SINGLE AXIOMS FOR THE CLASSICAL EOUIVALENTIAL CALCULUS

J. G. Peterson
(received 26 January, 1977)

1. INTRODUCTION. In the classical equivalential calculus EC(also

called Mtwo-valued equivalential calculus” [3] and "two-valued E-puren
[6]) we are concerned with those formulas of the propositional calculus
in which E is the sole connective, and which hold true when E is inter­
preted as classical equivalence (for a more precise formulation see §5).

This system may be equipped with either of two sets of rules of

inference.

By EC with ordinary detachment we shall mean EC with rules of

inference substitution and ordinary detachment (also known as the rule
of detachment or modus ponus).

By EC with reverse detachment we shall mean EC with rules of

inference substitution and reverse detachment (the rule to infer a
from Ea(3 and |3 ).

If no rule of inference is specified then the rules of substitution

and ordinary detachment are understood; if just ordinary or reverse
detachment is specified then the rule of substitution is understood.

The first to recognise that EC with ordinary detachment is

axiomatisable was Lesniewski who published in 1929 [2, p 16] the
following axiom set :
(A) EEEprEqpErq
and EEpEqrEEpqr.

(This paper won first prize in the New Zealand Mathematical Society essay
contest for students, 1975)

Math. Chronicle 6(1977) 21-47.

21
 After Lesniewski, Wajsberg I11J in 1932 published four simpler
axiom sets for this system. To Wajsberg belongs the credit of showing
that the system could be based upon a single axiom, as two of his four
sets of axioms contained only one member of fifteen letters apiece.

We will concern ourselves with the shortest possible single axioms

for EC either with ordinary or with reverse detachment.

In a paper published in 1939 [3] Lukasiewicz solved the problem of

the length of such an axiom in the case of ordinary detachment, showing
that :
Cl) EEpqEErqEpr
is a single axiom for EC with ordinary detachment and that no shorter
formula possesses this property. He also showed that each of '
•
(2) EEpqEEprErq
and (3) EEpqEErpEqr
is a single axiom for EC with ordinary detachment by deriving from each
of these formulas the axiom (1). The reverse derivation relies on the
completeness of the system proved by Lesniewski [2], -Lukasiewicz himself
proved the Post-completeness of the system which, with some further
argument, also suffices.

Lukasiewicz was credited with having shown that no other formula

with eleven letters is a single axiom for EC with ordinary detachment
but this is not so. In 1963 Meredith and Prior [6 ] showed that (3) can
be derived by substitution and ordinary detachment from :
(4) EEEpqrEqErp
and (5) EpEEqEprErq.
Again the reverse deduction rests on the completeness property.
Meredith and Prior claimed further, without proof, that :
(6 ) EpEEqErpEqr
(7) EEpEqrErEpq
(8 ) EEpqErEEqrp
(9') EEpqErEErqq
 (10) EEEpEqrrEqp
and (11) EEEpEqrpErp
are also single axioms for EC with ordinary detachment.

Prior [8 , Appendix I p307] later corrected (9r) to read

(9) EEpqErEErqp.

The above paper of Meredith and Prior concluded by showing that

from (4) using substitution and reverse detachment, ordinary detachment
could be obtained. That is, (4) remains a single axiom for EC when
ordinary detachment is replaced by reverse detachment.

The object of this paper is to completely answer the question which

of (l)-CH) is a single axiom for EC with ordinary detachment or with
reverse detachment. In doing so a self-contained treatment of the
equivalential calculus is provided.

In §2 an economical proof is provided that the formulas (l)-(5),

(7), (9) and (11) above are deductively equivalent under ordinary
detachment (cf [4, p262]) in the sense that from any one of them the
others may be derived by substitution and ordinary detachment. It is
then shown that these formulas are deductively equivalent under ordinary
detachment to the original axiom scheme (A) of Lesniewski, and hence
that each is a single axiom for EC with ordinary detachment.

The notion of a dual is then defined and used to show that each of
(8 ) and (10) is also an axiom for EC with ordinary detachment.

In §3 the assertion of Meredith regarding (6 ) is shown to be

incorrect : that in fact (6 ) is not an axiom for EC with ordinary
detachment.
In §4 it is shown that when ordinary detachment is replaced by
reverse detachment, (4), (8 )-(11) remain single axioms and are in
fact the shortest single axioms for EC with reverse detachment.
Furthermore an outline is given of the proof that (l)-(3), (5)-(7) are
not single axioms for EC with reverse detachment.

23
 In §5 a more precise formulation of the equivalential calculus is
given than that appearing at the beginning of this introduction. A
completeness theorem is proved showing that the theorems of the
equivalential calculus are precisely those formulas which take the
value 1 whenever the variables are interpreted as members of a group
in which every element is its own inverse.

The derivations given in this paper are simplified by the use of

Meredithrs condensed detachment operator D [8 , Appendix II, pp318-319].
These derivations were found with the aid of a computer program for
automatic theorem-proving based on iteration of an algorithm for the
above D. Comments on the application of automatic theorem-proving
techniques to this work appear in §6 .

