In finite-dimensional analysis, compactness is available essentially for free: the Heine-Borel theorem guarantees that every closed and bounded subset of $\mathbb{R}^n$ is compact, and the Bolzano-Weierstrass theorem ensures that every bounded sequence has a convergent subsequence. These facts underpin the entire machinery of optimisation, existence theory, and variational methods in finite dimensions. In infinite-dimensional [Banach spaces](/page/Banach%20Space), this machinery breaks down completely. The closed unit ball $\overline{B}_X := \{x \in X : \|x\|_X \le 1\}$ is **never** compact in the norm topology when $X$ is infinite-dimensional. Riesz's Lemma provides the obstruction: given any proper closed subspace $Y \subsetneq X$ and any $\varepsilon > 0$, there exists a unit vector $x \in X$ with $\operatorname{dist}(x, Y) > 1 - \varepsilon$. Iterating this construction produces a sequence of unit vectors $\{e_n\}_{n=1}^\infty$ with $\|e_m - e_n\|_X > 1/2$ for all $m \neq n$ — a bounded sequence with no Cauchy subsequence. Compactness fails, and with it the straightforward extraction of convergent subsequences. This is not an exotic pathology — it is the rule. Every $L^p$ space on a domain of positive measure, every [Sobolev space](/page/Sobolev%20Space), every space of continuous functions on a compact set, is infinite-dimensional. The existence theory for partial differential equations, the [calculus of variations](/page/Calculus%20of%20Variations), and optimisation in function spaces all depend on extracting limits from bounded sequences, and the norm topology cannot deliver. The Banach-Alaoglu theorem resolves this impasse by changing the topology. It asserts that the closed unit ball of the [dual space](/page/Dual%20Space) $X^*$ is compact — not in the norm topology, but in the [weak* topology](/page/Weak*%20Topology). This is the coarsest topology making every evaluation map $\operatorname{ev}_x: f \mapsto f(x)$ continuous, and it has far fewer open sets than the norm topology. The price of weak* convergence — requiring $f_n(x) \to f(x)$ for every $x \in X$, rather than $\|f_n - f\|_{X^*} \to 0$ — is exactly what buys back compactness. [example: Non-Compactness of the Unit Ball in $\ell^2$] Consider $X = \ell^2(\mathbb{N})$ and the standard basis $\{e_n\}_{n=1}^\infty$, where $e_n$ is the sequence with $1$ in position $n$ and $0$ elsewhere. Each $e_n$ lies in the closed unit ball: $\|e_n\|_{\ell^2} = 1$. However, for $m \neq n$: \begin{align*} \|e_m - e_n\|_{\ell^2}^2 = \|e_m\|_{\ell^2}^2 + \|e_n\|_{\ell^2}^2 = 2, \end{align*} so $\|e_m - e_n\|_{\ell^2} = \sqrt{2}$ for all $m \neq n$. No subsequence can be Cauchy, and the unit ball is not sequentially compact. Yet the same sequence $\{e_n\}_{n=1}^\infty$, viewed as elements of $(\ell^2)^* \cong \ell^2$ (via the Riesz representation), converges weak* to $0$: for any $x = (x_1, x_2, \ldots) \in \ell^2$, \begin{align*} e_n(x) = x_n \to 0 \quad \text{as } n \to \infty, \end{align*} since $\sum_{n=1}^\infty |x_n|^2 < \infty$ forces $x_n \to 0$. The sequence that is hopelessly spread out in the norm topology converges in the weak* topology. Banach-Alaoglu guarantees this phenomenon is not special to $\ell^2$ — it occurs in every dual space. [/example] The theorem is the starting point for a chain of results that governs compactness in infinite-dimensional analysis. Combined with separability, it yields the Sequential Banach-Alaoglu theorem: every bounded sequence in the dual of a separable space has a weak* convergent subsequence — the directly usable form for PDE existence proofs. Combined with [reflexivity](/page/Reflexive%20Space), it yields weak compactness of the unit ball of the original space (Kakutani's theorem), which is the foundation of the direct method in the calculus of variations. This page develops the theorem, its sequential form, its proof mechanism, and its