logic as mathematics hartley rogers, jr. henry cohn
TRANSCRIPT
LOGIC AS MATHEMATICS
Hartley Rogers, Jr.
Henry Cohn
© 2001 Hartley Rogers, Jr., Henry Cohn
This material is for private circulation only and may not be further distributed.
1.
LOGIC AS MATHEMATICS
§1. THE LANGUAGE L .
We describe and study an artificial language which we callL .
The syntaxof L concerns definitions, rules, relationships, and theorems about
the symbols and formulas ofL , where these definitions, rules, relationships, and
theorems make no reference to possible meanings which we may intend to associate
with those symbols and formulas.
The semanticsof L concerns definitions and relationships which describe
how we can, in fact, attribute mathematical meaning to the symbols and formulas of
L . In particular, we shall define certain mathematical objects, called structures, to
which formulas ofL can refer, and we shall give a mathematical definition for a
relationship of truthbetween formulas and structures. We customarily assert that this
relationship holds between a given formula and a given structure by saying that the
given formula “is true” of the given structure, or by saying, equivalently, that the
given structure “satisfies” the given formula. Accordingly, we shall sometimes refer
to this relationship oftruth between formulas and structures as the relationship of
satisfactionbetween structures and formulas.
§2. THE SYNTAX OF L.
We specify a finite alphabet of basic symbols.
Basic Symbols: Names of Symbols
x y z thevariable symbols
′ theprime
| thevertical bar
2.
¬ thenegation symbol
∧ ∨ → ↔ theconnective symbols:
conjunction, disjunction,
conditional, andbiconditional
∀ ∃ theuniversal quantifier symbol
and theexistential quantifier
symbol
= theidentity symbol
P therelation symbol
[ ] the left bracketand theright
bracket
Strings:A string is an arbitrary finite sequence of basic symbols. Thelength
of a given string is defined to be the length of this sequence. For example, the length
of the string ∧x|∧[↔′′∀∃ is 10. Two strings are said to be equalif they have the
same length and the same occurrences of the same symbols in the same order. A
string S1 is said to occur as a substringof a string S2 if S1 is equal to a segment of
adjacent symbols in S2. (It follows that every string occurs as a substring of itself and
that two strings are equal if and only if each occurs as a substring of the other.) We
also include under our definition ofstring the sequence of basic symbols of length
zero. We call this string theempty string.
Variables: A string consisting of a variable symbol followed by a finite
number (possibly zero) of primes followed by a single vertical bar is called a
variable(for example: y′′′|.) In working withL , we shall sometimes use a numerical
subscript to abbreviate the primes/bar in a variable (for example, we might
abbreviate y′′′|
3.
as y3.) The purpose of thebar in a variable is to mark the termination of the sequence
of primes in that variable and thus to avoid the undesirable syntactical ambiguity of
having one variable appear as an initial substring of another (and different) variable.
Quantifiers:A string which consists of a quantifier symbol followed by a
variable is called aquantifier(for example:∀y′′′|.)
NQ-strings:A string which is either empty or composed entirely of one or
more negation symbols and/or quantifiers is called anNQ-string(examples:¬¬¬,
∀x|∃y|∀z|, and¬∀x|∃y|¬∃x|¬ .)
Admissible sets of strings.A setA of strings is said to beadmissibleif A
satisfies the following four closure rules:
(1) For every choice of variables u and v, the strings composed as
[u=v] or as [uPv] must be inA. Strings of this form are calledatomic formulas
(examples: [y|=z|] or [x′′|Px′′|] .)
(2) If a string S is inA, then the string composed as¬S must also be
in A
(3) If the strings S1 and S2 are inA, then each of the four strings
composed as [S1∧S2], [S1∨S2], [S1→S2], and [S1↔S2] must be inA.
(4) If u is a variable and S is a string inA, then the string composed as
∀uS must be inA, and the string composed as∃uS must be inA.
Note from rule (1) that an admissible collection of strings must be infinite.
Well-formed formulas: Let S be the set of all strings. Evidently,S is an
admissible set. LetWbe the intersection of all the admissible subsets ofS (we include
4.
S itself as a subset ofS.) It follows thatWmust also be admissible since, as is easily
shown,W must satisfy each of the four closure rules.W is called the set ofwell-
formed formulas(“wffs”) of L (for example,¬[∀x|∃y|[y|Px|]→¬[z′|=z′|]] is a wff.)
Henceforth, we shall customarily abbreviate expressions for wffs by omitting vertical
bars (the preceding example would be abbreviated as¬∀x∃y[yPx]→¬[z′=z′] .) The
underlying string of basic symbols of a wff is easily restored from such an
abbreviated version since the presence and position of the missing bars will be
evident from the abbreviated version of the given wff.
SinceW is admissible, we have the following:
If θ is a wff, then the string composed as¬θ is also a wff. This new
wff is called thenegationof θ.
If ϕ andψ are wffs, then each of the four strings composed as [ϕ∧ψ],
[ϕ∨ψ], [ϕ→ψ], and [ϕ↔ψ] is also a wff. These new strings are known, respectively,
as theconjunction, thedisjunction, theconditional, and thebiconditional, of ϕ
followed byψ.
If u is a variable andθ is a wff, then each of the two strings composed
as∀uθ and as∃uθ is a wff. The first of these new strings is called theuniversal
quantification ofθ with respect tou, and the second is called theexistential
quantification ofθ with respect tou.
Reading wffs. In working with the syntax ofL , it is convenient to use certain
words or phrases of ordinary language as names for the connective symbols, the
quantifier symbols, the negation symbol, and the identity symbol, when these
symbols appear in a wff. Commonly used words and phrases include:
for ¬, “not”, “it is not the case that”;
5.
for ∧, “and”;
for ∨, “or”;
for →, “only if”
([S1→S2] can also be read as “if S1 then S2”);
for ↔, “if and only if”;
for ∀u, “for every u”, “for all u”;
for ∃u, “there exists a u such that”;
for =, “equals” , “is identical with”.
These words and phrases also have intuitive value. As we shall later see, they
suggest, for sufficiently simple wffs, the semantical relationships which we will
eventually define for those wffs. For more complex wffs, these words and phrases
have diminished intuitive value. Precise mathematical treatment of the semantical
relationships of such complex wffs requires careful attention to the positions and
nesting of brackets and to the use of certain inductive definitions. Example: the
preceding example of a wff can be read as “it is not the case that -- for every x there
exists a y such that yPx - only if - z′ is not equal to z′”. (where pauses are indicated
by hyphens.)
First inductive principle for wffs. In order to show that a given set of stringsB
includes all the wffs, it is enough to show thatB is admissible.This principle is
immediate from the above definition of wff. We use this principle to prove Theorems
1, 3, and 5 below.
Bracket-occurrences. A particular occurrence of a bracket (either a left-
bracket or right-bracket) in a string S is called abracket-occurrencein S. (Exercise:
6.
use the first inductive principle to conclude that the number of occurrences of a
bracket in a wff must be positive and even.)
Bracket-count: Given a bracket-occurrenceω in a string S, we inductively
define bS(ω), thecumulative bracket-count for the occurrenceω in the stringS, as
follows:
If ω is the initial (from the left) bracket-occurrence in S, then
bS(ω) = 0.
If ω is not the initial bracket-occurrence in S, letω′ be the closest
preceding bracket-occurrence. Then
bS(ω) = bS(ω′) + 1, if ω andω′ are both left brackets;
bS(ω) = bS(ω′) – 1, if ω andω′ are both right brackets;
bS(ω) = bS(ω′), if ω andω′ are brackets of opposite kinds.
For example, the successive bracket-counts in the wff
¬[∀x∃y[yPx]→¬[z=z]]are 0,1,1,1,1,0, and the successive bracket-counts in the
string [ [ [ ] [ ] ] [ ] ] are 0,1,2,2,2,2,1,1,1,0. The bracket-count for a wffθ is
sometimes referred to as thedepth functionfor θ, since it measures the depth at
which each bracket-occurrence lies in the nesting of the bracket-occurrences in the
wff θ. The maximum value of the depth function for a wffθ is called thedepthof θ.
Theorem 1. Letθ be a wff. Then the following statements must be true:
(a) the final basic symbol ofθ must be a right bracket; the initial bracket-
occurrence ofθ must be a left bracket; and the bracket-count at each these two
bracket-occurrences must be 0.
7.
(b) the bracket-count at each of the other bracket occurrences inθ must be
greater than 0.
Proof.Let B be the collection of all strings which satisfy (a) and (b). It is easy
to show thatB is admissible. We consider conditions (1) – (4) in the definition of
admissible set.(1) holds, since every atomic formula has exactly two bracket-
occurrences, each with bracket-count = 0. (2) is immediate, since the bracket-count
for ¬S must be the same as for S. For (3), we note that if strings S1 and S2 each
satisfy (a) and (b), it is evident from the definition ofbracket-countthat the string
composed as [S1→S2] must also satisfy (a) and (b), since the value of the bracket-
count in [S1→S2] for any given bracket-occurrence in S1 must be greater by 1 than
the value of the bracket-count in S1 for that bracket-occurrence; and similarly for any
given bracket-occurrence in S2. The same argument applies to strings composed as
[S1∧S2], [S1∨S2], or [S1↔S2]. The argument for (4) is the same as for (2). Hence, by
the inductive principle, every wff must satisfy (a) and (b).
Subformulas. Let ϕ andθ be wffs. We say thatϕ occurs as a subformulaof θ
if ϕ occurs as a substring ofθ. (Note that there may be more than one occurrence of
ϕ as a subformula ofθ.)