Finally in §7, the main results of the paper are summarised, and
mention is given to some outstanding problems concerning the equivalential
calculus and possible development of the application of the computer to
it .*

0)-( 5 ) , (7}-(ll) with ordinary detachment. We prove that (l)-(5),

(7), (9), (11) are deductively equivalent in the following way :
( 1 M 7 M 1 1 M 5 M 4 M 3 M 9 M 2 M 1 ) .

(1)=H7) : 1. EEpqEErqEpr
2. EEpEEarEsqEEsrp = Dl.l
. 3. EEpqEpq = D2.1

* Proofs of those results in §§2-4 for which proofs had not

previously been published, i.e. that (7)-(11) are single axioms for
EC with ordinary detachment, that (6 ) is not, and that (7)-(11) are
single axioms for EC with reverse detachment, have appeared in : J.G.
Peterson, Shortest Single Axioms for the Classical Equivalential
Calculus j Notre Dame Journal of Formal Logic, 27(1976), 257-271.
 4. EEpqEEprEEsqErs; = D2.2
5. EEpEqrEEqrp = D1.3
6. EEEpqrEErpq = D2.4
7. EEEpqErpErq = D5.1
3. EEEpqrErEpq = D5.5
9. EEpEEqrsEEsEEtrEqtp = D1.2
10. EEEEpqErpEEsrEtsEtq = D9.4
11. EEEEpqrrEEsqEps = D6.2
12. EpEqEpq = DIO.11
13. EEEEpqrEErpqEEsEtuEEtus = DD12.6.5
14. EEpEqrEEpqr = D7.13
15. EEEEpqrErEpqEEsEtuEEstu = DD12.8.14
★ 16. EEpEqrErEpq = D7.15

: 1. EEpEqrErEpq
2. EEpqEEpEqrr = Dl.l
3. EpEEqrEqErp = D1.2
4. EEpqEpEqEErEstEtErs = D3.1
5. EEpEEEqrEqErpss = D2.3
6. EEpqEpEqEErsEErEstt = D3.2
7. EEpEEEqrEqErpsEsEEtuEEtEuw = D6.5
8. EEpEEqrEEqErssEEtpt = D1.6
9. EEpEEqpqEErsEErEstt = D7.2
10. EpEEqpq = D5.2
11. EEEpEqrrEEsEEtstEpq = D1.9
12. EEpEEqpqEErsEsr = Dll.8
13. EEpqEqp = D12.10
14. EEpEqrEqErp = D13.1
15. EEpqEEqErpr = D1.14
* 16. EEEpEqrqErp = D13.15.

25
CU>C*] 1 . EEEpEqrqErp
2. EpEqEEErqpr = Dl.l
3. EpEEEqpEEErEstsEtrq = D2.1
4. EEEpEEEqErsrEsqEEEtEuvuEvtp = D3.1
5. EEpqEEEqErprEEEsEtutEus = D1.4
6. EEEEpEqrqErpEEEsEtutEus = D4.4
7. EEEpqErEEqEspsr = D4.5
8. EEEpEEEqprqEsrs = D7.5
9. EEpEqEErsEpEErEsrsq = D7.6
10. EEpEqEEErqprEEEsEtutEus = D3.3
11. EEEEpEEEqErsrEsqtpt = Dl.lO
12. EEEEpEqEEErqprEEsEtutEvEusv = D9.10
13. EEEpEEqErsrEsqp = D12.1
14. EEEEpEqrqEEEsEtutEusErp = D13.ll
15. EEpEqEErEEEsrtsEptq = D7.10
16. EEEEpEqEEErqprsEtEuEEEvusvt = D15.10
17. EEEpqErEEEsrqsp = D16.1
18. EEEEpqrpErq = D17.1
19. EpEEqEprErq = D14.18

The last two proofs establish the assertion, unproven in the

literature, that (7) and (11) are single axioms for EC with ordinary
detachment.

(5)=>(4) : 1. EpEEqEprErq
2. EEpEEqEErEqsEsrtEtp = Dl.l
* 3. EEEpqrEqErp = D2.1.

(4)=>(3) : 1. EEEpqrEqErp
2. ErEEqErpEpq = Dl.l
3. EEsEEEEpqrEqErptEts = D2.1
4. EEErpEEpqrq = D3.2
5. EEEpqqp = D4.2
6. EqEpEpq = D1.5

26
 7. EErEsEsrEqEpEpq = DD6 .6 . 6
8. EEsEsrEEqEpEpqr = D1.7
9. EEsrEEEaEpEpqrs = D1.8
10. EEEqEpEpqEEEqEpEpqrsEsr = D9.9
11. EEsEtEtsEEErpEEpqrq = DD6.4.6
12. EqEEEpEprqr = DIO.11
13. EEqEErEEEsEstrtpEpq = D2.12
14. EEsEEqEEEpEprqrtEts = D13.2
15. EEsEsEEpqrEqErp = D14.1
16. EEsEEpqrEEqErps = D1.15
17. EEEprEqpErq = D16.14
18. EEpqEErpEqr = D1.17.

