Given a family of [topological spaces](/page/Topology) $\{(X_\alpha, \tau_\alpha)\}_{\alpha \in A}$, we can form the Cartesian product $\prod_{\alpha \in A} X_\alpha$ as a set — its elements are functions $x: A \to \bigcup_{\alpha \in A} X_\alpha$ with $x(\alpha) \in X_\alpha$ for each $\alpha$. But what topology should this set carry? Two natural requirements guide the answer. First, the coordinate projections $\pi_\alpha: \prod_{\beta \in A} X_\beta \to X_\alpha$ should be [continuous](/page/Continuity) — a topology that forgets the structure of individual factors would be useless. Second, the topology should be no finer than necessary: imposing extra open sets beyond what these continuity requirements force would destroy [compactness](/page/Compact%20Space) and other limit-based properties in infinite products. These two requirements — continuity of projections and minimality — single out the **product topology**. Yet there is a competing candidate, the **box topology**, which declares *all* products of open sets (not just finitely many non-trivial ones) to be open. For finite index sets, the two topologies agree. For infinite products, they diverge dramatically: the box topology has too many open sets, and fundamental results like Tychonoff's theorem and the continuity of diagonal maps fail in it. [example: A Continuous Map That the Box Topology Breaks] Consider the sequence spaces $X_k = \mathbb{R}$ for $k \in \mathbb{N}$, so $\prod_{k=1}^\infty X_k = \mathbb{R}^{\mathbb{N}}$, the space of all real sequences. Define the map \begin{align*} f: \mathbb{R} &\to \mathbb{R}^{\mathbb{N}} \\ t &\mapsto (t, t, t, \ldots), \end{align*} sending each real number to the constant sequence. Each coordinate function $\pi_k \circ f: \mathbb{R} \to \mathbb{R}$ is simply the identity, which is continuous. Any reasonable topology on $\mathbb{R}^{\mathbb{N}}$ should therefore make $f$ continuous. In the **product topology**, $f$ is indeed continuous — this follows directly from the universal property, since each $\pi_k \circ f$ is continuous. In the **box topology**, however, $f$ fails to be continuous. Consider the open set \begin{align*} U = \prod_{k=1}^\infty \left(-\frac{1}{k}, \frac{1}{k}\right). \end{align*} This is open in the box topology (each factor is open in $\mathbb{R}$). The preimage is \begin{align*} f^{-1}(U) = \{t \in \mathbb{R} : t \in (-1/k, 1/k) \text{ for all } k \in \mathbb{N}\} = \bigcap_{k=1}^\infty \left(-\frac{1}{k}, \frac{1}{k}\right) = \{0\}. \end{align*} Since the singleton $\{0\}$ is not open in $\mathbb{R}$, the map $f$ is not continuous with respect to the box topology. The box topology declares too many sets open, so preimages of open sets can shrink to points. [/example] This example captures the central tension: the box topology respects the topology of each factor individually but ignores the global constraint that "most" coordinates should be unrestricted. The product topology resolves this by requiring that a basic open set restrict only finitely many coordinates. ## Definition [definition: Product Topology] Let $\{(X_\alpha, \tau_\alpha)\}_{\alpha \in A}$ be a family of [topological spaces](/page/Topology) indexed by a set $A$. The **product topology** (also called the **Tychonoff topology**) on the Cartesian product $\prod_{\alpha \in A} X_\alpha$ is the topology generated by the subbasis \begin{align*} \mathcal{S} = \left\{ \pi_\alpha^{-1}(U_\alpha) : \alpha \in A, \, U_\alpha \in \tau_\alpha \right\}, \end{align*} where $\pi_\alpha: \prod_{\beta \in A} X_\beta \to X_\alpha$ denotes the $\alpha$-th coordinate projection, defined by $\pi_\alpha(x) = x(\alpha)$. Equivalently, the product topology is the coarsest topology on $\prod_{\alpha \in A} X_\alpha$ for which every projection $\pi_\alpha$ is continuous. [/definition] A subbasis generates a topology by taking all finite intersections (to form a basis) and then all arbitrary unions. In this case, the basis consists of sets of the form \begin{align*} \bigcap_{i=1}^m \pi_{\alpha_i}^{-1}(U_{\alpha_i}) = \left\{ x \in \prod_{\alpha \in A} X_\alpha : x(\alpha_i) \in U_{\alpha_i} \text{ for } i = 1, \ldots, m \right\}, \end{align*} where $m \in \mathbb{N}$, $\alpha_1, \ldots, \alpha_m \in A$, and $U_{\alpha_i} \in \tau_{\alpha_i}$. Such a set is called a **cylinder set** (or **elementary open set**): it constrains finitely many coordinates to lie in prescribed open sets and places no restriction on the remaining coordinates. This finite restriction is the defining feature that distinguishes the product topology from the box topology. [remark: Notation for Products] When