An integral becomes an analytical working environment when it induces a norm and a complete function space. The $L^p$ spaces provide the standard language for approximation error, stability, duality, and weak formulations.

Let $(X,\mathcal M,\mu)$ be a measure space.

Definition and almost-everywhere equivalence

For $1\le p<\infty$,

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

The expression is initially a seminorm on measurable functions because $|f|_p=0$ means only that $f=0$ almost everywhere. The space $L^p(\mu)$ therefore consists of equivalence classes under

\[f\sim g \quad\Longleftrightarrow\quad f=g\ \text{almost everywhere}.\]

For $p=\infty$,

\[\|f\|_\infty =\operatorname*{ess\,sup}_{x\in X}|f(x)| =\inf\{M\ge0:|f|\le M\text{ almost everywhere}\}.\]

The essential supremum ignores null-set anomalies; the pointwise maximum does not.

Hölder’s inequality

Let $1<p<\infty$ and $q$ satisfy

\[\frac1p+\frac1q=1.\]

Then

\[\int_X|fg|\,d\mu \le\|f\|_p\|g\|_q.\]

After normalizing $|f|_p=|g|_q=1$, the inequality follows from Young’s inequality

\[ab\le\frac{a^p}{p}+\frac{b^q}{q}.\]

The endpoint cases pair $L^1$ with $L^\infty$. Hölder shows that multiplication defines a continuous bilinear map

\[L^p\times L^q\longrightarrow L^1.\]

Minkowski’s inequality

For $1\le p\le\infty$,

\[\|f+g\|_p\le\|f\|_p+\|g\|_p.\]

For $1<p<\infty$, write

\[|f+g|^p=|f+g|\cdot|f+g|^{p-1}\]

and apply Hölder separately to the $f$ and $g$ terms. Minkowski supplies the triangle inequality and therefore makes $L^p$ a normed vector space.

Completeness

The Riesz–Fischer theorem states that $L^p(\mu)$ is complete for $1\le p\le\infty$: every Cauchy sequence in the $L^p$ norm converges in that norm to an element of $L^p$.

Completeness is essential for analysis. An iterative algorithm may generate only approximants $f_n$; the Cauchy property guarantees that the limiting object remains inside the declared solution space.

Norm convergence implies convergence in measure when the measure is finite, and every $L^p$-convergent sequence has a subsequence converging almost everywhere. Full pointwise convergence need not follow.

Inclusion depends on the measure space

If $\mu(X)<\infty$ and $1\le p<q\le\infty$, then

\[L^q(\mu)\subset L^p(\mu),\qquad \|f\|_p \le\mu(X)^{\,1/p-1/q}\|f\|_q.\]

On an infinite-measure space there is no general inclusion in either direction. For instance on $(1,\infty)$, power functions $x^{-a}$ can belong to one $L^p$ space but not another depending on the exponent. Any statement comparing $L^p$ spaces must therefore name the underlying measure and whether its total mass is finite.

Duality and separability

For $1<p<\infty$ under standard hypotheses, every bounded linear functional on $L^p$ has the form

\[\Lambda_g(f)=\int_X f\,\overline g\,d\mu, \qquad g\in L^q.\]

Thus $(L^p)^*\cong L^q$. The case $p=1$ also has dual $L^\infty$ on the usual localizable settings, whereas the dual of $L^\infty$ is generally larger than $L^1$.

For Borel measures on separable metric spaces, and more generally for suitably countably generated $\sigma$-algebras, $L^p$ is separable when $1\le p<\infty$. $L^\infty$ is generally not separable. These qualifications matter when selecting countable approximation bases.

Sobolev spaces

For an open set $\Omega\subset\mathbb R^n$,

\[W^{k,p}(\Omega) =\{f\in L^p(\Omega): D^\beta f\in L^p(\Omega) \text{ for all }|\beta|\le k\},\]

where $D^\beta f$ is understood in the weak, distributional sense. A standard norm is

\[\|f\|_{W^{k,p}} =\left( \sum_{|\beta|\le k}\|D^\beta f\|_p^p \right)^{1/p}\]

for $p<\infty$, with the analogous maximum norm for $p=\infty$.

Sobolev spaces measure both field magnitude and derivative regularity. They are the natural domains of weak differential operators and finite-element formulations; they are not merely “smoother $L^p$ spaces.”

Selecting an error norm

Different norms encode different engineering objectives:

Norm Emphasis Typical interpretation
$L^1$ total absolute error robust aggregate discrepancy
$L^2$ energy and orthogonality least squares, projection, spectral methods
$L^\infty$ worst essential error tolerance envelopes
$W^{1,2}$ value and weak gradient elliptic energy, finite elements

A small $L^2$ error can coexist with a large localized pointwise error; a small $L^\infty$ error says nothing directly about derivative quality. The norm is part of the requirement, not an after-the-fact reporting choice.

For geometric approximation, the metric must also respect parameterization and physical measure. Measuring a surface error uniformly in parameter coordinates can overweight compressed regions and underweight stretched ones. Integrating with the surface Jacobian aligns the norm with physical area.