Locally Closed Subset¶
A subset of a topological space that is open within its own closure, equivalently the intersection of an ambient-open and ambient-closed set.
Core Idea¶
A subset \(E\) of a specified topological space \(X\) is locally closed when it is open in its closure \(\overline E^X\). Equivalently, there are a set \(U\) open in \(X\) and a set \(F\) closed in \(X\) such that \(E=U\cap F\). The relation is relative to the ambient topology: \(E\) can be neither open nor closed in \(X\) while being open within the smaller space \(\overline E^X\). Brian Conrad proves these equivalent tests, as well as the pointwise formulation: each point of \(E\) has an ambient neighborhood on which \(E\) is closed.[1]
The closed set supplies a support; the open set selects a region within it. One can always choose the closed support to be \(\overline E^X\), but other open–closed witnesses may exist. Writing \(U=X\setminus F_2\) also yields \(E=F_1\setminus F_2\) for closed \(F_1,F_2\) in \(X\). These are equivalent certificates of one subset property, not successive construction steps or separate abstractions.[1]
Structural Signature¶
Sig role-phrases: ambient topological space → selected subset → closed support and open window → relative-open closure test.
- Ambient topological space. The topology of \(X\) determines what counts as open, closed and the closure \(\overline E^X\). Without the ambient space, “locally closed” is under-specified; passing to a subspace can change the verdict.[1]
- Selected subset. The predicate concerns one \(E\subseteq X\). The related assertion that every subset of a given space satisfies a property is a different, whole-space classification, not another name for \(E\).[1][2]
- Closed support and open window. An ambient-closed \(F\) contains the relevant support while an ambient-open \(U\) selects exactly \(E=U\cap F\). The intersection is one piece. A union of several such pieces need not itself admit a single open–closed intersection witness.[1][3]
- Relative-open closure test. Independently of a chosen witness, check whether \(E\) is open as a subset of \(\overline E^X\). This test exposes the failure when \(E\) is dense in \(X\) but not open in \(X\): then \(\overline E^X=X\), so relative openness would require ambient openness.[1]
What It Is Not¶
It is not merely open or merely closed in \(X\). Every ambient-open set is locally closed by taking \(F=X\); every ambient-closed set is locally closed by taking \(U=X\). But \((0,1]\) is neither open nor closed in \(\mathbb R\) and is still locally closed. Thus Open Set and Closed Set describe witnesses or special cases, not upward genera for all locally closed subsets.[1]
It is not every finite union of locally closed subsets. In \(\mathbb R^2\), the open set \(\{(x,y):x\ne0\}\) and the closed singleton \(\{(0,0)\}\) are each locally closed. Their union is dense in \(\mathbb R^2\) but not open at the origin; therefore it is not locally closed. The separate word constructible has convention-sensitive technical meaning. In the Stacks Project's general-space definition it is formulated with retrocompact opens, while in a Noetherian space its constructible subsets are precisely finite unions of locally closed subsets. Do not omit that hypothesis when quoting the Stacks equivalence.[3][4]
It is not automatically a locally closed subscheme. A subset is only a collection of points with an induced topology; a subscheme also carries sheaf and ideal information. The Stacks Project constructs a reduced induced structure on a locally closed subset of a scheme, but that is an added scheme-theoretic choice, not a claim that the bare topological predicate determines every possible nonreduced structure.[5][6]
Finally, Submaximal Space is a distinct whole-space property concerning dense subsets. Its Wikipedia redirect to this article is candidate-discovery provenance, not evidence that the two identities are synonyms or that its candidate ID is covered here.[2]
Scope of Application¶
In ordinary point-set topology, half-open intervals provide easy cases where the ambient and relative notions differ. The predicate works in an arbitrary topological space; it does not require a metric, separation axiom or Noetherian condition. What must remain fixed is the ambient \(X\) against which closure and openness are measured.[1]
In algebraic geometry, Conrad describes subsets of an affine algebraic set of the form \(Z\cap\{f\ne0\}\), with \(Z\) Zariski closed and polynomial nonvanishing Zariski open. The punctured \(x\)-axis in the affine plane is such a subset. This is a genuine use of the same open-in-closed structure under a very different topology. When algebraic geometers attach a locally closed subscheme to such a set, that adds scheme structure beyond the topological classification.[1][5][6]
In a Noetherian topological space, finite unions of these pieces characterize constructible subsets under the Stacks Project convention. That use depends on the Noetherian qualification; it does not make every constructible set itself locally closed, nor does it license copying the statement unchanged into arbitrary spaces.[4][3]
