Professional Documents
Culture Documents
Sciencedirect: © 2015, Ifac (International Federation of Automatic Control) Hosting by Elsevier Ltd. All Rights Reserved
Sciencedirect: © 2015, Ifac (International Federation of Automatic Control) Hosting by Elsevier Ltd. All Rights Reserved
243
MICNON 2015
240 Sergei V. Gusev et al. / IFAC-PapersOnLine 48-11 (2015) 238–241
June 24-26, 2015. Saint Petersburg, Russia
The proof of theorem 2. The theorem 2 is a corollary Theorem 5. Let t = (t1 , . . . , tm ), A ∈ R(t)n×n , B ∈
of theorem 3. To see this let us note that due to Lagrange R(t)n×1 and G ∈ SMn+1 (R(t)). We consider the domain
celebrated theorem any positive rational number is a sum Θ ⊂ Rm consisting of the points θ = (θ1 , . . . , θm ) for which
of four squares of rational numbers. This implies that in the functions A(θ), B(θ), and G(θ) are defined. If the pair
the field of rational numbers the partial order defined (A(θ), B(θ)) is controllable for all θ ∈ Θ, then the following
in this section coincides with the unique linear order in statements are equivalent:
this field. In addition Pourchet (1971) had shown that
any non-negative polynomial in one variable with rational 1. The inequality
∗
coefficients is a sum of five squared polynomials with x x
rational coefficients. Thus, the field of rational numbers G(θ) ≥0
u u
has SOSP property.
is satisfied for all θ ∈ Θ, ω ∈ R, x ∈ Cn×1 , and u ∈ C such
that
4. HILBERT’S 17TH PROBLEM AND SOSP
PROPERTY OF THE FIELD OF RATIONAL iωx = A(θ)x + B(θ)u.
FRACTIONS.
2. There exists the matrix H ∈ SMn (R(t)) satisfying the
linear matrix inequality
One more example of the field that posses SOSP property
is closely connected with Hilbert’s 17th problem. H(θ)A(θ) + A(θ)∗ H(θ) H(θ)B(θ)
G(θ) ≥
B(θ)∗ H(θ) 0
for all θ ∈ Θ for which the matrix H(θ) is defined.
Hilbert’s 17th problem.
Given a multivariate polynomial with real coefficients that The essence of Theorem 5 can be putted in a nutshell
takes only non-negative values over the reals, can it be as follows: if the matrices A, B, and G in Yakubovich
represented as a sum of squares of rational functions? lemma depend rationally on the parameters, then the
linear matrix inequality has solution H which is also a
The positive answer on this question was given by Artin rational function of the parameters. The proof of theorem 5
in 1927. can be found in Gusev (2014).
Consider a field F of rational fractions in k variables,
F = R(µ), µ = (µ1 , . . . , µk ). One can get the following
5. EQUIVALENCE OF YAKUBOVICH STATEMENT
corollary of Artin’s result:
AND STATEMENT OF HILBERT’S 17TH PROBLEM.
Corollary . The field of rational fractions has SOSP prop-
erty. Hilbert formulated his problem for the field of reals. From
Proof. Take a polynomial in one variable p(λ) ∈ F [λ] with the Artin’s result it follows that the statement of Hilbert’s
coefficients from the field F. Suppose that for any f ∈ F, 17th problem holds for any real closed field. Artin also
there exist fractions qj ∈ F such that proved that the statement holds for the field of rational
numbers.
p(f ) = qj2 , (7)
The original formulation of the statement of Hilbert’s 17th
j
problem cannot hold in a field where positive elements
where qj depends of f. We should prove that in this case are not necessary sums of squares. But it can be shown
there exist polynomials pj ∈ F[λ] such that that accordingly generalized formulation holds in many
p= p2j . (8) such fields. This generalization of Hilbert’s 17th problem
j
is following:
The polynomial p can be considered as a rational fraction The generalized formulation of Hilbert’s 17th
p(λ, µ) in variables λ and µ. From (7) it follows that for all problem.
