How "thin" can a subset of a [topological space](/page/Topology) be? The question has two distinct answers depending on whether one measures size by **measure** or by **topology**. A set of [Lebesgue measure](/page/Lebesgue%20Measure) zero can still be [dense](/page/Dense%20Subset) — the [rationals](/page/Rational%20Number) $\mathbb{Q}$ have measure zero in $\mathbb{R}$ yet meet every open interval. A measure-zero set can be topologically large. The notion of **nowhere dense set** captures the opposite extreme: a set that is topologically negligible, so sparse that its [closure](/page/Closure) fails to contain even the smallest open region. Where a dense set intrudes into every open set, a nowhere dense set can be completely avoided within every open set. The distinction between "measure-theoretically thin" and "topologically thin" is not academic. It governs two of the most powerful impossibility arguments in analysis. The [Baire Category Theorem](/page/Baire%20Category%20Theorem) shows that a [complete metric space](/page/Complete%20Metric%20Space) cannot be expressed as a countable union of nowhere dense sets — a topological statement that has no analogue in measure theory (where a full-measure set can be a countable union of measure-zero sets). Conversely, the [Cantor set](/page/Cantor%20Set) is nowhere dense and has measure zero, but a "fat" Cantor set is nowhere dense and has positive measure. The two notions of smallness are genuinely independent. [example: A Set That Cannot Be Avoided] Consider $\mathbb{Q} \subset \mathbb{R}$ with the standard topology. The rationals are dense: every open interval $(a, b)$ with $a < b$ contains a rational number (by the [Archimedean property](/page/Archimedean%20Property)). Is $\mathbb{Q}$ topologically thin? The closure is $\overline{\mathbb{Q}} = \mathbb{R}$, and $\operatorname{int}(\overline{\mathbb{Q}}) = \operatorname{int}(\mathbb{R}) = \mathbb{R} \neq \varnothing$. So $\mathbb{Q}$ is **not** nowhere dense — its closure has nonempty interior (in fact, the interior of its closure is all of $\mathbb{R}$). Despite being countable and having measure zero, $\mathbb{Q}$ is topologically "thick" enough that its closure fills the entire line. Now consider $\mathbb{Z} \subset \mathbb{R}$. The closure is $\overline{\mathbb{Z}} = \mathbb{Z}$ (the integers are already [closed](/page/Closed%20Set) — each integer $n$ is isolated by the open set $(n - 1/2, n + 1/2) \cap \mathbb{Z} = \{n\}$). The interior of $\mathbb{Z}$ is empty: no open interval $(a,b)$ with $a < b$ is contained in $\mathbb{Z}$, because any such interval contains non-integer real numbers. So $\operatorname{int}(\overline{\mathbb{Z}}) = \varnothing$, and $\mathbb{Z}$ is nowhere dense. The contrast is instructive. Both $\mathbb{Q}$ and $\mathbb{Z}$ are countable, both have Lebesgue measure zero, but $\mathbb{Q}$ is dense while $\mathbb{Z}$ is nowhere dense. The difference lies entirely in how they interact with open sets: the rationals infiltrate every interval, while the integers leave gaps that open intervals can exploit. [/example] ## Definition The definition of a nowhere dense set involves two steps — first take the closure (to account for limit points), then check whether the result contains any open set. Taking the closure is essential: without it, the notion would conflate topological thinness with the failure to be closed. The set $\{1/n : n \in \mathbb{N}\}$ has empty interior in $\mathbb{R}$, but so does $\mathbb{Q}$, and the two sets have radically different topological behaviour. Taking the closure before checking the interior separates these cases: the closure of $\{1/n\}$ is $\{0\} \cup \{1/n : n \in \mathbb{N}\}$, which still has empty interior, while the closure of $\mathbb{Q}$ is all of $\mathbb{R}$. [definition: Nowhere Dense Set] Let $(X, \tau)$ be a [topological space](/page/Topology) and let $A \subset X$. The set $A$ is **nowhere dense** in $X$ if the [interior](/page/Interior) of its [closure](/page/Closure) is empty: \begin{align*} \operatorname{int}(\overline{A}) = \varnothing. \end{align*} [/definition] A set $A$ is nowhere dense precisely when $\overline{A}$ contains no nonempty open set. Since the interior of