COMPSCI 311: Introduction to Algorithms

Lecture 22: Intractability: SAT, NP

Ghazaleh Parvini

University of Massachusetts Amherst

Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time
Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

1. A polynomial number of standard computational steps
Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

1. A polynomial number of standard computational steps
2. A polynomial number of calls to a black box that solves
problem X
Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

1. A polynomial number of standard computational steps
2. A polynomial number of calls to a black box that solves
problem X

▶ Statement about relative hardness

Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

1. A polynomial number of standard computational steps
2. A polynomial number of calls to a black box that solves
problem X

▶ Statement about relative hardness

1. If Y ≤P X and X ∈ P , then Y ∈ P
Review: Polynomial-Time Reduction
▶ Y ≤P X: Problem Y is polynomial-time reducible to
Problem X,
solveY(yInput)
Construct xInput // poly-time
foo = solveX(xInput) // poly # of calls
return yes/no based on foo // poly-time

▶ . . . if any instance of Problem Y can be solved using

1. A polynomial number of standard computational steps
2. A polynomial number of calls to a black box that solves
problem X

▶ Statement about relative hardness

1. If Y ≤P X and X ∈ P , then Y ∈ P
2. If Y ≤P X and Y ∈
/ P then X ∈ /P
Reduction Strategies

▶ Reduction by equivalence
(Vertex-Cover ≤P Indept-Set and vice versa)

▶ Reduction to a more general case

(Vertex-Cover ≤P Set-Cover)

▶ Reduction by "gadgets"
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Clause C = x̄1 ∨ x2 ∨ x̄3 “or” of terms
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Clause C = x̄1 ∨ x2 ∨ x̄3 “or” of terms
Formula C1 ∧ C2 ∧ . . . ∧ Ck “and” of clauses
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Clause C = x̄1 ∨ x2 ∨ x̄3 “or” of terms
Formula C1 ∧ C2 ∧ . . . ∧ Ck “and” of clauses
Assignment (x1 , x2 , x3 ) = (1, 0, 1) assign 0/1 to each variable
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Clause C = x̄1 ∨ x2 ∨ x̄3 “or” of terms
Formula C1 ∧ C2 ∧ . . . ∧ Ck “and” of clauses
Assignment (x1 , x2 , x3 ) = (1, 0, 1) assign 0/1 to each variable
Satisfying assigment (x1 , x2 , x3 ) = (1, 1, 0) all clauses are “true”
Reduction by Gadgets: Satisfiability

▶ Can we determine if a Boolean formula has a satisfying

assignment?

(x1 ∨ x̄2 ) ∧(x̄1 ∨ x2 ∨ x̄3 ) ∧ (x2 ∨ x̄3 )

| {z }
"clause"

▶ Terminology

Variables x1 , . . . , x n
Term xi or x̄i variable or its negation
Clause C = x̄1 ∨ x2 ∨ x̄3 “or” of terms
Formula C1 ∧ C2 ∧ . . . ∧ Ck “and” of clauses
Assignment (x1 , x2 , x3 ) = (1, 0, 1) assign 0/1 to each variable
Satisfying assigment (x1 , x2 , x3 ) = (1, 1, 0) all clauses are “true”
Reduction by Gadgets: Satisfiability

SAT – Given boolean formula C1 ∧ C2 . . . ∧ Cm over variables

x1 , . . . , xn , does there exist a satisfying assignment?
Reduction by Gadgets: Satisfiability

SAT – Given boolean formula C1 ∧ C2 . . . ∧ Cm over variables

x1 , . . . , xn , does there exist a satisfying assignment?

3-SAT – Same, but each Ci has exactly three terms

Reduction by Gadgets: Satisfiability

SAT – Given boolean formula C1 ∧ C2 . . . ∧ Cm over variables

x1 , . . . , xn , does there exist a satisfying assignment?

3-SAT – Same, but each Ci has exactly three terms

2-SAT — each Ci has exactly two terms
Reduction by Gadgets: Satisfiability

SAT – Given boolean formula C1 ∧ C2 . . . ∧ Cm over variables

x1 , . . . , xn , does there exist a satisfying assignment?

3-SAT – Same, but each Ci has exactly three terms

2-SAT — each Ci has exactly two terms

Clicker. What is the strongest statement below that follows easily

from the definitions above?
A. 2-SAT ≤P 3-SAT ≤P SAT
B. 2-SAT ≤P SAT and 3-SAT ≤P SAT
C. SAT ≤P 3-SAT ≤P 2-SAT
Reduction by Gadgets: Satisfiability

Claim: 3-SAT ≤P IndependentSet.

