89
Fall 2006 Costas Busch - RPI 1 Reductions

Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Embed Size (px)

Citation preview

Page 1: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 1

Reductions

Page 2: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 2

Problem is reduced to problemX Y

If we can solve problem then we can solve problemX

Y

Page 3: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 3

Language is reduced tolanguage

There is a computable function (reduction) such that:f

BwfAw )(

A

B

Definition: A B

w )(wf

Page 4: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 4

Computable function : f

which for any string computes )(wfwThere is a deterministic Turing machineM

Recall:

Page 5: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 5

If: a: Language is reduced to b: Language is decidableThen: is decidable

Theorem:

Proof:

A BB

A

Basic idea:Build the decider for using the decider for

A

B

Page 6: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 6

Decider for B

Decider for A

compute

)(wf

)(wfw

accept

reject

accept

reject

(halt)

(halt)(halt)

(halt)

Inputstring

BwfAw )(

END OF PROOF

Reduction YES YES

NO NO

Page 7: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 7

Example:

}languages same the accept that

DFAs are and :,{ 2121 MMMMEQUALDFA

} language empty the

accepts that DFA a is :{

MMEMPTYDFA

is reduced to:

Page 8: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 8

Turing Machinefor reduction

DFADFA EMPTYMEQUALMM 21,

f21,MM M

MMf

21,

DFA

We only need to construct:

Page 9: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 9

21,MM M

MMf

21,

Let be the language of DFA Let be the language of DFA

1L

2L1M

2M

)()()( 2121 LLLLML

construct DFA by combining and so that:

M

DFA

1M 2M

Turing Machinefor reduction f

Page 10: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 10

)(21 MLLL

)()()( 2121 LLLLML

DFADFA EMPTYMEQUALMM 21,

Page 11: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 11

Decider

Decider for

compute M

Inputstring

DFAEQUAL

21,MM 21,MMf DFAEMPTY

YESYES

NONO

Reduction

Page 12: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 12

If: a: Language is reduced to b: Language is undecidableThen: is undecidable

Theorem (version 1):

A B

BA

(this is the negation of the previous theorem)

Proof:

Using the decider for build the decider forA

BSuppose is decidable B

Contradiction!

Page 13: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 13

Decider for B

Decider for A

compute

)(wf

)(wfw

accept

reject

accept

reject

(halt)

(halt)(halt)

(halt)

Inputstring

BwfAw )(

Reduction

END OF PROOF

If is decidable then we can build:B

CONTRADICTION!

YES YES

NO NO

Page 14: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 14

Observation:

In order to prove that some language is undecidablewe only need to reduce a known undecidable language to

B

BA

Page 15: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 15

State-entry problem

Input: M•Turing Machine•State q

Question: Does M

•Stringw

enter state qwhile processing input string ?w

Corresponding language:

} string input on state enters

that machine Turing a is :,,{

wq

MqwMSTATE TM

Page 16: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 16

Theorem:

(state-entry problem is unsolvable)

Proof: Reduce (halting problem) to (state-entry problem)

TMSTATE is undecidable

TMHALT

TMSTATE

Page 17: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 17

Decider for

YES

NO

wM,

state-entry problem decider

TMSTATEDeciderCompute

Reduction

wMf ,

wqM ,,ˆ

TMHALT

YES

NO

Given the reduction,if is decidable,then is decidable

TMSTATE

TMHALT

A contradiction!sinceis undecidable

Halting Problem Decider

TMHALT

Page 18: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 18

wM,Compute

Reduction

wMf ,wqM ,,ˆ

We only need to build the reduction:

TMHALTwM , TMSTATEqwM ,,ˆ

wMf ,

So that:

Page 19: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 19

Mqhalting

states

specialhalt state

Rxx ,

Construct from :M̂ M

A transition for every unused tape symbol of x

iq

iq

Page 20: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 20

M̂ halts on state qM halts

Mqhalting

states

specialhalt state

iq

Page 21: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 21

M̂ halts on state on inputq

M halts on input w

w

Therefore:

Equivalently:

END OF PROOF

TMHALTwM , TMSTATEqwM ,,ˆ

Page 22: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 22

Blank-tape halting problem

Input: MTuring Machine

Question: Does M halt when started with

a blank tape?

