Lp norms of trigonometric polynomials

Jordan Bell
April 3, 2014

1 Introduction

A trigonometric polynomial of degree n is an expression of the form

k=-nnckeikt,ck.

Using the identity eit=cost+isint, we can write a trigonometric polynomial of degree n in the form

a0+k=1nakcoskt+k=1nbksinkt,ak,bk.

For 1p< and for a 2π-periodic function f, we define the Lp norm of f by

fp=(12π02π|f(t)|p𝑑t)1/p.

For a continuous 2π-periodic function f, we define the L norm of f by

f=max0t2π|f(t)|.

If f is a continuous 2π-periodic function, then there is a sequence of trigonometric polynomials fn such that f-fn0 as n [31, p. 54, Corollary 5.4].

If 1p< and f is a continuous 2π-periodic function, then

fp=(12π02π|f(t)|p𝑑t)1/p(12π02πfp𝑑t)1/p=f.

Jensen’s inequality [16, p. 44, Theorem 2.2] (cf. [30, p. 113, Problem 7.5]) tells us that if ϕ:[0,) is convex, then for any function h:[0,2π][0,) we have

ϕ(12π02πh(t)𝑑t)12π02πϕ(h(t))𝑑t.

If 1p<q<, then ϕ:[0,) defined by ϕ(x)=xq/p is convex. Hence, if 1p<q< then for any 2π-periodic function f,

fp = (ϕ(fpp))1/q
= (ϕ(12π02π|f(t)|p𝑑t))1/q
(12π02πϕ(|f(t)|p)𝑑t)1/q
= (12π02π|f(t)|q𝑑t)1/q
= fq.

The Dirichlet kernel Dn is defined by

Dn(t)=k=-nneikt=1+2k=1ncoskt.

One can show [14, p. 71, Exercise 1.1] that

Dn1=4π2logn+O(1).

(On the other hand, it can quickly be seen that Dn=2n+1, and it follows from Parseval’s identity that Dn2=2n+1.)

Pólya and Szegő [27, Part VI] present various problems about trigonometric polynomials together with solutions to them. A result on L norms of trigonometric polynomials that Pólya and Szegő present is for the sum An(t)=k=1nsinktk. The local maxima and local minima of An can be explicitly determined [27, p. 74, no. 23], and it can be shown that [27, p. 74, no. 25]

An0πsintt𝑑t.

2 Lp norms

If 1p<q<, then [14, p. 123, Exercise 1.8] (cf. [7, p. 102, Theorem 2.6]) there is some C(p,q) such that for any trigonometric polynomial f of degree n, we have

fqC(p,q)n1p-1qfp.

This inequality is sharp [33, p. 230]: for 1p<q< there is some C(p,q) such that if Fn(t)=1nk=0n-1Dk(t) (Fn is called the Fejér kernel) then

Fnq>C(p,q)n1p-1qFnp.
Theorem 1.

Let 1pq. If f^(j)=0 for |j|>n+1 then

fq5(n+1)1p-1qfp.
Proof.

Let Kn(t)=j=-nn(1-|j|n+1)eijt, the Fejér kernel. From this expression we get |Kn(t)|Kn(0)=n+1. It’s straightforward to show that Kn(t)=1n+1(sinn+12tsin12t)2. Since sint2>tπ for 0<t<π, we get |Kn(t)|π2(n+1)t2, and thus we obtain

|Kn(t)|min(n+1,π2(n+1)t2).

Then, for any r1,

Knrr = 12π02π|Kn(t)|r𝑑t
12π0πn+1(n+1)r𝑑t+12ππn+12π(π2(n+1)t2)r𝑑t
= (n+1)r-12+121(n+1)r12r-1((n+1)2r-1-122r-1)
(n+1)r-12+121(n+1)r12r-1(n+1)2r-1
(n+1)r-1.

Hence Knr(n+1)1-1r.

