Abstract
As we know, one of the main goals of this book has been to find a parametrization of the unit sphere of spaces of polynomials endowed with different norms whose unit balls can be described in \(\mathbb {R}^3\), but mainly we have tried to obtain the extreme polynomials of the unit balls. We have also studied some of the extreme polynomials in arbitrary dimensions and we have even described some of the extreme polynomials of arbitrary degree. The reason behind this is that a full description of the extreme polynomials of the unit ball has, as a matter of fact, can be applied to obtain many sharp polynomial inequalities (as we will see in this final chapter).
Access provided by Autonomous University of Puebla. Download chapter PDF
Similar content being viewed by others
If the extreme polynomials of the unit ball are known, then we can simplify the problems that involve finding sharp inequalities between norms that depend on polynomials by using a simple consequence of the Krein-Milman Theorem.
Let X be a normed space. If C is a compact convex subset of X, then C coincides with the closed convex hull of its extreme points.
FormalPara Corollary 9.1If C is a convex body in a normed space X and \(f\colon C\rightarrow \mathbb {R}\) is a convex function that attains its maximum, then there exists an extreme point p ∈ C such that \(f(p)=\max \{f(x)\colon x\in C \}\).
The main idea to apply Corollary 9.1 is the following: Let B be a convex body in a normed space of polynomials and f be a convex function defined on B which attains its maximum and takes real values, then f attains its maximum at an extreme point of B by Corollary 9.1. Furthermore, if we have a full description of the extreme points of B, then we can find the maximum of f by evaluating f in the extreme points of B (this is the Krein-Milman Approach). This can be used in the case of norms of polynomials since it is known that the norm function is convex.
The rest of this chapter involves finding well known sharp inequalities for norms of polynomials that have appeared in this survey.
Let (X, ∥⋅∥) be a normed space and consider the normed space \(\mathcal {P}({ }^n X)\) (see the beginning of Sect. 5.5). Now, let us also consider the space of continuous symmetric n-linear forms of X denoted by \(\mathcal {L}_s({ }^n X)\) and endowed with the following norm:
for every \(L\in \mathcal {L}_s({ }^n X)\). By the beginning of Sect. 5.5, for every \(P\in \mathcal {P}({ }^n X)\), there exists a unique \(L\in \mathcal {L}_s({ }^n X)\) such that P(x) = L(x, …, x), for every x ∈ X, the polar of P.
9.1 Bernstein-Markov Type Inequalities
Bernstein type inequalities for polynomials are inequalities of the following form: if \(P\in \mathcal {P}({ }^n X)\), there exists a function Ψ(x) defined over C such that
where DkP denotes the k-th derivative of P (the optimal function Ψ(x) is known as the Bernstein function). On the other hand, Markov type inequalities are of the same fashion as Bernstein type inequalities but we are also taking the supremum of ∥DkP(x)∥ over all x ∈C (the optimal constant in Markov type inequalities is known as the Markov constant). The results of this section focus on finding the Bernstein function and the Markov constant that are known for the spaces that have been presented in this survey.
Theorem 9.2 (Araújo et al. [4])
Take \(\mathcal {P}_3(\mathbb {R})\) (see Sect.2.1). The Bernstein function for the inequality
is given by
The Bernstein function for the inequality
is given by
Theorem 9.3 (Muñoz et al. [47])
Let φ: [−1, 1] → [0, +∞) be defined by \(\varphi (x)=\sqrt {1-x^2}\). On the space \(\mathcal {P}_3^\varphi (\mathbb {R})\) (see Sect.2.1.1), the Bernstein function for the inequality
is given by
Theorem 9.4 (Muñoz et al. [48])
Let \(m,n\in \mathbb N\) be odd and such that m > n. On the space \(\mathcal {P}_{m,n,\infty }(\mathbb {R})\) (see Sect.3.1), the Bernstein function for the inequality
is given by
where λ 0 comes from Theorem 3.1 and
The Markov constant is given by
and equality is attained for the polynomials
In order to prove Theorem 9.4, we will prove first the following technical lemmas.
Lemma 9.1 (Muñoz et al. [48])
Let \(m,n\in \mathbb N\) be odd and such that m > n and let λ0 be the number from Theorem 3.1. We have
Proof
Recall from Lemma 3.1 that \(|\lambda _0|<\frac {n}{m}<1\) and consider the inequality
We will show when (9.1) holds. If 0 < x < 1, then inequality (9.1) is equivalent to m − n > mxn − nxm. Now, since the function x↦mxn − nxm is strictly increasing on (0, 1), the curves y = mxn − nxm and y = m − n intersect in, at most, one point which is x = 1. Hence, it is easy to check that the inequality m − n > mxn − nxm is satisfied on (0, 1), which implies that m − n > mxn − nxm holds when \(x\in \left (0,\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\right )\) and we have proven the first inequality of the lemma. The second inequality follows after doing some simple calculations and using the fact that λ0 satisfies \(n+m\lambda _0=(m-n)|\lambda _0|{ }^{\frac {m}{m-n}}\). □
Lemma 9.2 (Muñoz et al. [48])
Let \(m,n\in \mathbb N\) be odd and such that m > n and let λ0 be the number from Theorem 3.1. If we define the functions
then g(x) ≥ f(x) provided x satisfies
Proof
By symmetry, assume that x > 0. After some calculations, it is easy to check that the functions f and g intersect at the points \(x_1=\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\) and \(x_2=\left (|\lambda _0|\frac {1-|\lambda _0|{ }^{\frac {n}{m-n}}}{1-|\lambda _0|{ }^{\frac {m}{m-n}}}\right )^{\frac {1}{m-n}}\). By Lemma 9.1, the points x1 and x2 are not in the intervals \(\left (0,\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\right )\) or \(\left (\left (\frac {n}{m}\right )^{\frac {1}{m-n}},1\right )\). Hence, either f ≥ g or f ≤ g in each one of the previous intervals. Now, notice that f(1) < g(1) and \(f\left (\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\right )<g\left (\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\right )\). Indeed, the former is trivial and the latter is true because of the following reasoning. Notice that the inequality \(f\left (\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\right )<g\left (\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\right )\) is equivalent to \(\left |\frac {\lambda _0n}{m}+1\right |<\frac {1}{|\lambda _0|{ }^{\frac {m}{m-n}}}\left |\frac {\lambda _0n}{m}-\lambda _0\right |\). Moreover, it is also equivalent to \(|\lambda _0|{ }^{\frac {m}{m-n}}<|\lambda _0|\) which is satisfied since \(-1<-\frac {n}{m}<\lambda _0<0\) (see Lemma 3.1) and the proof is complete. □
Lemma 9.3 (Muñoz et al. [48])
Let \(m,n\in \mathbb N\) be odd and such that m > n and let λ0 be the number from Theorem 3.1. If we define the functions
then \(h(x)\geq \max \{f(x),g(x) \}\) provided x satisfies
Proof
Assume that \(\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\leq |x|\leq \left (\frac {n}{m}\right )^{\frac {1}{m-n}}\) holds, then it is enough to show that h(x) ≥ f(x) and h(x) ≥ g(x).
