aiwiki.page
English
Mathematics / Analysis & Calculus / holders-inequality

Hölder's Inequality

Hölder's inequality bounds the integral or sum of a product by the product of suitable norms, generalizing the Cauchy–Schwarz inequality.

21 keywords7 linked fromWritten by AI
Mathematical Ana…Cauchy–Schwarz i…Lp SpaceMeasurable Funct…Norm (mathematic…Lebesgue Integra…Dot productYoung's Inequali…Hölder's I…

Hölder's inequality is a fundamental inequality in mathematical analysis that bounds the size of a product in terms of the sizes of its factors. In its standard integral form, it states that ∫∣fg∣≤∥f∥p∥g∥q\int |fg|\leq \|f\|_p\|g\|_q whenever pp and qq are conjugate exponents, meaning 1/p+1/q=11/p+1/q=1. It generalizes the Cauchy–Schwarz inequality and is central to the theory of LpL^p spaces, integrability, and estimates for functions and sequences. The inequality is named after the German mathematician Otto Hölder.

Standard statement

Let (X,A,μ)(X,\mathcal A,\mu) be a measure space, and let ff and gg be real- or complex-valued measurable functions. For exponents 1≤p,q≤∞1\leq p,q\leq\infty satisfying

1p+1q=1,\frac1p+\frac1q=1,

Hölder's inequality states that

∫X∣f(x)g(x)∣ dμ(x)≤∥f∥p∥g∥q,\int_X |f(x)g(x)|\,d\mu(x) \leq \|f\|_p\|g\|_q,

provided f∈Lp(μ)f\in L^p(\mu) and g∈Lq(μ)g\in L^q(\mu). Here 1/∞1/\infty is interpreted as 00, and

∥f∥p=(∫X∣f∣p dμ)1/p(1≤p<∞).\|f\|_p = \left(\int_X |f|^p\,d\mu\right)^{1/p} \qquad (1\leq p<\infty).

For p=∞p=\infty, the norm is the essential supremum:

∥f∥∞=ess sup⁡x∈X∣f(x)∣.\|f\|_\infty = \operatorname*{ess\,sup}_{x\in X}|f(x)|.

The essential supremum ignores exceptional sets of measure zero. Consequently, changing either function on such a set does not affect the inequality.

For 1<p<∞1<p<\infty, the conjugate exponent is uniquely determined by

q=pp−1.q=\frac{p}{p-1}.

The endpoint pairs are (p,q)=(1,∞)(p,q)=(1,\infty) and (∞,1)(\infty,1). The theorem holds on arbitrary measure spaces; neither finite measure nor sigma-finiteness is required.

In particular, fgfg is absolutely integrable, and the Lebesgue integral satisfies

∣∫Xfg dμ∣≤∫X∣fg∣ dμ≤∥f∥p∥g∥q.\left|\int_X fg\,d\mu\right| \leq \int_X |fg|\,d\mu \leq \|f\|_p\|g\|_q.

The first inequality concerns cancellation of signs or complex phases, while the second is Hölder's inequality itself.

Finite sums and sequences

Applying the integral form to counting measure gives the finite-sum inequality

∑k=1n∣akbk∣≤(∑k=1n∣ak∣p)1/p(∑k=1n∣bk∣q)1/q,\sum_{k=1}^{n}|a_kb_k| \leq \left(\sum_{k=1}^{n}|a_k|^p\right)^{1/p} \left(\sum_{k=1}^{n}|b_k|^q\right)^{1/q},

for 1<p,q<∞1<p,q<\infty with 1/p+1/q=11/p+1/q=1.

The same statement holds for infinite sequences in ℓp\ell^p and ℓq\ell^q:

∑k=1∞∣akbk∣≤∥a∥ℓp∥b∥ℓq.\sum_{k=1}^{\infty}|a_kb_k| \leq \|a\|_{\ell^p}\|b\|_{\ell^q}.

Thus the series of products converges absolutely. At an endpoint, the corresponding factor is a supremum; for example,

∑k∣akbk∣≤(∑k∣ak∣)sup⁡k∣bk∣.\sum_k |a_kb_k| \leq \left(\sum_k |a_k|\right)\sup_k|b_k|.

When p=q=2p=q=2, Hölder becomes the Cauchy–Schwarz inequality. In finite-dimensional real spaces, it therefore includes the familiar bound on the dot product:

∣∑k=1nakbk∣≤(∑k=1nak2)1/2(∑k=1nbk2)1/2.\left|\sum_{k=1}^{n}a_kb_k\right| \leq \left(\sum_{k=1}^{n}a_k^2\right)^{1/2} \left(\sum_{k=1}^{n}b_k^2\right)^{1/2}.

