2
IEEE TRANSACTIONS ON COMPUTERS, FEBRUARY 1970 Define M'l(a)I(X, Y, S, { M'(y I x) }) by letting m«i(yI X) ai(yI x)mi(yI x) + (1 - ai(yI x))ni(yI x) where ai(yI x) [0, 1], i= 1,2,. * c (2) for each pair (x, y)EH(X, Y). Post-multiplying (2) by H and using (1), we get mi'(y x)H = oi(y | x)g;(y x) + (1 - ai(y x))9i(y I x) = gi(y x) i = 1, 2, . . .*, c where gi(y I x) _ ith row of the matrix G(y | x). Consequently, M'(y x)H= G(y| x) for every pair (x, y) EE(X, Y), and thus M'(a)EEG,(MO) for all ai(ylx) E[o, 1], i=l, 2, * , c, and the assertion is thus proved. Q.E.D. REFERENCES [1] J. W. Carlyle, "Reduced forms for stochastic sequential ma- chines," J. Math. Anal. Appi., vol. 7, pp. 167-175, 1963. [21 G. C. Bacon, "Minimal-state stochastic finite state systems," IEEE Trans. Circuit Theory (Correspondence), vol. CT-11, pp. 307-308, June 1964. [31 S. Even, "Comments on the minimization of stochastic ma- chines," IEEE Trans. Electronic Computers, vol. EC-14, pp. 634- 637, August 1965. [41 J. W. Carlyle, "State-calculable stochastic sequential machines, equivalences, and events," 1965 IEEE Conf. Rec. on Switching Theory and Logical Design, pp. 258-263. Solution to Harrison's Problem K. K. NAMBIAR Abstract-The complete solution to a number theoretic problem which has applications in relay contact networks is given. As a corol- lary, a conjecture by Harrison is shown to be true. Index Terms-Number theoretic problem, relay contact net- works, uniform loading. While commenting on a letter by Fielder [1] on Shannon's almost uniform distribution, Harrison [2] has shown that the only solution to the congruence 2= 3 mod (n - 1) is n = 2. After solving this congruence, Harrison states that "a problem of more difficulty is the following: a necessary and sufficient condition for uniform loading in relay contact networks of n variables is that n be a solution of the following congruence: 2n 2 mod (n -1)." 1 Harrison gives some partial results and a conjecture. In this note we show that his conjecture is true and give the complete solution to (1). Since (1) is trivially true for n=2, we will exclude this solution from further consideration. Manuscript received April 29, 1969. The author is with the Department of Electrical Engineering, Drexel Institute of Technology, Philadelphia, Pa. 19104. Case A, when n is even: We have to solve the congru- ence 2n " 2 mod (n - 1) or equivalently 2n-= 1 mod (n - 1) (2) since n -1 is odd. Let p5alp2a2 . . .p Pra be the prime power decomposition of n -1. Note that no pi can be equal to 2 since n -1 is odd. As a consequence of Euler's theorem [3 ], we can state that the solution to (2) is given by all those n for which the ratio n-i 4(n - 1) is an integer. Here 4(k) represents the well-known Euler function defined as the number of numbers of the sequence which are relatively prime to k. Since [3], n-I PIP2 .. *Pr o(n -1) (pl -1) (p2 -1) ... (pr -1) (3) we have to find those sets of pi which make the right- hand side of (3) an integer. But clearly, it is impossible to find such a set, since the numerator is odd and the denominator is even. Hence, we conclude that an even n cannot be a solution of (1). Case B, when n is odd: Putting n=2m+1, we have to solve the congruence 22m+l- 2 mod 2m or equivalently 22m = 1 mod m. (4) It is easy to see that m cannot be even, since the odd number (22m - 1) cannot be divided without a remainder by an even m. Let qlblq22 . . . q,be be the prime power decomposition of m. Note that no qi can be equal to 2, since we have already established that m cannot be even. The solution to (4) is given by all those m for which the ratio 2m 2qlq2 q5 ¢0(m) (q, -1) (q2 -1) . . .(q - 1) (5) is an integer. The only way to make the right-hand side of (5) an integer is to have s = 0 or s = 1 and qi = 3. Hence the general solution of (4) is m = 3k, k = 0,1, 2 Theorem: General solution to the congruence 2" _ 2 mod (n - 1) is n=1 + 2.3, k = 0,1, 2, **. 166

Solution to Harrison's Problem

  • Upload
    kk

  • View
    219

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Solution to Harrison's Problem

IEEE TRANSACTIONS ON COMPUTERS, FEBRUARY 1970

Define M'l(a)I(X, Y, S, { M'(yI x) }) by letting

m«i(yI X) ai(yI x)mi(yI x) + (1 - ai(yI x))ni(yI x)

where

ai(yI x) [0, 1], i= 1,2,. * c (2)

for each pair (x, y)EH(X, Y). Post-multiplying (2) by Hand using (1), we get

mi'(y x)H = oi(y | x)g;(y x) + (1 - ai(y x))9i(y I x)= gi(y x) i = 1, 2, . . .*, c

where

gi(y I x) _ ith row of the matrix G(y | x).