consequences. ## Definition The Banach-Alaoglu theorem concerns compactness in the weak* topology on a dual space. The key definitions are the dual space and the weak* topology, both of which are treated in detail on their own pages. We record the essential structure here. [definition: Weak-Star Topology on the Dual Space] Let $X$ be a [normed vector space](/page/Normed%20Vector%20Space) over $\mathbb{R}$ with norm $\|\cdot\|_X$, and let $X^* := \mathcal{L}(X, \mathbb{R})$ denote the [dual space](/page/Dual%20Space) of bounded linear functionals on $X$. The **weak* topology** on $X^*$, denoted $\sigma(X^*, X)$, is the coarsest [topology](/page/Topology) on $X^*$ making every evaluation map \begin{align*} \operatorname{ev}_x: X^* &\to \mathbb{R} \\ f &\mapsto f(x) \end{align*} continuous, for each $x \in X$. A net $(f_\alpha)$ in $X^*$ converges to $f \in X^*$ in the weak* topology if and only if $f_\alpha(x) \to f(x)$ for every $x \in X$. [/definition] The weak* topology is an [initial topology](/page/Product%20Topology): it is generated by the family of maps $\{\operatorname{ev}_x\}_{x \in X}$. A subbasis consists of sets of the form $\{f \in X^* : |f(x) - f_0(x)| < \varepsilon\}$ for fixed $x \in X$, $f_0 \in X^*$, and $\varepsilon > 0$. This is strictly coarser than the weak topology $\sigma(X^*, X^{**})$ on $X^*$ — which uses all of $X^{**}$ as test functionals — unless $X$ is [reflexive](/page/Reflexive%20Space), in which case the canonical embedding $J: X \to X^{**}$ is surjective and the two topologies on $X^*$ coincide. The norm topology, the weak topology, and the weak* topology on $X^*$ form a hierarchy: \begin{align*} \sigma(X^*, X) \subset \sigma(X^*, X^{**}) \subset \tau_{\|\cdot\|_{X^*}}, \end{align*} where each inclusion denotes "coarser than or equal to." Making a topology coarser adds no new open sets and therefore makes it *easier* for a set to be compact (there are fewer open covers to worry about). This is the geometric reason that the weak* unit ball can be compact even when the norm unit ball is not. ## The Banach-Alaoglu Theorem The central difficulty in infinite-dimensional analysis is that the closed unit ball fails to be compact. The Banach-Alaoglu theorem overcomes this by identifying the dual ball as a closed subset of a product of compact intervals — and products of compact spaces are compact by [Tychonoff's theorem](/theorems/953). [quotetheorem:212] More generally, for any $R > 0$, the closed ball $\{f \in X^* : \|f\|_{X^*} \le R\}$ is weak* compact. This follows by rescaling: the map $f \mapsto f/R$ is a weak*-homeomorphism from the ball of radius $R$ to the unit ball. ### The Proof Strategy: Embedding into a Product The proof of Banach-Alaoglu is a masterclass in converting a functional-analytic question into a topological one. The idea is to realise $B_{X^*}$ as a closed subset of a product of compact intervals, then invoke Tychonoff's theorem. For each $x \in X$, the operator norm bound $\|f\|_{X^*} \le 1$ forces $|f(x)| \le \|x\|_X$, so the value $f(x)$ is confined to the compact interval $D_x := [-\|x\|_X, \|x\|_X]$. The evaluation map \begin{align*} \Phi: B_{X^*} &\to \prod_{x \in X} D_x \\ f &\mapsto (f(x))_{x \in X} \end{align*} embeds $B_{X^*}$ into the product $P := \prod_{x \in X} D_x$. Tychonoff's theorem guarantees that $P$ is compact in the [product topology](/page/Product%20Topology). The product topology on $P$ has convergence defined by convergence in each coordinate: a net $(g_\alpha)$ converges to $g$ in $P$ if and only if $g_\alpha(x) \to g(x)$ for every $x \in X$. This is exactly the definition of weak* convergence, so $\Phi$ is a homeomorphism onto its image (with the subspace topology from $P$). The remaining step is to verify that $\Phi(B_{X^*})$ is **closed** in $P$. An element $g \in P$ is a function $g: X \to \mathbb{R}$ with $|g(x)| \le \|x\|_X$ for all $x$. The image $\Phi(B_{X^*})$ consists