Corresponding language:

}tape blank on started when halts

that machin aTuring is :{ eMMBLANKTM

Page 23: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 23

Theorem:

(blank-tape halting problem is unsolvable)

Proof: Reduce (halting problem) to (blank-tape problem)

TMBLANK is undecidable

TMHALT

TMBLANK

Page 24: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 24

Decider for

YES

NO

wM,

blank-tape problem decider

DeciderCompute

Reduction

wMf ,

TMHALT

YES

NO

Given the reduction,If is decidable,then is decidableTMHALT

A contradiction!sinceis undecidable

Halting Problem Decider

TMHALT

TMBLANK

TMBLANK

Page 25: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 25

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMHALTwM , TMBLANKM ˆ

wMf ,

So that:

Page 26: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 26

no

yes

Write on tape w

Tape is blank?

Run

with input

Construct from :M̂ wM,

If halts then halt

M

w

M

Page 27: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 27

M̂ halts when started on blank tape

M halts on input

no

yesM

Write on tape w

Tape is blank?

Run

with inputw

w

Page 28: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 28

END OF PROOF

M̂ halts when started on blank tape

M halts on inputw

TMHALTwM , TMBLANKM ˆ

Equivalently:

Page 29: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 29

If: a: Language is reduced to b: Language is undecidableThen: is undecidable

Theorem (version 2):

A B

BA

Proof:

Using the decider for build the decider for A

B

Suppose is decidable B

Contradiction!

Then is decidableB

Page 30: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 30

Suppose is decidableB

Decider for B

s

accept

reject

(halt)

(halt)

Page 31: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 31

Suppose is decidableB

Decider for B

s

accept

reject

(halt)

(halt)

Then is decidableB(we have proven this in previous class)

reject

accept(halt)

(halt)

Decider for BNO YES

YES NO

Page 32: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 32

Decider for B

Decider for A

compute

)(wf

)(wfw

accept

reject

accept

reject

(halt)

(halt)(halt)

(halt)

Inputstring

BwfAw )(

Reduction

If is decidable then we can build:B

CONTRADICTION!

YES YES

NO NO

Page 33: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 33

Decider for B

Decider for A

compute

)(wf

)(wfw

accept

reject accept

reject

(halt)

(halt)(halt)

(halt)

Inputstring

BwfAw )(

Reduction

END OF PROOFCONTRADICTION!

Alternatively:

NO YES

YES NO

Page 34: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 34

Observation:

In order to prove that some language is undecidablewe only need to reduce some known undecidable languagetoor to B

(theorem version 1)

(theorem version 2)

B

AB

Page 35: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 35

Undecidable Problems for Turing Recognizable languages

• is empty?L

L• is regular?

L• has size 2?

Let be a Turing-acceptable language L

All these are undecidable problems

Page 36: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 36

• is empty?L

L• is regular?

L• has size 2?

Let be a Turing-acceptable language L

Page 37: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 37

Empty language problem

Input: MTuring Machine

Question: Is )(ML empty?

Corresponding language:

} language empty the accepts

that machine aTuring is :{

MMEMPTYTM

?)( ML

Page 38: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 38

Theorem:

(empty-language problem is unsolvable)

is undecidable

Proof: Reduce (membership problem) to

(empty language problem)

TMA

TMEMPTY

TMEMPTY

Page 39: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 39

Decider for

YES

NO

wM,

empty problem decider

DeciderCompute

Reduction

wMf ,

M̂YES

NO

Given the reduction,if is decidable,then is decidable

membership problem decider

TMA

TMATMEMPTY

TMEMPTY

A contradiction!sinceis undecidable

TMA

Page 40: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 40

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMATwM , TMEMPTYM ˆ

wMf ,

So that:

Page 41: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 41

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

sTape of M̂

input string

accepts ?

troys

Construct from :M̂ wM,

Page 42: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 42

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

s

accepts ?

troys

Tape of M̂

Page 43: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 43

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

s

accepts ?

w

troys

Tape of M̂

Page 44: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 44

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

s

accepts ?

During this phase this area is not touched

working area of

wM

troys

Page 45: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 45

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

s

accepts ?

waltered

Simply check if entered an accept stateM

troys

Page 46: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 46

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

s

accepts ?

