Main Content

Advanced LMI Techniques

This last section gives a few hints for making the most out of the LMI Lab. It is directed toward users who are comfortable with the basics, as described in Tools for Specifying and Solving LMIs.

Structured Matrix Variables

Fairly complex matrix variable structures and interdependencies can be specified with lmivar. Recall that the symmetric block-diagonal or rectangular structures are covered by Types 1 and 2 of lmivar provided that the matrix variables are independent. To describe more complex structures or correlations between variables, you must use Type 3 and specify each entry of the matrix variables directly in terms of the free scalar variables of the problem (the so-called decision variables).

With Type 3, each entry is specified as either 0 or ±xn where xn is the n-th decision variable. The following examples illustrate how to specify nontrivial matrix variable structures with lmivar. The following examples show variable structures with uncorrelated and interdependent matrix variables.

Specify Matrix Variable Structures

Suppose that the variables of the problem include a 3-by-3 symmetric matrix X and a 3-by-3 symmetric Toeplitz matrix, Y, given by:


The variable Y has three independent entries, and thus involves three decision variables. Since Y is independent of X, label these decision variables n + 1, n + 2, and n + 3, where n is the number of decision variables involved in X. To retrieve this number, define the Type 1 variable X.

[X,n] = lmivar(1,[3 1]);
n = 6

The second output argument n gives the total number of decision variables used so far, which in this case is n = 6. Given this number, you can define Y.

Y = lmivar(3,n+[1 2 3;2 1 2;3 2 1]);

An equivalent expression to define Y uses the MATLAB(R) command toeplitz to generate the matrix.

Y = lmivar(3,toeplitz(n+[1 2 3]));

To confirm the variables, visualize the decision variable distributions in X and Y using decinfo.

lmis = getlmis; 
ans = 3×3

     1     2     4
     2     3     5
     4     5     6

ans = 3×3

     7     8     9
     8     7     8
     9     8     7

Specify Interdependent Matrix Variables

Consider three matrix variables, X, Y, and Z, with the following structure.


where x, y, z, and t are independent scalar variables. To specify such a triple, first define the two independent variables, X and Y, which are both Type 1.

[X,n,sX] = lmivar(1,[1 0;1 0]);
[Y,n,sY] = lmivar(1,[1 0;1 0]);

The third output of lmivar gives the entry-wise dependence of X and Y on the decision variables (x1,x2,x3,x4):=(x,y,z,t).

sX = 2×2

     1     0
     0     2

sY = 2×2

     3     0
     0     4

Using lmivar, you can now specify the structure of the Type 3 variable Z in terms of the decision variables x1=x and x4=t.

[Z,n,sZ] = lmivar(3,[0 -sX(1,1);-sY(2,2) 0]);

Because sX(1,1) refers to x1 and sY(2,2) refers to x4, this expression defines the variable Z as:


Confirm this results by checking the entry-wise dependence of Z on its decision variables.

sZ = 2×2

     0    -1
    -4     0

Complex-Valued LMIs

The LMI solvers are written for real-valued matrices and cannot directly handle LMI problems involving complex-valued matrices. However, complex-valued LMIs can be turned into real-valued LMIs by observing that a complex Hermitian matrix L(x) satisfies

L(x) < 0

if and only if


This suggests the following systematic procedure for turning complex LMIs into real ones:

  • Decompose every complex matrix variable X as

    X = X1 + jX2

    where X1 and X2 are real

  • Decompose every complex matrix coefficient A as

    A = A1 + jA2

    where A1 and A2 are real

  • Carry out all complex matrix products. This yields affine expressions in X1, X2 for the real and imaginary parts of each LMI, and an equivalent real-valued LMI is readily derived from the above observation.

For LMIs without outer factor, a streamlined version of this procedure consists of replacing any occurrence of the matrix variable X = X1 + jX2 by


and any fixed matrix A = A1 + jA2, including real ones, by


For instance, the real counterpart of the LMI system

MHXM < X,    X = XH > I(1)

reads (given the decompositions M = M1 + jM2 and X = X1 + jX2 with Mj, Xj real):


Note that X = XH in turn requires that X1=X1H and X2+X2T=0. Consequently, X1 and X2 should be declared as symmetric and skew- symmetric matrix variables, respectively.