Corollary 2. Consider given occurrences of a wffϕ and a wffψ as subformulas of a
wff θ. It must be the case that either: (i) the occurrence ofϕ is a substring of the
occurrence ofψ, or (ii) the occurrence ofψ is a substring of the occurrence ofϕ, or
(iii) the occurrences ofϕ andψ do not overlap (that is to say, there is no occurrence
of a basic symbol inθ which is common to the given occurrences ofϕ andψ.)
8.
Proof.This fact follows as a corollary to Theorem 1. It is enough to show that
if the occurrences forϕ andψ overlap inθ, then one of these occurrences must be a
substring of the other. Assume that the given occurrences ofϕ andψ overlap.
Case 1: The occurrences forϕ andψ share the same final bracket-occurrence.
Then, ifϕ andψ have the same length, then they must occur as the same substring of
θ, and hence each is a substring of the other. Ifϕ andψ do not have the same length,
then the occurrence of the shorter must be a substring of the occurrence of the longer.
Case 2: The final bracket-occurrences for the given occurrences ofϕ andψ in
θ are distinct. Letω1 be the final bracket-occurrence forϕ, and letω2 be the final
bracket-occurrence forψ. By Theorem 1, both must be right brackets. Without loss
of generality, we can assume thatω1 appears to the left ofω2 in θ. Since the
occurrences ofϕ andψ overlap, and since, by Theorem 1,ω1 is the final symbol inϕ,
ω1 must appear in the occurrence ofψ. Let ω1* be the initial bracket-occurrence inϕ,
and letω2* be the initial bracket-occurrence inψ. We consider three subcases:
(a) ω1* = ω2* . This is impossible because it implies that
bψ(ω1) = bϕ(ω1) = 0, contrary to Theorem 1 forψ.
(b) ω2* precedesω1* in θ. Thenω2* is not in ϕ, andϕ must be a substring of
ψ.
(c) ω1* precedesω2* in θ. Then the change in bracket-count fromω1* to ω1 in
θ must be 0 by Theorem 1 forϕ. This implies that the change fromω2* to ω1 must be
negative by Theorem 1 forϕ; but this implies that the change fromω2* to ω1 must be
negative inψ, since it is the same interval of count for bothϕ andψ. This contradicts
Theorem 1 forψ. Thus (b) is the only possible subcase, and Corollary 2 is proved.
Theorem 3. If a wffθ has more than two bracket-occurrences, then it has a unique
(one and only one) bracket-occurrenceω with the following three properties:ω is a
9.
right-bracket;bθ(ω) = 1; andω is immediately followed a connective. Furthermore,
the next bracket-occurrence afterω in θ must be a left-bracket.
Proof.Consider the setΣ of all wffs θ which satisfy Theorem 3. By Theorem
1, each of these wffs has a final right-bracket of depth 0. Furthermore,Σ must
contain all the atomic formulas and must be closed under negation and
quantification. It follows by the definition of bracket-count thatΣ is closed under
conjunction, since the new combined wff will have exactly two right brackets of
depth 1, the first of which must be directly followed by the conjunction symbol,
while the second must be directly followed by a final right bracket; similarly for
disjunction, conditional, and biconditional. ThusΣ is an admissible set, and
Theorem 3 follows by the inductive principle. If a wffθ has more than two bracket-
occurrences, then the unique connective-occurrence specified in Theorem 3 is
referred to as theprincipal connective ofθ.
Corollary 4. A given wffθ must be exactly one of the following: an atomic formula, a
quantification, a negation, a conjunction, a disjunction, a conditional, or a
biconditional.
Proof.Consider the initial symbol ofθ. If it is neither a negation symbol nor a
quantifier symbol and if the wff is not an atomic formula, then, by the inductive
principle, it must be a bracket and, by Theorem 3, there must be a unique principal
connective, say∧. Thusθ = [S1∧S2]. Moreover, S1 and S2 must both be wffs,
otherwise the collection of strings obtained by removingθ from W would be an
admissible set, contradicting our assumption thatθ is a wff. (A weaker version of
Corollary 4 can be obtained by putting “at least one” in place of “exactly one”. This
10.
weaker version is immediate by the inductive principle and does not require Theorem
3.)
Subformula-occurrences. By asubformula-occurrencein a given wffθ, we
mean a subformulaϕ coupled with a particular locationof ϕ in θ. (We can think of
this subformula-occurrence as the ordered pair (ϕ,i), where the occurrence begins at
the i th symbol ofθ.) We define Subocc(θ) to be the set of all subformula-occurrences
in θ. If a is a member of Subocc(θ), we define |a| to be the wff part ofa. Thus when
a andb are two different occurrences of the same wffϕ as a subformula ofθ, we
will have thata ≠ b but |a| = |b| = ϕ. Let a andb be members of Subocc(θ). We say
thata is contained inb if a occurs as a substring of the occurrenceb, and we say
thata is properly contained inb if a is contained inb anda ≠ b. (That is to say, ifa
is contained inb and |a| is shorter than |b|.) As an abbreviation for “a is properly
contained inb”, we write: a π b. We say thata is animmediate predecessorof b if
a π b and if for noc in Subocc(θ) is true thata π c π b. Finally, we say thatb
is animmediate successorof a if a is an immediate predecessor ofb.
Theorem 5. For any given wffθ, the relationπ has the following properties:
(i) It is never true thata π a.
(ii) If a π b andb π c , thena π c .
(iii) There exists a uniqueb in Subocc(θ) such that for alla ≠ b,
a π b, and for thisb, |b| = θ.
(iv) If |a| ≠ θ, a has exactly one immediate successor.
(v) Everya has either 0, 1, or 2 immediate predecessors.
(vi) If a has no immediate predecessor, then|a| must be an atomic
wff; if a has exactly 1 immediate predecessor, then|a| must be
11.
either a negation or a quantification; and ifa has 2
immediate predecessors, then|a| must be either a conjunction,
a disjunction, a conditional, or a biconditional.
(vii) Subocc(θ) is a finite set.
Proof. (i) and (ii) are immediate from the definitions. ((i) and (ii) assert that
the relationπ is astrict partial ordering.) For (iii), let b = (θ,1) be the unique
occurrence ofθ as a subformula of itself. ((iii) asserts that the partial orderingπ has
a maximum element.) (iv) follows from Corollary 2. (v) and (vi) follow from
Corollary 4. For (vii), let n = the length ofθ. For a pair (ϕ,i) there can be no more
than n possible values for i, and for each of these values, there can be no more than n
different substringsϕ. Hence there can be no more than n2 subformula-occurrences in
Subocc(θ).
Subformula trees. In Theorem 5, (iii) and (iv) assert that that the partial
ordering is atree,and (v) asserts that this tree is abifurcating tree(that is to say, no
element can have more than two immediate predecessors.) We refer to this tree as the
subformula treefor the given wffθ. The various subformula-occurrences are referred
to asnodesof the tree.
For a given wffθ, let a0 be a node with no predecessors. (This node must be
an occurrence of an atomic formula inθ.) Since the entire tree is finite, there exists a
unique finite sequencea0, a1, …, ak such that for all i, 0 <i < k–1,ai+1 is the
immediate successor ofai and |ak| = θ. This sequence is called thebranchof the tree
determined byb0, andao is called theterminal elementof this branch. If we take the
terminal elements in the same left-to-right order in which they appear inθ, we can
12.
obtain a uniquetree-diagramfor the subformula tree. For example, forθ = the wff
¬[∀x∀y[yPx]→¬[z=z]], we have the tree-diagram:
¬[∀x∀y[yPx]→¬[z=z]]
ÿ
∀x∀y[yPx]→¬[z=z]]
/ \
∀x∀y[yPx] ¬[z=z]
ÿ ÿ
∀y[yPx] [z=z]
ÿ
[yPx]
In this example, there are two branches; the left branch has 5 nodes, and the right
branch has 4.
The rank of a wff. We define therank of a wff to be the maximum length of a
branch in its subformula tree, where we take the distance from a subformula-
occurrence to its immediate successor to be 1 (thus an atomic formula has rank 0.)
We now have three measures of complexity for a wff:length, rank,anddepth. In the
above example, the given wff has length = 25 (including bars), rank = 4, and depth =
1. It is easy to show that for any wff, we must always have:depth <rank < length.
Corollary 6. Every wff has a finite rank.
Proof. From Theorem 5 (vii).
13.
As a further corollary to Theorem 5, we have:
Second inductive principle for wffs. Let P be an alleged property of wffs. LetP(r) be
the assertion that all wffs of rank <r have the property. If we can show (i) thatP(0),
and (ii) that for all r > 0, P(r) impliesP(r + 1), then we can conclude that all wffs
have the property.
Defining a function onWby tree-induction. Let f be a function from the setW
of all wffs into some specified collectionQof mathematical objects. (Qmight be the
setN of non-negative integers, or it might be a set of algebraic objects of a more
complex kind.) We say f istree-inductiveif there exist eightdefining functionsh0,
h¬, h∧, h∨, h→, h↔, h∃, and h∀ with the following properties:
(1) h0: W0 → Q, where W0 is the set of atomic wffs, and h0 has the property
that for any wffϕ in W0, f(ϕ) = h0(ϕ). Thus h0 can be seen as prescribing the
behavior of f onW0.
(2) h¬: Q → Q, with the property that f(¬ϕ) = h¬(f(ϕ)). Thus h¬ prescribes
the value of f for¬ϕ, given the value of f forϕ.
(3) h∧: Q × Q → Q, with the property that f([ϕ∧ψ]) = h∧(f(ϕ), f(ψ)). Here
h∧
prescribes the value of f for [ϕ∧ψ], given the values of f forϕ andψ.
(4) – (6) Similarly to (3) for h∨, h→, and h↔.