Now check input string

troys

Page 47: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 47

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

saccepts ?

The only possible accepted string

troys

t r o y

Page 48: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 48

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

saccepts ?

troys

accepts }{)ˆ( troyMLM w

does notaccept

M w )ˆ(ML

Page 49: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 49

Therefore:

acceptsM w )ˆ(ML

Equivalently:

TMATwM , TMEMPTYM ˆ

END OF PROOF

Page 50: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 50

• is empty?L

L• is regular?

L• has size 2?

Let be a Turing-acceptable language L

Page 51: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 51

Regular language problem

Input: MTuring Machine

Question: Is )(ML a regular language?

Corresponding language:

language} regular a accepts

that machine aTuring is :{ MMREGULARTM

Page 52: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 52

Theorem:

(regular language problem is unsolvable)

is undecidable

Proof: Reduce (membership problem) to (regular language problem)

TMA

TMREGULAR

TMREGULAR

Page 53: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 53

Decider for

YES

NO

wM,

regular problem decider

DeciderCompute

Reduction

wMf ,

M̂YES

NO

Given the reduction,If is decidable,then is decidable

membership problem decider

TMA

TMA

TMREGULAR

A contradiction!sinceis undecidable

TMATMREGULAR

Page 54: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 54

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMATwM , TMREGULARM ˆ

wMf ,

So that:

Page 55: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 55

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If has the form

s

sTape of M̂

input string

accepts ?

nnba

Construct from :M̂ wM,

s

Page 56: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 56

skip inputstring s

write on tape

run

on input

w Mw

M w

yesaccepts ?

accepts }0:{)ˆ( nbaML nnM w

does notaccept

M w )ˆ(ML

then accept

If has the form

s

nnbas

not regular

regular

Page 57: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 57

Therefore:

acceptsM w )ˆ(ML

Equivalently:

TMATwM , TMREGULARM ˆ

END OF PROOF

is not regular

Page 58: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 58

• is empty?L

L• is regular?

L• has size 2?

Let be a Turing-acceptable language L

Page 59: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 59

Does have size 2?

Size2 language problem

Input: MTuring Machine

Question: )(ML

Corresponding language:

strings} two exactly accepts

that machine aTuring is :{2 MMSIZE TM

?2|)(| ML

(accepts exactly two strings?)

Page 60: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 60

Theorem:

(regular language problem is unsolvable)

is undecidable

Proof: Reduce (membership problem) to (size 2 language problem)

TMA

TMSIZE 2

TMSIZE 2

Page 61: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 61

Decider for

YES

NO

wM,

size2 problem decider

DeciderCompute

Reduction

wMf ,

M̂YES

NO

Given the reduction,If is decidable,then is decidable

membership problem decider

TMA

TMA

A contradiction!sinceis undecidable

TMA

TMSIZE 2

TMSIZE 2

Page 62: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 62

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMATwM , TMSIZEM 2ˆ

wMf ,

So that:

Page 63: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 63

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

sTape of M̂

input string

accepts ?

Construct from :M̂ wM,

},{ albanytroys

Page 64: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 64

skip inputstring s

write on tape

run

on input

w Mw

M w

yesaccepts ?

accepts },{)ˆ( albanytroyML M w

does notaccept

M w )ˆ(ML

2 strings

0 strings

then accept

If

s

},{ albanytroys

Page 65: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 65

Therefore:

acceptsM w )ˆ(ML

Equivalently:

TMATwM , TMSIZEM 2ˆ

END OF PROOF

has size 2

Page 66: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 66

RICE’s Theorem

• is empty?L

L• is regular?

L• has size 2?

Undecidable problems:

This can be generalized to all non-trivialproperties of Turing-acceptable languages

Page 67: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 67

Non-trivial property:

A property possessed by some Turing-acceptable languages but not all

: is empty?LExample:

L

}{troyL YES

NO

},{ albanytroyL NO

P

1P

Page 68: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 68

: is regular?

More examples of non-trivial properties:

L

}0:{ nbaL nn

YES

NO

L2P

}0:{ naL nYES

: has size 2?L3PL

}{troyL },{ albanytroyL

NO

YES

NO

Page 69: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 69

Trivial property:

A property possessed by ALL Turing-acceptable languages

P

: has size at least 0?LExamples: 4PTrue for all languages

: is accepted by some Turing machine?