The last two proofs are those published by Meredith [6] and
reproduced here for completeness.

1. EEpqEErpEqr
2. EEpEqrEEEsqErsp = Dl.l
3. EEEpEqrEEsqpErs = D2.1
4. EEpEEqErsEEtrqEEstp = D1.3
5. EpEqEpq = D3.2
6. EEEpqEErqpr = D2.5
7. EpEEqErEqrp = D5.5
8. Epp = D6.2
9. EEEpqrEErqp = D4.7
10. EEpqEqp = D1.8
11. EEpEEpqrErq = D9.6
12. EEpqErEErqp = DIO.11

1. EEpqErEErqp
2. EpEEpEqEEqrsEsr = Dl.l
3. EpEEpEEqErEErstEtsq = D1.2
4. EEEEpqErEErqpEsEEstuEut = D2.1
5. EEEEpqErEErqpEEsEtEEtuvEvus = D3.1
 6. EEEEEpqrqpr = D4.4
7. EEEEEEEpqrqprEsEEstuEut = D2.6
8. EpEEpqEEEErsqsr = D1 . 6
9. EpEEpEEqrEEEEstrtsq = D1 .8
10. EEEEpqErEErqpEEstEEEEuvtvus = D9.1 •
11. EpEEpEqrEEEEEEstutsuEvEEvrq = D1.7
12. EEEEpEEpqrsErqs = D5.7
13. EEEpqEEEErspsrq = DIO.7
14. EEpEEpqErqr = D7.ll
15. EEEpqpq = D14.1
16. EEpEEpqrErq = D5.14
17. EEEpEEpqrsEErqs = 012.14
18. EEpqEEEqrpr = D13.16
19. EEpEEpEEqrsaEsr = D5.18
20. EEEpqErEErsEpsq = D18.14
21. EEEpqEErprq = 018.15
22. EEpqEEEEErqpsrs = D17.18
23. EEpqEqp = D21.16
24. EEpEEqrErpq = D21.19
25. EEEpEEpqErqsEsr = 020.24
26. EEEEpqprErq = 021.24
27. EEEpqrF.pErq = 024.22
28. EEpqEEprEqr = D26.25
29. EEEpqErqEpr = D23.28
30. EEpqEEprErq = 027.29

This proof establishes that (9) is a single axiom for EC with

ordinary detachment as claimed by Meredith [6 ].

(2)=»(1) : 1. EEpqEEprErq
2. EEEpqrErEEpsEsq = Dl.l
3. EEpEEqrErsEEEqstEtp = D2.2
4. EEEpqrErEpq = D3.1

28
 5. EEEEpqrsEsErEpq = D1.4
6. EEEpqEqrEpr = D4.1
* 7. EEpqEErqEpr = D5.6.

This is the derivation given by Lukasiewicz [3], slightly

improved.

It is now established that (l)-(5), (?) > (.9) and (11) are
deductively equivalent. We show that they are deductively equivalent
to the original axiom scheme (A) of Lesniewski and hence are single
axioms for EC with ordinary detachment.

1. EEpqEErqEpr
2. EEpEEqrEsqEEsrp = Dl.l
3. EEpqEpq = D2.1

CaI
li
4. EnpEqrEEqrp o

5. EEEpqErpEra = D4.1
6. EEEpqrEErpq = D2.D2.2
7. Epp = D5.3
8. EEpqEqp = D1.7
9. EEpEqrEErqp = D1.8
10. EEEpqrErEqp = DS.9
11. EEEpqrEpEqr = DIO.6
12. EEpEqrEEpqr = D8.ll
13. EEEprEqpErq = DlO.l.

This proof slightly condenses the relevant portions of the proof

given by Lukasiewicz [3].

(A)=>(3) : 1. EEpEqrEEpqr
2. EEEprEqpErq
3. EEpqEqp = D2.1
* 4. EEpqEErpEqr = D3.2.
We may now use the fact that (9) and (11) are single axioms for EC
with ordinary detachment, and the duality notions defined below, to prove
that (8 ) and (10) are single axioms for EC with ordinary detacliment.

29
Definition. By the length, of a formula a , denoted Z(cO, we mean
the number of primitive symbols in a.

Definition. The dual of a formula a is defined by induction

on the length of a :

(i) if Z-(a) = 1 then = a

(ii) if Z(a.) > 1 then a = E(?y for some unique formulas

8 and y and ad = Ey^

Definition. We call the rule : from (3 and Ea(3 deduce a, the rule
of reverse detachment.

Lemma, a is an axiom for EC with ordinary detachment iff is an

axiom for EC with reverse detachment.

Proof. It is easily seen that IB is derivable from a by ordinary

detachment iff is derivable from a1^ by reverse detachment. The
lemma is proved by noting (§5 Corollary 3) that {3 belongs to EC iff
P belongs to EC.

