Collectionwise Normal Space¶

Separate every discrete indexed family of closed sets at once by matching pairwise-disjoint open neighborhoods, extending ordinary two-set normality to arbitrary family size.

Version: v1 · 2026-08-30 · History
Domain-specific #: 1498
Origin domain: mathematics
Subdomain: general topology
Aliases: Collectionwise Normality, Collection Wise Normal Space, Collection Wise Normality

Core Idea¶

A topological space \(X\) is collectionwise normal when every discrete indexed family \((F_i)_{i\in I}\) of closed subsets can be expanded to a matching pairwise-disjoint family of open sets \((U_i)_{i\in I}\):

\[ F_i\subseteq U_i\quad(i\in I), \qquad U_i\cap U_j=\varnothing\quad(i\ne j). \]

The input word discrete is a family condition. Every point of \(X\) must have a neighborhood meeting at most one \(F_i\).^[1] It is stronger than saying the \(F_i\) are pairwise disjoint and stronger than local finiteness, but it does not say that \(X\) itself has the discrete topology. The output is also indexed: \(U_i\) must contain the corresponding \(F_i\), so the axiom cannot be discharged by finding unrelated disjoint open sets. Engelking's standard equivalence theorem permits the output family to be required to be discrete, not merely pairwise disjoint.^[2]^[3]

The abstraction upgrades ordinary normality from a finite separation guarantee to an arbitrary-family guarantee. A normal space can separate two disjoint closed sets and, by finite repetition, any finite discrete family. Under the convention used here, ordinary normality separates every countable discrete family of closed sets; unrestricted collectionwise normality removes the cardinal bound. Thus countably collectionwise normal remains standard terminology but is equivalent to normality under this convention. Nothing in that result licenses uncountable iteration: previously chosen neighborhoods can accumulate and exhaust the room needed for later members. Collectionwise normality packages exactly the missing unrestricted simultaneous choice. That stronger guarantee is why the condition appears in metrization theory: every metrizable space has it, and Bing proved that a Moore space is metrizable exactly when it is collectionwise normal.^[4]^[5]^[6]

Separation conventions must remain visible. Some authors include \(T_1\) or Hausdorffness in the phrase; others define only the family-expansion axiom. This entry uses the latter, convention-neutral core and states extra separation hypotheses whenever a theorem requires them. Under the matching neutral convention for normality, collectionwise normality implies normality by applying the axiom to a two-member discrete closed family. It also implies collectionwise Hausdorffness when the relevant singleton family is closed, but neither converse holds in general.^[7]^[6]

Structural Signature¶

Sig role-phrases:

the topology-bearing carrier — a space \((X,\tau)\) whose open and closed subsets supply the ambient separation language
the arbitrary index set — \(I\), with no finite or countable restriction in the base property
the discrete closed input family — \((F_i)_{i\in I}\), locally isolated as a family and closed member by member
the neighborhood discreteness test — every point of \(X\) has a neighborhood meeting at most one \(F_i\)
the indexed open witnesses — \((U_i)_{i\in I}\), one open neighborhood assigned to each input member
the containment matching — \(F_i\subseteq U_i\) under the same index, preserving which witness belongs to which closed set
the simultaneous disjointness verdict — \(U_i\cap U_j=\varnothing\) for all distinct indices, established at once rather than sequentially
the equivalent discrete-expansion form — the witnesses may be chosen as a discrete open family, a theorem stronger in appearance than bare pairwise disjointness
the convention and consequence contract — explicit handling of \(T_1\)/Hausdorff assumptions and of implications to normality, collectionwise Hausdorffness, and metrization only under their stated hypotheses

Each role is necessary. Dropping closedness changes the tested class; dropping family discreteness asks for impossible separation of overlapping or accumulating sets; dropping the shared index loses the expansion; dropping simultaneity collapses the property to a sequence of locally plausible but globally incompatible choices. The result is a universal existence axiom, not an algorithm that canonically selects one unique family of witnesses.