of those $g$ that are additionally **linear**. Linearity is a closed condition: for any fixed $x, y \in X$ and $\alpha, \beta \in \mathbb{R}$, the set \begin{align*} \{g \in P : g(\alpha x + \beta y) = \alpha g(x) + \beta g(y)\} \end{align*} is closed (it is the preimage of $\{0\}$ under the continuous map $g \mapsto g(\alpha x + \beta y) - \alpha g(x) - \beta g(y)$, which involves only finitely many coordinate projections). Since $\Phi(B_{X^*})$ is the intersection of all such sets over all $x, y \in X$ and $\alpha, \beta \in \mathbb{R}$, it is an intersection of closed sets, hence closed. A closed subset of a compact space is compact, so $\Phi(B_{X^*})$ is compact, and therefore $B_{X^*}$ is weak* compact. [remark: Role of Tychonoff's Theorem] The proof depends essentially on [Tychonoff's theorem](/theorems/953) for an arbitrary (typically uncountable) product. Since the index set is $X$ itself — an infinite-dimensional vector space — the product $\prod_{x \in X} D_x$ has uncountably many factors. Tychonoff's theorem for arbitrary products is equivalent to the Axiom of Choice, so the Banach-Alaoglu theorem in its full generality is not provable in ZF alone. For separable $X$, the theorem can be proved without the full Axiom of Choice, using only the countable Axiom of Choice (via the sequential version below). This logical distinction rarely matters in practice, but it explains why some authors state the theorem only for separable spaces. [/remark] ### What the Theorem Does Not Say Several common misreadings of Banach-Alaoglu deserve explicit correction. **It does not say that $B_{X^*}$ is norm-compact.** In infinite-dimensional spaces, the norm-closed unit ball $B_{X^*}$ is never norm-compact (by Riesz's Lemma applied to $X^*$, which is itself infinite-dimensional when $X$ is). The compactness is only in the weak* topology, which is strictly coarser than the norm topology. **It does not give sequential compactness in general.** Compactness in the weak* topology means that every *net* has a convergent subnet — not that every *sequence* has a convergent subsequence. In non-metrizable topological spaces, compactness and sequential compactness are logically independent. The weak* topology on $B_{X^*}$ is metrizable if and only if $X$ is [separable](/page/Separable), and it is only in this case that Banach-Alaoglu yields convergent subsequences from sequences. For non-separable $X$ (such as $X = \ell^\infty(\mathbb{N})$), the theorem guarantees compactness of $B_{X^*}$ but does **not** guarantee that every bounded sequence in $X^*$ has a weak* convergent subsequence. **It does not say anything about $B_X$ (the unit ball of $X$ itself).** The theorem concerns the dual ball $B_{X^*}$, not the ball of the original space. For weak compactness of $B_X$, one needs reflexivity — this is the content of [Kakutani's theorem](/theorems/897), discussed below. ## Sequential Compactness and Separability For applications in PDE theory and the calculus of variations, one works with sequences, not nets. The abstract compactness of $B_{X^*}$ guaranteed by Banach-Alaoglu is not directly usable unless it can be upgraded to sequential compactness. This upgrade requires an additional structural hypothesis on $X$: **separability**. The mechanism is metrizability. When $X$ is separable, the weak* topology on bounded subsets of $X^*$ becomes metrizable, and in metrizable compact spaces, compactness is equivalent to sequential compactness. ### Metrizability of the Weak* Topology on Bounded Sets [quotetheorem:495] The construction deserves scrutiny. Each term $|f(x_k) - g(x_k)|/(1 + |f(x_k) - g(x_k)|)$ is bounded by $1$, so the geometric weights $2^{-k}$ guarantee absolute convergence. The function $t \mapsto t/(1+t)$ is a metric-preserving transformation of $[0, \infty)$: it is strictly increasing, subadditive, and vanishes only at $0$. These properties ensure