MIT 6.

972 Algebraic techniques and semidefinite optimization March 16, 2006

Lecture 11
Lecturer: Pablo A. Parrilo Scribe: ???

In this lecture we continue our study of SOS polynomials. After presenting a couple of applications,
we focus here on the dual side, and provide a natural probabilistic interpretation of the corresponding
problem. We further present some recent results on the density of the cone of SOS polynomials relative
to that of the nonnegative polynomials.

1 SOS applications
1.1 Lyapunov functions
The possibility of reformulating conditions for a polynomial to be a sum­of­squares as an SDP is very
useful, since we can use the SOS property in a control context as a convenient sufficient condition for
polynomial nonnegativity. Recent work has applied the sum­of­squares approach to the problem of
finding a Lyapunov function for nonlinear systems [Par00, PP02]. This approach allows one to search
over affinely parametrized polynomial or rational Lyapunov functions for systems with dynamics of the
form
ẋi (t) = fi (x(t)) for all i = 1, . . . , n
where the functions fi are polynomials or rational functions. Then the condition that the Lyapunov
function be positive, and that its Lie derivative be negative, are both directly imposed as sum­of­squares
constraints in terms of the coefficients of the Lyapunov function.
As an example, consider the following system:

ẋ = −x + (1 + x)y
ẏ = −(1 + x)x.

Using SOSTOOLS [PPP05] we easily find a quartic polynomial Lyapunov function, which after rounding
(for purely cosmetic reasons) is given by

V (x, y) = 6x2 − 2xy + 8y 2 − 2y 3 + 3x4 + 6x2 y 2 + 3y 4 .

It can be readily verified that both V (x, y) and (−V̇ (x, y)) are SOS, since
⎡ ⎤T ⎡ ⎤⎡ ⎤
x 6 −1 0 0 0 x ⎡ ⎤T ⎡ ⎤⎡ ⎤
⎢−1 8 x 10 1 −1 1 x
⎢y⎥ 0 0 −1⎥ ⎢y⎥
⎢y⎥ ⎢1 2 1 −2 ⎥⎢ y ⎥
−V˙ = ⎢
⎢ 2⎥ ⎢ ⎥ ⎢ 2⎥
⎢x ⎥
⎢0 0 3 0 0⎥
V =⎢ ⎥ ⎢x ⎥ ,
⎥ ⎢ ⎥ ⎥ ⎢ ⎥⎢ ⎥,
⎢ ⎣ x2 ⎦ ⎣−1 1 12 0 ⎦ ⎣ x2 ⎦
⎣xy ⎦ ⎣0 0 0 6 0 ⎦ ⎣xy ⎦
xy 1 −2 0 6 xy
y2 0 −1 0 0 3 y2

and the matrices in the expression above are positive definite. Similar approaches may also be used for
finding Lyapunov functionals for certain classes of hybrid systems.

1.2 Entangled states in quantum mechanics

The state of a finite­dimensional quantum system can be described in terms of a positive semidefinite
Hermitian matrix, called the density matrix. An important property of a bipartite quantum state ρ is
whether or not it is separable, which means that it can be written as a convex combination of tensor
products of rank one matrices, i.e.,
� �
ρ= pi (xi xTi ) ⊗ (yi yiT ), pi ≥ 0, pi = 1,
i i

11­1
 n1 n2
where for simplicity we have restricted ρ, xi , yi to be real. Here xi ∈ Rn1 , yi ∈ Rn2 , and ρ ∈ S+ .
How to recognize if a state is entangled or not?

Complete ToDo

2 Moments
Consider a nonnegative measure µ on R (or if you prefer, a real­valued random variable X). We can
then define the moments, which are the expectation of powers of X.
�
µk := E[X ] = xk dµ
k
(1)

What constraints, if any, should the µk satisfy? Is is true that for any set of numbers µ0 , µ1 , . . . , µk ,
there always exists a nonnegative measure having exactly these moments?
It should be apparent that some conditions are required. For instance, consider (1) for an even value
of k. Since the measure µ is nonnegative, it is clear that in this case we have µk ≥ 0.
However, that’s clearly not enough, and more restrictions should hold. A simple one can be derived by
recalling the relationship between the first and second moments and the variance of a random variable,
i.e., var(X) = E[X 2 ] − E[X]2 = µ2 − µ12 . Since the variance is always nonnegative, we should have
µ2 − µ21 ≥ 0.
How to systematically derive conditions of this kind? Notice that the previous inequality can be
obtained by noticing that for all a, b,
� �T � �� �
a 1 µ1 a
0 ≤ E[(a + bX)2 ] = a2 + 2abE[X] + b2 E[X 2 ] = ,
b µ1 µ2 b

which implies that the 2 × 2 matrix above must be positive semidefinite. Interestingly, the inequality
obtained earlier is exactly equal to the determinant of this matrix.
Exactly the same procedure can be done for higher­order moments. Proceeding this way, we have
that the higher order moments must always satisfy:
⎡ ⎤
1 µ1 µ2 ··· µd
⎢µ1
⎢ µ2 µ3 · · · µd+1 ⎥
⎥
⎢µ2
⎢ µ3 µ4 · · · µd+2 ⎥
⎥ � 0. (2)
⎢ .. .. .. . . . ⎥
.
⎣. . . . . ⎦
µd µd+1 µd+2 ··· µ2d

Notice that the diagonal elements correspond to the even­order moments, which should obviously be
nonnegative.