Consequently, M'(y x)H= G(y| x) for every pair (x, y)EE(X, Y), and thus M'(a)EEG,(MO) for all ai(ylx)E[o, 1], i=l, 2, * , c, and the assertion is thusproved. Q.E.D.

REFERENCES[1] J. W. Carlyle, "Reduced forms for stochastic sequential ma-

chines," J. Math. Anal. Appi., vol. 7, pp. 167-175, 1963.[21 G. C. Bacon, "Minimal-state stochastic finite state systems,"

IEEE Trans. Circuit Theory (Correspondence), vol. CT-11, pp.307-308, June 1964.

[31 S. Even, "Comments on the minimization of stochastic ma-chines," IEEE Trans. Electronic Computers, vol. EC-14, pp. 634-637, August 1965.

[41 J. W. Carlyle, "State-calculable stochastic sequential machines,equivalences, and events," 1965 IEEE Conf. Rec. on SwitchingTheory and Logical Design, pp. 258-263.

Solution to Harrison's ProblemK. K. NAMBIAR

Abstract-The complete solution to a number theoretic problemwhich has applications in relay contact networks is given. As a corol-lary, a conjecture by Harrison is shown to be true.

Index Terms-Number theoretic problem, relay contact net-works, uniform loading.

While commenting on a letter by Fielder [1] onShannon's almost uniform distribution, Harrison [2] hasshown that the only solution to the congruence

2= 3 mod (n - 1)

is n = 2. After solving this congruence, Harrison statesthat "a problem of more difficulty is the following: anecessary and sufficient condition for uniform loadingin relay contact networks of n variables is that n be asolution of the following congruence:

2n 2 mod (n -1)." 1

Harrison gives some partial results and a conjecture. Inthis note we show that his conjecture is true and givethe complete solution to (1). Since (1) is trivially truefor n=2, we will exclude this solution from furtherconsideration.

Manuscript received April 29, 1969.The author is with the Department of Electrical Engineering,

Drexel Institute of Technology, Philadelphia, Pa. 19104.

Case A, when n is even: We have to solve the congru-ence

2n" 2 mod (n - 1)

or equivalently

2n-=1 mod (n - 1) (2)

since n -1 is odd. Let p5alp2a2 . . .pPra be the prime powerdecomposition of n -1. Note that no pi can be equal to2 since n -1 is odd. As a consequence of Euler's theorem[3 ], we can state that the solution to (2) is given by allthose n for which the ratio

n-i

4(n - 1)is an integer. Here 4(k) represents the well-known Eulerfunction defined as the number of numbers of thesequence

which are relatively prime to k. Since [3],

n-I PIP2.. *Pr

o(n -1) (pl -1)(p2 -1) ... (pr -1)(3)

we have to find those sets of pi which make the right-hand side of (3) an integer. But clearly, it is impossibleto find such a set, since the numerator is odd and thedenominator is even. Hence, we conclude that an evenn cannot be a solution of (1).

Case B, when n is odd: Putting n=2m+1, we haveto solve the congruence

22m+l- 2 mod 2m

or equivalently22m = 1 mod m. (4)

It is easy to see that m cannot be even, since the oddnumber (22m - 1) cannot be divided without a remainderby an even m. Let qlblq22 . . . q,be be the prime powerdecomposition of m. Note that no qi can be equal to 2,since we have already established that m cannot beeven. The solution to (4) is given by all those m forwhich the ratio

2m 2qlq2 q5

¢0(m) (q, -1)(q2 -1) . . .(q - 1)(5)

is an integer. The only way to make the right-hand sideof (5) an integer is to have s = 0 or s = 1 and qi = 3. Hencethe general solution of (4) is

m = 3k, k = 0,1, 2

Theorem: General solution to the congruence

2" _ 2 mod (n - 1)

is

n=1 + 2.3, k = 0,1, 2,**.

166

Page 2: Solution to Harrison's Problem

SHORT NOTES

Harrison has conjectured [2 ]that (1) implies

n 1 mod 6.