Firstly, notice that the function xn − xm is strictly increasing on the interval \(\left (0,\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\right )\) since the derivative is positive. Hence, the maximum of x↦xn − xm on \(\left (0,\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\right )\) is attained at \(x=\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\) with value \(\frac {(m-n)n^{\frac {n}{m-n}}}{m^{\frac {m}{m-n}}}\). Thus, \(x^n-x^m\leq \frac {(m-n)n^{\frac {n}{m-n}}}{m^{\frac {m}{m-n}}}\) for \(\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\leq |x|\leq \left (\frac {n}{m}\right )^{\frac {1}{m-n}}\), which implies after rearranging the inequality that f(x) ≤ h(x).
Secondly, notice that the inequality \(\frac {mn}{n+m\lambda _0}x^{n-1}|x^{m-m}+\lambda _0|\leq n\left (\frac {n}{m}\right )^{\frac {n}{m-n}}\frac {1}{|x|}\) is equivalent to \(\frac {m}{n+m\lambda _0}|x^m+\lambda _0 x^n|\leq \left (\frac {n}{m}\right )^{\frac {n}{m-n}}\). Since the derivative of xm + λ0xn is only 0 when x = 0 or \(x=\pm \left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}}\), we have that xn + λ0xn is monotone on the interval \(\left [\left (\frac {|\lambda _0|n}{m}\right )^{\frac {1}{m-n}},\left (\frac {n}{m}\right )^{\frac {1}{m-n}}\right ]\). Hence, it is enough to evaluate xn + λ0xn at the endpoints of the interval and after some simple evaluations notice that the proof is complete. □
Proof (of Theorem 9.4 )
Notice that the Bernstein function on the space \(\mathcal {P}_{m,n,\infty }(\mathbb {R})\) is given by
However it is enough to find the above supremum over the set of extreme points of the unit ball by Corollary 9.1.
We know from Theorem 3.3 that the set of extreme points of Bm,n,∞ is
Observe that the extreme polynomials P(x) = ±1 are irrelevant to find the Bernstein function. Hence we focus our attention on the extreme polynomials
where \(t\in \left [\frac {n}{m-n},\frac {n}{n+m\lambda _0}\right ]\). Thus,
Let us define \(R(t)=mx^{n-1}\left [tx^{m-n}-\left (\frac {n}{m-n}\right )^{\frac {m-n}{m}}t^{\frac {n}{m}}\right ]\). Notice that the above supremum is attained at either \(t=\frac {n}{m-n}\), or \(t=\frac {n}{n+m\lambda _0}\), or at a critical point of R(t) inside the open interval \(\left (\frac {n}{m-n},\frac {n}{n+m\lambda _0}\right )\). It is easy to show that there exists only one critical point of R(t) which is \(t_0=\frac {n}{m-n}\left (\frac {n}{m}\right )^{\frac {m}{m-n}}\frac {1}{|x|{ }^m}\) and
Now, notice that the series of inequalities \(\frac {n}{m-n}\leq t_0\leq \frac {n}{n+m\lambda _0}\) is equivalent to
Hence, after some easy calculations, we have
where, after evaluating the function R in the above points, we have
and
By applying Lemmas 9.2 and 9.3 the result follows. □
Theorem 9.5 (Muñoz et al. [48])
Let \(m,n\in \mathbb N\) be such that m > n, m is odd and n is even. On the space \(\mathcal {P}_{m,n,\infty }(\mathbb {R})\), the Bernstein function for the inequality
is given by
The Markov constant is given by
and equality is attained for the polynomials
Theorem 9.6 (Muñoz et al. [48])
Let \(n\in \mathbb {N}\) be odd. On the space \(\mathcal {P}_{2n,n,\infty }(\mathbb {R})\) , the Bernstein function for the inequality
is given by
The Markov constant is given by 4n and equality is attained for the polynomials
Theorem 9.7 (Muñoz et al. [48])
Let \(n\in \mathbb {N}\) be even. On the space \(\mathcal {P}_{2n,n,\infty }(\mathbb {R})\) , the Bernstein function for the inequality
is given by
The Markov constant is given by 8n and equality is attained for the polynomials
Theorem 9.8 (Muñoz et al. [47])
Let \(m,n\in \mathbb {N}\) be such that m is odd, n is even and m > n. On the normed subspace of \(\mathcal {P}_{m,n,\infty }(\mathbb {R})\) given by trinomials that are bounded by the linear mapping φ(x) = |x| over the interval [−1, 1], the Bernstein function for the inequality
is given by
where \(t_1\in \mathbb {R}\) is the unique solution of
on the interval \(\left (\frac {1}{\sqrt [n]{2(n+1)}},\frac {1}{\sqrt [n]{n+1}}\right )\). The Markov constant is given by m + n + 1 and equality is attained for the polynomials
Theorem 9.9 (Muñoz et al. [47])
On the normed subspace of \(\mathcal {P}_{2,1,\infty }(\mathbb {R})\) given by trinomials that are bounded by the linear mapping φ(x) = |x| over the interval [−1, 1], the Bernstein function for the inequality
is given by
Theorem 9.10 (Muñoz et al. [49])