a set is the largest open set it contains, $\operatorname{int}(\overline{A}) = \varnothing$ means that $\overline{A}$ — even after adding all limit points of $A$ — remains too sparse to swallow any open region. An immediate consequence of the definition is that every subset of a nowhere dense set is nowhere dense. If $B \subset A$ and $A$ is nowhere dense, then $\overline{B} \subset \overline{A}$, so $\operatorname{int}(\overline{B}) \subset \operatorname{int}(\overline{A}) = \varnothing$. ## Equivalent Characterisations The definition involves a two-step operation — closure, then interior — which makes direct verification cumbersome. In practice, one rarely computes $\operatorname{int}(\overline{A})$ explicitly. Instead, one uses equivalent characterisations that are more amenable to the arguments at hand. The most useful characterisations rephrase "nowhere dense" as a condition on the complement, a condition on open subsets, or a density condition. [quotetheorem:1084] Each characterisation captures a different aspect of topological thinness. Characterisation (ii) gives a global perspective: $A$ is nowhere dense if and only if after removing $A$ and all its limit points from $X$, what remains is still dense. The "holes" that $\overline{A}$ punches in $X$ do not isolate any open region from the complement. This is the characterisation most useful in arguments involving the [Baire Category Theorem](/page/Baire%20Category%20Theorem), where one works with complements of closed nowhere dense sets, which are precisely the dense open sets. Characterisation (iii) is the workhorse for direct verification. It says that $A$ can be "locally avoided": within any open region, no matter how small, there is a sub-region that misses $A$ entirely. This is the formal negation of density — a dense set meets every nonempty open set, while a nowhere dense set can be dodged within every nonempty open set. The equivalence of (iii) and (iv) reveals that the closure step in the definition is automatically enforced by the "open subset" quantifier. To see why (iii) implies (iv), suppose $H$ is a nonempty open set with $H \cap A = \varnothing$. We claim $H \cap \overline{A} = \varnothing$ as well. If some point $p$ belonged to both $H$ and $\overline{A}$, then $H$ would be an open neighbourhood of $p$ disjoint from $A$ — but the [neighbourhood characterisation of closure](/page/Closure) requires every open neighbourhood of a point in $\overline{A}$ to meet $A$. This contradiction shows $H \cap \overline{A} = \varnothing$. In short: an open set that avoids $A$ automatically avoids $\overline{A}$. [remark: Closed Nowhere Dense Sets] When $A$ is itself closed, the definition simplifies: $A$ is nowhere dense if and only if $\operatorname{int}(A) = \varnothing$, because $\overline{A} = A$. For closed sets, "nowhere dense" and "having empty interior" are the same condition. For non-closed sets, these conditions differ: the set $(0,1) \cap \mathbb{Q}$ has empty interior in $\mathbb{R}$, but it is not nowhere dense because $\overline{(0,1) \cap \mathbb{Q}} = [0,1]$ and $\operatorname{int}([0,1]) = (0,1) \neq \varnothing$. [/remark] ## Nowhere Dense Sets in $\mathbb{R}$ The real line provides a rich collection of examples that reveal the boundaries of the concept — what qualifies as nowhere dense, what does not, and how nowhere denseness interacts with other properties like cardinality, measure, and closure. ### Discrete and Isolated Sets The simplest nowhere dense sets are those whose points are "spread apart." If every point of $A$ is isolated — surrounded by a neighbourhood containing no other point of $A$ — then $A$ is closed (its complement is open) and has empty interior, hence is nowhere dense. The integers $\mathbb{Z}$, the set $\{1/n : n \in \mathbb{N}\}$, and any finite subset of $\mathbb{R}$ fall into this category. [example: The Sequence $\{1/n\}$ is Nowhere Dense] Let $A = \{1/n : n \in \mathbb{N}\} \subset \mathbb{R}$. The closure is $\overline{A} = \{0\} \cup \{1/n : n \in \mathbb{N}\}$: the only limit point is $0$, since for any $x \notin A \cup \{0\}$, one can find $\varepsilon > 0$ such that $(x - \varepsilon, x + \varepsilon)$ contains no element of $A$. To show $\operatorname{int}(\overline{A}) = \varnothing$, let $(a,b)$ be any open interval with $a < b$. If $(a,b) \subset \overline{A}$, then every point in $(a,b)$ belongs to $\{0\} \cup \{1/n : n \in \mathbb{N}\}$. But $(a,b)$ is uncountable (it contains the interval and thus uncountably many real numbers), while $\overline{A}$ is countable. This is a contradiction, so no open interval is contained in $\overline{A}$, and $\operatorname{int}(\overline{A}) = \varnothing$. [/example] ### The Cantor Set: Uncountable yet Nowhere Dense A natural question is whether nowhere dense sets must be "small" in the sense of cardinality. The integers and the sequence $\{1/n\}$ are both countable. Can a nowhere dense set be uncountable? The [Cantor set](/page/Cantor%20Set) provides an emphatic affirmative. [example: The Cantor Set is Nowhere Dense] The standard Cantor set $C \subset [0,1]$ is constructed iteratively. Begin with $C_0 = [0,1]$. At each stage $k$, remove the open middle third of every closed interval in $C_{k-1}$ to obtain $C_k$. The Cantor set is the intersection: \begin{align*} C := \bigcap_{k=0}^\infty C_k. \end{align*} **$C$ is closed:** Each $C_k$ is a finite union of closed intervals, hence closed. The intersection of closed sets is closed, so $C$ is closed. Therefore $\overline{C} = C$. **$C$ has empty interior:** Suppose $(a, b) \subset C$ for some $a < b$. At stage $k$ of the construction, each remaining interval has length $3^{-k}$. Choose $k$ large enough that $3^{-k} < b - a$. Then $(a, b)$ is longer than any single interval in $C_k$, so $(a,b)$ must intersect one of the removed open intervals (which partition the complement of $C_k$ within $[0,1]$). But removed points do not belong to $C$, contradicting $(a, b) \subset C$. Hence $\operatorname{int}(C) = \varnothing$. **$C$ is uncountable:** The Cantor set consists of all numbers in $[0,1]$ whose ternary expansion uses only the digits $0$ and $2$. The map sending such a number $\sum_{k=1}^\infty a_k 3^{-k}$ (with each $a_k \in \{0, 2\}$) to $\sum_{k=1}^\infty (a_k/2) \cdot 2^{-k}$ is a surjection from $C$ onto $[0,1]$, because every number in $[0,1]$ has a binary expansion. Hence $|C| \ge |[0,1]| = |\mathbb{R}|$. So the Cantor set is a closed, uncountable, nowhere dense subset of $\mathbb{R}$ with Lebesgue measure $\mathcal{L}^1(C) = 0$. It demonstrates that nowhere dense sets can be as large as the continuum. [/example] ### Nowhere Dense Sets of Positive Measure The Cantor set has measure zero, which might suggest that nowhere dense sets must be measure-theoretically negligible as well. This suggestion is false — and its failure is one of the most important lessons about the independence of topology and measure. [example: A Fat Cantor Set] Fix a parameter $0 < \alpha < 1$. Modify the Cantor set construction as follows: at stage $k$, remove from each of the $2^{k-1}$ remaining intervals a centred open interval of length $\alpha \cdot 3^{-k}$ (rather than $3^{-k}$). Let $C_\alpha$ denote the resulting compact set. The total length removed is: \begin{align*} \sum_{k=1}^\infty 2^{k-1} \cdot \frac{\alpha}{3^k} = \frac{\alpha}{3} \sum_{k=0}^\infty \left(\frac{2}{3}\right)^k = \frac{\alpha}{3} \cdot 3 = \alpha. \end{align*} Therefore $\mathcal{L}^1(C_\alpha) = 1 - \alpha > 0$. **$C_\alpha$ is nowhere dense:** The argument is the same as for the standard Cantor set. At each stage, we remove an open interval from the middle of every remaining closed interval. The remaining closed intervals at stage $k$ each have length at most $(1 - \alpha/3)^k / 2^{k-1}$, which tends to $0$ as $k \to \infty$. For any open interval $(a,b)$ with $a < b$, choosing $k$ sufficiently large ensures that $(a,b)$ is wider than any remaining interval, so $(a,b)$ must overlap a removed portion and cannot lie entirely in $C_\alpha$. Since $C_\alpha$ is closed, $\operatorname{int}(C_\alpha) = \operatorname{int}(\overline{C_\alpha}) = \varnothing$. By choosing $\alpha$ close to $0$, we obtain nowhere dense sets of measure arbitrarily close to $1$. The fat Cantor set $C_{1/2}$ has measure $1/2$ — half the interval $[0,1]$ is occupied by a nowhere dense set. This shows that topological thinness (nowhere dense) and measure-theoretic thinness (small measure) are independent properties. [/example] ## Meagre Sets and the Baire Category Framework A single nowhere dense set is thin, but how thin is a countable union of nowhere dense sets? The integers $\mathbb{Z}$ are nowhere dense, and so is each translate $\mathbb{Z} + r$ for $r \in \mathbb{R}$. Yet $\bigcup_{q \in \mathbb{Q}} (\mathbb{Z} + q)$ contains all rationals, which are dense — so a countable union of nowhere dense sets can be dense. The question becomes: *can a countable union of nowhere dense sets fill up the entire space?