NP-complete Languages

Fall 2005 Costas Busch - RPI 1

 Polynomial Time Reductions

Polynomial Computable function f :

There is a deterministic Turing machine M

such that for any string w computes f (w)
in polynomial time: O (| w |k
)

Fall 2005 Costas Busch - RPI 2

 Observation:

| f (w ) | O (| w | ) k

since, M cannot use

more than O (| w |k ) tape space
in time O (| w |k )

Fall 2005 Costas Busch - RPI 3

Definition:
Language A
is polynomial time reducible to
language B

if there is a polynomial computable

function f such that:

w  A  f ( w)  B

Fall 2005 Costas Busch - RPI 4

Theorem:
Suppose that A is polynomial reducible to B.
If B  P then A P .

Proof:
Let M be the machine that decides B
in polynomial time
Machine M  to decide A in polynomial time:
On input string w : 1. Compute f (w)
2. Run M on input f (w)
3. If f (w )  B acccept w
Fall 2005 Costas Busch - RPI 5
 Example of a polynomial-time reduction:

We will reduce the

The 3CNF-satisfiability problem

to

The clique problem

Fall 2005 Costas Busch - RPI 6

 3CNF formula: literal
( x1  x2  x3 )  ( x3  x5  x6 )  ( x3  x6  x4 )  ( x4  x5  x6 )

clause
Each clause has three literals

Language:
3CNF-SAT ={ w : w is a satisfiable
3CNF formula}

Fall 2005 Costas Busch - RPI 7

 A 5-clique in graph G

Language:
CLIQUE = { G, k  : graph G
contains a k-clique}
Fall 2005 Costas Busch - RPI 8
Theorem: 3CNF-SAT is polynomial time
reducible to CLIQUE

Proof: give a polynomial time reduction

of one problem to the other

Fall 2005 Costas Busch - RPI 9

 Transform formula to graph.
Example:
( x 1  x 2  x 4 )  (x 1  x 2  x 4 )  ( x 1  x 2  x 3 )  ( x 2  x 3  x 4 )
Clause 2
Create Nodes: x1 x2 x4 Clause 3

x1 x1

Clause 1 x2 x2

x4 Clause 4 x3

x2 x3 x4

Fall 2005 Costas Busch - RPI 10

 ( x 1  x 2  x 4 )  (x 1  x 2  x 4 )  ( x 1  x 2  x 3 )  ( x 2  x 3  x 4 )

x1 x2 x4

x1 x1

x2 x2

x4 x3

x2 x3 x4

Add link from a literal  to a literal in every

other clause, except the complement of literal 
Fall 2005 Costas Busch - RPI 11
 ( x 1  x 2  x 4 )  (x 1  x 2  x 4 )  ( x 1  x 2  x 3 )  ( x 2  x 3  x 4 )

x1 x2 x4

x1 x1

x2 x2

x4 x3

x2 x3 x4

Resulting Graph
Fall 2005 Costas Busch - RPI 12
( x 1  x 2  x 4 )  (x 1  x 2  x 4 )  ( x 1  x 2  x 3 )  ( x 2  x 3  x 4 )  1
x1  1
x2  0 x1 x2 x4
x3  0
x1 x1
x4  1

x2 x2

x4 x3

x2 x3 x4

The formula is satisfied if and only if

the Graph has a 4-clique End of Proof
Fall 2005 Costas Busch - RPI 13
 NP-complete Languages
We define the class of NP-complete
languages
Recursive (decidable)

NP

NP-complete

Fall 2005 Costas Busch - RPI 14

 A language L is NP-complete if:

• L is in NP, and

• Every language in NP
is reduced to L in polynomial time

Fall 2005 Costas Busch - RPI 15

Observation:

If a NP-complete language
is proven to be in P then:

P  NP

Fall 2005 Costas Busch - RPI 16

 Recursive (decidable)

NP

NP-complete

Fall 2005 Costas Busch - RPI 17

 Recursive (decidable)

NP

P
? NP-complete

Fall 2005 Costas Busch - RPI 18

 An NP-complete Language
Cook-Levin Theorem:
Language SAT (satisfiability problem)
is NP-complete

