Propositional

Approaches to
First-Order
Theorem Proving
David A. Plaisted
UNC Chapel Hill
May 2004
 History of AI
 Early emphasis on general
methods
 Newell Shaw Simon GPS
 Robinson 1965 resolution
 Cordell Green question answering
 Shift to specialized techniques
 Feigenbaum Expert Systems
 Is logic a suitable basis for AI?
04/09/25
 Approaches to AI
 Weak vs. strong methods in AI
 Declarative vs. procedural knowledge
 My interest: general logic-based approaches

04/09/25
 Aristotle on Deduction
A deduction is speech (logos)
in which, certain things having
been supposed, something
different from those supposed
results of necessity because of
their being so. (Prior Analytics
I.2, 24b18-20)
 Proof
 Proof is the idol before whom the
pure mathematician tortures
himself.
-- Sir Arthur Eddington
 You may prove anything by
figures. --Thomas Carlyle
 What is now proved was once
only imagined. -- William Blake
 Proof
 You cannot demonstrate
an emotion or prove an
aspiration. -- John Morley
 Prove all things; hold

fast that which is good. --

Bible, I Thessalonians
 Logic

 No, no, you're not thinking;

you're just being logical. --
Niels Bohr
 Logic is one thing and

commonsense another. --
Elbert Hubbard, The Note
Book, 1927
 Theorem Proving
 Potentially a key technology for AI
 Brittleness problem for expert
systems
 An unsolved problem
 Weak versus strong methods
 Problems with resolution
 Impact on entire field
 Importance of space versus time
 Theorem Proving on a
Computer
 Speed and accuracy of computers
 People get tired and make
mistakes
 How do people prove theorems?
 Potential applications
 Hardware verification
 Software verification
 AI and expert systems
 Robots
 Deductive Databases
 Semantic web and query
answering
 Mathematics research
 Education
Current theorem provers
 Largely syntactic
 Resolution or ME (tableau) based
 First-order provers are often
poor on non-Horn clauses
 Rarely can solve hard problems
 Human interaction needed for
hard problems

04/09/25
 How do humans prove
theorems?
 Semantics
 Case analysis
 Sequential search through space
of possible structures
 Focus on the theorem
People versus computers
 In a few areas computers are faster
 Propositional calculus
 Equational logic

 Geometry

 More to come in the future

 In general people are much better.

Why?
 Humans use semantics
 Computers use syntax in most cases
 The future

 Will provers soon be much more

powerful than they are now?
 Will they ever be much more
powerful than humans?
 Organization of the talk
 History of ATP
 Contributions of Martin Davis
 Contributions of Alan Robinson
 Achievements of Provers

 Propositional Calculus
 Propositional Resolution
 Horn Clauses
 Davis and Putnam’s Method
 The Satisfiability Threshold
 Propositional Calculus (continued)
 Performance Obtained
 Applications

 Semantics in Theorem Proving

 First Order Logic
 Clause form and Herbrand’s
theorem
 Criteria for evaluating provers
 Resolution
 Otter
 Model elimination
 Matings
 Propositional approaches to first
order logic
 Clause Linking
 Disconnection Calculus
 Disconnection Calculus Theorem
Prover
 First-Order DPLL Method
 Replacement Rules
 Definitions
 OSHL with semantics
 Comments on CADE system
competition
 David Hilbert

 Hilbert’s goal was to mechanize

mathematics. “Hilbert’s Program.”
 Goedel showed that this is
impossible.
 Automatic theorem proving tries to
mechanize what can be mechanized.
 Martin Davis
 Theorem Proving on Computers
 Davis and Putnam’s Method
 Clause Form Refutational
Theorem Proving
 Foreshadowing of Resolution
 Alan Robinson