* This question is the starting point for the theory of Baire category, and the answer — that it cannot, in complete metric spaces — is one of the deepest results in general topology. [definition: Meagre Set] Let $(X, \tau)$ be a topological space. A subset $M \subset X$ is **meagre** (or of **first category**) if it can be expressed as a countable union of nowhere dense sets: \begin{align*} M = \bigcup_{k=1}^\infty A_k, \quad \text{where each } A_k \text{ is nowhere dense in } X. \end{align*} A set that is not meagre is said to be of **second category**. A set whose complement is meagre is called **co-meagre** (or **residual**). [/definition] The terminology "first category" and "second category" is due to Baire (1899), and reflects a classification of sets by topological size. Meagre sets are the topologically negligible sets — the analogue of measure-zero sets in the measure-theoretic world. Co-meagre sets are the topologically large sets — the analogue of sets of full measure. The analogy between meagre/co-meagre and measure-zero/full-measure is suggestive but imperfect. The rationals $\mathbb{Q}$ are meagre in $\mathbb{R}$ (as a countable union of singletons, each of which is nowhere dense), yet $\mathbb{Q}$ has measure zero — so both notions agree on the smallness of $\mathbb{Q}$. But the fat Cantor set $C_{1/2}$ is nowhere dense (hence meagre) yet has positive measure. Conversely, the complement $[0,1] \setminus C_{1/2}$ is open and dense (hence co-meagre in $[0,1]$) but has measure only $1/2$. A set can be topologically large and measure-theoretically small, or vice versa. [example: The Rationals Are Meagre but Not Nowhere Dense] The set $\mathbb{Q}$ is not nowhere dense in $\mathbb{R}$: as shown in the opening example, $\overline{\mathbb{Q}} = \mathbb{R}$ and $\operatorname{int}(\mathbb{R}) = \mathbb{R} \neq \varnothing$. However, $\mathbb{Q}$ is meagre. Write $\mathbb{Q} = \{q_1, q_2, q_3, \ldots\}$ (the rationals are countable). Each singleton $\{q_k\}$ is nowhere dense: $\overline{\{q_k\}} = \{q_k\}$ and $\operatorname{int}(\{q_k\}) = \varnothing$ (no open interval is contained in a single point). Therefore: \begin{align*} \mathbb{Q} = \bigcup_{k=1}^\infty \{q_k\} \end{align*} is a countable union of nowhere dense sets, hence meagre. This illustrates a crucial distinction: a set can be dense and meagre simultaneously. Being meagre means the set is a countable union of nowhere dense sets, but the union itself may be far from nowhere dense. The rationals are the prototypical example of a "large set built from small pieces" — each piece is topologically negligible, but together they cover a dense subset. [/example] ### The Baire Category Theorem The Baire Category Theorem is the fundamental result connecting nowhere denseness and completeness. It asserts that in sufficiently nice spaces — complete metric spaces and locally compact Hausdorff spaces — the meagre sets genuinely are "small": they cannot cover the entire space. [quotetheorem:630] The two forms are equivalent via complementation. If $\{G_k\}$ are dense open sets and $\bigcap G_k$ fails to be dense, then the closed sets $F_k := X \setminus G_k$ (each nowhere dense, since $F_k$ is closed with empty interior) cover a nonempty open set, contradicting the category form. Conversely, if $X = \bigcup F_k$ with each $F_k$ nowhere dense, then $G_k := X \setminus \overline{F_k}$ is a dense open set (by characterisation (ii) of nowhere dense sets), and $\bigcap G_k \subset X \setminus \bigcup F_k = \varnothing$, contradicting the dense $G_\delta$ form. The completeness hypothesis is essential — the theorem fails