(7) Let V be the set of all variables in L. Then h∃: V × Q → Q. Here f(∃uϕ)
= h∃(u, f(ϕ)), and we see that h∃ prescribes the value of f for∃uϕ, given the variable u
and the value of f forϕ.
(8) Similarly, we have h∀: V × Q → Q, and f(∀uϕ) = h∀(u, f(ϕ)).
14.
Examples. (a) LetQ= N, the set of non-negative integers. Let f(θ) = the rank
of θ. f is tree-inductive, since it is prescribed by the eight defining functions: h0(ϕ) =
0; h¬(n) = n + 1; h∧(m, n) = h∨(m, n) = h→(m, n) = h↔(m, n) = max{m, n} + 1; and
h∃(u, n) = h∀(u, n) = n + 1.
(b) Let Q = N. Let f(θ) = the depth ofθ. f is tree-inductive; the
defining functions are the same as for example (a), except that h¬(n) = h∃(u, n) =
h∀(u, n) = n.
(c) Let Q = N. Let f(θ) = the length ofθ. For any variable u, let
ln(u) = the length of the string u. f is tree-inductive, since it is prescribed by the
following eight defining functions: h0([uPv]) = h0([u=v]) = ln(u) + ln(v) + 3; h¬(n) =
n + 1; h∧(m, n) = h∨(m, n) = h→(m, n) = h↔(m, n) = m + n + 3; h∃(u, n) = h∀(u, n) =
ln(u) + n + 1.
Theorem 7. For any eight defining functions of the forms described above, and for
any wffθ: (i) there is a uniquefunctionfθ from the nodes of the subformula tree ofθ
to Qsatisfying the inductive conditions prescribed by those defining functions; (ii)
for any other occurrence ofθ as a node in the subformula tree of some other wffψ,
we havefψ(θ) = fθ(θ); and (iii) if we definef(θ) = fθ(θ), we see that this functionf
from W to Q is uniquelydefined by these eight defining functions.
Proof. If (i) fails, consider a node of lowest rank at which there is either no
value of fθ or more than one value for fθ; the contradiction is immediate. For (ii) and
(iii), we see that if we have two different subformula occurrences ofθ in two
possibly different subformula trees, then the subtrees determined by these two
occurrences ofθ must be identical.
15.
We can summarize the result of Theorem 7 as:
Principle of tree-induction for wffs. Given a setQand any eight defining functions
as described for Theorem 7, there exists a unique functionf from W(the set of all
wffs) into Q such that for any wffθ, f satisfies the prescribed conditions as they
apply toθ.
As an application of Theorem 7, we have:
Exchange principle for tree-induction: Let f be a function defined onW by tree-
induction; letθ be a wff; letϕ occur as a subformula ofθ; let ψ be any wff such that
f(ψ) = f(ϕ); and letθ′ be a wff which results when the stringψ is substituted for the
string ϕ at one or more occurrences ofϕ in θ. Thenf(θ′) = f(θ). [In other words, if
we replace any subformula by another wff with the same f-value, the the f-value of
the entire wff must remain unchanged.]
Proof.The tree ofθ′ results from replacing the subtree forϕ by the tree ofψ
at one or more nodes in the tree ofθ. Since f(ψ) = f(ϕ), the inductive steps at all
unreplaced nodes in the tree ofθ remain unaffected by the replacement..
16.
§3. ALGORITHMS AND DECISION PROCEDURES
A symbolic functionf is a function (or “mapping”) whose inputsare certain
finite strings of symbols from some finite set Ai of symbols (theinput alphabet for f)
and whose outputsare finite strings of symbols from some (possibly different) finite
set Ao of symbols (theoutput alphabetfor f.) We shall use the following notations in
connection with a symbolic function f.Si will be the set of all finite strings of
symbols from Ai; D will be an infinite subset ofSi consisting of those members ofSi
which will
serve as possible inputs to f; andSo will be the set of allfinite strings of symbols
from Ao. D is called thedomain of f. Thus asymbolic functionf from D into So is a
subset ofD × So; such that for every member s ofD, there is a unique member t ofSo
such that (s,t) is a member of f. We can then express the assertion that (s,t) is a
member of f by writing: f (s) = t.
Examples. (1) Let f be the symbolic function which has, as its domain, the set
S of all finite strings of basic symbols ofL , and which gives, as corresponding
output for each input string S, the binary expression for the length of S. Here,Si = S
= D; Ao consists of the two symbols ‘0’ and ‘1’; andSo is the set of all finite strings
from Ao.
(2) Let g be the function which has, as its domain, the set of all wffs
of L of depth > 0, and which gives, as corresponding output for each input wff, the
symbol for the principle connective of that wff. Here Ao = { ∧,∨,→,↔}.
(3) Let h be the function which has, as its domain, the setS of all
finite strings of basic symbols ofL , and which gives, as corresponding output for
each input S, the symbol ‘1’ when S is a wff and the symbol ‘0’ when S is not a wff.
17.
Let D be the domain for a given symbolic function f, and letC be a subset of
D. If a symbolic function f gives output ‘1’ for each input fromC, and output ‘0’ for
each input which is inD but not fromC, we say that f is thedecision functionfor C
on D. In example 3 above, h is the decision function forW on S.
Algorithms. [The following introduction to algorithms is informal; a more
formal treatment will appear in a later in this text.] By an “algorithm” for a symbolic
function f, mathematicians have traditionally meant a finite set of instructionsby
means of which one can carry out, for any input s for f, a well-defined and
deterministic hand-computation producing f(s). We now sharpen this to a narrower
and more specific definition: Analgorithmfor a symbolic function f is a finite,
dedicated, digital computerwhich accepts any input s for f and eventually produces
the corresponding output f(s). (Evidently, a general-purpose computer, together with
an appropriate fixedprogram, can also be viewed as an algorithm for f.) We also
make these assumptions: (1) Given a specific input s, the computation determined by
the algorithm is carried out in a sequence ofcomputation-steps, where the nature of
each step is determined by the current internal state of the computer and its memory.
(2) The content and duration of a computation-step can be defined as the actions
which occur during a single cycle of the computer’s timing clock. (3) For each input
s, the computer produces the output f(s) after a finite number of computation-steps,
where the number of steps depends on the particular input. (4) An input s is given to
the computer as a finite sequence of input symbols on a tape. Thisinput tapeis read
at the rate of one-symbol-per-step as the computation proceeds. The output f(s) is
printed on anoutput tapeat the rate of one-symbol-per-step as the computation
terminates. (5) Additional memory (blank and erasable) is provided to the computer,
in finite increments, as needed during a computation.
18.
Note that a given symbolic function may have more than one algorithm, and that one
algorithm for a given symbolic function may be significantly more efficient (take
fewer steps) than another algorithm for that function.
Decision procedures. For a given domainD and subsetC, an algorithm for
the decision function forC on D is called adecision procedure forC on D.
Computability. Let f be a given symbolic function. If there is an algorithm for
f, we say that f iscomputable. If the decision function forC on D is computable,
we say, equivalently, that thedecision problem forC on D is solvable.
Examples. In example 1 above, in order to compute f for a given input string
in S, we need only count the number of symbols in that string and give the result in
binary notation. It is easy to write a single finite program (for a general purpose
computer) which will do this for all strings inS. Hence, by definition, f is
computable. In example 2 above, in order to compute g(ϕ) for a given wff ϕ of depth
> 0, our algorithm need only calculate the bracket count forϕ and find the symbol in
ϕ that immediately follows the first right-bracket of depth 1. We give this symbol as
output. Thus g is computable. We now turn to example 3. In Theorem 8 below, we
show that the decision problem forW on S is solvable.
Theorem 8. There exists an algorithm such that for any string S inL as input, this
algorithm produces output‘1’ when S is a wff, and produces output ‘0’ when S is not
a wff.
Proof.Let S be a given string. Let n be the length of S. We give an informal
outline of the algorithm:
Part 1. Make a list of all substrings of S which happen to be atomic
19.
formulas; call these wffsα1, α2, …, αk. Make a list of all substrings of S which
happen to be variables; call these variables u1, u2, …, up. Each of these two lists must
be finite.
Part 2.The computation now proceeds in stages(where each stage can be
accomplished in a finite number of computation-steps.
Stage 0.See if S occurs in the listα1, …, αk. If so, we have found
that S is a wff of rank 0; we end the computation and give output ‘1’. If not, we have
found that S cannot be a wff of rank 0, and we go on to stage 1.
Stage 1.Make a list of all wffs of rank <1 which can be constructed
from α1, α2,. …, αk with repetitions permitted and with new quantifier variables (if
any) restricted to u1, …, up. There can only be finitely many wffs on this new list. See
if S occurs in this list. If so, we have found that S is a wff of rank 1; we end the
computation and give output ‘1’. If not, we have found that S cannot be a wff of rank
< 1, we go on to stage 2.
……………..………………………………………………………………………..
Stage m+1, for m <n – 2 . Make a list of all wffs of rank <m + 1
which can be constructed from the list of wffs of rank <m made in stage m, with
repetitions permitted and any new quantifier variables restricted to the uj, …, up.
There can be only finitely many wffs on this new list. See if S occurs in this list. If
so, we have found that S is a wff of rank m + 1; we end the computation and give
output ‘1’. If not, we have found that S cannot be a wff of rank <m + 1; we go on to
stage m + 2. If necessary, proceed in this way through stage m = n – 1.
Finally, we have:
Stage n.End computation; give output ‘0’.
20.
To complete our proof of Theorem 8, we note, from the definitions of
subformula-treeandrank, that for any m, a wff of length m must have rank < m.
Since S has length n, it follows that if S is a wff , then S must have been constructed
by stage n – 1, and our algorithm must have given output ‘1’. If we do not find that S
is a wff by stage n – 1, we know that S is not a wff, and we give output ‘0’ at stage n.