Proof:
Part1: SAT is in NP
(we have proven this in previous class)
Part2: reduce all NP languages
to the SAT problem
in polynomial time
Fall 2005 Costas Busch - RPI 19
Take an arbitrary language L  NP

We will give a polynomial reduction of L to SAT

Let M be the NonDeterministic

Turing Machine that decides L in polyn. time

For any string w we will construct

in polynomial time a Boolean expression  (M ,w )
such that: w  L   (M ,w ) is satisfiable

Fall 2005 Costas Busch - RPI 20

All computations q0

of M on string w qi qj

depth
| w | n
n k

…
…

… accept
…
reject accept

(deepest leaf)
reject
Fall 2005 Costas Busch - RPI 21
Consider q0

an accepting qi qj

computation
depth

n k

…
…

… accept
…
reject accept

(deepest leaf)
reject
Fall 2005 Costas Busch - RPI 22
Computation path Sequence of
q0 Configurations
qi
initial state
qj
1: q0  1 2  n
2:   1 qi  2  n
…


qm
nk  x :   1  l qa  l1  n k
qa
accept state
accept

w   1 2  n
Fall 2005 Costas Busch - RPI 23
 n
Machine M Tape
w
n k
n k

2n k

Maximum working space area on tape

during n time steps
k

Fall 2005 Costas Busch - RPI 24

 Tableau of Configurations
1: #    q0  1  2   n    #
2: #     1 qi  2   n    #
……

Accept configuration
x: #      1  2  3  qa  l1  nk #

indentical rows
nk : #      1  2  3  qa  l1  nk #
nk nk
2n  3k
Fall 2005 Costas Busch - RPI 25
 Tableau Alphabet

C  { # }  {tape alphabet}  {set of states}

 { # }  {1 , ,  r }  {q1 , , qt }

Finite size (constant)

Fall 2005 Costas Busch - RPI 26

For every cell position i, j
and
for every symbol in tableau alphabet s C

Define variable xi , j ,s

Such that if cell i , j contains symbol s

Then xi , j ,s  1
Else xi , j ,s  0
Fall 2005 Costas Busch - RPI 27
 Examples:
nk  3
1: #    q0  1  2   n    #
2: #     1 qi  2   n    #

x1,1,#  1 x 2,n k 3,q  1

i

x1,1,  0 x 2,n k 3,#  0

Fall 2005 Costas Busch - RPI 28

  (M , w ) is built from variables x i , j ,s

 (M ,w )  cell  start   accept  move

When the formula is satisfied,

it describes an accepting computation
in the tableau
of machine M on input w
Fall 2005 Costas Busch - RPI 29
 makes sure that every
 cell cell in the tableau contains
exactly one symbol

 
cell  
 
   xi , j ,s   xi , j ,s  xi , j ,t

all i , j  s C  s ,t C
 

  s t 

Every cell contains Every cell contains

at least one symbol at most one symbol
Fall 2005 Costas Busch - RPI 30
 Size of cell :

 
cell  
 
    xi , j ,s    xi , j ,s  xi , j ,t
all i , j  s C  s ,t C
 

  s t 

2
( 2n 3)k  ( |C |  | C |2 )

 O (n 2k
)

Fall 2005 Costas Busch - RPI 31

 start makes sure that the tableau
starts with the initial configuration

 start  x1,1,#  x1,2,  

 x1,n k 1,  x1,n k 2,q  x1,n k 3,  
0 1

 x1,2n k 2,  x1,2n k 2,#

Describes the initial configuration

in row 1 of tableau

Fall 2005 Costas Busch - RPI 32

 Size of  start :

 start  x1,1,#  x1,2,  

 x1,n k 1,  x1,n k 2,q  x1,n k 3,  
0 1

 x1,2n k 2,   x1,2n k 3,#

2n  3  O (n )
k k

Fall 2005 Costas Busch - RPI 33

 makes sure that the computation
accept leads to acceptance

 accept   xi , j ,q
all i , j
all q F
Accepting states

An accept state should appear somewhere

in the tableau
Fall 2005 Costas Busch - RPI 34
 Size of  accept :

 accept   xi , j ,q
all i , j
all q F

(2n  3)  O (n
k 2 2k
)