Let Vn(t)=2K2n+1(t)-Kn(t), the de la Vallée Poussin kernel. Then

Vnr2K2n+1r+Knr2(2n+2)1-1r+(n+1)1-1r5(n+1)1-1r.

For |j|n+1 we have Vn^(j)=1, and one thus checks that Vn*f=f. Take 1q+1=1p+1r. By Young’s inequality we have

fq=Vn*fqVnrfp5(n+1)1p-1qfp.

Let Xn={a0+k=1nakcoskt+bksinkt:ak,bk}, the real vector space of real valued trigonometric polynomials of degree n, have norm

fXn=max{|a0|,|a1|,,|an|,|b1|,,|bn|}.

Let Yn,p be the same vector space with the Lp norm. Ash and Ganzburg [1] give upper and lower bounds on the operator norm of the map i:XnYn,p defined by i(f)=f.

Bernstein’s inequality [14, p. 50, Exercise 7.16] states that for 1p, if f is a trigonometric polynomial of degree n, then

fpnfp.

In the other direction, if fC1 then

12π02πf(s)𝑑s+12π0tsf(s)𝑑s+12πt2π(s-2π)f(s)𝑑s=12π02πf(s)𝑑s+12π02πsf(s)𝑑s-t2πf(s)𝑑s=12π02πf(s)𝑑s+12πsf(s)|02π-12π02πf(s)𝑑s-f(s)|t2π=f(t).

Hence

|f(t)| 12π02π|f(s)|𝑑s+12π0ts|f(s)|𝑑s+12πt2π(2π-s)|f(s)|𝑑s
12π02π|f(s)|𝑑s+0t|f(s)|𝑑s+t2π|f(s)|𝑑s
= f1+2πf1,

so

ff1+2πf1.

This is an instance of the Sobolev inequality [26].

It turns out that for a trigonometric polynomial the mass cannot be too concentrated. More precisely, the number of nonzero terms of a trigonometric polynomial restricts how concentrated its mass can be. Let dμ=dt2π. Thus μ([0,2π])=1. A result of Turán [20, p. 89, Lemma 1] states that if λ1,,λN and T(t)=n=1Nbneiλnt, bn, then for any closed arc I[0,2π],

T(2eμ(I))N-1maxtI|T(t)|.

Nazarov [11, p. 452] shows that there is some constant A such that if E is a closed subset of [0,2π] (not necessarily an arc), then

T^1(Aμ(E))NmaxtE|f(T)|.

Nazarov [23] proves that there exists some constant C such that if 0q2 and μ(E)13, then

TqeC(N-1)(1-μ(E)2π)(12πE|T(t)|q𝑑t)1/q.

These results of Turan and Nazarov are examples of the uncertainty principle [9], which is the general principle that a constrain on the support of the Fourier transform of a function constrains the support of the function itself.

In [10], Hardy and Littlewood present inequalities for norms of 2π-periodic functions in terms of certain series formed from their Fourier coefficients. Let ck, k, be such that ck0 as k±, and define c0*,c1*,c-1*,c2*,c-2*, to be the absolute values of the ck ordered in decreasing magnitude. For real r>1, define

Sr*(c)=(k=-ck*r(|k|+1)r-2)1/r.

For instance, if ck=1 for -NkN and ck=0 for |k|>N, then Sr*(c)=(1+2k=2N+1kr-2)1/r. Hardy and Littlewood state the result [10, p. 164, Theorem 2] that if 1<p2 then there is some constant A(p) such that for any sequence c, with ck0 as k±, if f(t)=k=-ckeikt and fp< then

Sp*(c)A(p)fp.

A proof of this is given in Zygmund [35, vol. II, p. 128, chap. XII, Theorem 6.3]. Asking if this inequality holds for p=1 suggests the following question that Hardy and Littlewood pose at the end of their paper [10, p. 168]: Is there a constant A such that for all distinct positive integers mk,k=1,,N, we have

k=1Ncosmkt1>AlogN?