the index set is $A = \{1, \ldots, n\}$, we write $X_1 \times X_2 \times \cdots \times X_n$ for $\prod_{k=1}^n X_k$. When all factors are the same space $X$, we write $X^A$ for $\prod_{\alpha \in A} X$. In particular, $\mathbb{R}^n = \prod_{k=1}^n \mathbb{R}$ with the product topology coincides with the standard Euclidean topology, and $\mathbb{R}^{\mathbb{N}} = \prod_{k=1}^\infty \mathbb{R}$ is the space of all real sequences. [/remark] For comparison, we record the competing topology. [definition: Box Topology] Let $\{(X_\alpha, \tau_\alpha)\}_{\alpha \in A}$ be a family of topological spaces. The **box topology** on $\prod_{\alpha \in A} X_\alpha$ is the topology generated by the basis \begin{align*} \mathcal{B}_{\text{box}} = \left\{ \prod_{\alpha \in A} U_\alpha : U_\alpha \in \tau_\alpha \text{ for every } \alpha \in A \right\}. \end{align*} [/definition] In the box topology, a basic open set allows *every* coordinate to be independently restricted. In the product topology, a basic open set restricts only *finitely many* coordinates — all other factors are the full space $X_\alpha$. Since every product-basic open set is also box-basic open (take $U_\alpha = X_\alpha$ for the unrestricted coordinates), the product topology is always coarser than the box topology: \begin{align*} \tau_{\text{product}} \subset \tau_{\text{box}}. \end{align*} For finite index sets, every box-basic open set restricts only finitely many coordinates (because there are only finitely many to restrict), so the two topologies agree. The discrepancy emerges only for infinite products. ## Continuous Maps and the Universal Property The product topology is not merely "a natural choice" — it is the *unique* topology characterized by a universal property. This is the deeper reason why the product topology, rather than the box topology, is the correct construction: it is the categorical product in the category of topological spaces and [continuous maps](/page/Continuity%20(Metric%20Spaces)). The key question is: given topological spaces $Z$ and $\{X_\alpha\}_{\alpha \in A}$, when is a map $f: Z \to \prod_{\alpha \in A} X_\alpha$ continuous? The universal property answers: precisely when each "component" $\pi_\alpha \circ f: Z \to X_\alpha$ is continuous. [Continuity](/page/Continuity) of a map into a product reduces to checking continuity of each coordinate function separately. [quotetheorem:962] The universal property encodes a precise sense in which the product topology is *minimal*. Any topology $\tau$ on $\prod_{\alpha \in A} X_\alpha$ that makes all projections continuous must contain every subbasis element $\pi_\alpha^{-1}(U_\alpha)$, and hence must contain the product topology. If $\tau$ is strictly finer than the product topology (as the box topology is, for infinite products), then the "if" direction of property (2) can fail: a map whose components are all continuous need not be continuous into $(\prod X_\alpha, \tau)$. The opening example demonstrated exactly this failure for the box topology. [explanation: Why the Box Topology Violates the Universal Property] The box topology makes all projections continuous (since it is finer than the product topology), so condition (1) holds. However, condition (2) fails for infinite products. The diagonal map $\Delta: X \to X^{\mathbb{N}}$ defined by $\Delta(x) = (x, x, x, \ldots)$ has continuous components ($\pi_k \circ \Delta = \mathrm{id}_X$ for all $k$), yet $\Delta$ is not continuous in the box topology, as shown in the opening example. This means the box topology is *not* the categorical product: it is a legitimate topology on the underlying set $\prod X_\alpha$, but it does not satisfy the universal mapping property that defines products in the category of topological spaces. As a consequence, many constructions that rely on the universal property — building continuous maps into products by specifying components, recognizing product spaces via their mapping properties — fail in the box topology. [/explanation] ### Projections: Open but Not Closed The projections $\pi_\alpha$ are continuous by definition, but they have an additional property that is sometimes surprising: they are **open maps** (they send open sets to open sets). However, they are generally **not** closed maps. [quotetheorem:955] The openness of projections is essential for transferring local properties from the product to factors. For instance, it implies that a product of [Hausdorff spaces](/page/Topology) is Hausdorff: given distinct points $x \neq y$ in $\prod X_\alpha$, there exists some index $\alpha_0$ with $x(\alpha_0) \neq y(\alpha_0)$. Since $X_{\alpha_0}$ is Hausdorff, we find disjoint open sets $U, V \subset X_{\alpha_0}$ separating $x(\alpha_0)$ and $y(\alpha_0)$. The preimages $\pi_{\alpha_0}^{-1}(U)$ and $\pi_{\alpha_0}^{-1}(V)$ are disjoint open sets in the product separating $x$ and $y$. However, projections fail to be closed maps in general. The failure is instructive. [example: A Projection That Is Not Closed] Take $X_1 = X_2 = \mathbb{R}$ with the standard topology, so the product is $\mathbb{R}^2$. Consider the set \begin{align*} C = \{(x, y) \in \mathbb{R}^2 : xy = 1\}, \end{align*} the graph of the function $y = 1/x$. This is a closed subset of $\mathbb{R}^2$: it is the preimage of the closed set $\{1\}$ under the continuous map $(x, y) \mapsto xy$. The projection onto the first coordinate is \begin{align*} \pi_1(C) = \{x \in \mathbb{R} : \exists\, y \in \mathbb{R} \text{ with } xy = 1\} = \mathbb{R} \setminus \{0\}. \end{align*} This set is not closed in $\mathbb{R}$. The projection $\pi_1$ maps the closed set $C$ to the non-closed set $\mathbb{R} \setminus \{0\}$. Geometrically, the hyperbola $xy = 1$ "escapes to infinity" as $x \to 0$, and this asymptotic behavior prevents the image from being closed. The projection loses the limit point $0$ because there is no point $(0, y)$ on the curve — the sequence $(1/n, n)$ lies in $C$ and its first coordinates converge to $0$, but the corresponding second coordinates diverge. [/example] This failure has deep consequences. The **Tube Lemma** and the **closed map lemma** for compact spaces arise precisely as corrections to this failure, and Tychonoff's theorem relies essentially on compactness to recover closedness properties. ### Convergence in Product Spaces The product topology admits a concrete description of convergence: a net (or sequence, when applicable) converges in the product if and only if it converges in each coordinate. [quotetheorem:956] This characterisation follows directly from the universal property: the identity map from the net into the product is "continuous" (in the net-convergence sense) precisely when each coordinate is. For the space $\mathbb{R}^{\mathbb{N}}$ of real sequences, product-topology convergence of a sequence of sequences $(x_k)_{k=1}^\infty$ means that the $j$-th entry $x_k(j)$ converges for each fixed $j$ — this is exactly **pointwise convergence**. In the box topology, coordinatewise convergence is necessary but *not sufficient*. The box topology demands a kind of "uniform" convergence across all coordinates simultaneously, which is a strictly stronger requirement. [example: Coordinatewise Convergence Failing in the Box Topology] Consider the sequence of points in $\mathbb{R}^{\mathbb{N}}$ defined by \begin{align*} x_k = \left(\frac{1}{k}, \frac{1}{k}, \frac{1}{k}, \ldots \right) \end{align*} for $k \in \mathbb{N}$. Each coordinate converges to $0$: $\pi_j(x_k) = 1/k \to 0$ for every $j$. In the product topology, $x_k \to \mathbf{0} = (0, 0, 0, \ldots)$. In the box topology, this convergence fails. The open set $U = \prod_{j=1}^\infty (-1/j, 1/j)$ contains $\mathbf{0}$, but for each $k$, the point $x_k$ fails to lie in $U$ as soon as $j > k$ (since $1/k \ge 1/j$ for $j \le k$ but $1/k < 1/j$ requires $j < k$). More precisely, $x_k \in U$ requires $1/k < 1/j$ for all $j$, which fails for $j \ge k$. Hence no tail of the sequence lies in $U$, and $x_k \not\to \mathbf{0}$ in the box topology. [/example] ## The Box Topology: Pathologies of the Wrong Choice The opening examples have already illustrated two failures of the box topology: the diagonal map loses continuity, and coordinatewise convergent sequences may not converge. These are symptoms of a deeper problem. The box topology is simply too fine — it has too many open sets — for the standard theorems of general topology to hold on infinite products. We collect the most important pathologies here to motivate why the product topology is the standard choice. ### Failure of Compactness The most catastrophic failure is the loss of Tychonoff's theorem. Let $X_k = \{0, 1\}$ with the discrete topology for each $k \in \mathbb{N}$. Each $X_k$ is finite, hence compact. The product $\{0, 1\}^{\mathbb{N}}$ in the product topology is compact by Tychonoff's theorem — it is homeomorphic to the Cantor set. In the box topology, $\{0, 1\}^{\mathbb{N}}$ is **discrete**: every singleton $\{x\}$ is open, because $\{x\} = \prod_{k=1}^\infty \{x(k)\}$ and each $\{x(k)\}$ is open in the discrete topology. Since $\{0, 1\}^{\mathbb{N}}$ is uncountable (it bijects with $\mathcal{P}(\mathbb{N})$, or equivalently with $\mathbb{R}$ via binary