Necessary and sufficient, multivariate case ToDo

Remark 1. For unbounded intervals, the SDP conditions characterize the closure of the set of moments,
but not necessarily the whole set. As an example, consider the set of moments given by µ = (1, 0, 0, 0, 1),
corresponding to the Hankel matrix ⎡ ⎤
1 0 0
⎣0 0 0⎦ .
0 0 1

11­2
Although the matrix above is PSD, it is not hard to see that there is no nonnegative measure corresponding
to those moments. However, the parametrized atomic measure given by
ε4 1 ε4 1
µε = · δ(x + ) + (1 − ε4 ) · δ(x) + · δ(x − )
2 ε 2 ε
has as first five moments (1, 0, ε2 , 0, 1), and thus as ε → 0 the corresponding Hankel matrix is the one
given above.

2.1 Nonnegative measures on intervals

Just like we did for the case of polynomials nonnegative on intervals, we can similarly obtain a necessary
and sufficient characterization for moments. For simplicity, we present below only one particular case,
corresponding to the interval [−1, 1].
Lemma 2. There exists a nonnegative measure in [−1, 1] with moments (µ0 , µ1 , . . . , µ2d+1 ) if and only
if ⎡ ⎤ ⎡ ⎤
µ0 µ1 µ2 ··· µd µ1 µ2 µ4 · · · µd+1
⎢µ1
⎢ µ2 µ3 · · · µd+1 ⎥
⎥ ⎢ µ2
⎢ µ3 µ5 · · · µd+2 ⎥ ⎥
⎢µ2
⎢ µ 3 µ 4 · · · µd+2 ⎥ ± ⎢
⎥ ⎢ µ 3 µ 4 µ 6 · · · µd+3 ⎥ ⎥ � 0. (3)
⎢ .. .. .. .. . ⎥ ⎢ ..
. .. .. .. .. ⎥
⎣. . . . . ⎦ ⎣ . . . . . ⎦
µd µd+1 µd+2 · · · µ2d µd+1 µd+2 µd+3 · · · µ2d+1
Notice that the necessity is clear, since it follows from consideration of the quadratic form (in the
ai ):
� � � d �d
�d i 2
0 ≤ E (1 ± X)( i=0 ai X ) = (µj+k ± µj+k+1 )aj ak ,
j=0 k=0

where the first inequality follows since 1 ± X is always nonnegative, since X is supported on [−1, 1].
Notice the similarities (in fact, the duality) with the conditions for polynomial nonnegativity.

2.2 The moment curve

An appealing geometric interpretation of the set of valid moments is in terms of the so­called moment
curve, which is the parametric curve in Rd+1 given by t �→ (1, t, t2 , . . . , td ). Indeed, it is easy to see that
every point on the curve corresponds to a Dirac measure, where all the probability is concentrated on a
given point. Thus, every finite (or infinite) measure on the interval corresponds to a point in the convex
hull. In Figure 1 we present an illustration of the set of valid moments, for the case d = 3.

3 Bridging the gap

What to do in the cases where the set of nonnegative polynomials is no longer equal to the SOS ones? As
we will see in much more detail later, it turns out that we can approximate any semialgebraic problem
(including the simple case of a single polynomial being nonnegative) by sum of squares techniques.
As a preview, and a hint at some of the possibilities, let’s consider how to prove nonnegativity of a
particular polynomial which is not a sum of squares. Recall that the Motzkin polynomial was defined
as:
M (x, y) = x4 y 2 + x2 y 4 + 1 − 3x2 y 2 .
and is a nonnegative polynomial that is not SOS. We can try multiplying it by another polynomial which
is known to be positive, and check whether the resulting product is SOS. In this case, multiplying by
the factor (x2 + y 2 ), we can find the decomposition
(x2 + y 2 ) · M (x, y) = y 2 (1 − x2 )2 + x2 (1 − y 2 )2 + x2 y 2 (x2 + y 2 − 2)2 ,

11­3
 1

0.8

0.6

0.4

0.2

µ3
−0.2

−0.4

−0.6

−0.8 1
0.5
−1
1 0
0.5 −0.5
0 −1
µ1
µ2

Figure 1: Set of valid moments (µ1 , µ2 , µ3 ) of a probability measure on [−1, 1]. This is the convex hull
of the moment curve (t, t2 , t3 ), for −1 ≤ t ≤ 1. An explicit SDP representation is given in (3).

which clearly certifies that M (x, y) ≥ 0.

More details will follow...

References
[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.
[PP02] A. Papachristodoulou and S. Prajna. On the construction of Lyapunov functions using the
sum of squares decomposition. In Proceedings of the 41th IEEE Conference on Decision and
Control, 2002.

[PPP05] S. Prajna, A. Papachristodoulou, and P. A. Parrilo. SOSTOOLS: Sum of squares optimization

toolbox for MATLAB, 2002­05. Available from http://www.cds.caltech.edu/sostools and
http://www.mit.edu/~parrilo/sostools.

11­4