Hölder inequality

1 Statement of Hölder’s inequality

Hölder’s inequality controls the size of an integral (or sum) of a pointwise product of two measurable functions by the product of appropriate Lebesgue norms.

1.1 Continuous (integral) form

Let \((X,\mathcal{M},\mu)\) be a measure space, and let \(p,q\in[1,\infty]\) satisfy \[ \frac1p+\frac1q=1 \quad (\text{with } 1/\infty:=0). \] If \(f\in L^p(X)\) and \(g\in L^q(X)\), then the product \(fg\) is integrable and \[

\int_X	f(x)g(x)	\,d\mu(x)\le \|f\|_{L^p(X)}\,\|g\|_{L^q(X)}.

\]

1.2 Discrete (sum) form

On a counting measure space (or for sequences), the same statement becomes: for sequences \((a_n)\in \ell^p\) and \((b_n)\in \ell^q\) with \(1/p+1/q=1\), \[

\sum_n	a_n b_n	\le \left(\sum_n	a_n	^p\right)^{1/p}\left(\sum_n	b_n	^q\right)^{1/q}.

\] The case \(p=\infty\) or \(q=\infty\) is interpreted using the corresponding sup-norm.

1.3 The conjugate-exponent condition

The exponents \(p\) and \(q\) are called conjugate (or Hölder-conjugate) when they satisfy \(1/p+1/q=1\). This relation ensures the inequality is dimensionally consistent with the scaling behavior of \(L^p\) norms and with the convexity underlying the proof.

1.4 Typical assumptions on measurable functions

A common formulation assumes \(f\) and \(g\) are measurable and belong to the relevant Lebesgue spaces \(L^p\) and \(L^q\). Under these conditions the right-hand side is finite, and the left-hand side becomes well-defined (finite or \(+\infty\) only if one of the norms diverges).

1.5 Equality cases and sharpness

For \(1<p,q<\infty\), equality (up to measure-zero modifications) occurs precisely when \(f\) and \(g\) have aligned magnitudes in the sense that there exists a constant \(c\ge 0\) such that \[

f(x)	^p = c\,	g(x)	^q

\quad \text{for almost every } x \text{ where } f,g \text{ are nonzero}. \] In addition, \(f\) and \(g\) may differ by signs/phases that match the equality in the absolute-value product. The constant \(1\) in the inequality is optimal.

2 Functional-analytic setting

Hölder’s inequality is best understood inside the theory of Lebesgue spaces and their duality properties.

2.1 Lebesgue spaces L^p and norms

For \(1\le p<\infty\), the norm is \[

\|f\|_{L^p}=\left(\int_X	f	^p\,d\mu\right)^{1/p},

\] while for \(p=\infty\), \[

\|f\|_{L^\infty}=\operatorname*{ess\,sup}_{x\in X}	f(x)	.

\]

2.1.1 Integrability requirements for the product

The inequality implies that whenever \(f\in L^p\) and \(g\in L^q\) with conjugate exponents, the product \(fg\) lies in \(L^1\). In particular, the mapping \((f,g)\mapsto fg\) is controlled from \(L^p\times L^q\) into \(L^1\).

2.1.1.1 Borderline behavior when p or q approaches 1 or infinity

When \(p\to 1^+\) and \(q\to\infty\), Hölder’s inequality tends to the estimate \[

\int	fg	\le \|f\|_{L^1}\|g\|_{L^\infty},

\] which reflects the fact that \(L^\infty\) functions act as pointwise multipliers on \(L^1\). Conversely, if \(p=\infty\) then \(q=1\) and the same type of control holds with the roles reversed. These endpoint versions rely on essential supremum and do not require integrability of the higher power of the bounded function.

2.2 Measure space formulations

Hölder’s inequality applies broadly: it holds on \(\sigma\)-finite measure spaces and, with suitable conventions, on more general spaces as well. Its proof typically uses only measurability, the definition of \(L^p\) norms, and convexity tools that do not depend on geometric structure.

2.3 Hölder conjugates (p, q) and notation

One frequently writes \(q=p'\) or \(p=q'\), where the Hölder conjugate \(p'\) of \(p\in(1,\infty)\) is defined by \[ p'=\frac{p}{p-1}. \]

This notation streamlines formulas, for example \(\|fg\|_{L^1}\le \|f\|_{L^p}\|g\|_{L^{p'}}\).

3 Consequences and corollaries

Many standard estimates in analysis can be viewed as particular cases or direct applications of Hölder’s inequality.

3.1 Cauchy–Schwarz as a special case

Choosing \(p=q=2\) yields \[

\int_X	f(x)g(x)	\,d\mu(x)\le \|f\|_{L^2}\|g\|_{L^2},

\] which is exactly the Cauchy–Schwarz inequality.

