Theory of Computation

Lecture #6

Sarmad Abbasi

Virtual University

Sarmad Abbasi (Virtual University) Theory of Computation 1 / 39

Lecture 6: Overview

Prove the equivalence of enumerators and TMs.

Dovetailing
Definition of Algorithm
The Church-Turing Thesis.
Encodings.
Hilbert’s 10th problem.
Matijasevič’s Theorem.

Sarmad Abbasi (Virtual University) Theory of Computation 2 / 39

Enumerators

Enumerator’s are another model of computation. They do not take any

input and only produce output. Let us define them formally: An
enumerator E = (Q, Σ, Γ, q0 , δ).
1 Q is the set of states.
2 Σ is the output alphabet and #,  6∈ Σ.
3 Γ is the work alphabet and  ∈ Γ.
4 q0 is the start state.
5 δ is the transition function.
Well what is the domain and range of δ.

δ : Q × Γ 7→ Q × Γ{L, R} × (Σ ∪ {, #})

Sarmad Abbasi (Virtual University) Theory of Computation 3 / 39

Enumerators

How does the a configuration of E look like:

αqβ, w1 #w2 # · · · wk −1 #x
where α, β ∈ (Γ ∪ )∗ and x ∈ (Σ∗ ∪ Σ∗ #)

Sarmad Abbasi (Virtual University) Theory of Computation 4 / 39

Enumerators

Write the definition of the yields relation for enumerators.

We say that E outputs a string w if it reaches a configuration

αqβ, w1 #w2 · · · wk −1 #w#

Sarmad Abbasi (Virtual University) Theory of Computation 5 / 39

Enumerators

L(E) the language enumerated by E is the set of all strings that it

outputs.

Sarmad Abbasi (Virtual University) Theory of Computation 6 / 39

Enumerators

Last time we outlined a proof of the following theorem.

A language L is Turing-recognizable if and only if some enumerator

enumerates it.

Sarmad Abbasi (Virtual University) Theory of Computation 7 / 39

Enumerators

One side was easy:

Let L be a language enumerated by an enumerator E. The L is Turing

recognizable.

To prove this we are given that an enumerator E enumerates L. We

have to show that some Turing machine M accepts L. The informal
description of the Turing machine is as follows:
1 Input x
2 Run the Enumerator E each time it outputs a string w check if
w = x. If w = x go to accept state. Otherwise, continue.

Sarmad Abbasi (Virtual University) Theory of Computation 8 / 39

Enumerators
The other side was a bit tricky. But let us look at the details.

If L is Turing recognizable then there is an enumerator E that

enumerates L.
Since L is Turing recognizable therefore some TM M accepts L. We
can use M as a subroutine in our enumerator E. Here is our first
attempt.
Recall that we know how to generate strings in lexicographical order as

, 0, 1, 00, 01, 10, 11, . . . ,

1 For i = 1, . . . ,
2 Simulate the machine M on xi
3 If M accepts xi output xi .
Why is this wrong?

Sarmad Abbasi (Virtual University) Theory of Computation 9 / 39

Enumerators

Lets say
1 M accepts .
2 M rejects 0.
3 M accepts 1.
4 M loops forever on 00.
5 M accepts 01.
Then the enumerator will never output 01. Whereas, 01 ∈ L. So, we
have to get around this problem.

Sarmad Abbasi (Virtual University) Theory of Computation 10 / 39

Dovetailing

The solution is given by dovetailing:

1 For i = 1, 2, 3, . . . ,
2 For j = 1, 2, . . . , i
3 Simulate M on xj for i steps only.
4 If M accepts xj output xj .

Sarmad Abbasi (Virtual University) Theory of Computation 11 / 39

Dovetailing

Suppose xt ∈ L. Then M accepts xt and it must accept xt is some finite

number of steps. Lets say this number is k . Then we claim that when
i = max(k , t) the string xt will be outputted. That is because M is
simulated for on
x1 , . . . , xi
for i steps. Since, i ≥ t hence

xt ∈ {x1 , . . . , xi }

and since i ≥ k hence M will accept xt in k steps and xt will be

outputted.

Sarmad Abbasi (Virtual University) Theory of Computation 12 / 39

Dovetailing

Hence all strings that are in L are enumerated by this enumerators. To

see if x 6∈ L then E will not enumerate it, we simply observe that E only
enumerates strings that are accepted by M.