Computational complexity. As we shall see, the decision procedure described
in Theorem 8 can require a large amount of computation time (as measured by
number of computation-steps) for an input string of only moderate size. An algorithm
of this kind is said to have a high level ofcomputational complexity. We make the
following definitions. Let f be a given symbolic function fromD into So. Assume
that we have a particular algorithm for f. Let s be any given finite string inD. We
define ||s|| to be the length of s, and we define #(s) to be the number of computation-
steps used by our algorithm to compute f (s). For any given n > 0, we then define
T (n) = )s(#maxn||s|| ≤
.
The function T(n) is positive and non-decreasing. It is called thetime-complexity
functionfor the given algorithm. It is customary to compare time-complexity
functions by comparing their respective asymptotic rates of growth. An algorithm
with complexity function T(n) is said to haveat-most-polynomial growthif there is a
positive integer k such that .)n(T
nlim
k
n∞=
∞→. We abbreviate this last limit by
writing ∞→)n(T
nk
(we use analogous abbreviations in other cases below.) An
21.
algorithm with complexity function T(n) is said to haveat-most-exponential growth
if there is a real number a > 1, such that ∞→)n(T
an
It is said to haveat-least-
exponential growthif there is a real number a > 1 such that ∞→na
)n(T. An algorithm
with complexity function T(n) is said to havesuperexponential growthif for every
real number a > 1, ∞→na
)n(T. It is easily shown, from these definitions, that an
algorithm which has at-most-polynomial growth cannot have at-least-exponential
growth and that an algorithm which has at-most-exponential growth cannot have
superexponential growth.
An algorithm which has at-most-polynomial growth is conventionally viewed
as “fast”. An algorithm which has at-least-exponential growth is conventionally
viewed as “slow”. An algorithm which has superexponential growth is sometimes
viewed as “impracticably slow”.
Theorem 9. The algorithm given in the proof of Theorem 8 has superexponential
growth.
Proof.Let T(n) be the time complexity function for the algorithm given in the
proof of Theorem 8. We prove the following lemma, from which Theorem 9 will
directly follow.
Lemma. Forn > 15, T(n) >1n22
−
.
Proof of lemma.Consider the string s = [x|Py|]∨[x|=y|]. Note that s is not a
wff (it fails Theorem 1) and that the length of s is 15. Applying the algorithm to s,
we
22.
find that the computation must continue through stage 14. Note that the list for stage
0 has 2 atomic wffs and that 2 =022 . If we consider only connections which use the
single connective∨, and if we disregard all uses of quantification, we see that the list
at stage 1 must contain at least22 =122 wffs, and that the list at stage 2 must contain
at least(22)2 =222 wffs. By the same argument, at stage 14 we must have at least
1422 wffs in the new list. Since listing and comparing each of the wffs in this list with
the original input string must require at least one computation-step, we have T(15) >1422 . Similarly, for any input s with n >15, we have T(n) >
1n22−
. This proves the
lemma.
To complete the proof of Theorem 9, we need only show that for every real a > 1,
∞→− n2 a2
1n
. This is easily proved by use of L’Hopital’s rule together with the fact
that ∞→∞→ )n(flogifonlyandif)n(f . (L’Hopital’s rule asserts that if f and g are
differentiable, increasing, positive-valued functions on the positive real numbers,
then ∞=�∞=∞→∞→ g
flim
'g
'flim
xx.)
Note that T(15) >1422 is already a very large number. If we perform one
computation-step per nanosecond and blindly apply the prescribed algorithm to the
input [x|Py|]∨[x|=y|], the computation will require at least 104910 centuries. Note also
that, however absurd this algorithm may seem, it has served a useful purpose for us:
it has provided a proof that the decision problem forW on S is, in fact, solvable.
23.
We now turn to the task of finding a faster algorithm for the decision function
for W on S. The following theorem will be useful for this purpose.
Theorem 10. Letθ be a wff. For m >0, let km be the number of bracket-occurrences
of depth m inθ. Then k0 = 2, and for each m > 0, we have: (a) km = 4q, for some q
> 0; and (b) the bracket-occurrences of depth m determine q non-overlapping
substrings ofθ such that each substring begins with a bracket-occurrence of depth
m, ends with a bracket-occurrence of depth m, and includes two other bracket-
occurrences of depth m. Furthermore, (c) in each such substring, the first bracket-
occurrence of depth m is a left-bracket, the second bracket-occurrence of depth m is
a right-bracket, the third bracket-occurrence of depth m is a left-bracket, and the
fourth and last bracket-occurrence of depth m is a right-bracket. Furthermore, (d)
within in each of these substrings, all bracket-occurrences of depth≠ m have depth
greater than m, while (e) between any two such substrings, there is at least one
bracket-occurrence of depth less than m.
Proof. It will be enough to show that the set of all strings satisfying the
properties asserted forθ in Theorem 10 is admissible. Evidently, every atomic
formula satisfies these properties, and the negation or quantification of a string
having these properties must have these properties. It remains to show, for example,
that the conjunction of any two strings having these properties must also have these
properties. This is immediate when we note that the depths of all bracket-occurrences
in the two component strings (of the conjunction) are increased by 1 when the
conjunction is formed, that the new conjunction must have exactly four brackets of
depth 1, and that (for (e)) any two four-bracket segments of depth m > 1 must either
24.
lie in the same component string and already satisfy (e) or else lie each in a separate
component string and be separated by a bracket-occurrence of depth = 1.
Corollary 11. For the quantity km in Theorem 10, we have the bound: km < 2m+1.
Proof.This is immediate from the inductive construction in the proof of
Theorem 10.
We now use Theorem 10 to get an improved algorithm for the decision
function forW on S.
Theorem 12. There exists an algorithm of at-most-polynomial growth for the
decision function forW on S.
Proof.We first describe an algorithm and observe that it is a decision
procedure for a certain setC in S. We then show thatC = W. Finally, we show that
the algorithm has at-most-polynomial growth.
Let S be a given string of basic symbols ofL .
Part 1 of the algorithm.The computer makes a single pass through S
from the left. For S to be a wff, each of the following must hold: (a) There must be at
least one bracket-occurrence in S; the first bracket-occurrence in S must be a left-
bracket; and the string of symbols which precedes this first bracket-occurrence must
be a possibly empty NQ-string. (b) During this pass, the computer also considers
each substring of S whose first and last symbols are brackets and which contains no
other bracket-occurrence. There are four cases: (i) left-bracket then left-bracket– in
this case, the substring between must be a possibly empty NQ-string; (ii) left-bracket
then right-bracket– in this case, the substring between and including these brackets
must be an atomic wff; (iii) right-bracket then left-bracket– in this case, the
25.
substring between must be a single connective followed by a possibly empty NQ-
string; (iv) right-bracket then right-bracket– in this case, the substring between must
be empty. The last bracket-occurrence in S must be a right-bracket and must be the
last symbol in S. During this first pass through S, the computer also keeps a running
bracket-count, verifies that the conditions in Theorem 1 hold, and notes the highest
value of the bracket-count that occurs.
If S fails to meet any of the requirements in part 1, the calculation terminates
with output ‘0’. If the pass succeeds, the computer goes on to part 2.
Part 2.Let d be the maximum value of the bracket-count in part 1.
For each m, 1 <m < d, the computer makes a further pass through S to verify that
conditions b, c, d, and e in Theorem 10 are satisfied for the bracket-occurrences in S
of depth m.. If any of these passes fails, the computation ends with output ‘0’. If
each pass succeeds, the computation ends with output ‘1’.
If a string S of basic symbols ofL gives output ‘1’, we say that S is
acceptable. Hence, by definition, the algorithm provides a decision procedure for the
set of acceptable strings in S. In order to prove that the algorithm is a decision
procedure forW, it remains to prove two propositions:
Lemma13. Every wff is an acceptable string.
Lemma14. Every acceptable string is a wff.
Proof of Lemma 13.By a proof identical to the proof of Theorem 10,
except that “acceptable strings” replaces “strings satisfying the properties asserted for
θ in Theorem 10”, we find that the set of acceptable strings is admissible. Hence, by
the first inductive principle, every wff is an acceptable string.
Proof of Lemma14.Let S be an acceptable string. We prove Lemma
14 by mathematical induction on the depth of S, where thedepthof an arbitrary
string S is defined to be the maximum value of the bracket-count of S.
26.
Let P(m) be the assertion that for everyl , 0 < l < m , every acceptable string
of depthl is a wff..
Basis step. We showP(0). If S is acceptable and has depth 0, part 1
of the algorithm assures that S is an atomic formula preceded by a possibly empty
NQ-string. Hence S is a wff.
Inductive step. Assume thatP(m) is true We proveP(m+1). Let S be
acceptable and have depth m + 1. By part 2 of the decision procedure, there must be
exactly four bracket-occurrences of depth 1, and by part 1, the second occurrence
must be followed by an adjacent connective. Let S1 be the substring of S contained.
between and including the first and second occurrences of depth 1, and let S2 be the
substring of S contained between and including the third and fourth occurrences of
depth 1. By part 2 of the decision procedure, S1 and S2 must each have depth <m.
Hence, by our assumption thatP(m) holds, each of S1 and S2 must be a wff.. Hence,
by part 1, S must be consist of a possibly empty NQ-string followed by the
conjunction, disjunction, conditional, or biconditional of the wffs S1 and S2. Hence S
is a wff.
It remains to show that the algorithm described above has at most polynomial
growth. We accomplish this by considering our computation procedure in more
detail.