Fall 2005 Costas Busch - RPI 35

move makes sure that the tableau
give a valid sequence
of configurations

move is expressed in terms of

legal windows

Fall 2005 Costas Busch - RPI 36

 Tableau

Window
a q1 b
q2 a c

2x6 area of cells

Fall 2005 Costas Busch - RPI 37

 Possible Legal windows

a q1 b a q1 b
a  b, R q2 a c a a q2

b  c, L
q1 q2 a a q1 a b a
a a b a b q2
b  a,R

Legal windows obey the transitions

Fall 2005 Costas Busch - RPI 38

 Possible illegal windows

a b a a q1 b
a  b, R a a a q1 a a

q1 b  c, L q2 b q1 b
q2 b q2
b  a,R

Fall 2005 Costas Busch - RPI 39

 move   (window (i, j) is legal)
all i, j

window (i,j) is legal:

j j j
i a q1 b i a q1 b i a a q1
q2 a c a a q2 a a b 
((is legal)  (is legal)  (is legal))
all possible legal windows
in position (i , j )

Fall 2005 Costas Busch - RPI 40

 j
i a q1 b
q2 a c
(is legal)

Formula:
xi , j ,a  xi , j 1,q1  xi , j ,b
 xi 1, j ,q2  xi 1, j 1,a  xi 1, j 2,c

Fall 2005 Costas Busch - RPI 41

Size of move :
Size of formula for a legal window
in a cell i,j: 6

x Number of possible legal windows

in a cell i,j: at most | C |6

x Number of possible cells: (2n  3)

k 2

 6 | C |6 (2n k  3)2  O (n 2k )
Fall 2005 Costas Busch - RPI 42
Size of  (M ,w ) :

 (M ,w )  cell  start   accept  move

O (n 2k )  O (n k )  O (n 2k
)  O (n 2k
)

 O (n 2k
)
it can also be constructed in time O (n 2k )
polynomial in n
Fall 2005 Costas Busch - RPI 43
  (M ,w )  cell  start   accept  move

we have that:

w  L   (M ,w ) is satisfiable

Fall 2005 Costas Busch - RPI 44

 Since,
w  L   (M ,w ) is satisfiable
and
 (M ,w ) is constructed
in polynomial time

L is polynomial-time reducible to SAT

END OF PROOF
Fall 2005 Costas Busch - RPI 45
Observation 1:
The  (M ,w ) formula can be converted
to CNF (conjunctive normal form) formula
in polynomial time

 (M ,w )  cell  start   accept  move

Already CNF
NOT CNF
But can be converted to CNF
using distributive laws
Fall 2005 Costas Busch - RPI 46
 Distributive Laws:

P  (Q  R )  (P  Q )  (P  R )

P  (Q  R )  (P  Q )  (P  R )

Fall 2005 Costas Busch - RPI 47

Observation 2:
The  (M ,w ) formula can also
be converted to a 3CNF formula
in polynomial time

 a1  a2    al 
conver
t
 a1  a2  z1   (z1  a3  z 2 )  (z 2  a4  z 3 )    (zl 3  al 1  zl )

Fall 2005 Costas Busch - RPI 48

 From Observations 1 and 2:

CNF-SAT and
3CNF-SAT are
NP-complete languages

(they are known NP languages)

Fall 2005 Costas Busch - RPI 49

 Theorem:
If: a. Language A is NP-complete
b. Language B is in NP
c. A is polynomial time reducible to B
Then: B is NP-complete
Proof:
Any language L in NP
is polynomial time reducible to A.
Thus, L is polynomial time reducible to B
(sum of two polynomial reductions,
gives a polynomial reduction)
Fall 2005 Costas Busch - RPI 50
 Corollary: CLIQUE is NP-complete

Proof:
a. 3CNF-SAT is NP-complete
b. CLIQUE is in NP (shown in last class)
c. 3CNF-SAT is polynomial reducible to CLIQUE
(shown earlier)

Apply previous theorem with

A=3CNF-SAT and B=CLIQUE
Fall 2005 Costas Busch - RPI 51