You are on page 1of 6

Proof-détours in classical logic

Adrian Rezuş
July 16–20, 2017; rev. October 31, 2017

i

c 2017 Adrian Rezuş (Nijmegen, The Netherlands)

c 2017 équivalences (Nijmegen, The Netherlands) [(pdf)LATEX]

draft: July 16–20, 2017; revised: October 31, 2017


printed in the netherlands
In what follows, if not already standard – Barendregt (1981, 1984), Baader &
Nipkow (1998, 1999), TeReSe (2003) – , the terminology is as in Rezuş (2017b).
The proofs of some statements can be found in Rezuş (2017, 2017a).
§1 An ‘extended’ λ-calculus [Rezuş (2017b)]. In the background, we have an
abstract grammar with two syntactic categories – scalars and witness terms –
that are both atomic and substitutive (in the sense of Haskell B. Curry). The
inner structure of the scalars (the scalar ‘constructors’) can be ignored. The
régime of metavariables is fixed by:

atomic scalars :: u, v, w
scalar terms :: s, t := u | ...
atomic witness terms :: x, y, z
witness terms :: a, b, c, d, e, f

while the witness terms (w-terms, for short) are given explicitly by:
R
a, b, c, d, e, f := x | c ? a | ∂z.e | ≺a,c | (x,z).e | ↓t (c) | Σ(u,z).e.
R
Let ∂ Σ be the following scalar mono-fold based on [∂,?], with ‘character-
istic’ equations (‘postulates’):

(β∂) ` c ? (∂z.e[z]) = e[z:=c],


(η∂) ` ∂z.(z?a) = a (z not free in a),
R R
(β ) ` ≺a,b ? ( (x,y).c[x,y]) = c[x:=a,y:=b],
R R
(η ) ` (x,y).(≺x,y ? c) = c (x,y not free in c),

(βΣ) ` ↓t (a) ? (Σ(u,x).c[u,x]) = c[u:=t,x:=a]),


(ηΣ) ` Σ(u,x).(↓u (x) ? c) = c (u, x not free in c),
R R R
and ∂ Σζ := ∂ Σ + (ζ ) + (ζΣ), where
R R
(ζ ) ` ∂z.e[z] = (x,y).e[z:=≺x,y] (x, y fresh for e[z]),
(ζΣ) ` ∂z.e[z] = Σ(u,x).e[z:=↓u (x)] (u, x fresh for e[z]).

So, beyond the underlying mono-fold ∂ based on [∂,?], characterised by (β∂)


and (η∂) (a copy of the pure extensional λ-calculus), we have two kinds of pairs
R(term-forms: ≺a,c and ↓t (c) resp.), and two dyadic abstractors (term-forms:
(x,z).e and Σ(u,z).e, resp.).1 The compatibility conditions are as expected.
1 Note that the scalar syntax replicates a segment of the ‘pure’ syntax. Syntactically, one

can unify the notation for the pairs and the dyadic ‘split’-abstractors
R resp., by writing <a,c>
≡ ≺a,c, and <t,c> ≡ ↓t (c), resp., as well as S(x,z).e ≡ (x,z).e, and S(u,z).e ≡ Σ(u,z).e,
resp., since [1] the declarations of the metavariables make things notationally unambiguous,
and [2] the equational behaviour of the (pair-split) teams is ‘isomorphic’, so to speak.

