Construction of inversive congruential pseudorandom number generators with maximal period length

Construction of inversive congruential pseudorandom number generators with maximal period length

Journal of Computational North-Holland and Applied Mathematics -ri) (1992) 345-349 345 CAM 1115 Construction of inversive congruential orandom num...

382KB Sizes 0 Downloads 102 Views

Journal of Computational North-Holland

and Applied Mathematics -ri) (1992) 345-349

345

CAM 1115

Construction of inversive congruential orandom number generators with maximal perio Jiirgen Eichenauer-Herrmann Fachbereich Mathematik, Technische Hochschule, Darmstadt, Germany Received 25 March 1991 Revised 21 May 1991

Abstract Eichenauer-Herrmann, J., Construction of inversive congruential pseudorandom number generators maximal period length, Journal of Computational and Applied Mathematics 40 (1992) 345-349.

with

The inversive congruential method for generating uniform pseudorandom numbers is a particularly attractive alternative to linear congruential generators with their well-known inherent deficiencies like the unfavourable coarse lattice structure in higher dimensions. In the present paper the modulus in the inversive congruential method is chosen as a power of an arbitrary odd prime. The existence of inversive congruential generators with maximal period length is proved by a new constructive characterization of these generators. Keywords: Pseudorandom iength.

numbers,

inversive congruential

method, prime power modulus, maximal period

1. Introduction The linear congruential method for generating uniform pseudorandom numbers in the interval [0, 1) shows a lot of undesirable regularities which are due to the linearity of the underlying recursion (cf. [12,13,18,19]). Therefore several nonlinear congruential generators have been proposed and analysed (cf. [ l-l 1,14- 17,193). A particularly promising method is based on achieving nonlinearity by employing the operation of multiplicative inversion with respect to a given modulus. In case the modulus is a prime (cf. [1,3,6,14,16,17]) or a power of two (cf. [5,8,10,16]), several results on the corresponding inversive congruential sequences are available. When the modulus iq a power of an arbitrary odd prime, inversive congruential sequences with maximal period length have been characterized (cf. [ll]) and their statistical independence pra;nerties have been analysed (cf. [7]). However, the existence of inversive Correspondence to. Dr. J. Eichenauer-Herrmann, straBe 7, W-6100 Darmstadt, Germany. 0377-0427/92/$05.00

Fachbereich

Mathematik,

Technische Hochschule, SchloBgarten-

0 1992 - Elsevier Science Publishers B.V. All rights reserved

,346

congruenti sequences with maximal period length could not yet be proved for moduli being a power of an arbitrary odd prime. This was the motivation for the present paper, where inversive congruential generators with maximal period length are explicitly constructed. Let p a 3 be a prime, and let m 2 2 be an integer. For integers CI, 6 with a f 0 (mod p) a sequence (x,),,, of integers with xn + 0 (mod p) for n 2 0 is called an inversive congruentiai sequmce if the recursion x ?z+\ =ax;‘+b

(mod pm),

n20,

is satisfied. where X,-’ denotes the multiplicative

inverse of x, mol;Jo

p”. Let

h = min(n 2 I Ix,, =x0 (mod pj\ be the period length of (x,1, z ,, medulo p. A method for computing the period length A is described in [3]. In the present paper the existence of inversive congn;ential sequences with maximal period length hpm- ’ modulo pm is prover; by an explicit construction. In the second section the main result is stated precisely. Its comprehensive proof is sketched in the third section.

2. Sequences with maximal period length First, a characterization of inversive congruential sequences with maximal period length hP m-1 modulo pm is given, which can easily be deduced from [al, Theorem 121. Proposition 1. (a) Suppose that A = 1, i.e., x0’- bx, - a = 0 (XXX! p), and 8m> 3. Then the incersire congruential sequence (xnJn a 0 has maximal period length pm- 1 modulo pm if and only if (I) (11)

a=2

(mod3)

and

Q = -x0’ (mod p)

xz-b.y,-a=6 and

(mod9)

xi-bx,-af0

forp=?

(mod p2)

or

forp>5.