From the lemma, (8 )= (11)^ and (10) = (9)*^ are single axioms for EC
with reverse detachment since (11) and (9) are for EC with ordinary
detachment. We now show that from each of (S) and (10) with ordinary
detachment we may deduce reverse detachment.

We start with (8 ), Ea(3 and 3 :

1_ EEpqErEEqrp
2. EaB
3. P
4. EpEEEqEErqspEsr = Dl.l
5. EEEpEEqprEsEEEtEEutvsEvuErq = D4.4
6. EEEpEqErprq = D5.1
7. EpEEqpEErEqEsrs = D1.6
8. EEppEEqEpErqr = D7.3
 9. EEpEaEqpq = D8.2
10. a = D6.2,

and derive a, i.e. reverse detachment holds.

We start with CIO), Ecc(3 and (3

1. EEEpEqrrEqp
2. Eccj3
3. P
4. EpEqErEpErq = Dl.l
5. EEpEqErEpErqEsEtEuEsEut = DDDD4.4
6. EEpEpqq = D1.5
7. EpEqEpq = D6.4
8. EEpEqEpqEcc(3 = DD7.7.2
oo
a
I—1
9. Epa
ti

10. a = D9.3,

and derive a , i.e. reverse detachment holds.

Thus every formula which may be derived from (8 ) or from (10) using
reverse detachment may be derived from it using ordinary detachment.
Since (8 ) and (10) are single axioms for EC with reverse detachment,
this proves that (8 ) and (10) are single axioms for EC with ordinary
detachment.
We have therefore shown that each of (l)-(5), (7)-(11) is a single
axiom for EC with ordinary detachment.

3. (6) with ordinary detachment. It can be shown that the formula (6 )

is not a single axiom for EC with ordinary detachment as was asserted
by Meredith and Prior [6 ]. The proof given here exhibits a logical
matrix in which (6 ) holds but (1) does not. This concept of validity
ander logical matrices is well established in the literature, but for
the purpose of this proof the necessary results are summarised below.

31
Definition. By a groupoid we mean a pair <£,#> where. G is a
non-empty set and g: <x3]f>- f-+ gxy is a binary operation on G.

Thus <F,E> is a groupoid where F is the set of formulas of

the equivalential calculus and E is defined by E: <a,j3>f-vEap.

Definition. An endomorphism of the groupoid <F,E> is called a

substitution.

Definition. We say a set D c F is closed under substitution if

o(D) c D for every substitution a , and closed under ordinary
detachment if c € D and Ea(3 € D => p € D.

Definition. By a matrix we mean a triple <G,T,g> with <G3g> a

groupoid and T c G.

A matrix <G,T,g> is normal if and gxy € F =» u £ V.

We say a formula a holds (or is valid) in <G,T,g> if
h{_a) 6 F for every homomorphism h: <F,E> -> <G,g>.

Ir is clear that the set of formulas which hold in a matrix is

closed under substitution, and that if the matrix is normal, the set
of formulas which hold in it is also closed under ordinary detachment.
We may thus conclude that if a formula holds in a normal matrix, then
all formulas which may be derived from it by applications of substitution
ana ordinary detachment also hold in this matrix. Therefore if (6 )
holds in a normal matrix in which (1 ) does not hold, (1 ) is not derivable
from (6 ) and so (6 ) is not a single axiom for EC with ordinary detachment.

Consider the normal matrix <{0,1,2}, {0} , g> where g is

defined by the table :
9 0 1 2

0 Q 2 1

1 1 0 2

2 2 1 0 .

32
 This table defines right subtraction in th.e abelian group 7L 3

so we may write Qc-y] for gxzf where: € {0 ,1 ,2 }.

Let 7i be a homomorphism from <F,E> to <{0,1,2},^> then

S.(EpEEaErpEqr)
= Ti(EpEEqErpEqr)
= gh Cp)ggh (q)gh (r)htp)gh {cfiaCr)
= (p) -h (q)Cr) ~h (p) +?i(q) -fa(r) ]
= 0 using results from the abelian group Z3 .

Thus (6 ) holds in the above matrix.

However if h: <F,E> -+ <{0,1,2}, g> is a homomorphism such that

Mp) = faC°l) = 0 and h Cr) = 1 then

h (EEpqEErqEpr)
= ^(p)faCq)^(r)?:Cq)9rfa(p)fa(r)
= gg 0 0 g g l O g O l
=g 0 g 1 2
=g 0 2

=1*0.

Thus Cl) does not hold in the matrix as required.

4.(1)-(T1) with reverse detachment. It can be verified that (4) = C4)d ,

(8 )=Cll)d, (9 )=(10 )d , (10 )=(9 )d and Cll)=(S)d . Thus the duals of
(4), (8 )-Cll) are single axioms for EC with ordinary detachment. From
the lemma in §2 it follows that (4), (8 )-(ll) are single axioms
for EC with reverse detachments

Note that the above proof in respect of (4) is a simpler

alternative to Meredith's proof [6 ], using only an observation about
the structure of the axiom.