Let n be the length of the input string S. We first show that the first pass, described
in part 1 of the procedure, can be accomplished in exactly n steps. In order to achieve
this, the computer must be able to carry out several simultaneous tasks as it moves
through S: (i) It must verify that the symbols, which precede the first bracket form a
possibly empty NQ-string. (ii) It must verify that the symbols which lie between a
left- bracket and a succeeding left-bracket form a possibly empty NQ-string. (iii) It
must verify that the symbols which lie between a right-bracket and a succeeding left-
bracket form a connective followed by a possibly empty NQ-string. (iv) It must
verify
27.
that the that the symbols which lie between a left-bracket and a succeeding right-
bracket form the interior portion of an atomic wff. (v) It must maintain a running
bracket-count to signal when the bracket-count returns to 0. (vi) It must maintain a
record of the highest bracket count so far. (vii) It must check that the first bracket
exists and is a left-bracket, and that the final symbol is a right-bracket.
The computer does all this by using several registers whose operations (to test
for an NQ-string, for example) can be repeated during the first pass. This NQ-string
register, for example, contains at most one symbol at any time and shows ‘E’ (for
empty string) at the beginning of a test. If it shows E, it accepts only from the set N1
= { ¬, ∀, ∃} and shows that symbol; if it contains¬, it accepts and shows only from
N1; if it contains∀ or ∃, it accepts and shows only from N2 = {x,y,z}; if it contains x,
y, or z, it accepts and shows only from N3 = { ′, |}; if it shows ′, it accepts and shows
only from N3, if it contains |, it accepts and shows only from N1; if presented with an
unacceptable symbol, it shows the symbol F and ceases to operate. If it shows | or¬
or E when a bracket appears, this indicates that it has currently found a possibly
empty NQ-string. If it shows any other symbol, this indicates that an NQ-string has
not been completed. Similar single-symbol registers can be used to test for an atomic
wff, and to show whether the current bracket count is positive. At the beginning of
each computation step, the content of the single-symbol registers is fed to the
computer’s central processing unit and helps determine the actions to be taken during
that step. We observe that part 1 requires at most n computation steps.
If the string S is not rejected during this first pass, the computer goes on to
part 2. The maximum value d of the bracket-count has been preserved in memory
and, in accord with part 2, the computer proceeds to carry out d passes through S, one
for each value of m between 0 and d. Several more single-symbol registers can be
used
28.
for these tests, as in part 1. The number of registers required is easily shown to be
independent of n, d, and m. The value of m starts at 0 and increases by 1 with each
pass; the final pass occurs when m = d. We note that each pass in part 2 can be
carried out from left to right or from right to left, since the conditions in Theorem 10
are symmetrical with regard to direction through S. Hence the successive passes can
alternate in direction. It follows that part 2 requires at most n×d computation-steps.
Since d <n, we have the upper-bound: T(n) <n2 + n. Applying the definition
of at-most- polynomial growthwith exponent k = 3, we have
∞→+
=+
≥
n
11
n
nn
n
)n(T
n2
33
. This completes the proof of Theorem 12.
(It is not hard to show that d <(1/9)n. Hence the above bound can be improved to
T(n) < (n2/9) + n.)
Invariance of complexity categories. It is a significant feature of the
complexity categories defined above (at-most-polynomial, at-least-exponential, at-
most-exponential,andsuperexponential) that relatively major changes in a given
complexity function are required to move that function out of the categoryat-least-
exponentialand into the categoryat-most-polynomial, or out of the category
superexponentialand into the categoryat-most-exponential.This implies that
relatively major changes in the fundamental nature of an algorithm will be required
to move that algorithm from one category to another. The following definitions and
theorem make this clear.
Let T(n) be a given complexity function, and let f(n) be a non-decreasing,
positive-valued function. The function T′(n) = T(n)× f(n) is called aslow-downof
T(n) by f(n). (It is the complexity that would result from taking an algorithm with
complexity T(n) and replacing each individual computation-step, for each input of
29.
length n, by a procedure which requires f(n) steps.) Similarly, the function T′′(n) =
T(n)/ f(n) is called aspeed-upof T(n) by f(n). (It is the complexity that would result
from taking an algorithm with complexity T(n) and, for each input of length n,
replacing every successive block of f(n) computation-steps by a single step.) If f(n)
has at-most-polynomial growth, we say that T′(n) is apolynomial slow-downof T(n)
and that T′′(n) is apolynomial speed-up ofT(n). Similarly forexponential slow-down
andexponential speed-up.
The following theorem is easily proved from these definitions.
Theorem 15. (a) If T(n) has at-most-polynomial growth, then a polynomial slow-
down of T(n) must still have at-most-polynomial growth.
(b) If T(n) has at-least-exponential growth, then a polynomial speed-
up of T(n) must still have at-least-exponential growth.
(c) If T(n) has superexponential growth, then an exponential speed-up
of T(n) must still have superexponential growth.
The complexity categories considered here are far from exhaustive. For example,
there exist algorithms with complexity functions that lie above all the at-most-
polynomial functions and below all the at-least-exponential functions, where “f
above g” or “g below f” means that f/g→ ∞ .
30.
§4. THE SEMANTICS OF L.
Occurrences of variables. The following syntacticalterminology will be used
in connection with the semanticsof the languageL . An occurrence of a variable v in
a wff θ is said to be a boundoccurrence if it occurs in a subformula (ofθ) of the form
∀vϕ or of the form∃vϕ. An occurrence of v inθ which is not a bound occurrence is
said to be a freeoccurrence inθ. Evidently, it is possible for a variable v to have both
free and bound occurrences in the same wff. If all occurrences of all variables in a
wff θ are bound,θ is said to be a closedwff. Closed wffs are also called sentences.
For example, [∀x∃y[xPy]→[xPx]] is not closed, since x has a free occurrence (in
fact, there are two free occurrences and two bound occurrences of x), while
[∀x∃y[xPy]→∃x[xPx]] is a sentence (with five bound occurrences of x and two
bound occurrences of y.)
Structures forL : Let D be an arbitrary, non-empty set. Let R⊆ D × D, where
D × D is the set of all ordered pairs (a,b) such that a∈ D and b∈ D. Then the
ordered pair (D, R) is called a structurefor L .
Examples:
(1) Let D be the setR of real numbers. Let R = {(s,t)
| (s,t)∈ R × R and s < t}. We refer to R as <R. Then (R, <R) is a structure forL , the
strict linear ordering of the real numbers.
(2) Let D be the setQ of rational numbers. Let R =
{(p,q) | (p,q)∈ Q × Q and p < q} = <Q. Then (Q, <Q) is a structure forL , thestrict
linear ordering of the rational numbers.
(3) Let D be the setZ of integers. Let R =
31.
{(m,n) | (m,n)∈ Z × Z and m < n} = <Z. Then (Z, <Z) is a structure forL , thestrict
linear ordering of the integers.
(4) Let D be the setZ. Let R = {(m,n) | (m,n)∈ Z × Z and
m divides n}=÷Z . (Here, “m divides n” means that n = mq for some integer q.) Then
(Z, ÷Z) is a structure forL , thepartial ordering of the integers under divisibility.
Satisfaction forL : Let α = (D,R) be a structure forL , and letσ be a
sentence (closed wff) forL . We wish to give a mathematical definition for the
concept that “σ is true ofα“ when we think of the variables ofL as ranging over D
and think of the relation symbol P as R. We will use the terminology: “α satisfiesσ“
for our desired concept, and we will write it:α |= σ. In order to obtain a
mathematical definition of “α |= σ” which accords with our intuition, we make a
wider definition which applies to wffs with free as well as bound occurrences of
variables. Recall thatV is the infinite set consisting of all the variables ofL . Let s be
a function (“mapping”) from the infinite setV to the set D. (Thus s takes variables of
L as inputs and gives members of D as outputs.) Such a function s is called an
assignmentfor the structureα, since s assigns to each variable a member of D. (We
allow a single member of D to be assigned to more than one variable, but each
variable must have a unique member of D assigned to it.) LetAα be the set of all
assignments forα, and letAα* be the set of all subsets ofAα. We use tree-induction to
define a function Sα: W → Aα* , where, for any wffθ, Sα(θ) is a member ofAα
*.
Sα(θ) will be calledthe set of assignments forα which satisfyθ. In the definition of
the function Sα which follows, note that the requirements for tree-induction, in terms
of the eight special functions of Theorem 7, are satisfied. The tree-induction goes as
follows:
32.
(1a)ϕ = [u=v]. Then Sα(ϕ) = {s ∈ Aα | s(u) = s(v)}.
(1b) ϕ = [uPv]. Then Sα(ϕ) = {s ∈ Aα | (s(u),s(v))∈ R}.
(2) Sα(¬ϕ) = {s ∈ Aα | s ∉ Sα(ϕ)}.
(3) Sα([ϕ∧ψ]) = Sα(ϕ) ∩ Sα(ψ).
(4) Sα([ϕ∨ψ]) = Sα(ϕ) ∪ Sα(ψ).
(5) Sα([ϕ→ψ]) = Sα(ψ)∪{s ∈ Aα | s ∉ Sα(ϕ)}.
(6) Sα([ϕ↔ψ]) = (Sα(ϕ)∩ Sα(ψ))∪({s∈Aα| s∉Sα(ϕ)} ∩{s∈Aα | s∉Sα(ψ)}).
Let v be a variable ofL , let c∈ D, and let s∈ Aα. We define the assignment svc ∈ Aα
by: (i) s vc (u) = s(u) for variables u different from v; and (ii) sv
c (v) = c.
(7) Sα(∃uϕ) = {s∈Aα | for some c∈D, scv ∈ Sα(ϕ)}.
(8) Sα(∀uϕ) = {s∈Aα | for every c∈D, scv ∈ Sα(ϕ)}.
For every wffθ of L , for any given structureα, and for any assignment s∈
Aα, we now writeα |= θ [s] if s ∈ Sα(θ), and we say thatα satisfiesθ under the
assignments.