3.2 Bounding bilinear forms

If \(B(f,g)\) is the bilinear form \[ B(f,g)=\int_X f(x)g(x)\,d\mu(x), \] then Hölder gives continuity on Lebesgue spaces: \[

B(f,g)	\le \|f\|_{L^p}\|g\|_{L^q}.

\] Thus \(B\) defines a bounded bilinear functional when \(p,q\) are conjugate.

3.3 Norm estimates for multiplication operators

For fixed \(g\in L^q\), define the multiplication operator \(M_g(f)=fg\). Hölder implies \[

\|M_g(f)\|_{L^1}\le \|f\|_{L^p}\|g\|_{L^q},

\] so \(M_g: L^p\to L^1\) is bounded. In other contexts, one combines Hölder with embeddings to obtain bounds into different target spaces.

3.4 Embeddings and basic estimates

Hölder can also support basic comparisons between norms when combined with measure information (e.g., finite measure spaces). It is commonly used to show that certain \(L^p\) inclusions hold under appropriate hypotheses.

3.4.1 Integrability transfer between L^p spaces

On spaces of finite measure, if \(r>s\), then \(L^r\subseteq L^s\). A standard route uses Hölder by writing \(	f	^s=	f	^r\cdot	f	^{s-r}\) and controlling the latter factor using measure bounds. The exact constants depend on \(\mu(X)\).

4 Generalizations

Hölder’s original two-factor statement extends naturally to more factors, mixed norms, and weighted settings.

4.1 Generalized Hölder inequality for multiple factors

For measurable functions \(f_1,\dots,f_m\) and exponents \(p_1,\dots,p_m\in[1,\infty]\) satisfying \[ \sum_{j=1}^m \frac{1}{p_j}=1, \] one has \[

\int_X \prod_{j=1}^m	f_j	\,d\mu \le \prod_{j=1}^m \|f_j\|_{L^{p_j}}.

\] The proof can be built by iterative application of the two-factor inequality.

4.2 Mixed-norm and product-space versions

On product measure spaces \(X\times Y\), one often uses mixed Lebesgue norms (e.g., \(L^p_xL^r_y\)). Hölder can be applied separately in each variable, producing inequalities that control integrals by products of mixed norms. These forms are widely used in PDEs and harmonic analysis to handle anisotropy.

4.3 Weighted Hölder inequalities

If weights \(w>0\) are introduced, one works with weighted norms such as \(\|f\|_{L^p(w)}=(\int	f	^p w\,d\mu)^{1/p}\). A typical weighted Hölder inequality takes the form

\[

\int	f g	\,d\mu \le \|f\|_{L^p(w)}\|g\|_{L^q(w^{-\frac{q}{p}})},

\] with suitable conventions at endpoints. Weighted variants are central when dealing with non-uniform densities or adapted function spaces.

4.4 Hölder in other settings (e.g., sequences and norms)

The inequality extends to settings where “integral” is replaced by another positive functional, such as a sum over indices, or more generally by abstract integration with respect to a measure. It also appears in algebraic forms controlling products in normed spaces, provided the functionals correspond to integrals/sums.

5 Related inequalities and relationships

Hölder sits in a network of inequalities that reflect convexity, duality, and interpolation.

5.1 Young’s inequality and its link to Hölder

Young’s inequality for products states that for conjugate exponents \(p,q\) and nonnegative \(a,b\), \[ ab \le \frac{a^p}{p}+\frac{b^q}{q}. \]

Integrating this pointwise inequality with \(a=	f	\) and \(b=	g	\) yields Hölder after applying the normalization given by the norms. Conversely, Hölder can be proved using Young-type convexity arguments.

5.2 Minkowski’s inequality connection

Minkowski’s inequality concerns \(\|f+g\|_{L^p}\) and is closely tied to Hölder via duality and the convexity of \(t\mapsto t^p\) for \(p\ge 1\). While it controls sums rather than products, it is frequently paired with Hölder in estimates.

5.3 Interpolation perspective

Hölder is compatible with interpolation theory, where bounds in \(L^p\) scale between endpoints. In many applications, the structural form of Hölder’s inequality supports estimates of operators across a range of \(p\)-values.

5.4 Duality of L^p spaces

A central relation is that for \(1<p<\infty\), the dual of \(L^p\) is \(L^{p'}\), meaning every bounded linear functional on \(L^p\) can be represented as integration against an \(L^{p'}\) function. Hölder provides the fundamental bound needed to show that such functionals are continuous.

6 Applications

Hölder’s inequality appears throughout analysis because it offers a robust mechanism for estimating nonlinear terms.

6.1 Estimates in PDE (energy-type bounds)

In PDE, nonlinearities often take the form of products, such as \(u\nabla u\), \(u^2\), or forcing terms involving multiplication. Hölder is used to control these terms in integral norms, enabling energy estimates and establishing bounds that prevent blow-up in iterative schemes.

6.2 Harmonic analysis bounds

In harmonic analysis, one frequently estimates integrals involving convolutions or maximal functions, where decompositions lead to products of functions living in different \(L^p\) spaces. Hölder then converts the problem into norm estimates that can be handled by other tools.