By the lemma of §2 , in order to show that none of C1)~C3), (5)-C7)

is a single axiom for EC with reverse detachment, it will be enough to

33
show that the duals of those egressions are not single axioms for EC
with ordinary-detachment.

In order to do this, use is -made of a result of J. Kalman Cl],

that every formula derivable from an axiom set by substitution and
ordinary detachment is a substitution instance of a formula derivable
from that axiom set by condensed detachment. By §5 Corollary 2
(Characterisation of Theorems) it is clear that the formula :
(a) EEppEqq
is not a substitution instance of another formula in EC up to isomorphism.
Thus if (a) cannot be derived by condensed detachment from a formula in
2C, that formula is not an axiom for EC with ordinary detachment.

Now an algorithm exists for condensed detachment (see §6 for a

brief description of this algorithm in terms of Robinson's unification
algorithm) and in particular it can be determined if condensed
detachment may be applied to any particular ordered pair of formulas.

Applying the algorithm it is easy to see that no condensed de­

tachments can be performed with (1 )^, (5)^ or (6 )^ as sole axiom.
Similarly systematic computer evaluation shows that from (2)^, (3)^
and (7)^ only four, three and seven distinct formulas respectively
can be derived by condensed detachment and in no case is (a) derived.

Thus none of (l)1^, (2)^, (3)^, (5)^, (6 )^, or (7)^ is an axiom

for EC with ordinary detachment as required.

In addition it is clear that (4 )d} (8 )^, (9)^, (10)^ and (11)^

are the shortest possible single axioms for EC with reverse detachment
since, by the lemma, the dual of a shorter single axiom for EC with
reverse detachment would be a shorter single axiom for EC with ordinary
detachment than that given by lukasiewicz.

34
5. The Completeness Theorem.. The classical completeness result for
the equivalential calculus EC Is that a is a theorem of EC if and only
if a holds in the normal matrix for equivalence defined as follows:

M = <{0,1}, {1}, *> where ’ is given by

• 1 0

1 1 0
0 0 1

The result proved here is that a is a theorem of EC if and only

if a holds in the matrix of every Boolean group as defined below.
In proving that this result is equivalent to the classical completeness
result, the characterisation of the theorems of the equivalential
calculus known to Lesniewski [2, p26, p29] is proved.

Definition. A Boolean group is a group in which every element is its

own inverse.

Thus ’ 1 0 is a Boolean group. (In fact, up to

1 1 0
0 0 1

isomorphism, the only two element Boolean group).

Definition. The normal matrix <A, {1}, *>, where <A, •> is a
Boolean group with identity 1 , is called the matrix of the Boolean
group <A, •> .

We denote by M the matrix of the two element Boolean group.

Definition. We call a a Boolean group tautology and write j= a

if a holds in the matrix of every Boolean group in the sense of §3.

We write M 1= a if a holds in M .

35
Definition... We call a a theorem of the equivalential calculus and
write f a if a may Be derived by substitution and detachment from:

CA) EEEprEqpErq
and EEpEqrEEpqr or equivalently from any one of the
formulas 0-)_0>), (7)-Cll).

Lemma 1 . If an algebra <A, g> satisfies the identities

Ci) gxgyz = ggxyz

(ii) ggxxx = x
Ciii) gxx = gyy

then it is a Boolean group.

Proof: <AS g> is a semigroup by Ci) •

We may write gxx=gyy= ...=1 by (iii) whence
gxl=glx=x by (i) and (ii), thus 1 is an identity.

Also by (iii) every element x has an inverse namely x itself.

Thus <A3g> is a Boolean group.

Lemma 2. All Boolean groups are commutative.

Proof: Let <A3 •> be a Boolean group, xty € A then

#
xy=xy1=xyyxyx
=xlxyx
=xxyx
=1yx
=yx .

Lemma 3. (i) |_ EEpEqrEEpr

(ii) |- EEpqEqp
(iii) Epp
(iv) |_ EEpqEEprEqr
Cv) EEpqEErpErq
Cvi) EEppEqq
Cvii) f— EEEpppp

36
 1.. EEpEqrEEpqr CA1
2.. EEEprEqpErq CA1
3. EEpqEErqEpr dl
4. EEpqEqp D2.1
5. Epp D2.4
6. EEEpqErqEpr Dl .3
7. EEpqEEprEqr D4.6
8. EEEpqEprEqr Dl .7
9. EEpqEErpErq D4.8
10. EEEpqqp Dl .4
11. EpEEpqq D4.10
12. EEpEpqq Dl.11
13. EpEqEqp D4.12
14. EEppEqq DD13.5.5.

Note that in substituting p for q in 10 we obtain

* 10'. EEEpppp.

Lemma 4. If I— Ea(3 then h-Efta .