 Resolution in First-Order Logic

 Unification in a Clause Form
Refutational Prover
 Many non-resolution methods are
still in this tradition
 First reasonably powerful
theorem prover for first-order
logic
Achievements of Provers
 Robbins Problem Solution
 Proof of Cantor’s Theorem
 Hardware Verification
 Prolog
 Constraints
 Quasigroup existence and
nonexistence
 Equivalential calculus axiom systems
 Euclidean and non-Euclidean
geometry
Achievements of Provers
 Verification of communication
networks
 Basketball scheduling
 Planning
 RRTP and description logic
 Propositional Calculus
 Formulae are composed of
Boolean variables p,q,r, … and
Boolean connectives:
  (conjunction, “and”)
  (disjunction, “or”)
  (negation, “not”)
  (implication, “if then”)
  (equivalence, “if and only if”)
 Example formula
 pqp
 Interpretation:
 “It is raining” and “It is Tuesday”
implies “It is raining.
 Another interpretation:
 “All birds are green” and “All fish are
purple” implies “All birds are green.”
 Both interpretations make the
formula true.
 The formula is valid (true in all
interps.)
 Another example formula:
 pqp
 Interpretation:
 2=2  3=3  2  2
 Another interpretation:
 2=2  3  3  2  2
 The first interpretation makes
the formula false.
 The second makes it true.
 The formula is not valid.
 Truth Tables
Truth Table for Conjunctions
1st Conjunct 2nd Conjunct Statement
A B A B
true true true
true false false
false true false
false false false

Truth Table for Disjunctions

1st Disjunct 2nd Disjunct Statement
A B A B
true true true
true false true
false true true
false false false
 Truth Table for Conditionals
Antacedent Consequent Statement
A B A B
true true true
true false false
false true true
false false true

Truth Table for Equivalences

Antacedent Consequent Statement
A B A B
true true true
true false false
false true false
false false true

Truth Table for Negations

Negation
Statement
A A
true false
false true
 Interpretations assign meanings to
symbols.
 In Boolean logic interpretations
assign truth values (true, false) to the
symbols.
 An interpretation in Boolean logic is
called a valuation.
 Thus a valuation I is an assignment of
truth values (true or false) to each
variable in a formula
 A valid formula
P Q P Q P  Q P
T T T T
T F F T
F T F T
F F F T

A satisfiable invalid formula

P Q P Q (P  Q)  Q
T T T T
T F F F
F T F T
F F F F
 An unsatisfiable formula: P  P

P P P P

T F F

F T F
 Testing Validity
 Using truth tables is exponential
 Resolution
 Davis and Putnam’s Method
 Local Search Methods
 Conjunctive Normal
Form
 Any propositional formula can be
put into conjunctive normal form
(clause form).
 Example:
 (p  q  r)  (p  r)  (q  r)

 Represent as sets:
 {p, q, r}, {p, r}, {q, r}



clause clauseclause
 Conjunctive Normal
Form
 A formula in conjunctive normal
form is unsatisfiable if for every
interpretation I, there is a clause
C that is false in I.
 A formula in cnf is satisfiable if
there is an interpretation I that
makes all clauses true.
 Binary Resolution Step
 For any two clauses C1 and C2, if there is
a literal L1 in C1 that is complementary
to a literal L2 in C2, then delete L1 and L2
from C1 and C2 respectively, and
construct the disjunction of the
remaining clauses. The constructed
clause is a resolvent of C1 and C2.
 Examples of Resolution Step
 C1=a b, C2=b c
 Complementary literals : b,b
 Resolvent: ac

 C1=a bc, C2=b d

 Complementary literals : b, b

 Resolution in Propositional
Logic
1. a b c a b c
2. b b
3. c  d  e c  d  e
4. e  f ef
5. d   f d
f
 Resolution in Propositional Logic
(continued)
 First, the goal to be
a
proved, a , is negated a  b c