Assuming, for instance, that MC5×5, the LMI system (Equation 1) would be specified as follows:

M1=real(M), M2=imag(M) 
bigM=[M1 M2;-M2 M1] 

% declare bigX=[X1 X2;-X2 X1] with X1=X1' and X2+X2'=0:

[X1,n1,sX1] = lmivar(1,[5 1]) 
[X2,n2,sX2] = lmivar(3,skewdec(5,n1)) 
bigX = lmivar(3,[sX1 sX2;-sX2 sX1])

% describe the real counterpart of (1.12):

lmiterm([1 1 1 0],1) 
lmiterm([-1 1 1 bigX],1,1) 
lmiterm([2 1 1 bigX],bigM',bigM) 
lmiterm([-2 1 1 bigX],1,1)

lmis = getlmis

Note the three-step declaration of the structured matrix variable bigX,


  1. Specify X1 as a (real) symmetric matrix variable and save its structure description sX1 as well as the number n1 of decision variables used in X1.

  2. Specify X2 as a skew-symmetric matrix variable using Type 3 of lmivar and the utility skewdec. The command skewdec(5,n1) creates a 5-by–5 skew-symmetric structure depending on the decision variables n1 + 1, n1 + 2,...

  3. Define the structure of bigX in terms of the structures sX1 and sX2 of X1 and X2.

See Structured Matrix Variables for more details on such structure manipulations.

Specifying cTx Objectives for mincx

The LMI solver mincx minimizes linear objectives of the form cTx where x is the vector of decision variables. In most control problems, however, such objectives are expressed in terms of the matrix variables rather than of x. Examples include Trace(X) where X is a symmetric matrix variable, or uTXu where u is a given vector.

The function defcx facilitates the derivation of the c vector when the objective is an affine function of the matrix variables. For the sake of illustration, consider the linear objective


where X and P are two symmetric variables and x0 is a given vector. If lmsisys is the internal representation of the LMI system and if x0, X, P have been declared by

x0 = [1;1] 
X = lmivar(1,[3 0]) 
P = lmivar(1,[2 1]) 
lmisys = getlmis

the c vector such that cTx=Trace(X)+x0TPx0 can be computed as follows:

n = decnbr(lmisys) 
c = zeros(n,1)

for j=1:n, 
	[Xj,Pj] = defcx(lmisys,j,X,P) 
	c(j) = trace(Xj) + x0'*Pj*x0 

The first command returns the number of decision variables in the problem and the second command dimensions c accordingly. Then the for loop performs the following operations:

  1. Evaluate the matrix variables X and P when all entries of the decision vector x are set to zero except xj: = 1. This operation is performed by the function defcx. Apart from lmisys and j, the inputs of defcx are the identifiers X and P of the variables involved in the objective, and the outputs Xj and Pj are the corresponding values.

  2. Evaluate the objective expression for X:= Xj and P:= Pj. This yields the j-th entry of c by definition.

In our example the result is

c = 

Other objectives are handled similarly by editing the following generic skeleton:

n = decnbr( LMI system ) 
c = zeros(n,1) 
for j=1:n, 
	[ matrix values ] = defcx( LMI system,j,
matrix identifiers) 
	c(j) = objective(matrix values) 

Feasibility Radius

When solving LMI problems with feasp, mincx, or gevp, it is possible to constrain the solution x to lie in the ball

xTx < R2

where R > 0 is called the feasibility radius. This specifies a maximum (Euclidean norm) magnitude for x and avoids getting solutions of very large norm. This may also speed up computations and improve numerical stability. Finally, the feasibility radius bound regularizes problems with redundant variable sets. In rough terms, the set of scalar variables is redundant when an equivalent problem could be formulated with a smaller number of variables.

The feasibility radius R is set by the third entry of the options vector of the LMI solvers. Its default value is R = 109. Setting R to a negative value means “no rigid bound,” in which case the feasibility radius is increased during the optimization if necessary. This “flexible bound” mode may yield solutions of large norms.