Let \(m,n\in \mathbb {N}\) be with different parity and such that m > n. On the space \(\mathcal {P}_{m,n,2}(\mathbb {R})\), the Bernstein function for the inequality
is given by
The Markov constant is given by
Remark 9.1
On Theorem 9.10, notice that if we consider n = 1, then we have Bernstein’s function and Markov’s constant for the space \(\mathcal {P}_2(\mathbb {R})\) (see Sect. 2.1) which are given, respectively, by
and
with equality attained for the polynomials
Theorem 9.11 (Muñoz et al. [46])
Take \(\mathcal {P}({ }^2 \Delta )\) (see Sect.4.1). The Markov constant for the inequality
is given by
and equality is attained for the polynomials
The Bernstein function for the inequality
is given by
The Markov constant is given by 6 and equality is attained for the polynomials
Theorem 9.12 (Gámez et al. [23])
Take \(\mathcal {P}({ }^2 \Box )\) (see Sect.4.2). The Bernstein function for the inequality
is given by
where α 0 is the unique root of the equation
in the interval \(\left [\frac {3-\sqrt {5}}{2},\frac {12-3\sqrt {3}}{13} \right ]\) . The Markov constant is given by
and equality is attained for the polynomials
The Bernstein function for the inequality
is given by
The Markov constant is given by 3 and equality is attained for the polynomials
Theorem 9.13 (Araújo et al. [2])
Take \(\mathcal {P}\left ({ }^2 D\left (\frac {\pi }{4}\right )\right )\) (see Sect.4.3). The Bernstein function for the inequality
is given by
where
-
(a)
\(0\leq y\leq \frac {\sqrt {2}-1}{2}x\) or \(\left (4\sqrt {2}-5 \right )x\leq y\leq x\),
-
(b)
\(\frac {\sqrt {2}-1}{2}x\leq y\leq \left (\sqrt {2}-1 \right )x\),
-
(c)
\(\left (\sqrt {2}-1 \right )x\leq y\leq \left (4\sqrt {2}-5 \right )x\).
The Markov constant is
and equality is attained for the polynomials
The Bernstein function for the inequality
is given by
The Markov constant is given by
and equality is attained for the polynomials
Theorem 9.14 (Jiménez et al. [34])
Take \(\mathcal {P}\left ({ }^2 D\left (\frac {\pi }{2}\right )\right )\) . The Bernstein function for the inequality
is given by
The Markov constant is given by \(2\sqrt {5}\) and equality is attained for the polynomials
The Bernstein function for the inequality
is given by
The Markov constant is given by 4 and equality is attained for the polynomials
Theorem 9.15 (Jiménez et al. [34])
On \(\mathcal {P}({ }^2 \ell _p^2)\) for p ∈{1, 2, ∞} (see Sects.4.3, 5.1, and 5.2), the Markov constant in the inequality
is
-
(i)
4 if p = 1,
-
(ii)
2 if p = 2,
-
(iii)
\(2\sqrt {2}\) if p = ∞.
9.2 Polarization Constants
It is easy to see just by the definition of the norms defined on \(\mathcal {P}({ }^n X)\) and \(\mathcal {L}_s({ }^n X)\) that: for every \(P\in \mathcal {P}({ }^n X)\),
where L is the polar of P. But furthermore, the converse is also true, i.e., there exists C ≥ 1 such that ∥L∥≤ C∥P∥. In particular, we have the following result that can be applied for any normed space X.
Theorem 9.16 (Martin [42])
Let X be a normed space. If \(P\in \mathcal {P}({ }^n X)\) , then
where L is the polar of P.
Notice that throughout this survey we have considered the norm over the space of n-homogeneous polynomials to be, not only defined over the unit ball of a certain normed space, but also over a convex body of a normed space. To be more precise, let X be a normed space and take C a convex body in X. We define the following norm over the space of continuous n-homogeneous polynomials of X: for every continuous n-homogeneous polynomial P,
and we also define the following norm over the space of symmetric n-linear forms of X: for every symmetric n-linear form L,
Notice that the condition “every continuous n-homogeneous polynomial P has a unique continuous symmetric n-linear form L (the polar of P) such that P(x) = L(x, …, x)” is purely algebraic. Therefore, it does not depend on the topology that we consider over the space of n-homogenous polynomials or over the space of symmetric n-linear forms.
It is easy to see by the definition of the above norms that ∥P∥C ≤∥L∥C. However, the reverse inequality as in Martin’s Theorem is not true as it can be seen later on. Furthermore, there is not yet an analogous version of Martin’s Theorem when the norm is defined over an arbitrary convex body. Thus it is still an open problem to find a result similar to the one of Martin’s Theorem when we consider the norm defined over other convex bodies apart from the unit ball of X.
