Computing functions with Turing

machines
 Turing Machines with Outputs
• When we begin the computation the tape
contains the input.
• When the TM accepts (halts) return what is
written in the tape.
• TM doesn’t reject in any input.
 Number representation
• Decimal: 12
• Binary: 1100
• Unary: 111111111111

The unary is space consuming so generally we

prefer binary.
When we don’t care about resources it is more
convenient to use unary (easier to manipulate
with TMs).
 Total and Partial Functions
• A function f : N → N is total (or just function)
when f(n) is defined for every n
Example: f(n) = 2n
f

0 0

1 2

N 2 4
N
3 6

4 8
… …
 Total and Partial Functions
• A function f : N → N is called partial when f(n)
is defined for some n.
Example: f(n) = logn
f

1 0
0 2 1
3
N 6 5 4 2
N
7 3
9
8 3
… …
Computable and Partially Computable
Functions
• A (total) function f : N → N is (total)
computable if we can find a Turing Machine
that computes it (given any number n in Unary
as input in the tape, after completing the
computation the tape contains f(n) in Unary).
• A partial function f : N → N is said to be
partially computable if there is a Turing
Machine that partially computes it (if f is
defined for n then the machine should output
f(n), else it should loop for ever).
 Example: f(n) = 2n is computable
We design a TM that computes f(n).
High Level Program:
• The tape is divided into input and output (output is
right after the first blank after the input)
• Repeat:
– Erase one 1 from the input.
– Pass along the rest of the input
– Pass the blank that separates the input from the output.
– Pass along the output until you reach the end (blank).
– write two 1s.
– Go to the beginning of the input.
• Until the input is erased (accept).
 Example: f(n) = 2n is computable
The machine for f(n) = 2n

1→1,R 1 → 1, R
1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: ε
… …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: ε
… …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = 2n is computable
• Test input: 11
… 1 1 1 1 …

1→1,R 1 → 1, R

1→□,R □→□,R
q0 q1 q2

□→□,L □→1,L
qf q5 q4 q3

1 → 1, L 1 → 1, L
 Example: f(n) = logn is partially
computable.
We design a TM that partially computes f(n).
High Level Program:
• Add a $ in the beginning and the end of the input (this will
make the task of going from one state to another easier)
• Repeat:
– For every two ones in the input erase the first and leave the
second there.
– If there was an odd > 1 number of 1s then loop for ever (the
function is undefined)
– If the number of 1 is even then pass the $ sign
– Pass along the output until you reach the end (blank).
– Write one 1 in the output (after the $).
– Go to the beginning of the input.
• Until there is only one 1 in the input.
• Erase the two $ signs and accept.
 Example: f(n) = logn is partially
computable.

qf
$E

1→1,R 1→□,R

$R 1 ev 1→1,R od
1→1,R
□→$,L

1→□,R

□→$,R L
$L 0 ou
1→1,L $
1 → 1, L □→□,R □→□,L 1 → 1, L
 Example: f(n) = logn is partially
computable.
Try to run the machine by hand
Test inputs:
• ε (shouldn’t accept)
• 1 (should accept)
• 11 (should accept)
• 111 (shouldn’t accept)
• 111111 (shouldn’t accept)
• 11111111 (should accept)
 k-ary functions
• A function f might have more than one
parameters.
• f: N x N x … x N → N is called k-ary function
k times
Examples:
• + : N x N → N (addition of integers) is a binary
function.
 Representation of a k-tuple in a DTM
A k-tuple (x1 , x2 , …, xk) can be represented in a
Turing Machine by the following way:
x1 x2 xk

… 1 1 … 1 0 1 … 1 0 … 1 … 1 …
head

• Example: (2, 4)

… 1 1 0 1 1 1 1 …
 Binary addition is computable
• High level program:
– Remove the 0 from the middle and make the 1s in
the input consecutive.
• TM for this function:

1 → 1, R 1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 0 1 1 1 …

1 → 1, R 1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 0 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 0 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Binary addition is computable
• Test input: (2, 3) = 110111