The above definition ofsatisfactionby tree-induction is a central concept of
mathematical logic, and is due to Alfred Tarski. To help establish the reader’s
intuitive understanding of this concept, we repeat the definition ofsatisfactionfor L
in a different but equivalent form:
(1a) α |= [u=v] [s] if and only if (‘iff’) s(u) = s(v);
(1b) α |= [uPv] [s] iff (s(u),s(v))∈ R;
(2) α |= ¬ϕ [s] iff α |≠ ϕ [s];
(3) α |= [ϕ∧ψ] [s] iff α |= ϕ [s] andα |= ψ [s];
(4) α |= [ϕ∨ψ] [s] iff either α |= ϕ [s] or α |= ψ [s] (or both)
(5) α |= [ϕ→ψ] [s] iff either α |≠ ϕ [s] or α |= ψ [s] (or both);
33.
(6) α |= [ϕ↔ψ] [s] iff either: α |= ϕ [s] andα |= ψ [s] or else:
α |≠ ϕ [s] andα |≠ ψ [s].
(7) α |= ∃vϕ [s] iff for somec in D, α |= ϕ [svc ];
(8) α |= ∀vϕ [s] iff for every c in D, α |= ϕ [svc ].
This latter definition has the form of an induction by rank. In (7), for example, if∃vϕ
has rank r + 1, thenϕ has rank r, and we are inductively assuming thatα |= ϕ[s] has
already been defined for each s∈ Aα.
Facts about satisfaction:
Theorem 16. If assignmentss ands′ agree on the variables which occur free inθ,
then: α |= θ [s] iff α |= θ [s′].
Proof. Induction by rank: Ifθ has rank 0 (an atomic wff), the statement is
immediate by definition (cases (1a) and (1b)). In case (2), we note that the set of
variables occurring free does not change. In cases (3) – (6), we note that if s and s′
agree on the variables occurring free in [ϕ∨ψ] (for example) then s and s′ must agree
on the variables free inϕ andon the variables free inψ. In case (7), we assume that
the statement of the theorem holds for the wffϕ for all pairs of assignments s and s′
which agree on the free variables ofϕ; it then follows, for each such pair, that for any
c in D, α |= ϕ ]s[ cv iff α |= ϕ ]s[ c
v′ , and from this it follows that for any pair s and s′
which agree on the free variables of∃vϕ: α |= ∃vϕ [s] iff
α |= ∃vϕ [s′]. The proof is similar for case (8).
The following two corollaries are immediate.
Corollary 17. Ifσ is a sentence, then for any two assignmentss ands′ : α |= σ [s]
iff α |= σ [s′].
34.
Corollary 18. Ifσ is a sentence, then for every structureα, α |= σ[s] for everys iff α
|= σ[s] for somes.
Further comments and definitions. Thus, for a sentencein L , we can now
define: α |= σ to mean thatα |= σ[s] for all s (iff α |= σ[s] for some s). In this case,
we say thatσ is trueof α. If α |≠ σ, then for no assignment s can we haveα |= σ[s],
and we say thatσ is falsefor α.
For a wff ϕ which is not closed, the statement that “ϕ is true ofα” is
sometimes used to mean thatα |= ϕ [s] for all assignments s. In this case, whereϕ is
not closed, if we letϕ* be a sentence obtained by prefixing universal quantifiers toϕ
for all the variables free inϕ, we have that “ϕ is true ofα” iff α |= ϕ*.
For any setΦ of sentencesof L , (whereΦ is possibly infinite), we defineα |=
Φ to mean thatα |= σ for every sentenceσ in Φ, and we then say thatα satisfiesΦ.
Equivalently, we say thatα is a model forthe set of sentencesΦ.
Proofs about satisfaction. Our mathematical definition of satisfactionallows
us to prove, as a theorem of mathematics, that a given structureα satisfies (or does
not satisfy) a given wff under a given assignment s. We give a simple example: Letα
= (R, <R); and letσ be the sentence∃x∀y[yPx]. We shall prove that the set of all s
such thatα |= σ [s] is empty and hence conclude thatσ is false ofα. Observe that the
subformula tree forσ has 3 nodes. Starting at the bottom, we consider, at each node,
the set of satisfying assignments for that node:
Bottom nodeϕ0 = [yPx]: Here we have Sα(ϕ0) = {s∈Aα| s(y) <R s(x)}.
Next nodeϕ1 = ∀y[yPx]: Here we have Sα(ϕ1) = {s∈Aα| for all c in R,
)y(scy <R s(x)}. From the mathematical definition of <R, we see that this set is
empty.
35.
Final nodeϕ2 = ∃x∀y[yPx]: Here we have Sα(ϕ2) = {s∈Aα| for some
d in R, dxs ∈ Sα(ϕ1)}. Since Sα(ϕ1) is empty, Sα(ϕ2) must be empty. Henceα |≠ σ.
Semantical equivalence:Let ϕ andψ be given wffs (not necessarily
sentences.) We say thatϕ andψ are semantically equivalentif for everystructureβ,
Sβ(ϕ) = Sβ(ψ); that is to say,ϕ andψ are semantically equivalent if, for every
structureβ, ϕ andψ are satisfied by exactly the same assignmentsfor β; or
equivalently, if for everyβ and every assignment s forβ, β |= ϕ [s] iff β |= ψ [s)].
Theorem 19. If given wffsϕ andψ are semantically equivalent, it follows that if a
wff θ has a given occurrence ofϕ as a subformula, and ifθ′ is the result of
substitutingψ for that occurrence ofϕ, thenθ andθ′ are semantically equivalent. (In
particular, ifθ andθ′ are sentences, then for every structureβ: β |= θ iff β |= θ′ .)
Proof.For any given structureβ, and for any given wffθ, the theorem is true
by the exchange principle for tree-induction.
Let α be a given structure. We say that given wffsϕ andψ areα-equivalentif
Sα(ϕ) = Sα(ψ), that is to say, if for every assignment s forα, α |= ϕ [s] iff α |= ψ [s].
Theorem 20. Ifϕ andψ are α-equivalent, it follows that if a wffθ has a given
occurrence ofϕ as a subformula, and ifθ′ is the result of substitutingψ for that
occurrence ofϕ, thenθ andθ′ are α-equivalent.(In particular, ifθ andθ′ are
sentences, then:α |= θ iff α |= θ′.)
Proof.As for Theorem 18.
36.
Theorem 21 . The following semantical equivalences hold. (This is a partial list of
well-known examples.) In each case, the equivalence holds for any choices of the
indicated subformulas.
(1) [ϕ↔ψ] eq [[ϕ→ψ]∧[ψ→ϕ]].
(2) [ϕ→ψ]eq [¬ϕ∨ψ].
(3) ∀vϕ eq ¬∃v¬ϕ.
(4) ∃vϕ eq ¬∀v¬ϕ.
(5) ¬[ϕ∧ψ] eq [¬ϕ∨¬ψ]. [(5) and (6) are known as
(6) ¬[ϕ∨ψ] eq [¬ϕ∧¬ψ]. deMorgan’s laws.]
(7) ¬¬ϕ eq ϕ.
(8) [ϕ∧ψ] eq [ψ∧ϕ]. [(8) and (9) are
(9) [ϕ∨ψ] eq [ψ∨ϕ]. commutativity laws.]
(10) [ϕ∧[ψ1∨ψ2]] eq [[ϕ∧ψ1]∨[ϕ∧ψ2]]. [ Distributivity law.]
(11) [ϕ∨[ψ1∧ψ2]] eq [[ϕ∨ψ1]∧[ϕ∨ψ2]]. [ ′′ ′′ .]
(12) [ϕ1∧[ϕ2∧ϕ3]] eq [[ϕ1∧ϕ2]∧ϕ3]. [Associativity law.]
(13) [ϕ1∨[ϕ2∨ϕ3]] eq [[ϕ1∨ϕ2]∨ϕ3]. [ ′′ ′′ .]
(14) ∃v[ϕ∨ψ] eq [∃vϕ∨∃vψ].
(15) Provided thatv does not occur free inϕ:
∃v[ϕ∧ψ] eq [ϕ∧∃vψ].
(16) Provided thatv does not occur free inϕ:
∃vϕ eq ϕ.
(17) [u=v] eq [v=u].
(18) Let ϕ have a free occurrence ofv but no bound occurrences ofu andv,
37.
and defineϕ uv to be the result of substitutingu for all occurrences ofv
in ϕ. Then:
∃v[[u=v]∧ϕ] eq ϕ uv .
(19) Let ϕ have no occurrences ofu and letϕ vu be as in (18). Then
∃vϕ eq ∃uϕ vu , [rules foralphabetic change
∀vϕ eq ∀uϕ vu . of bound variable]
Proof.The proof for each equivalence uses ordinary mathematical reasoning
together with the definition of satisfiabilityas given above. For example, for (7) we
have:β |= ¬¬θ[s] ⇔ β |≠ ¬θ[s] ⇔ β |= θ[s]; similarly, for (14), we have:β |=
∃v[ϕ∨ψ][s] ⇔ β |= [ϕ∨ψ][s vc ] for some c⇔ for some c, either:β |= ϕ [sv
c ] or β |=
ψ[svc ] ⇔ either: for some c,β |= ϕ[sv
c ] or for some c,β |= ψ[svc ] ⇔ either:β |=
∃vϕ[s] or β |= ∃vϕ[s] ⇔ β |= [∃vϕ∨∃vψ][s].
Theorem 22. The followingα-equivalences hold forα = (R, <R) ):
(20) ¬[u=v] eq. [[uPv]∨[vPu]] .
(21) ¬[uPv] eq. [[u=v]∨ vPu]] .