expansions), the box topology produces an uncountable discrete space, which is not compact — the open cover by singletons has no finite subcover. ### Failure of Connectedness A similar pathology afflicts connectedness. Take $X_k = \mathbb{R}$ for each $k \in \mathbb{N}$. Each factor is connected. In the product topology, $\mathbb{R}^{\mathbb{N}}$ is connected (as we will prove later). In the box topology, $\mathbb{R}^{\mathbb{N}}$ is **not connected**. Define \begin{align*} B = \left\{x \in \mathbb{R}^{\mathbb{N}} : \sup_{k \in \mathbb{N}} |x(k)| < \infty \right\} \end{align*} (the set of bounded sequences) and its complement $U = \mathbb{R}^{\mathbb{N}} \setminus B$ (the unbounded sequences). Both $B$ and $U$ are nonempty. We claim $B$ is open in the box topology: for any bounded sequence $x$ with $|x(k)| \le M$, the set $\prod_{k=1}^\infty (x(k) - 1, x(k) + 1)$ is a box-open neighbourhood of $x$ contained in the set of sequences bounded by $M + 1$. By a similar argument (replacing "bounded" with "unbounded to the same degree"), $U$ can be shown to be open. This produces a disconnection of $\mathbb{R}^{\mathbb{N}}$ in the box topology. ### When the Two Topologies Agree Despite these pathologies, the box topology and product topology coincide for finite products. If $A = \{1, \ldots, n\}$, every subset of $A$ is finite, so every box-basic open set $\prod_{k=1}^n U_k$ already restricts only finitely many coordinates. The two bases are identical, and the topologies agree. [quotetheorem:957] In particular, $\mathbb{R}^n$ with the product topology (viewing $\mathbb{R}^n = \mathbb{R} \times \cdots \times \mathbb{R}$) is the same as $\mathbb{R}^n$ with the standard Euclidean topology. Open rectangles $(a_1, b_1) \times \cdots \times (a_n, b_n)$ form a basis for both topologies, and the product-topology open sets are exactly the Euclidean open sets. ## Tychonoff's Theorem Compactness is the single most important topological property for analysis: it provides convergent subsequences, attainment of suprema, and finite covers. The question of whether compactness is preserved under arbitrary products is therefore fundamental. For finite products, the answer is straightforward: if $X_1, \ldots, X_n$ are compact, then $X_1 \times \cdots \times X_n$ is compact. The standard proof uses the Tube Lemma to reduce an open cover of the product to a cover of a "tube" around each slice. But the Tube Lemma is a tool for finitely many coordinates — it cannot be iterated over infinitely many factors. The extension to arbitrary products is **Tychonoff's theorem**, one of the most consequential results in general topology. Its proof requires the Axiom of Choice (it is in fact equivalent to it), and it holds only for the product topology — not for the box topology, as the earlier counterexample with $\{0, 1\}^{\mathbb{N}}$ demonstrated. [quotetheorem:953] The "only if" direction is immediate: each projection $\pi_\alpha$ is continuous and surjective, so $X_\alpha = \pi_\alpha(\prod X_\beta)$ is the continuous image of a compact space, hence compact. The substance lies in the "if" direction. The proof strategy most commonly used (due to Alexander) exploits the **Alexander Subbasis Theorem**: a space is compact if and only if every cover by subbasis elements has a finite subcover. Since the subbasis of the product topology consists of sets $\pi_\alpha^{-1}(U_\alpha)$, compactness of the product reduces to a combinatorial argument about covers by such "single-coordinate" sets, where the compactness of each factor can be applied directly. ### Significance and Applications Tychonoff's theorem is far more than an abstract curiosity — it is the engine behind many foundational results across analysis and topology. **The Banach-Alaoglu theorem.** The closed unit ball $B_{X^*}$ of the dual of a [normed space](/page/Normed%20Vector%20Space) $X$ is compact in the [weak* topology](/page/Weak*%20Topology). The proof embeds $B_{X^*}$ into a product of compact intervals $\prod_{x \in B_X} [-\|x\|, \|x\|]$ via the evaluation map, applies Tychonoff's theorem to conclude the product is compact, and verifies that $B_{X^*}$ is a closed subset of this compact product. **Existence of ultrafilters.** Tychonoff's theorem (or equivalently, the Axiom of Choice) implies that every filter on a set can be extended to an ultrafilter. Ultrafilters provide an alternative proof of Tychonoff's theorem and underlie the construction of ultrapowers in model theory. **Stone-Cech compactification.** The compactification $\beta X$ of a completely regular space $X$ is constructed by embedding $X$ into a product $[0, 1]^{C(X, [0, 1])}$, which is compact by Tychonoff's theorem. **Probability theory.