(b) Suppose that A > 2, i.e., xi - bx, - a f 0 (mod p>. Then the inversive congruential sequence (x,), a 0 hn.o maximal period length Ap”- ’ module pm if and only if x, f x, (mod p2). Proposition 1 shows that inversive congruential sequences with maximal hp”-’ can be characterized by simple explicit conditions for A = 1, whereas crucial condition x, f x, (mod p2) is only implicit. The difficulty in evaluating arises from the nonlinearity of the underlying recursion. Nevertheless, in the result an explicit characterization is established.

period length for A > 2 the this condition following main

Theorem 2. Suppose that A 2 2. Let c be an integer with x, = x0 + pt (mod p2 ). For integers cy, #%let (Y,!. 20 be a sequence of integers with y, =x0 (mod p) and Yn-L1 . = (a +par)y;’

+b +p/? (mod pm),

n 20.

3en the inversive congruential sequence ( y,), ~ C has maximal period length A pm - ’ module pm if and only if A(xg - &x0

-

a)(ba - 2ap) + [a(4a + b2) f 0 (mod p).

J. Eichenauer-Herrmann

3. Proof of Thwem

/ Pseudorandom number generator

347

2

(i) First, observe that the sequence ( y,), , 0 is well defined since y, =x, (mod I;) for n 3 0. In the following presentation the integer ma&ix A=

(

;

a 01

plays an important role. Let I denote the unit matrix and let (q,Jn 3 o be a sequence of integers with q0 = 0, qr - 1 and 4, =lyn+ +aqn_2 for n 2 2. Then A” =aq,_,I+q,A and -’

X,=-((4,+1xo+a4~j(4~Xo+Q4n-1)

(mod pm)

for n 2 1 can easily be proved by induction. Therefore x0 +~5

=q

= (qA+Ixo

+ w&7,x0

one obtains

+ w,_,)-’

(mod

p2),

which yields qh( xu”- 5x0 -a) Now, the assumption hence

+pt(q*x,

+aq,_,)

= 0 (mod p’).

A 3 2, i.e., xi - bx, - a f 0 (mod p), implies that qA=puo (mod p2) and

Ah = aq,_,Z +pv,A

(mod p”),

where v. denotes an Integer with v. = -5(x: (ii) Let vl, v2 be integers with 1/I = h(4a -Fb2)-l(2a

+ b@qh_r

- !xo - a!-‘aq, _ 1 (mod p). (mod p)

and v2 = h(4a + b2)-‘(2@ Plate th:t 4.a + b2 f 0 (mod pi

- bcx)q,_, follows

(mod p).

from part (ii) of the Theorem

be an integer matrix with (4a + b2)B = (ICY+ b@A + (2ap - ba)I

(mod

Then, according to (i), one obtains (A+pB)A=AA+pAAh-lB=aq,,_,I+p(uoA+haq,_,A-lB) =aq,d+p( =aq,_,I+p

uoA+au,I+av,A-‘) b:), + au, uo + u2

p).

in 131.Furthermore,

let

J. Eichenauer-Herrmann / Pseudorandom number generator

,343

iet ( zJn 2 o be a sequence of integers with

(iii) Subsequently,

z. 15(1 -p&)y,

(mod p’)

-p/S3

and z n+1= (V +P&k

+ u‘ +p&)((l

+P&)z,

+P&) -’ (mod P’),

n a 0.

ote that the sequence (z,!, 2o is well defined since z,, =xX, (mod p) for n 2 0. Furthermore, let ri”‘, . . . , ypJ be integers with (A +pB)”

=

for n 2 0. Then z, =

($)z~

+ y~))(y~)zo

+ yr))-’

(mod p”)

for N 2 0 can easily be proved by induction. Hence, it follows from (ii) that z, = ((WA-l +P(6v,+av,))zo+Pa(VO+VZ)) l

+ “2J~n

{Pi+J

= (aqA-lz,

+

aq,_,

+pjav,

- 6~2))

+p[(6vo+av,)zo+ajvoi

c2>3)

-P(aq,-,)-2[(vo+

- b?*-J1

-1

+o+av,

+,I)

=zo-p(aq,_,)-1(vo+vI)(z~-6zo-a) =zzo +-pa-‘(4a

+ 6’)-I[

A(x,’ -6x,

- a)(ba

- 2@)

+ ea(4a

+

b’)] (mod p’).

