lmisolver - linear matrix inequation solver
The syntax the function evalfunc must be as follows:
[LME,LMI,OBJ]=evalfunct(X) where X is a list of matrices, LME, LMI are lists and OBJ a real scalar.
lmisolver solves the following problem:
minimize f(X1,X2,...,Xn) a linear function of Xi's
under the linear constraints: Gi(X1,X2,...,Xn)=0 for i=1,...,p and LMI (linear matrix inequalities) constraints:
Hj(X1,X2,...,Xn) > 0 for j=1,...,q
The functions f, G, H are coded in the Scilab function evalfunc and the set of matrices Xi's in the list X (i.e. X=list(X1,...,Xn) ).
The function evalfun must return in the list LME the matrices G1(X),...,Gp(X) (i.e. LME(i)=Gi(X1,...,Xn), i=1,...,p). evalfun must return in the list LMI the matrices H1(X0),...,Hq(X) (i.e. LMI(j)=Hj(X1,...,Xn) , j=1,...,q). evalfun must return in OBJ the value of f(X) (i.e. OBJ=f(X1,...,Xn) ).
lmisolver returns in XLISTF , a list of real matrices, i. e. XLIST=list(X1,X2,..,Xn) where the Xi's solve the LMI problem:
Defining Y,Z and cost by:
[Y,Z,cost]=evalfunc(XLIST) , Y is a list of zero matrices, Y=list(Y1,...,Yp) , Y1=0, Y2=0, ..., Yp=0 .
Z is a list of square symmetric matrices, Z=list(Z1,...,Zq) , which are semi positive definite Z1>0, Z2>0, ..., Zq>0 (i.e. spec(Z(j)) > 0),
cost is minimized.
lmisolver can also solve LMI problems in which the Xi's are not matrices but lists of matrices. More details are given in the documentation of LMITOOL.
//Find diagonal matrix X (i.e. X=diag(diag(X), p=1) such that
//A1'*X+X*A1+Q1 < 0, A2'*X+X*A2+Q2 < 0 (q=2) and trace(X) is maximized
n=2;A1=rand(n,n);A2=rand(n,n);
Xs=diag(1:n);Q1=-(A1'*Xs+Xs*A1+0.1*eye());
Q2=-(A2'*Xs+Xs*A2+0.2*eye());
deff('[LME,LMI,OBJ]=evalf(Xlist)','X=Xlist(1),LME=X-diag(diag(X));...
LMI=list(-(A1''*X+X*A1+Q1),-(A2''*X+X*A2+Q2)),OBJ= -sum(diag(X)) ');
X=lmisolver(list(zeros(A1)),evalf);X=X(1)
[Y,Z,c]=evalf(X)
lmitool ,