We are able to define now what is known as the n-polarization constant of a space of continuous n-homogeneous polynomials on a convex body. Let X be a normed space and C ⊂ X a convex body. Let \(\mathcal {P}({ }^n\mathbf {C})\) be the space of n-homogeneous polynomials on X bounded on C endowed with the norm defined by
Similarly, if L is the polar of \(P\in \mathcal {P}({ }^n\mathbf {C})\) we define
We define the n-polarization constant \(\text{c}_{\text{pol}}(\mathcal {P}({ }^n\mathbf {C}))\) of \(\mathcal {P}({ }^n\mathbf {C})\) as the following value:
Furthermore, assume that there exists \(P\in \mathcal {P}({ }^n\mathbf {C})\) such that
where L is the polar of P, then we say that P is an extremal polynomial for \(\text{c}_{\text{pol}}(\mathcal {P}({ }^n\mathbf {C}))\).
The following results show the known exact values of the polarization constants of the spaces of homogeneous polynomials that have been dealt with in this survey (most of them use the Krein-Milman approach, specially those whose norm involve convex bodies different from the unit ball).
Theorem 9.17 (Muñoz et al. [46])
If Δ is the simplex defined in Sect.4.1, then \(\mathit{\text{c}}_{\mathit{\text{pol}}}(\mathcal {P}({ }^2\Delta ))=3\). Furthermore, P(x, y) = ±(x2 + y2 − 6xy) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}(\mathcal {P}({ }^2\Delta ))\).
Proof
The result follows from the Markov constant in Theorem 9.11 for the inequality ∥DP(x, y)∥Δ≤ Ψ(x, y)∥P∥Δ since
for all \((x,y),(u,v)\in \mathbb {R}^2\) and where L is the polar of P. □
Theorem 9.18 (Gámez et al. [23])
If \(\Box \) is the unit square defined in Sect.4.2, then \(\mathit{\text{c}}_{\mathit{\text{pol}}}(\mathcal {P}({ }^2\Box ))=\frac {3}{2}\). Furthermore, P(x, y) = ±(x2 + y2 − 3xy) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}(\mathcal {P}({ }^2\Box ))\).
Theorem 9.19 (Araújo et al. [2])
If \(D\left (\frac {\pi }{4}\right )\) is the circular sector defined in Sect.4.3, then \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2D\left (\frac {\pi }{4}\right )\right )\right )=2+\frac {\sqrt {2}}{2}\). Furthermore, \(P(x,y)=\pm (x^2+(5+4\sqrt {2})y^2-(4+4\sqrt {2})xy)\) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2D\left (\frac {\pi }{4}\right )\right )\right )\).
Theorem 9.20 (Jiménez et al. [34])
If \(D\left (\frac {\pi }{2}\right )\) is the circular sector defined in Sect.4.3, then \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2D\left (\frac {\pi }{2}\right )\right )\right )=2\). Furthermore, P(x, y) = ±(x2 + y2 − 4xy) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2D\left (\frac {\pi }{2}\right )\right )\right )\).
Theorem 9.21 (Sarantopoulos [53])
Let 1 ≤ p ≤∞. We have \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2\ell _p^2\right )\right )=2^{\frac {|p-2|}{2}}\) (see Sect.5). Furthermore, P(x, y) = ±(x2 − y2) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2\ell _p^2\right )\right )\).
Remark 9.2
It is important to mention that, although we know the extreme polynomials on the spaces \(\ell _p^2\), the proof of Theorem 9.21 in [53] does not use the Krein-Milman approach but a direct approach. It involves obtaining a sharper bound C than that of Martin’s bound for every polynomial and then finding a polynomial P such that ∥L∥C = C∥P∥C, where L is the polar of P.
An interesting question started by Harris in 1975 related to polarization constants for polynomials on ℓp spaces states that, in a complex setting we have
For the previous estimate consult [32] or [20] for a more modern and accessible exposition. The question as to whether \(\text{c}_{\text{pol}}({\mathcal P}({ }^n\ell _\infty ^n({\mathbb C})))= \frac {n^{\frac {n}{2}}(n+1)^{\frac {n+1}{2}}}{2^nn!}\) remains unsolved nowadays.
Theorem 9.22 (Kim [37])
Let w ∈ (0, 1).
-
(a)
If \(w\leq \sqrt {2}-1\), then \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )=\frac {2\left (1+w^2\right )}{\left (1+w\right )^2}\) (see Sect.6.1). Furthermore, \(P(x,y)=\pm \left (\frac {4}{(1+w)^2}xy\right )\) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )\).
-
(b)
If \(\sqrt {2}-1<w\), then \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )=1+w^2\). Furthermore, \(P(x,y)=\pm \left (x^2-y^2\right )\) are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )\).
Theorem 9.23 (Kim [39])
Let \(w=\frac {1}{2}\). We have \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {H}_{1/2}^2\right )\right )=\frac {5}{4}\) (see Sect.6.2). Furthermore,
and
are extremal polynomials for \(\mathit{\text{c}}_{\mathit{\text{pol}}}\left (\mathcal {P}\left ({ }^2 \mathcal {H}_{1/2}^2\right )\right )\).
9.3 Unconditional Constants
Let us denote by xα the monomial
where \(\mathbf {x}=(x_1,\ldots ,x_m)\in \mathbb {K}^m\) (\(\mathbb {K}=\mathbb {R}\) or \(\mathbb {C}\)) and α = (α1, …, αm) with \(\alpha _k\in \mathbb {N}\cup \{0\}\) for every k ∈{1, …, m}. For P(x) =∑|α|≤naαxα (where |α| = α1 + ⋯ + αm) a polynomial of degree n on \(\mathbb {K}^m\), we define the modulus |⋅| of P by |P|(x) =∑|α|≤n|aα|xα. If C is a convex body in \(\mathbb {R}^m\), we denote by \(\mathcal {P}({ }^n \mathbf {C})\) the space of n-homogeneous polynomials on \(\mathbb {R}^m\) endowed with the norm ∥P∥C (see Sect. 9.2). Let \(\mathcal {B}_n=\{{\mathbf {x}}^\alpha \colon |\alpha |\leq n \}\) be the canonical basis of \(\mathcal {P}({ }^n \mathbf {C})\). The unconditional constant of \(\mathcal {B}_n\) is equal to the best possible constant C (denoted by \(C_{\text{unc}}(\mathcal {P}({ }^n \mathbf {C}))\)) in the inequality
The following results show all the exact values of the unconditional constants that are known of the spaces that have been presented on this survey.