and added to the b c b

clause set.
 The derivation of c

c d

indicates that the e f d e

database of clauses
d f  d
is inconsistent.
f f

 Horn clauses
 At most one positive literal
 Basis of Prolog
 Satisfiability can be tested in linear
time
 Resolution is fast for Horn clauses
 Resolution is very slow for non Horn
clauses
 Horn clauses: p  q  r, p  q   r, r
 Non Horn clause: p  q  r
 Hard problems are usually non-Horn
 DPLL (Davis and Putnam’s
Method)
(Purity rule omitted)
1. If no clauses in KB, return T (Satisfiable)
2. If a clause in KB is empty (FALSE), return F
(Unsatisfiable)
3. If KB has a unit clause C with prop. p, then
return DPLL(KB,p←polarity(p,C))
4. Choose an uninstantiated variable p
5. If DPLL(KB, p←TRUE) returns T, return T
6. If DPLL(KB, p←FALSE) returns T, return T
7. Return F
 DPLL Example
{p,r},{p,q,r},
{p,r}
p=T p=
F
{T,r},{T,q,r}, {F,r},{F,q,r},
{T,r} {F,r}
SIMPLIF SIMPLI
Y FY
{q,r} {r},
{r} SIMPLI
{} FY
 DPLL Viewed Abstractly
 The call DPLL(KB, p←TRUE) is
testing interpretations where p is
TRUE
 The call DPLL(KB, p←FALSE) is
testing interpretations where p is
FALSE
 In this way, interpretations are
examined in a sequential manner
 For each interpretation, a reason is
found that the formula is false in it
 Such a sequential search of
interpretations is very fast
 DPLL (Davis and Putnam’s
method), contiued
 DPLL does a backtracking search
for a model of the formula
 DPLL is much faster than
propositional resolution for non-
Horn clauses
 Very fast data structures developed
 Popular for hardware verification
 Local search can be much faster but
is incomplete
 “Systematic methods can now
routinely solve verification
problems with thousands or tens
of thousands of variables, while
local search methods can solve
hard random 3SAT problems with
millions of variables.”
 (from a conference
announcement)
 NP Complete but Easy
 How can the satisfiability problem
be so easy when it is NP complete?
 If there are many clauses the proof
is likely to be short and can be
found quickly
 If there are few clauses there are
likely to be many interpretations
and one is likely to be found quickly
 The hard problems are in the middle
at the “satisfiability threshold”
 First Order Logic
 Formulae may contain Boolean
connectives and also variables x,
y, z, …, predicates P,Q,R, …,
function symbols f,g,h, …, and
quantifiers  and  meaning “for
all” and “there exists.”
 Example: x(P(x)  yQ(f(x),y))
 Individual Constants
 Formulae can also contain
constant symbols like a,b,c which
can be regarded as functions of
no arguments.
 Example: x(P(x)  Q(x,c))
 Consider the formula yxP(x,y) 
xyP(x,y). Let the domain be the
set of people, and let P(x,y) be “x
loves y”.
 The formula then is interpreted as
“if there exists y such that for all x, x
loves y, then for all x, there exists y
such that x loves y.” In other words,
if there is someone that everyone
loves, then everyone loves someone.
 The formula is true under this
interpretation.
 In fact this formula is true under all
interpretations, and is a valid formula.
 Consider this formula: xyP(x,y) 
yxP(x,y). Under the same
interpretation, this formula becomes “If
for all x, there exists y such that x loves
y, then there exists y such that for all x,
x loves y.”
 In other words, if everyone loves
someone, then there is someone that
everyone loves.
 This formula is false under this
interpretation and is not a valid formula.
 Clauses
 An atom is a predicate symbol followed
by arguments, as, P(a, f(x)).
 A literal is an atom or its negation, as,
P(a,f(x)).
 A clause is a disjunction of literals,
often written as a set.
 Example: {p(x), p(f(x))} for p(x) 
p(f(x))
 A conjunction of clauses is also written
as a set, as, {C1, C2, C3} signifying C1
C2  C3.
 Substitutions
 A substitution  is an assignment
of terms to variables.
 If C is a clause then C  is C with
the substitution applied uniformly.
 Thus {P(x)}{x  f(a)} is {P(f(a))}.
 C  is called an instance of C. If C
 has no variables, it is called a
ground instance of C.
 Semantics
 Gelernter 1959 Geometry Theorem
Prover
 Adapt semantics to clause form:
 An interpretation (semantics) I is an
assignment of truth values to
literals so that I assigns opposite
truth values to L and L for atoms L.
 The literals L and L are said to be
complementary.
 Semantics
 We write I C (I satisfies C) to