** Kolmogorov's existence theorem constructs a probability measure on an infinite product space $\mathbb{R}^{\mathbb{N}}$ (or more generally $\mathbb{R}^T$ for an arbitrary index set $T$). The underlying topology is the product topology, and Tychonoff's theorem (combined with regularity of measures) ensures that the cylinder-set construction extends to a full measure. ### Tychonoff's Theorem and the Axiom of Choice The equivalence between Tychonoff's theorem and the Axiom of Choice (proved by Kelley, 1950) means that any proof of the full theorem must use some form of Choice. The finite version — compactness of $X_1 \times \cdots \times X_n$ — requires no choice and can be proved by induction using the Tube Lemma. [example: The Cantor Space as a Tychonoff Product] The **Cantor set** $\mathcal{C} \subset [0, 1]$ — constructed by iteratively removing middle thirds — is homeomorphic to the product space $\{0, 2\}^{\mathbb{N}}$, where $\{0, 2\}$ carries the discrete topology. The homeomorphism sends a sequence $(a_k)_{k=1}^\infty \in \{0, 2\}^{\mathbb{N}}$ to the point \begin{align*} \phi: \{0, 2\}^{\mathbb{N}} &\to \mathcal{C} \\ (a_k)_{k=1}^\infty &\mapsto \sum_{k=1}^\infty \frac{a_k}{3^k}. \end{align*} Since $\{0, 2\}$ is finite and discrete, it is compact. By Tychonoff's theorem, $\{0, 2\}^{\mathbb{N}}$ is compact in the product topology. Since $\phi$ is a continuous bijection from a compact space to a Hausdorff space, it is a homeomorphism. This gives a "coordinate-free" proof that the Cantor set is compact, totally disconnected, and has no isolated points — properties inherited from the product structure. The product perspective also reveals the Cantor set's universality: every compact [metrizable](/page/Metrizable%20Space) space is a continuous image of $\{0, 2\}^{\mathbb{N}}$, and every totally disconnected compact metrizable space embeds into it. [/example] ## Connectedness and Separability of Products ### Connectedness The question of whether [connectedness](/page/Connectedness) is preserved by products is more subtle than it first appears. For a finite product $X \times Y$, the proof that connectedness is preserved uses the fact that slices $X \times \{y\}$ and $\{x\} \times Y$ are connected and share a common point, so their union is connected. This argument extends by induction to any finite product. For infinite products, induction is no longer available. The proof requires a different strategy: one shows that the "sections through a fixed point" are dense in the product, and that the closure of a connected set is connected. [quotetheorem:963] The proof proceeds as follows. Fix a point $p \in \prod_{\alpha \in A} X_\alpha$. For any finite subset $F \subset A$, define the "section through $p$ at $F$" to be the subspace \begin{align*} S_F = \left\{ x \in \prod_{\alpha \in A} X_\alpha : x(\alpha) = p(\alpha) \text{ for all } \alpha \notin F \right\}. \end{align*} This subspace is homeomorphic to the finite product $\prod_{\alpha \in F} X_\alpha$, which is connected (by the finite case). All sections $S_F$ share the common point $p$, so their union $\bigcup_F S_F$ is connected. One then verifies that $\bigcup_F S_F$ is dense in $\prod_{\alpha \in A} X_\alpha$ in the product topology: every basic open set (which restricts only finitely many coordinates) intersects some $S_F$. Since the closure of a connected set is connected, the full product is connected. This argument fails entirely in the box topology. In the box topology, the sections $S_F$ are not dense — there are box-open sets (restricting infinitely many coordinates) that no section intersects. As shown in the pathologies section, $\mathbb{R}^{\mathbb{N}}$ with the box topology is disconnected. ### Path Connectedness If each factor is [path connected](/page/Path%20Connectedness), the product is also path connected. Given two points $x, y \in \prod_{\alpha \in A} X_\alpha$, choose a path $\gamma_\alpha: [0, 1] \to X_\alpha$ from $x(\alpha)$ to $y(\alpha)$ in each factor. Define $\gamma: [0, 1] \to \prod_{\alpha \in A} X_\alpha$ by $\gamma(t)(\alpha) = \gamma_\alpha(t)$. By the universal property, $\gamma$ is continuous because each coordinate function $\pi_\alpha \circ \gamma = \gamma_\alpha$ is continuous. So $\gamma$ is a path from $x$ to $y$. ### Separability A [topological space](/page/Topology) is [separable](/page/Separable) if it contains a [countable](/page/Countable%20Set) dense subset. The product of separable spaces is separable — but only when the index set is not too large and