PreprintPDF Available
Preprints and early-stage research may not have been peer reviewed yet.
On Reductions of Hintikka Sets for Higher-Order Logic
Alexander Steen1, Christoph Benzmüller2
1University of Luxembourg, FSTM, alexander.steen@uni.lu
2Freie Universität Berlin, FB Mathematik und Informatik,c.benzmueller@fu-berlin.de
January 28, 2021
Abstract
Steen’s (2018) Hintikka set properties for Church’s type theory based on primitive equality are reduced
to the Hintikka set properties of Brown (2007). Using this reduction, a model existence theorem for Steen’s
properties is derived.
1 Introduction and Preliminaries
Abstract consistency properties and Hintikka sets play an important role in the study (e.g., of Henkin-complete-
ness) of proof calculi for Church’s type theory [7, 2], aka. classical higher-order logic (HOL). Technically quite
different definitions of these terms have been used in the literature, since they depend on the primitive logical
connectives assumed in each case. The definitions of Benzmüller, Brown and Kohlhase [3, 4], for example, are
based on negation, disjunction and universal quantification, while Steen [8], in the tradition of Andrews [1],
works with primitive equality only. Despite their conceptual relationship, important semantical corollaries that
are implied by these syntax related Hintikka properties (such as model existence theorems) can hence not be
directly transferred between formalisms.
Brown [6, 5] presents generalised abstract consistency properties in which the primitive logical connectives
can vary. In this paper we show that the properties of Steen can be reduced to those of Brown.1Theorem 1
that we establish in this paper paves way for the convenient reuse of (e.g., model existence) results from the
work of Brown [6, 5] in the context of Steen’s setting.
Paper structure. In the remainder of this introduction we recapitulate some relevant (syntactic) notions on
HOL, mainly to clarify our notation; for further details on HOL as relevant for this paper. §2 we present the
Hintikka properties as used by Brown, and in §3 we give the related properties as used by Steen. In §3 we then
show various lemmas that are implied in Steen’s setting, and it are (some of ) those lemmas which prepare the
main reduction result of this paper (Theorem 1), which is given in §4. A model existence theorem for Hintikka
sets as defined by Steen is then derived in §5.
Equality conventions. Different notions of equality will be used in the following: If a concept is defined (as
an abbreviation), the symbol := is used. Primitive equality, written =τ, refers to a logical constant symbol
from the HOL language such that sτ=τtτis a term of the logic (assuming sτand tτare, where τis a
type annotation), cf. details further below. Leibniz-equality, written ˙=, is a defined term; usually it stands
for λXτ. λYτ.P.(P X)(P Y ), where is a (primitive) logical connective. Meta equality denotes
set-theoretic identity between objects. Finally, ?, for ?⊆ {β, η }is used for syntactic equality modulo β,η
and βη-conversion, respectively (as in the related work, α-conversion is taken as implicit).
Syntax of HOL. The set Tof simple types is freely generated from the base types oand ιby juxtaposition.
The types oand ιrepresent the type of Booleans and individuals, respectively. A type ντ represents the type
of a total function from objects of type τto objects of type ν.
1In an earlier reduction attempt we tried to reduce Steen’s [8] abstract consistency properties to those of Benzmüller, Brown and
Kohlhase [3]. This attempt had gaps that could not easily be closed (as was pointed out by an unknown reviewer). The mapping
that we applied in this initial reduction attempt between the two formalisms replaced primitive equations (in Steen’s Hintikka sets)
by Leibniz equations (to obtain Hintikka sets in the style of Benzmüller, Brown and Kohlhase). It thereby introduced additional
universally quantified formulas which in turn triggered the applicability of abstract consistency conditions whose validity could
not be ensured by referring to those of Steen. In this work we therefore instead map to the structurally better suited, generalised
conditions of Brown [6, 5], which enables us to circumvent the problems induced by our previous reduction attempt to the work of
Benzmüller, Brown and Kohlhase.
1
Let Στbe a set of constant symbols of type τ∈ T and let Σ := Sτ∈T Στbe the union of all typed symbols,
called a signature. Let further Vdenote a set of (typed) variable symbols. From these the terms of HOL are
constructed by the following abstract syntax (τ , ν ∈ T ):
s, t ::= cτΣ|Xτ∈ V | (λXτ. sν)ντ |(sν τ tτ)ν
The terms are called constants,variables,abstractions and applications, respectively. The set of all terms of
type τover a signature Σis denoted Λτ(Σ), and Λc
τ(Σ) is used for closed terms, respectively. The notion of free
and bound variables are defined as usual, and a term tis called closed if tdoes not contain any free variables.
It is assumed that the set Vcontains countably infinitely many variable symbols for each type τ∈ T .
The type of a term is written as subscript but may be dropped by convention if clear from the context
(or if not important). Also, parentheses are omitted whenever possible, and application is assumed to be left-
associative. Furthermore, the scope of an λ-abstraction’s body reaches as far to the right as is consistent with
the remaining brackets. Nested applications s t1. . . tnmay also be written in vector notation s tn.
Variants of HOL often differ regarding the choice of the primitive logical connectives in the signature Σ.
In any case, s6=tis used in the remainder as an abbreviation for ¬(s=t). Also, for simplicity, binary
logical connectives may be written in infix notation; e.g., the term poqoformally represents the application
(ooo poqo). Furthermore, binder notation is used for universal and existential quantification: The term Xτ. so
is used as a short-hand for Πτ(λXτ. so), where Πτis a constant symbol. Finally, Leibniz-equality, denoted ˙=,
is defined as ˙= := λXτ. λYτ.P.(P X)(P Y ). A Σ-formula sois a term from soΛo(Σ) of type oand a
Σ-sentence if it is a closed Σ-formula. The reference to Σmay be omitted if clear from the context.