Clarity¶
The word locally does not mean “open in the ambient space near every point.” It says that each point of \(E\) has a neighborhood on which \(E\) is closed, equivalently that \(E\) is open within its own closure. The two descriptions initially look opposite; Conrad's proof shows they identify the same interface between an open window and a closed support. The chosen ambient space must accompany every such statement.[1]
This distinction resolves three common confusions. A half-open interval need not be globally open or closed. A finite collection of locally closed strata is not necessarily one locally closed stratum. And a locally closed locus of points does not, without extra data, say which scheme structure or nilpotent information is present there.[1][4][6]
Manages Complexity¶
To decide local closedness, one need not test an unrelated neighborhood construction at every point. Compute \(\overline E^X\) and ask a single relative-openness question, or exhibit one \(U\cap F\) witness. Each certificate collapses infinitely many local checks into an ambient set relation. The two certificates are useful in different presentations: closure is natural for an already described \(E\), while \(U\cap F\) is natural when equations and nonvanishing conditions specify the set.[1]
The compression has a limit. If an object is described as a finite union of pieces \(U_i\cap F_i\), recording each piece may be necessary; compressing the union into one locally closed label can be false. Similarly, treating a topological support as though it carried a unique arbitrary subscheme structure discards information rather than managing it.[3][4][6]
Abstract Reasoning¶
Start with \(E\subseteq X\). If a candidate closed support \(F\) and open condition \(U\) are apparent, verify both are ambient-open/closed and that \(E=U\cap F\). If not, compute \(\overline E^X\) and test whether its complement of \(E\) is closed within that closure. Failure of both tests excludes the predicate. A particularly fast disproof occurs when \(E\) is dense in \(X\) but not ambient-open: its closure is all of \(X\), leaving no smaller support in which it could be relatively open.[1]
Once the predicate holds, one can reason locally on an open part of a closed support; that does not imply compactness, smoothness, equidimensionality or an automatically chosen scheme structure. If a later argument uses finite stratification, check whether it needs one locally closed set or a union of them, and whether a Noetherian constructibility theorem is actually available.[1][3][4][6]
Knowledge Transfer¶
The definition transfers literally from Euclidean to Zariski topology: in both, fix \(X\), form a closure, then require the selected subset to be open inside it. What changes is which sets the topology deems open and closed, so the same point-set description cannot be moved between topologies without retesting it. Algebraic geometry also adds a separate question about scheme structure that ordinary topology does not answer.[1][5]
Outside topology, “open within a closed support” can be an illuminating analogy, but the named mathematical predicate requires genuine ambient open and closed sets. Whether the portable skeleton merits a prime node is a future-prime question. This entry does not claim cross-domain reach by metaphor alone, nor does it turn its Open Set and Closed Set operands into strict parents.
Examples¶
A half-open Euclidean interval¶
Let \(X=\mathbb R\) with the usual topology and \(E=(0,1]\). Put \(U=(0,\infty)\), open in \(\mathbb R\), and \(F=(-\infty,1]\), closed in \(\mathbb R\). Then \(E=U\cap F\). Its closure in \(\mathbb R\) is \([0,1]\), and \((0,1]=(0,\infty)\cap[0,1]\) is open in that closure. Yet \(E\) is not ambient-open because no neighborhood of $1$ lies in \(E\), and is not ambient-closed because $0$ lies in its closure but not in \(E\). This is local closedness without either simpler ambient classification.[1]
Mapped back: Ambient topological space → usual \(\mathbb R\); selected subset → \((0,1]\); closed support and open window → \(F=(-\infty,1]\) and \(U=(0,\infty)\); relative-open closure test → \(E\) is open in \(\overline E^{\mathbb R}=[0,1]\).
A punctured Zariski-closed line¶
Let \(X=\mathbb A^2_{\mathbb C}\) with its Zariski topology. Write \(F=V(y)\) for the \(x\)-axis and \(U=D(x)=\{x\ne0\}\) for the principal open nonvanishing locus. Then \(E=F\cap U\) is the \(x\)-axis with its origin removed. The line \(V(y)\) is its Zariski closure and \(E\) is open inside that line. It is not closed in the plane because the origin is a closure point. Nor is it open in the plane: a nonempty Zariski-open set in the irreducible affine plane cannot be confined to the proper closed line \(V(y)\). Conrad's general algebraic form is precisely closed \(Z\) intersect nonvanishing \(\{f\ne0\}\).[1]
Mapped back: Ambient topological space → Zariski \(\mathbb A^2_{\mathbb C}\); selected subset → punctured \(x\)-axis; closed support and open window → \(V(y)\) and \(D(x)\); relative-open closure test → \(E\) is open in \(\overline E^X=V(y)\).