Sarmad Abbasi (Virtual University) Theory of Computation 13 / 39

Dovetailing

A minor point to notice is that E enumerates a given string many many

times. If this bothers you, you can modify E so that it enumerates each
string only once.
Now, E keeps a list of all the strings it has enumerated and checks the
list before enumerating every string if it has been enumerated before.

Show that if L is Turing recognizable then there is an enumerator E

that enumerates L. Furthermore, each string of L is enumerated by E
exactly once.

Sarmad Abbasi (Virtual University) Theory of Computation 14 / 39

Definition of an algorithm

Intuitive notion of algorithms

equals
Turing machine algorithms

Another way to put it:

An algorithm is a Turing machine that halts on all inputs.

An algorithm is a decider.

Sarmad Abbasi (Virtual University) Theory of Computation 15 / 39

Church-Turing Thesis

The Church-Turing thesis states that the intuitive notion of algorithms

is equivalent to the mathematical concept of a Turing machine.
It gives the formal definition for answering questions like Hilbert’s tenth
problem.

Sarmad Abbasi (Virtual University) Theory of Computation 16 / 39

Encodings

Note that we always take inputs to be strings over some alphabet Σ∗ .

Suppose Σ = {a, b, c, d} we can map
1 a to 00; that is e(a) = 00.
2 b to 01; that is e(a) = 01.
3 c to 10; that is e(a) = 10.
4 d to 01; that is e(a) = 11.
This way we can encode any alphabet to {0, 1}. Note that this maps
Σ∗ to {0, 1}∗ in a natural way.

e(w1 w2 . . . wn ) = e(w1 )e(w2 ) · · · e(wn ).

Sarmad Abbasi (Virtual University) Theory of Computation 17 / 39

Encodings

Suppose

L ⊂ Σ∗
we can define
Le = {e(w) : w ∈ L}.
Designing a TM for M or designing a TM for Le is equally difficult or
easy.

Sarmad Abbasi (Virtual University) Theory of Computation 18 / 39

Encodings

We want to solve problems where the input is an interesting object

such as a graph, or a matrix. However, our definitions only allow us
inputs which are strings. But, this is usually not a restriction. Let us
look at a few examples:

L = {(i, j) : i 2 = j}
(1, 1), (2, 4), (3, 9) are in L and (1, 3), (2, 5), (3, 1) are not in L.
L is a subset of N × N.

Sarmad Abbasi (Virtual University) Theory of Computation 19 / 39

Encodings

What we can do is to encode the pairs into strings. Let us say we

reserve 11 to represent a 1, 00 to represent a 0 and 01 will be the
separator. Then we can encode

(2, 4) to 110001110000

We represent the encoding of a pair (i, j) by < (i, j) >.

Then we can solve the following problem

L0 = {< (i, j) >: i 2 = j}

This is a language over {0, 1}∗ .

Sarmad Abbasi (Virtual University) Theory of Computation 20 / 39

Encodings

Let us look at another example:

D = {A : A is a n × n with integer entires and det(A) = 0}

Sarmad Abbasi (Virtual University) Theory of Computation 21 / 39

Encodings

How can we encode this problem so that it becomes a language over

{0, 1}∗ . Lets say we have the matrix
 
2 5
3 6

Sarmad Abbasi (Virtual University) Theory of Computation 22 / 39

Encodings

First we can write it as linearly as 2, 5, 3, 6. Now, we can write all the

numbers in binary where 00 represents a 0, 11 represents a 1 and 01
represents the separator ,. So
 
2 5
= 11000111001101111101111100
3 6

Sarmad Abbasi (Virtual University) Theory of Computation 23 / 39

Encodings

Now, we can define the following language

D 0 = {hAi : A is a n × n with integer entires and det(A) = 0}

Sarmad Abbasi (Virtual University) Theory of Computation 24 / 39

Encodings

Similarly we can encode graphs into strings: Let us do an example:

Consider the graph shown in the following figure:

Sarmad Abbasi (Virtual University) Theory of Computation 25 / 39

Encodings
This graph can be written as a vertex list followed by an edge list as
follows:

(1, 2, 3, 4)((1, 2), (2, 3), (3, 1), (1, 4)).