In the following, variables are denoted by capital letters such as Xτ, Yτ, Zτ, and, more specifically, the
variable symbols Po, Qoand Fντ , Gν τ are used for predicate or Boolean variables and variables of functional
type, respectively. Analogously, lower case letters sτ, tτ, uτdenote general terms and fντ , gντ are used for terms
of functional type.
Semantics of HOL. The semantics of HOL, including the notions of Σ-models and Σ-Henkin models, is not
discussed here; cf. [3, 8, 6, 5] for details.
2 Hintikka sets as defined by Brown [6]
In the formulation of HOL as employed by Brown [6], the set of primitive logical connectives is not fixed and
may be chosen arbitrarily from the set {>o,o,¬oo,ooo,ooo ,ooo ,ooo} ∪ {Πo( )|τ T } ∪ {Σo()|τ
T } ∪ {=τ
oτ τ |τ∈ T }. All remaining constant symbols from Σare called parameters. In this presentation, the
notation from Brown is adapted and we apply the conventions from above. For example, we denote general
terms with lower case symbols sand tinstead of upper case letters as used by Brown, and instead of wff τ(Σ),
which Brown uses to denote the set of internal terms of type τ, we use Λτ(Σ).
Brown distinguishes between internal terms, the elements of Λτ(Σ), and external propositions (meta-level
propositions) in a set prop(Σ), which are defined as follows [6, Def. 2.1.20] (we use the color red to visually
highlight Brown’s meta-level connectives and their associated Hintikka set properties in the remainder of this
paper):
If sΛo(Σ), then sprop(Σ).
If α∈ T and s, t Λα(Σ), then [s=
·t]prop(Σ).
>
·prop(Σ).
If sprop(Σ), then [¬
·s]prop(Σ).
If s, t prop(Σ), then [s
·t]prop(Σ).
If sprop(Σ), then [
·Xαs]prop(Σ).
Closed propositions sprop(Σ) are also called sentences; we then write ssent(Σ).
Brown introduces the following Hintikka set properties for external propositions.
Properties of extensional Hintikka sets H[6, Def. 5.5.4]:
~
c:s /∈ H or ¬
·s /∈ H.
~
βη :If s∈ H, then s∈ H.
~
:¬
·>
·/∈ H.
2
~
¬:If ¬
·¬
·s∈ H, then sH.
~
:If s
·t∈ H, then s∈ H or t∈ H.
~
:If ¬
·(s
·t)∈ H, then ¬
·s∈ H and ¬
·t∈ H.
~
:If
·Xτs∈ H, then [t/x]s∈ H for every closed term tΛc
τ(Σ).
~
:If ¬
·(
·Xτs)∈ H, then there is a parameter pτΣτsuch that ¬
·([p/X]s)∈ H.
~
]:If s∈ H, then s]∈ H. Also, if ¬
·s∈ H, then ¬
·s]∈ H.
~
m:If ¬
·h sn∈ H and h tn∈ H, then there is an iwith 1insuch that ¬
·si=
·ti∈ H.
~
dec :If pis a parameter and ¬
·(p sn)=
·ι(p tn)∈ H, then there is an i,1in, s.t. ¬
·si=
·ti∈ H.
~
b:If ¬
·(s=
·ot)∈ H, then {s, ¬
·t}⊆Hor {¬
·s, t}⊆H.
~
f:If ¬
·(f=
·ντ g)∈ H, then there is a parameter pτΣτsuch that ¬
·(f p =
·νg p)∈ H.
~
o
=:If s=
·ot∈ H, then {s, t}⊆Hor {¬
·s, ¬
·t}⊆H.
~
=:If f=
·ντ g∈ H, then (f u =
·νg u)∈ H for every closed term uΛc
τ(Σ).
~
r
=:¬
·(s=
·is)/∈ H.
~
u
=:Suppose (s=
·it)∈ H and ¬
·(u=
·iv)∈ H. Then ¬
·(s=
·iu)∈ H or ¬
·(t=
·iv)∈ H. Also, ¬
·(s=
·iv)∈ H
or ¬
·(t=
·iu)∈ H.
The collection of all sets of sentences satisfying all these properties is called Hintβfb(Σ).
3 Hintikka sets as defined by Steen [8]
In the formulation of HOL as employed by Steen [8], the equality predicate, denoted =τ, for each type τ, is
assumed to be the only logical connective present in the signature Σ, i.e., {=τ|τ∈ T } Σ. All (potentially)
remaining constant symbols from Σare called parameters. Such signatures are also referred to as Σ=. A formu-
lation of HOL based on equality as sole logical connective originates from Andrew’s system Q0, cf. [1] and the
references therein. The usual logical connectives are defined as follows (technically our formulation is a modi-
fication of the one used by Andrews [1], since the order of terms in defining equations is swapped in many cases):
>o:= =o
ooo =ooo
o(ooo)(ooo)=o
ooo
o:= (λPo. P ) =oo (λPo.>)
¬oo := λPo. P =o
ooo := λPo. λQo.(λFooo. F > >) =o(ooo)(λFooo. F P Q)
ooo := λPo. λQo.¬(¬P∧ ¬Q)
ooo := λPo. λQo.¬PQ
ooo := λPo. λQo. P =oQ
Πτ
o():= λP. P =λXτ.>
The (primitive and defined) connectives of this formulation of HOL are written in blue in the following, as
are the properties below.
Properties for acceptable Hintikka sets [8, Def. 3.15]
~
c:s /∈ H or ¬s /∈ H.
~
βη : If sβη tand s∈ H, then t∈ H.