Structural Tensions¶
T1: One-piece clarity versus finite-union coverage. A single \(U\cap F\) witness makes the local-closedness test compact and supports reasoning on one relative-open piece. A finite union of such pieces reaches more subsets in Noetherian constructible geometry, but loses the guarantee that the union itself has one locally closed witness; one must track the pieces and their frontiers separately. Insisting on one piece can exclude useful constructible loci, while calling every finite union one piece can make a false topological claim. Diagnostic: Does the proof need one relatively open piece, or only a finite decomposition into pieces?[1][4]
T2: Topological portability versus scheme-theoretic information. The bare subset test applies across topologies and avoids extra sheaf choices, but cannot distinguish structures with the same underlying points. Retaining scheme-theoretic structure permits claims about ideals and nilpotents, at the cost of additional data that the topological predicate alone does not supply. Neither side subsumes the other: use the subset when point-set localization is enough, and specify a subscheme when the argument needs its structure. Diagnostic: Would changing the sheaf or nilpotent structure while keeping the same point set change the conclusion?[5][6]
Structural–Framed Character¶
This entry sits near the structural end of the spectrum within a mathematical frame: once an ambient topology is fixed, the open-in-closure criterion is an exact set relation, not an evaluative judgment. Evaluative weight: “locally closed” asserts membership in a formal class, not desirability or quality. Human-practice dependence: people choose which topology and presentation to study, but the truth of \(E=U\cap F\) then follows from set relations rather than convention or institutional assessment. Institutional origin: the term belongs to mathematical topology and geometry; no policy body makes a particular subset locally closed. Vocabulary travel: loose uses of “locally closed” outside topology do not carry the open/closed-set axioms, whereas Euclidean and Zariski uses do. Import versus recognition: importing the name to another field would require a genuine topology and closure operator; recognizing a merely similar bounded-window pattern is not enough.[1]
Its character: formally structural after the ambient topology is declared, yet domain-specific because that topology, relative openness and closure are indispensable to the named predicate. The tempting portable “open within closed support” skeleton is held as a future-prime question, not asserted as an established parent.
Structural Core vs. Domain Accent¶
The candidate portable core is the relationship “select an accessible region within a containing support.” But in this entry its terms are not generic access and support: \(U\) must be topologically open in \(X\), \(F\) topologically closed in \(X\), and \(E\) open in \(\overline E^X\). Those axioms are what make the equivalence proof work. Merely seeing a bounded region or an inclusion rule in another domain does not instantiate the locally closed subset.[1]
No live prime has been checked as a strict parent carrying that entire relation. The more portable open-within-closed-support pattern is therefore an explicit future-prime question, while this typed mathematical node remains domain-specific. Its domain accent is the ambient topology and closure-relative criterion; algebraic geometry's sheaf-bearing locally closed subscheme is an additional, distinct layer rather than a hidden ingredient of the core.[5][6]
Instantiates / Related Primes¶
No strict prime parent is proposed. Live Open Set and Closed Set supply the two operands of the \(U\cap F\) certificate and are special cases when \(F=X\) or \(U=X\), respectively. They cannot be strict parents of a non-open, non-closed instance such as \((0,1]\). Live Quasi-projective Variety is a more specifically algebraic related identity; its projective-variety conditions do not define arbitrary locally closed subsets of arbitrary spaces. This node is staged unparented pending independent DAG review.
Neighborhood in Abstraction Space¶
Locally Closed Subset sits in a crowded region of the domain-specific corpus (38th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.
Family — Point-Set Topology Foundations (17 abstractions)
Nearest neighbors
- Open Set — 0.89
- Clopen Set — 0.89
- Interior — 0.88
- Subspace Topology — 0.88
- A-paracompact Space — 0.88
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- Open Set / Closed Set: either is locally closed in its ambient space, but neither exhausts locally closed subsets.[1]
- Constructible Subset: the finite-union relationship requires the stated Noetherian condition in the Stacks equivalence, and a union need not be one locally closed subset.[3][4]
- Locally Closed Subscheme: adds scheme structure; a reduced induced structure can be chosen, but it is not contained in the bare subset predicate.[5][6]
- Submaximal Space: a whole-space property involving dense subsets, separately held despite the Wikipedia redirect.[2]
References¶
[1] Brian Conrad, “Dimension”, Stanford Math 145 handout, p. 1 (definition, equivalences and Zariski \(Z\cap\{f\ne0\}\) example). registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s ↩t ↩u
[2] Rostam Mohamadian, “Locally closed sets and submaximal spaces”, arXiv:2205.07191 (2022), abstract and introduction. registry ↩a ↩b ↩c
[3] The Stacks Project, “Constructible subsets”, Definition 5.15.1 and terminology note. registry ↩a ↩b ↩c ↩d ↩e ↩f
[4] The Stacks Project, “Constructible sets in Noetherian spaces”, Lemma 5.16.1. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g
[5] The Stacks Project, “Immersions of schemes”, §26.10, Definition 26.10.2. registry ↩a ↩b ↩c ↩d ↩e ↩f
[6] The Stacks Project, “Reduced induced schemes”, §26.12, Lemma 26.12.4 and Remark 26.12.6. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h