Thread Subject: optimization------linprog doesn't accept equality constraints

hi, all:

i use linprog solving a linear problem. however, looking
into the solution, it seems like the linprog algorithm
ignored the equality constraints. a part code of the
problem is as following:

[p,N] = size(Y);
ue = eye(p,p);
A = kron(Y',ue);
sigma = -ones(p*N,1);
A = [A,sigma];
sigma1 = zeros(1,p*p);
sigma1 = [sigma1,1];
A = -[A;sigma1];
b = zeros(p*N+1,1);

u1 = ones(1,p);
Aeq = kron(ue,u1);
sigma0 = zeros(p,1);
Aeq = [Aeq,sigma0];
beq = Wm;

f = zeros(p*p,1);
f = [f;1];

[Q,fval] = linprog(f,A,b,Aeq,beq);

where p is 3,4,5..... and N = 1000,10000... and Y>0

if N is very small, like less than 10, it's OK.
however, if N is great than 100, or even big,
then it seems like the `linprog` algorithm doesn't take the
equality constraints into consider.

could anyone give me some help?
thank you.

Jun

"Jun Li" <twofishleft@gmail.com> wrote in message
<fven6c$hfh$1@fred.mathworks.com>...

>
> could anyone give me some help?
> thank you.
>

Could you provide Y and Wm?

Bruno

thank you so much for your reply.

yes, such as

p = 3; % number of sources
a = ones(p,1); % mean weighr of each souce
N=1000; % Number of samples
x=diag(a)*rand(pmember,N);
x=x./repmat(sum(x),pmember,1);

M = rand(p,p);
Y= M*x;

u=ones(p,1);
Wm=inv(Y(:,[1:p])')*u;

this is one case of Y, and Wm.

"Jun Li" <twofishleft@gmail.com> wrote in message
<fven6c$hfh$1@fred.mathworks.com>...


> if N is very small, like less than 10, it's OK.
> however, if N is great than 100, or even big,
> then it seems like the `linprog` algorithm doesn't take the
> equality constraints into consider.
>

It seems like linprog fails and cannot converge to the
solution with large number of variables. Alternative: find a
better LP solver.

Bruno

TOMLAB /CPLEX will take care of any issues you have. If the solver
cannot solve the problem you can use the conflict refiner to figure
things out.

Best wishes, Marcus
Tomlab Optimization Inc.
http://tomopt.com/

"Marcus M. Edvall" <edvall@gmail.com> wrote in message
<3132d65e-443e-4ef9-a8ed-be3921435447@34g2000hsf.googlegroups.com>...
> TOMLAB /CPLEX will take care of any issues you have.

I wonder if you have tried OP's problem. I have tested it
using another LP code (interior point method) that has
passed hundred of problem on NETLIB test set and NEMS
problems (number variables up to 35000), and it obviously
also fails here.

I'm sure any numerical SW - including TOMLAB - has its own
limit.

Bruno

"Marcus M. Edvall" <edvall@gmail.com> wrote in message
<3132d65e-443e-4ef9-a8ed-be3921435447@34g2000hsf.googlegroups.com>...
> TOMLAB /CPLEX will take care of any issues you have.

I wonder if you have tried OP's problem. I have tested it
using another LP code (interior point method) that has
passed hundred of problem on NETLIB test set and NEMS
problems (number variables up to 35000), and it obviously
also fails here.

I'm sure any numerical SW - including TOMLAB - has its own
limit.

Bruno