McGehee, Pigno and Smith [18] prove that there is some K such that for all N, if n1,,nN are distinct integers and c1,,cN satisfy |ck|1, then

k=1Nckeinkt1>KlogN.

Thus

k=1Ncosmkt1=12k=1Neimkt+e-imkt112Klog(2N).

For k2, define TN(t)=n=1Neinkt. Since TN=N, for each p1 we have TNpN. Hua’s lemma [22, p. 116, Theorem 4.6] states that if ϵ>0, then

TN2k=O(N1-k2k+ϵ).

Hua’s lemma is used in additive number theory. The number of sets of integer solutions of the equation

f(x1,,xn)=N,arxrbr

is equal to (cf. [12, p. 151])

a1x1b1anxnbn01e2πi(f(x1,,xn)-N)t𝑑t.

Borwein and Lockhart [4]: what is the expected Lp norm of a trigonometric polynomial of order n? Kahane [13, Chapter 6] also presents material on random trigonometric polynomials.

Nursultanov and Tikhonov [25]: the sup on a subset of 𝕋 of a trigonometric polynomial f of degree n being lower bounded in terms of f, n, and the measure of the subset.

3 ℓᵖ norms

For a 2π-periodic function f, we define f^: by

f^(k)=12π02πe-iktf(t)𝑑t.

For 1p<, we define the p norm of f^ by

f^p=(k=-|f^(k)|p)1/p,

and we define the norm of f^ by

f^=maxk|f^(k)|.

Parseval’s identity [31, p. 80, Theorem 1.3] states that f2=f^2.

If 1p<, then

f^(+f^p+)1/p=f^p.

If 1p<q<, then, since for each k, |f^(k)|f^q1,

1=(k=-(|f^(k)|f^q)q)1/q(k=-(|f^(k)|f^q)p)1/q=f^pp/qf^qp/q.

Hence for 1p<p,

f^qf^p.

For 1p<, if f is a trigonometric polynomial of degree n then

f^p=(k=-nn|f^(k)|p)1/p(k=-nnf^p)1/p=(2n+1)1/pf^.

For 1p<q<, we have [30, p. 123, Problem 8.3] (this is Jensen’s inequality for sums)

(k=-nn12n+1|f^(k)|p)1/p(k=-nn12n+1|f^(k)|q)1/q,

i.e.

(2n+1)-1/pf^p(2n+1)-1/qf^q.

Hence for 1<p<q<,

f^p(2n+1)1p-1qf^q.

For any t,

|f(t)|=|k=-f^(k)eikt|k=-|f^(k)eikt|=k=-|f^(k)|=f^1.

Hence

ff^1.

For any k,

|f^(k)|=|12π02πe-iktf(t)𝑑t|12π02π|f(t)|𝑑t=f1.

Hence

f^f1.

The Hausdorff-Young inequality [32, p. 57, Corollary 2.4] states that for 1p2 and 1p+1q=1, if fLp then

f^qfp.

The dual Hausdorff-Young inequality [32, p. 58, Corollary 2.5] states that for 1p2 and 1p+1q=1, if fLq then

fqf^q.

A survey on the Hausdorff-Young inequality is given in [6])

For M+1kM+N, let ak and let S(t)=k=M+1N+1akeikt. Let t1,,tR, and let δ be such that if rs then

tr-tsδ,

where t=mink|t-k| is the distance from t to a nearest integer. The large sieve [19] is an inequality of the form

r=1R|S(2πtr)|2Δ(N,δ)k=M+1M+N|ak|2.

A result of Selberg [19, p. 559, Theorem 3] shows that the large sieve is valid for Δ=N-1+δ-1.

Kristiansen [15]

Boas [2]

For F:/n, its Fourier transform F^:/n (called the discrete Fourier transform) is defined by

F^(k)=1nj=0n-1F(j)e-2πijk/n,0kn-1,

and one can prove [31, p. 223, Theorem 1.2] that

F(j)=k=0n-1F^(k)e2πikj/N,0jn-1.