the product carries the product topology. [quotetheorem:946] The countable dense subset is constructed explicitly. If $D_k \subset X_k$ is a countable dense subset, consider the set of "eventually constant" sequences: those $x \in \prod_{k=1}^\infty X_k$ for which $x(k) \in D_k$ for all $k$ and $x(k) = d_k$ (a fixed element of $D_k$) for all but finitely many $k$. This set is countable (a countable union of finite products of countable sets) and dense (every basic open set, which restricts only finitely many coordinates, intersects it). [example: Failure of Separability for Uncountable Products] The [separability](/page/Separable) result requires a countable index set. Consider the uncountable product $\{0, 1\}^{\mathbb{R}}$, where $\{0, 1\}$ has the discrete topology. By Tychonoff's theorem, this space is compact. However, it is **not** separable in the product topology when $|\mathbb{R}|$ is uncountable. To see this, for each $r \in \mathbb{R}$, define $e_r \in \{0, 1\}^{\mathbb{R}}$ by $e_r(s) = 1$ if $s = r$ and $e_r(s) = 0$ if $s \neq r$. Consider the open set $U_r = \pi_r^{-1}(\{1\})$. Then $e_r \in U_r$ but $e_s \notin U_r$ for $s \neq r$. Any dense subset must contain at least one point in each $U_r$ that separates it from $U_s$ for $s \neq r$. A careful argument using the $\Delta$-system lemma shows that no countable set can be dense. More precisely, $\{0, 1\}^{\mathbb{R}}$ has the **countable chain condition** (every family of pairwise disjoint open sets is countable) but is not separable — illustrating that these properties, which are equivalent in metrizable spaces, diverge in general topological spaces. [/example] ## Hausdorff and Regularity Properties Separation axioms transfer cleanly to products. This is a consequence of the product topology being defined by the projections, which "import" the separation properties of each factor. [quotetheorem:958] The Hausdorff property was already verified in the discussion after projections: two distinct points differ in some coordinate, and separation in that coordinate lifts to the product. Regularity and complete regularity follow by similar coordinate-based arguments. The notable omission is **normality** (T4). The product of two normal spaces need not be normal. The **Sorgenfrey plane** — the product $\mathbb{R}_\ell \times \mathbb{R}_\ell$ of the Sorgenfrey line (the real line with the lower limit topology) with itself — is the classical example: $\mathbb{R}_\ell$ is normal, Lindelof, and perfectly normal, but $\mathbb{R}_\ell \times \mathbb{R}_\ell$ is not normal. This failure is one of the reasons why complete regularity, rather than normality, is the preferred separation axiom for infinite-dimensional topology and functional analysis. ## Metrizability of Countable Products When can a product space be metrized — that is, when does its topology arise from a metric? For finite products of metrizable spaces, [metrizability](/page/Metrizable%20Space) is immediate: if $d_k$ is a metric on $X_k$, then $d(x, y) = \max_{1 \le k \le n} d_k(x_k, y_k)$ (or the Euclidean combination $(\sum d_k^2)^{1/2}$) metrizes the product topology on $X_1 \times \cdots \times X_n$. For infinite products, the situation is more delicate. The "max" construction breaks down because the [supremum](/page/Supremum%20and%20Infimum) of infinitely many distances may be infinite. However, if the index set is countable and we truncate the metrics, we can build a single metric that captures coordinatewise convergence. [quotetheorem:959] Several features of this metric deserve comment. **Why the truncation.** The function $t \mapsto t/(1+t)$ maps $[0, \infty)$ bijectively onto $[0, 1)$, so each summand is at most $2^{-k}$. The series converges absolutely and uniformly, ensuring $D$ is a well-defined finite-valued function. Without truncation, $\sum 2^{-k} d_k(x(k), y(k))$ could diverge if the metrics $d_k$ are unbounded. **Convergence characterisation.** A sequence $x_m \to x$ in the metric $D$ if and only if $\bar{d}_k(x_m(k), x(k)) \to 0$ for each $k$, which holds if and only if $d_k(x_m(k), x(k)) \to 0$ for each $k$. This is precisely coordinatewise convergence, matching the product topology. **The role of countability.