HiGHS icon indicating copy to clipboard operation
HiGHS copied to clipboard

"WARNING: Solution with objective XXX has untransformed violations: bound = ..." seems to accurately predict integer-infeasibility in the MIP.

Open jeffreydeankelly2 opened this issue 2 years ago • 5 comments

I have been researching primal heuristics with HiGHS' MIP and I have noticed that when the following message occurs:

WARNING: Solution with objective 95.5311 has untransformed violations: bound = 1.073 (col 1533); integrality = 1.386e-14; row = 0

Many other similar warning messages subsequently follow and these only appear when the MIP problem is truly integer-infeasible.

Does this make sense and should it be an error message for integer-infeasibility?

jeffreydeankelly2 avatar Dec 07 '23 14:12 jeffreydeankelly2

My understanding is that messages like this indicate an error in presolve and/or postsolve.

Specifically, the MIP solver is working on a presolved problem, but when it thinks that it has found an improving integer feasible solution, it "untransforms" it to the space of the original problem to check its feasibility and get the objective. Column/row bound or integrality violations can't be expected to be zero, since postsolve involves floating-point operations, but the should be very small, If they are meaningful like this, then it looks like an error in presolve and/or postsolve.

If you can let me have the instance that yields this warning, I can investigate.

jajhall avatar Dec 07 '23 15:12 jajhall

Yes, here is the HiGHS MPS file of the problem.

On Thu, Dec 7, 2023 at 10:49 AM Julian Hall @.***> wrote:

My understanding is that messages like this indicate an error in presolve and/or postsolve.

Specifically, the MIP solver is working on a presolved problem, but when it thinks that it has found an improving integer feasible solution, it "untransforms" it to the space of the original problem to check its feasibility and get the objective. Column/row bound or integrality violations can't be expected to be zero, since postsolve involves floating-point operations, but the should be very small, If they are meaningful like this, then it looks like an error in presolve and/or postsolve.

If you can let me have the instance that yields this warning, I can investigate.

— Reply to this email directly, view it on GitHub https://github.com/ERGO-Code/HiGHS/issues/1543#issuecomment-1845588752, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALHONEFUQ63SUWY6D6XBKG3YIHQPFAVCNFSM6AAAAABALFELBWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNBVGU4DQNZVGI . You are receiving this because you authored the thread.Message ID: @.***>

--


I M P L - " M a k i n g O p t i m i z a t i o n a n d E s t i m i z a t i o n S m a r t e r"

Jeffrey D. Kelly Industrial Algorithms Limited - i n d u s t r i @ l g o r i t h m s Email: @.*** Phone: (647) 917-4675 (IMPL) Making Industrial AI (Algorithms & Integration) Real!


This email and any files transmitted with it are confidential, proprietary and intended solely for the individual or entity to whom they are addressed. If you have received this email in error please delete it immediately.

jeffreydeankelly2 avatar Dec 07 '23 16:12 jeffreydeankelly2

I can't reproduce this behaviour with the latest version of HiGHS. What version are you using?

jajhall avatar Dec 08 '23 10:12 jajhall

1.6.0 for Windows from https://github.com/JuliaBinaryWrappers/HiGHS_jll.jl/releases

On Fri, Dec 8, 2023 at 5:37 AM Julian Hall @.***> wrote:

I can't reproduce this behaviour with the latest version of HiGHS. What version are you using?

— Reply to this email directly, view it on GitHub https://github.com/ERGO-Code/HiGHS/issues/1543#issuecomment-1846937184, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALHONEDOCDXRDSGABHLH263YILUXNAVCNFSM6AAAAABALFELBWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQNBWHEZTOMJYGQ . You are receiving this because you authored the thread.Message ID: @.***>

--


I M P L - " M a k i n g O p t i m i z a t i o n a n d E s t i m i z a t i o n S m a r t e r"

Jeffrey D. Kelly Industrial Algorithms Limited - i n d u s t r i @ l g o r i t h m s Email: @.*** Phone: (647) 917-4675 (IMPL) Making Industrial AI (Algorithms & Integration) Real!


This email and any files transmitted with it are confidential, proprietary and intended solely for the individual or entity to whom they are addressed. If you have received this email in error please delete it immediately.

jeffreydeankelly2 avatar Dec 08 '23 10:12 jeffreydeankelly2