that $d$ satisfies the triangle inequality and separates points: if $d(f, g) = 0$, then $f(x_k) = g(x_k)$ for all $k$, and since $\{x_k\}$ is dense and $f - g$ is continuous, $f = g$ on all of $X$. The equivalence of $d$-convergence and weak* convergence on $B_R$ rests on two arguments. In one direction, $d(f_n, f) \to 0$ implies $f_n(x_k) \to f(x_k)$ for every $k$, and density of $\{x_k\}$ combined with the uniform bound $\|f_n\|_{X^*} \le R$ extends this to $f_n(x) \to f(x)$ for all $x \in X$ (by an $\varepsilon/3$ argument). In the other direction, weak* convergence gives $f_n(x_k) \to f(x_k)$ for each $k$, and dominated convergence on the series (with dominating sequence $2^{-k}$) gives $d(f_n, f) \to 0$. The restriction to bounded sets is essential. The weak* topology on the entire space $X^*$ is *never* metrizable when $X$ is infinite-dimensional — it is not even first-countable at the origin (a basic neighbourhood is determined by finitely many test points, and $X$ cannot be covered by countably many such finite sets). Metrizability holds only on norm-bounded subsets. ### The Sequential Banach-Alaoglu Theorem The combination of Banach-Alaoglu (compactness) and weak* metrizability on bounded sets (metrizability) immediately yields the sequential version. [quotetheorem:496] Moreover, the weak* limit $f$ satisfies the norm bound $\|f\|_{X^*} \le \liminf_{k \to \infty} \|f_{n_k}\|_{X^*}$, a consequence of the weak* lower semicontinuity of the dual norm. The proof is immediate from the preceding results. Let $R := \sup_n \|f_n\|_{X^*}$. By Banach-Alaoglu, the ball $B_R$ is weak* compact. By the metrizability theorem, $B_R$ is metrizable in the weak* topology. A compact metrizable space is sequentially compact (by the standard diagonal/total-boundedness argument), so $\{f_n\}$ has a subsequence converging in the $d$-metric, hence converging weak*. The norm bound on the limit follows from weak* lower semicontinuity of the norm: for each $x$ with $\|x\|_X \le 1$, $|f(x)| = \lim_k |f_{n_k}(x)| \le \liminf_k \|f_{n_k}\|_{X^*}$. **Separability cannot be dropped.** Consider $X = \ell^\infty(\mathbb{N})$, which is not separable. Define $f_n \in (\ell^\infty)^*$ by $f_n(x) = x_n$ for $x = (x_1, x_2, \ldots) \in \ell^\infty$. Each $f_n$ has $\|f_n\|_{(\ell^\infty)^*} = 1$, so the sequence is bounded. If a subsequence $f_{n_k}$ converged weak* to some $f$, then for every $x \in \ell^\infty$, $x_{n_k} \to f(x)$. But taking $x = \mathbb{1}_S$ where $S = \{n_{2k}\}_{k=1}^\infty$ gives $f_{n_k}(x) = \mathbb{1}_S(n_k)$, which alternates between $0$ and $1$ and does not converge. Thus, no subsequence of $\{f_n\}$ converges weak* — sequential compactness fails without separability. [example: Extracting a Weak* Limit in $L^\infty$] Let $X = L^1(0, 1)$, which is separable (the rational step functions are countable and dense). Its dual is $X^* = L^\infty(0, 1)$. Consider the bounded sequence $g_n \in L^\infty(0, 1)$ defined by \begin{align*} g_n(t) := \sin(2\pi n t). \end{align*} We have $\|g_n\|_{L^\infty} = 1$ for all $n$. By the Sequential Banach-Alaoglu theorem, some subsequence converges weak* in $L^\infty$. In fact, the entire sequence converges: for any $h \in L^1(0, 1)$, \begin{align*} \int_0^1 h(t) \sin(2\pi n t) \, d\mathcal{L}^1(t) \to 0 \quad \text{as } n \to \infty, \end{align*} by the Riemann-Lebesgue lemma. So $g_n \overset{*}{\rightharpoonup} 0$ in $L^\infty(0, 1)$. The weak* limit $g = 0$ satisfies $\|g\|_{L^\infty} = 0 < 1 = \liminf_n \|g_n\|_{L^\infty}$: the norm strictly decreases in the limit. This is consistent with the weak* lower semicontinuity of the norm but shows that equality need not hold — oscillating sequences can "lose mass" in the weak* limit, just as they do in weak limits. [/example] ## Reflexivity and Weak Compactness Banach-Alaoglu gives compactness of the dual ball $B_{X^*}$ in the weak* topology. But in PDE theory one typically works with function spaces $X$ (such as [Sobolev spaces](/page/Sobolev%20Space) $W^{1,p}(U)$ for $1 < p < \infty$), not with their duals. The question is: **when is the unit ball of $X$ itself compact in a weak topology?