row-scaling=1 they can force it on.

coderabbitai · 2026-07-22T21:42:31Z

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

@coderabbitai resume to resume automatic reviews.
@coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

▶️ Resume reviews
🔍 Trigger review

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: d66026d8-dfc0-470f-aaf6-b06405397c4e

📥 Commits

Reviewing files that changed from the base of the PR and between a6bcbfa and 378c96b.

📒 Files selected for processing (11)

.gitignore
CONTRIBUTING.md
ci/test_cpp.sh
cpp/include/cuopt/mathematical_optimization/constants.h
cpp/include/cuopt/mathematical_optimization/pdlp/solver_settings.hpp
cpp/src/dual_simplex/scaling.cpp
cpp/src/dual_simplex/simplex_solver_settings.hpp
cpp/src/math_optimization/solver_settings.cu
cpp/src/pdlp/solve.cu
cpp/tests/dual_simplex/unit_tests/solve_barrier.cu
datasets/quadratic_programming/download_qplib_test_dataset.sh

🚧 Files skipped from review as they are similar to previous changes (10)

cpp/src/math_optimization/solver_settings.cu
ci/test_cpp.sh
cpp/src/pdlp/solve.cu
cpp/src/dual_simplex/simplex_solver_settings.hpp
cpp/include/cuopt/mathematical_optimization/constants.h
.gitignore
CONTRIBUTING.md
datasets/quadratic_programming/download_qplib_test_dataset.sh
cpp/src/dual_simplex/scaling.cpp
cpp/tests/dual_simplex/unit_tests/solve_barrier.cu

📝 Walkthrough

Walkthrough

Changes

The QPLIB downloader now supports S3 with HTTP fallback and is wired into CI and contributor instructions. QCQP Ruiz equilibration is configurable, and barrier scaling uses both row and column imbalance ratios with QPLIB regression tests.

QPLIB scaling regression

Layer / File(s)	Summary
QPLIB dataset acquisition and wiring `datasets/quadratic_programming/download_qplib_test_dataset.sh`, `ci/test_cpp.sh`, `CONTRIBUTING.md`, `.gitignore`	Adds S3-first QPLIB downloads with HTTP fallback, CI and documentation commands, and an ignore rule for downloaded instances.
QCQP Ruiz configuration plumbing `cpp/include/cuopt/mathematical_optimization/constants.h`, `cpp/include/cuopt/mathematical_optimization/pdlp/solver_settings.hpp`, `cpp/src/dual_simplex/simplex_solver_settings.hpp`, `cpp/src/math_optimization/solver_settings.cu`, `cpp/src/pdlp/solve.cu`	Declares, validates, initializes, and forwards the QCQP Ruiz equilibration setting into barrier solver settings.
Column-aware Ruiz scaling regression `cpp/src/dual_simplex/scaling.cpp`, `cpp/tests/dual_simplex/unit_tests/solve_barrier.cu`	Uses row and column norm ratios for automatic Ruiz-scaling decisions and adds QPLIB_8515 barrier tests for enabled and forced-off behavior.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Suggested reviewers:msarahan, chris-maes

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 18.18% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change to the Ruiz equilibration skip heuristic.
Description check	✅ Passed	The description matches the implemented scaling fix, dataset support, CI/docs updates, and regression test.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

Create PR with unit tests

_{Comment @coderabbitai help to get the list of available commands.}

coderabbitai

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@datasets/quadratic_programming/download_qplib_test_dataset.sh`:
- Line 1: Update the dataset download script to enable strict shell error
handling and ensure a failed wget for QPLIB_8515.lp exits with a nonzero status
instead of being masked by continue.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 4a58299f-8732-465d-aa64-49ef9aeee586

📥 Commits

Reviewing files that changed from the base of the PR and between 077e514 and d43e238.

