You are on page 1of 54

1

Turing Machines Variants


Zeph Grunschlag
2
Announcement
Midterms not graded yet
Will get them back Tuesday
3
Agenda
Turing Machine Variants
Non-deterministic TMs
Multi-Tape
4
Input-Output Turing Machines
Input/output (or IO or transducing) Turing
Machines, differ from TM recognizers in that
they have a neutral halt state q
halt
instead of
the accept and reject halt states. The TM is
then viewed as a string-function which takes
initial tape contents u to whatever the non
blank portion of the tape is when reaching
q
halt
. If v is the tape content upon halting,
the notation f
M
(u) = v is used.
If M crashes during the computation, or enters
into an infinite loop, M is said to be
undefined on u.

5
Input-Output Turing Machines
When f
M
crashes or goes into an infinite
loop for some contents, f
M
is a partial
function
If M always halts properly for any
possible input, its function f is total
(i.e. always defined).

6
TM Notations
There are three ways that Sipser uses to
describe TM algorithms.
1) High level pseudocode which explains how
algorithm works without the technical
snafoos of TM notation
2) Implementation level describe how the TM
operates on its tape. No need to mention
states explicitly.
3) Low-level description. One of:
Set of complete goto style instructions
State diagram
Formal description: spell out the 7-tuple
7
High-Level TM
Example
Let's for example describe a Turing
Machine M which multiplies numbers
by 2 in unary:
M = "On input w = 1
n

For each character c in w
copy c onto the next available b
blank space"
8
Implementation-Level TM
Example
The idea is to carry out the high level
description by copying each character
to the end. We also need to keep track
of which characters have already been
copied (or were copies themselves) by
distinguishing these characters. One
way is to use a different character, say
X.
EG: Lets see how 11111 is transformed.
9
Implementation-Level TM
Example
So round by round, tape transformed as
follows:
11111
X1111X
XX111XX
XXX11XXX
XXXX1XXXX
XXXXXXXXXX
1111111111
10
Implementation-Level TM
Example
Implementation level describes what
algorithm actually looks like on the Turing
machine in a way that can be easily turned
into a state-diagram
Some useful subroutines:
fast forward
move to the right while the given condition holds
rewind
move to the left while the given condition holds
May need to add extra functionality
Add $ if need to tell when end of tape is
11
Implementation-Level TM
Example
M = "On input w = 1
n

1. HALT if no input
2. Write $ in left-most position
3. Sweep right and write X in first blank
4. Sweep left through X-streak and 1-streak
5. Go right
6. If read X, go right and goto 9.
Else, replace 1 by X, move right.
7. If read X [[finished original w]] goto 8
Else, goto 3
8. Sweep to the right until reach blank, replace by X
9. Sweep left replacing everything non-blank by 1
10. HALT
12
Implementation-Level TM
Example
At the low level the Turing Machine is
completely described, usually using a
state diagram:
0
halt
R
2
1$,R
$|XR
1|XR
3 4
1L
1X,R
1,L
X,L
XL 1L
5 6
8
1R
XR
XR
9
X1,L
$1,L
$R
9
X1,L
13
Non-Deterministic TMs
A non-Deterministic Turing Machine N allows
more than one possible action per given
state-tape symbol pair.
A string w is accepted by N if after being put
on the tape and letting N run, N eventually
enters q
acc
on some computation branch.
If, on the other hand, given any branch, N
eventually enters q
rej
or crashes or enters an
infinite loop on, w is not accepted.
Symbolically as before:
L(N) = { x S* | accepting config. y, q
0
x * y }
(No change needed as need not be function)
14
Non-Deterministic TMs
Recognizers vs. Deciders
N is always called a non-deterministic
recognizer and is said to recognize
L(N); furthermore, if in addition for all
inputs and all computation branches, N
always halts, then N is called a non-
deterministic decider and is said to
decide L(N).

15
Non-Deterministic TM
Example
Consider the non-deterministic method:

void nonDeterministicCrossOut(char c)
while()
if (read blank) go left
else
if (read c)
cross out, go right, return
OR go right // even when reading c
OR go left // even when reading c

16
Non-Deterministic TM
Example
Using randomCross() put together a non-
deterministic program:

1. while(some character not crossed out)
nonDeterministicCrossOut(0)
nonDeterministicCrossOut(1)
nonDeterministicCrossOut(2)
2. ACCEPT

Q: What language does this non-deterministic
program recognize ?
17
Non-Deterministic TM
Example
A: {x {0,1,2}* | x has the same no.
of 0s as 1s as 2s }

Q: Suppose q is the state of the TM while running
inside nonDeterministicCrossOut(1) and q is
the state of the TM inside
nonDeterministicCrossOut(2).
Suppose that current configuration is
u = 0XX1Xq12X2
For which v do we have u v ?
18
Non-Deterministic TM
Example
A: 0XX1Xq12X2
0XX1qX12X2 | 0XX1X1q2X2 | 0XX1XXq 2X2
These define 3 branches of computation
tree:



Q: Is this a non-deterministic TM decider?
0XX1XXq 2X2
0XX1Xq12X2
0XX1X1q2X2 0XX1qX12X2
19
Non-Deterministic TM
Example
A: No. This is a TM recognizer, but not a decider.
nonDeterministicCrossOut() often enters an
infinite branch of computation since can see-
saw from right to left to right, etc. ad infinitum
without ever crossing out anything. I.e.,
computation tree is infinite!
Note: If you draw out state-diagrams, you will see
that the NTM is more compact, than TM
version so there are some advantages to non-
determinism! Later, will encounter examples
of efficient nondeterministic programs for
practically important problems, with no known
efficient counterpart: The P vs. NP Problem.
20
NTMs
Konigs Infinity Lemma
For Problem 3.3 in Sipser the following fact is
important:
If a NTM is a decider then given any input,
there is a number h such that all computation
branches involve at most h basic steps. I.e.,
computation tree has height h. Follows from:
Konigs Infinity Lemma: An infinite tree with
finite branching at each node must contain an
infinitely long path from the root.
Or really, the contrapositive is used: A tree
with no infinite paths, and with finite
branching must itself be finite.
21
Konigs Infinity Lemma
Proof Idea
Idea is to smell-out where the infinite
part of the tree is and go in that
direction:


22
Konigs Infinity Lemma
Proof Idea
Idea is to smell-out where the infinite
part of the tree is and go in that
direction:



23
Konigs Infinity Lemma
Proof Idea
Idea is to smell-out where the infinite
part of the tree is and go in that
direction:




24
Konigs Infinity Lemma
Proof Idea
Idea is to smell-out where the infinite
part of the tree is and go in that
direction:





25
Konigs Infinity Lemma
Proof Idea
Idea is to smell-out where the infinite
part of the tree is and go in that
direction:






26
Konigs Infinity Lemma
Proof
Proof. Given an infinite tree with finite
branching construct an infinite path
inductively:
Vertex v
0
: Take the root.
Edge v
n
v
n+1
: Suppose v
0
v
1
v
n-1
v
n
has been constructed and that the
subtree from v
n
is infinite. Then one of
v
n
s finite no. of children, call it v
n+1
,
must have an infinite subtree, so add
the edge v
n
v
n+1.

27
Multi-tape TMs
Often its useful to have several tapes
when carrying out a computations. For
example, consider a two tape I/O TM
for adding numbers (we show only how
it acts on a typical input)
28
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
Input string
29
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
30
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
31
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
32
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
33
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$
34
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$ 1
35
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$ 1 0
36
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$ 1 0 1
37
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$ 1 0 1 1
38
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1 1
$ 1 0 1 1
39
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0 1
$ 1 0 1 1
40
Multi Tape TM
Addition Example
$ 1 0 1 $ 1 0
$ 1 0 1 1
41
Multi Tape TM
Addition Example
$ 1 0 1 $ 1
$ 1 0 1 1
42
Multi Tape TM
Addition Example
$ 1 0 1 $
$ 1 0 1 1
43
Multi Tape TM
Addition Example
$ 1 0 1
$ 1 0 1 1
44
Multi Tape TM
Addition Example
$ 1 0 0
$ 1 0 1 0
45
Multi Tape TM
Addition Example
$ 1 0 0
$ 1 0 0 0
46
Multi Tape TM
Addition Example
$ 0 0 0
$ 1 0 0 0
47
Multi Tape TM
Addition Example
0 0 0 0
$ 0 0 0 0
48
Multi Tape TM
Addition Example
1 0 0 0
1 0 0 0 0
49
Multi Tape TM
Addition Example
1 0 0 0
1 0 0 0 0
50
Multi Tape TM
Addition Example
1 0 0 0
1 0 0 0 0
51
Multi Tape TM
Addition Example
1 0 0 0
1 0 0 0 0
52
Multi Tape TM
Addition Example
1 0 0 0 0
1 0 0 0 0
Output
string
HALT!
53
Multitape TMs
Formal Notation
NOTE: Sipsers multitape machines cannot
pause on one of the tapes as above example.
This isnt a problem since pausing 1-tape
machines can simulate pausing k-tape
machines, and non-pausing 1-tape machines
can simulate 1-tape pausing machines by
adding dummy R-L moves for each pause.
Formally, the d-function of a k-tape machine:
k k k
} R L, { : Q Q
54
Multitape TMs
Conventions
Input always put on the first tape
If I/O machine, output also on first
tape
Can consider machines as string-
vector generators. E.g., a 4 tape
machine could be considered as
outputting in (S*)
4

You might also like