** The answer involves a precise interplay between Banach-Alaoglu, the canonical embedding $J: X \to X^{**}$, and the notion of [reflexivity](/page/Reflexive%20Space). ### From Banach-Alaoglu to Kakutani The canonical embedding $J: X \to X^{**}$, defined by $J(x)(f) := f(x)$ for $x \in X$ and $f \in X^*$, is an isometric injection. Applying Banach-Alaoglu to the [Banach space](/page/Banach%20Space) $X^*$ (viewed as a normed space with dual $X^{**}$) gives weak* compactness of $B_{X^{**}}$ in the topology $\sigma(X^{**}, X^*)$. When $X$ is **reflexive** — meaning $J$ is surjective — the identification $X \cong X^{**}$ converts the weak* topology $\sigma(X^{**}, X^*)$ on $X^{**}$ into the [weak topology](/page/Weak%20Topology) $\sigma(X, X^*)$ on $X$. Under this identification, Banach-Alaoglu for $B_{X^{**}}$ becomes weak compactness of $B_X$. When $X$ is **not** reflexive, $J(B_X)$ is a proper subset of $B_{X^{**}}$; it is dense by [Goldstine's Lemma](/theorems/898) but not closed, hence not compact. The compactness does not transfer. This is the content of Kakutani's theorem, which gives a clean topological characterisation of reflexivity. [quotetheorem:897] Kakutani's theorem transforms reflexivity from an algebraic property (surjectivity of $J$) into a topological one (weak compactness of the unit ball). The forward direction follows from Banach-Alaoglu as sketched above. The converse is more subtle: if $B_X$ is weakly compact, then $J(B_X)$ is weak* compact in $X^{**}$ (since $J$ is weak-to-weak* continuous), hence weak* closed. But Goldstine's Lemma says $J(B_X)$ is weak* dense in $B_{X^{**}}$. A subset that is simultaneously dense and closed must be the entire space, so $J(B_X) = B_{X^{**}}$, giving $J(X) = X^{**}$. This result explains the structural role of reflexivity in existence theory. The spaces $L^p(U)$ for $1 < p < \infty$ and the Sobolev spaces $W^{k,p}(U)$ for $1 < p < \infty$ are all reflexive, so their unit balls are weakly compact. The spaces $L^1(U)$ and $L^\infty(U)$ are not reflexive, and their unit balls are not weakly compact — bounded sequences in $L^1$ can concentrate into measures, and bounded sequences in $L^\infty$ can oscillate without producing weak limits in the space. ### The Logical Chain The complete picture of compactness in infinite-dimensional spaces involves four theorems, each building on the last: 1. **Banach-Alaoglu** (all normed spaces): $B_{X^*}$ is weak* compact. 2. **Goldstine's Lemma** (all normed spaces): $J(B_X)$ is weak* dense in $B_{X^{**}}$. 3. **Kakutani** (Banach spaces): $X$ is reflexive $\Leftrightarrow$ $B_X$ is weakly compact. 4. **Eberlein-Smulian** (Banach spaces): weak compactness $\Leftrightarrow$ weak sequential compactness for bounded sets. Combining (3) and (4) yields the result most directly used in applications: *a [Banach space](/page/Banach%20Space) is reflexive if and only if every bounded sequence has a [weakly convergent](/page/Weak%20Topology) subsequence.* ### Goldstine's Lemma: The Missing Piece [quotetheorem:898] Goldstine's Lemma quantifies the gap between a space and its bidual. The unit ball $B_X$, mapped into $X^{**}$, fills out $B_{X^{**}}$ up to weak* closure. Whether this dense subset is actually the whole ball is precisely the question of reflexivity. In the reflexive case, the dense set is closed (hence equal to $B_{X^{**}}$); in the non-reflexive case, it is a proper dense subset — the closure adds new elements that do not come from $X$. [example: Goldstine's Lemma for $c_0$ and $\ell^\infty$] Let $X = c_0(\mathbb{N})$, the space of real sequences converging to $0$, with the supremum norm. The dual is $X^* = \ell^1(\mathbb{N})$, and