Error while using sedumi solver with YALMIP

[Hesam]

Good morning,

While using YALMIP and SEDUMI to solve a simple SDP problem, I received an error message:

Nope, unexplained crash in SeDuMi! (could be memory issues or wrong binary)

then I used sdpsettings('debug',1) and I got the following error message:

Undefined function or variable "y_s".

Error in callsedumi (line 75)

Primal = y_s;

Error in solvesdp (line 350)

    eval(['output = ' solver.call '(interfacedata);']);

Error in optimize (line 31)

[varargout{1:nargout}] = solvesdp(varargin{:});

Error in LMI_test (line 8)

optimize(F,P(1,1),sdpsettings('solver','SEDUMI','debug',1));

 

How can I resolve this issue?

Thanks and best regards,

Hessam.

[Johan Löfberg]

Impossible to answer without code. Most often due to memory, or that the problem is too trivial. In case you cannot post reproducible code, you will have to put a break in callsedumi and step through to see the crash

[Hesam]

It does not work in general on my system. 

Here is the code:

A = [-1 2 0;-3 -4 1;0 0 -2];

P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0];

F = [F, trace(P) == 1];

optimize(F,P(1,1),sdpsettings('solver','SEDUMI','debug',1))

Thanks!

[Johan Löfberg]

so does sedumi even work?

sedumi(1,1,1)

SeDuMi 1.32 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.

Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500

eqs m = 1, order n = 2, dim = 2, blocks = 1

nnz(A) = 1 + 0, nnz(ADA) = 1, nnz(L) = 1

 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec

  0 :            8.00E+00 0.000

  1 :   7.87E-01 5.00E-01 0.000 0.0625 0.9900 0.9900   1.40  1  1  2.9E-01

  2 :   1.00E+00 1.65E-04 0.000 0.0003 0.9999 0.9999   1.09  1  1  

iter seconds digits       c*x               b*y

  2      0.6   Inf  1.0000000000e+00  1.0000000000e+00

|Ax-b| =   0.0e+00, [Ay-c]_+ =   0.0E+00, |x|=  1.0e+00, |y|=  1.0e+00

[Hesam]

I doubt it works.

I had worked already with YALMIP and SEDUMI for several years. But today I installed latest YALMIP and SEDUMI versions on my new laptop.

The complete error message that I receive from previous code is as follows:

SeDuMi 1.32 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.

Alg = 2: xz-corrector, theta = 0.250, beta = 0.500

Put 1 free variables in a quadratic cone

eqs m = 6, order n = 9, dim = 21, blocks = 4

nnz(A) = 24 + 0, nnz(ADA) = 36, nnz(L) = 21

 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec

  0 :            1.71E+01 0.000

 

SeDuMi had unexplained problems, maybe due to linear dependence?

YALMIP tweaks the problem (adds 1e6 magnitude bounds on all variables) and restarts...

 

SeDuMi 1.32 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.

Alg = 2: xz-corrector, theta = 0.250, beta = 0.500

Put 1 free variables in a quadratic cone

eqs m = 6, order n = 21, dim = 33, blocks = 4

nnz(A) = 36 + 0, nnz(ADA) = 36, nnz(L) = 21

 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec

  0 :            1.61E+07 0.000

Nope, unexplained crash in SeDuMi! (could be memory issues or wrong binary)

Make sure you have a recent and compiled version

 

 

For better diagnostics, use sdpsettings('debug',1)

 

 

Undefined function or variable "y_s".

Error in callsedumi (line 75)

Primal = y_s;

Error in solvesdp (line 350)

    eval(['output = ' solver.call '(interfacedata);']);

Error in optimize (line 31)

[varargout{1:nargout}] = solvesdp(varargin{:});

I have no idea what's wrong.

[Johan Löfberg]

Was this the output from sedumi(1,1,1) which I want you to run?

[Hesam]

No sorry, it was from the same simple SDP problem.

The result from sedumi(1,1,1) is as follows:

>> sedumi(1,1,1)

SeDuMi 1.32 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.

Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500

eqs m = 1, order n = 2, dim = 2, blocks = 1

nnz(A) = 1 + 0, nnz(ADA) = 1, nnz(L) = 1

 it :     b*y       gap    delta  rate   t/tP*  t/tD*   feas cg cg  prec

  0 :            8.00E+00 0.000

Error using maxstep

Too many input arguments.

Error in wregion (line 99)

        maxt1 = min(maxstep(dx,xc,uxc,K), maxstep(dz,zc,uzc,K));

Error in sedumi (line 477)

    [xscl,yNxt,zscl,y0Nxt, w,relt, dxmdz,err, wr] = ...

 

Thanks.

[Johan Löfberg]

so you have a function maxstep in your path which confuses sedumi

>> which maxstep -all

C:\work\solvers\sedumi\maxstep.m

[Hesam]

Thanks!

[Sarnavi Mahesh]

Good Evening, 

Was just following the instructions to remove a similar problem, using sedumi(1,1,1) gives this output:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

SeDuMi 1.32 by AdvOL, 2005-2008 and Jos F. Sturm, 1998-2003.

Alg = 2: xz-corrector, Adaptive Step-Differentiation, theta = 0.250, beta = 0.500

Undefined function or variable 'qrK'.

Error in sdinit (line 86)

vfrm.s = qrK(d.u,K);

Error in sedumi (line 401)

[d, v,vfrm, y,y0, R] = sdinit(A,b,c,dense,K,pars);

 

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I am not sure as to why it so, I have the qrK function in the same directory. Would you be able to help ? 

[Johan Löfberg]

Sounds like you don't have the suitable mex files for your computer, i.e. you haven't downloaded a version with binaries, and/or you have to compile sedumi

>> which qrK

C:\work\solvers\sedumi\qrK.mexw64

[Sarnavi Mahesh]

I downloaded from this github page, https://github.com/sqlp/sedumi

All the compiled files are readily available here, right ? or I am downloading from an incorrect repo ?

Additionally, if i try to compile sedumi on my local machine, it says some binaries are missing. 

[Johan Löfberg]

So what is your operating system and matlab version, and thus mex format (run mexext in matlab)

[Sarnavi Mahesh]

Its OS X El Capitan version 10.11.6, I am running matlab 2017b, and for mex file compilation matlab is using xcode 8.1

mexext gives 'mexmaci64'

[Johan Löfberg]

Then I don't what you are doing,. When I download the zip it has a file qrK.mexmaci64, so if the directory is in your path it should be found

[Sarnavi Mahesh]

i know, that's why it is more puzzling, the downloaded zip already contains this file. thanks. 

[Johan Löfberg]

show me

dir(fileparts(which('sedumi')))

[Sarnavi Mahesh]

did it. Saved the workspace in the above mat file, contains the struct generated by  

dir(fileparts(which('sedumi')))

Accessing the struct, i can see that qrK.mexmaci64 is present. 

[Johan Löfberg]

Then I think you should open in issue in the sedumi issue tracker