MIT 6.

972 Algebraic techniques and semidefinite optimization May 2, 2006

Lecture 19
Lecturer: Pablo A. Parrilo Scribe: ???

Today we continue with some additional aspects of quantifier elimination. We will then recall the
Positivstellensatz and its relations with semidefinite programming. After introducing copositive matrices,
we present Pólya’s theorem on positive forms on the simplex, and the associated relaxations. Finally,
we conclude with an important result due to Schmüdgen about representation of positive polynomials
on compact sets.

1 Certificates

Talk about certificates in QE ToDo

2 Psatz revisited
Recall the statement of the Positivstellensatz.
Theorem 1 (Positivstellensatz). Consider the set S = {x ∈ Rn | fi (x) ≥ 0, hi (x) = 0}. Then,
⎧
⎨ f + h = −1
S=∅ ⇔ ∃f, h ∈ R[x] s.t. f ∈ cone{f1 , . . . , fs }
h ∈ ideal{h1 , . . . , ht }
⎩

Once again, since the conditions on the polynomials f, h are convex and affine, respectively, by re­
stricting their degree to be less than or equal to a given bound d we have a finite­dimensional semidefinite
programming problem.

2.1 Hilbert 17th problem

As we have seen, in the general case nonnegative multivariate polynomials can fail to be a sum of squares
(the Motzkin polynomial being the classical counterxample). As part of his famous list of twenty­three
problems that he presented at the International Congress of Mathematicians in 1900, David Hilbert
asked the following1 :

17. Expression of definite forms by squares. A rational integral function or form

in any number of variables with real coefficient such that it becomes negative for no real
values of these variables, is said to be definite. The system of all definite forms is invariant
with respect to the operations of addition and multiplication, but the quotient of two definite
forms in case it should be an integral function of the variables is also a definite form. The
square of any form is evidently always a definite form. But since, as I have shown, not every
definite form can be compounded by addition from squares of forms, the question arises
which I have answered affirmatively for ternary forms whether every definite form may not
be expressed as a quotient of sums of squares of forms. At the same time it is desirable, for
certain questions as to the possibility of certain geometrical constructions, to know whether
the coefficients of the forms to be used in the expression may always be taken from the realm
of rationality given by the coefficients of the form represented.
1 This text was obtained from http://aleph0.clarku.edu/ djoyce/hilbert/, and corresponds to Newson’s translation
~
of Hilbert’s original German address. In that website you will also find links to the current status of the problems, as well
as the original German text.

19­1
 In other words, can we write every nonnegative polynomial as a sum of squares of rational functions?
As we we show next, this is a rather direct consequence of the Psatz. Of course, it should be clear (and
goes without saying) that we are (badly) inverting the historical order! In fact, much of the motivation
for the development of real algebra came from Hilbert’s question.
How can we use the Psatz to prove that a polynomial p(x) is nonnegative? Clearly, p is nonnegative
if and only if the set {x ∈ Rn | p(x) < 0} is empty. Since our version of the Psatz does not allow for
strict inequalities (there are slightly more general, though equivalent, formulations that do), we’ll need
a useful trick discussed earlier (“Rabinowitch’s trick”). Introducing a new variable z, the nonnegativity
of p(x) is equivalent to the emptiness of the set described by

−p(x) ≥ 0, 1 − zp(x) = 0.

The Psatz can be used to show that this holds if and only if there exist polynomials s0 , s1 , t ∈ R[x, z]
such that
s0 (x, z) − s1 (x, z) · p + t(x, z) · (1 − zp) = −1,
where s0 , s1 are sums of squares. Replace now z → 1/p(x), and multiply by p2k (where k is sufficiently
large) to obtain
s̃0 − s̃1 · p = −p2k ,
where s̃0 , s̃1 are sums of squares in R[x]. Solving now for p, we have:

s̃0 (x) + p(x)2k s̃1 (x)(s̃0 (x) + p(x)2k )

p(x) = = ,
s̃1 (x) s̃21 (x)

and since the numerator is a sum of squares, it follows that p(x) is indeed a sum of squares of rational
functions.

3 Copositive matrices and Pólya’s theorem

An interesting class of matrices are the so­called copositive matrices, which are those for which the
associated quadratic form is nonnegative on the nonnegative orthant.
Definition 2. A matrix M ∈ S n is copositive is it satisfies

xT M x ≥ 0, for all xi ≥ 0.