~
r
=:(s6=s)/∈ H.
~
s
=: If u[s]p∈ H and s=t∈ H then u[t]p∈ H.
~
+
b: If s=t∈ H, then {s, t}⊆H or {¬s, ¬t}⊆H.
~
b: If s6=t∈ H, then {s, ¬t} ⊆ H or {¬s, t}⊆H.
~
+
f: If fν τ =gντ ∈ H, then f s=g s ∈ H for any closed term sΛc
τ(Σ).
~
f: If fν τ 6=gντ ∈ H, then f w6=g w ∈ H for some parameter wΣτ.
~
m: If s, t are atomic and s, ¬t∈ H, then s6=t∈ H.
~
d: If h sn6=h tn∈ H, then there is an iwith 1insuch that si6=ti∈ H.
The collection of all sets satisfying these properties is called H. Every element H ∈ His called acceptable.
Definition 1. A set Hof formulas is called saturated iff s∈ H or ¬s∈ H for every closed formula s.
3
Derived properties
Lemma 1 (Basic properties).Let H ∈ H. Then it holds that
(a) /∈ H
(b) ¬>/∈ H
(c) >=/∈ H
(d) If so=>∈Hor >=so∈ H (so6= ∈ H or 6=so∈ H), then {so,>} ⊆ H ({so,¬⊥} ⊆ H)
(e) If so=⊥∈Hor =so∈ H (so6=> ∈ H or >6=so∈ H), then {¬so,¬⊥} ⊆ H ({¬so,>} ⊆ H)
(f) If ¬⊥ ∈ H, then >∈H
(g) If > ∈ H, then ¬⊥ ∈ H
(h) If s=t∈ H and t=u∈ H, then s=u∈ H.
Proof. Let H ∈ Hbe an acceptable Hintikka set.
(a) Assume ⊥∈H. By definition of it holds (λP. P)=(λP . >)∈ H. Hence, by ~
+
fand ~
βη , it follows
that w=>∈Hfor any closed term w. Taking w¬>we obtain ¬>=>∈H. But then, ~
+
bgives us a
contradiction to ~
c. Hence /∈ H.
(b) Assume ¬> ∈ H. By definition of it holds (=6==)∈ H, which contradicts ~
r
=. Hence, ¬>/∈ H.
(c) Assume >=⊥∈H. Applying ~
+
bgives us that either {>,⊥} ⊆ H or {¬>,¬⊥} ⊆ H. Either case is
impossible by either (a) or (b) of this lemma. Hence, >=/∈ H.
(d) Let so=> ∈ H or >=so∈ H. In both cases it follows by ~
+
bthat either {s, >} ⊆ H or {¬s, ¬>} ⊆ H.
Since the latter case contradicts (b) from above, it follows that {s, >} ⊆ H. The negative cases are
analogous using ~
b.
(e) Let so=⊥ ∈ H or =so∈ H. In both cases it follows by ~
+
bthat either {s, ⊥} ⊆ H or {¬s, ¬⊥} ⊆ H.
Since the former case contradicts (a) from above, it follows that {¬s, ¬⊥} ⊆ H. The negative case is
analogous using ~
b.
(f) Let ¬⊥ ∈ H. Then, by definition of ,(λP. P )6=(λP. >)∈ H. By ~
fand ~
βη it holds that p6=> ∈ H
for some parameter p. By ~
bit follows that either {p, ¬>} ⊆ H or {¬p, >} ⊆ H. Since the former case
is ruled out by (b) from above, the latter case yields the desired result.
(g) Let > ∈ H, that is, =o=ooo=o∈ H. By ~
+
fand ~
βη it follows that (so=to)=(so=to)∈ H for every two
closed formulas s, t. For st¬it follows that (¬=¬)=(¬=¬)∈ H, and hence, by ~
+
b, either
¬=¬⊥∈Hor ¬6=¬⊥∈H. Since the latter case is ruled out by ~
r
=, it follows that ¬=¬⊥∈H.
Again, by ~
+
b, it follows that either ¬⊥∈Hor ¬(¬)∈ H. The latter case is impossible by ~
r
=since
¬(¬)(6=)and hence ¬⊥ ∈ H.
(h) Let s=t∈ H and t=u∈ H. By ~
s
=it follows directly that s=u∈ H.
Lemma 2 (Properties of usual connectives).Let H ∈ H. Then it holds that
(a) If ¬¬so∈ H, then s∈ H
(b) If (soto)∈ H, then s∈ H or t∈ H.
(c) If (soto)∈ H, then s∈ H and t∈ H.
(d) If ΠαF∈ H, then F s ∈ H for every closed term s.
(e) If ¬ΠαF∈ H, then ¬(F w)∈ H for some parameter wΣ.
Proof. Let H ∈ Hbe an acceptable Hintikka set.
4
(a) Let ¬¬so∈ H. By definition of ¬and ~
βη it holds (s6=)∈ H. Hence, by ~
b, either {s, ¬⊥} ⊆ H,
or {¬s, ⊥} ⊆ H. As the latter case is impossible by Lemma 1(a), it follows that {s, ¬⊥} ⊆ H and, in
particular, that s∈ H.
(b) Let (soto)∈ H. By definition of ,¬and ~
βη it holds λP. P > >6=λP. P (¬s) (¬t)∈ H. Hence, by
~
fand ~
βη , it follows that (p> >)6=p(¬s) (¬t)for some parameter pΣ. By ~
deither (i) >6=¬s∈ H
or (ii) >6=¬t∈ H. Hence, by ~
b, applied to both cases, it holds that either (i) ¬¬s∈ H, or (ii) ¬¬t∈ H
(because ¬>/∈ H by Lemma 1(b)). It follows that s∈ H or t∈ H by (a) of this lemma.