It is clear from the general solution that this conjectureis true provided we exclude the solution n =3 and thetrivial solution n = 2.

REFERENCES[1] D. C. Fielder, "On Shannon's almost uniform distribution,"

IEEE Trans. Electronic Computers (Correspondence), vol.EC-13, pp. 53-54, February 1964.

[2] M. A. Harrison, "A remark on uniform distribution," IEEETrans. Electronic Computers (Correspondence), vol. EC-13, pp.630-631, October 1964.

[3] J. Hunter, Number Theory. London: Oliver and Boyd, 1964.

General Repetitive Events and MachinesB. G. REYNOLDS

Abstract-A finite automaton is a general repetitive machine ifthere is a path in the state graph from every final state to the startstate. In this paper general repetitive events are defined in terms ofcertain factorization properties of the associated regular expressions,and a one-to-one onto correspondence is shown between generalrepetitive machines and general repetitive events.

Index Terms-Automaton, finite automaton, general repetitiveevent, general repetitive machine, repetitive event, repetitive ma-chine.

INTRODUCTION

Although there are techniques for going from regularexpressions to state graphs and vice versa [1 ], [2 ], thereis no general theory relating forms of regular expressionsto state graph structures. Insight into such relationshipshas been developed through investigations of particularclasses of events and automata [3 ]- [5 ]. In keeping withthis approach we define in this paper a class of autom-ata and a class of events and show the correspondencebetween them. The class of automata was defined byReynolds and Cutlip [6]. The reader is referred toSection 2 of that paper for the notation and terminologyused here.

GENERAL REPETITIvE EVENTS AND MACHINES

The following definition is taken from [6].Definition 1: An automaton A = (S, M, so, F) is a

general repetitive machine (GRM) iff for each state si Fthere is a tape xE* such that M(si, xi) =so.

If for some tape x, M(s, x) so for all s(F, A is arepetitive machine (RM1).A general repetitive machine is thus an automaton

having a path from every final state to the start state.The class of repetitive machines forms a proper subclassof the class of general repetitive machines; in an RM asingle tape x will take A from any final state to s0, but

Manuscript received February 27, 1969; revised August 21, 1969.Material for this paper was taken from the author's Ph.D. disserta-tion [7]. This work was supported by the U. S. Air Force Office ofScientific Research under Contract AFOSR-67-1023.

The author is with Texas Instruments Incorporated, Dallas, Tex.75222.

some GRMs require different tapes for different finalstates.

It was also shown in [6] that every GRM is eitherstrongly connected or consists of a single nonfinal deadstate SD and a strongly connected subset S-{ SD }. Notethat the results of this study apply to strongly con-nected machines, since these are a proper subclass ofGRMs.We now define repetitive and general repetitive

events. To avoid cumbersome notation, we make nonotational distinction between an event and a regularexpression denoting that event.

Definition 2: A repetitive event (RE) is a regular eventP such that, for some tape xG i*,

1) P can be written P=P(xP)*, and2) for any events R and Q, if RCP and RxQCP

then QCP.

The tape x is called a return tape. The class of REs isdenoted by R.

Definition 3: A general repetitive event (GRE) is aregular event R with the properties

1) R = UJ 1 Pi, where each Pi is an RE with returntape xi,

2) for 1 < i, j < m, Pi can be written Pi = (Pjxj) *Pi,and

3) for 1<i, j<m and for any events Ri and Q, ifRiCPi and RixiQCP1 then Qcpj.

The class of GREs is denoted by G.Observe that an RE is a GRE, and hence R is a proper

subclass of G. Also, no finite event belongs to eitherclass.A lemma is needed for later use.Lemma 1: If P is an event and x is a tape, the follow-

ing are equivalent:P = P(xP)*P = (Px)*PPxPC P.

(1)(2)

(3)Proof: The equivalence of (1) and (2) is seen as

follows:

P(xP)* = P[A + xP + xPxP + * * ]

= P + PxP + PxPxP+

= [A + Px+ PxPx+ * *P

= (Px)*P.

That (1) implies (3) is trivial. Conversely, if PxPCP,then (PxP)xPCPxPCP, or P(xP)2CP, and in likefashion we observe that P(xP)nCP for n=0, 1, 2, * *.*where (xP)O =A. Hence P(xP)*CP. Obviously PCP(xP) * so P=P(xP)*. Q.E.D.

The following theorem relates GREs to GRMs. Allautomata are assumed to be reduced.

Theorem 1: Let A (R) = (S, M, so, F) be a finite autom-aton. Then A (R) is a GRM if and only if R is a GRE.

167