Proof: 1. Eap
2. EEpqEqp Lemma 3(ii)
* 3. Epcc D2.1 .

Theorem 1. (Completeness Theorem) |=a «* |- a .

Proof: ( ') Let h be a homomorphism from <F, E> to <At •> a

Boolean group (cf §3) then
Tz(EEpqEErqEpr) = fo(p)?iCq)frCr)ft(q)?i(p)ft(:r) = 1
by the commutativity and associativity and self inverse property
of <A, •> .

Thus (1) holds in the matrix of any Boolean group.

Now I— a a is derivable from Cl) by substitution and modus ponens,
*» a holds in the matrix of any Boolean group (cf §3) ,
• f» a .

37
 C. =? 1 |jlvre define the. relation.^ qn the s^t of formulas f
of tRe equivalential calculus by*
a ~ p « \— Ea^ .

Then I— Eaa since I— Epp by Lemma 3 (iii)>

I— Ea(3 =» |— E(3a by Lemma 4,
and |— Ea{3 and I— E(3y =* I— Ea(3 and 1— EEa^Eay by Lemma 3(v)
=» I— Eay ,
thus ~ is an equivalence relation.

Also h- Eaj3 =» I— EEyocEy [3 by Lemma 3(v) and

I— EEayEpy by Lemma 3(iv) ,
thus ~ is a congruence relation.

The quotient algebra <F, E>/~ is a Boolean group since it

satisfies (i), (ii) and (iii) of Lemma 1 by (i), (vii) and (vi)
respectively of Lemma 3. The identity of <F, E>/~ is Epp/~ .

Let N aQ then aQ holds in the matrix of any Boolean group,

in particular in the matrix <F/~ , {Epp/~},E> . Thus for any

homomorphism h from <F, E> to <F, E>/~
h(aQ) = Epp/~ .

Let h be the quotient map from <F, E> to <F, E>/~ then
h(a ) = a /~
o o
i.e. Epp ~ <2q

i.e. 1— EEppaQ

but I— Epp by Lemma 2(iii)

thus 1— aQ as required.

Corollary 1. No variable in a occurs an odd number of times => (— a

Proof: If all variables in a occur an even numfier of times and
h:<F, E> -+ <A,g> is a homomorphism to a Boolean group <A,g> then
7i(a) is a product of elements of A such that each element of A
occurring in this product occurs an even number of times. Thus by
the commutativity and self-inverse properties of <A,g>, the elements
of A may be reduced in pairs to 1, giving h(a) = 1.
Thus f= a and so by the completeness theorem (- a as required.
Lemma 5 M a =» no variable occurs in a an odd number of times.

Proof: Let pj occur in a n times where n is odd. • Let <A,•>

be the two element Boolean group and let h from <F,E> to <A,•>
be a homomorphism such that ^t(pi) = 0 and h{p) = 1. p 4 Pi .

Then ft(a) = 0n = 0 since n is odd. Thus M a .

Corollary 2. (Characterisation of Theorems)

|— a ~ no variable occurs in a an odd number of times.

Proof: ( » ) [ — a = » f = a by the completeness theorem,

» M |= a ,
=» no variable occurs in a an odd number of
times by Lemma 5.
( «=» ) By Corollary 1.
Corollary 3. f— a « f- ad .
Proof: By Corollary 2.
Theorem 2. J— a ~ M \= a .

Proof: ( « • ) [ — a = » ) = a by the completeness theorem,

=» M f= a .
C ♦* ) M a =» no variable occurs in a an odd number of
times by Lemma 5,
■» |— a by Corollary 1.

39
6„ Computer Tachniques.s An essential tool for the finding of the
derivations- ih the previous- wortL has been a computer program for
iterating an algorithm for Mereditfi^s- condensed detachment operator
D Ccf §1). Briefly this program operates in the following way. To
detach y from Ea(3 , Robinsonrs unification algorithm [9, p32] is
used to unify y and a , and the most general unifier which results
is applied to |3 , yielding D Eap.y. As every theorem provable by
substitution and detachment is a substitution instance of a theorem
provable by condensed detachment [l], this program may be used for
theorem-proving in any theory with substitution and detachment as
sole rules of inference.

The program is coded in algol and run on the Auckland University

B6700 computer. It is about 245 lines long.

It is common, in the field of automatic theorem-proving, to

require the theorem-prover to completely solve a problem put to it.
In this work however it is acceptable if the program finds enough
material for the researcher to derive, by hand calculation, the
required result. A complete computer solution is not required. The
application of the program is especially interesting in view of the
fact that proofs for some of the results for which it has searched
have not previously appeared.

The comments below concern the application of the program to the

results recorded in §2.

The search space (cf[7, pl7]) may be formalised as follows.

We start with an axiom set. On the set of theorems which may be
derived from this set we define a "depth function” by assigning
depth 0 to the axioms and depth n+1 to a theorem whose deepest
parent has depth n. The standard procedure of systematically using
every theorem as a parent with all the theorems generated before it,
is a breadth-first search technique (cf[7» p45]) with respect to this
formalisation.