(c) Let (soto)∈ H. By definition of and ~
βη it holds (λP. P > >)=(λP. P s t)∈ H. Hence, by ~
+
f
and ~
βη , it follows that (w> >)=(w s t)∈ H for every closed term w. By ~
βη , using wλx. λy. x
and wλx. λy. y, it holds >=s∈ H and >=t∈ H, respectively. Application of Lemma 1(d) yields the
desired result.
(d) Let Παs∈ H. By definition of Παand ~
βη it holds s=λx. >∈ H. Hence, by ~
+
fand ~
βη , it follows
that s t=> ∈ H for every closed term t. Application of Lemma 1(d) yields the desired result.
(e) Let ¬Παs∈ H. By definition of Παand ~
βη it holds that s6=(λx. >)∈ H. Hence, by ~
fand ~
βη , it
follows that (s p)6=> ∈ H for some parameter p. Application of Lemma 1(e) yields the desired result.
Lemma 3 (Properties of Leibniz equality).Let H ∈ H. Then it holds that
(a) If s˙= t∈ H, then s=t∈ H.
(b) If ¬(s˙= t)∈ H, then s6=t∈ H.
(c) ¬(s˙= s)/∈ H.
(d) If u[s]p∈ H and s˙= t∈ H, then u[t]p∈ H.
(e) If s˙= t∈ H, then t˙= s∈ H.
(f) If s˙= t∈ H and t˙= u∈ H, then s˙= u∈ H.
Proof. Let H ∈ Hbe an acceptable Hintikka set.
(a) Let (s˙= t)∈ H. By definition of ˙= and ~
βη we have λP. (P s)(P t)=λP. >∈ H, and hence, by
~
+
fand ~
βη , it holds that (w s)(w t)=>∈Hfor every closed term w. Then, (w s)(w t)∈ H by
Lemma 1(d). By definition of and ~
βη it holds that ¬(w s)(w t)∈ H and hence, by Lemma 2(b),
that ¬(w s)∈ H or (w t)∈ H. For w(λX. s=X)it follows by ~
βη that ¬(s=s)∈ H or (s=t)∈ H.
Since the former case contradicts ~
r
=, it follows that (s=t)∈ H.
(b) Let ¬(s˙= t)∈ H. By definition of ˙= and ~
βη we have λP. (P s)(P t)6=λP. >∈ H. Hence, by
~
fand ~
βη , it holds that (p s)(p t)6=> ∈ H for some parameter p. By Lemma 1(e) it follows that
¬(p s)(p t)∈ H. Then, by Lemma 2(a) and 2(c), it follows that ¬¬(p s)∈ H and ¬(p t)∈ H.
Moreover, (p s)∈ H by Lemma 2(a). By ~
mit then follows that (p s)6=(p t)∈ H, and finally, by ~
d,
that s6=t∈ H.
(c) Assume ¬(s˙= s)∈ H. By (b) above it follows that s6=s∈ H which contradicts ~
r
=. Hence, ¬(s˙= s)/∈ H.
(d) Let u[s]p∈ H and s˙= t∈ H. By (a) above it holds that s=t∈ H and thus by ~
s
=it follows that u[t]p∈ H.
(e) Let (s˙= t)∈ H. By definition of ˙= and ~
βη we have λP. (P s)(P t)=λP. >∈ H. Hence, by
~
+
fand ~
βη , it holds that (w s)(w t)=>∈Hfor every closed term w. Then, (w s)(w t)∈ H by
Lemma 1(d). By definition of and ~
βη it holds that ¬(w s)(w t)∈ H and hence by Lemma 2(b)
that ¬(w s)∈ H or (w t)∈ H. For w(λX. t ˙= s), it follows by ~
βη that ¬(t˙= s)∈ H or (t˙= s)∈ H.
Assume ¬(t˙= s)∈ H. Since by Lemma 3(a) it holds that s=t∈ H, it follows by ~
s
=that ¬(t˙= t)∈ H,
which contradicts (c) above. Hence, (t˙= s)∈ H.
(f) Let (s˙= t)∈ H and (t˙= u)∈ H. By (d) above it follows that (s˙= u)∈ H.
5
Lemma 4 (Sufficient conditions for saturatedness).Let H ∈ H. Then it holds that
(a) If > ∈ H, then His saturated.
(b) If ¬s∈ H for some closed term s, then His saturated.
(c) If st∈ H for some closed terms s, t, then His saturated.
(d) If st∈ H for some closed terms s, t, then His saturated.
(e) If ΠτP∈ H for some closed term Pτo, then His saturated.
(f) If s˙= t∈ H for some closed terms s, t, then His saturated.
Proof. Let H ∈ Hbe an acceptable Hintikka set.
(a) Let > ∈ H, that is, =o=ooo=o∈ H. By ~
+
fand ~
βη it follows that (so=to)=(so=to)∈ H for every two
closed formulas s, t. For stcfor some closed term c, it follows that (c=c)=(c=c)∈ H and thus, by
~
+
band ~
r
=, it holds that c=cH. By ~
+
bit follows that c∈ H or ¬c∈ H. Hence, His saturated.
(b) If ¬s∈ H for some closed term s, then s=⊥∈H. By ~
+
b, it follows that either {s, ⊥} ⊆ H or
{¬s, ¬⊥} ⊆ H. Since the former case is ruled out by Lemma 1(a), it follows that ¬⊥ ∈ H. By Lemma 1(f )
it follows that >∈Hand by (a) above it follows that His saturated.
(c) If st∈ H for some closed terms s, t, then by definition of and ~
βη we have ¬(¬s∧¬t)∈ H. An
application of (b) yields the desired result.