What It Is Not¶

Not ordinary normality with a plural noun. Normality controls a pair of disjoint closed sets. Collectionwise normality controls every discrete closed family of arbitrary cardinality, and that quantifier increase is substantive.^[7]
Not Hausdorffness. Hausdorff separation concerns two points. The present input members may be large closed subsets, and the witnesses must work simultaneously across the whole indexed family.
Not collectionwise Hausdorffness. That weaker property tests closed discrete sets of points—families of singletons. Collectionwise normality permits arbitrary closed members.^[6]
Not a discrete ambient space. “Discrete” modifies the family \((F_i)\), not \(X\). Connected metric spaces such as \(\mathbb R^2\) are collectionwise normal.
Not pairwise disjointness of the input. A pairwise-disjoint family can accumulate. The intervals \((1/(n+1),1/n)\) in \(\mathbb R\) are pairwise disjoint but every neighborhood of zero meets infinitely many.
Not local finiteness. A locally finite family can overlap, whereas a discrete family cannot. The related locally-finite open-expansion property is called expandability.
Not paracompactness. Hausdorff paracompactness is a sufficient route to collectionwise normality, not the same covering property.
Not metrizability. A metric supplies enough distance control to build the disjoint expansion, but collectionwise normality alone supplies no metric. Bing's result needs a Moore-space development as well.^[4]
Not automatically hereditary. Requiring every subspace to be collectionwise normal defines the stronger hereditarily collectionwise normal property.
Not a finite or countable test. Countably collectionwise normal and \(\kappa\)-collectionwise normal restrict the allowed family sizes and are genuine variants.

Scope of Application¶

Collectionwise normality belongs to general and set-theoretic topology, where the size and arrangement of closed families control metrization, covering, extension, and subspace results. Its technical vocabulary remains literal in these settings and should not be exported as a loose synonym for “separate many things.”

Separation-axiom hierarchies — locates the gap between normality, collectionwise Hausdorffness, collectionwise normality, and their hereditary or cardinal-bounded refinements.
Metrization theory — combines with developments, uniform bases, or related cover structures in criteria that recover a compatible metric.^[4]^[5]
Moore-space theory — supplies the exact strengthening of normality that Bing's metrization theorem needs; the normal Moore-space problem shows why ordinary normality cannot simply be substituted.^[6]
Paracompactness and covering properties — receives sufficient conditions from Hausdorff paracompactness and interacts with locally finite refinements, while remaining distinct from them.
Set-theoretic topology — exposes where normality-to-collectionwise-normality implications depend on additional topological or set-theoretic assumptions.^[7]^[6]
Dimension and extension theory — discrete open expansions are used to coordinate indexed local constructions without unintended intersections; the family-level guarantee is the reusable input.

The boundary is equally important: ordinary metric examples often make the property look automatic because distance supplies a uniform construction. The node becomes discriminating precisely in nonmetrizable spaces and in theorems where a pairwise separation axiom is too weak to coordinate an arbitrary family.

Clarity¶

The name makes one hidden quantifier visible. “Normal” answers whether two closed sets can be separated; “collectionwise” asks whether the answer stays yes when the problem is presented as one arbitrary indexed family. That change blocks the invalid proof strategy “separate the sets one pair at a time.” For infinitely many members, locally correct choices can accumulate, and revising one neighborhood can destroy disjointness already established elsewhere. The collectionwise axiom asserts the existence of a globally compatible assignment in one statement.

It also separates three notions often blurred by informal language. A pairwise-disjoint family has no shared points; a locally finite family is met only finitely often near each point; a discrete family is met at most once near each point. Only the last is the input condition. On the output side, pairwise-disjoint open witnesses suffice by definition, while the theorem that they can be selected as a discrete open family gives a stronger-looking normal form. Naming the roles lets a reader ask the right debugging questions: Are the inputs closed? Is the family truly discrete at accumulation points? Does each witness contain its assigned member? Was disjointness achieved simultaneously? Which \(T_1\) or Hausdorff convention is in force?