** The metric $D$ relies on assigning a weight $2^{-k}$ to the $k$-th factor, which requires the factors to be enumerated by $\mathbb{N}$. For uncountable products of nontrivial spaces, the product topology is generally not metrizable. Indeed, if each $X_\alpha$ contains at least two points and $A$ is uncountable, then $\prod_{\alpha \in A} X_\alpha$ is not first-countable (no point has a countable neighbourhood basis), hence not metrizable. [example: The Space $\mathbb{R}^{\mathbb{N}}$ as a Metric Space] The space $\mathbb{R}^{\mathbb{N}}$ of all real sequences, equipped with the product topology, is metrized by \begin{align*} D(x, y) = \sum_{k=1}^\infty 2^{-k} \cdot \frac{|x(k) - y(k)|}{1 + |x(k) - y(k)|}. \end{align*} This metric space is complete (since $\mathbb{R}$ is complete), separable (since $\mathbb{R}$ is separable), and not locally compact (the closed ball $\overline{B}_D(0, r)$ is not compact for any $r > 0$). It is homeomorphic to the irrational numbers — a classical result of Alexandroff and Urysohn. In functional analysis, this space serves as the model for **[Frechet spaces](/page/Fr%C3%A9chet%20Space)** whose topology is defined by countably many seminorms. Completeness follows from the general fact that a countable product of complete metric spaces is complete in the product metric $D$: if $(x_m)$ is Cauchy in $D$, then each coordinate sequence $(x_m(k))$ is Cauchy in $\bar{d}_k$ (hence in $d_k$), and the coordinatewise limit $x(k) = \lim_m x_m(k)$ defines a point $x$ with $D(x_m, x) \to 0$. [/example] ## Function Spaces and Initial Topologies The product topology is a special case of a more general construction — the **initial topology** — and this perspective reveals the product topology's ubiquity across mathematics. ### Initial Topologies Given a set $X$ and a family of maps $f_\alpha: X \to Y_\alpha$ into topological spaces $Y_\alpha$, the **initial topology** on $X$ with respect to $\{f_\alpha\}$ is the coarsest topology making every $f_\alpha$ continuous. It is generated by the subbasis $\{f_\alpha^{-1}(U) : U \in \tau_{Y_\alpha}\}$. The product topology on $\prod_{\alpha \in A} X_\alpha$ is exactly the initial topology with respect to the projections $\{\pi_\alpha\}$. This identification is not merely a restatement — it reveals that many familiar topologies in analysis are "product topologies in disguise." ### Pointwise Convergence as the Product Topology Let $X$ be a set and $Y$ a topological space. The set $Y^X = \{f: X \to Y\}$ of all functions from $X$ to $Y$ is, as a set, the Cartesian product $\prod_{x \in X} Y$ (each function $f$ assigns to each $x \in X$ an element $f(x) \in Y$). Under this identification, the product topology on $Y^X$ is the **topology of pointwise convergence**: a net $f_\lambda \to f$ if and only if $f_\lambda(x) \to f(x)$ for every $x \in X$. This is the natural starting point for function-space topologies, but it is often too weak for analytical purposes. [example: Pointwise vs. Uniform Convergence on $C([0, 1])$] Consider the function space $C([0,1]; \mathbb{R})$ — the set of [continuous functions](/page/Continuity%20(Metric%20Spaces)) $f: [0,1] \to \mathbb{R}$. This is a subset of $\mathbb{R}^{[0,1]} = \prod_{x \in [0,1]} \mathbb{R}$. The product topology on $\mathbb{R}^{[0,1]}$ gives pointwise convergence. However, the subspace $C([0,1]; \mathbb{R})$ is **not closed** in this topology: a pointwise limit of continuous functions need not be continuous. The sequence $f_k(x) = x^k$ on $[0,1]$ converges pointwise to the discontinuous function $f(x) = 0$ for $x \in [0,1)$ and $f(1) = 1$. The topology of **uniform convergence** — induced by the supremum norm $\|f - g\|_\infty = \sup_{x \in [0,1]} |f(x) - g(x)|$ — is strictly finer than the pointwise topology and makes $C([0,1]; \mathbb{R})$ a complete metric space (hence closed in itself). Under uniform convergence, limits of continuous functions are continuous. This illustrates the general principle: the product topology (pointwise convergence) is the coarsest useful topology on a function space, but it typically lacks the completeness and convergence properties that analysis demands. Stronger topologies — uniform convergence, Sobolev norms, $L^p$ norms — are introduced to restore these properties, at the cost of having fewer compact sets. [/example] ### Weak Topologies as Initial Topologies In functional analysis, the [weak topology](/page/Topology) $\sigma(X, X^*)$ on a Banach space $X$ is the initial topology with respect to the evaluation maps $\{x \mapsto f(x) : f \in X^*\}$. The [weak* topology](/page/Weak*%20Topology) $\sigma(X^*, X)$ on the dual $X^*$ is the initial topology with respect to $\{\hat{x}: X^* \to \mathbb{R}\}_{x \in X}$, where $\hat{x}(f) = f(x)$. In both cases, the initial topology construction embeds the space into a product of copies of $\mathbb{R}$ (one for each functional or each point), and the topology is the restriction of the product topology. The Banach-Alaoglu theorem, which asserts weak* compactness