(d) If st∈ H for some closed terms s, t, then by definition of and ~
βη it holds (λg. g s t)=(λg. g > >)∈ H.
By ~
+
fand ~
βη it follows that s=>∈H(take λx. λy. x). By ~
+
b, it follows that either {s, >} ⊆ H or
{¬s, ¬>} ⊆ H. Since the latter case is ruled out by Lemma 1(b), it follows that > ∈ H. An application
of (a) above yields the desired result.
(e) If Πτs∈ H for some closed terms s, then by definition of Πτand ~
βη it holds that s=(λx. >)∈ H.
By ~
+
fand ~
βη it follows that (s w)=> ∈ H for every closed term w. By ~
+
b, it follows that either
{(s w),>} ⊆ H or {¬(s w),¬>} ⊆ H. Since the latter case is ruled out by Lemma 1(b), it follows that
> ∈ H. An application of (a) above yields the desired result.
(f) Let s˙= t∈ H. By definition of Πτand ~
βη it holds that ΠλP. (P s)(P t)∈ H. An application of (e)
yields the desired result.
Corollary 1. Let H ∈ Hand let s6=t∈ H or ¬(s˙= t)∈ H for some closed terms s, t. Then, His saturated.
Proof. As (s6=t)¬(s=t), both cases are a special instance of Lemma 4(b).
Lemma 5 (Saturated sets properties).Let H ∈ Hand let Hbe saturated. Then it holds that
(a) If s=t∈ H, then s˙= t∈ H.
(b) If s=t∈ H then t=s∈ H.
(c) s=s∈ H for every closed term s.
(d) s˙= s∈ H for every closed term s.
Proof. Let H ∈ Hand let Hbe saturated.
(a) Let s=t∈ H and assume s˙= t /∈ H. Since His saturated we have ¬(s˙= t)∈ H. Then, by Lemma 3(b),
it follows that s6=t∈ H, and thus {s=t, s6=t}⊆H, which contradicts ~
c. Hence, s˙= t∈ H.
(b) Let s=t∈ H and assume t=s /∈ H. Since His saturated we have t6=s∈ H. Then, by ~
s
=, it follows that
t6=t∈ H which contradicts ~
r
=. Hence, t=s∈ H.
(c) Let sbe a closed term of some type and assume that s=s /∈ H. Since His saturated we have that s6=s∈ H.
Since this contradicts ~
r
=it follows that s=s∈ H.
(d) Let sbe a closed term of some type and assume that s˙= s /∈ H. Since His saturated we have that
¬(s6=s)∈ H. Since this is impossible by Lemma 3(c) it follows that s˙= s∈ H.
6
Lemma 6 (Properties of negated equalities).Let H ∈ H. It holds that
(a) If s6=t∈ H, then t6=s∈ H.
(b) If ¬(s˙= t)∈ H, then ¬(t˙= s)∈ H
(c) If s6=t∈ H, then ¬(s˙= t)∈ H.
Proof. Let H ∈ H.
(a) Let s6=t∈ H and assume that t6=s /∈ H. By Corollary 1 it follows that His saturated and hence we have
that t=s∈ H. By saturation and Lemma 5(b) it follows that s=t∈ H, and thus {s6=t, s =t}⊆H,
which contradicts ~
c. Hence, t6=s∈ H.
(b) Let ¬(s˙= t)∈ H and assume ¬(t˙= s)/∈ H. By Corollary 1 it follows that His saturated and hence we
have that t˙= s∈ H. Then, by Lemma 3(d), it follows that s˙= t∈ H, and thus {¬(s˙= t), s ˙= t}⊆H,
which contradicts ~
c. Hence, ¬(t˙= s)∈ H.
(c) Let s6=t∈ H and assume ¬(s˙= t)/∈ H. By Corollary 1 it follows that His saturated and hence we have
that (s˙= t)∈ H. Then, by Lemma 3(a), it follows that s=t∈ H, and thus {s=t, s 6=t} ⊆ H, which
contradicts ~
c. Hence, ¬(s˙= t)∈ H.
Definition 2 (Leibniz-free).Let Sbe a set of formulas. Sis called Leibniz-free iff s˙= t /Sfor any terms s, t.
Corollary 2 (Impredicativity Gap).Let H ∈ H.His saturated or Leibniz-free.
Proof. Assume that His not Leibniz-free. Then, there exists some formula s˙= t∈ H. An application of
Lemma 4(f) yields the desired result.
Summary of properties of equality and Leibniz-equality. The following table contains an overview of
the implied properties of =and ˙= , respectively. A property that holds unconditionally is marked with X, a
property that holds for saturated Hintikka sets is marked with sat..
Property ?=?˙=
s?s∈ H sat.sat.
¬(s?s)/∈ H X X
If s?t∈ H and t?u∈ H, then s?u∈ H X X
If s?t∈ H, then t?s∈ H sat.X
If u[s]p∈ H and s?t∈ H, then u[t]p∈ H X X
4 Reduction of H(Steen) to Hintβfb (Brown)
We reduce the notion of Hintikka sets of Steen to the notion of Hintikka sets by Brown. Conceptually, every
formula from a set H ∈ His first translated to its red equivalent under a signature Σ⊆ {¬} ∪ {=τ|τ T }
containing no further primitive logical connectives. Note that this involves mapping a primitive connective to a
primitive connective (i.e. =to =) as well as mapping a defined connective to a primitive connective (i.e. ¬to ¬).
In a second step, the red formulas are translated into their meta-counterparts, i.e. into external propositions.