Theorem 9.24 (Grecu et al. [30])
If \(m,n\in \mathbb {N}\) with m > n, then
(see Sect.3.1) where λ0 comes from Theorem 3.1, and equality is attained for the polynomials
respectively.
Remark 9.3 (Grecu et al. [30])
In Theorem 9.24 it can be seen that for every \(k\in \mathbb {N}\) with k > 1 and every \(n\in \mathbb {N}\) even we have
which is independent of n.
Theorem 9.25 (Grecu et al. [30])
On the space \(\mathcal {P}({ }^2\Delta )\) (see Sect.4.1) we have
and equality is attained for the polynomials
Theorem 9.26 (Gámez et al. [23])
On the space \(\mathcal {P}({ }^2\Box )\) (see Sect.4.2) we have
and equality is attained for the polynomials
Theorem 9.27 (Gámez et al. [23])
On the space \(\mathcal {P}\left ({ }^2 D\left (\frac {\pi }{4}\right )\right )\) (see Sect.4.3) we have
and equality is attained for the polynomials
Theorem 9.28 (Jiménez et al. [34])
On the space \(\mathcal {P}\left ({ }^2 D\left (\frac {\pi }{2}\right )\right )\) we have
and equality is attained for the polynomials
Theorem 9.29 (Grecu et al. [30])
On the spaces \(\mathcal {P}({ }^2\ell _1^2)\), \(\mathcal {P}({ }^2\ell _2^2)\) and \(\mathcal {P}({ }^2\ell _\infty ^2)\) (see Sects.4.3, 5.1, and 5.2) we have, respectively, the unconditional constants given by
with equality attained for the polynomials
respectively.
Proof
We will prove the result for the space \(\mathcal {P}({ }^2\ell _1^2)\) since the other cases can be done analogously. By Theorem 5.2, we know that the extreme polynomials of the unit ball of \(\mathcal {P}({ }^2\ell _1^2)\) are
-
(a)
P(x, y) = ±x2 ± y2 ± 2xy,
-
(b)
\(P(x,y)=\pm \frac {\sqrt {4|t|-t^2}}{2}(x^2-y^2)+txy\), where |t|∈ (2, 4].
Notice that if P is as in (a), then \(\||P|\|{ }_{\ell _1^2}=\|P\|{ }_{\ell _1^2}=1\). Hence, it is enough to consider polynomials of type (b). If P is as in (b), then P attains its norm in \(\ell _1^2\) at \(\left (\frac {1}{2},\frac {1}{2}\right )\). Thus,
□
Theorem 9.30 (Araújo et al. [2])
Let 1 < p < ∞ with p ≠ 2 and take \(\mathcal {P}({ }^2 \ell _p^2)\) (see Sects.5.3 and 5.4). Let us define the function
and set \(M_f=\sup \left \{f(\alpha )\colon \alpha \in \left [2^{-\frac {1}{p}},1 \right ] \right \}\), we have that \(C_{\mathit{\text{unc}}}(\mathcal {P}({ }^2 \ell _p^2))=M_f\).
Theorem 9.31 (Kim [37])
Let 0 < w < 1.
-
(a)
If \(w\leq \sqrt {2}-1\), then \(\mathit{\text{c}}_{\mathit{\text{unc}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )=\frac {1+w^2+\sqrt {2\left (1+w^4\right )}}{\left (1+w\right )^2}\) (see Sect.6.1) and equality is attained for the polynomials \(P(x,y)=\pm \left (\frac {4}{(1+w)^2}xy\right )\).
-
(b)
If \(\sqrt {2}-1<w\), then \(\mathit{\text{c}}_{\mathit{\text{unc}}}\left (\mathcal {P}\left ({ }^2 \mathcal {O}_w^2\right )\right )=\frac {1+w^2+\sqrt {(1+w^2)^2+4w^2}}{2}\) and equality is attained for the polynomials
$$\displaystyle \begin{aligned} P(x,y)=\pm(\alpha x^2-\alpha y^2\pm\sqrt{\alpha(1-\alpha)}xy), \end{aligned}$$where \(\alpha =\frac {1}{2}+\frac {1+w^2}{2\sqrt {(1+w^2)^2+4w^2}}\).
Theorem 9.32 (Kim [39])
Let \(w=\frac {1}{2}\). Then, \(\mathit{\text{c}}_{\mathit{\text{unc}}}\left (\mathcal {P}\left ({ }^2 \mathcal {H}_{1/2}^2\right )\right )=\frac {3}{2}\) (see Sect.6.2) and equality is attained for the polynomials \(P(x,y)=\pm \left (x^2+\frac {1}{4}y^2+xy\right )\) and \(Q(x,y)=\pm \left (x^2+\frac {3}{4}y^2+xy\right )\).