the bidual is $X^{**} = \ell^\infty(\mathbb{N})$. The canonical embedding $J: c_0 \to \ell^\infty$ is the inclusion map. Goldstine's Lemma asserts that the unit ball of $c_0$ is weak* dense in the unit ball of $\ell^\infty$ (with the weak* topology $\sigma(\ell^\infty, \ell^1)$). Concretely, this means: for any $y = (y_1, y_2, \ldots) \in \ell^\infty$ with $\|y\|_{\ell^\infty} \le 1$, any finite collection $a_1, \ldots, a_m \in \ell^1$, and any $\varepsilon > 0$, there exists $x \in c_0$ with $\|x\|_{c_0} \le 1$ and \begin{align*} \left|\sum_{n=1}^\infty a_j(n)(x_n - y_n)\right| < \varepsilon \quad \text{for } j = 1, \ldots, m. \end{align*} The truncation $x^{(N)} := (y_1, \ldots, y_N, 0, 0, \ldots) \in c_0$ satisfies $\|x^{(N)}\|_{c_0} \le \|y\|_{\ell^\infty} \le 1$ and $\sum_n a_j(n)(x^{(N)}_n - y_n) = -\sum_{n > N} a_j(n) y_n \to 0$ as $N \to \infty$, since $a_j \in \ell^1$. So $J(x^{(N)}) \to y$ in $\sigma(\ell^\infty, \ell^1)$. The constant sequence $y = (1, 1, 1, \ldots) \in \ell^\infty$ is the weak* limit of elements of $c_0$, but $y \notin c_0$. This exhibits the non-reflexivity of $c_0$: the image $J(B_{c_0})$ is dense but not closed in $B_{\ell^\infty}$. [/example] ## Applications to PDE Existence Theory The primary consumer of the Banach-Alaoglu theorem is the existence theory for partial differential equations, where the standard strategy has three steps: (1) establish a uniform bound on a sequence of approximate solutions, (2) extract a weakly (or weak*) convergent subsequence, (3) verify that the limit solves the equation. Step (2) is where Banach-Alaoglu enters. ### The Direct Method of the Calculus of Variations The prototype application is the **direct method**: to find a minimiser of an energy functional $\mathcal{E}: X \to \mathbb{R} \cup \{+\infty\}$ over a reflexive Banach space $X$. **Step 1 (Coercivity and boundedness).** Show that $\mathcal{E}(u) \to +\infty$ as $\|u\|_X \to \infty$, so that any minimising sequence $(u_n)$ with $\mathcal{E}(u_n) \to \inf \mathcal{E}$ is bounded: $\|u_n\|_X \le C$. **Step 2 (Subsequence extraction).** Since $X$ is reflexive and $(u_n)$ is bounded, the Eberlein-Smulian theorem (which rests on Banach-Alaoglu and Kakutani) provides a subsequence $u_{n_k} \rightharpoonup u^*$ weakly in $X$. **Step 3 (Weak lower semicontinuity).** Show that $\mathcal{E}(u^*) \le \liminf_k \mathcal{E}(u_{n_k}) = \inf \mathcal{E}$, so $u^*$ is a minimiser. The structure of this argument explains *why* reflexivity is needed: Step 2 requires that bounded sequences have weakly convergent subsequences, and Kakutani's theorem says this is exactly reflexivity. It also explains *why* the calculus of variations focuses on $L^p$ spaces with $1 < p < \infty$: these are reflexive, while $L^1$ and $L^\infty$ are not. ### Weak* Compactness for Non-Reflexive Spaces When the natural function space is not reflexive — $L^1$, $L^\infty$, the space of Radon measures $\mathcal{M}(\Omega)$ — one cannot use weak compactness directly. Instead, one embeds the problem into a dual space and uses weak* compactness via Sequential Banach-Alaoglu. [example: Weak* Limits of Bounded Sequences of Measures] Let $\Omega \subset \mathbb{R}^n$ be a bounded open set. Consider a sequence of nonnegative Radon measures $\{\mu_k\}_{k=1}^\infty$ on $\overline{\Omega}$ with uniformly bounded total variation: \begin{align*} \sup_k \|\mu_k\|_{\mathcal{M}(\overline{\Omega})} = \sup_k \mu_k(\overline{\Omega}) \le C. \end{align*} The space $\mathcal{M}(\overline{\Omega})$ is the dual of $C(\overline{\Omega})$ (by the Riesz-Markov-Kakutani theorem), and $C(\overline{\Omega})$ is separable (by the Stone-Weierstrass theorem, polynomials with rational coefficients are countable and dense). By the Sequential Banach-Alaoglu theorem, there exist a subsequence $\mu_{k_j}$ and a measure $\mu \in \mathcal{M}(\overline{\Omega})$ with \begin{align*} \int_{\overline{\Omega}} \phi \, d\mu_{k_j} \to \int_{\overline{\Omega}} \phi \, d\mu \quad \text{for every } \phi \in C(\overline{\Omega}). \end{align*} This is the starting point for concentration-compactness arguments, for proving existence of minimisers in geometric measure theory, and for the theory of Young measures. [/example] ### Galerkin Approximations and Weak Limits A second major application pattern arises in Galerkin methods for evolution equations. One constructs finite-dimensional approximate solutions $u_n$ (by projection onto an $n$-dimensional subspace), establishes uniform energy estimates $\|u_n\|_X \le C$ in a reflexive space $X$ (typically a Sobolev space), and then passes to the limit using weak compactness. The bound and Banach-Alaoglu (via Eberlein-Smulian) produce a subsequence $u_{n_k} \rightharpoonup u$ weakly in $X$; the weak formulation of the equation, which is linear in the test functions, passes to the limit, and $u$ solves the equation. This pattern is used to prove existence of weak solutions for elliptic boundary value problems (the Lax-Milgram theorem provides a more direct route for coercive problems, but Galerkin methods extend to nonlinear and time-dependent equations), parabolic equations (where one works in spaces like $L^2(0, T; H^1_0(U))$), and systems arising in fluid mechanics. ## Standard Arguments with the Banach-Alaoglu Theorem The following techniques appear repeatedly in functional analysis and PDE theory. Each uses Banach-Alaoglu (or its sequential form) as a key ingredient. ### Diagonal Extraction for Countable Families When one needs weak* convergence simultaneously for a countable family of sequences, the standard device is **diagonal extraction**. Suppose $X$ is separable and $\{f_n^{(j)}\}_{n=1}^\infty$ is a bounded sequence in $X^*$ for each $j \in \mathbb{N}$. By Sequential Banach-Alaoglu, extract a subsequence of $(f_n^{(1)})$ converging weak*; from this, extract a further subsequence of $(f_n^{(2)})$ converging weak*; continue. The diagonal subsequence $f_{n_k}^{(k)}$ converges weak* simultaneously for all $j$. This technique is ubiquitous in regularity theory, where one applies difference quotient estimates in each coordinate direction and extracts a single subsequence converging in all directions simultaneously. ### Bounded-to-Weak* Closure of Convex Sets Mazur's theorem states that the weak closure of a convex set in a Banach space equals its norm closure. The weak* analogue is: a convex subset $C \subset X^*$ is weak* closed if and only if $C \cap B_R$ is weak* closed for every $R > 0$. The proof uses Banach-Alaoglu: since $B_R$ is weak* compact, $C \cap B_R$ is weak* compact whenever it is weak* closed, and compactness provides the sequential characterisation (in the separable case) needed to verify closure. This criterion — known as the **Krein-Smulian theorem** — is essential for verifying that the range of an adjoint operator is weak* closed, a key step in proving surjectivity results for operators between Banach spaces. ### Upgrading Weak* to Weak Convergence The Banach-Alaoglu theorem is a statement about dual spaces — it gives compactness of $B_{X^*}$, not of $B_X$. In a reflexive space $X$, the canonical embedding $J: X \to X^{**}$ is surjective, so one can identify $X$ with $X^{**}$ and $B_X$ with $B_{X^{**}}$. Under this identification, the weak* topology $\sigma(X^{**}, X^*)$ on $X^{**}$ corresponds to the weak topology $\sigma(X, X^*)$ on $X$. Applying Banach-Alaoglu to the normed space $X^*$ (whose dual is $X^{**}$) gives weak* compactness of $B_{X^{**}}$, which translates to weak compactness of $B_X$. At the level of sequences: if $X$ is reflexive and $(x_n) \subset X$ is bounded, then $J(x_n)$ is a bounded sequence in $X^{**}$. By Banach-Alaoglu applied to $X^{**}$ (with separability of $X^*$ providing the sequential version), a subsequence satisfies $J(x_{n_k}) \overset{*}{\rightharpoonup} \Phi$ in $\sigma(X^{**}, X^*)$. Since $J$ is surjective, $\Phi = J(x)$ for some $x \in X$, and $J(x_{n_k}) \overset{*}{\rightharpoonup} J(x)$ translates to $x_{n_k} \rightharpoonup x$ in $\sigma(X, X^*)$ — weak convergence in the original space. This is the mechanism by which Banach-Alaoglu, a theorem about dual spaces, yields weak convergence results for the original space in the reflexive case. [example: Extracting Weak Limits in Sobolev Spaces] Let $U \subset \mathbb{R}^n$ be a bounded open set with $C^1$ boundary, and let $1 < p < \infty$. Suppose $\{u_k\}_{k=1}^\infty \subset W^{1,p}(U)$ satisfies the uniform bound \begin{align*} \|u_k\|_{W^{1,p}(U)} \le M \quad \text{for all } k \in \mathbb{N}. \end{align*} Since $W^{1,p}(U)$ is reflexive (it is the product $L^p \times (L^p)^n$ with an equivalent norm, and products of reflexive spaces are reflexive), the Eberlein-Smulian theorem provides a subsequence $u_{k_j} \rightharpoonup u$ weakly in $W^{1,p}(U)$. Explicitly, this means: \begin{align*} \int_U u_{k_j} \phi \, d\mathcal{L}^n &\to \int_U u \phi \, d\mathcal{L}^n \quad \text{for every } \phi \in L^q(U), \\ \int_U \partial_{x_i} u_{k_j} \psi \, d\mathcal{L}^n &\to \int_U \partial_{x_i} u \psi \, d\mathcal{L}^n \quad \text{for every } \psi \in L^q(U), \end{align*} where $q = p/(p-1)$ is the conjugate exponent. The Rellich-Kondrachov theorem upgrades the convergence of $u_{k_j}$ itself (not its derivatives) to strong convergence in $L^r(U)$ for $r < p^* = np/(n-p)$ when $p < n$. But the derivatives $\nabla u_{k_j}$ converge only weakly in $L^p$ — this is the best that Banach-Alaoglu can provide without additional structural information. [/example] ## Comparison: Weak*, Weak, and Strong Compactness The relationship between the three topologies on a dual space — and the corresponding notions of compactness — is a source of frequent confusion. The following summary clarifies the distinctions. | Property | Norm topology | Weak topology $\sigma(X^*, X^{**})$ | Weak* topology $\sigma(X^*, X)$ | |---|---|---|---| | $B_{X^*}$ compact? | Never (if $\dim X = \infty$) | Iff $X^*$ is reflexive | **Always** (Banach-Alaoglu) | | Sequential compactness of $B_{X^*}$? | Never (if $\dim X = \infty$) | Iff $X^*$ is reflexive (Eberlein-Smulian) | If $X$ is separable | | Metrizability of $B_{X^*}$? | Always | If $X^*$ is separable | If $X$ is separable | | Convergence tested by | $\|f_n - f\|_{X^*} \to 0$ | $\Phi(f_n) \to \Phi(f)$ for all $\Phi \in X^{**}$ | $f_n(x) \to f(x)$ for all $x \in X$ | The table reveals why the weak* topology is the natural setting for compactness arguments: it is the only topology that provides compactness of $B_{X^*}$ without any structural assumption on $X$. The price is that weak* convergence is the weakest form of convergence, and many nonlinear operations (such as composition with nonlinear functions, or products of weakly convergent sequences) do not pass to the limit. [remark: Weak* Compactness Does Not Imply Weak Compactness] The inclusion $\sigma(X^*, X) \subset \sigma(X^*, X^{**})$ means that every weakly compact subset of $X^*$ is automatically weak* compact, but the converse fails. For instance, $B_{\ell^\infty}$ is weak* compact in $\sigma(\ell^\infty, \ell^1)$ (Banach-Alaoglu, since $\ell^\infty = (\ell^1)^*$), but $B_{\ell^\infty}$ is **not** weakly compact (since $\ell^\infty$ is not reflexive). The distinction matters in practice: weak compactness is needed for the direct method in the original space, while weak* compactness suffices for dual-space arguments. [/remark] ## References - H. Brezis, *Functional Analysis, Sobolev Spaces and Partial Differential Equations* (2011). - J. Conway, *A Course in Functional Analysis* (1990). - L.C. Evans, *Partial Differential Equations* (2010). - W. Rudin, *Functional Analysis* (1991). - E. Kreyszig, *Introductory Functional Analysis with Applications* (1978).