inequalitySep 2, 2026

A simple proof of a sharp Hardy-type inequality for positive sequences

I
idriss-bado
Member · Research discussion
Votes
4
Answers
3
Views
9
4

Soit (an)n≥1(a_n)_{n\geq 1} une suite de réels positifs telle que

∑n=1∞an2<∞et∑n=1∞n2an2<∞. \sum_{n=1}^{\infty} a_n^2<\infty \qquad\text{et}\qquad \sum_{n=1}^{\infty} n^2a_n^2<\infty.

Montrer que

(∑n=1∞an)4≤π2(∑n=1∞an2)(∑n=1∞n2an2). \left(\sum_{n=1}^{\infty} a_n\right)^4 \leq \pi^2 \left(\sum_{n=1}^{\infty} a_n^2\right) \left(\sum_{n=1}^{\infty} n^2a_n^2\right).

Replies

3 answers
3
I
idriss-bado
Research member
Sep 2, 2026

Fix x>0x>0. Write

an=ann2+x2⋅1n2+x2. a_n = a_n \sqrt{n^2+x^2} \cdot \frac{1}{\sqrt{n^2+x^2}}.

By Cauchy--Schwarz,

(∑n=1∞an)2≤(∑n=1∞an2(n2+x2))(∑n=1∞1n2+x2). \left( \sum_{n=1}^\infty a_n \right)^2 \le \left( \sum_{n=1}^\infty a_n^2 (n^2+x^2) \right) \left( \sum_{n=1}^\infty \frac{1}{n^2+x^2} \right).

Now, since t↦(t2+x2)−1t \mapsto (t^2+x^2)^{-1} is decreasing, we have

∑n=1∞1n2+x2<∫0∞dtt2+x2=π2x. \sum_{n=1}^\infty \frac{1}{n^2+x^2} < \int_0^\infty \frac{dt}{t^2+x^2} = \frac{\pi}{2x}.

Therefore,

(∑n=1∞an)2<π2x(x2∑n=1∞an2+∑n=1∞n2an2). \left( \sum_{n=1}^\infty a_n \right)^2 < \frac{\pi}{2x} \left( x^2 \sum_{n=1}^\infty a_n^2 + \sum_{n=1}^\infty n^2 a_n^2 \right).

Set

A=∑n=1∞an2,B=∑n=1∞n2an2. A = \sum_{n=1}^\infty a_n^2, \qquad B = \sum_{n=1}^\infty n^2 a_n^2.

Then

(∑n=1∞an)2<π2(xA+Bx). \left( \sum_{n=1}^\infty a_n \right)^2 < \frac{\pi}{2} \left( x A + \frac{B}{x} \right).

We now choose x>0x>0 to minimize the right-hand side. The expression xA+B/xx A + B/x attains its minimum at x=B/Ax = \sqrt{B/A}, and the minimum value is

2AB. 2\sqrt{AB}.

Hence

(∑n=1∞an)2<πAB. \left( \sum_{n=1}^\infty a_n \right)^2 < \pi \sqrt{AB}.

Squaring gives

(∑n=1∞an)4<π2AB, \left( \sum_{n=1}^\infty a_n \right)^4 < \pi^2 AB,

which is exactly the desired inequality.

To see that the constant π2\pi^2 is optimal, take

an(ε)=εn2+ε2,ε>0. a_n^{(\varepsilon)} = \frac{\sqrt{\varepsilon}}{n^2 + \varepsilon^2}, \qquad \varepsilon > 0.

A straightforward asymptotic computation gives, as ε→0+\varepsilon \to 0^+,

∑n=1∞an(ε)∼π2,∑n=1∞(an(ε))2∼π4ε,∑n=1∞n2(an(ε))2∼π4ε. \sum_{n=1}^\infty a_n^{(\varepsilon)} \sim \frac{\pi}{2}, \qquad \sum_{n=1}^\infty (a_n^{(\varepsilon)})^2 \sim \frac{\pi}{4\varepsilon}, \qquad \sum_{n=1}^\infty n^2 (a_n^{(\varepsilon)})^2 \sim \frac{\pi}{4\varepsilon}.

Thus the ratio

(∑n=1∞an(ε))4(∑n=1∞(an(ε))2)(∑n=1∞n2(an(ε))2)⟶π2. \frac{\left( \sum_{n=1}^\infty a_n^{(\varepsilon)} \right)^4} {\left( \sum_{n=1}^\infty (a_n^{(\varepsilon)})^2 \right) \left( \sum_{n=1}^\infty n^2 (a_n^{(\varepsilon)})^2 \right)} \longrightarrow \pi^2.

So any constant smaller than π2\pi^2 would fail for sufficiently small ε\varepsilon. The constant is therefore best possible.

This is essentially Carleson's inequality for sequences: the parameter xx acts as a cut-off, and the integral comparison gives exactly the sharp constant after optimization.

3
I
idriss-bado
Research member
Sep 2, 2026

Proof.