Manages Complexity¶

Without the abstraction, a proof involving \((F_i)_{i\in I}\) carries a quadratic—and for infinite \(I\), globally coupled—obligation: choose one neighborhood per member, preserve every containment, and prevent every pair of witnesses from intersecting. Solving those constraints sequentially is unsafe because the remaining feasible region changes after every choice. Collectionwise normality compresses the whole compatibility problem to one reusable certificate on the ambient space. Once \(X\) is known to have the property and the input family is certified discrete and closed, the indexed disjoint expansion can be invoked without rebuilding it for that theorem.

The compression produces a clean branching discipline. If the family has two members, ordinary normality is the cheaper tool. If it is an arbitrary discrete closed family, collectionwise normality is the relevant certificate. If the members are singletons, collectionwise Hausdorffness may suffice. If only countable families occur, a cardinal-bounded variant may be enough. If every subspace must inherit the result, the hereditary variant is the actual requirement. This prevents both under-specification—using normality where family coordination is needed—and over-specification—demanding a metric or paracompactness when the proof consumes only a disjoint open expansion.

Abstract Reasoning¶

Diagnostic — certify the input before invoking the axiom. Verify that each \(F_i\) is closed and that every point has a neighborhood meeting at most one family member. Pairwise disjointness is not enough; one accumulation point can invalidate the premise. The inference is then direct: discrete closed input family plus collectionwise-normal ambient space yields matching pairwise-disjoint open witnesses.

Constructive — exploit a metric when one is available. For a nontrivial discrete closed family in a metric space, put \(H_i=\bigcup_{j\ne i}F_j\). Discreteness makes the family locally finite, so \(H_i\) is closed. Distance functions then define

\[ U_i=\{x:d(x,F_i)<\tfrac13 d(x,H_i)\}. \]

Each \(U_i\) is open and contains \(F_i\). If a point lay in both \(U_i\) and \(U_j\), then \(d(x,F_i)<d(x,F_j)/3\) and \(d(x,F_j)<d(x,F_i)/3\), an impossibility. This move explains, rather than merely cites, why metrizability is sufficient.

Boundary-drawing — choose the weakest adequate separation property. Ask whether the theorem's objects are two closed sets, a closed discrete set of points, or an arbitrary discrete closed family. Route respectively toward normality, collectionwise Hausdorffness, or collectionwise normality. Then ask whether the family-size or subspace quantifier requires a countable, \(\kappa\)-bounded, or hereditary variant.

Predictive — combine independent ingredients in a metrization criterion. A Moore-space development controls local bases across scales; collectionwise normality controls simultaneous separation. Bing's theorem predicts a compatible metric from the conjunction, not from either label alone.^[4] The order matters: verify the development and the family-level separation, then conclude metrizability.

Counterexample discipline — refuse infinite induction. A finite proof of normal separation does not extend by “continue forever.” At a limit stage, the union or closure of earlier choices can meet later inputs. The correct move is to demand a theorem providing the entire indexed expansion or to strengthen the ambient hypothesis to collectionwise normality.

Knowledge Transfer¶

Within point-set, set-theoretic, and metrization topology, the mechanism transfers intact. The same five roles—discrete closed family, arbitrary index, open expansion, same-index containment, simultaneous disjointness—recur in Moore-space metrization, covering-property arguments, extension theorems, and hereditary refinements. A proof can move between these subfields without translating the identity because open, closed, discrete family, and neighborhood keep their formal meanings.

Outside topology, only a loose coordination pattern travels: many claimants need protected regions assigned simultaneously without overlap. That residue does not carry the topology-specific premise or conclusions. Replacing closed sets with departments, radio channels, or database locks changes what “neighborhood,” “open,” and “discrete” mean and removes the separation-axiom theorems. Such cases may instantiate generic allocation, isolation, or constraint-satisfaction patterns, but they are not Collectionwise Normal Spaces. The honest transfer policy is therefore full mechanism within topology, structural analogy outside it.