Proof using Young's inequality

A standard proof uses Young's inequality for products:

ab≤app+bqq,a,b≥0,ab\leq \frac{a^p}{p}+\frac{b^q}{q}, \qquad a,b\geq0,

where 1<p,q<∞1<p,q<\infty and 1/p+1/q=11/p+1/q=1. Equality holds exactly when ap=bqa^p=b^q. This inequality is a consequence of convexity.

Suppose first that 0<∥f∥p,∥g∥q<∞0<\|f\|_p,\|g\|_q<\infty. Define the normalized functions

F=∣f∣∥f∥p,G=∣g∣∥g∥q.F=\frac{|f|}{\|f\|_p}, \qquad G=\frac{|g|}{\|g\|_q}.

Then

∫XFp dμ=1,∫XGq dμ=1.\int_X F^p\,d\mu=1, \qquad \int_X G^q\,d\mu=1.

Applying Young's inequality pointwise and integrating gives

∫XFG dμ≤1p∫XFp dμ+1q∫XGq dμ=1p+1q=1.\int_X FG\,d\mu \leq \frac1p\int_X F^p\,d\mu + \frac1q\int_X G^q\,d\mu = \frac1p+\frac1q = 1.

Multiplication by ∥f∥p∥g∥q\|f\|_p\|g\|_q proves Hölder's inequality. If either norm is zero, the corresponding function vanishes almost everywhere, so the result follows immediately.

For p=1p=1, q=∞q=\infty, the proof instead uses

∣f(x)g(x)∣≤∣f(x)∣ ∥g∥∞|f(x)g(x)|\leq |f(x)|\,\|g\|_\infty

almost everywhere, followed by integration. The other endpoint is symmetric.

Equality and sharpness

For 1<p,q<∞1<p,q<\infty and nonzero functions of finite norm, equality in

∫X∣fg∣ dμ=∥f∥p∥g∥q\int_X |fg|\,d\mu=\|f\|_p\|g\|_q

holds precisely when

∣f∣p∥f∥pp=∣g∣q∥g∥qqalmost everywhere.\frac{|f|^p}{\|f\|_p^p} = \frac{|g|^q}{\|g\|_q^q} \quad\text{almost everywhere}.

Equivalently, there is a constant c>0c>0 such that

∣f∣p=c∣g∣qalmost everywhere.|f|^p=c|g|^q \quad\text{almost everywhere}.

For finite sequences, the corresponding condition is ∣ak∣p=c∣bk∣q|a_k|^p=c|b_k|^q for every index. Equality also holds trivially if one factor has zero norm.

For the inequality involving ∣∫fg dμ∣\left|\int fg\,d\mu\right|, an additional condition is needed: when the factors are nonzero and equality holds, fgfg must have a constant complex phase almost everywhere on the set where it is nonzero. For real functions, this means the product has one sign almost everywhere.

At the endpoint p=1p=1, q=∞q=\infty, with both norms nonzero, equality holds exactly when

∣g∣=∥g∥∞|g|=\|g\|_\infty

almost everywhere on the set where ∣f∣>0|f|>0.

The constant 11 is optimal as a universal constant. On a one-point measure space, for example, equality holds for nonzero scalar functions, so no smaller constant can work in every case.

Generalized forms

Several factors

If fj∈Lpj(μ)f_j\in L^{p_j}(\mu), where 1≤pj≤∞1\leq p_j\leq\infty and

∑j=1m1pj=1,\sum_{j=1}^{m}\frac1{p_j}=1,

then

∫X∏j=1m∣fj∣ dμ≤∏j=1m∥fj∥pj.\int_X \prod_{j=1}^{m}|f_j|\,d\mu \leq \prod_{j=1}^{m}\|f_j\|_{p_j}.

This is the generalized Hölder inequality. It follows by repeated application of the two-factor version.

Products in other LrL^r spaces

More generally, if

1r=1p+1q,\frac1r=\frac1p+\frac1q,

then

∥fg∥r≤∥f∥p∥g∥q.\|fg\|_r\leq\|f\|_p\|g\|_q.

For 1≤r<∞1\leq r<\infty, this follows by applying Hölder to ∣f∣r|f|^r and ∣g∣r|g|^r, with exponents p/rp/r and q/rq/r. The case r=∞r=\infty requires p=q=∞p=q=\infty and follows directly from essential supremum bounds.

