I ran the LP file in CPLEX 12.10.0 interactive, both with default parameter settings and with NumericalEmphasis and ScaInd set to 1. Neither threw an error. Both solved the problem in a little over a minute. The solutions differ: the run with default parameter settings got a "solution" with a better objective value (154.97 v. 127.34) but with slack bound violations as large as 1.9e-5, as opposed to 6.8e-13 with numerical emphasis on. The numerical emphasis run saw a lot of "suspicious" bases but no unstable or ill-posed ones.
Then I ran the model in the CPLEX 12.8.0 interactive optimizer. No matter what combination of parameter settings I tried (default, numerical emphasis and scaling on, numerical emphasis on and preprocessing linear set to 0) I got error 3019 after about 100 nodes or so. The clone logs contained nothing. So maybe this is a bug in 12.8 that is fixed in 12.10, or maybe there is a numerical error that 12.10 is better at working around. I'm pasting the problem statistics below. They look pretty benign to me, except that there's an eight orders of magnitude spread in the linear coefficients of the lone quadratic constraint. So maybe that's the problem, maybe it's a (since fixed) bug, or maybe it's something else. If upgrading to 12.10 is an option, I'd recommend that.
Problem name : model_error.lp
Objective sense : Maximize
Variables : 7210 [Nneg: 6600, Free: 600, Binary: 10]
Objective nonzeros : 600
Linear constraints : 33021 [Less: 33010, Greater: 10, Equal: 1]
Nonzeros : 132030
RHS nonzeros : 11
Quadratic constraints: 1 [Less: 1]
Linear terms : 600
Quadratic terms : 600
RHS nonzeros : 1
Variables : Min LB: 0.000000 Max UB: 1.000000
Objective nonzeros : Min : 1.000000 Max : 1.000000
Linear constraints :
Nonzeros : Min : 0.008434000 Max : 803.0247
RHS nonzeros : Min : 5.000000 Max : 803.0247
Quadratic constraints:
Linear terms : Min : 0.0003607243 Max : 17252.45
Quadratic terms : Min : 0.003192250 Max : 420.6109
RHS nonzeros : Min : 926803.8 Max : 926803.8
------------------------------
Paul Rubin
------------------------------
Original Message:
Sent: Wed May 20, 2020 03:05 AM
From: Prasanna R
Subject: Failure to solve MIP subproblem in SOCP
Hi Xavier,
I have tried almost all you have listed. Still the error persists.
I have tried the following:
1) Set NumericalEmphasis to 1 and have set ScaInd (scaling parameter) to 1
2) I have improved coefficients in the data by rounding off 5 decimal places to 2
3) I ran on a different machine that too didn't work.
I am a bit skeptical about the code being correct which I checked and as far as i know I could not locate any error. And like I mentioned before only for some values of p, I encounter this error. Even the .sav and .lp files throw the same error in Interactive optimizer.
I am attaching .lp as .txt file for your perusal. Thank you for your help.
------------------------------
Prasanna R
Original Message:
Sent: Fri May 15, 2020 06:24 AM
From: Xavier Nodet
Subject: Failure to solve MIP subproblem in SOCP
Hello Prasanna,
There are a number of messages in this group that relate to Error 3019, including:
- Cplex Error 3019
- Error 3019 on a set of problems
- Error 3019: Failure to solve MIP subproblem
- a problem related to CPLEX Error 3019
I suggest you have a look at them, and come back to explain what you tried among all the suggestions given there, or why they don't apply.
------------------------------
Xavier
Original Message:
Sent: Fri May 15, 2020 02:55 AM
From: Prasanna R
Subject: Failure to solve MIP subproblem in SOCP
Hi,
I am encountering this error while solving a Second order cone program. This is a facility location type model, where for some values of p (the number of facilities to locate) it solves fine, while for some other values I encounter this error: CPLEX Error 3019: Failure to solve MIP Subproblem.
I am using Big M in this formulation, however the Big M used is the tightest one (means I can't reduce it any further without affecting feasibility). Kindly give me some suggestions as to overcome this problem. Any help will be greatly appreciated. Thank you
I am using Cplex 12.8 concert technology in C++
Regards,
------------------------------
Prasanna R
------------------------------
#DecisionOptimization