Topological Epistemology

Origin. McKinsey, Tarski (1940s), recent: Baltag, Bezhanishvili. Topology for knowledge. Open sets as evidence. Interior as knowable. Spatial epistemology.

Models. Knowledge via topological interior. Evidence as open sets. Knowability = being in interior. Connects topology and epistemology. Geometric intuition for knowledge.

Formalism.

Topological space: (X, τ) where X = states, τ = topology. Open sets: observable evidence. Interior int(A) = largest open subset. Closure cl(A) = smallest closed superset.

Knowledge as interior: [[Kφ]] = int([[φ]]). Know φ: have evidence for φ. Evidence = open neighborhood. Inside = supported by evidence.

Properties: K distributes: K(φ ∧ ψ) ↔ Kφ ∧ Kψ. Kφ → φ (factivity — int(A) ⊆ A in every topological space). Kφ → KKφ (positive introspection). S4 corresponds to topology.

Evidence semantics: E(φ) = open neighborhoods supporting φ. Combined evidence: intersection. Weaker evidence: superset. Evidence ordering.

Belief via closure: [[Bφ]] = cl(int([[φ]])). Believe φ: evidence consistent with φ. Weaker than knowledge. The KD45 belief laws require extremally disconnected spaces (Baltag, Bezhanishvili, Özgün, Smets).

Knowability: ◇Kφ: φ could be known. Not all truths knowable. Fitch's paradox addressed. Topological constraints.

Learning: Evidence shrinks neighborhoods. More evidence: smaller opens. Limit: singleton (complete knowledge). Dynamic topology.

Symbols.

Symbol	Unicode	Name	Meaning
int	—	Interior operator	Largest open subset; interprets K
cl	—	Closure operator	Smallest closed superset
τ	U+03C4	Topology	Family of open sets on X
Kφ	—	Knowledge	int([[φ]])
X	—	State space	Underlying set of the space
[[φ]]	—	Truth set	States where φ holds
B	—	Belief	cl(int([[φ]]))
E	—	Evidence	Open neighbourhoods supporting φ
◇	U+25C7	Diamond	Possibility; ◇Kφ is knowability
∧	U+2227	Conjunction	K distributes over it
↔	U+2194	Biconditional	Distribution law for K
→	U+2192	Material conditional	Factivity and introspection axioms
⊆	U+2286	Inclusion	int(A) ⊆ A gives factivity
∘	U+2218	Composition	cl∘int, the belief operator
ℝ	U+211D	Real line	A space validating exactly S4
ℚ	U+211A	Rationals	Another space validating exactly S4

Metatheory. McKinsey and Tarski (1944) proved S4 sound and complete for the interior semantics over the class of all topological spaces, and complete already for any dense-in-itself metric space, so ℝ, ℚ and the Cantor set each validate exactly S4. Relational semantics is a special case: Alexandrov spaces, where arbitrary intersections of opens are open, correspond exactly to preorders, and every S4 Kripke frame is one — general topologies are strictly more permissive, which is what makes the interior reading a generalization rather than a rephrasing. Interpreting belief as cl∘int, Baltag, Bezhanishvili, Özgün and Smets showed the KD45 laws hold exactly on extremally disconnected spaces, those in which the closure of every open set is open, and that on that class knowledge validates S4.2 rather than S5 — the topological counterpart of Stalnaker's result that strong belief forces knowledge to be S4.2. Moss and Parikh's subset space logic, with a second modality for shrinking neighbourhoods, is decidable and finitely axiomatized (Dabrowski, Moss and Parikh, 1996).

Applies to. Epistemology. Modal logic. Learning theory. Formal verification.

Limitations. Knowledge as interior is S4, so positive introspection is imposed and negative introspection is unavailable: ¬Kφ → K¬Kφ would require the complement of every open set to be open, forcing a partition topology in which opens and closeds coincide, the interior operator is just a relational S5 modality, and the topological structure does no work. Factivity holds by construction, since int(A) ⊆ A in every space, so the account can represent no false knowledge and says nothing about justified false belief. Logical omniscience persists — interior distributes over finite intersection, so knowledge is closed under conjunction and under consequence. Which topology represents an agent's evidence is settled outside the theory, and the choice determines the logic: extremal disconnectedness is stipulated to obtain KD45 belief, not derived. And the knowability principle still collapses by Fitch's argument unless ◇ is reread as a dynamic operator over evidence-gathering, which changes the language rather than answering within it.