Professional Documents
Culture Documents
1 s2.0 S0019995868910966 Main PDF
1 s2.0 S0019995868910966 Main PDF
1 s2.0 S0019995868910966 Main PDF
207
208 JONES
such t h a t m > l a n d V ( X , ,i x , i + 1 ) i s t r u e f o r i = 1 , - - - , m - - 1.
We will Call these sequences trajectories and we write them as:
x~ - ~ x, --* ... -~ z., or (4.1) -~ (~.') -~ ... -~ (~.').
214 JONES
where u~ and v~ are obtained from u~ and v~ b y deleting e and e' (if
present), and
(bl if b~ C A
v
h(b )= aaifb = all i= ,n
if b~ = e
The 6n -{- 2-ary transition predicate Vz(% 6) ,where ~, and ~ represent
sequences of 3n ~ 1 strings is defined by:
Vz(% ~) is true iff there is an I D a such t h a t ~ = ~, and either
(1) ~ = ~ for some ID/~ such t h a t a [- ~, or
(2) ~ is accepting and ~ consists of 3n -k 1 ~'s. |
:LEMMA 11. If Z accepts W(~2~) directly (i.e., without explicit trans-
216 JONES
For example, if W = {(a, b), (b, a), (b, ix)}, then W ( x l , x~)
(xl = a A x 2 - - - b) V (xl = b) A (x2 = a) V (xl = b / ~ x 2 = ix).
LEMMA 13. V z is in MAT, for any Turing machine Z = ( A , P ) .
Proof. Let P be 1-I~, 2 . I 2 , . . . , 1.Iz. Define the predicates ID,
Final and V~ as follows, where ~ and ~ are abbreviations for the sequences:
r , u ~ , b ~ , v l , - - . , u ~ , b ~ , v ~ a n d s , u~,t b l ! , v xt , - . . , u ~ t, b ~ ,t v ~ ,t r e s p e c -
tively.
(a) ID (~) is true iff 3, = ~ for some ID a of Z
(b) Final (% ~) is true iff ~, = a for an accepting I D a, and
is all X's.
(c) V~(~', ~) (for 1 < i = l) is true iff thereare I D ' s a and ~ such
t h a t 3, = ~ and ~ = ~, and a ~ f~ by applying instruction
i.I~.
It suffices to show t h a t these are in 5~[AT, since
V z ( 7 , ~) ~ V~=~ V,(7, ~) V Final (% ~).
I D ( ~ ) is in M A T by the following:
Now let F ( r ) hold iff/~ is Ac. This is finite, and so is in MAT. Then
we have: Final (% ~) ~-~ ID(3,) A F ( r ) A s = ix A v l ' = ix A bl' = ixA
• - . A v~' = ix.
Thus I D and Final are in MAT. For the V~ we have nine cases from
Definition 1. To shorten the definitions define Matcht by the following,
CLASSES OF AUTOMATA AND TRANSITIVE CLOSURE 217
Note: The step "compare the positions of heads 3 and 4" can
be done as follows:
1. Move the head on tape 1 to the right end.
2. Do the following steps repeatedly, until e' is found on tape 3
and/or 4:
Move the head on tape 1 left one symbol; move the heads on
tapes 3 and 4 right one symbol.
3. If e' is encountered on tapes 3 and 4 simultaneously, the heads
were in the same positions; otherwise they were not.
4. Restore the original head positions by repeating the following
until e' is scanned on tape 1:
Move the head on tape 1 right one symbol; move the heads on
tapes 3 and 4 left one symbol. |
I~EMM~- 25. There are nondeterministic machines RAND0, RAND1,
and RAND2 such that:
(a) RAND1 is linear bounded, and RAND: is two-way nonwriting.
(b) RAND~ (i = 0, 1 or 2) will when given the eight tapes above,
generate a string Xj+l such that Ri(x~+l , xl) is true, and then
stop.
CLASSES OF AUTOMATA AND TRANSITIVE CLOSURE 225
(c) For each xi+i such that Ri(xi+l, xl) is true, R A N D i has a
computation which generates it.
Proof. The nondeterministic jump instruction "Jp, q" is used for
randomness. I n R A N D 2 , it is clear that xi+l can be obtained by moving
the "bound heads" on tapes 6 and 8, since it is required that xj+~ Px.
For RAND~, it is necessary first to do such movement to determine the
size of xj+l. Since ] xj+~ ] -< [ x I, no expansion is required. RAND1 will
then write a random string x~.+~ within these bounds.
RAND0 can be very similar to R A N D y , but with one difference:
RAND0 must be able to expand tapes 6, 7 and 8 when determining the
size of x~+~.
Using these guidelines, construction of R A N D 0 , RAND~, and RAND2
is straightforward. |
Proof of Theorem 22. Let Zb and RAND~ be as above, for i = 0,.1, 2.
Construct Z' to act as follows:
I. Move the head on tape 4 to the right end, so that xi = x.
II. Apply I~AND~, to yield a candidate xj+~ on tapes 6, 7, 8.
III. Apply Zb to evaluate V(xj, xs+~).
IV. If true, compare xi+~ and y (x~+~ can be found intact on tape
6). If false, stop, rejecting•
V. If xj+~ = y, accept the input.
VI. If unequal, copy xj+~ in place of x j , (i.e. tape 7 to tape 3),
and go to II.
The above will generate the trajectory x = Xl -~ x2 --~ . . . , putting
xj- on tape 3 and xj+l on tape 7, f o r j = 1, 2, . . . .
Steps I through V are either elementary or have been covered by
Lemmas, and so can be done b y the same type of automaton as Z. Step
VI involves copying, which can be done directly by a Turing machine
or a linear-bounded automaton, and indirectly by head movements for a
nonwriting automaton, since xj.+~ is a substring of x.