… 1 1 1 1 1 …

1 → 1, R 1 → 1, R

0→1,R □→□,L 1→□,R

q0 q1 q2 qf
 Predicates
• Predicate: A Boolean (yes-no) function.
• P : N → {0,1}
• Examples
Unary predicate
1, n  0
P : N  {0,1}, P(n)  
 0, else
k-ary predicate
1, x  y
Q : N  N  {0,1}, Q( x, y)  
 0, else
• Partial Predicates: Predicates with unique
value that are not defined for some input.
 Computable Predicates
• A predicate P is computable if it can be
computed by a Turing Machine
– a) There is a Turing Machine that, given its
parameters n1, …, nk as input it outputs P(n1, …, nk)
– b) There is a Turing Machine that decides P (in
other words accepts if the output is 1 and rejects if
the output is 0.
• The 2 definitions are equivalent. We use the
second one.
 P(n) is computable
• High level program:
– If the tape is empty accept else reject
• Turing Machine:

□→□,R
q0 qf
 Q(x,y) is computable
• High level program:
- Remove one 1 from both x and y. If you can do
that for all the 1s in x then accept, else reject.
• TM for this predicate:
0→0,R
1 → 1, R x → x, R x → x, L

1→x,R 0→0,R 1→x,L □→□,R

q0 q1 q2 q3 qf

1 → 1, L

x → x, R q4
1 → 1, L
 Computable Sets, Characteristic
function
• The characteristic function χΑ of a set A is
defined as follows:
1, x  A
 A ( x)  
0, x  A

• A set is computable if its characteristic

function is computable.
 Computable Languages
• A language is computable if its characteristic
function is computable (we can use a Turing
Machine to decide membership)
• For example {anbn , n≥0} is a computable
language because there is a Turing Machine
that, given any string in Σ* it decides whether
the string belongs in the language or not.
 Partially Computable Languages
• Partially Computable Predicates: There is a
Turing Machine that for output 1 it accepts
(halts) and for ↑ it loops for ever.
• Partially Computable Languages: The
characteristic function is partially computable
(languages that are accepted by Turing
Machines).
 Counting Infinite sets
• We say that two infinite sets A, B are of the
same size if there is an one to one and onto
function from A to B (or from B to A)
• We say that an infinite set A is at most as large
as another infinite set B if we can find an one
to one function from A to B.
 Examples
|A| = |B| |A| ≤ |B|
a1 b1 a1 b1

A a2 b2 B A a2 b2 B
a3 b3 a3 b3
… … …
b4
…
 Countable Sets
• A set is countable if you can find an one to one and
onto correspondence with the natural numbers
(intuitively this means that it has the same number
of elements as the natural numbers)

a 1

b 2

A c 3
N
d 4

e 5
… …
 The set of the even numbers is
countable
• There is an one to one an onto function from
E (the even numbers) to N: f(n) = n/2

0 0

2 1

E 4 2 N
6 3

8 4
… …
 N x N is countable
0 1 8 9 24
(0,0) (0,1) (0,2) (0,3) (0,4) . . .

3 (1,0) (1,1)2 (1,2)7 (1,3)10 (1,4)23

4 (2,0) (2,1) (2,2)6 (2,3)11 (2,4)22

5
15(3,0) (3,1) (3,2) (3,3)12 (3,4)21
14 13
16(4,0) (4,1) (4,2) (4,3) (4,4)20
17 18 19
.
.
.
.
.
.
 The set of Turing Machines is
countable
• Every Turing Machine can be given a unique
number in binary as follows:
– Give to the states a number: q0 is 1, q1 is 11 etc…
– Give to the symbols of the stack a unique number:
a is 1, b is 11, c is 111 etc…
– Assign 1 to L, 11 to R and 111 to S
– For each δ(q, a) = (q’, b, H) give the binary number
11…1011...1011…1011…1011…1
q a q’ b H
where H is L, R or S.
 The set of Turing Machines is
countable
• Every Turing Machine can be given a unique
number in binary as follows:
– To obtain the number of the machine combine
each number for the arrows together (separated
with 00).
– The number starts with q0 00 qf 00
• The number associated with the Turing
Machine M is denoted as <M>
 Example
• The Turing Machine MP that computes the
Predicate
?
P(n) := n=0 has the following number