9.4 Bohnenblust–Hille and Hardy–Littlewood Constants
We begin by considering the following constants which are closely related to the Bohnenblust–Hille and Hardy–Littlewood constants as we will see. Let α = (α1, …, αn) with \(n\in \mathbb {N}\) and let us consider the standard notation |α| = |α1| + ⋯ + |αn|. Let \(\mathcal {P}({ }^m \mathbb {K}^n)\) denote the vector space of m-homogeneous polynomials on \(\mathbb {K}^n\) (where \(\mathbb {K}=\mathbb {R}\) or \(\mathbb {C}\)). Notice that if \(P\in \mathcal {P}({ }^m \mathbb {K}^n)\), then P can be written as
where \(a_\alpha \in \mathbb {K}\) and \({\mathbf {x}}^\alpha =x_1^{\alpha _1}\cdots x_n^{\alpha _n}\) for \(\mathbf {x}=(x_1,\ldots ,x_n)\in \mathbb {K}^n\). If |⋅| is a norm on \(\mathbb {K}^n\), then |⋅| induces a norm on \(\mathcal {P}({ }^m\mathbb {K}^n)\) called the polynomial norm and it is given by
where BX is the unit ball of the normed space \(X=(\mathbb {K}^n,|\cdot |)\). The space \(\mathcal {P}({ }^m \mathbb {K}^n)\) endowed with the polynomial norm is denoted by \(\mathcal {P}({ }^m X)\). Besides the polynomial norm, there are other interesting norms on \(\mathcal {P}({ }^m\mathbb {K}^n)\) such as the ℓq-norms on the coefficients, i.e., if \(P\in \mathcal {P}({ }^m \mathbb {K}^n)\) and 1 ≤ q ≤∞, then
Let us represent by ∥⋅∥p the polynomial norm of the space \(\mathcal {P}({ }^m\ell _p^n(\mathbb {K}))\), where 1 ≤ p ≤∞. Since the space \(\mathcal {P}({ }^m \mathbb {K}^n)\) is finite dimensional, we have that the norms |⋅|q and ∥⋅∥p (1 ≤ q, p ≤∞) are equivalent, i.e., there exist k, K > 0 such that
for any \(P\in \mathcal {P}({ }^m \mathbb {K}^n)\). Notice that the unit balls of the spaces \((\mathcal {P}({ }^m\mathbb {K}^n),|\cdot |{ }_q)\) and \(\mathcal {P}({ }^m \ell _p^n(\mathbb {K}))\), denoted by \(\mathsf {B}_{|\cdot |{ }_q}\) and \(\mathsf {B}_{\|\cdot \|{ }_p}\), respectively, satisfy that the mapping \(\mathsf {B}_{|\cdot |{ }_q}\ni P\rightarrow \|P\|{ }_p\) is bounded by \(\frac {1}{k}\) and the mapping \(\mathsf {B}_{\|\cdot \|{ }_p}\ni P\rightarrow |P|{ }_q\) is bounded by K. Moreover, the continuity of such mappings and the compactness of \(\mathsf {B}_{|\cdot |{ }_q}\) and \(\mathsf {B}_{\|\cdot \|{ }_p}\) satisfy the following maxima.
Definition 9.1
Let 1 ≤ q, p ≤∞. We define the following constants
From now on, we are interested in calculating the exact values of km,n,q,p and Km,n,q,p when we are considering polynomials whose coefficients are real numbers (we will consider real polynomials and complex polynomials with real coefficients separately). To do so, we will be applying the Krein-Milman approach to the mappings \(\mathsf {B}_{|\cdot |{ }_q}\ni P\rightarrow \|P\|{ }_p\) and \(\mathsf {B}_{\|\cdot \|{ }_p}\ni P\rightarrow |P|{ }_q\). Hence, we will need, for instance, the extreme points of the unit ball \(\mathsf {B}_{|\cdot |{ }_q}\). It is well known that the extreme points of \(\mathsf {B}_{|\cdot |{ }_q}\) are
where {e1, …, em+1} stands for the canonical basis of \(\mathbb {R}^{m+1}\) and \(\mathsf {S}_{|\cdot |{ }_q}\) is the unit sphere of \((\mathbb {R}^{m+1},|\cdot |{ }_q)\).
The above problem is an extension of the polynomial Bohnenblust–Hille and Hardy–Littlewood constants problem. The m-Bohnenblust–Hille constant for polynomials is, in fact, an upper bound on \(K_{m,n,\frac {2m}{m+1},\infty }\). It was proved in [8] that if \(q\geq \frac {2m}{m+1}\), then there exists a constant Dm,q > 0 depending only on m and q such that
for any \(P\in \mathcal {P}({ }^m\ell _\infty ^n(\mathbb {K}))\) and every \(n\in \mathbb {N}\). Furthermore, any constant in the latter inequality for \(q<\frac {2m}{m+1}\) depends necessarily on n. By construction, notice that any viable choice of Dm,q satisfies \(D_{m,q}\geq \sup \{K_{m,n,q,\infty }\colon n\in \mathbb {N} \}\). This construction allows us to define the Bohnenblust-Hille constants depending on the field (\(\mathbb {R}\) or \(\mathbb {C}\)) since there are substantial differences.
Definition 9.2
The m-Bohnenblust-Hille constant for polynomials on \(\mathbb {K}\) is defined as
If \(n\in \mathbb {N}\) is fixed, then we define (m, n)-Bohnenblust-Hille constant for polynomials on \(\mathbb {K}\) as
Also, if we consider a subset E of \(\mathcal {P}({ }^m\ell _\infty ^n(\mathbb {K}))\) for some \(n\in \mathbb {N}\), then we define the (m, E)-Bohnenblust-Hille constant for polynomials on \(\mathbb {K}\) as
It is easy to see that
for all \(n\in \mathbb {N}\). A similar result to that of Bohnenblust-Hille for values of p different from ∞ can also be obtained. The proofs of the following results can be found in [1, 18]. There exist constants Cm,p and Dm,p independent of n such that
for all \(P\in ({ }^m \ell _p^n(\mathbb {K}))\) and every \(n\in \mathbb {N}\). If p = ∞, then we simply put \(\frac {2mp}{mp+p-2m}=\frac {2m}{m+1}\). Moreover, the exponents \(\frac {p}{p-m}\) for m < p ≤ 2m and \(\frac {2mp}{mp+p-2m}\) for 2m ≤ p ≤∞ are optimal in the sense that any constant H that satisfies
for all \(P\in ({ }^m \ell _p^n(\mathbb {K}))\) depends necessarily on n. The above construction allows us to define the following constants.