As opposed to positive semidefiniteness, which can be checked in polynomial time, the recognition
problem for copositive matrices is an NP­hard problem. The set of copositive is a closed convex cone,
for which checking membership is a difficult problem.
There are many interesting applications of copositive matrices. Among others, we mention:
• Consider a graph G, with A being its the adjacency matrix. The stability number α of the graph
G is equal to the cardinality of its largest stable set. By a result of Motzkin and Straus, it is known
that it can be obtained as:
P
1
= min xT (I + A)x
α(G) xi ≥0, i xi =1

This implies that α(G) ≤ γ if and only if the matrix γ · (I + A) − eeT is copositive.
• Another interesting application of copositive matrices is in the performance analysis of queueing
networks; see e.g. [KM96]. Modulo some (important) details, the basic idea is to use a quadratic
function xT M x as a Lyapunov function, where the matrix M is copositive and x represents the
lengths of the queues.

19­2
 Add details + others ToDo

An important related result is Pólya’s theorem on positive forms on the simplex:

Theorem 3 (P´ olya). Consider a homogeneous polynomial �n in n variables of degree d, that is strictly
positive in the unit simplex Δn := {x ∈ Rn | xi ≥ 0, i=1 xi = 1}. Then, for large enough k, the
polynomial (x1 + · · · + xn )k p(x) has nonnegative coefficients.
We can provide a natural hierarchy of sufficient conditions for a matrix to be copositive. Completeness
of this hierarchy follows directly from Pólya’s theorem [Par00].
There are some very interesting connections between Pólya’s result and a foundational theorem in
probability known as De Finetti’s exchangeability theorem.

4 Positive polynomials
The Positivstellensatz allows us to obtain certificates of the emptiness of a basic semialgebraic set,
explicitly given by polynomials.
What if we want to apply this for optimization? As we have seen, it is relatively straightforward to
convert an optimization problem to a family of feasibility problems, by considering the sublevel sets, i.e.,
the sets {x ∈ Rn | f (x) ≤ γ}.
In the general case of constrained problems, however, using the Psatz we will require conditions that
are not linear in the unknown parameter γ (because we need products between the contraints), and this
presents a difficulty to the direct use of SDP. Notice nevertheless, that the problem is certainly an SDP
for any fixed value of γ, and it thus quasiconvex (which is almost as good, except for the fact that we
cannot use “standard” SDP solvers to solve it directly, but rather rely on methods such as bisection).
Theorem 4 ([Sch91]). If p(x) is strictly positive on K = {x ∈ Rn | fi (x) ≥ 0}, and K is compact, then
p(x) ∈ cone{f1 , . . . , fs }.
In the next lecture we will describe the basic elements of Schmüdgen’s proof. His approach com­
bines both algebraic tools (using the Positivstellensatz to prove the boundedness of certain operators)
and functional analysis (spectral measures of commuting families of operators and the Hahn­Banach
theorem). We will also describe some alternative versions due to Putinar, as well as a related purely
functional­analytic result due to Megretski.
For a comprehensive treatment and additional references, we mention [BCR98, Mar00, PD01] among
others.

References
[BCR98] J. Bochnak, M. Coste, and M­F. Roy. Real Algebraic Geometry. Springer, 1998.
[KM96] P. R. Kumar and S. P. Meyn. Duality and linear programs for stability and performance
analysis of queuing networks and scheduling policies. IEEE Trans. Automat. Control, 41(1):4–
17, 1996.

[Mar00] M. Marshall. Positive polynomials and sums of squares. Dottorato de Ricerca in Matematica.
Dept. di Mat., Univ. Pisa, 2000.
[Par00] P. A. Parrilo. Structured semidefinite programs and semialgebraic geometry methods in robust­
ness and optimization. PhD thesis, California Institute of Technology, May 2000. Available at
http://resolver.caltech.edu/CaltechETD:etd­05062004­055516.

19­3
[PD01] A. Prestel and C. N. Delzell. Positive polynomials: from Hilbert’s 17th problem to real algebra.
Springer Monographs in Mathematics. Springer, 2001.
[Sch91] K. Schmüdgen. The K­moment problem for compact semialgebraic sets. Math. Ann., 289:203–
206, 1991.

19­4