Examples¶

Canonical — metric expansion and separated vertical lines¶

Let \((X,d)\) be a metric space and let \((F_i)_{i\in I}\) be a discrete family of nonempty closed sets with at least two members. For each \(i\), set \(H_i=\bigcup_{j\ne i}F_j\). A discrete family is locally finite, and a locally finite union of closed sets is closed, so \(H_i\) is closed. Because \(F_i\cap H_i=\varnothing\), every point of \(F_i\) has positive distance from \(H_i\). Define

\[ U_i=\{x\in X:d(x,F_i)<\tfrac13 d(x,H_i)\}. \]

The distance functions are continuous, so \(U_i\) is open, and \(F_i\subseteq U_i\). For \(i\ne j\), membership of \(x\) in \(U_i\) would give \(d(x,F_i)<d(x,F_j)/3\), since \(F_j\subseteq H_i\); membership in \(U_j\) would give the reverse strict inequality with the same factor. Thus the \(U_i\) are pairwise disjoint. A one-member family is handled by choosing \(U_i=X\).

In the visible special case \(X=\mathbb R^2\), take \(F_n=\{n\}\times\mathbb R\) for \(n\in\mathbb Z\). The family is discrete: around any point, a ball of radius less than \(1/2\) meets at most one vertical integer line. The open strips \(U_n=(n-1/3,n+1/3)\times\mathbb R\) contain their assigned lines and are pairwise disjoint.

Mapped back: the topology-bearing carrier is the metric topology on \(X\); \(I\) is arbitrary in the general construction and \(\mathbb Z\) in the visible case; the \(F_i\) are the discrete closed inputs; point-to-set distance constructs the indexed open witnesses; the unchanged index gives containment matching; the two incompatible distance inequalities certify simultaneous disjointness; and the result exhibits every metric space as collectionwise normal.^[5]

Applied in metrization theory — Bing's Moore-space test¶

Suppose a topologist is given a Moore space \(X\): its topology has a development, a sequence of open covers that supplies increasingly fine local control. A development resembles metric-scale information, but it does not by itself produce a metric. The second audit asks whether every discrete closed family \((F_i)\) has a matching disjoint open expansion. If yes, \(X\) is collectionwise normal, and Bing's theorem concludes that \(X\) is metrizable. Conversely, a metrizable space has both ingredients.^[4]

The distinction is operationally decisive. Replacing the second audit with ordinary normality changes the claim into the historical normal Moore-space conjecture. Devlin and Shelah review why that substitution is not a theorem of ZFC and give models with normal nonmetrizable Moore spaces; they also state the exact collectionwise-normal version as Bing's theorem.^[6] Thus a candidate proof that reports only “\(X\) is normal” has not discharged the family-level obligation.

Mapped back: \(X\) is the topology-bearing carrier; each arbitrary index set and discrete closed family forms a separation test; collectionwise normality supplies the same-index open expansion and global disjointness; the Moore development supplies the independent local-scale role; the explicit convention distinguishes normal from collectionwise normal; and the combined consequence contract licenses the metric conclusion exactly where Bing's hypotheses, rather than the weaker normality label, are satisfied.

Structural Tensions¶

T1: Pairwise disjointness versus family discreteness. Disjoint members can still accumulate at a point; discrete families cannot. Diagnostic: does every point have a neighborhood meeting at most one input member, including points outside the union?

T2: Finite separation versus arbitrary coordination. Normality handles every fixed finite discrete family, while limit stages can destroy a sequential construction. Diagnostic: is the index set genuinely bounded and finite, or does the proof need one simultaneous choice for arbitrary \(I\)?

T3: Pairwise-disjoint output versus discrete output. A discrete open expansion is stronger in appearance, though equivalent by theorem. Diagnostic: is the proof invoking Engelking's equivalence, or silently treating two nonidentical family conditions as definitions?

T4: Neutral identity versus separation convention. Some texts include \(T_1\) or Hausdorffness in the name, while others do not. Diagnostic: which convention is declared, and are theorem-specific separation hypotheses written rather than assumed?

