BHK Interpretation
Origin. Brouwer, Heyting, Kolmogorov (1920s-1930s). Proofs as constructions. Meaning-as-use for connectives. Foundation for intuitionistic logic semantics. Basis for Curry-Howard.
Models. Propositions interpreted by proof conditions. A proof of A → B transforms proofs of A to proofs of B. Constructive existence: provide witness. No truth independent of proof.
Formalism.
Conjunction: A proof of A ∧ B is a pair ⟨p, q⟩ where
- p is a proof of A
- q is a proof of B
Disjunction: A proof of A ∨ B is a pair ⟨i, p⟩ where
- i ∈ {0, 1} indicates which disjunct
- if i=0: p is a proof of A
- if i=1: p is a proof of B
Implication: A proof of A → B is a construction f that transforms any proof of A into a proof of B.
Negation: ¬A ≡ A → ⊥ A proof of ¬A transforms proofs of A to absurdity.
Universal: A proof of ∀x.A(x) is a construction f that given any object a, yields a proof of A(a).
Existential: A proof of ∃x.A(x) is a pair ⟨a, p⟩ where
- a is an object (witness)
- p is a proof of A(a)
Absurdity: ⊥ has no proof. From ⊥, anything follows (ex falso).
Non-validity: A ∨ ¬A: requires deciding A constructively. ¬¬A → A: double negation elimination fails.
Symbols.
| Symbol | Unicode | Name | Meaning |
|---|---|---|---|
| ⟨ ⟩ | U+27E8, U+27E9 | Angle brackets / Pairing | ⟨p, q⟩ proves a conjunction; ⟨a, p⟩ proves an existential |
| π | U+03C0 | Pi / Projection | π₁ and π₂ recover the components of a pair |
| ∧ | U+2227 | Conjunction | Proved by a pair of proofs |
| ∨ | U+2228 | Disjunction | Proved by a tag naming a disjunct plus a proof of it |
| → | U+2192 | Implication | Proved by a construction sending proofs to proofs |
| ¬ | U+00AC | Negation | Not primitive: ¬A abbreviates A → ⊥ |
| ⊥ | U+22A5 | Absurdity | The proposition stipulated to have no proof |
| ∀ | U+2200 | Universal quantifier | Proved by a construction defined on every object |
| ∃ | U+2203 | Existential quantifier | Proved by a witness together with a proof about it |
| ∈ | U+2208 | Element of | The disjunction tag ranges over i ∈ {0, 1} |
| ≡ | U+2261 | Identical to / Definition | Definitional abbreviation, as in ¬A ≡ A → ⊥ |
| λ | U+03BB | Lambda | The Curry–Howard term-former matching → |
| f | — | Construction | The primitive the interpretation leaves unexplained |
Metatheory. BHK stipulates what counts as a proof rather than assigning truth values, so its content is fixed only when "construction" is made precise, and each precisification is a theorem with its own reach. Kleene's number realizability (1945) reads constructions as indices of partial recursive functions and validates Heyting arithmetic — but it validates more besides, since Church's thesis CT₀ and Markov's principle are realizable and not intuitionistically provable, so realizability strengthens the interpretation rather than merely rendering it. Under Curry–Howard the clauses become typing rules: pairing for ∧, injection for ∨, abstraction for →, and dependent sums and products for ∃ and ∀, which is what Martin-Löf's type theory adopts as a definition. Kreisel's theory of constructions attempted a direct formalization and required a second clause, demanding that the construction come with a proof that it does what is claimed. The predicted properties do hold: Heyting arithmetic has the disjunction and existence properties.
Applies to. Intuitionistic foundations. Type theory semantics. Constructive mathematics. Proof assistants. Program extraction.
Limitations. The interpretation is circular where it matters most: "a construction transforming any proof of A into a proof of B" presupposes the notion of proof it purports to explain, and quantifies over the unbounded totality of proofs of A. Making that precise forces the second clause — the construction must be accompanied by a proof that it works — and the attempts by Kreisel and Goodman to formalize it did not settle the matter. No precisification is neutral: realizability validates Church's thesis and Markov's principle, Kripke semantics validates neither, and both are complete for intuitionistic logic, so BHK underdetermines its own semantics. The clauses say nothing about ⊥ beyond stipulating that it has no proof, so ex falso is imposed rather than derived — which is why minimal logic remains a coherent alternative reading of the same clauses.
© 2026 Lingenic LLC