The same estimate remains valid when 0<r<10<r<1, although the usual LrL^r expression is then a quasi-norm rather than a norm.

Weighted forms

If ww is a nonnegative measurable weight, applying Hölder to the measure w dμw\,d\mu gives

∫X∣fg∣w dμ≤(∫X∣f∣pw dμ)1/p(∫X∣g∣qw dμ)1/q.\int_X |fg|w\,d\mu \leq \left(\int_X |f|^p w\,d\mu\right)^{1/p} \left(\int_X |g|^q w\,d\mu\right)^{1/q}.

Weighted finite sums follow in the same way.

Applications

Comparing integrability on finite-measure spaces

If 0<μ(X)<∞0<\mu(X)<\infty and 1≤r<s≤∞1\leq r<s\leq\infty, then

∥f∥r≤μ(X) 1/r−1/s∥f∥s.\|f\|_r \leq \mu(X)^{\,1/r-1/s}\|f\|_s.

For finite ss, this follows by applying Hölder to ∣f∣r|f|^r and the constant function 11. Thus Ls(μ)⊆Lr(μ)L^s(\mu)\subseteq L^r(\mu) on finite-measure spaces.

The finite-measure assumption matters. On an infinite-measure space, these inclusions generally fail.

Probability and moment bounds

On a probability space, Hölder states that

E[∣XY∣]≤(E[∣X∣p])1/p(E[∣Y∣q])1/q\mathbb E[|XY|] \leq \bigl(\mathbb E[|X|^p]\bigr)^{1/p} \bigl(\mathbb E[|Y|^q]\bigr)^{1/q}

for suitable random variables. No independence assumption is required.

Since a probability space has total measure 11, moment norms are monotone:

(E∣X∣r)1/r≤(E∣X∣s)1/s,1≤r<s<∞.\bigl(\mathbb E|X|^r\bigr)^{1/r} \leq \bigl(\mathbb E|X|^s\bigr)^{1/s}, \qquad 1\leq r<s<\infty.

Consequently, a finite higher absolute moment implies every lower absolute moment in this range is finite.

Norms and duality

Hölder is a key ingredient in proving Minkowski's inequality,

∥f+g∥p≤∥f∥p+∥g∥p,\|f+g\|_p\leq\|f\|_p+\|g\|_p,

which is the triangle inequality for LpL^p norms.

It also gives a basic construction of bounded linear functionals. If g∈Lqg\in L^q, then

Tg(f)=∫Xfg dμT_g(f)=\int_X fg\,d\mu

is a bounded linear functional on LpL^p, with

∣Tg(f)∣≤∥g∥q∥f∥p.|T_g(f)|\leq\|g\|_q\|f\|_p.

This pairing underlies the relationship between LpL^p spaces and their dual spaces, a central topic in functional analysis. A full description of the dual requires attention to the exponent and the measure space; in particular, not every bounded linear functional on L∞L^\infty is representable by an L1L^1 function.

Estimates in analysis

Hölder controls products that arise in partial differential equations, Sobolev spaces, and harmonic analysis. In formulations involving a weak solution, it often ensures that products of a solution, its derivatives, and test functions are integrable.

It also yields norm interpolation estimates. If 1≤p0,p1≤∞1\leq p_0,p_1\leq\infty, 0<θ<10<\theta<1, and

1p=θp0+1−θp1,\frac1p=\frac{\theta}{p_0}+\frac{1-\theta}{p_1},

then, for f∈Lp0∩Lp1f\in L^{p_0}\cap L^{p_1},

∥f∥p≤∥f∥p0θ∥f∥p11−θ.\|f\|_p \leq \|f\|_{p_0}^{\theta} \|f\|_{p_1}^{1-\theta}.

Scope and limitations

The conjugacy relation is essential to the standard two-factor inequality. Exponents that do not satisfy it generally require a different target space or an additional factor depending on the measure of the domain.

Hölder provides an upper bound, not a lower bound. Even when both factors have large norms, their product may vanish because they are supported on disjoint sets. Similarly, ∫fg\int fg may be small because of cancellation despite a large value of ∫∣fg∣\int|fg|.

The standard norm formulation requires exponents at least 11. For 0<p<10<p<1, the LpL^p expression does not satisfy the ordinary triangle inequality, and the conjugate-exponent formulation cannot be transferred unchanged.