Reduction by Gadgets: Satisfiability

Claim: 3-SAT ≤P IndependentSet.

Reduction:
▶ Given 3-SAT instance Φ = ⟨C1 , . . . , Cm ⟩, we will construct an
independent set instance ⟨G, m⟩ such that G has an
independent set of size m iff Φ is satisfiable
▶ Return Yes if solveIS(⟨G, m⟩) = Yes
Reduction

▶ Idea: construct graph G where independent set will select one

term per clause to be true

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3
Reduction

▶ Idea: construct graph G where independent set will select one

term per clause to be true

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

▶ One node per term

Reduction

▶ Idea: construct graph G where independent set will select one

term per clause to be true

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

▶ One node per term

▶ Edges between all terms in same clause (select at most one)
Reduction

▶ Idea: construct graph G where independent set will select one

term per clause to be true

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

▶ One node per term

▶ Edges between all terms in same clause (select at most one)
▶ Edges between a literal and all of its negations (consistent
truth assignment)
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if G has an independent set of size m, then ⟨C1 , . . . , Cm ⟩ is

satisfiable
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if G has an independent set of size m, then ⟨C1 , . . . , Cm ⟩ is

satisfiable
▶ Suppose S is an independent set of size m
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if G has an independent set of size m, then ⟨C1 , . . . , Cm ⟩ is

satisfiable
▶ Suppose S is an independent set of size m
▶ Assign variables so selected literals are true. Edges from terms
to negations ensure non-conflicting assignment.
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if G has an independent set of size m, then ⟨C1 , . . . , Cm ⟩ is

satisfiable
▶ Suppose S is an independent set of size m
▶ Assign variables so selected literals are true. Edges from terms
to negations ensure non-conflicting assignment.
▶ Set any remaining variables arbitrarily
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if G has an independent set of size m, then ⟨C1 , . . . , Cm ⟩ is

satisfiable
▶ Suppose S is an independent set of size m
▶ Assign variables so selected literals are true. Edges from terms
to negations ensure non-conflicting assignment.
▶ Set any remaining variables arbitrarily
▶ At most one term per clause is selected. Since m are selected,
every clause is satisfied.
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
▶ Consider any satsifying assignment of ⟨C1 , . . . , Cm ⟩
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
▶ Consider any satsifying assignment of ⟨C1 , . . . , Cm ⟩
▶ Let S consist of one node per triangle corresponding to true
literal in that clause. Then |S| = m.
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
▶ Consider any satsifying assignment of ⟨C1 , . . . , Cm ⟩
▶ Let S consist of one node per triangle corresponding to true
literal in that clause. Then |S| = m.
▶ For (u, v) within clause, at most one endpoint is selected
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
▶ Consider any satsifying assignment of ⟨C1 , . . . , Cm ⟩
▶ Let S consist of one node per triangle corresponding to true
literal in that clause. Then |S| = m.
▶ For (u, v) within clause, at most one endpoint is selected
▶ For edge (xi , x̄i ) between clauses, at most one endpoint is
selected, because xi = 1 or x̄i = 1, but not both
Correctness

(x̄1 ∨ x2 ∨ x3 ) ∧ (x1 ∨ x̄2 ∨ x3 ) ∧ (x̄1 ∨ x̄2 ∨ x̄3 )

x̅ 1 x1 x̅ 1

x2 x3 x̅ 2 x3 x̅ 2 x̅ 3

Claim: if ⟨C1 , . . . , Cm ⟩ is satisfiable, then G has an independent

set of size m
▶ Consider any satsifying assignment of ⟨C1 , . . . , Cm ⟩
▶ Let S consist of one node per triangle corresponding to true
literal in that clause. Then |S| = m.
▶ For (u, v) within clause, at most one endpoint is selected
▶ For edge (xi , x̄i ) between clauses, at most one endpoint is
selected, because xi = 1 or x̄i = 1, but not both
▶ Therefore S is an independent set
Reductions So Far

Partial map of problems we can use to solve others in polynomial

time, through transitivity of reductions:

3-SAT

Indept-Set SAT ▶ Y X
means Y ≤P X.
Vertex-Cover

Set-Cover
Toward a Definition of NP

Remember our problem hierarchy:

EXP

NP

What is special about the mystery problems (NP)?

P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”:
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
Problem classes
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
Problem classes
▶ P: Decision problems for which there is a polynomial time
algorithm.
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
Problem classes
▶ P: Decision problems for which there is a polynomial time
algorithm.

▶ NP: Decision problems for which there is a polynomial time

certifier.
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
Problem classes
▶ P: Decision problems for which there is a polynomial time
algorithm.

