abs() is funky

[zoharl3]

Sometimes, using abs() or norm(., 1) causes infeasibility, e.g.:

x = sdpvar( n, 1, 'full' );
obj = norm( x, 1 );

However, doing it manually works fine:

t = sdpvar( n, 1 );
obj = sum( t );
con = [ dl <= t; -dl <= t ];

I think it happened both with gurobi for MILP and mosek for SOCP.
If you are interested, then I can try to come up with an example. But it's not a priority--less convenient, but I can do it manually.

[johanlofberg]

You would have to supply an example, as that is definitely not expected behavior and I've never seen such an issue before (norm(.,1) derives exactly the model you show)

btw using 'full' when defining a vector is redundant

[johanlofberg]

You change gurobi presolve options
https://yalmip.github.io/command/sdpsettings/

[Derrickzc]

You change gurobi presolve options https://yalmip.github.io/command/sdpsettings/

I only learn how to close in guorbi process.

%%params.Presolve=0; %% close presolve
%%rst = gurobi(model,params);

But in yalmip, I think I should use something like:
ops = sdpsettings ('verbose', 1, 'solver', 'gurobi', 'params.Presolve', '0')

[Derrickzc]

You change gurobi presolve options https://yalmip.github.io/command/sdpsettings/

Or like
ops = sdpsettings ('verbose', 1, 'solver', 'gurobi', 'gurobi.presolve', '0')?

[johanlofberg]

no according to the link it would be 'gurobi.presolve',0

[Derrickzc]

You change gurobi presolve options https://yalmip.github.io/command/sdpsettings/

Or like ops = sdpsettings ('verbose', 1, 'solver', 'gurobi', 'gurobi.presolve', '0')?

In case anyone find this answer, plus
ops = sdpsettings ('verbose', 1, 'solver', 'gurobi', 'gurobi.MIPGap', 1e-3)?

[zoharl3]

An advice for future reference:

https://support.gurobi.com/hc/en-us/community/posts/4407464930321-pre-solve-is-buggy