The Subspace Topology

The subspace topology is not a curiosity confined to point-set exercises; it is the invisible scaffolding underneath almost every geometric argument. In real analysis, every theorem about continuous functions on closed intervals $[a,b]$ — the Extreme Value Theorem, the Intermediate Value Theorem — is implicitly a theorem about the subspace topology $[a,b] \subset \mathbb{R}$, since compactness and connectedness of $[a,b]$ are properties of that induced topology, feeding directly into Compact Spaces and Open Covers and Connected Spaces. In differential geometry, a manifold-with-boundary or a submanifold $M \subset \mathbb{R}^n$ (a sphere, a torus, a curve) is topologized exactly by the subspace topology inherited from Euclidean space, before any smooth structure is layered on top; understanding which sets are relatively open on the sphere (spherical caps, for instance) is a direct application of the definition given here. In complex analysis, domains and their boundaries, and Riemann surfaces realized as subsets of $\mathbb{C}^2$ or projective space, again rely on the subspace topology to make sense of open sets and continuity. In algebraic topology and algebraic geometry, the Zariski topology on a variety is frequently studied via its subspace topology inherited from affine or projective space, and the notion of a "closed subvariety" is precisely the closed-set characterization proved above. In computer science and data analysis, subspace topologies underlie the formal treatment of simplicial complexes and finite topological models used in digital topology and persistent homology, where a data set is treated as a subspace of a larger ambient metric space. In cryptography and coding theory, finite topological spaces built as subspaces of product spaces appear in the study of automata and formal language topologies, though there the emphasis is combinatorial rather than geometric.

:::quiz[Which of the following is open in the subspace $A = [1,3]$ of $\mathbb{R}$ with the usual topology?]

• $[1, 2)$ ✓
• $\{1\}$
• $[1,3]$ minus the single point $2$, i.e. $[1,2)\cup(2,3]$
• $(2,3]$

Ask the tutor: Why does the subspace topology on $\mathbb{Q}$ fail to be discrete even though $\mathbb{Q}$ is countable and totally disconnected?

Think of $\tau$ as a flashlight that illuminates certain regions of $X$: exactly the open sets. To topologize a subset $A$, we do not ask which parts of $A$ are illuminated by $X$'s own light in some intrinsic sense; instead we take every possible open region of $X$, shine it onto $A$, and record the shadow $U \cap A$. The collection of all such shadows is declared open in $A$. This immediately explains the key phenomenon: a set can be a "shadow of an open set" without itself being open in the big space. Consider $A = [0,1] \subset \mathbb{R}$. The set $U = (-1, 0.5)$ is open in $\mathbb{R}$, and its shadow on $A$ is

which is therefore open in the subspace $[0,1]$ — even though $[0,0.5)$ itself is neither open nor closed in $\mathbb{R}$. The endpoint $0$ "counts as interior" in $A$ because, from the point of view of someone living inside $[0,1]$, there is nothing to the left of $0$ to worry about. A second key phenomenon: subsets that look topologically trivial in the ambient space can become richly structured once given the subspace topology. The integers $\mathbb{Z} \subset \mathbb{R}$ inherit the discrete topology — every singleton $\{n\}$ is open in $\mathbb{Z}$, because $\{n\} = (n - 0.5,\, n+0.5) \cap \mathbb{Z}$. So a "thin," seemingly formless subset of $\mathbb{R}$ becomes a space in which every point is isolated.

:::svg[Shadows of open sets: how $(-1, 0.5)$ and $(0.5, 1.5)$ restrict to $A=[0,1]$]

Example 1: The half-open subspace $[0,1)$ of $\mathbb{R}$, and openness of $[0, 0.3)$ versus $[0.7, 1)$. Let $X = \mathbb{R}$ with the usual topology (open sets are arbitrary unions of open intervals), and let $A = [0,1)$. We investigate which of the following subsets of $A$ are open in the subspace topology:

Claim: $V_1 = [0,0.3)$ is open in $A$. Take $U_1 = (-0.5, 0.3)$, which is open in $\mathbb{R}$ (an open interval). Then

since the part of $(-0.5, 0.3)$ below $0$ is cut off by $A$, and the part at or above $0.3$ was never in $U_1$. Since $U_1 \in \tau$, $V_1 \in \tau_A$: it is open in $A$. Claim: $V_2 = [0.7, 1)$ is open in $A$. Take $U_2 = (0.7 - \epsilon, 2)$ for any small $\epsilon>0$, e.g. $U_2 = (0.65, 2)$, open in $\mathbb{R}$. Then

Careful — this is wrong as stated: $(0.65, 2) \cap [0,1) = [0.65,1)$ (using that $A$ starts at $0$, but the interval $(0.65,2)$ already excludes values below $0.65$, so the intersection is $(0.65,1)$, not including $0.65$). Let us redo this correctly. We want $U_2 \cap A = [0.7,1)$ exactly. Choose $U_2 = [0.7, 1) \cup (1, 2)$? That is not open. Instead choose $U_2 = (0.6, 1.5) \setminus \{0.6\}$— unnecessarily complicated. The clean choice: since $0.7$ is an interior point candidate only because $A$ does not extend below where we need it to, but here $0.7$ is interior to $\mathbb{R}$ itself, so we must be honest: to realize $[0.7,1)$ as $U \cap A$ we need $U$ to contain $0.7$ as an interior point of $\mathbb{R}$ and to stop exactly at $1$ from below while $A$ itself stops at $1$. Take $U_2 = (0.7 - \delta,\ 1+\delta)$ for any $\delta$ with $0 < \delta < 0.3$, say $\delta = 0.1$, so $U_2 = (0.6, 1.1)$. Then