╨
indicate that semantics I makes the
clause C true.
 If C is a ground clause then I satisfies
C if I satisfies at least one of its
literals.
 Otherwise I satisfies C if I satisfies all
ground instances D of C. (Herbrand
interpretations.)
 If I does not satisfy C then we say I
falsifies C.
 Example Semantics
 Specify I by interpreting symbols
 Interpret predicate p(x,y) as x = y
 Interpret function f(x,y) as x + y
 Interpret a as 1, b as 2, c as 3
 Then p(f(a,b),c) interprets to TRUE
but p(a,b) interprets to FALSE
 Thus I satisfies p(f(a,b),c) but I
falsifies p(a,b)
 Obtaining Semantics
 Humans using mathematical
knowledge
 Automatic methods (finite
models)
 Trivial semantics
 Herbrand’s Theorem
 A set S of clauses is unsatisfiable if
there is a finite unsatisfiable set T
of ground instances of S.
 The basis of uniform proof
procedures.
 Example: S = {{p(a)},{p(x),
p(f(x))}, {p(f(f(a)))}}
 T = {{p(a)},{p(a), p(f(a))},
{p(f(a)), p(f(f(a)))}, {p(f(f(a)))}}
{p(a)} {p(x), p(f(x))}
{p(f(f(a)))}

{p(a)}
{p(a), p(f(a))}
{p(f(a)), p(f(f(a)))}

{p(f(f(a)))}
 Criteria to evaluate
provers
 Don’t know versus don’t care
nondeterminism
 Clauses generated by need or
possibility
 Instantiation by unification or by
semantics or neither
 Clauses selected by semantics
 Goal sensitivity
 Space versus time
 Resolution Principle
 Steps for resolution refutation proofs
 Put the premises or axioms into clause
form.
 Add the negation of what is to be proved, in

clause form, to the set of axioms.

 Resolve these clauses together, producing

new clauses that logically follow from them.

 Produce a contradiction by generating the

empty clause.
 This is possible if and only if the theorem is

valid. (Completeness)
 Prove that “Fido will die.” from the
statements “Fido is a dog.”,
“All dogs are animals.”
and “All animals will die.”
 Changing premises to predicates
 (x) (dog(X)  animal(X))
 dog(fido)

 Modus Ponens and {fido/X}

 animal(fido)
 (Y) (animal(Y)  die(Y))

 Modus Ponens and {fido/Y}

 die(fido)
 Equivalent Reasoning by Resolution
 Convert predicates to clause form

Predicate form Clause form

1.(x) (dog(X)  animal(X)) dog(X) 
animal(X)
2.dog(fido) dog(fido)
3.(Y) (animal(Y)  die(Y)) animal(Y)  die(Y)

 Negate the conclusion

4.die(fido) die(fido)
 Equivalent Reasoning by
Resolution(continued)

dog(X)  animal(X) animal(Y)  die(Y)

{Y/X}
dog(fido) dog(Y)  die(Y)
{fido/Y}

die(fido) die(fido)

Resolution proof for the “dead dog” problem

 Skolemization
 Skolem constant
 (X)(dog(X)) may be replaced by dog(fido)
where the name fido is picked from the
domain of definition of X to represent that
individual X.
 Skolem function
 If the predicate has more than one
argument and the existentially quantified
variable is within the scope of universally
quantified variables, the existential variable
must be a function of those other variables.
 (X)(Y)(mother(X,Y))

(X)mother(X,m(X))
 (X)(Y)(Z)(W)(foo (X,Y,Z,W))