Definition 9.3
Let m < p ≤∞. The (m, p)-Hardy-Littlewood constant for polynomials on \(\mathbb {K}\) is defined as
for m < p ≤ 2m, and
for 2m ≤ p ≤∞. If \(n\in \mathbb {N}\) is fixed, then we define the (m, n, p)-Hardy-Littlewood constant for polynomials on \(\mathbb {K}\) as
for m < p ≤ 2m, and
for 2m ≤ p ≤∞. Also, if we consider a subset E of \(\mathcal {P}({ }^m\ell _\infty ^n(\mathbb {K}))\) for some \(n\in \mathbb {N}\), then we define
for m < p ≤ 2m, and
for 2m ≤ p ≤∞.
Notice that \(D_{\mathbb {K},m}=D_{\mathbb {K},m,\infty }\). So essentially the Hardy-Littlewood constants are in fact a generalization of the Bohnenblust-Hille constants. But furthermore, the constants Km,n,q,p are also a generalization of the Hardy-Littlewood constants since \(C_{\mathbb {K},m,p}(n)=K_{m,n,\frac {p}{p-m},p}\) for m < p ≤ 2m and \(D_{\mathbb {K},m,p}(n)=K_{m,n,\frac {2mp}{mp+p-2m},p}\) for 2m ≤ p ≤∞. Hence we have
This section is about providing some of the constants km,n,q,p, Km,n,q,p, and in particular, the Hardy-Littlewood and Bohnenblust-Hille constants, that have been obtained through the Krein-Milman approach.
9.4.1 On the Complex Case
Assume that \(\mathbb {K}=\mathbb {C}\).
Theorem 9.33 (Jiménez et al. [33])
Let \(E_{\mathbb {R}}\) be the real subspace of \(\mathcal {P}({ }^2 \ell _\infty ^2(\mathbb {C}))\) given by \(\{az^2+bw^2+czw\colon (a,b,c)\in \mathbb {R}^3 \}\) . We have
with extremal polynomials
9.4.2 On the Real Case
Assume that \(\mathbb {K}=\mathbb {R}\). All the results that are presented have been obtained for the cases when m = n = 2.
Theorem 9.34 (Jiménez et al. [33])
Let \(f\colon \left [\frac {1}{2},1 \right ]\rightarrow \mathbb {R}\) be given by
We have
where
In particular, the exact value of f(t0) is given by
where
and
Moreover, the following polynomials are extremal
Theorem 9.35 (Araújo et al. [3])
If q, p ∈{1, ∞}, then
with extremal polynomials given, respectively, by
Theorem 9.36 (Araújo et al. [3])
If q, p ∈{1, ∞}, then
with extremal polynomials given, respectively, by
Theorem 9.37 (Araújo et al. [3])
For every q ∈ [1, ∞), let \(f_{q,1}\colon [2,4]\rightarrow \mathbb {R}\) and \(f_{q,\infty }\colon \left [\frac {1}{2},1\right ]\rightarrow \mathbb {R}\) be given by
We have
In particular, K2,2,q,1 = 4 and \(K_{2,2,q,\infty }=2^{\frac {1}{q}}\) for every q ≥ 2, with extremal polynomials given, respectively, by
Remark 9.4 (Araújo et al. [3])
The exact value of the maximum of the functions fq,1 and fq,∞ or the points of attainment of the maximum seems to be a much harder task. However, by using the symbolic calculus tool of MATLAB, we are able to obtain the exact values where the functions reach its maximum for certain values of q. For instance, for \(q=\frac {4}{3}\), the maximum of fq,1(t) and fq,∞(t) is attained at
and
respectively. Also, for \(q=\frac {3}{2}\), the maximum of fq,1(t) is attained at
where
And also for \(q=\frac {3}{2}\), the maximum of fq,∞(t) is attained at
where
and
Theorem 9.38 (Araújo et al. [3])
If p ∈ (1, ∞), then
with extremal polynomials given, respectively, by
Theorem 9.39 (Araújo et al. [3])
For every q ≥ 1 and p ≥ 2, let \(f_{q,p}\colon [0,1]\rightarrow \mathbb {R}\) be given by
We have
See also [13] in connection to the previous result.
Corollary 9.2 (Araújo et al. [3])
For 4 ≤ p ≤∞, we have
Theorem 9.40 (Araújo et al. [3])
If q > 1, then
with extremal polynomials given by
where \(a_0=\left (1+2^{\frac {2(1-q)}{q-2}}\right )^{-\frac {1}{2}}\).
Theorem 9.41 (Araújo et al. [3])
If q, p > 2, then
If fq,p is as in Theorem 9.39 and q, p > 2, then the following polynomials are extremal
Corollary 9.3 (Araújo et al. [3])
If p ≥ 2, then
with extremal polynomials given by
Corollary 9.4 (Araújo et al. [3])
For 2 < p ≤ 4, we have
It is important to mention that Corollary 9.4 was first proven in [13].
Corollary 9.5 (Araújo et al. [3])
We have
with all extremal polynomials given by
with α, β ≥ 0 and α4 + β4 = 1.
Theorem 9.42 (Araújo et al. [3])
For p > 2, let \(f_{1,p}\colon \left [0,\frac {1}{2}\right ]\rightarrow \mathbb {R}\) be defined by
We have
Remark 9.5 (Araújo et al. [3])
The exact calculation of the above supremum seems to be a harder task. However, by using the symbolic calculus tool of MATLAB, we can obtain the exact value of the supremum of f1,p(t) as well as the point where it attains its maximum for certain values of p. For p = 4, the function f1,4(t) attains its maximum on \(\left [0,\frac {1}{2}\right ]\) at \(t=\frac {3-2\sqrt {2}}{6}\) and, therefore, \(K_{2,1,4}=\sqrt {6}\).