▶ NP: Decision problems for which there is a polynomial time

certifier.
▶ A solution can be “certified” in polynomial time.
P and NP

Intuition. For many “hard” decision problems, at least one thing is

“easy”: if the correct answer is Yes, there is an easy proof
▶ Independent set: show an independent set of size at least k
▶ SAT: show a satisfying assignment
Problem classes
▶ P: Decision problems for which there is a polynomial time
algorithm.

▶ NP: Decision problems for which there is a polynomial time

certifier.
▶ A solution can be “certified” in polynomial time.
▶ NP = “non-deterministic polynomial time”
Solver vs. Certifier
Let X be a decision problem and s be problem instance
(e.g., s = ⟨G, k⟩ for Independent Set)
Solver vs. Certifier
Let X be a decision problem and s be problem instance
(e.g., s = ⟨G, k⟩ for Independent Set)

Poly-time solver. Algorithm A(s) such that A(s) = Yes iff correct
answer is Yes, and running time polynomial time in |s|
yes/no

s
Instance
Solver vs. Certifier
Let X be a decision problem and s be problem instance
(e.g., s = ⟨G, k⟩ for Independent Set)

Poly-time solver. Algorithm A(s) such that A(s) = Yes iff correct
answer is Yes, and running time polynomial time in |s|
yes/no yes/no

A C

s s t
Instance Instance Hint

Poly-time certifier. Algorithm C(s, t) such that for every instance s,

there is some t such that C(s, t) = Yes iff correct answer is Yes,
and running time is polynomial in |s|.
Solver vs. Certifier
Let X be a decision problem and s be problem instance
(e.g., s = ⟨G, k⟩ for Independent Set)

Poly-time solver. Algorithm A(s) such that A(s) = Yes iff correct
answer is Yes, and running time polynomial time in |s|
yes/no yes/no

A C

s s t
Instance Instance Hint

Poly-time certifier. Algorithm C(s, t) such that for every instance s,

there is some t such that C(s, t) = Yes iff correct answer is Yes,
and running time is polynomial in |s|.
▶ t is the “certificate” or hint; size must also be polynomial in |s|
Certifier Example: Independent Set

Input s = ⟨G, k⟩.

Problem: Does G have an independent set of size at least k?
Idea: Certificate t = an independent set of size k
Certifier Example: Independent Set

Input s = ⟨G, k⟩.

Problem: Does G have an independent set of size at least k?
Idea: Certificate t = an independent set of size k

CertifyIS( ⟨G, k⟩, t)

if |t| < k return No
for each edge e = (u, v) ∈ E do
if u ∈ t and v ∈ t return No
Return Yes
Certifier Example: Independent Set

Input s = ⟨G, k⟩.

Problem: Does G have an independent set of size at least k?
Idea: Certificate t = an independent set of size k

CertifyIS( ⟨G, k⟩, t)

if |t| < k return No
for each edge e = (u, v) ∈ E do
if u ∈ t and v ∈ t return No
Return Yes

Polynomial time?
Certifier Example: Independent Set

Input s = ⟨G, k⟩.

Problem: Does G have an independent set of size at least k?
Idea: Certificate t = an independent set of size k

CertifyIS( ⟨G, k⟩, t)

if |t| < k return No
for each edge e = (u, v) ∈ E do
if u ∈ t and v ∈ t return No
Return Yes

Polynomial time? Yes, linear in |E|.

Example: Independent Set

▶ Independent Set ∈ P?
Example: Independent Set

▶ Independent Set ∈ P?
▶ Unknown. No known polynomial time algorithm.
Example: Independent Set

▶ Independent Set ∈ P?
▶ Unknown. No known polynomial time algorithm.

▶ Independent Set ∈ NP?

Example: Independent Set

▶ Independent Set ∈ P?
▶ Unknown. No known polynomial time algorithm.

▶ Independent Set ∈ NP?

▶ Yes. Easy to certify solution in polynomial time.
Example: 3-SAT

Input: formula Φ on n variables.

Problem: Is Φ satisfiable?
Idea: Certificate t = the satisfying assignment
Example: 3-SAT

Input: formula Φ on n variables.

