93 lines
2.7 KiB
Mathematica
93 lines
2.7 KiB
Mathematica
|
function y = blkdiag(varargin)
|
||
|
|
%BLKDIAG (overloaded)
|
||
|
|
|
||
|
|
if nargin<2
|
||
|
|
y=varargin{1};
|
||
|
|
return
|
||
|
|
end
|
||
|
|
|
||
|
|
% Get dimensions
|
||
|
|
n = zeros(length(varargin),1);
|
||
|
|
m = zeros(length(varargin),1);
|
||
|
|
isasdpvar = zeros(length(varargin),1);
|
||
|
|
%Symmetric = zeros(nargin,1);
|
||
|
|
for i = 1:length(varargin)
|
||
|
|
if isa(varargin{i},'sdpvar')
|
||
|
|
isasdpvar(i) = 1;
|
||
|
|
n(i)=varargin{i}.dim(1);
|
||
|
|
m(i)=varargin{i}.dim(2);
|
||
|
|
else
|
||
|
|
[n(i) m(i)] = size(varargin{i});
|
||
|
|
end
|
||
|
|
end
|
||
|
|
|
||
|
|
% Find all free variables used
|
||
|
|
all_lmi_variables = [];
|
||
|
|
for i = 1:length(varargin)
|
||
|
|
all_lmi_variables = [all_lmi_variables getvariables(varargin{i})];
|
||
|
|
end
|
||
|
|
all_lmi_variables = unique(all_lmi_variables);
|
||
|
|
|
||
|
|
% Create an SDPVAR
|
||
|
|
y=sdpvar(1,1,'rect',all_lmi_variables,[]);
|
||
|
|
|
||
|
|
|
||
|
|
% Some indexation tricks
|
||
|
|
msums = cumsum([0; m]);
|
||
|
|
nsums = cumsum([0; n]);
|
||
|
|
summ=sum(m);
|
||
|
|
sumn=sum(n);
|
||
|
|
indextable = reshape(1:sumn*summ,sumn,summ);
|
||
|
|
|
||
|
|
is = [];
|
||
|
|
js = [];
|
||
|
|
ss = [];
|
||
|
|
for j = 1:length(varargin)
|
||
|
|
nnindex = indextable(1+nsums(j):nsums(j+1),1+msums(j):msums(j+1));
|
||
|
|
if isasdpvar(j)
|
||
|
|
this_uses = find(ismembc(all_lmi_variables,varargin{j}.lmi_variables));
|
||
|
|
mindex = [1 this_uses+1];
|
||
|
|
|
||
|
|
[a,b,d] = find(varargin{j}.basis.');
|
||
|
|
is = [is(:);reshape(mindex(a),[],1)];
|
||
|
|
js = [js(:);reshape(nnindex(b),[],1)];
|
||
|
|
ss = [ss(:);d(:)];
|
||
|
|
else
|
||
|
|
[a,b,d] = find( varargin{j}(:).');
|
||
|
|
is = [is;ones(length(a),1)];
|
||
|
|
js = [js;reshape(nnindex(b),[],1)];
|
||
|
|
ss = [ss(:);d(:)];
|
||
|
|
end
|
||
|
|
end
|
||
|
|
|
||
|
|
y.basis = sparse(js,is,ss,sum(m)*sum(n),1+length(all_lmi_variables));
|
||
|
|
y.dim(1) = sumn;
|
||
|
|
y.dim(2) = summ;
|
||
|
|
% Reset info about conic terms
|
||
|
|
y.conicinfo = [0 0];
|
||
|
|
|
||
|
|
y = unfactor(y);
|
||
|
|
% Update the factors
|
||
|
|
doublehere = 0;
|
||
|
|
for i = 1:length(varargin)
|
||
|
|
if isa(varargin{i},'sdpvar')
|
||
|
|
if length(varargin{i}.leftfactors)==0
|
||
|
|
y = flush(y);
|
||
|
|
return
|
||
|
|
end
|
||
|
|
for j = 1:length(varargin{i}.leftfactors)
|
||
|
|
y.rightfactors{end+1} = [zeros(size(varargin{i}.rightfactors{j},1),sum(m(1:1:i-1))) varargin{i}.rightfactors{j} zeros(size(varargin{i}.rightfactors{j},1),sum(m(i+1:1:end)))];
|
||
|
|
y.leftfactors{end+1} = [zeros(sum(n(1:1:i-1)),size(varargin{i}.leftfactors{j},2)); varargin{i}.leftfactors{j}; zeros(sum(n(i+1:1:end)),size(varargin{i}.leftfactors{j},2))];
|
||
|
|
y.midfactors{end+1} = varargin{i}.midfactors{j};
|
||
|
|
end
|
||
|
|
elseif isnumeric(varargin{i})
|
||
|
|
here = length(y.midfactors)+1;
|
||
|
|
y.rightfactors{here} = [zeros(m(i),sum(m(1:1:i-1))) eye(m(i)) zeros(m(i),sum(m(i+1:1:end)))];
|
||
|
|
y.leftfactors{here} = [zeros(sum(n(1:1:i-1)),size(varargin{i},1)); eye(size(varargin{i},1)); zeros(sum(n(i+1:1:end)),size(varargin{i},1))];
|
||
|
|
y.midfactors{here} = varargin{i};
|
||
|
|
end
|
||
|
|
end
|
||
|
|
y = cleandoublefactors(y);
|
||
|
|
|
||
|
|
|