(X)(Y)(W)(foo(X,Y,f(X,Y),W))
 Resolution on the predicate calculus
 A literal and its negation in parent
clauses produce a resolvent only if they
unify under some substitution .  is
then applied to the resolvent before
adding it to the clause set.
 C = dog(X) animal(X)
1

C2 = animal(Y) die(Y)
Resolvent : dog(Y) die(Y) {Y/X}
C1 = p(X)  q(f(X)) C2 = q(Y)  r(g(Y))
Resolvent: p(X)  r(g(f(X)))
 “Lucky student”
1. Anyone passing his history exams
and winning the lottery is happy
 X(pass(X,history)  win(X,lottery) 
happy(X))
2. Anyone who studies or is lucky can
pass all his exams.
 XY(study(X)  lucky(X)  pass(X,Y))
3. John did not study but he is lucky
 study(john)  lucky(john)
4. Anyone who is lucky wins the
lottery.
 X(lucky(X)  win(X,lottery))
 Clause forms of “Lucky student”
1. pass(X,history)  win(X,lottery)
happy(X)
2. study(X)  pass(Y,Z)
lucky(W) pass(W,V)
3. study(john)
lucky(john)
4. lucky(V)  win(V,lottery)
5. Negate the conclusion “John is
happy”
happy(john)
 Resolution refutation for the
“Lucky Student” problem

pass(X, history) win(X,lottery) happy(X) win(U,lottery) lucky(U)

{U/X}
pass(U, history) happy(U) lucky(U) happy(john)

{john/U}
lucky(john) pass(john,history) lucky(join)

{}
pass(john,history) lucky(V) pass(V,W)

{john/V,history/W}
lucky(john) lucky(john)

{}
 Evaluating resolution
 Clauses generated by possibility
(bad)
 Don’t care nondeterminism (good)
 Unification based (good?)
 No semantics (bad)
 Uses a large amount of space (bad)
 Often not goal sensitive (bad)
 Refinements
 Many refinements of resolution
have been developed in an attempt
to improve its performance
 Set of support
 Hyper resolution
 Ancestry filter form
 Unit preference
 …
 Otter
 PROBLEM SEC CLAUSES KEPT
 LCL064-1.in 0.14 1080844 8604
 LCL064-2.in 0.00 9448 1954
 LCL065-1.in 0.00 2992 653
 LCL066-1.in 0.00 1452 306
 LCL067-1.in 0.14 492984 9283
 LCL068-1.in 0.29 569577 9593
 LCL069-1.in 0.00 3577 288
 LCL070-1.in 0.14 427166 8840
 LCL071-1.in 0.29 449389 8941
 LCL072-1.in 0.00 161139 6280
 Hyper Linking
 Separates instantiation and
inference
 Given S, selects clauses C and D in S
and literals L in C and M in D, and
generates instances C’ and D’ so that
L’ and M’ are complementary. Then
C’ and D’ are added to S.
 Periodically S is tested for
unsatisfiability using DPLL.
 Hyper Linking
Problem Input OTTER Hyper
Clauses (sec) Linking
Ph5 45 38606.76 1.8
Ph9 297 >24 hrs 2266.6
Latinsq 16 >24 hrs 56.4
Salt 44 1523.82 28.0
Zebra 128 >24 hrs 866.2
 Eliminating Duplication with the
Hyper-Linking Strategy, Shie-Jue
Lee and David A. Plaisted,
Journal of Automated Reasoning
9 (1992) 25-42.
 Later propositional
strategies
 Billon’s disconnection calculus,
derived from hyper-linking
 Disconnection calculus theorem
prover (DCTP), derived from
Billon’s work
 FDPLL
 Performance of DCTP on
TPTP, 2003
 First in EPS and EPR (largely
propositional)
 Third in FNE (first-order, no
equality) solving same number as
best provers
 Fourth in FOF and FEQ (all first-
order formulae, and formulae with
equality)
 Not tuned to 50 categories!
 Definition Detection