📒 Files selected for processing (6)

.gitignore
CONTRIBUTING.md
ci/test_cpp.sh
cpp/src/dual_simplex/scaling.cpp
cpp/tests/dual_simplex/unit_tests/solve_barrier.cu
datasets/quadratic_programming/download_qplib_test_dataset.sh

ramakrishnap-nv

A minor suggestion

ramakrishnap-nv · 2026-07-22T21:44:15Z

 !datasets/quadratic_programming
 !datasets/quadratic_programming/**
+# downloaded QPLIB instances (see download_qplib_test_dataset.sh)
+datasets/quadratic_programming/qplib/


Suggested change
datasets/quadratic_programming/qplib/
!datasets/quadratic_programming/qplib/
Should this be

I think this is correct because we want to ignore the downloaded instances, while not ignoring the checked-in instances.

rg20

Thanks for the improvements!

mlubin · 2026-07-23T14:41:51Z

/ok to test d43e238

mlubin · 2026-07-24T01:29:06Z

/ok to test 92e20ef

coderabbitai

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@cpp/tests/dual_simplex/unit_tests/solve_barrier.cu`:
- Around line 266-269: Update the test around solve_lp and the existing solution
object to first assert optimal termination and verify the objective matches the
319.9999 reference value, then retain the iteration-count assertion. Ensure the
numerical correctness checks occur before validating that more than 50 steps
were taken.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

Push a commit to this branch (recommended)
Create a new PR with the fixes

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 6e24c9fd-4b16-4a8e-86ba-67ccfdf64b5d

📥 Commits

Reviewing files that changed from the base of the PR and between 5ea14b8 and 92e20ef.

📒 Files selected for processing (3)

cpp/include/cuopt/mathematical_optimization/constants.h
cpp/src/math_optimization/solver_settings.cu
cpp/tests/dual_simplex/unit_tests/solve_barrier.cu

🚧 Files skipped from review as they are similar to previous changes (2)

cpp/include/cuopt/mathematical_optimization/constants.h
cpp/src/math_optimization/solver_settings.cu

mlubin · 2026-07-24T01:38:48Z

/ok to test a6bcbfa

The skip heuristic in scaling() only measured row-norm imbalance, so a QP/SOCP with balanced rows but severely imbalanced columns (e.g. QPLIB_8515, where free variables appear only with ~1e-8 coefficients against O(1) rows) received no scaling at all. The resulting augmented KKT system was so ill-conditioned that barrier terminated at a suboptimal point while reporting convergence (320.13 vs the true optimum 320.00 on QPLIB_8515, non-deterministically across runs). With the column check, Ruiz runs and the instance solves to 319.999989 in 14 iterations instead of 338. Adds QPLIB_8515 as a downloaded test dataset (new download_qplib_test_dataset.sh, wired into ci/test_cpp.sh and CONTRIBUTING.md) and a full-solve regression test that asserts the optimal objective matches the QPLIB reference value. Signed-off-by: Miles Lubin <mlubin@nvidia.com>

The skip heuristic now checks both row-norm and column-norm imbalance, but the comment still described only row-norm imbalance. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Miles Lubin <mlubin@nvidia.com>

Signed-off-by: Miles Lubin <mlubin@nvidia.com>

Drop the redundant force-on solve from the toggle test (it re-treads the equilibrated path); force-off now asserts a >50-iteration blow-up, and the default-path test bounds the equilibrated solve to <30 iterations. Remove the redundant Python registry test (already covered by test_solver_settings) and trim the redundant "hidden from --help" note from the QCQP comments. Signed-off-by: Miles Lubin <mlubin@nvidia.com>