40
 With each pair of theorems capable of generating two new theorems,
it will be seen that the search space widens rapidly with increasing
depth. The only limiting factors are that detachment is not always
possible (since unification sometimes fails), a generated theorem may
be the same as one already generated, or its length may exceed a bound
determined by the dimension of the array in which the theorems are
stored in the computer (a restriction called the "length heuristic"
by some authors, e.g. Siklossy et al [10, pll]) .

The program had at its disposal 64,000 words of storage and a

maximum of 2,700 seconds of processing time.

The breadth-first technique applied to axiom (1) yielded 600

theorems in 168 seconds including theorems (2), (3), (5) and (10) ;
from these 600 theorems the derivation included in §2 from (1) to (7)
was easily constructed. This technique applied to (2) and (3) again
yielded 600 theorems very quickly, reaching only depth 5 on each
occasion. Again the derivations in §2 (i.e to (1) and (9) respectively)
were easily constructed from the generated theorems. Except for the
axioms indicated above, and (4) and (5) which are closely linkly
(cf §2), none of the 10 axioms yielded enough information in a breadth-
first search for a derivation to another axiom to be constructed.

Axiom (3) may be used as an example of the inability of the breadth-

first search to reach a great depth in the search space. In generating
the first 600 theorems of length at most 27 no theorem after theorem 39
in the search is used by the theorem prover as a parent ; however all
theorems occurring before theorem 167 in the search are at depth 4.
Thus the theorems at depth 5 appear to number several thousand and a
search to greater depth would appear to require such vast storage
resources as to be impracticable. Since the length of the longest
proof in §2 is 30 and the goal is at depth 14 in this proof, it is
clear that a search to greater depth is required, and hence some
heuristic considerations are desirable.

41
 It can be proven that the breadth-first search procedure produces
a path of minimal length to the goal (cf [7, p45]), and most research on
heuristic methods aims to preserve this property. For the present
search space, the property minimised by the breadth-first procedure is
the depth of the goal, whereas our aim in §2 was rather to minimise the
length of proof and the minimal depth proofs found by breadth-first
search were considerably altered to this end. Hence, in considering
heuristic methods, we were not constrained to preserve minimal path
length.

The best heuristic method found in this work involves the maximum
length of the stored theorems.

It is found that, as the search continues and more theorems are

generated, and hence become available for use as parents, the maximum
length of the theorems which are accepted for storage by the program
can be reduced without the program exhausting the pairs of parents
available to it for generation, and terminating. For example when the
breadth-first method is applied to (9) with theorems of length at most
19 stored, the program terminates after generating only two theorems.
With the storage length increased to 23 the breadth-first search generates
600 theorems to depth 11 in 600 seconds. However if after 50 theorems
of length at most 27 have been generated, only theorems of length at most
15 are stored, depth 13 is reached in the first 600 theorems generated
in 412 seconds, but more importantly 120 of the 600 theorems are of length
11 compared with 25 of the 600 with a constant maximum of 23, a con­
siderable advantage for subsequent hand calculation.

The particular heuristic used in finding the proofs for §2

accepted theorems of maximum length 27 for the first 50 theorems and
theorems of maximum length 15 thereafter. The program also gave pre­
ference to theorems of length 11 for one of the parents when using this
heuristic.

42
 An idea of the efficiency of this heuristic can be obtained if we
define the concept of Mpenetrance" as follows:

Penetrance = Proof length obtained/theorems generated.

This modifies the usual definition [7, p72] which is in terms of path
length rather than proof length.

To derive axiom (2) from axiom (9) the theorems necessary to allow
a simple hand calculation of (2) were derived with a proof length of 27
while 207 theorems were generated,
i.e. Penetrance = 0.13.
The penetrance for the first 18 theorems of this proof was 0.30.

The breadth first method proved these first 18 theorems while

generating 1000 theorems,
i.e. Penetrance = 0.018.

The heuristic method also found a proof of length 14 of axiom (2)

from axiom (3) in generating 642 theorems,
i.e. Penetrance = 0.022.
The penetrance for the first 9 theorems of this proof was 0.34.

The breadth-first method proved these first 9 theorems in

generating 781,
i.e. Penetrance = 0.011.

Variations of the simple heuristic outlined above involving

cutting down to a maximum length of 11 for the stored theorems at some
further point have been successfully used. So far, however, these
have been applied by trial and error to the search spaces of individual
axioms. To find a derivation of an axiom from (10) has proved a
difficult problem and has been solved only by the latter technique.

It may be argued that an ultimate goal of automatic theorem-proving

is to develop theorem-provers capable of proving results beyond the
unaided resources of the mathematician. Against this standard the

43
above program measures well. Although, the ten single axioms of §2
had all been given in the literature, proofs for five of them had not
appeared, and the proof in §3 that (6) is not a single axiom leaves
Meredith's other claims in respect of (7)-(11) in need of verification.
Except, perhaps, for experts in the field, such verification would have
been impossible without the computer program.