Proof.As in Theorem 21, but using the specific structureα.
Definition: For any given structureβ, let Tβ = { σ | σ is a sentence ofL and
β |= σ}. T β is called thefirst-order theory of the structureβ.
Remark. Is there an algorithm for deciding, for any given pair of wffs inL , whether
or not that pair are semantically equivalent? The answer is NO. If two wffs happen
to be semantically equivalent, can we always find an elementary proof, in terms of
38.
assignments, for this fact? The answer is YES. If two wffs happen notto be
semantically equivalent, can we always find a proof, at some mathematical level, of
this fact. The answer is NO. The affirmative answer to the second of these questions
will follow from the principal theorem of §7 below.
39.
§5. A DECISION PROCEDURE FOR Tαααα, WHERE αααα = (R, <R).
We first enlargeL to a new languageL + by adding two new basic symbols: t
and f , by including the strings [t] and [f] as atomic wffs, and by augmenting the
definition of satisfactionto include: Sα([t]) = Aα and Sα([f]) = ∅ (the empty set.)
The decision procedure: Given an arbitrary sentenceσ of L, we use some of
the equivalences (1) - (21) above to transformσ into a succession ofα-equivalent
sentences inL + as follows:
(A) Use (1) and (2) to eliminate↔ and→.
(B) Use (3) to eliminate∀.
(C) Choose a subformula∃vθ, whereθ, considered by itself, has no
bound occurrences of any variables. We proceed to “eliminate” this occurrence of∃v
by a chain of equivalences as described in (D) - (I):
(D) Use (5), (6), and (7) to distribute any¬ symbols intoθ and to
remove any pairs¬¬.
(E) Use (20) and (21) to remove any remaining¬ symbols.
(F) Let θ′ be the result of (D) and (E) onθ. Use (8), (9), and (10) to
expressθ′ as a repeated disjunction of repeated conjunctions.
(G) Use (12) and (13) to express this disjunction as
[ϕ1∨[ϕ2∨[ϕ3∨...∨[ϕk-1∨ϕk]...]]] and to express each conjunctionϕi as
[ψi,1∧[ψi,2∧...∧[ψi,m-1∧ψi,m]...]] where all theψi,j containing v come last.
40.
(H) Use (14) and (15) to distribute∃v through this disjunction and, as
far as possible, through each conjunction. This may create additional occurrences of
∃v.
(I) Use (16), (17), and (18) together with the following α-
equivalences to complete the elimination of all the above occurrences of∃v. (This
completes (C) for its given occurrence of∃v and reduces by one the total number of
quantifiers appearing after (B).
(22) ∃v[v=u] eq [t].
(23) ∃v[vPv] eq [f].
(24) ∃v[[vPu1]∧[ ... ∧[vPuk]] ..] eq [t].
(25) ∃v[[w 1Pv]∧[ ... ∧[wmPv]]..] eq [t].
(26) ∃v[[[w 1Pv]∧... [wmPv]] ∧ [[vPu1]∧... [vPuk]]] eq
[[[w 1Pu1]∧...∧[w1Puk]]∧...∧[[w mPu1]∧...∧[wmPuk]] .
(J) In this way, repeating (C) as needed, we can eliminate all
occurrences of∃. The closed wff which remains must have [t] and [f] as its only
atomic subformulas. We can now use the following equivalences to reduce this wff
to a single atomic wff, which must be [t] or [f].
(27) [ϕ∧[t]] eq. ϕ.
(28) [ϕ∨[t]] eq. [t].
41.
(29) [ϕ∧[f]] eq. [f].
(30) [ϕ∨[f]] eq. ϕ.
(31) ¬[f] eq. [t].
(32) ¬[t] eq. [f].
(K) If the final wff (of L +) is [t], we see that Sα(σ) = Sα([t]) = Aα and
hence thatα |= σ, andσ is in Tα. If the final wff is [f], we see that Sα(σ) = Sα([f]) =
∅ and hence thatα |≠ σ, andσ is not in Tα.
Example. We apply the procedure toσ = ∀x∀y[[xPy]∨[yPx]]. For ease of
reading, we omit brackets whose presence in the underlying unabbreviated wff is
clear from context. Main stages of the procedure are indicated on the left. A few
numerals from Theorem 20 are also shown.
(A) not applicable
(B) ¬∃x¬¬∃y¬[ … ]
(C) Move to eliminate∃y.
(D) ¬∃x∃y[¬[xPy]∧¬[yPx]] {(5)(7)}
(E) ¬∃x∃y[[x=y ∨ yPx]∧[y=x ∧ xPy]]
(F)(G) ¬∃x∃y[ [x=y ∧ y=x]∨[yPx ∧ y=x]∨[x=y ∧ xPy]∨[yPx ∧ xPy] ]
(H) ¬∃x[ ∃y[x=y∧y=x] ∨ ∃y[yPx∧y=x] ∨ ∃y[x=y∧xPy] ∨ ∃y[yPx∧xPy] ]
(I) ¬∃x[ [x=x] ∨ [xPx] ∨ [xPx] ∨ [xPx] ] {(18); (8)(18); (17)(18); (26)}
42.
(C) Move to eliminate∃x.
(H) ¬[ ∃x[x=x] ∨ ∃x[xPx] ∨ ∃x[xPx] ∨ ∃x[xPx] ]
(I) ¬[ t ∨ f ∨ f ∨ f ] {(22);(23);(23);(23)}
(J) ¬[t] (30)
(J) [f] (32)
(K) We conclude thatσ is not in Tα.
Remarks. The algorithm described above is said to “solve the decision
problem for Tα with α = (R, <R)”. The general method of this decision procedure is
known as the “elimination of quantifiers method”. Unfortunately, this algorithm is
“slow” in the sense that it requires exponential time to complete its calculations, No
“fast” (polynomial time) algorithm is known for this decision problem.
43.
§6. THE LANGUAGES P AND 2P
The languageL is commonly known as the language offirst-order logic with
identity for a single binary relation.So far, we have presented the syntax and
semantics ofL . We next present the syntax and semantics of a language which we
shall callP. P is commonly known as the language ofpropositional logic.
The syntax of P.As with L , we begin with a finite alphabet ofbasic symbols.
These are: p q r variable symbols
′ prime
| vertical bar
¬ negation
∧ ∨ → ↔ connectives
[ ] brackets
As with L , astring is defined to be a finite (possibly empty) sequence of
basic symbols, and avariable is defined to be a string consisting of a variable symbol
followed by a finite number (possibly zero) of primes followed by a single vertical
bar. AnN-string is a possibly empty string composed entirely of negations. For every
variable v, the string composed as [v] is anatomic formula. A set A of strings is
admissibleif it is closed under the following three closure rules:
(1) Every atomic formula is inA. (HenceA must be infinite.)
(2) If string S is inA, then the string composed as¬S must be inA.
(3) If the strings S1 and S2 are inA, then so also must be the strings composed
as [S1∧S2], [S1∨S2], [S1→S2], and [S1↔S2].
44.
Well-formed formulas: Let SP be the set of all strings of basic symbols forP.
SP is an admissible set. LetWP be the intersection of all admissible subsets ofSP.
ThenWP is itself admissible and is called the set ofwell-formed formulas(“wffs”) of
P. Example: [[[p|]→[q′|]]→[r′′|]] . Henceforth, we will usually abbreviate wffs by
omitting vertical bars and brackets in each atomic formula and by using subscript
numerals for each string of primes (the preceding example would be abbreviated:
[[p→q1]→r2].)
A first inductive principleholds forP just as forL . A bracket-countfor a
string inP is defined exactly as forL . Subformulasare defined as forL . Theorem 1,
Corollary 2, and Theorem 3then hold word-for-word forP, and their proofs are the
same as forL . Corollary 4also holds with the words “a quantification” deleted.
Subformula occurrences, andsubformula treesare defined forP as forL , as
arelength, depth, and rank.Theorem 5, Corollary 6, and the second inductive
principlehold for P just as forL . Definition by tree-inductionis carried out forP as
for L , except that the twodefining functionsh∃ and h∀ are omitted. Theorem 7holds
for P as forL , except that the reference to “eight” defining functions becomes a
reference to “six”. Moreover, the principle of tree-inductionand the exchange
principle for tree-inductionhold for P exactly as forL .
Decision procedures for wffs inP. Theorems 8, 9, and 10, along with
Corollary 11, Theorem 12, and Lemmas 13and 14are proved forP as forL , but with
N-stringsin place ofNQ-strings.
The semantics of P.In the semantics ofP, we have no counterpart to the
concept ofstructurein the semantics ofL . Instead, we only use assignments, where
45.
anassignment(for P) is a function s:VP → {0,1}, with VP = the set of all variables in
P. We can think of this function s as assigning “truth” or “falsity” to every variable in
P, where, for a variable v, s(v) = 1 indicates that v is assignedtruth and s(v) = 0
indicates that v is assignedfalsity. In connection withP, the integers 0 and 1 are
sometimes referred to astruth-values.
We use tree-induction to define, for every wffθ, S(θ) = the set of assignments which
satisfyθ. Let AP be the set of all assignments. The tree-induction for S goes as
follows: (1) ϕ = [v]. Then S(ϕ) = {s ∈ AP | s(v) = 1},
(2) S(¬ϕ) = {s ∈ AP | s∉ S(ϕ)},
(3) S([ϕ∧ψ]) = S(ϕ) ∩ S(ψ),
(4) S([ϕ∨ψ]) = S(ϕ) ∪ S(ψ),
(5) S([ϕ→ψ]) = S(ψ) ∪ {s ∈ AP | s∉ S(ϕ)},
(6) S([ϕ↔ψ]) = (S(ϕ)∩S(ψ))∪({s∈AP|s∉S(ϕ)} ∩{s∈AP|s∉S(ψ)}).