The tri-state switch in scaling() was hand-aligned in a way clang-format rejects; reformat to satisfy ci/check_style.sh. Signed-off-by: Miles Lubin <mlubin@nvidia.com>

mlubin · 2026-07-24T13:55:17Z

/ok to test 378c96b

mlubin · 2026-07-24T15:33:47Z

/merge

mlubin force-pushed the fix-ruiz-skip-column-imbalance branch 4 times, most recently from 8ba9578 to ed675a1Compare July 21, 2026 21:41

mlubin added bug Something isn't working non-breaking Introduces a non-breaking change labels Jul 21, 2026

yuwenchen95 reviewed Jul 22, 2026
View reviewed changes

chris-maes reviewed Jul 22, 2026
View reviewed changes

mlubin marked this pull request as ready for review July 22, 2026 21:37

mlubin requested review from a team as code owners July 22, 2026 21:37

mlubin requested review from chris-maes, msarahan, ramakrishnap-nv and rg20 July 22, 2026 21:38

coderabbitaiBot reviewed Jul 22, 2026
View reviewed changes

Comment threaddatasets/quadratic_programming/download_qplib_test_dataset.sh

ramakrishnap-nv approved these changes Jul 22, 2026
View reviewed changes

rg20 approved these changes Jul 23, 2026
View reviewed changes

coderabbitaiBot reviewed Jul 24, 2026
View reviewed changes

Comment threadcpp/tests/dual_simplex/unit_tests/solve_barrier.cu

mlubinand others added 3 commits July 24, 2026 09:54

Update stale comment in Ruiz equilibration skip heuristic
9022180
The skip heuristic now checks both row-norm and column-norm imbalance, but the comment still described only row-norm imbalance. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Signed-off-by: Miles Lubin <mlubin@nvidia.com>

feat(lp): register qcqp_hyper_ruiz_equilibration parameter
b8ba060
Signed-off-by: Miles Lubin <mlubin@nvidia.com>

mlubin added 3 commits July 24, 2026 09:54

feat(lp): honor qcqp_hyper_ruiz_equilibration in barrier scaling
8d7e747
Signed-off-by: Miles Lubin <mlubin@nvidia.com>

style(lp): apply clang-format to Ruiz equilibration switch
378c96b
The tri-state switch in scaling() was hand-aligned in a way clang-format rejects; reformat to satisfy ci/check_style.sh. Signed-off-by: Miles Lubin <mlubin@nvidia.com>

mlubin force-pushed the fix-ruiz-skip-column-imbalance branch from a6bcbfa to 378c96bCompare July 24, 2026 13:55

rapids-botBot merged commit 0d93a5a into NVIDIA:mainJul 24, 2026
92 checks passed

mlubin deleted the fix-ruiz-skip-column-imbalance branch July 24, 2026 15:34

	datasets/quadratic_programming/qplib/
	!datasets/quadratic_programming/qplib/

Uh oh!

Conversation

mlubin commented Jul 21, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Improvements

Neutral

Regressions

Uh oh!

copy-pr-botBot commented Jul 21, 2026

Uh oh!

mlubin commented Jul 21, 2026

Uh oh!

github-actionsBot commented Jul 21, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Test Summary

Uh oh!

yuwenchen95Jul 22, 2026

Choose a reason for hiding this comment

Uh oh!

mlubinJul 22, 2026

Choose a reason for hiding this comment

Uh oh!

chris-maesJul 22, 2026

Choose a reason for hiding this comment

Uh oh!

coderabbitaiBot commented Jul 22, 2026• edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Reviews paused

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

coderabbitaiBot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ramakrishnap-nv left a comment

Choose a reason for hiding this comment

Uh oh!

ramakrishnap-nvJul 22, 2026

Choose a reason for hiding this comment

Uh oh!

mlubinJul 23, 2026

Choose a reason for hiding this comment

Uh oh!

rg20 left a comment

Choose a reason for hiding this comment

Uh oh!

mlubin commented Jul 23, 2026

Uh oh!

mlubin commented Jul 24, 2026

Uh oh!

coderabbitaiBot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

mlubin commented Jul 24, 2026

Uh oh!

mlubin commented Jul 24, 2026

Uh oh!

mlubin commented Jul 24, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

mlubin commented Jul 21, 2026•
edited
Loading

github-actionsBot commented Jul 21, 2026•
edited
Loading

coderabbitaiBot commented Jul 22, 2026•
edited
Loading