1
R R
From Rezuş (2017b) we know that ∂ Σζ (resp. its ζ-freeR subsystem ∂ Σ)
is consistent, qua equational theory – formally: Cons(∂ Σ(ζ) ) –, since it is
(properly) contained in ∂πΠ, the scalar extension of the cartesian fold ∂π.2
R
§2 Reduction. Let us focusR on the ζ-free system ∂ Σ first. As expected, the
conversion (equality) of ∂ Σ can be generated by a ‘notion of reduction’ –
terminology as in Barendregt (1981, 1984) – by orienting the equalities above
from left to right.
Let 7→ be the associated notion of reduction, 7−→ be its compatible closure,
R closure of 7−→. (Usually,  is called ‘reduc-
and  be the reflexive-transitive
tion’.) Then the equality of ∂ Σ is the symmetric closure of  and we have
also the expected
R
Theorem. The notion of reduction of ∂ Σ (i.e., the relation 7→) is confluent.
Proof. As ever. [Hint. Examine critical pairs – Baader & Nipkow (1998, 1999),
TeReSe (2003), etc. – or else use the ‘residuation-method’ of ‘parallel’ reduc-
tions, as in Takahashi (1989, 1995).] 
From this, we have, once more, a
R
Corollary. Cons(∂ Σ).
§3 Art déco (‘typing’). We revert to ‘typing’ next. Our ‘types’ are formulas
of classical first-order logic, otherwise just decorations on witness terms. (The
‘typing’ is rigid, à la Church, thus.) The scalars are supposed to represent
‘individual’ terms (as mentioned before, the inner structure of the scalars – to
be specified explicitly for any particular first-order theory – can be ignored).
The underlying first-order logic signature is based on [f,¬,→,∀], i.e., falsum,
(classical) negation, material implication and the (classical) universal quantifier,
i.e., we have an abstract grammar
atoms :: P[t] (where t is a finite sequence of scalar terms),
formulas :: A, B, C, ... := f | P[t] | ¬A | A→C | ∀u.C[u].
¯
We write, for convenience A9C :=R ¬(A→C) and ∀u.A[u] := ¬(∀u.A[u]). The
resulting witness theory is called ∂ Σ[f,¬,→,∀].
§3.1 The primitive rules of inference. Where Γ̂ stands for an arbitrary
R proof-
context, the witness operators (the primitive rules of inference) of ∂ Σ[f,¬,→,∀]
are:
(∂) Γ̂ ` ∂x:¬C.e[x] : C, if Γ̂[x:¬C] ` e[x] : f [reductio ad absurdum],
(?) Γ̂ ` c ? a : f, if Γ̂ ` c : ¬C, Γ̂ ` a : C [‘law of non-contradiction’],
R R
( ) Γ̂ ` (x:A,y:¬C).e[x,y] : A→C, if Γ̂[x:A][y:¬C] ` e[x,y] : f,
(π) Γ̂ ` ≺a,c : A9C, if Γ̂ ` a : A, and Γ̂ ` c : ¬C,
(Σ) Γ̂ ` Σ(u,x:¬C[u]).e[u,x] : ∀u.C[u], if Γ̂[u][x:¬C ` e[u,x] : f,
¯
(Π) Γ̂ ` ↓t (c) : ∀u.C[u], if Γ̂ ` c : ¬C[u:=t].
2 Note,
R
however, that although ∂ Σ(ζ) is a sub-cartesian fold, it in not cartesian (we can
simulate projections for the primitive pair ≺...,..., yet the resulting pairing is not ‘surjective’).

2
§3.2 The classical ‘détour eliminations’. Under the decoration above,
R the prim-
itive βη-conditions – meant to define the witness operators of ∂ Σ[f,¬,→,∀] –
state proper détour eliminations, if viewed as reduction ‘rules’. In the spirit of
Chrysippus, the founder of classical logic – cf. Rezuş (2009) –, a proof-détour
(formally: a ‘redex’) points out to a way of coping with a contradiction.
R
§3.3 The ζ-rules. What about the ζ-conditions of ∂ Σζ [f,¬,→,∀]? Consider
the decorated witness terms (∂-abstractions) of the form a := ∂z: ¬C.e[z] (where
e[z] : f, for [z: ¬C]), and call them reductio-terms. If C is an atom or f then
a is said to be atomic, otherwise it is complex. A witness term containing no
complex
R reductio-terms is said to be déco-normal [d-normal, for short]. Finally,
let ∂ Σ? [f,¬,→,∀] be the d-normal fragment of ∂ Σζ [f,¬,→,∀].
R

If we are reading the ζ-conditions as ‘(improper) reductions’ – oriented from


left to right –, they can serve to eliminate (recursively) complex reductio-terms
in favour of atomic ones:
R R
(ζ ) ` ∂z:(A9C).e[z] 7→ζ (x:A,y:¬C).e[z:=≺x,z], x,y fresh for e,
¯
(ζΣ) ` ∂z:(∀u.C[u]).e[z] 7→ζ Σ(u,x:¬C).e[z:=↓u (x)], u,x fresh for e.