Since $[0,1)$ contains no points below $0$, and $(0.6,1.1)$ contains no points $\le 0.6$, the intersection is $(0.6, 1)$ — this still does not include $0.6$, and does not equal $[0.7,1)$ either, since it includes points in $(0.6, 0.7)$ that we don't want. The issue: $0.7$ is not an endpoint like $0$ is; there is no ambient open set whose intersection with $A$ starts exactly at $0.7$ and is closed there, because near $0.7$, $A$ looks just like $\mathbb{R}$ (both extend on both sides), so any open $U \ni 0.7$ contains an interval around $0.7$, forcing $U \cap A$ to contain points below $0.7$ too. Hence: Corrected claim: $V_2 = [0.7,1)$ is <strong>not</strong> open in $A$. Suppose toward contradiction $[0.7,1) = U \cap A$ for some open $U \subseteq \mathbb{R}$. Then $0.7 \in U$, and since $U$ is open in $\mathbb{R}$, there is $\epsilon > 0$ with $(0.7-\epsilon, 0.7+\epsilon) \subseteq U$. Choose $\epsilon' = \min(\epsilon, 0.1)$; then $0.7 - \epsilon'/2 \in U$ and also $0.7 - \epsilon'/2 \in A = [0,1)$ (since it's between $0$ and $1$), so $0.7 - \epsilon'/2 \in U \cap A = [0.7,1)$. But $0.7 - \epsilon'/2 < 0.7$, contradicting $0.7-\epsilon'/2 \in [0.7,1)$. Hence no such $U$ exists, and $V_2$ is not open in $A$. This is an important corrective moral: only sets touching the "boundary" of $A$ that is missing from $A$ can gain openness for free; $0$ is such a boundary point of $A=[0,1)$ (since points immediately below $0$ are outside $A$), but $0.7$ is an interior point of the ambient line, not a boundary artifact of $A$, so $[0.7,1)$ behaves in $A$ exactly as it does in $\mathbb{R}$: not open. Claim: $V_3 = (0.2,0.3)$ is open in $A$. This is immediate since $(0.2,0.3)$ is already open in $\mathbb{R}$, and $(0.2,0.3) \cap A = (0.2,0.3)$ (as $(0.2,0.3) \subset [0,1)$ already), so $V_3 = U \cap A$ with $U = (0.2,0.3) \in \tau$.

Conclusion of Example 1: in the subspace $A = [0,1)$, open sets include ordinary open intervals inside $A$, together with "left-clipped" intervals $[0, b)$ for $b \le 1$ — because $0$ is a boundary point of $A$ excluded from below — but not right-clipped intervals like $[c, 1)$ for $0 < c < 1$, since $1 \notin A$ provides no such exemption on the right at any interior point $c$.

Example 2: Closed sets in $A = \mathbb{Q}$, the rationals, and a set that is clopen in $\mathbb{Q}$ but neither open nor closed in $\mathbb{R}$. Let $X = \mathbb{R}$ with the usual topology, and $A = \mathbb{Q}$, the rational numbers, with the subspace topology. Fix the irrational number $\sqrt{2}$ and consider

Claim: $V$ is open in $\mathbb{Q}$. Since $U = (-\infty, \sqrt2)$ is open in $\mathbb{R}$ and $V = U \cap \mathbb{Q}$, $V \in \tau_{\mathbb{Q}}$ directly from the definition. Claim: $V$ is also closed in $\mathbb{Q}$. By Part 2 of the Subspace Topology Theorem (i.e., the theorem proved above), $V$ is closed in $\mathbb{Q}$ iff $V = D \cap \mathbb{Q}$ for some $D$ closed in $\mathbb{R}$. Take $D = (-\infty, \sqrt2]$, which is closed in $\mathbb{R}$ (complement is the open set $(\sqrt2,\infty)$). Then

because $\sqrt2 \notin \mathbb{Q}$, so intersecting $(-\infty,\sqrt2]$ with $\mathbb{Q}$ gives exactly the same rationals as intersecting the open version $(-\infty,\sqrt2)$ with $\mathbb{Q}$ (the single extra endpoint $\sqrt2$ is irrational, hence not in $\mathbb{Q}$, hence doesn't change the intersection). So $V = D \cap \mathbb{Q}$ with $D$ closed in $\mathbb{R}$, hence $V$ is closed in $\mathbb{Q}$. Conclusion: $V$ is clopen (both open and closed) in the subspace $\mathbb{Q}$, even though as a subset of $\mathbb{R}$ it is neither open (it contains no interval around any rational close to $\sqrt2$ from the — wait, it is in fact open in $\mathbb{R}$? Let's check: is $\mathbb{Q}\cap(-\infty,\sqrt2)$ open in $\mathbb{R}$? No — its interior in $\mathbb{R}$ is empty, since every open interval contains irrationals not in $V$ and also points beyond... actually every nonempty open interval contains points of $\mathbb{Q}\cap(-\infty,\sqrt2)$ and points not in it arbitrarily close, so $V$ has empty interior in $\mathbb{R}$; it is not open in $\mathbb{R}$) nor closed in $\mathbb{R}$ (its closure in $\mathbb{R}$ is all of $(-\infty,\sqrt2]$, which is strictly larger than $V$, since every real number in $(-\infty,\sqrt2)$, rational or not, is a limit of rationals in $V$). This example demonstrates vividly how the subspace topology can manufacture clopen sets — a phenomenon central to the theory of Connected Spaces, since a space with a nontrivial proper clopen subset is by definition disconnected: $\mathbb{Q}$ is disconnected by the partition $\mathbb{Q} = V \sqcup W$ into two disjoint nonempty clopen pieces.