Formally, we define the mapping as follows: Let Σbe given as introduced above. We define H.:=
{(s[¬\¬])[=\=]|sH}, where s[l\r]denotes that term in Λc(Σ)that is obtained by replacing all occurrences
of lin sby r. Obviously, if sΛc
o(Σ)then (s[¬\¬])[=\=]Λc
o(Σ), and thus H.Λc
o(Σ).
As the Hintikka properties of Brown are defined in terms of external propositions, we proceed by constructing
a set H]
.prop(Σ)from H.by enriching H.with its external counterparts: To that end, for any term
soΛo(Σ), we denote by s]prop(Σ)the term that is constructed by replacing the primitive connective at
head position by its external equivalent2, i.e. we have that (¬s)]is equal to ¬
·sand that (s=ot)]is equal to
s=
·ot. If there is no connective at head position, the term is left unchanged, i.e., (p sn)]equals p snand (X sn)]
equals X snwhenever pΣis a parameter and Xis a variable.
Then, H]
.is defined inductively as follows. For H ∈ H, let H]
.be the smallest set of external propositions
over Σsuch that
2This corresponds to the original definition of s]by Brown [6, Def. 2.1.38].
7
(1) if so∈ H, then s∈ H]
., and
(2) if so∈ H, then s]∈ H]
., and
(3) if ¬
·s∈ H]
., then ¬
·s]∈ H]
..
Intuitively, this translation adopts the translated object-level terms of H(clause (1)), and additionally aug-
ments the set corresponding meta-level terms by replacing connectives at head positions by its meta-connective
equivalent (clause (2)). Also, connectives directly under a (meta-)negation are considered (clause (3)).3Deep
translations are then implicitly provided by the Hintikka closure properties.
We provide simple examples of unsaturated Hintikka sets H,H.and H]
.:
Let H:= {s|sβη tfor t∈ {a=ιb, a =ιa, b =ιb, b =ιa}}. Then H.={s|sβη tfor t∈ {a=ιb, a =ι
a, b =ιb, b =ιa}} and H]
.=H.∪ {s|sβη tfor t∈ {a=
·ιb, a =
·ιa, b =
·ιb, b =
·ιa}}.
Let H:= {s|sβη tfor t∈ {a=ιp(¬), a =ιa, p(¬)=ιp(¬), p(¬)=ιa}}. Then H.={s|sβ η tfor t
{a=ιp(¬), a =ιa, p(¬)=ιp(¬), p(¬)=ιa}} and H]
.=H.∪ {s|sβη tfor t∈ {a=
·ιp(¬), a =
·ιa, p(¬)=
·ι
p(¬), p(¬)=
·ιa}}.
We now show that if H ∈ Hthen H]
.Hintβfb (i.e., H]
.fulfils all ~
from §2).
Theorem 1 (Reduction of Hto Hintβfb).If H ∈ H, then there exists an extensional Hintikka set H0Hintβfb
such that H]
.≡ H0.
Proof. Let H ∈ Hbe a Hintikka set according §3, i.e., fulfilling all ~
properties.
We verify every ~
-property for H]
.:
~
c: Assume that both s∈ H]
.and ¬
·s∈ H]
.. Then, by definition, sHand ¬sH. As this contradicts ~
c,
it follows that s/∈ H]
.or ¬
·s/∈ H]
..
~
βη : Let s∈ H]
.. Then, by definition, s∈ H. Since sβη sit follows by ~
βη that s∈ H. By definition, we
then have s∈ H]
..
~
:¬
·>
·/∈ H]
.is vacuously true, as >
·is never the result of any term translation.
~
¬: Let ¬
·¬
·s∈ H]
.. Then, by definition, ¬¬s∈ H. By Lemma 2(a) it follows that s∈ H and hence s∈ H]
..
~
: Vacuously true, as
·is never the result of any term translation.
~
: Vacuously true, as
·is never the result of any term translation.
~
: Vacuously true, as
·is never the result of any term translation.
~
: Vacuously true, as
·is never the result of any term translation.
~
]: Holds by definition of H]
..
~
m: Let ¬
·p sn∈ H]
.and let p tn∈ H]
.for some parameter p. Then, by definition, ¬p sn∈ H and
p tn∈ H. By ~
mit follows that p tn6=p sn∈ H. By ~
dit then follows that there is some i,
1in, such that ti6=si∈ H. By Lemma 6(a) it follows that si6=ti∈ H and hence ¬
·(si=
·ti)∈ H]
..
~
dec: Let pbe a parameter and let ¬
·(p sn)=
·ι(p tn)∈ H]
.. Then, by definition, ¬(p sn)=ι(p tn)∈ H.
By ~
dit follows that there is some i,1in, such that ¬(si=ti)∈ H and hence ¬
·(si=
·ti)∈ H]
..
~
b: Let ¬
·(s=
·ot)∈ H]
.. Then, by definition, ¬(s=ot)∈ H. By ~
bit follows that {s, ¬t}⊆Hor {¬s, t} ⊆ H.
Hence we have that {s,¬
·t}⊆H]
.or {¬
·s,t}⊆H]
..
~
f: Let ¬
·(f=
·ντ g)∈ H]
.. Then, by definition, ¬(f=ντ g)∈ H. By ~
fit follows that ¬(f p =νg p)∈ H for
some parameter pτ. Hence we have that ¬
·(f p =
·νg p)∈ H]
..
~
o
=: Let s=
·ot∈ H]
.. Then, by definition, s=ot∈ H. By ~
+
bit follows that {s, t}⊆Hor {¬s, ¬t}⊆H.
Hence we have that {s,t}⊆H]
.or {¬
·s,¬
·t} ⊆ H]
..
3Clauses (2) and (3) intuitively reflect property ~
]of Hintikka sets by Brown [6, Def. 5.5.1].