Posons

>A:=∑n≥1an2,>>B:=∑n≥1n2an2.> > A:=\sum_{n\ge1}a_n^2, > \qquad > B:=\sum_{n\ge1}n^2a_n^2. >

Fixons x>0x>0. Pour tout n≥1n\ge1, écrivons

>an>=>an2(n2+x2)> >1(n2+x2)1/4> >1(n2+x2)1/4.> > a_n > = > \sqrt{a_n^2(n^2+x^2)} > \, > \frac{1}{(n^2+x^2)^{1/4}} > \, > \frac{1}{(n^2+x^2)^{1/4}}. >

Appliquons l'inégalité de Hölder généralisée avec les exposants

>(2,4,4),>>12+14+14=1.> > (2,4,4), > \qquad > \frac12+\frac14+\frac14=1. >

Il vient

>∑n≥1an>≤>(>∑n≥1an2(n2+x2)>)1/2>(>∑n≥11n2+x2>)1/4>(>∑n≥11n2+x2>)1/4>=>(B+x2A)1/2>(>∑n≥11n2+x2>)1/2.>\begin{align*} > \sum_{n\ge1}a_n > &\le > \left( > \sum_{n\ge1}a_n^2(n^2+x^2) > \right)^{1/2} > \left( > \sum_{n\ge1}\frac1{n^2+x^2} > \right)^{1/4} > \left( > \sum_{n\ge1}\frac1{n^2+x^2} > \right)^{1/4}\\ > &= > \left(B+x^2A\right)^{1/2} > \left( > \sum_{n\ge1}\frac1{n^2+x^2} > \right)^{1/2}. > \end{align*}

Par conséquent,

>(∑n≥1an)2>≤>(B+x2A)>∑n≥11n2+x2.> > \left(\sum_{n\ge1}a_n\right)^2 > \le > (B+x^2A) > \sum_{n\ge1}\frac1{n^2+x^2}. >

Comme la fonction

>t⟼1t2+x2> > t\longmapsto \frac1{t^2+x^2} >

est positive et décroissante sur [0,+∞)[0,+\infty),

>∑n≥11n2+x2>≤>∫0∞dtt2+x2>=>π2x.> > \sum_{n\ge1}\frac1{n^2+x^2} > \le > \int_0^\infty\frac{dt}{t^2+x^2} > = > \frac{\pi}{2x}. >

Ainsi

>(∑n≥1an)2>≤>π2>(>Bx+Ax>).> > \left(\sum_{n\ge1}a_n\right)^2 > \le > \frac{\pi}{2} > \left( > \frac{B}{x}+Ax > \right). >

Le membre de droite est minimal lorsque

>x=BA.> > x=\sqrt{\frac BA}. >

Pour ce choix,

>Bx+Ax>=>2AB.> > \frac{B}{x}+Ax > = > 2\sqrt{AB}. >

D'où

>(∑n≥1an)2>≤>πAB.> > \left(\sum_{n\ge1}a_n\right)^2 > \le > \pi\sqrt{AB}. >

En élevant au carré,

>>(∑n≥1an)4>≤>π2>(∑n≥1an2)>(∑n≥1n2an2)>.> > \boxed{ > \left(\sum_{n\ge1}a_n\right)^4 > \le > \pi^2 > \left(\sum_{n\ge1}a_n^2\right) > \left(\sum_{n\ge1}n^2a_n^2\right) > }. >

∎

3
I
idriss-bado
Research member
Sep 2, 2026

A convenient way to recover the sharp constant is to use the one-sided Fourier spectrum. Let

f(t)=∑n≥1aneint. f(t)=\sum_{n\ge1}a_ne^{int}.

For any x>0x>0,

∣f(0)∣2=∣∑n≥1n2+x2 an 1n2+x2∣2 |f(0)|^2 = \left|\sum_{n\ge1} \sqrt{n^2+x^2}\,a_n\, \frac1{\sqrt{n^2+x^2}} \right|^2

and Cauchy--Schwarz gives

∣f(0)∣2≤(∑n≥1(n2+x2)an2)(∑n≥11n2+x2). |f(0)|^2 \le \left(\sum_{n\ge1}(n^2+x^2)a_n^2\right) \left(\sum_{n\ge1}\frac1{n^2+x^2}\right).

Since

∑n≥11n2+x2≤∫0∞dtt2+x2=π2x, \sum_{n\ge1}\frac1{n^2+x^2} \le \int_0^\infty\frac{dt}{t^2+x^2} = \frac{\pi}{2x},

Parseval yields

∣f(0)∣2≤14x(∥f′∥22+x2∥f∥22). |f(0)|^2 \le \frac1{4x} \left(\|f'\|_2^2+x^2\|f\|_2^2\right).

Choosing

x=∥f′∥2∥f∥2 x=\frac{\|f'\|_2}{\|f\|_2}

gives the one-sided Agmon inequality

  ∥f∥∞2≤12∥f∥2∥f′∥2.   \boxed{\; \|f\|_\infty^2 \le \frac12\|f\|_2\|f'\|_2 . \;}

Here the factor 1/21/2 comes from the fact that only positive Fourier frequencies occur.

Now

f(0)=∑n≥1an,∥f∥22=2π∑n≥1an2,∥f′∥22=2π∑n≥1n2an2. f(0)=\sum_{n\ge1}a_n, \qquad \|f\|_2^2=2\pi\sum_{n\ge1}a_n^2, \qquad \|f'\|_2^2=2\pi\sum_{n\ge1}n^2a_n^2.

Therefore

(∑n≥1an)2≤π(∑n≥1an2)1/2(∑n≥1n2an2)1/2, \left(\sum_{n\ge1}a_n\right)^2 \le \pi \left(\sum_{n\ge1}a_n^2\right)^{1/2} \left(\sum_{n\ge1}n^2a_n^2\right)^{1/2},

and hence

(∑n≥1an)4≤π2(∑n≥1an2)(∑n≥1n2an2). \boxed{ \left(\sum_{n\ge1}a_n\right)^4 \le \pi^2 \left(\sum_{n\ge1}a_n^2\right) \left(\sum_{n\ge1}n^2a_n^2\right). }

Your answer

You need a verified account to contribute. Create an account or log in to join the discussion.