T5: Sufficient class versus exact classification. Metric and Hausdorff paracompact spaces provide the property, but the property itself supplies neither a metric nor all of paracompactness. Diagnostic: is a one-way sufficient theorem being mistaken for an equivalence?

T6: Base property versus graded variants. Countable, \(\kappa\)-bounded, strong, and hereditary versions alter family-size, witness, or subspace quantifiers. Diagnostic: which quantifier has changed, and has the variant name been preserved?

T7: Existence certificate versus canonical algorithm. The axiom ensures some compatible expansion, but typically not a unique or functorial choice. Diagnostic: does the downstream argument need mere existence, or an additional rule selecting witnesses coherently across changes of input?

T8: Autonomy versus reduction. The node is built from Topological Space, Open Set, Closed Set, indexing, and family discreteness, yet their loose conjunction does not assert simultaneous disjoint expansion. Diagnostic: if the proof only needs topology or one open witness, reduce to those nodes; if it consumes the universal indexed-family guarantee, preserve Collectionwise Normal Space.

Structural–Framed Character¶

Collectionwise Normal Space is structural. Its evaluative weight is nil: a space either satisfies the quantified expansion axiom under a stated convention or it does not. “Normal” here is technical vocabulary, not praise or a judgment that the space behaves properly.

It is not human-practice-bound. Closed subsets, open neighborhoods, family discreteness, and the universal quantifier are mathematical objects. Human choice affects notation and separation conventions, but not whether a fixed formal statement is true of \((X,\tau)\).

Its institutional origin is general and set-theoretic topology, yet the property is not maintained by an institution or social rule. Its import-versus-recognize pattern is recognition inside any topological space: once \(\tau\) and the quantified family axiom are fixed, no agent must enact the property for it to hold.

Its vocabulary travels literally through separation axioms, Moore spaces, metrization, and covering theory, but not outside topology. Open and closed sets, discrete families, and topological neighborhoods are constitutive, so the node remains domain-specific even though its formal shape is clean.

Its character: a purely structural, topology-bound family-separation axiom that upgrades pairwise closed-set separation to one globally compatible open expansion for every discrete indexed collection.

Structural Core vs. Domain Accent¶

The structural core is a quantified compatibility guarantee: for every admissible indexed input family, assign one containing witness to each member so that all witnesses satisfy a global nonintersection constraint. That core explains the node's problem-solving force—local feasible choices are replaced by one simultaneous existence certificate.

The domain accent is not removable decoration. “Admissible” means discrete family in the neighborhood sense; inputs must be closed subsets of one topological space; witnesses must be open in the same topology; containment uses the original index; disjointness is set intersection; and the licensed consequences belong to the normality and metrization hierarchies. Remove those roles and one gets a broad allocation or separation pattern, not collectionwise normality. The concept therefore fails the prime bar while remaining a stable, reusable abstraction across topology.

Collectionwise Normal Space is related to the prime topology because every role and consequence is invariantly stated in the open-set structure, but it is not another prime-level account of topology. It uses set_and_membership through indexed families, subset containment, unions, and intersections. Its simultaneous witness assignment resembles constraint satisfaction: every local containment constraint and every global nonintersection constraint must be satisfied together.

The word discrete also makes discreteness a useful comparison, not a DAG parent. The live prime characterizes separated states or an ambient discrete structure; here discreteness is a premise on one family inside a space that may be connected. Likewise, Open Set and Closed Set are domain-specific constituents, while Topological Space is the exact taxonomic live parent. The developed node's autonomous contribution is the universal same-index disjoint-expansion guarantee that none of those constituents entails alone.

Relationships to Other Abstractions¶

Current abstraction Collectionwise Normal Space Domain-specific

Parents (1) — more general patterns this builds on

Collectionwise Normal Space is a kind of Topological Space Domain-specific

Collectionwise Normal Space is related to the prime topology because every role and consequence is invariantly stated in the open-set structure, but it is not another prime-level account of topology.