8
~
=: Let f=
·ντ g∈ H]
.. Then, by definition, f=ν τ g∈ H. By ~
+
fit follows that f s =νg s ∈ H for every
closed term sτΛc
τ(Σ). Hence we have that f s =
·νg s ∈ H]
.for every closed sτΛc
τ(Σ).
~
r
=: Assume ¬
·(s=
·is)∈ H]
.. Then, by definition, ¬(s=is)∈ H, contradicting ~
r
=. Hence we have that
¬
·(s=
·is)/∈ H]
..
~
u
=: Let (s=
·it)∈ H]
.and ¬
·(u=
·iv)∈ H]
.. Then, by definition, (s=it)∈ H and ¬(u=iv)∈ H. By ~
mit
follows that (s=it)6=o(u=iv)∈ H. By ~
dwe then have s6=iu∈ H or t6=iv∈ H. Hence, ¬
·(s=
·iu)∈ H]
.or
¬
·(t=
·iv)∈ H]
.. For the second half, we know by Corollary 1 that His saturated and hence by Lemma 5(b)
it holds that (t=is)∈ H. Applying ~
s
=yields t6=iu∈ H or s6=iv∈ H, and thus ¬
·(t=
·iu)∈ H]
.or
¬
·(s=
·iv)∈ H]
..
We claim that this result analogously holds if the original definitions of Andrews for the logical connectives
are assumed instead of the slightly modified ones introduced in §3 and used in [8]; a technical proof remains
future work.
5 Use Case: Bridging Model Existence
In this section, we apply the above reduction to derive a model existence theorem for Steen’s Hintikka sets.
Informally, we proceed as follows: There exists an extensional {=,¬}-model M ∈ Mβbf such that M |=H]
..
Because we constructed H]
.as an extensional Hintikka set based on negation and equality, we know that the
domain of Booleans in Mis bivalent. In order to get a model solely based on equality (as required in the notion
of Steen), we subsequently restrict Mto terms over {=τ|τ T }. Finally, we find an extensional model over
frames isomorphic to it.4
First, we summarize important results by Brown [6] used in this reduction.
Theorem 2 (Model Existence for Extensional Hintikka Sets [6, Theorem 5.7.17]).Let Hbe an extensional
Σ-Hintikka set (i.e., H ∈ Hintβbf(Σ)). There is an extensional Σ-model M(i.e., M ∈ Mβbf(Σ)) such that
M |=H.
Theorem 3 (Property b[6, Theorem 3.3.7]).Let Σbe a signature and Mbe an Σ-model. Suppose either
>,Σor ¬Σ. Then Msatisfies biff Dohas two elements.
Theorem 4 (Isomorphic Models over Frames [6, Theorem 3.5.6]).Let M=D,@,E, vMβbf (Σ)be an
extensional Σ-model such that Dohas two elements. There is an isomorphic Σ-model Mh=Dh,@h,Eh, vh
over frames, in particular Dh
o={T , F }and vhis the identity.
Now we infer a model existence theorem for Steen’s Hintikka properties by bridging to those of Brown: Let
H ∈ Hbe a Hintikka set (according to Steen) over a signature Σ=. Let H]
.be the translated set according to
§4. By Theorem 1 it follows that H]
.is an extensional Σ-Hintikka set, for Σ:= {¬} ∪ {=τ|τ T } ∪ {p|p
Σ=is parameter}. By Theorem 2 it follows that there is an extensional model M=D,@,E, vMβbf(Σ)
such that M |=H]
.. Since ¬Σit follows by Theorem 3 that Dois bivalent.5
Now we eliminate ¬from the signature Σto get an extensional model over {=τ|τ T }; we refer to this
signature as Σ=, i.e., let Σ=:= Σ\ {¬}. To that end, let M0:= D,@,E|Λ(Σ=), v.M0is an extensional
Σ=-model, in particular Dois bivalent [6, Theorem 3.3.15]. Now by Theorem 4 it follows that there is an
Σ=-model over frames Mh= (Dh,@h,Eh, vh)isomorphic to M0.
We now construct our desired Henkin model M/for Hover Σ=as follows: M/:= (D/,I/), where
D/:= Dh, and
I/:= c7→ Eh(c)for all cΣ=.
It is easy to see that M/is a Henkin model; in particular I(=τ)(a,b) = Tiff abfor every a,b∈ Dh
τ,
τ T , by property L=τ(q)of Mhfor q≡ E(=τ)∈ Dh
oτ τ . Finally we need to verify that M/|=Hindeed
holds. Let so∈ H. By definition M/|=sif and only if kskM/,g Tfor every g. An induction over the
structure of syields the desired result: If sis an equality of the form (l=τr), then kl=τrkM/,g Tiff
4We take a slight indirection here: We first assume negation is part of the red signature, to make sure there exists an element
in n∈ Doo that is the interpretation of negation. Sadly, it seems there is currently no easier way to enforce its existence; a more
convenient way would be to show that there is an extensional model that satisfies L¬(n) without having negation in the signature,
cf. [6, 3] for details on these semantic L-properties.
5The fact that Msatisfies property bfollows directly from the fact that M ∈ Mβbf(Σ).
9
klkM/,g ≡ krkM/,g. Since l=τr∈ H]
., we know M |=l=τrand, consequently, that Eϕ(l)≡ Eϕ(r)for every
assignment ϕ. It follows that Eh
ϕ(l[=\=]) ≡ Eh
ϕ(r[=\=]) and hence, by definition of M/and the induction
hypothesis, I/(=τ)(krkM/,g,krkM/,g ) = Tand thus kl=τrkM/,g T, for every g. For parameters pΣ=,
the proposition follows directly. For complex formulas of the form (s sn)and for abstractions, we apply the
induction hypotheses to every sub-formula. This construction yields:
Theorem 5 (Bridged Model Existence).Let H ∈ Hbe a Σ=-Hintikka set. Then there exists a Σ=-Henkin
model Msuch that M |=H.