Problem OSHL Otter Otter

Time Time Clauses
P1 0.3 0.03 51
P2 2.3 1000+ 41867
P3 11.25 1000+ 27656
P4 1.35 1000+ 105244
P5 2.0 1000+ 54660
 Replacement Rules with
Definition Detection, David A.
Plaisted and Yunshan Zhu, in
Caferra and Salzer, eds.,
Automated Deduction in
Classical and Non-Classical
Logics, LNAI 1761 (1998) 80-94.
 Structure of OSHL
 Goal sensitivity if semantics chosen
properly
 Choose initial semantics to satisfy axioms
 Use of natural semantics
 For group theory problems, can specify a
group
 Sequential search through possible
interpretations
 Thus similar to Davis and Putnam’s method
 Propositional Efficiency

 Constructs a semantic tree

Ordered Semantic Hyperlinking (Oshl)

 Reduce first-order logic problem to

propositional problem
 Imports propositional efficiency into
first-order logic
 The algorithm
 Imposes an ordering on clauses
 Progresses by generating instances and
refining interpretations

I0 I1 I2 I3 …

D0 D1 D2 T
unsatisfiable
 OSHL
 I0 is specified by the user
 Di is chosen so that Ii falsifies Di
 Di is an instance of a clause in S
 Ii is chosen so that Ii satisfies Dj for
all j < i
 Let Ti be {D0,D1, …, Di-1}.
 Ii falsifies Di but satisfies Ti
 When Ti is unsatisfiable OSHL stops

and reports that S is unsatisfiable.

 Rules of OSHL
(C1,C2, …, Cn), D minimal contradict I
(C1,C2, …, Cn,D)

(C1,C2, …, Cn), Cn not needed

(C1,C2, …, Cn-1,D)

(C1,C2, …, Cn,D), max resolution possible

(C1,C2, …, Cn-1,res(Cn,D,L))
 Example
()
({-p1,-p2,-p3})
({-p1,-p2,-p3},{-p4,-p5,-p6})
({…},{…},{-p7})
({…},{…},{-p7},{p3,p7})
({…},{-p4,-p5,-p6},{p3})
({-p1,-p2,-p3},{p3})
({-p1,-p2})
 Number of Clauses
Generated
 Problem #clauses, Otter Oshl+semantics
 GRP005-1 57 3
 GRP006-1 62 7
 GRO007-1 85 22
 GRP018-1 266 16
 GRP019-1 267 15
 GRP020-1 265 18
 GRP021-1 264 19
 GRP023-1 79 22
 GRP032-3 83 14
 GRP034-3 141 30
 GRP034-4 222 6
 GRP042-2 21 15
 GRP043-2 80 81
 GRP136-1 0 8
 GRP137-1 0 8
 Engineering Issue
 OSHL generates about 10 clauses
per second
 Otter generates more than a
million clauses per second
 A factor of 100,000 in
engineering!
 Need to look at search space
sizes rather than times
 Evaluating OSHL
 Clauses generated by need (good)
 Don’t care nondeterminism (good)
 Instantiates using semantics (good)
 Goal sensitive (good)
 Space efficient (good)
 No unification (bad?)
 Need for more engineering
 TPTP library by Geoff Sutcliffe &
Christian Suttner
 Thousands of problems for theorem
provers
 Used to benchmark first order theorem
provers
 Contains 6973 theorems at present

 CASC competition by Sutcliffe et al.

 Every year: who has the fastest/most
accurate first order theorem prover on
the planet?
 Uses blind test from the TPTP library
 Current chamption: Vampire
 By Voronkov and Riazonov in Manchester
 CADE System
Competition
 The issue of 50 categories
 The 300 seconds issue
 Summary
 Efficiency of DPLL
 First-Order Theorem Proving
 Resolution
 Propositional Approaches
 Clause Linking
 DCTP and the CADE Competition
 Semantics
 OSHL