In finite-dimensional spaces like $\mathbb{R}^n$, the Bolzano-Weierstrass theorem guarantees that every bounded sequence has a convergent subsequence. This sequential compactness is the engine behind most existence arguments in analysis: to find a minimiser of a functional, one extracts a limit from a minimising sequence and checks it solves the problem. In infinite-dimensional [Banach spaces](/page/Banach%20Space) — the natural setting for PDEs — this completely breaks down. The closed unit ball of an infinite-dimensional Banach space is never compact in the norm topology, and a bounded sequence can oscillate indefinitely without accumulating anywhere.

The remedy is to work with a coarser topology, called the **weak topology**, in which far more [sequences](/page/Sequence) converge. Rather than demanding that $x_n$ approach $x$ in norm, we only ask that every continuous linear functional evaluates to the right limit. This is precisely weak enough to restore compactness (in reflexive spaces) while remaining strong enough to pass to [limits](/page/Limit) in the linear equations arising in PDE theory.

## Motivation

[motivation] ### The Compactness Crisis in Infinite Dimensions The Bolzano-Weierstrass theorem states that every bounded sequence in $\mathbb{R}^n$ has a convergent subsequence. The proof relies on the Heine-Borel property: closed bounded subsets of $\mathbb{R}^n$ are compact. In infinite-dimensional Banach spaces, Heine-Borel fails catastrophically. The following result, due to Riesz, quantifies the failure: if $Y$ is a proper closed subspace of a Banach space $X$, then for every $\varepsilon > 0$ there exists $x \in X$ with $\|x\|_X = 1$ and $\mathrm{dist}(x, Y) > 1 - \varepsilon$. Iterating this with a nested sequence of subspaces produces a sequence in the closed unit ball $\overline{B}(0, 1) \subseteq X$ with $\|x_m - x_n\|_X > 1/2$ for all $m \neq n$. Such a sequence has no convergent subsequence. The unit ball is therefore not (sequentially) compact in the norm topology of any infinite-dimensional Banach space. This is not a pathology — it is the generic situation in PDE theory. The natural spaces $L^p(U)$, $W^{k,p}(U)$, and $C^k(\overline{U})$ are all infinite-dimensional, and every bounded sequence of approximate solutions lives in a non-compact ball. ### What Goes Wrong With Minimising Sequences The loss of compactness directly obstructs the most natural existence strategy: the **direct method**. To solve a variational problem $\inf_{u \in A} I(u)$ for some functional $I$ and admissible set $A$, one constructs a minimising sequence $\{u_n\}$ with $I(u_n) \to \inf I$. The energy bound $I(u_n) \leq C$ typically implies $\|u_n\|_X \leq C'$ for some Banach norm, but without compactness of the ball, there is no reason for $\{u_n\}$ to have a norm-convergent subsequence. Even if the functional $I$ is well-behaved, the argument stalls because we cannot produce a candidate limit. ### The Fix: Test With Functionals The insight is to weaken what "convergence" means. Instead of requiring $\|x_n - x\| \to 0$ — which asks the sequence to be close in every direction simultaneously — we only ask that $f(x_n) \to f(x)$ for each continuous linear functional $f \in X^*$ separately. This is a strictly weaker condition: each functional tests the sequence along a single "direction" in the dual, and we demand convergence one direction at a time rather than uniformly. The resulting notion — **weak convergence** — admits far more convergent sequences. In reflexive Banach spaces (which include all $L^p$ and $W^{k,p}$ spaces for $1 < p < \infty$), every bounded sequence has a weakly convergent subsequence, restoring the compactness that the direct method requires. The price is that weak limits are harder to work with: the norm can drop, nonlinear expressions need not pass to the limit, and strong convergence must be recovered separately when needed. Understanding this trade-off — what passes to the weak limit and what does not — is the core of the theory. [/motivation]

## Weak Convergence

[definition:Weak Topology] Let $X$ be a real Banach space with continuous dual $X^*$, consisting of all bounded linear functionals \begin{align*} f: X &\to \mathbb{R} \\ x &\mapsto f(x). \end{align*} The **weak topology** on $X$ is the coarsest topology under which every evaluation map $x \mapsto f(x)$ is continuous, for all $f \in X^*$. [/definition]

The weak topology is strictly coarser than the norm topology whenever $X$ is infinite-dimensional: every weakly [open set](/page/Open%20Set) is norm-open (since each $f \in X^*$ is norm-continuous), but not conversely. The open unit ball $B(0,1)$ is norm-open but not weakly open — every weakly open neighbourhood of the origin contains a translate of a closed subspace of finite codimension, and therefore contains points of arbitrarily large norm.

