Solver not found (penbmi)

[zwjmosquito]

Hi all,

I installed penlab from their website and add the path to Matlab. I also run the test code and it shows 'ok' for all the cases.  

Running sample LMI, BMI and PMI problems...

control1            : ok

theta1              : ok

truss1              : ok

mcp100              : ok

bmi_example         : ok

bmi_f4e             : ok

pmi_example         : ok

pmi_AC1             : ok

pmi_NN4             : ok

But when use it in my code like this:

sol = optimize(F,lmd1,sdpsettings('solver','penbmi'));

it shows that 'Solver not found (penbmi)'

I'm getting so confused. Is anyone knows how to solve this problem? Thanks!!

[Johan Löfberg]

You talk about penlab, but then you write penbmi in sdpsettings...

YALMIP looks for the file penlab.m in your path. If it is in the path, it should work

[zwjmosquito]

Thanks for your quick response. I was thinking penbmi is a solver contains in the 'penlab'... Sorry I'm a beginner of Yalmip.

So when I use bmibnb as the global optimization solver and it shows:

* Upper solver     : none

is it also because I don't install penbmi? 

[Johan Löfberg]

No, you have explicitly selected 'none' as upper bound solver. Yalmip never selects that automatically

[zwjmosquito]

The way I use bmibnb is 

optimize(F,lmd1,sdpsettings('solver','bmibnb')); 

I did not select solver for upper and lower bound. Then Yalmip automatically use Mosek as Lower solver and it doesn't get any upper solver. 

[Johan Löfberg]

Really?

yalmip('clear')

x = sdpvar(1,1);

y = sdpvar(1,1);

t = sdpvar(1,1);

A0 = [-10 -0.5 -2;-0.5 4.5 0;-2 0 0];

A1 = [9 0.5 0;0.5 0 -3 ; 0 -3 -1];

A2 = [-1.8 -0.1 -0.4;-0.1 1.2 -1;-0.4 -1 0];

K12 = [0 0 2;0 -5.5 3;2 3 0];

F = [x>=-0.5, x<=2, y>=-3, y<=7];

F = [F, A0+x*A1+y*A2+x*y*K12-t*eye(3)<=0];

options = sdpsettings('solver','bmibnb');

optimize(F,t,options)

ans = 

    solvertime: 0

          info: 'No suitable solver'

       problem: -2

    yalmiptime: 0.0730

[zwjmosquito]

I tried to run your code, and the interesting thing is, now it uses penlab as an upper solver!

* Starting YALMIP global branch & bound.

* Branch-variables : 2

* Upper solver     : penlab

* Lower solver     : MOSEK

* LP solver        : MOSEK

But when I run my code again, it's still:

Starting YALMIP global branch & bound.

* Branch-variables : 5

* Upper solver     : none

* Lower solver     : MOSEK

* LP solver        : MOSEK

Any idea how this happens?

[Johan Löfberg]

We have to see your code

[zwjmosquito]

function [Xopt,LMDopt] = lambda_upper(K1,K2,A1,C1,A2,C2 )

% Calculate the value of lambda s.t. X > Phi(K1,K2,X)

F1 = A1 + K1*C1;

F2 = A2 + K2*C2;

N = length(A1);

Y = sdpvar(N,N,'symmetric');

Z1 = Y*K1;

Z2 = Y*K2;

lmd1 = sdpvar(1,1);

lmd2 = sdpvar(1,1);

F = [ 0 <= Y <= eye(N)];

F = [F, [Y sqrtm(lmd1)*(Y*A2 + Z2*C2) sqrtm(lmd2)*(Y*A1 + Z1*C1); sqrtm(lmd1)*(Y*A2 + Z2*C2) Y zeros(N,N); sqrtm(lmd2)*(Y*A1 + Z1*C1) zeros(N,N) Y] > 0];

F = [F, 0 <= lmd1 <= 1, 0 <= lmd2 <= 1, lmd1 + lmd2 == 1];

%LMDopt = bisection(F,lmd1,sdpsettings('solver','bmibnb'))

sol = optimize(F,lmd1,sdpsettings('solver','bmibnb'));

% Analyze error flags

if sol.problem == 0

else

 display('Hmm, something went wrong!');

 sol.info

 yalmiperror(sol.problem)

end

Xopt = value(Y);

LMDopt = value(lmd1);

end

[Johan Löfberg]

OK, appears to select none in some scenarios. Didn't know that

* Starting YALMIP global branch & bound.

* Branch-variables : 5

* Upper solver     : none

* Lower solver     : MOSEK

* LP solver        : GUROBI

 Node       Upper      Gap(%)       Lower    Open

    1 :    3.283E-16     0.00     -1.000E-10   2  Improved solution  

* Finished.  Cost: 3.2825e-16 Gap: 1e-08

* Timing: 0% spent in upper solver (0 problems solved)

*         2% spent in lower solver (11 problems solved)

*         60% spent in LP-based domain reduction (6 problems solved)

[zwjmosquito]

Thank you anyway. Finally I transfer it into a feasible problem and solved it using cvx! 

[Johan Löfberg]

A feasible one?

If you can solve it in cvx, you can always solve it using YALMIP as YALMIP solves a larger class of problems, so you must have created a much easier model then when you gave it to cvx

[zwjmosquito]

Yes, I assume a value for lambda, then make the problem into a feasible problem - which is, when lambda = something, can we find a Y such that the positive condition holds. And for this problem I think Yalmip also should work but cvx is much simple to understand (I still don't get how to choose a suitable solver in Yalmip because there are so many solvers).

[Johan Löfberg]

OK, was a bit confused by your choice of the word feasible then. The original problem was feasible also (a solution exists), and what you meant was that you didn't have a suitable solver setup.

Recommended solver with minimal setup time for simple LMIs with YALMIP is SDPT3. If you install that, it will automatically be picked by YALMIP when you solve LMIs