1 101 represents ).
2 110 represent (.
3 011 represents ,.
4 000 represents 0
5 111 represents 1.
and now it is easy to encode this whole graph into a 0/1 string.
The encoding starts as

110111011111000011111111011111

000000101110110111011111000101011110 · · ·
which is very long but not conceptually difficult.
Sarmad Abbasi (Virtual University) Theory of Computation 26 / 39
Encodings

Thus suppose we have

L = {G : G is connected }.

We can instead look at

L0 = {hGi : G is connected }.

This should not be surprising for you.

As computer scientist you know that everything inside a computer is
represented by 0’s and 1’s.

Sarmad Abbasi (Virtual University) Theory of Computation 27 / 39

Hilbert’s Tenth Problem

Recall from the first lecture:

Hilbert’s tenth problem was about diaphantine equations. Let us cast it
in modern language.

Devise an algorithm that would decide if a (multi-variate) polynomial

has integral roots

Here is an example of a multivariate polynomial:

6x 3 yz 2 + 3xy 2 − x 3 − 10.

Sarmad Abbasi (Virtual University) Theory of Computation 28 / 39

Hilbert’s Tenth Problem

Hilbert asked to study the language

H = {hpi : p is a polynomial with integral roots}.

Once again hi mean encodings.

Is there a Turing machine M that halts on all inputs and

L(M) = H

Sarmad Abbasi (Virtual University) Theory of Computation 29 / 39

Matijasevič Theorem

Theorem (Matijasevič 1970)

H is not decidable. There is no Turing machine M that halts on all
inputs with
L(M) = H.

Sarmad Abbasi (Virtual University) Theory of Computation 30 / 39

Matijasevič Theorem

Let look at a simpler language:

H1 = {hpi : p is a polynomial in x and has an integral root}

Lets see that H1 is Turing decidable:

Sarmad Abbasi (Virtual University) Theory of Computation 31 / 39

Matijasevič Theorem

Here is the description of a TM

1 On input p.
2 for i = 0, 1, 2, . . . ,
3 evaluate the polynomial on x = i and x = −i. If it evaluates to 0
accept.

Sarmad Abbasi (Virtual University) Theory of Computation 32 / 39

Matijasevič Theorem

Note the the above machine is not a decider. It is only a recognizer.

Show that H1 is also Turing decidable.

Sarmad Abbasi (Virtual University) Theory of Computation 33 / 39

How to describe a Turing machine

There are three ways to do it.

1 formal description.
2 implementation description.
3 high-level description.

Sarmad Abbasi (Virtual University) Theory of Computation 34 / 39

How to describe a Turing machine

In a formal description you have to specify the whole TM using the

7-things needed. That is you have to specify

(Q, Σ, Γ, q0 , qa , qr , δ).
We have seen examples of this in the last two lectures.

Sarmad Abbasi (Virtual University) Theory of Computation 35 / 39

How to describe a Turing machine
Implementation description: In this you refer to the tapes of the Turing
machine and explain how it moves and performs its task.
Let look at an example

L = {w#w : w ∈ {0, 1}∗ }.

A TM can be described as follows:
1 Scan the input and make sure it is of the form

w1 #w2

where w1 , w2 ∈ {0, 1}∗ . Or in other words make sure there is only

one # in the input.
2 Zigzag over the input matching symbols of w1 and w2 .
3 If all symbols match then accept. If any symbol fails to match
reject.

Sarmad Abbasi (Virtual University) Theory of Computation 36 / 39

How to describe a Turing machine

In a high level description we do not refer to the tapes. We simply

describe the Turing machine as an algorithm. We simply give
psuedo-code for the algorithm.
Suppose we wish to show that

L = {hji : i 2 = j for some i}.

A description of TM can be given as follows:

1 On input j.
2 For i = 1, . . . , j
3 check if i 2 = j if it is accept.
4 Reject
Note that in this case, we are assuming many things.

Sarmad Abbasi (Virtual University) Theory of Computation 37 / 39

How to describe a Turing machine

We are assuming that we can build a machine (subroutine) that can

multiply numbers, check for equality etc.

Sarmad Abbasi (Virtual University) Theory of Computation 38 / 39

How to describe a Turing machine

When you give a high-level description you must be convinced that you
can convert it into a formal one in principle. Although it will take a very
long time to do so.

Sarmad Abbasi (Virtual University) Theory of Computation 39 / 39