It is then obvious that we can always apply the ζ-rules first, before attempt-
ing to apply any proper reduction ‘rule’, i.e., we have the rather trivial
R
Theorem. The witness terms of ∂ Σζ [f,¬,→,∀] are d-normalisable.
Proof. Straightforward. 
Finally, the reader can establish by herself the expected
Theorem. In ∂ Σ? [f,¬,→,∀], all witness terms are bounded.
R

Proof. Mutatis mutandis, as, e.g., in Rezuş (1990). Cf. also López-Escobar
(1990).3 
In other words, the (decorated) witness terms of ∂ Σ? [f,¬,→,∀] are strongly
R

normalisable.
§4 Remarks. (1) Our use of ζ-conditions parallels closely the use of the cor-
responding (ζγ)-conditions in λγ-calculi – Rezuş (1990) –, as first suggested
by Dag Prawitz’s in his PhD Diss. (Stockholm 1965). Cf. also with the Kol-
mogorov β-only calculi of Rezuş (2017c). (2) Incidentally, in the above one could
have also used an alternative ‘classical’ decoration based on the primitive sig-
nature [f,M,∀], where M stands for the ‘Sheffer functor’ nand (incompatibility),
and ¬A := (A M A). (3) Since the methods of proof hinted at here are rather
standard, the reader can easily establish analogous results for the
R corresponding
witness theories based on di-folds, equationally equivalent to ∂ Σ(ζ )[f,¬,→,∀],
described in Rezuş (2017b). (4) For a recent update on the earliest history of
normalisation proofs for classical logic, see Rezuş (2017d).
§5 Acknowledgement. The author is indebted to J. Roger Hindley (Swansea,
Wales, UK) for comments on a previous draft of this note.
3 As a matter of fact, in view of the confluence result mentioned earlier, it is enough to

establish weak normalisability for the corresponding terms.

3
References
[1] Franz Baader, and Tobias Nipkow (1998, 1999) Term Rewriting and All That,
Cambridge University Press, Cambridge 1998 [first edition], 1999 [paperback].
[2] Henk Barendregt (1981, 1984) The Lambda Calculus. Its Syntax and Semantics,
North Holland, Amsterdam etc. 1981 [first edition], 1984 [second edition].
[3] Edgar George Kenneth López-Escobar (1990) Remarks on the Church-Rosser property,
The Journal of Symbolic Logic 55 (1), [March] 1990, pp. 106–112.
[4] Dag Prawitz 1965 Natural Deduction, A Proof-Theoretical Study Almqvist & Wik-
sell, Uppsala etc. 1965. [Acta Universitatis Stockholmiensis: Stockholm Studies in Phi-
losophy 3]. (Originally PhD Diss., Stockholm 1965.)
[5] Adrian Rezuş (1990) Classical Proofs. λ-calculus Methods in Elementary Proof The-
ory [Lecture Notes, University of Nijmegen 1986–1987], Nijmegen [August 5] 1990 [un-
published]. (Preprint available online @ www.equivalences.org.)
[6] — (2009) An Ancient Logic (Chrysippus and His Modern Readers I), LAP – Lambert
Academic Publishing, Saarbrücken [December] 2016 [isbn: 978-3-330-01661-3].
[7] — (2017) Lukasiewicz, Jaśkowski and Natural Deduction (Curry-Howard for Classical
Logic), Nijmegen, April 2017 [forthcoming].
[8] — (2017a) Witness theory for classical logic (Inferential systems), Nijmegen, April 2017
[forthcoming].
[9] — (2017b) Cartesian folds and witness theory, Nijmegen, June 2017 [forthcoming].
[10] — (2017c) On the Kolmogorov λγ-calculus, Nijmegen, June 2017 [forthcoming].
[11] — (2017d) Review of: Jan von Plato Saved from the Cellar. Gerhard Gentzen’s Short-
hand Notes on Logic and Foundations of Mathematics, Springer International Publish-
ing [CH], [April] 2017 [isbn: 978-3-319-42119-3] [Sources and Studies in the History of
Mathemarics and Physical Sciences], Studia Logica [forthcoming].
[12] Masako Takahashi (1989) Parallel reductions in lambda-calculus, Journal of Symbolic
Computation 7 (2), [February] 1989, pp. 113–123.
[13] — (1995) Parallel reductions in lambda-calculus, Information and Computation
118 (1), [April] 1995, pp. 120–127.
[14] TeReSe [Marc Bezem, Jan Willem Klop, and Roel de Vrijer (eds.)] (2003) Term
Rewriting Systems, Cambridge University Press, Cambridge 2003 [Cambridge Tracts
in Theoretical Computer Science 55].

You might also like