of the dual ball, is therefore a direct application of Tychonoff's theorem: the dual ball embeds as a closed subset of a product of compact intervals. This connection between [weak topologies](/page/Weak%20Topology) and products also explains why the weak topology on an infinite-dimensional Banach space is not metrizable on the full space (the "index set" $X^*$ is uncountable), while the weak* topology on the dual ball $B_{X^*}$ of a **separable** Banach space *is* metrizable (because countably many functionals suffice to generate the topology on bounded sets). ## Working with Product Spaces This section collects the standard techniques and constructions that arise repeatedly when working with the product topology. ### The Tube Lemma The Tube Lemma is the key technical tool for finite products of compact spaces. It compensates for the failure of projections to be closed maps by showing that, around a compact "slice," an open set can be fattened into a "tube." [quotetheorem:960] The Tube Lemma is the essential ingredient in the proof that a finite product of compact spaces is compact (which then provides the base case for Tychonoff's theorem by induction). It fails without the compactness hypothesis on $Y$. [example: Failure of the Tube Lemma Without Compactness] Take $X = Y = \mathbb{R}$ and consider the open set \begin{align*} W = \{(x, y) \in \mathbb{R}^2 : |x| < 1/(1 + |y|)\}. \end{align*} One verifies that $W$ is open: it is the preimage of $(-\infty, 0)$ under the [continuous function](/page/Continuity%20(Metric%20Spaces)) $(x, y) \mapsto |x| - 1/(1 + |y|)$. The slice $\{0\} \times \mathbb{R} \subset W$ since $|0| = 0 < 1/(1 + |y|)$ for all $y$. However, no tube $(-\varepsilon, \varepsilon) \times \mathbb{R}$ is contained in $W$: for any $\varepsilon > 0$, taking $y$ large enough so that $1/(1 + |y|) < \varepsilon$, the point $(\varepsilon/2, y)$ lies in the tube but not in $W$ (since $\varepsilon/2 > 1/(1 + |y|)$ for $|y|$ sufficiently large). The failure occurs because $Y = \mathbb{R}$ is not compact — the "width" of $W$ around the slice shrinks to zero as $|y| \to \infty$. [/example] ### Constructing Maps into Products The universal property provides the standard method for constructing continuous maps into a product space: specify each coordinate function and verify that each is continuous. Given topological spaces $Z$ and $\{X_\alpha\}_{\alpha \in A}$, and continuous maps $g_\alpha: Z \to X_\alpha$, there is a unique continuous map \begin{align*} g: Z &\to \prod_{\alpha \in A} X_\alpha \\ z &\mapsto (g_\alpha(z))_{\alpha \in A} \end{align*} satisfying $\pi_\alpha \circ g = g_\alpha$. This map is called the **product map** or **diagonal map** induced by the family $\{g_\alpha\}$. In practice, this is how one proves continuity of maps into $\mathbb{R}^n$: a function $f: X \to \mathbb{R}^n$ is continuous if and only if each component function $f_k: X \to \mathbb{R}$ is continuous. It is also how one constructs embeddings into products — for instance, the embedding of a compact Hausdorff space $X$ into $[0, 1]^{C(X, [0, 1])}$ used in the Stone-Cech compactification. ### Restricting to Subproducts If $B \subset A$ is a subset of the index set, the **sub-projection** $\pi_B: \prod_{\alpha \in A} X_\alpha \to \prod_{\alpha \in B} X_\alpha$ is continuous and open. The product $\prod_{\alpha \in A} X_\alpha$ can be decomposed as \begin{align*} \prod_{\alpha \in A} X_\alpha \cong \prod_{\alpha \in B} X_\alpha \times \prod_{\alpha \in A \setminus B} X_\alpha, \end{align*} and this homeomorphism is the standard tool for reducing arguments about arbitrary products to arguments about two factors. ### Closed Sets in Products While the product of open sets $\prod_{\alpha \in A} U_\alpha$ (with $U_\alpha = X_\alpha$ for all but finitely many $\alpha$) forms a basic open set, the product of closed sets $\prod_{\alpha \in A} C_\alpha$ is always closed in the product topology — even without the "finitely many" restriction. This follows because $\prod C_\alpha = \bigcap_{\alpha \in A} \pi_\alpha^{-1}(C_\alpha)$, and each $\pi_\alpha^{-1}(C_\alpha)$ is closed (preimage of a closed set under a continuous map), so the intersection is closed. This asymmetry — open products require the finite restriction, closed products do not — is a recurring source of confusion and is worth internalizing. ## References 1. Munkres, J. R., *[Topology](/page/Topology)* (2000). 2. Willard, S., *General Topology* (1970). 3. Kelley, J. L., *General Topology* (1955). 4. Engelking, R., *General Topology* (1989). 5. Bourbaki, N., *General Topology, Chapters 1-4* (1995).