The weak topology is an initial topology — the coarsest making a specified family of maps continuous — so its convergence behaviour is completely determined by the family. The following theorem makes this precise and provides the sequential criterion that is used in practice. We write $x_n \rightharpoonup x$ to denote weak convergence.

[quotetheorem:255]

The forward direction is immediate from [continuity](/page/Continuity): if $x_n \to x$ in the weak topology, then each $f \in X^*$ (being weakly continuous by construction) maps the convergent sequence to a convergent sequence. The reverse direction uses the fact that every weak neighbourhood of $x$ contains a finite intersection of sub-basic [sets](/page/Set) $\{y : |f_i(y) - f_i(x)| < \varepsilon\}$ for finitely many $f_1, \ldots, f_m \in X^*$, and pointwise convergence in each $f_i$ ensures $x_n$ eventually enters the intersection. The theorem reduces checking convergence in the (abstractly defined) weak topology to checking countably many real-number limits — one for each $f \in X^*$.

Strong (norm) convergence always implies weak convergence: if $\|x_n - x\|_X \to 0$ then $|f(x_n) - f(x)| \leq \|f\|_{X^*}\|x_n - x\|_X \to 0$ for every $f \in X^*$, by the definition of the operator norm. The converse is false in general — the oscillation example below exhibits a sequence that converges weakly to zero while maintaining unit norm.

### Weak Convergence in $L^p$ Spaces

In the abstract setting, weak convergence tests against an arbitrary $f \in X^*$. In $L^p$ spaces for $1 < p < \infty$, the [Riesz Representation Theorem](/theorems/221) identifies the dual $L^p(U)^*$ isometrically with $L^q(U)$ (where $1/p + 1/q = 1$), via the pairing $f_v(u) = \int_U u \, v \, d\mathcal{L}^n$ for $v \in L^q(U)$. Under this identification, weak convergence $u_n \rightharpoonup u$ in $L^p(U)$ is equivalent to the integral condition

\begin{align*} \lim_{n \to \infty} \int_U u_n(x)\, v(x) \, d\mathcal{L}^n(x) = \int_U u(x)\, v(x) \, d\mathcal{L}^n(x) \quad \text{for every } v \in L^q(U). \end{align*}

This concrete characterisation is what is used in practice: to check weak convergence in $L^p$, one tests the sequence against every function in the conjugate space $L^q$, and "convergence" means that every such integral converges.

The case $p = 1$ is subtler. The dual of $L^1(U)$ is $L^\infty(U)$, so weak convergence in $L^1$ requires $\int u_n v \, d\mathcal{L}^n \to \int u v \, d\mathcal{L}^n$ for all $v \in L^\infty$. This is a strictly stronger condition than convergence of [integrals](/page/Integral) against continuous [functions](/page/Function) alone, and the failure of reflexivity for $L^1$ means that bounded sequences in $L^1$ need not have weakly convergent subsequences (see Problem 1).

### Weak Sequential Compactness

The central compactness theorem for weak convergence restores, in reflexive spaces, the bounded-implies-subsequentially-convergent property that fails in the norm topology.

[quotetheorem:214]

The hypothesis of reflexivity is essential. A Banach space $X$ is reflexive if and only if the canonical isometric embedding $J: X \hookrightarrow X^{**}$ defined by $J(x)(f) := f(x)$ is surjective — that is, every element of the bidual is already an evaluation functional at some point of $X$. The proof of the theorem works by applying the [Banach-Alaoglu theorem](/theorems/212) to the sequence $\{J(x_n)\}$ in $X^{**}$ (which lies in a weak*-compact ball) and then using surjectivity of $J$ to pull the limit back to $X$. The class of reflexive spaces covers the most important cases in PDE theory: $L^p(U)$ and $W^{k,p}(U)$ are reflexive for $1 < p < \infty$. The theorem fails for $L^1(U)$, $L^\infty(U)$, and $C(\overline{U})$, none of which are reflexive — bounded sequences in these spaces can concentrate, oscillate, or escape to the [boundary](/page/Boundary) without converging weakly.