Hierarchy paths (5) — routes to 3 parentless roots

Collectionwise Normal Space → Topological Space → Closure

Show alternative paths (4)

Neighborhood in Abstraction Space¶

Collectionwise Normal Space sits in a sparse region of the domain-specific corpus (62^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — General Topology & Separation (12 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-09-08

Not to Be Confused With¶

Normal space. Separates two disjoint closed sets, not every arbitrary discrete family. Tell: can the stated axiom be applied to an uncountable indexed closed family in one step?
Hausdorff space. Separates two distinct points. Tell: are the inputs points in a pair, or arbitrary closed sets in a family?
Collectionwise Hausdorff space. Separates a closed discrete set point by point. Tell: must each input member be a singleton?
Paracompact space. Requires locally finite refinements of open covers. Tell: is the quantified input an open cover or a discrete closed family?
Metrizable space. Has a topology induced by a metric. Tell: has an actual distance function been obtained, or only an open expansion?
Moore space. Has a development. Tell: is local base refinement supplied without the family-level separation axiom?
Discrete space. Every singleton is open. Tell: does “discrete” modify the whole ambient space or only the tested family?
Pairwise-disjoint family. Members do not intersect but may accumulate. Tell: what happens at a point outside the union, such as zero for intervals approaching zero?
Locally finite family. Each point meets only finitely many nearby members, possibly more than one. Tell: is “finitely many” sharpened to “at most one”?
Expandable space. Expands locally finite closed families to locally finite open families. Tell: are both the premise and conclusion local-finiteness conditions rather than discrete/disjoint ones?
Countably collectionwise normal space. Tests only countable discrete families. Tell: is the cardinality of \(I\) bounded by \(\aleph_0\)?
\(\kappa\)-collectionwise normal space. Tests families only up to a specified cardinal. Tell: is a cardinal parameter part of the property?
Hereditarily collectionwise normal space. Requires the property in every subspace. Tell: is the quantifier over subspaces present?
Strongly collectionwise normal space. Adds a stronger expansion/separation condition. Tell: is “strongly” defined by an extra formal witness requirement rather than used for emphasis?
Disjoint open expansion. Names one output family, not the ambient-space property. Tell: is the phrase describing witnesses for a particular \((F_i)\), or universally quantifying over all such families?
Discrete open expansion. Names the equivalent stronger-form witness family. Tell: is the open family itself locally isolated, and is this one construction being confused with the global axiom?

References¶

[1] P. S. Aleksandrov, “Normal space,” Encyclopedia of Mathematics. https://encyclopediaofmath.org/wiki/Normal_space. Verified 2026-08-26. registry ↩

[2] Ryszard Engelking, General Topology, revised and completed edition, Heldermann, 1989, Theorem 5.1.17. Bibliographic record: https://books.google.com/books/about/General_Topology.html?id=K3spAQAAMAAJ. Verified 2026-08-26. registry ↩

[3] π-Base, property P88, “Collectionwise normal.” https://topology.pi-base.org/properties/P88. Verified 2026-08-26. registry ↩

[4] R. H. Bing, “Metrization of Topological Spaces,” Canadian Journal of Mathematics 3 (1951), 175–186. https://doi.org/10.4153/CJM-1951-022-3. Verified 2026-08-26. registry ↩a ↩b ↩c ↩d ↩e

[5] “Metrizable space,” Encyclopedia of Mathematics. https://encyclopediaofmath.org/wiki/Metrizable_space. Verified 2026-08-26. registry ↩a ↩b ↩c

[6] Keith J. Devlin and Saharon Shelah, “A Note on the Normal Moore Space Conjecture,” Canadian Journal of Mathematics 31 (1979), 241–251. https://doi.org/10.4153/CJM-1979-025-8. Verified 2026-08-26. registry ↩a ↩b ↩c ↩d ↩e ↩f

[7] Mary Ellen Rudin, “Normality versus Collectionwise Normality,” in Handbook of Set-Theoretic Topology, North-Holland, 1984, pp. 687–732. https://doi.org/10.1016/B978-0-444-86580-9.50018-5. Verified 2026-08-26. registry ↩a ↩b ↩c