Let s be an assignment, and letϕ be a wff. If s∈ S(ϕ), we say that ssatisfies
ϕ, and we write: |=ϕ[s]. Moreover, for each assignment s, we define a function
s*: WP → {0,1} as follows: s*(ϕ) = 1 if |= ϕ[s], and s*(ϕ) = 0 if |≠ ϕ[s]. Thus s
satisfiesϕ if s*(ϕ) = 1, and s does not satisfyϕ if s*(ϕ) = 0.
Theorem 23. If assignmentss ands′ (for P) agree on the variables of a wffθ (of P),
thens*(θ) = s′*(θ), that is to say: |= θ[s] iff |= θ[s′].
Proof.For any wffϕ, cases (1) to (6) above enable us to define the functions
s* and s′* on the subformula tree ofϕ by tree-induction. Since, by hypothesis, s* and
s′* must agree on the terminal elements of this tree, we must have s*(ϕ) = s′*(ϕ).
46.
Semantical equivalenceis defined forP in the same way as forL . We say that
wffs ϕ andψ are semantically equivalent if S(ϕ) = S(ψ), and an exact analogue to
Theorem 19 is immediate. Furthermore, the semantical equivalences (1), (2), and (5)-
(13) in Theorem 21 are also immediate.
The truth-tablealgorithm. Let s be an assignment, and letϕ be a wff. If we
know the values of s(v) for each variable v inθ, we can calculate s*(ϕ) by a simple
algorithm. Consider the subformula tree ofθ. We assign the value s*(ϕ) to each node
ϕ on this tree by the following inductive procedure. Each terminal node (in the tree)
is an occurrence of a variable inθ. To each occurrence of a variable v as a terminal
node, we assign the value s*(v) = s(v).. The assignment of 0 or 1 to successor nodes
in the tree is then uniquely determined and easily calculated by cases (1) – (6) above.
The algorithm is known as thetruth-table algorithm, because each of the conditions
(2) – (6) can be described by a simple table of values; for example:
ϕ ψ ϕ → ψ
1 1 1
1 0 0
0 1 1
0 0 1 .
Definitions. If a wff θ of P has s*(θ) = 1 for all assignments s inAP, we say
thatθ is valid. If s*(θ) = 1 for someassignment s inAP, we say thatθ is satisfiable.A
wff of P which is valid is also called atautology.
47.
Theorem 24. There exists a decision procedure for deciding whether or not a given
wff of P is a tautology.
Proof.We use the truth-table algorithm, and carry out the algorithm for each
possible assignment of 0 or 1 to the variables of the given wff. If there are n distinct
variables in the given wff, then 2n applications of the truth-table algorithm will be
required. It is not difficult to show that this procedure has at-least-exponential growth
and hence does not have at-most-polynomial growth. The problem of whether or not
there exists a polynomial-growth decision procedure for tautologies inP is
equivalent to the “P = NP” problem, a major unsolved problem in mathematics at
the present time.
Theorem 25. Letθ be a wff ofP. Then:
(a) θ is valid iff ¬θ is not satisfiable, and
(b) θ is satisfiable iff ¬θ is not valid.
Proof. Immediate.
Definition.Let Φ be a (possibly infinite) set of wffs ofP and letθ be a wff of
P. We say thatΦ semantically impliesθ if every assignment which satisfies all the
wffs in Φ also satisfiesθ, and we express this fact aboutΦ andθ by writing Φ |= θ.
Theorem 26. (The “compactness theorem” forP.) If Φ |= θ, then there must exist a
finite subset∆ of Φ such that∆ |= θ.
(We shall later extend this theorem toL , where it is harder to prove.) This extension
to L is one of the “great theorems” of elementary logic upon which many
applications depend.)
Proof.Let s be an assignment and letΨ be a set of wffs. We write |=Ψ[s] to
mean that |=ψ [s] for every wff ψ in Ψ. To prove a given assertion of the form
48.
“if A then B”, it is enough to prove the equivalent assertion (known as the
contrapositiveof the first assertion) “if not-B then not-A”. In the present case, not-B
asserts that for every finite subset∆ ⊆ Φ, there exists an assignment s such that
|= ∆∪{¬θ}[s], while not-A asserts that there exists an assignment s such that
|= Φ∪{ ¬θ}[s]. We assume thatΦ is infinite. Letϕ0 = ¬θ, and let {ϕ1, ϕ2, … } be a
listing of the wffs ofΦ. For m > 0, defineψm = [ϕ0∧[ϕ1∧[ϕ2∧ … ∧ϕm].]], and, by
assumption, let sm be an assignment such that |=ψm[sm]. Let v1, v2, …be a list of all
the variables ofP (for example, p, q, r, p′, q′, …). For any given m, let nm be the
maximum value of n such that vn occurs inψm. Let k(m) = max{m, nm}. Let ms be
the finite assignment defined on the domain {v1, v2, … vk(m)} such that for all i <
k(m), )v(s)v(s imim = . Note that 'mm sands need not agree on their common
domain, since sm and sm′ may differ. Note also that as m varies, there are infinitely
many distinct finite mappings sˆ m. We define an assignment s inductively as follows:
Basis step.s(v1) = 0 if {m | s m(v1) = 0} is infinite; and s(v1) = 1 otherwise. In
the latter case {m | sˆ m(v1) = 1} must be infinite.
Induction step.s(vk+1) = 0 if {m | for i < k, sm(vi) = s(vi) and sm(vk+1) = 0} is
infinite; and s(vk+1) = 1 otherwise. In the latter case, {m | for i <k, sm(vi) = s(vi) and
sm(vk+1) = 1} must be infinite).
From this construction it follows directly that |=ψm for each m, and hence
that |=Φ[s] and |=¬θ[s]. This proves the theorem.
We now extendP by introducing quantifiers over the variables ofP. The
resulting language is2P,commonly known as the language ofsecond-order
propositional logic
49.
The syntax of 2P. We enlargeP in a straightforward way by adding∃ and∀
as new basic symbols. We defineadmissiblesets for2P by adding a fourth closure
rule:
(4) If u is a variable and if the string S is in an admissible setA, then the
strings composed as∃uS and∀uS must also be inA. The wffs of2P are then defined
as the members of the intersection of all admissible sets. Thefirst induction principle
holds as forP. Subformulasandsubformulatrees are defined as before, and
analogues of Theorems, Corollaries, and Lemmas 1-14 are easily proved.NQ-strings
are defined as forL , and free and bound occurrences are defined as before.
The semantics of 2P.The set of assignments for2P is the same as the setAP
defined forP. The set S(θ) of all assignments satisfying a given wffθ is defined by a
tree induction which uses the same rules (1) – (6) as forP togther with the rules:
(7) S(∃uϕ) = {s∈AP | either 1u
0u sors is in S(ϕ)}.
(8) S(∀uϕ) = {s∈AP | both 1u
0u sands are in S(ϕ)}. .
As before, we write |=ϕ[s] for s satisfiesϕ. Free and bound occurrences of
variables, andsentences, are defined as forL . Theorem 23 holds for2P and is
proved in the same way as froL . Validity andsatisfiabilityof sentences are defined
as forL , and analogues to Theorem 16 and Corollaries 17 and 18 [with references to
structureα deleted] are proved in the same way as before. Finally,semantical
equivalencesare defined as forL , and the equivalences (1) – (16) and (19) of
Theorem 21 are proved for2P in the same way as before.
Reading wffs. A subformula of the form∃vθ can be read as “there exist a
truth-value for v such thatθ (is true)”. Similarly,∀vθ can be read as “for either truth-
50.
value of v,θ (is true).
A decision procedure for the valid sentences of2P. As we have noted, there
is no decision procedure for the valid sentences ofL . If we try elimination of
quantifiers as in§5,we find that at stage (I) we have no rules for proceeding further.
For the case of2P, however, we have such rules and can get a decision algorithm. To
carry out this algorithm, we extend the language2P to the language2P+ by adding
the atomic wffs [t] and [f] with the semantic rules: S([t]) =AP and S([f]) =∅. We
then see that:∃u[u] eq. [t],
∃u[v] eq [v] for v different from u,
∃u¬[u] eq. [t],
∃u¬[v] eq [v] for v different from u,
∃u[[u]∨¬[u]] eq. [t],
∃u[[u]∧¬[u]] eq. [f].
Example. We are given:∀p∃q∀r[[p→q]→r] (with brackets omitted as
usual).We proceed in the standard way.
¬∃p¬∃q¬∃r¬[¬[¬p∨q]∨r]
………….[[p∧¬q]∨r]
……..∃r[[¬p∨q]∧¬r]
……… ∃r[[¬p∧¬r]∨[q∧¬r]]
51.
……..[[¬p∧∃r¬r] ∨[q∧∃r¬r]]
……..[[¬p∧t]∨[q∧t]]
…∃q¬[¬p∨q]
…∃q[p∧¬q]
¬∃p¬[p∧∃q¬q]
¬∃p¬[p∧t]
¬∃p¬p
¬t
f
This decision procedure for2P can also be used for the languageP. Let ϕ be a wff of
P, and let v1, … vk be the distinct variables inϕ, listed in alphabetical order. (This
order was described in Theorem 26.) Then the sentence of2P written as
∀v1∀v2…∀vkϕ is called theuniversal closureof ϕ, and the sentence written as
∃v1∃v2…∃vkϕ is called theexistential closureof ϕ. When we apply the decision
procedure of2P to the universal closure ofϕ, we have a single computation which
produces “t” if and only ifϕ is a tautology inP. Similarly, applying it to the
existential closure ofϕ, we get “t” if and only ifϕ is a satisfiable wff inP.