References
N. Albuquerque, F. Bayart, D. Pellegrino, J.B. Seoane-Sepúlveda, Optimal Hardy-Littlewood type inequalities for polynomials and multilinear operators. Israel J. Math. 211(1), 197–220 (2016)
G. Araújo, P. Jiménez-Rodríguez, G.A. Muñoz-Fernández, J.B. Seoane-Sepúlveda, Polynomial inequalities on the π/4-circle sector. J. Convex Anal. 24(3), 927–953 (2017)
G. Araújo, P. Jiménez-Rodríguez, G.A. Muñoz-Fernández, J.B. Seoane-Sepúlveda, Equivalent norms in polynomial spaces and applications. J. Math. Anal. 445, 1200–1220 (2017)
G. Araújo, G.A. Muñoz-Fernández, D.L. RodríguezVidanes, J.B. Seoane-Sepúlveda, Sharp Bernstein inequalities using convex analysis techniques. Math. Inequal. Appl. 23(2), 725–750 (2020). https://doi.org/10.7153/mia-2020-23-61
H.F. Bohnenblust, E. Hille, On the absolute convergence of Dirichlet series. Ann. Math. (2) 32(3), 600–622 (1931)
W. Cavalcante, D. NúñezAlarcón, D. Pellegrino, The optimal Hardy-Littlewood constants for 2-homogeneous polynomials on \(\ell _{p}(\mathbb {R}^{2})\) for 2 < p < 4 are 22∕p. J. Math. Anal. 445, 1200–1220 (2017)
V. Dimant, P. Sevilla-Peris, Summation of coefficients of polynomials on ℓp spaces. Publ. Mat. 60(2), 289–310 (2016)
S. Dineen, Extreme integral polynomials on a complex Banach space. Math. Scand. 92(1), 129–140 (2003)
J.L. Gámez-Merino, G.A. Muñoz-Fernández, V.M. Sánchez, J.B. Seoane-Sepúlveda, Inequalities for polynomials on the unit square via the Krein-Milman theorem. J. Convex Anal. 20(1), 125–142 (2013)
B.C. Grecu, G.A. Muñoz-Fernández, J.B. Seoane-Sepúlveda, Unconditional constants and polynomial inequalities. J. Approx. Theory 161(2), 706–722 (2009). https://doi.org/10.1016/j.jat.2008.12.001
L.A. Harris, Bounds on the derivatives of holomorphic functions of vectors, in Analyse fonctionnelle et applications (Comptes Rendus Colloq. Analyse, Inst. Mat., Univ Federal Rio de Janeiro, Rio de Janeiro, 1972) (Hermann, Paris, 1975), pp. 145–163. Actualités Aci. Indust., No. 1367. MR0477773
P. Jiménez-Rodríguez, G.A. Muñoz-Fernández, M. MurilloArcila, J.B. Seoane- Sepúlveda, Sharp values for the constants in the polynomial Bohnenblust-Hille inequality. Linear Multilinear Algebra 64(9), 1731–1749 (2015)
P. Jiménez-Rodríguez, G.A. Muñoz-Fernández, D. Pellegrino, J.B. Seoane-Sepúlveda, Bernstein-Markov type inequalities and other interesting estimates for polynomials on circle sectors. Math. Inequal. Appl. 20(1), 285–300 (2017). https://doi.org/10.7153/mia-20-21
S.G. Kim, Polarization and unconditional constants of \(\mathcal {P}({ }^{2}d_{\ast }(1, w)^{2})\). Commun. Korean Math. Soc. 29(3), 421–428 (2014). https://doi.org/10.4134/CKMS.2014.29.3.421
S.G. Kim, Extreme 2-homogeneous polynomials on the plane with a hexagonal norm and applications to the polarization and unconditional constants. Stud. Sci. Math. Hungar. 54(3), 362–393 (2017)
M. Krein, D. Milman, On extreme points of regular convex sets. Stud. Math. 9(1), 133–138 (1940)
R.S. Martin, Contributions to the Theory of Functionals. Dissertation (Ph.D.), Cal. Inst. of Tech. (1932). https:/doi.org/10.7907/9rtv-x659
G.A. Muñoz-Fernández, S. Gy Révész, J.B. Seoane-Sepúlveda, Geometry of homogeneous polynomials on non symmetric convex bodies. Math. Scand. 105(1), 147–160 (2009). https://doi.org/10.7146/math.scand.a-15111
G.A. Muñoz-Fernández, V.M. Sánchez, J.B. Seoane-Sepúlveda, Estimates on the derivative of a polynomial with a curved majorant using convex techniques. J. Convex Anal. 17(1), 241–252 (2010)
G.A. Muñoz-Fernández, V.M. Sánchez, J.B. Seoane-Sepúlveda, Lp-analogues of Bernstein and Markov inequalities. Math. Inequal. Appl. 14(1), 135–145 (2011). https://doi.org/10.7153/mia-14-11
G.A. Muñoz-Fernández, Y. Sarantopoulos, J.B. Seoane-Sepúlveda, An application of the Krein-Milman theorem to Bernstein and Markov inequalities. J. Convex Anal. 15(2), 299–312 (2008)
Y. Sarantopoulos, Estimates for polynomial norms on Lp(μ) spaces. Math. Proc. Camb. Philos. Soc. 99(2), 263–271 (1986)
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2022 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this chapter
Cite this chapter
Ferrer, J., García, D., Maestre, M., Muñoz, G.A., Rodríguez, D.L., Seoane, J.B. (2022). Applications. In: Geometry of the Unit Sphere in Polynomial Spaces. SpringerBriefs in Mathematics. Springer, Cham. https://doi.org/10.1007/978-3-031-23676-1_9
Download citation
DOI: https://doi.org/10.1007/978-3-031-23676-1_9
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-23675-4
Online ISBN: 978-3-031-23676-1
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)