Timeline for Linearizing a constraint with square root of a variable
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 2, 2020 at 7:03 | comment | added | Johan Löfberg | McCormick envelope would almost never give a solution to the original problem. That's why it is called a relaxation, and normally is used iteratively as a core step in a global solver. If you solve the convex model I described here, it should not skyrocket as an SOCP is solved in roughly the same order of magnitude as an LP. | |
| Jul 2, 2020 at 2:19 | comment | added | tcokyasar | Well, I tried the McCormick envelope and it doesn’t provide the exact solution. So, I had to go with MIQCP formulation and as you said, the solution time for some instances skyrocketed. | |
| Jul 1, 2020 at 6:06 | comment | added | Johan Löfberg | If you can solve it as a convex problem, you should do so. Keeping the nonconvex model can cause it to go from seconds to days in terms of solution time | |
| Jun 30, 2020 at 21:13 | comment | added | tcokyasar | I guess we do not need a lot of linearization procedures after all. I was using Gurobi 8 and just realized that Gurobi 9.0 does not mind about PSD anymore. | |
| Jun 30, 2020 at 21:11 | vote | accept | tcokyasar | ||
| Jun 30, 2020 at 18:13 | comment | added | Johan Löfberg | Convex quadratic constraint? Of course. So does Mosek and Cplex. If you use a modelling language (such as YALMIP in MATLAB, disclaimer: developed by me) you simply write b>=T*mu.^2 + Asqrt(T)*mu or similar and you are done | |
| Jun 30, 2020 at 18:04 | comment | added | tcokyasar | Johan, do you know if Gurobi supports this? If yes, can you please provide a link to an example? Thanks! | |
| Jun 30, 2020 at 17:47 | history | answered | Johan Löfberg | CC BY-SA 4.0 |