One can also prove Parseval’s identity for the Fourier transform on /n [31, p. 223, Theorem  1.2]. It states

k=0n-1|F^(k)|2=1nj=0n-1|F(j)|2.

Let P(t)=k=0n-1akeikt. Define F:/n by

F(j)=k=0n-1ake2πikj/n,0jn-1.

(That is, F^(k)=ak.) We then have

k=0n-1|ak|2=1nj=0n-1|F(j)|2=1nj=0n-1|P(2πjn)|2.

Thus

P2=(1nj=0n-1|P(2πjn)|2)1/2.

The Marcinkiewicz-Zygmund inequalities [35, vol. II, p. 28, chap. X, Theorem 7.5] state that there is a constant A such that for 1p, if f is a trigonometric polynomial of degree n then

(12n+1k=02n|f(2πk2n+1)|p)1/pA(2π)1/pfp,

and for each 1<p< there exists some Ap such that if f is a trigonometric polynomial of degree n then

fpAp(12n+1k=02n|f(2πk2n+1)|p)1/p.

Máté and Nevai [17, p. 148, Theorem 6] prove that for p>0, if Sn is a trigonometric polynomial of degree n then

Sn((1+np)e2)1/pSnp.

Máté and Nevai [17] prove a version of Bernstein’s inequality for 0<p<1, and their result can be sharpened to the following [34]: For 0<p<1, if Tn is a trigonometric polynomial of order n then

TnpnTnp.

Let suppf^={k:f^(k)0}. A subset Λ of is called a Sidon set [28, p. 121, §5.7.2] if there exists a constant B such that for every trigonometric polynomial f with suppf^Λ we have

f^1Bf.

Let B(Λ) be the least such B. A sequence of positive integers λk is said to be lacunary if there is a constant ρ such that λk+1>ρλk for all k. If λk is a lacunary sequence, then {λk} is a Sidon set [21, p. 154, Corollary 6.17]. If Λ is a Sidon set, then [28, p. 128, Theorem 5.7.7] (cf. [21, p. 157, Corollary 6.19]) for any 2<p<, for every trigonometric polynomial f with suppf^Λ we have

fpB(Λ)pf2,

and

f22B(Λ)f1.

Let 0<p<. A subset E of is called a Λ(p)-set if for every 0<r<p there is some A(E,p) such that for all trigonometric polynomials f with suppf^E we have

fpA(E,p)f2.

Λ(p) sets were introduced by Rudin, and he discusses them in his autobiography [29, Chapter 28]. A modern survey of Λ(p)-sets is given by Bourgain [5].

Bochkarev [3] proves various lower bounds on the L1 norms of certain trigonometric polynomials. Let ck, k1. If there are constants A and B such that

A(logk)sk|ck|B(logk)sk,k1,

then [3, p. 58, Theorem 19]

k=1nckeik2t1{(logn)s-12,s>12,loglogn,s=12.

If P(t)=k=0nakeikt with ak{-1,1}, then by the Cauchy-Schwarz inequality and Parseval’s identity we have

P1=12π02π1|P(t)|𝑑t12P2=1P^2=n+1.

Newman [24] shows that in fact we can do better than what we get using the Cauchy-Schwarz inequality and Parseval’s identity:

P1<n+0.97.

A Fekete polynomial is a polynomial of the form k=1l-1(kl)zk, l prime, where (kl) is the Legendre symbol. Let Pl(t)=k=1l-1(kl)eikt. Erdélyi [8] proves upper and lower bounds on (1|I|I|Pl(t)|q𝑑t)1/q, q>0, where I is an arc in [0,2π].