Problem: Is Φ satisfiable?
Idea: Certificate t = the satisfying assignment

Certify3SAT( ⟨Φ⟩, t)
▷ Check if t makes Φ true
P, NP, EXP

EXP

NP

▶ 3SAT and Independent Set are in NP, as are many other

problems that are hard to solve, but easy to certify!
P, NP, EXP

EXP

NP

▶ 3SAT and Independent Set are in NP, as are many other

problems that are hard to solve, but easy to certify!
▶ Claim: P ⊆ NP
▶ Claim: NP ⊆ EXP
▶ Both straightforward to prove, but not critical right now.
NP-Complete

NP-
complete
NP

▶ NP-complete = a problem Y ∈ NP with the property that

X ≤P Y for every problem X ∈ NP!
NP-Complete

CIRCUIT-SAT

3-SAT SAT VC SC .... IS

▶ Cook-Levin Theorem: In 1971, Cook and Levin independently

showed that particular problems were NP-Complete.
▶ We’ll look at Circuit-SAT as canonical NP-Complete
problem.
Circuit-SAT

Problem: Given a circuit built of And, Or, and Not gates with
some hard-coded inputs, is there a way to set remaining inputs so
the output is 1?
Circuit-SAT

Problem: Given a circuit built of And, Or, and Not gates with
some hard-coded inputs, is there a way to set remaining inputs so
the output is 1?

Satisfiable? Yes. Set inputs: 1, 1, 0.

Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint
Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint
Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint

▶ s is Yes instance ⇔ ∃ t such that C(s, t) outputs Yes

Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint

▶ s is Yes instance ⇔ ∃ t such that C(s, t) outputs Yes

▶ Construct a circuit where s is hard-coded, and circuit is
satsifiable iff ∃ t that causes C(s, t) to output Yes
Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint

▶ s is Yes instance ⇔ ∃ t such that C(s, t) outputs Yes

▶ Construct a circuit where s is hard-coded, and circuit is
satsifiable iff ∃ t that causes C(s, t) to output Yes
▶ s is Yes instance ⇔ circuit is satisfiable
Circuit-SAT

Cook-Levin Theorem Circuit-SAT is NP-Complete.

Proof Idea: encode arbitrary certifier C(s, t) as a circuit

▶ If X ∈ NP, then X has a poly-time certifier C(s, t):
yes/no

s t
Instance Hint

▶ s is Yes instance ⇔ ∃ t such that C(s, t) outputs Yes

▶ Construct a circuit where s is hard-coded, and circuit is
satsifiable iff ∃ t that causes C(s, t) to output Yes
▶ s is Yes instance ⇔ circuit is satisfiable
▶ Algorithm for Circuit-Sat implies an algorithm for X
A Circuit-SAT reduction

See Independent Set example in other slides

A Circuit-SAT reduction
▶ Vertex Cover – Does G have VC of size at most k? (Counting
gadget is an example for v3 , v4 only)

v3 v4

v1 v2
A Circuit-SAT reduction
▶ Vertex Cover – Does G have VC of size at most k? (Counting
gadget is an example for v3 , v4 only)

v3 v4
2?
v1 v2
v4 ∧ ¬ ∧ 1?

v3 ∨ ¬ 0?

v2 ∨ ∧

v1 ∨ ∧ Cover?

∨ ∧

∨
Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Want to prove problem X is NP-complete

Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Want to prove problem X is NP-complete

▶ Check X ∈ NP.
Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Want to prove problem X is NP-complete

▶ Check X ∈ NP.
▶ Choose known NP-complete problem Y .
Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Want to prove problem X is NP-complete

▶ Check X ∈ NP.
▶ Choose known NP-complete problem Y .
▶ Prove Y ≤P X.
Proving New Problems NP-Complete

Fact: If Y is NP-complete and Y ≤P X, then X is NP-complete.

Want to prove problem X is NP-complete

▶ Check X ∈ NP.
▶ Choose known NP-complete problem Y .
▶ Prove Y ≤P X.
Clicker

It’s easy to show that 3-SAT ≤P Circuit-SAT. What can we

conclude from this?
A. 3-SAT is NP-complete.
B. 3-SAT is in NP.
C. If 3-SAT is NP-complete, then Circuit-SAT is also
NP-complete.
Proving New Problems NP-Complete

Theorem: 3-SAT is NP-Complete.

Proving New Problems NP-Complete

Theorem: 3-SAT is NP-Complete.

▶ In NP?
Proving New Problems NP-Complete

Theorem: 3-SAT is NP-Complete.

▶ In NP? Yes, check satisfying assignment in poly-time.
Proving New Problems NP-Complete

Theorem: 3-SAT is NP-Complete.

▶ In NP? Yes, check satisfying assignment in poly-time.
▶ Can show that Circuit-SAT ≤P 3-SAT (next time)
NP-Complete Problems: Preview

Circuit-SAT

Constraint satisfaction
3-SAT

Packing
Indept-Set Ham-Cycle Graph-Coloring
Subset-Sum
Ham-Path
Partitioning
Vertex-Cover Traveling-Salesperson 0-1 Knapsack

Sequencing Numerical
Set-Cover

Covering