Acknowledgements.
We thank an anonymous reviewer for very valuable feedback to this work (cf. footnote 1).
References
[1] Peter B. Andrews. An Introduction to Mathematical Logic and Type Theory. Applied Logic Series. Springer,
2002.
[2] Christoph Benzmüller and Peter Andrews. Church’s type theory. In Edward N. Zalta, editor, The Stanford
Encyclopedia of Philosophy, pages pp. 1–62 (in pdf version). Metaphysics Research Lab, Stanford University,
summer 2019 edition, 2019.
[3] Christoph Benzmüller, Chad Brown, and Michael Kohlhase. Higher-order semantics and extensionality.
Journal of Symbolic Logic, 69(4):1027–1088, 2004. Preprint: http://christoph-benzmueller.de/papers/
J6.pdf.
[4] Christoph Benzmüller, Chad Brown, and Michael Kohlhase. Semantic techniques for cut-elimination in
higher-order logics. SEKI Report SR-2004-07, Saarland University, Saarbrücken, Germany, 2004. Preprint:
http://christoph-benzmueller.de/papers/R37.pdf.
[5] Chad E. Brown. Set Comprehension in Church’s Type Theory. PhD thesis, Department of Mathematical
Sciences, Carnegie Mellon University, 2004.
[6] Chad E. Brown. Automated Reasoning in Higher-Order Logic. Set Comprehension and Extensionality in
Church’s Type Theory, volume 10 of Studies in Logic. College Publications, 2007.
[7] Alonzo Church. A formulation of the simple theory of types. J. Symb. Log., 5(2):56–68, 1940.
[8] Alexander Steen. Extensional paramodulation for higher-order logic and its effective implementation Leo-III.
PhD thesis, Freie Universität Berlin, Berlin-Dahlem, Germany, 2018. Published as DISKI – Dissertations in
Artificial Intelligence, Vol. 345, AKA Verlag and IOS Press, 2018. ISBN 978-1-61499-919-5.
10
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
In this report we present a variety of model existence theorems for classical higher-order logics. The extend the well-known abstract consistency methods to higher-order logics and pinpoint the contribution of extensionality. The results reported here weaken the saturation precondition needed in earlier model existence theorems, making them appliccable to cut-free higher-order logics. This report reprints the article "Higher Order Semantics and Extensionality" under review by the Journal of Symbolic Logic and develops the cut-free extensions in detail.
Article
Full-text available
In this paper we re-examine the semantics of classical higher-order logic with the purpose of clarifying the role of extensionality. To reach this goal, we distinguish nine classes of higher-order models with respect to various combinations of Boolean extensionality and three forms of functional extensionality. Furthermore, we develop a methodology of abstract consistency methods (by providing the necessary model existence theorems) needed to analyze completeness of (machine-oriented) higher-order calculi with respect to these model classes.
Article
Automation of classical higher-order logic faces various theoretical and practical challenges. On a theoretical level, powerful calculi for effective equality reasoning from first-order theorem proving cannot be lifted to the higher-order domain in a simple manner. Practically, implementations of higher-order reasoning systems have to incorporate procedures that often have high time complexity or are not decidable in general. In my dissertation, both the theoretical and the practical challenges of designing an effective higher-order reasoning system are studied. The resulting system, the automated theorem prover Leo-III, is one of the most effective and versatile systems, in terms of supported logical formalisms, to date.
Article
Of the languages of symbolic logic, typed lambda-calculus (a formation of higher-order logic) is the most suitable for use in automated mathematical systems. This book begins with a study of propositional calculus and first-order logic, then provides a complete introduction to typed lambda-calculus. CONTENTS: Introduction. Propositional Calculus. First-Order Logic. Probability and Refutability. Further Topics in First-Order Logic. Type Theory. Formalized Number Theory. Incompleteness and Undecidability. Appendix: Summary of Theorems. Bibliography. Index.
Article
Many mathematical and computational concepts can be represented in a natural way using higher-order logic. Consequently, higher-order logic has become an important topic of research. /Automated Reasoning in Higher-Order Logic/ presents both a theoretical analysis of fragments of higher-order logic as well as a complete automated search procedure for an extensional form of higher-order logic. The first part of the book provides a detailed presentation of the theory (syntax and semantics) of fragments of higher-order logic. The fragments differ in the amount of extensionality and set comprehension principles included. Three families of sequent calculi are defined and proven sound and complete with respect to appropriate model classes. Using the model constructions in the book, different versions of Cantor's theorem are determined to not be provable in certain fragments. In fact, some versions of Cantor's theorem are independent of other versions (in sufficiently weak fragments). In the second part of the book, an automated proof procedure for extensional type theory is described. Proving completeness of such a higher-order search procedure is a nontrivial task. The book provides such a completeness proof by first proving completeness of the ground case and then proving appropriate lifting results. /Automated Reasoning in Higher-Order Logic/ is an essential document for researchers in higher-order logic and higher-order theorem proving. The book is also essential reading for programmers implementing or extending higher-order search procedures. Users of higher-order theorem provers can use the book to improve their understanding of the underlying logical systems.
Church's type theory
  • Christoph Benzmüller
  • Peter Andrews
Christoph Benzmüller and Peter Andrews. Church's type theory. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy, pages pp. 1-62 (in pdf version). Metaphysics Research Lab, Stanford University, summer 2019 edition, 2019.