The program has proved useful in finding proofs of length up to 30

without the benefit of any problem reducing techniques (i.e. techniques
for splitting a problem into subproblems which may be solved independently
each with fewer steps than the original). It should be noted again that
complete computer solutions were not required, however from the point of
view of proving new results the technique used of applying the researcher*
intuition, in the form of hand calculation, to the computer output
appears just as acceptable as the conventional approach of attempting
to include this intuition in the program.

7. Concluding Remarks. This paper has proved that ten theorems of the
classical equivalential calculus are shortest single axioms for it with
substitution and ordinary detachment as rules of inference. It has also
been shown that ten theorems (namely the duals of the single axioms above)
are shortest single axioms for the classical equivalential calculus with
substitution and reverse detachment as rules of inference. Five theorems
appear on both lists. It is not known whether any more shortest single
axioms exist for the equivalential calculus with either set of rules of
inference.

There are 630 theorems of the correct length to be shortest single

axioms, but of these 286 have been shown not to be single axioms with
substitution and ordinary detachment in a manner similar to that used
in §4 to prove that the duals of (l)-(3), (5)-(7) are not such single
axioms. Of the remainder, approximately half may be rejected because
they satisfy the matrices of abelian groups under left or right
subtraction or the normal matrix for implication* In addition onany
likely candidates have been tested by computer without success. There
remain over 100 possible shortest single axioms with substitution and
ordinary detachment but none that look particularly promising. As
shown in §2 the problem for rules of inference substitution and reverse
detachment is reducible to that for substitution and ordinary detachment.

The other contribution of this paper has been the completeness

theorem showing that the theorems of the equivalential calculus, as
axiomatised by Lesniewski, are those formulas which hold in the matrix
of any Boolean group. The characterisation of these theorems known to
Lesniewski [2, p26, p29] has then been proved as a corollary. This
completeness result has been proved to be equivalent to the classical
completeness result that the theorems of the equivalential calculus are
those formulas which hold in the normal matrix for equivalence. This
last result appears in the literature [2].

The use of the computer in this work provides an area for further
development. The theorem-prover described in §6 is susceptible of
some improvement. While this program has proved successful in finding
proofs when these exist, there is some doubt about its effectiveness in
enabling formulas which are not single axioms to be rejected. As no
algorithm for solving this problem is known (it is not even known if this
problem, which might be called the "rejection problem11, is solvable), we
may desire to base an assessment of the potential of a formula on the
theorems proved from it by the theorem-prover.

Apart from improving the heuristics used in the theorem prover of

§6, a more powerful theorem-prover might be constructed, capable of
adding to the rules of inference (initially just the condensed detach­
ment operator D) new rules previously proved as lemmas by the theorem-
prover .

45
 A further development might be to adapt the theorem-prover for
interactive use. In some cases the best strategy is only apparent
when the theorem-prover has proceeded some distance into the search
space.

An opportunity to extend the application of the computer to this

work is provided by the abovementioned "rejection problem"; One method
of solving instances of this problem is to exhibit matrices in which
given formulas hold but the axioms do not. The test of whether given
formulas hold in given matrices appears a good candidate for mechanisation.

REFERENCES

1. J .A. Kalman, An algorithm for Meredithrs condensed detachment,

(abstract), Journal of Symbolic Logic, 39(1974), 206.

2. S. Lesniewski, Grundziige eines neuen Systems der Grundlagen

der Mathematik, Fundamenta Mathematicae, 14(1929), 1-81.

3. Jan -Lukasiewicz, Der Aquivalenzenkalkul3 Collectanea Logica,

1(1939), 145-169. English translation in [5], pp88-115
and [4], pp250-277.

4. Jan Lukasiewicz, Jan Lukasiewicz : Selected Works3 ed. by

L. Borkowski, North Holland Publishing Co., Amsterdam, 1970.

5. S. McCall, Polish Logic3 1920-19393 Oxford University Press,

Oxford, 1967.

6. C .A. Meredith and A.N. Prior, Axiomatics of the Pro-positional

Calculus, Notre Dame Journal of Formal Logic, 4(1963), 171-187.

7. N.J. Nilsson, Problem Solving Methods in Artificial Intelligence

McGraw-Hill, New York, 1971.

8. A.N. Prior, Formal Logic, 2nd ed., Oxford University Press,

Oxford, 1971.

46
9. J.A. Robinson, A Machine-Oriented Logic Based on the Resolution
Principle, Journal of the Association for Computing Machinery,
12(1965), 23-41.

10. L. Siklossy, A.Rich and V. Marinov, Breadth-First Search : Some

Surprising Results, Artificial Intelligence, 4(1973), 1-27.

11. M. Wajsberg, Ein neues Axiom des Aussagenkalkuls in der Symbolik

von Sheffer, Monatshefte fur Mathematik und Physik, 39(1932), 259-262.

University of Auckland.

47