References

  • [1] J. M. Ash and M. Ganzburg (1999) An extremal problem for trigonometric polynomials. Proc. Amer. Math. Soc. 127 (1), pp. 211–216. External Links: ISSN 0002-9939, Document, Link, MathReview (V. A. Baskakov) Cited by: §2.
  • [2] R. P. Boas (1947) Inequalities for the coefficients of trigonometric polynomials. II. Nederl. Akad. Wetensch., Proc. 50, pp. 759–762 = Indagationes Math. 9, 369–372 (1947). External Links: MathReview (A. C. Offord) Cited by: §3.
  • [3] S. V. Bochkarev (2006) Multiplicative inequalities for the L1-norm: applications to analysis and number theory. Proc. Steklov Inst. Math. 255, pp. 49–64. External Links: ISSN 0371-9685, MathReview (Hans-Jürgen Schmeisser) Cited by: §3.
  • [4] P. Borwein and R. Lockhart (2001) The expected Lp norm of random polynomials. Proc. Amer. Math. Soc. 129 (5), pp. 1463–1472. External Links: ISSN 0002-9939, Document, Link Cited by: §2.
  • [5] J. Bourgain (2001) Λp-sets in analysis: results, problems and related aspects. In Handbook of the Geometry of Banach Spaces, volume I, W. B. Johnson and J. Lindenstrauss (Eds.), pp. 195–232. External Links: Document, Link, MathReview (Kathryn Hare) Cited by: §3.
  • [6] P. L. Butzer (1994) The Hausdorff-Young theorems of Fourier analysis and their impact. J. Fourier Anal. Appl. 1 (2), pp. 113–130. External Links: ISSN 1069-5869, Document, Link, MathReview (H. Triebel) Cited by: §3.
  • [7] R. A. DeVore and G. G. Lorentz (1993) Constructive approximation. Die Grundlehren der mathematischen Wissenschaften, Vol. 303, Springer. Cited by: §2.
  • [8] T. Erdélyi (2012) Upper bounds for the Lq norm of Fekete polynomials on subarcs. Acta Arith. 153 (1), pp. 81–91. External Links: ISSN 0065-1036, Document, Link, MathReview (D. S. Lubinsky) Cited by: §3.
  • [9] G. B. Folland and A. Sitaram (1997) The uncertainty principle: a mathematical survey. J. Fourier Anal. Appl. 3 (3), pp. 207–238. External Links: ISSN 1069-5869, Document, Link, MathReview (Michael Cowling) Cited by: §2.
  • [10] G. H. Hardy and J. E. Littlewood (1948) A new proof of a theorem on rearrangements. J. London Math. Soc. 23 (3), pp. 163–168. External Links: ISSN 0024-6107, MathReview (R. Salem) Cited by: §2.
  • [11] V. Havin and B. Jöricke (1994) The uncertainty principle in harmonic analysis. Ergebnisse der Mathematik und ihrer Grenzgebiete, 3. Folge, Vol. 28, Springer. Cited by: §2.
  • [12] L. K. Hua (1982) Introduction to number theory. Springer. Note: Translated from the Chinese by Peter Shiu Cited by: §2.
  • [13] J. Kahane (1985) Some random series of functions. second edition, Cambridge University Press. Cited by: §2.
  • [14] Y. Katznelson (2004) An introduction to harmonic analysis. third edition, Cambridge Mathematical Library, Cambridge University Press. Cited by: §1, §2, §2.
  • [15] G. K. Kristiansen (1974) Proof of an inequality for trigonometric polynomials. Proc. Amer. Math. Soc. 44, pp. 49–57. External Links: ISSN 0002-9939, MathReview (M. Marden) Cited by: §3.
  • [16] E. H. Lieb and M. Loss (2001) Analysis. second edition, Graduate Studies in Mathematics, Vol. 14, American Mathematical Society, Providence, RI. Cited by: §1.
  • [17] A. Máté and P. G. Nevai (1980) Bernstein’s inequality in Lp for 0<p<1 and (C, 1) bounds for orthogonal polynomials. Ann. of Math. (2) 111 (1), pp. 145–154. External Links: ISSN 0003-486X, Document, Link Cited by: §3.