6.3 Convergence results using norm control

If a sequence \((f_n)\) converges in \(L^p\) and \((g_n)\) is bounded in \(L^q\), Hölder can be used to show convergence of \(\int f_n g_n\) or convergence in \(L^1\) for products under additional assumptions. It therefore plays a role in compactness arguments and in proving continuity of nonlinear maps.

6.4 Operator theory and boundedness proofs

When analyzing linear or multilinear operators, one often reduces boundedness to estimates of the form \(\int	K(x)	\,	f(x)	\) or \(\int	(Tf)(x)g(x)	\). Hölder provides the canonical route from kernel or multiplier information to operator norms.

7 Proofs

Multiple proofs exist, each highlighting a different structural aspect: convexity, duality, or optimization.

7.1 Proof via convexity (Young’s inequality)

Assume \(1<p,q<\infty\) and \(f,g\) are nonnegative (apply to absolute values). Normalize so that \(\|f\|_{L^p}=\|g\|_{L^q}=1\) by replacing \(f\) with \(f/\|f\|_{L^p}\) and similarly for \(g\). Young’s inequality gives, pointwise,

\[ f(x)g(x)\le \frac{f(x)^p}{p}+\frac{g(x)^q}{q}. \] Integrating and using the normalization yields \[ \int fg \le \frac{1}{p}\int f^p + \frac{1}{q}\int g^q = \frac{1}{p}+\frac{1}{q}=1. \]

Undoing the normalization gives the general bound \(\int	fg	\le \|f\|_p\|g\|_q\).

7.2 Proof using the supremum characterization of L^p dual norms

For \(1<p<\infty\), one can use the characterization \[

\|h\|_{L^p} = \sup\left\{\left	\int_X h(x)\,\phi(x)\,d\mu(x)\right	:\ \|\phi\|_{L^{p'}}\le 1\right\}.

\]

Apply this with \(h=f\) and \(\phi=\operatorname{sgn}(g)	g	\) normalized appropriately. The resulting estimate directly implies Hölder’s inequality and simultaneously reflects the duality structure of \(L^p\) spaces.

7.3 Proof using scaling and optimization

Consider the normalized inequality \[

\int	f g	\le C\,\|f\|_p\|g\|_q

\] and show that the best constant is \(C=1\) by optimizing over scalar dilations \(f\mapsto \alpha f\), \(g\mapsto \beta g\). The constraint \(1/p+1/q=1\) ensures that the objective and constraints scale compatibly, and the optimization reduces to a convex maximization problem whose solution corresponds to the equality condition.

7.4 Discrete-to-continuous transfer arguments

One approach proves Hölder for sequences first (using finite sums and discrete convexity), then extends to integrals by approximating functions with simple functions and using monotone or dominated convergence to pass to the limit. This method emphasizes that the core inequality is measure-agnostic and relies on positivity and approximation.

8 Examples

Concrete examples illustrate how Hölder works, how to verify it, and why the equality condition matters.

8.1 Simple power-function examples

On \([0,1]\) with Lebesgue measure, take \(f(x)=x^{-1/p}(\log(1/x))^{-2/p}\) for \(x\in(0,1/2]\) (and modified on the rest to avoid issues), and define \(g\) so that \(	f	^p\) and \(	g	^q\) are proportional. This setup ensures \(f\in L^p\), \(g\in L^q\), and the product satisfies integrability, yielding an explicit estimate after computing norms.

8.2 Verifying Hölder for characteristic functions

Let \(A\subset X\) have finite measure and set \(f=\mathbf{1}_A\). Then \(\|f\|_{L^p}=\mu(A)^{1/p}\). If \(B\subset X\) is another measurable set and \(g=\mathbf{1}_B\), then \(fg=\mathbf{1}_{A\cap B}\) and

\[ \int \mathbf{1}_{A\cap B}\,d\mu=\mu(A\cap B)\le \mu(A)^{1/p}\mu(B)^{1/q}, \] which is Hölder in this special case. It can be viewed as a bound on overlap measure via sizes of the sets.

8.3 Worked estimates with explicit norms

Consider \(X=\mathbb{R}\) with Lebesgue measure and define \(f\) and \(g\) as Gaussians or compactly supported smooth functions. Their \(L^p\) norms are finite, and Hölder provides an immediate bound for \(\int	f g	\). In applications, such worked examples serve as sanity checks for more abstract arguments.

8.4 Examples illustrating equality conditions

Equality for \(1<p,q<\infty\) requires proportionality of \(	f	^p\) and \(	g	^q\). For instance, on a measure space with atoms, one can choose \(f\) and \(g\) so that on each atom the magnitudes satisfy \(	f	^p=c	g	^q\). Then the inequality becomes an equality, demonstrating sharpness and aligning with the convexity equality case in the proof.