Suppose T~(V)(x, y) is true, so there is a trajectory x = xl--~ X2 --+
• .. --~ xk = y, such that V(xi, xj+l) holds f o r j < k. Then Z' w i l l b e
given eight tapes initially containing (x, x, x, x, y, x, x, x ) and Scanned
at their left ends. Step I sets xj = x, and step II can generate x2 on
tapes 6, 7, 8, by Lemma 25. Zb then accepts (Xl, x2) and x~ is compared
to y. If unequal, x2 is copied onto tapes 2, 3, 4. A new string x~ is theri
generated on tapes 6, 7, and 8. Since T~(V)(x, y ) i s truel there is a cor-
rect sequence of choices of x2, x3, -.- , xk by R A N D y , and so (x, y)
will ultimately be accepted in at least one computation•
Conversely, if T~(V)(x, y) is false, there is no sequence of choices
226 - JO)rEs
Proof. Enuml merely has to add i to x~.+l if possible within its bounds,
and it is well k~own that linear-bounded automata can add. Enum2
can do its Enumeration by moving its "bound heads" on tapes 6 and 8,
so construction is again straightforward. |
Proof of Theorem 23. We only need consider i = 1 or 22 since Corollary
26 handles TMz. By Theorem 21 we may assume that Z strongly accepts
V(x, y). Let Zb be as in Lemma 24, and Enum~ as in Lemma 28. Con-
struct Z' to accept T~(V), as follows:
I. Move the head on tape 4 to the right end, so that xi = x.
II. Move the head on tape 8 to the left end, so that x j+l = ~.
III. (a) Apply Zb to evaluate V(xj, xj+~).
(b) If V(xs, xj+~) is true, go to step IV.
(c) If false, copy the original xj'+l from tape 6 to tape 7 (it
may have been destroyed by Zb).
(d) Apply Enumi to xi+~, to obtain the next trial value in the
list E~(x~), if any.
(e) Go to III (a) if there is a next xj+~ in E~(xi).
(f) Reject if the present X~+l is the last string in E~(x~).
IV. Compare xj+l to y(xj+~ can be found on tape 6).
V. If equal, accept the input.
VI. If not equal, copy xj+~ onto x~, and go to II.
By the Lemmas, each step can be performed by the required deter-
ministic type of automaton. The computations for Z' are very similar
to those of Theorem 22. The only essential difference is in steps Ii and
III, where Z' of Theorem 22 picks an x~.+~ at random, our present Z'
enumerates all the possible values of x5+1 (for a given x j), trying each
one until it finds the one (if any) for which V(x~, x~+~)is true.
We see as before that Z' accepts Tid(V). |
THEOREM 29. TM = TR0, LBA = TR1, and MTA2 = TR2;
TMd = TRod, LBAd = T R y , and MTA2~ = TR2d.
Proof. By Theorems 14 and 16, TM ___ TR0, LBA ___ T R t , MTA
TR2, TM~ ~ TR0, LBA~ ~ TR~a, and MTA2d c T R y . By definition
of the TRi classes, Lemma 17 and 19, and Theorems 22 and 23, the re-
verse containments also hold. |
The following theorem shows that these classes are also closed under
appropriate quantifiers. Proof is omitted, as the same results are prove~l
in [S], [M] and [KS]. Independent and much simpler proofs can easily
be constructed using Theorem 29 and transitive closure.
THE0i~EM 30. Define the bounded quantifier operations Y<, 3<, Wr
228 ZONES
and 3 e as follows:
(Vz)<yR(~, y) ~ Vz( I z [ <- ] Y -* R ( ~ , y))
(3z)<~R(~., y) e-~ 3z( [ z I =< [ Y A R ( ~ , y))
( V z ) v ~ ( ~ , y) e-~ Y z ( z P y --->R(~, , y) )
(3z)eyR(2~ , y) ~ ~ z ( z P y /~ R(£~ , y) ).
then
(i) TR0/s closed under 3.
(ii) TR1 and TR~ are closed under V< and 3 < .
(iii) TR~ and TR2d are closed under Ve and 3 ~ .
CO~OLLAaY 31.
(i) TRu contains the rudimentary predicates.
(ii) T I ~ contains the S-rudimentary predicates.
Results (i) and (ii) are the main theorems of 5/Iyhill [M] and Kreider
and Ritchie [KR], respectively.
5. CONCLUSIONS AND FURTHER POSSIBILITIES
The goal of this paper has been to develop a set of tools which faith-
fully describe automata, and which are more convenient, comprehensible,
and flexible than the more customary transition tables and programs.
The success of this approach will be measured in terms of its utility and
the insights which it gives into the nature and powers of automata of
:~rious types.
It is clear that this approach can be extended to cover other types of
n/Ult~tape automata, for example by adding different bounding relations
R~(x, y). The author strongly suspects that Ritchie's classes F~ [R] can
be c0mpletely described by appropriate bounding conditions R. In any
case, the methods used in Lemma 19 and Theorems 21, 22 and 23 appear
to be quite generally extendable to various types of two-way automata.
•/(s a final remark, it should be noted that it is quite simple to impose
eoaditioi£s on V(x, y) to make it correspond to the derivation rules of
formal grammars, so that ~ormal languages could be studied by these
tools.
l~c~i.v~D: February 27, 1968; revised July 26, 1968.
REFERENCES
B. Biic~, J. R. (1960), Weak second-order arithmetics and finite ~u~omata de-
sign. Zeitschr. f. Ma~h. Logik und Grundl. d. Math. B6, $66-92.
CLASSES OF AUTOMATA AND TRANSITIVE CLOSURE 229