for incomplete metric spaces. [example: The Baire Category Theorem Fails in $\mathbb{Q}$] The rational numbers $\mathbb{Q}$ with the metric $d(x,y) = |x - y|$ form an incomplete metric space. We have already shown that $\mathbb{Q} = \bigcup_{k=1}^\infty \{q_k\}$ is a countable union of nowhere dense sets, so $\mathbb{Q}$ is meagre in itself. In a complete metric space, this is impossible. More concretely, the dense $G_\delta$ form fails. For each $q \in \mathbb{Q}$, the set $G_q := \mathbb{Q} \setminus \{q\}$ is open and dense in $\mathbb{Q}$ (removing one point from a dense set leaves a dense set, and the complement of a point in $\mathbb{Q}$ is open since $\{q\}$ is closed). But: \begin{align*} \bigcap_{q \in \mathbb{Q}} G_q = \bigcap_{q \in \mathbb{Q}} (\mathbb{Q} \setminus \{q\}) = \varnothing. \end{align*} A countable intersection of dense open subsets of $\mathbb{Q}$ is empty — violating the conclusion of the Baire Category Theorem. The completeness of $\mathbb{R}$ is what prevents this collapse: in $\mathbb{R}$, the sets $\mathbb{R} \setminus \{q\}$ are dense and open, and their countable intersection is $\mathbb{R} \setminus \mathbb{Q}$, which is dense (the irrationals are dense in $\mathbb{R}$). [/example] The theorem also holds for [locally compact](/page/Locally%20Compact%20Space) [Hausdorff spaces](/page/Hausdorff%20Space), which need not be metrizable. The key property shared by complete metric spaces and locally compact Hausdorff spaces is that both have a "nested sequence" argument available: one can construct a decreasing sequence of closed sets with nonempty intersection (via Cantor's intersection theorem in the complete metric case, via compactness in the locally compact case). ## Co-Meagre Sets and Generic Properties The Baire Category Theorem acquires its power through the notion of a **generic property**: a property that holds for all elements of a co-meagre set. Since co-meagre sets are large (their complements are meagre, hence topologically negligible), a generic property holds for "most" elements in the topological sense. The term "generic" is used loosely in mathematics, but in the Baire category framework it has a precise meaning: a property $P$ is generic in a complete metric space $X$ if the set $\{x \in X : P(x) \text{ holds}\}$ is co-meagre, equivalently, if its complement is meagre. Since a co-meagre set contains a dense $G_\delta$ set (by the Baire Category Theorem), a generic property holds on a dense subset — in fact, on a "thick" dense subset that is stable under countable intersection. [definition: Co-Meagre Set] Let $(X, \tau)$ be a topological space. A subset $R \subset X$ is **co-meagre** (or **residual**) if its complement $X \setminus R$ is meagre. Equivalently, $R$ is co-meagre if it contains a countable intersection of dense open sets (a dense $G_\delta$ set). [/definition] Co-meagre sets are closed under countable intersection: if $R_1, R_2, R_3, \ldots$ are co-meagre, then each $X \setminus R_k$ is meagre, and $X \setminus \bigcap_{k=1}^\infty R_k = \bigcup_{k=1}^\infty (X \setminus R_k)$ is a countable union of meagre sets, hence meagre (a countable union of countable unions of nowhere dense sets is again a countable union of nowhere dense sets). Therefore $\bigcap R_k$ is co-meagre. This stability under countable intersection is what makes the notion of "generic" robust: if countably many properties are each generic, then all of them hold simultaneously on a co-meagre set. The most celebrated application of generic properties is the following result, which asserts that the "typical" continuous function is far more pathological than the smooth functions one encounters in calculus. [example: Generically, Continuous Functions Are Nowhere Differentiable] Consider the [Banach space](/page/Banach%20Space) $C([0,1])$ equipped with the supremum norm $\|f\|_\infty = \sup_{x \in [0,1]} |f(x)|$. This is a complete metric space. The set of functions in $C([0,1])$ that are differentiable at **at least one point** turns out to be meagre. For each $M \in \mathbb{N}$ and each $x_0 \in [0,1]$, define: \begin{align*} E_{M, x_0} := \left\{ f \in C([0,1]) : \exists \delta > 0 \text{ such that } |h| < \delta \implies \left|\frac{f(x_0 + h) - f(x_0)}{h}\right| \le M \right\}. \end{align*} This is the set of continuous functions whose difference quotient at $x_0$ is eventually bounded by $M$ — a necessary condition for differentiability at $x_0$ with derivative of absolute value at most $M$. Each $E_{M, x_0}$ is closed in $C([0,1])$ (the boundedness condition on difference quotients is preserved under uniform limits) and has empty interior. To see the latter, take any $f \in C([0,1])$ and $\varepsilon > 0$. One can construct $g \in C([0,1])$ with $\|f - g\|_\infty < \varepsilon$ such that the difference quotients of $g$ at $x_0$ exceed $M$: add to $f$ a small sawtooth function of amplitude less than $\varepsilon$ and slope exceeding $M + \|f\|_\infty$. Then $g \notin E_{M, x_0}$, so $E_{M, x_0}$ contains no open ball. Hence $E_{M, x_0}$ is nowhere dense. The set of functions differentiable at $x_0$ is contained in $\bigcup_{M=1}^\infty E_{M, x_0}$, a countable union of nowhere dense sets, hence meagre. Taking a further countable union over a dense set of points $x_0$ (say, $x_0 \in \mathbb{Q} \cap [0,1]$), the set of functions differentiable at **any rational point** is still meagre (a countable union of meagre sets is meagre). A more careful argument (using a countable cover that accounts for all points) shows that the set of functions differentiable at even a single point is meagre. By the Baire Category Theorem, the co-meagre complement — functions that are **nowhere differentiable** — is dense in $C([0,1])$. In the sense of Baire category, the "generic" continuous function is nowhere differentiable. This does not contradict the density of smooth (hence everywhere differentiable) functions in $C([0,1])$: both the smooth functions and the nowhere differentiable functions are dense, but the smooth functions form a meagre set while the nowhere differentiable functions form a co-meagre set. [/example] ## Stability Properties of Nowhere Dense Sets Working with nowhere dense sets requires understanding which operations preserve nowhere denseness and which destroy it. These stability properties determine the scope of Baire category arguments and reveal the algebraic structure of the ideal of meagre sets. ### Operations That Preserve Nowhere Denseness [quotetheorem:1085] Property (i) is immediate from the monotonicity of closure: if $B \subset A$ and $\operatorname{int}(\overline{A}) = \varnothing$, then $\overline{B} \subset \overline{A}$, so $\operatorname{int}(\overline{B}) \subset \operatorname{int}(\overline{A}) = \varnothing$. Property (ii) follows from the fact that interior distributes over finite unions in one direction: $\operatorname{int}(\overline{A_1} \cup \overline{A_2}) \subset \operatorname{int}(\overline{A_1 \cup A_2})$. Since $\overline{A_1 \cup A_2} = \overline{A_1} \cup \overline{A_2}$, we need $\operatorname{int}(\overline{A_1} \cup \overline{A_2}) = \varnothing$. Suppose $U$ is a nonempty open set contained in $\overline{A_1} \cup \overline{A_2}$. Since $A_1$ is nowhere dense, $U$ contains a nonempty open set $V$ with $V \cap \overline{A_1} = \varnothing$ (by characterisation (iv)). Then $V \subset \overline{A_2}$, so $V \subset \operatorname{int}(\overline{A_2})$, contradicting $\operatorname{int}(\overline{A_2}) = \varnothing$. The argument extends by induction to any finite union. Property (iii) is a consequence of the idempotency of closure: $\overline{\overline{A}} = \overline{A}$. If $A$ is nowhere dense, then $\operatorname{int}(\overline{A}) = \varnothing$, and $\overline{\overline{A}} = \overline{A}$, so $\operatorname{int}(\overline{\overline{A}}) = \operatorname{int}(\overline{A}) = \varnothing$, meaning $\overline{A}$ is nowhere dense. ### Countable Unions Can Escape Nowhere Denseness Property (ii) does not extend to countable unions — this is precisely the gap that the theory of meagre sets is designed to fill. A countable union of nowhere dense sets is meagre but need not be nowhere dense. [example: A Countable Union of Nowhere Dense Sets That Is Dense] Each singleton $\{q\}$ for $q \in \mathbb{Q}$ is nowhere dense in $\mathbb{R}$. Yet $\mathbb{Q} = \bigcup_{q \in \mathbb{Q}} \{q\}$ is dense in $\mathbb{R}$: $\overline{\mathbb{Q}} = \mathbb{R}$ and $\operatorname{int}(\overline{\mathbb{Q}}) = \mathbb{R} \neq \varnothing$. The union has far more topological substance than any of its summands. This phenomenon — a countable union of negligible sets becoming non-negligible — is familiar from measure theory (a countable union of measure-zero sets has measure zero), but the analogy breaks: in measure theory, countable unions of null sets are null, while in Baire category, countable unions of nowhere dense sets need not be nowhere dense. The correct category-theoretic analogue of "null" is "meagre," and meagre sets are closed under countable unions by definition. [/example] ## Nowhere Dense Sets and Continuous Functions The concept of nowhere dense sets arises naturally in the study of [continuous functions](/page/Continuity), where the following question appears: if a continuous function vanishes on a dense set, must it vanish everywhere? The answer is yes, and the proof rests on the observation that the zero set of a continuous function is closed. But a more refined question — what can the zero set of a continuous function look like? — leads directly to nowhere dense sets. [quotetheorem:1086] Statement (i) warns against a natural but false intuition. Continuous preimages of nowhere dense sets need not be nowhere dense, because a continuous function can "stretch" a thin set into a thick one. [example: Preimage of a Nowhere Dense Set Under a Continuous Map] Let $f: \mathbb{R} \to \mathbb{R}$, $x \mapsto 0$, be the constant zero function. The singleton $\{0\}$ is nowhere dense in $\mathbb{R}$. But $f^{-1}(\{0\}) = \mathbb{R}$, which is not nowhere dense (in fact, $\operatorname{int}(\overline{\mathbb{R}}) = \mathbb{R}$). For a less degenerate example, consider the [Cantor function](/page/Cantor%20Set) $f: [0,1] \to [0,1]$, which is continuous, nondecreasing, and maps the Cantor set $C$ onto all of $[0,1]$. The set $\{1/2\}$ is nowhere dense in $[0,1]$, but $f^{-1}(\{1/2\}) = [1/3, 2/3]$, which has nonempty interior. [/example] Statement (ii) provides a positive result: open maps preserve nowhere denseness in the forward direction. This is useful in the theory of topological groups and quotient spaces, where quotient maps are open. ## Working with Nowhere Dense Sets: Standard Techniques This section collects the argument patterns that recur in proofs involving nowhere dense sets and Baire category. These techniques form the toolkit for applying the abstract theory to concrete problems. ### Verifying Nowhere Denseness via the Open Subset Criterion The most common method for proving that a set $A$ is nowhere dense uses characterisation (iii): show that every nonempty open set $G$ contains a nonempty open set $H$ with $H \cap A = \varnothing$. In metric spaces, this reduces to: for every open ball $B(x, r)$, find a sub-ball $B(y, s) \subset B(x, r)$ with $B(y, s) \cap A = \varnothing$. [example: A Closed Set With Empty Interior Is Nowhere Dense] Let $F \subset \mathbb{R}$ be a closed set with $\operatorname{int}(F) = \varnothing$. We verify that $F$ is nowhere dense using the open subset criterion. Let $(a,b)$ be any nonempty open interval. Since $\operatorname{int}(F) = \varnothing$, the interval $(a,b)$ is not contained in $F$, so there exists $x_0 \in (a,b) \setminus F$. Since $F$ is closed, $\mathbb{R} \setminus F$ is open, so there exists $\varepsilon > 0$ with $(x_0 - \varepsilon, x_0 + \varepsilon) \subset \mathbb{R} \setminus F$. Setting $\delta = \min(\varepsilon, x_0 - a, b - x_0) > 0$, the interval $H := (x_0 - \delta, x_0 + \delta)$ satisfies $H \subset (a,b)$ and $H \cap F = \varnothing$. This argument shows that for closed sets, the concepts "empty interior" and "nowhere dense" coincide. For non-closed sets, the closure step in the definition is essential. [/example] ### Building Baire Category Arguments The standard structure of a Baire category argument has three steps: **Step 1 (Decompose into nowhere dense sets).** Express the "exceptional set" $E$ — the set where some pathological or undesirable condition holds — as a countable union $E = \bigcup_{k=1}^\infty E_k$. **Step 2 (Verify nowhere denseness).