L5P

True for allTuring-acceptable languages

Page 70: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 70

We can describe a property as the setof languages that possess the property

P

: is empty?LExample:

L

}{troyL YES

NO

},{ albanytroyL NO

P

}{P

If language has property then PL PL

Page 71: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 71

: has size 1?LP

}{a

},{ a

NO

NO

YES

Example: Suppose alphabet is }{a

}{aa},{ aa

}{aaa}{},{ aaa

}},{},{},{},{},{{ aaaaaaaaaaP

},,{ aaaNO },,{ aaaaaaaaa

Page 72: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 72

Non-trivial property problem

Does have the non-trivial property ?

Input: MTuring Machine

Question: )(ML

Corresponding language:

})( is, that , property

trivial-non the has )( that such

machine aTuring is :{

PMLP

ML

MMPROPERTYTM

?)( PML P

Page 73: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 73

Rice’s Theorem: TMPROPERTY is undecidable

(the non-trivial property problem is unsolvable)

Proof: Reduce (membership problem)

to

TMA

TMPROPERTY TMPROPERTYor

Page 74: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 74

We examine two cases:

Case 1:

Case 2:

P

P

Examples: : is empty?)(MLP

: is regular?)(MLP

: has size 2?)(MLPExample:

Page 75: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 75

Let be the Turing machine thataccepts

Case 1: P

Since is non-trivial, there is a Turing-acceptable languagesuch that:

PX

PX

XM

X

Page 76: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 76

Reduce (membership problem) to

TMA

TMPROPERTY

Page 77: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 77

Decider for

YES

NO

wM,

Non-trivial property problem decider

DeciderCompute

Reduction

wMf ,

M̂YES

NO

Given the reduction,if is decidable,then is decidable

membership problem decider

TMA

TMA

A contradiction!sinceis undecidable

TMA

TMPROPERTY

TMPROPERTY

Page 78: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 78

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMATwM , TMPROPERTYM ˆ

wMf ,

So that:

Page 79: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 79

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

sTape of M̂

input string

accepts ?

Construct from :M̂ wM,

Xs

Page 80: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 80

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

saccepts ?

Xs

For this phase we can run machinethat accepts , with input string

XMX s

Page 81: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 81

skip inputstring s

write on tape

run

on input

w Mw

M w

yesaccepts ?

accepts XML )ˆ(M w

does notaccept

M w )ˆ(ML

P

P

then accept

If

s

Xs

Page 82: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 82

Therefore:

acceptsM w PML )ˆ(

Equivalently:

TMATwM , TMPROPERTYM ˆ

Page 83: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 83

Let be the Turing machine thataccepts

Case 2: P

Since is non-trivial, there is a Turing-acceptable languagesuch that:

PX

PX

XM

X

Page 84: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 84

Reduce (membership problem) to

TMA

TMPROPERTY

Page 85: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 85

Decider for

YES

NO

wM,

Non-trivial property problem decider

DeciderCompute

Reduction

wMf ,

M̂YES

NO

Given the reduction,if is decidable,then is decidable

membership problem decider

TMA

TMA

A contradiction!sinceis undecidable

TMA

TMPROPERTY

TMPROPERTY

Page 86: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 86

wM,Compute

Reduction

wMf ,M̂

We only need to build the reduction:

TMATwM , TMPROPERTYM ˆ

wMf ,

So that:

Page 87: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 87

skip inputstring s

write on tape

run

on input

w Mw

M w

yes

then accept

If

s

sTape of M̂

input string

accepts ?

Construct from :M̂ wM,

Xs

Page 88: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 88

skip inputstring s

write on tape

run

on input

w Mw

M w

yesaccepts ?

accepts XML )ˆ(M w

does notaccept

M w )ˆ(ML

P

P

then accept

If

s

Xs

Page 89: Fall 2006Costas Busch - RPI1 Reductions. Fall 2006Costas Busch - RPI2 Problem is reduced to problem If we can solve problem then we can solve problem

Fall 2006 Costas Busch - RPI 89

Therefore:

acceptsM w PML )ˆ(

Equivalently:

TMATwM , TMPROPERTYM ˆ

END OF PROOF