Theory of Computation
Arik Chen
arik.cj@gmail.com
Turing machine Grammar R.E.
M=(Q,E,A,T,q0,qa,qr) G=(V,T,P,S) recursive is a subset of R.E.
Q: a set of state V is the alphabet; if L is R.E, ~L may not R.E.
E: input alphabet set. T is all the terminal If L is recursive, then ~L is also.
A: tape alphabet set, large than E. S is a special non-terminal start L is recursive iff both ~L and L are
T: transition function symbol R.E
q0: start state qa: accept state e.g. cAd->aABd
qr:reject state
PDA CFG CFL
P=(Q,E,A,T,q0,F,) G=(V, T, P, S) some context free language is not
A: the stack alphabet P is the finite set of production. regular ().
T: transition relation ((state, tape, e.g. A->AB
stack), (state', stack))
q0: the initial state
F: a set of final states,
accept:(f,e,a) or (q, e, e)
FA regular expression regular language
A={Q,E,T,q0,F} all regular language are context
In DFA, T maps input (state, free language.
symbol) to one state
In NFA, T maps input (state,
symbol) to a set of states
Automata and Language
Relation of Languages
Finite Automata
NFA
DFA
Finite Automata
NFA vs DFA
Theorem: for each NFA, there is and equivalent DFA
the key idea is to view a NFA as occupying not a single
state but a set of state.
#state of DFA= 2#state of NFA (cause state explosion?)
see proof by induction (P71)
When state is considered, NFA is exponentially better
than NFA
Regular Expression
regular expression to FA
Theorem:the class of languages accepted by FA is
closed under union, concatenation, * , complementary,
intersection.
FA to regular expression
DFA to regular expression
NFA to regular expression
proof: L(FA) is regular language
Theorem: Accept by FA <=> a language is regular.
State Minimization
remove unreachable state
find equivalent classes
Theorem: there is a DFA accepts L with the precious
number of states as the equivalent classes of L .
Theorem: L is regular <=> equivalent classes is finite
for example {anbn}, [e],[a],[aa]...
Extreme of FA/regular
pumping lemma
if L is regular language
then there exists a constant n
each w longer than n in L can (exists y) break into three
strings w=xyz and satisfy that
y is not empty
|xy| is not longer than n
xykz is also in L
when there are finite number of string in language L,
pumping lemma is useless.
finite equivalent class
if we want show a language L is not regular, you should
prove that pumping lemma is not satisfied or the number
of equivalent class is infinite.
Context Free Grammar
CFL
Chomsky Normal Form
All production of the form A->BC or A->a
step 1: Eliminate nullable production
step 2: Elimitnate unit production
convert to production of length 2 or length 1
CFL closure property
close under union, concatenation, and star
not close under intersection or complementation(P147)
intersection of CFL and regular language is CFL.
parse tree
Q: Can we decide whether a language is CFL? CSL can not
be decide but semi decide.
Push-Down Automata
NPDA
the idea is use stack to remember the first half for further
check.
PDA
FA vs PDA
every FA can be viewed as a PDA without operate the
stack
PDA vs CFL
Each CFL is accepted by some PDA.
map productions to transition relation
if a language is accept by a PDA, it is a CFL.
map transition relation to productions.
Q: non-deterministic PDA is powerful than deterministic PDA?
Extreme of CFL/PDA
Pumping Theorem
G is a CFL
Any string longer than Fanout|non-terminal|
can be (exist) rewritten as w=uvxyz
either v or y is nonempty
uvnxynz is also in L(G)
to prove not CFL, show non is exist.
{anbncn}
{an:n is a prime}
CFG vs regular
there are CFL not regular (anbn), but all regular language
are context free.
state => non-terminal
symbol =>terminal
arrow=>production
Right linear Grammar <=> regular language
every regular language can be accept by right linear G
Language accept by right linear G is regular.
proof by construct CFG<=>NFA
Grammar
The rules decide what L(language) is exactly generated.
if not belong to this L, non-terminals can not be erased
L(G) iff L is RE
L(G)<=>L(M)
(q, u1a1w1)|-M (q', u2a2w2) iff u2a2q'w2=>u1a1qw1
If a string can be generate by the grammar, then it can be
searched by applying all the rules. If the string cannot, then
the search will never stop.
What can/cannot be computed?
Algorithm device excludes non-halt TM.
Church-Turing thesis: assert algorithm corresponds to a
certain Turing machine
Therefore, some computational problem cannot be solved.
That is computation task that cannot be performed by
Turing machines are impossible, hopeless, undecidable.
How to encoding Turing Machine?
if Turing Machine can receive encoding of Turing Machine
then there would be interesting operations.
Universal Turing Machine
U('M', 'W')='M(W)'
encode states
encode alphabet
encode transition function
Implement with 3-tape machine
undecidable problem
halt problem
diagonal(x)
a: if halt(x,x)
then goto a
else halt
Does diagonal(diagonal) halt?
d(d) halt when halt(d,d) return no which means d(d) not
halt, a contradiction.
That is to say, there can be no program halt(P,X) to
solve this problem: to tell whether arbitrary program can
halt or not
Halt Problem
Exist M semi-decide W, which means halt("M","W") is semi-
decided.
halt(M, W) recursive?
if yes. same contradiction may happen.
therefore NOT recursive
Decidable Problem
show decidable (recursive)
build a decider (TM)
reduction
know a solvable problem, and solved by Mk
then use Mk to construnct Muk to solve unknown
problem.
1. encode(reduce, map) input of Muk to input of MK
2. use Mk to work on the mapped input.
Undecidable Problem
show undecidable
reduction
find a known problem to be hard (unsolvable)
An unknown problem can be assumed decidable,
decided by Muk
1. encode instance (input) of known problem to input
of Muk
2. show that the problem is solved
however, known problem is unsolvable, so Muk is
impossible.
Does M halt on empty string?