Well-Posedness Issues

The LMI solvers used in the LMI Lab are based on interior-point optimization techniques. To compute feasible solutions, such techniques require that the system of LMI constraints be strictly feasible, that is, the feasible set has a nonempty interior. As a result, these solvers may encounter difficulty when the LMI constraints are feasible but not strictly feasible, that is, when the LMI

L(x) ≤ 0

has solutions while

L(x) < 0

has no solution.

For feasibility problems, this difficulty is automatically circumvented by feasp, which reformulates the problem:

Find x such that

L(x) ≤ 0(2)


Minimize t subject to

Lx < t × I.

In this modified problem, the LMI constraint is always strictly feasible in x, t and the original LMI Equation 2 is feasible if and only if the global minimum tmin of Equation 2 satisfies

tmin ≤ 0

For feasible but not strictly feasible problems, however, the computational effort is typically higher as feasp strives to approach the global optimum tmin = 0 to a high accuracy.

For the LMI problems addressed by mincx and gevp, nonstrict feasibility generally causes the solvers to fail and to return an “infeasibility” diagnosis. Although there is no universal remedy for this difficulty, it is sometimes possible to eliminate underlying algebraic constraints to obtain a strictly feasible problem with fewer variables.

Another issue has to do with homogeneous feasibility problems such as

ATP + P A < 0, P > 0

While this problem is technically well-posed, the LMI optimization is likely to produce solutions close to zero (the trivial solution of the nonstrict problem). To compute a nontrivial Lyapunov matrix and easily differentiate between feasibility and infeasibility, replace the constraint P > 0-by-P > αI with α > 0. Note that this does not alter the problem due to its homogeneous nature.

Semi-Definite B(x) in gevp Problems

Consider the generalized eigenvalue minimization problem

Minimize λ subject to

A(x) < λB(x), B(x) > 0, C(x) <0.(3)

Technically, the positivity of B(x) for some xRn is required for the well-posedness of the problem and the applicability of polynomial-time interior-point methods. Hence problems where


with B1(x) > 0 strictly feasible, cannot be directly solved with gevp. A simple remedy consists of replacing the constraints

A(x) < B(x), B(x) > 0



where Y is an additional symmetric variable of proper dimensions. The resulting problem is equivalent to Equation 3 and can be solved directly with gevp.

Efficiency and Complexity Issues

As explained in Tools for Specifying and Solving LMIs, the term-oriented description of LMIs used in the LMI Lab typically leads to higher efficiency than the canonical representation

A0 + x1A1 + ... + xNAN < 0.(4)

This is no longer true, however, when the number of variable terms is nearly equal to or greater than the number N of decision variables in the problem. If your LMI problem has few free scalar variables but many terms in each LMI, it is therefore preferable to rewrite it as Equation 4 and to specify it in this form. Each scalar variable xj is then declared independently and the LMI terms are of the form xjAj.

If M denotes the total row size of the LMI system and N the total number of scalar decision variables, the flop count per iteration for the feasp and mincx solvers is proportional to

  • N3 when the least-squares problem is solved via Cholesky factorization of the Hessian matrix (default) [2].

  • M-by-N2 when numerical instabilities warrant the use of QR factorization instead.

While the theory guarantees a worst-case iteration count proportional to M, the number of iterations actually performed grows slowly with M in most problems. Finally, while feasp and mincx are comparable in complexity, gevp typically demands more computational effort. Make sure that your LMI problem cannot be solved with mincx before using gevp.

Solving M + PTXQ + QTXTP < 0

In many output-feedback synthesis problems, the design can be performed in two steps:

  1. Compute a closed-loop Lyapunov function via LMI optimization.

  2. Given this Lyapunov function, derive the controller state-space matrices by solving an LMI of the form

    M + PTXQ + QTXTP < 0(5)

    where M, P, Q are given matrices and X is an unstructured m-by-n matrix variable.

It turns out that a particular solution Xc of Equation 5 can be computed via simple linear algebra manipulations [1]. Typically, Xc corresponds to the center of the ellipsoid of matrices defined by Equation 5.

The function basiclmi returns the “explicit” solution Xc:

Xc = basiclmi(M,P,Q)

Since this central solution sometimes has large norm, basiclmi also offers the option of computing an approximate least-norm solution of Equation 5. This is done by

X = basiclmi(M,P,Q,'Xmin')

and involves LMI optimization to minimize ∥X ∥.

Related Topics