Hi,

The dualize in the last version of Yalmip is so great. The computation time on solving the dualized prblem is only half of the original one! That's great!

However, that's only happened in the variable matrix without multi-block. My model is a kind of semidfinite relaxtions of combinatorial problems which have been solved by SDPA or sedumi. And I have all the data of 'mDIM,nBLOCK,bLOCKsTRUCT,c,F'.

My question is when nBLOCK=1, that's no problem. But when nBLOCK >1, the message is :
---------------------------------------
ans =

yalmiptime: 0.6720
solvertime: 0.7950
info: 'Infeasible problem (SeDuMi-1.3)'
problem: 1
dimacs: [NaN NaN NaN NaN NaN NaN]
-------------------------------------------

Need help~~~ Thanks a lot.


Bess

Send me code and data so I can see what you are trying to do

If you have blocks, you should do something like

X1 = sdpvar(n)
X2 = sdpvar(n)
X = blkdiag(X1,X2)

constraints = [X1>0,X2>0]

etc

Note, the individual blocks have to be used, if you use X>0 and dualize will not exploit structure (I think...)

Thanks for your reply.
The code is below, and the data is attached.
-------------
clear all;

load ieee4mb;


index = size(bLOCKsTRUCT);
for i=1:index(1)
X{i} = SDPVAR(bLOCKsTRUCT(i),bLOCKsTRUCT(i));
end

% object function
obj = 0;
for i=1:index(1)
if ~isempty(F{i,1})
QX = X{i};
QF = F{i,1};
obj = -obj - QF('*QX(;
end
end

FF = [];
for i=1:index(1)
FF = [FF,X{i}>=0];
end;

Cindex = size(F); % constraints, the 1st one is belong to the object function
for j=2:Cindex(2)
CF =0;
for i=1:index(1)
if ~isempty(F{i,j})
QX = X{i};
QF = F{i,j};
CF = CF + QF('*QX(;
end
end;
FF = [FF,CF==c(j-1)];
end


FF

%% auto dual
options = sdpsettings('verbose',1,'solver','sedumi');
[D,d] = dualize(FF,obj);
solvesdp(D,-d,options)
---------------------------

send the code or put it in a code block. as you can see, it gets destroyed in plain text

...and no data is attached

Try again~~~

how I can send the data to you? it doesn't work when i try to attch the data