** Show that each $E_k$ is nowhere dense. Typically, one shows $E_k$ is closed (or contained in a closed set) with empty interior. The closure step often uses a compactness or sequential convergence argument; the empty interior step often uses a perturbation or density argument (exhibiting, in every open set, an element not in $E_k$). **Step 3 (Apply Baire).** Conclude by the Baire Category Theorem that $E$ is meagre, hence $X \setminus E$ is co-meagre and in particular dense. The "generic" element of $X$ avoids $E$. [example: The Uniform Boundedness Principle via Baire Category] Let $X$ be a Banach space, let $Y$ be a normed vector space, and let $\{T_\alpha\}_{\alpha \in \mathcal{A}} \subset \mathcal{L}(X, Y)$ be a family of bounded linear operators that is **pointwise bounded**: for every $x \in X$, $\sup_{\alpha \in \mathcal{A}} \|T_\alpha x\|_Y < \infty$. **Claim:** The family is uniformly bounded: $\sup_{\alpha \in \mathcal{A}} \|T_\alpha\|_{\mathcal{L}(X,Y)} < \infty$. **Step 1.** For each $k \in \mathbb{N}$, define: \begin{align*} E_k := \left\{ x \in X : \sup_{\alpha \in \mathcal{A}} \|T_\alpha x\|_Y \le k \right\} = \bigcap_{\alpha \in \mathcal{A}} \{ x \in X : \|T_\alpha x\|_Y \le k \}. \end{align*} Each $E_k$ is closed (an intersection of closed sets — each $\{x : \|T_\alpha x\|_Y \le k\}$ is the preimage of $(-\infty, k]$ under the continuous function $x \mapsto \|T_\alpha x\|_Y$). The pointwise boundedness hypothesis says that for every $x \in X$, there exists $k$ with $\sup_\alpha \|T_\alpha x\| \le k$, i.e., $x \in E_k$. Therefore $X = \bigcup_{k=1}^\infty E_k$. **Step 2.** By the Baire Category Theorem (since $X$ is a Banach space, hence a complete metric space), $X$ is not meagre in itself. Since $X = \bigcup_{k=1}^\infty E_k$ and each $E_k$ is closed, at least one $E_k$ must have nonempty interior (otherwise all $E_k$ would be nowhere dense, and their union would be meagre). Fix $k_0$ such that $\operatorname{int}(E_{k_0}) \neq \varnothing$, and choose $x_0 \in X$ and $r > 0$ with $B(x_0, r) \subset E_{k_0}$. **Step 3.** For any $x \in X$ with $\|x\| \le 1$, the point $x_0 + rx$ belongs to $B(x_0, r) \subset E_{k_0}$, so $\|T_\alpha(x_0 + rx)\| \le k_0$ for all $\alpha$. By linearity: \begin{align*} \|T_\alpha(rx)\| = \|T_\alpha(x_0 + rx) - T_\alpha(x_0)\| \le \|T_\alpha(x_0 + rx)\| + \|T_\alpha(x_0)\| \le k_0 + k_0 = 2k_0. \end{align*} Dividing by $r$: $\|T_\alpha x\| \le 2k_0/r$ for all $x$ with $\|x\| \le 1$ and all $\alpha$. Hence $\sup_\alpha \|T_\alpha\| \le 2k_0/r < \infty$. [/example] ### The Complement Technique To prove that a set $A$ is nowhere dense, it is sometimes easier to work with the complement. By characterisation (ii), $A$ is nowhere dense if and only if $X \setminus \overline{A}$ is dense. When $A$ is closed, this simplifies to: $A$ is nowhere dense if and only if $X \setminus A$ is dense. Showing that an open set is dense can be more natural than showing that a closed set has empty interior — one constructs approximating sequences rather than hunting for intervals. [example: Using the Complement to Verify Nowhere Denseness] Let $A = \{(x, y) \in \mathbb{R}^2 : y = 0\}$ be the $x$-axis in $\mathbb{R}^2$. The set $A$ is closed. To show $A$ is nowhere dense, we verify that $\mathbb{R}^2 \setminus A = \{(x,y) \in \mathbb{R}^2 : y \neq 0\}$ is dense: for any $(x_0, y_0) \in \mathbb{R}^2$ and any $\varepsilon > 0$, the point $(x_0, y_0 + \varepsilon/2)$ belongs to $\mathbb{R}^2 \setminus A$ (since $y_0 + \varepsilon/2 \neq 0$ when $y_0 \neq 0$, and if $y_0 = 0$, then $\varepsilon/2 \neq 0$) and satisfies $|(x_0, y_0 + \varepsilon/2) - (x_0, y_0)| = \varepsilon/2 < \varepsilon$. More generally, any proper affine subspace of $\mathbb{R}^n$ (a translate of a proper linear subspace) is closed and nowhere dense, because its complement is open and dense. [/example] ## References 1. Baire, R., *Sur les fonctions de variables reelles*, Annali di Matematica Pura ed Applicata (1899). 2. Oxtoby, J. C., *Measure and Category* (2nd ed., 1980). 3. Munkres, J. R., *Topology* (2nd ed., 2000). 4. Rudin, W., *Functional Analysis* (2nd ed., 1991). 5. Kelley, J. L., *General Topology* (1955). 6. Kechris, A. S., *Classical Descriptive Set Theory* (1995).