<MP> = 100110010101101011

□→□,R
q0 qf
 Uncountable Sets
• Countable sets are infinite.
• However there are some sets that are considered
“even larger”.
• There is no way to enumerate them.
• Diagonalization method: Suppose that there is an
enumeration of all the elements of the set.
• Obtain a new element by taking different parts of
each element and changing them.
• The new element is not in the enumeration.
Contradiction!!!
 The set of N →{0,1} Predicates is
not countable
Suppose that it was. An enumeration of all the
predicate would be the following:
1 2 3 4 5 6
1 0 1 0 1 1 1 f(1) = 1
2 1 0 1 1 0 0 . . . f(2) = 1
3 0 0 1 0 1 1 f(3) = 0
4 1 1 1 0 1 1 f(4) = 1
.
.
.

The predicate f(n) = 1 - fn(n) is not in the enumeration.

 Not all predicates are computable
1. Turing Machines are countable
2. Predicates are uncountable
Thus there is a predicate for which there is no
Turing Machine that decides it.
A predicate that is not computable
• The Halting Problem: Given a Turing Machine M and
an input x does M halt with input x?
• Important problem: If a machine doesn’t stop then
we don’t know for sure if it is going to accept the
input or not.
• It turns out that this problem is not solvable!
• In other words we can prove that the predicate Halt
is not computable (there is no Turing Machine that
takes as input the pair (<M>, x) and decides if M is
going to accept on x.
 Predicate Halt is not computable
 1, M ( M ) halts
• We will prove that the predicate H ( M )  
0, M ( M ) loops
is not computable.
• Suppose that there is a TM HTM that decides H.
In other words, HTM(<M>) halts if H(<M>) = 1 else loops.
• Take the machine H’TM which does the opposite: halts on
input M if H(<M>) = 0 else loops.
• Run H’TM with input the machine H’TM itself.
– H’TM (<H’TM>) is going to halt if H(<H’TM>) = 0 which means that
H’TM (<H’TM >) loops.
– H’TM (<H’TM>) is going to loop if H(<H’TM>) = 1 which means
that H’TM (<H’TM >) halts.
 Important Facts
• This method is called self-reference
• Thesis Turing-Church implies that any other
machine or program can do exactly whatever a
Turing Machine can do.
• There is no machine or program that can
decide whether or not a machine or program is
going to halt.
• Computers CANNOT decide the halting
problem.
Halting problem for Pascal programs
• Suppose that there is a Pascal program Halt
that takes as input another Pascal program p
and decides whether this program is going to
halt or not (outputs true if p halts, else false).
• Create a new Pascal program Halt’ by adding
to Halt the following code:
– while Halt(p)=true do a:=1;
• This makes Halt’(p) to loop if p halts and vice
versa.
Halting problem for Pascal programs
• Of course Halt’ is a proper Pascal program so
why not giving it as input to Halt’ (self-
reference).
• Now:
– If Halt’(Halt’) halts then Halt(Halt’) = true so
Halt’(Halt’) is going to loop .
– If Halt’(Halt’) loops then Halt(Halt’) = false so
Halt’(Halt’) is going to halt
Contradiction!!!
 Universal Turing Machine
• We can create a Turing machine U that takes a
pair of numbers (<M>, x) as input and then
simulates M running on input x:
– If M accepts on x then U accepts on (<M>, x)
– If M rejects on x then U rejects on (<M>, x)
– If M loops on x then U loops on (<M>, x)
• This machine is called a Universal Turing
Machine.
Predicate H is partially computable
 1, M ( M ) halts
• The partial predicate
H (  M )  
, M ( M ) loops
is partially computable.
• Run the Universal Turing machine U with input
(<M>,<M>).
 A predicate that is not partially
computable
, M ( M ) halts
• Consider the predicate H ( M )  
0, M ( M ) loops
• H is not partially computable
• Suppose that there was a TM U’ that could partially
compute H .
• Idea: Run both machines U and U’ on input (<M>,<M>).
At some point one of them will halt.
– If U halts then accept
– If U’ halts then reject
• But this decides the Halt predicate. Contradiction!
• Difficult part: Simulate in one machine the concurrent
running of U and U’.