  • [18] O. C. McGehee, L. Pigno, and B. Smith (1981) Hardy’s inequality and the L1 norm of exponential sums. Ann. of Math. (2) 113 (3), pp. 613–618. External Links: ISSN 0003-486X, Document, Link, MathReview (John Fournier) Cited by: §2.
  • [19] H. L. Montgomery (1978) The analytic principle of the large sieve. Bull. Amer. Math. Soc 84 (4), pp. 547–567. Cited by: §3.
  • [20] H. L. Montgomery (1994) Ten lectures on the interface between analytic number theory and harmonic analysis. CBMS Regional Conference Series in Mathematics, American Mathematical Society, Providence, RI. Cited by: §2.
  • [21] C. Muscalu and W. Schlag (2013) Classical and multilinear harmonic analysis, volume 1. Cambridge Studies in Advanced Mathematics, Cambridge University Press. Cited by: §3.
  • [22] M. B. Nathanson (1996) Additive number theory: the classical bases. Graduate Texts in Mathematics, Vol. 164, Springer. Cited by: §2.
  • [23] F. L. Nazarov (2000) Complete version of Turan’s lemma for trigonometric polynomials on the unit circumference. In Complex analysis, operators, and related topics: the S. A. Vinogradov memorial volume, V. P. Havin and N. K. Nikolski (Eds.), Operator Theory: Advances and Applications, Vol. 113, pp. 239–246. External Links: MathReview (V. A. Baskakov) Cited by: §2.
  • [24] D. J. Newman (1960) Norms of polynomials. Amer. Math. Monthly 67, pp. 778–779. External Links: ISSN 0002-9890, MathReview (F. Herzog) Cited by: §3.
  • [25] E. Nursultanov and S. Tikhonov (2013) A Sharp Remez Inequality for Trigonometric Polynomials. Constr. Approx. 38 (1), pp. 101–132. External Links: ISSN 0176-4276, Link Cited by: §2.
  • [26] T. Oh and Á. Bényi The Sobolev inequality on the torus revisited. Publ. Math. Debrecen. Cited by: §2.
  • [27] G. Pólya and G. Szegő (1976) Problems and theorems in analysis, volume II. Die Grundlehren der mathematischen Wissenschaften, Vol. 216, Springer. Note: Translated from the German by C. E. Billigheimer Cited by: §1.
  • [28] W. Rudin (1962) Fourier analysis on groups. Interscience Tracts in Pure and Applied Mathematics, Interscience Publishers. Cited by: §3.
  • [29] W. Rudin (1997) The way I remember it. History of Mathematics, Vol. 12, American Mathematical Society, Providence, RI. External Links: ISBN 0-8218-0633-5, MathReview (R. L. Cooke) Cited by: §3.
  • [30] J. M. Steele (2004) The Cauchy-Schwarz master class: an introduction to the art of mathematical inequalities. MAA Problem Books Series, Cambridge University Press. Cited by: §1, §3.
  • [31] E. M. Stein and R. Shakarchi (2003) Fourier analysis: an introduction. Princeton Lectures in Analysis, Vol. I, Princeton University Press. Cited by: §1, §3, §3.
  • [32] E. M. Stein and R. Shakarchi (2011) Functional analysis: introduction to further topics in analysis. Princeton Lectures in Analysis, Vol. IV, Princeton University Press. Cited by: §3.
  • [33] A. F. Timan (1963) Theory of approximation of functions of a real variable. International Series of Monographs on Pure and Applied Mathematics, Vol. 34, Pergamon Press, Oxford. Note: Translated from the Russian by J. Berry Cited by: §2.
  • [34] M. von Golitschek and G. G. Lorentz (1989) Bernstein inequalities in Lp, 0p+. Rocky Mountain J. Math. 19 (1), pp. 145–156. External Links: ISSN 0035-7596, Document, Link, MathReview (Jöran Bergh) Cited by: §3.
  • [35] A. Zygmund (2002) Trigonometric series, volumes I & II combined. third edition, Cambridge Mathematical Library, Cambridge University Press. External Links: ISBN 0-521-89053-5 Cited by: §2, §3.