λ ∈ R and µ ∈ Rk p(λ, µ) ≥ 0. Then from Artin’s result it Given a multivariable polynomial with coefficients from
follows that some ordered field F, all values of which over the field F
p= rj2 , (9) are sums of squares. Can this polynomial be represented
j as a sum of squares of rational functions with coefficients
where rj are rational functions in λ and µ. To complete from F?
the prove we need the following result: It is clear that this formulation of Hilbert’s 17th problem
Theorem 4. (Artin—Cassels—Pfister). If the polynomial coincides with original one for the field of reals and for
in one variable is the sum of squares of rational functions, any field where any positive element is sum of squares.
then it is sum of squares of polynomials. The presented generalized formulation of Hilbert’s 17th
problem is a special case of problem that was formulated
The rational fractions rj can be considered as rational and investigated by Zeng (1988, 1998). From corollary of
fractions in variable λ with coefficients from the field F. theorem 3 in Zeng (1998) it follows that the generalized
Applying the theorem 4 to (9) we get (8). statement of Hilbert’s 17th problem holds in an ordered
field if and only if the field has SOSP property.
Now, using theorem 3, we can conclude that Yakubovich
statement holds when the field F is a field of rational Summing up above mentioned result and the Theorem 3
fractions. In fact, this result can be strengthened. we get the following statement.
244
MICNON 2015
June 24-26, 2015. Saint Petersburg, Russia
Sergei V. Gusev et al. / IFAC-PapersOnLine 48-11 (2015) 238–241 241
Theorem 6. The generalized Yakubovich statement and
g2(n+1−i) ,
if i = j,
the generalized statement of Hilbert’s 17th problem are 1
Gi,j = g2n+2−i−j , if |i − j| = 1,
equivalent.
2
0 otherwise
REFERENCES Taking into account that any vector z = (z1 , z2 , . . . , zn+1 )∗ ,
Gelig, A.K., Leonov, G.A., and Yakubovich, V.A. (1978). satisfying (A.1), has the form z = zn+1 (λn , λn−1 , . . . , 1)∗ ,
Stability of nonlinear systems with nonunique equilib- we get inequality
rium state. Fizmatgiz, Moscow. z ∗ Gz = zn+1
2
g(λ)0 (A.3)
Gusev, S.V. (2014). Kalman–Popov–Yakubovich lemma
for all λ, z, satisfying (A.1). Thus, there exists H ∈
for ordered fields. Automation and Remote Control, 75,
18–33. HMn (Fc ) such that (A.2) is fulfilled. Let G̃ = Re(G +
Gusev, S.V. and Likhtarnikov, A.L. (2006). Kalman– i(M ∗ HN − N ∗ HM )). Due to (A.2) G̃0 and, conse-
Popov–Yakubovich lemma and the S-procedure: A his- quently, there exist vectors hj ∈ F(n+1)×1 such that
torical essay. Automation and Remote Control, 67, 77–
G̃ = hj h∗j . (A.4)
121.
j
Iwasaki, T. and Hara, S. (2005). Generalized KYP lemma:
Unified frequency domain inequalities with design appli- From the equality in (A.3) it follows that g(λ) =
cations. IEEE Trans. Automat. Control, 50, 41–59. , 1)∗ . After substitution G̃
(λn , λn−1 , . . . , 1)G̃(λn , λn−1 , . . .
2
Kalman, R. (1963). Lyapunov functions for the problem of from (A.4) we obtain g(λ) = j fj (λ), where fj (λ) =
Lur’e in automatic control. Proc. Nat. Acad. Sci. USA, n
(λ , λ n−1
, . . . , 1)hj . This proves that the field F has SOSP
49, 201–205. property.
Kalman, R. (1970). New algebraic methods in the stability
theory. Proc. 5th Int. Conf. Nonlinear Oscillations, 2,
189–199.
Pourchet, Y. (1971). Sur la representation en somme de
carres des polynomes a une indeterminantee sur un corps
de nombres algbriques. Acta Arithm., 19, 89–104.
Tanaka, T. and Langbort, C. (2011). Symmetric formula-
tion of the Kalman-Yakubovich-Popov lemma and exact
losslessness condition. Proc. 50th IEEE CDC, 5645–
5652.
Yakubovich, V.A. (1962). Solution of some matrix inequal-
ities appearing in the automatic control theory. Dokl.
Acad. Nauk SSSR, 143, 1304–1307.
Zeng, G. (1988). A characterization of preodered fields
with the weak Hilbert property. Proc. American Math-
matical Society, 104, 335–342.
Zeng, G. (1998). A problem about the weak Hilbert
property. Acta Mathematica Sinica, New Series, 14,
481–486.
245