(iv) Finally, a short calculation and an induction argument show that yn: 3 (1 +PP3)Zn

+Ppa

(mod

P’),

n 3 0,

where the relations ar = & + a& - 6p4 (mod p) and p = PI + & (mod p) are used. Therefore y’hfy, ‘\mod p’) if and only if t, f z. (mod p2), i.e., according to (iii), if and only if A(xi -6x,

- a)(ba

- 2ap)

+ ca(4a

+

b2) f 0 (mod p).

Hence, the desired result follows from Proposition

l(b).

terenrces ill J. Eichenauer. H. Grothe and J. Lehn, Marsaglia’s lattice test and non-linear congruential

pseudo random number generators, Metrika 35 (1988) 241-250. if21J. Eichenauer, H. Grothe, J. Lehn and A. Topuzoj$~, A multiple recursive non-linear congruential pseudo random number generator, Manuscripta Math. 59 (1987) 331-346. 31 J. Eichenauer and J. Lehn, A non-linear congruential pseudo random number generator, Statist. Papers 27 t !986) 315-326. 141 J. Eichenauer and J. Lehn, On the structure of quadratic congruential sequences, Manuscripta Math. 58 (1987) 129-140.

J. Eichenauer-Herrmann / Pseudorandom number generator

349

[5] J. Eichenauer, J. Lehn and A. Topuzoglu, A nonlinear con E -:lential pseudorandom number generator with power of two modulus, Math. Comp. 51 (1988) 757-759. [6] J. Eichen auer-Herrmann, Inversive congruential pseudorandom numbers avoid the planes, Math, Camp. 56 (1991) 297-301. [7] J. Eichenauer-Herrmann, On the discrepancy of inversive congruential pseudorandom numbers with prime power modulus, Manuscripta Math. 71 0991) 153-161. [8] J. Eichenauer-Herrmann, H. Grothe, H. Niederreiter and A. Topuzoglu, On the lattice structure of a nonlinear generator with modulus 2*, J. Comput. Appl. Math. 31 (1) i1990) 81-85. [9] J. Eichenauer-Herrmann and H. Niederreiter, On the discrepancy of quadratic congruential pseudorandom numbers, J. Comput. Appf. Math. 34 (2) (1991) 243-249. [lo] J. Eichenauer-Herrmann and H. Niederreiter, Lower bounds for the discrepancy of inversive congruential pseudorandom numbers with power of two modulus, Math. Comp., to appear. [ll] J. Eichenauer-Herrmann and A. Topuzoglu, On the period length of congruential pseudorandom number sequences generated by inversions, .I. Comput. Appl. Math. 31(l) (1990) 87-96. [12] G. Marsaglia, Random numbers fall mainly in the planes, Proc. Nat. Acad. Sci. 61 (1968) 25-28. [13] G. Marsaglia, Regularities in congruential random number generators, Numer. Math. 16 (1970) 8-10. [14] H. Niederreiter, Remarks on nonlinear congruential pseudorandom numbers, Metrika 35 (1988) 321-328. [ 151 H. Niederreiter, Statistical independence of nonlinear congruential pseudorandom numbers, Monatsh. Math. 106 (1988) 149-159. [16] H. Niederreite:, The serial test for congruential pseudorandom numbers generated by inversions, Math. Camp. 52 (1989) 135-144. [17] H. Niederreiter, Lower bounds for the discrepancy of inversive congruential pseudorandom numbers, Math. Comp. 55 (1990) 277-287. [18] H. Niederreiter, Recent trends in random number and random vector generation, Ann. Oper. Res. 31 (1991) 323-346. [19] H. Niederreiter, Nonlinear methods for pseudorandom number and vector generation, in: Proc. Internat. Rorkshtg on Computationally-Intensive Methods in Simulation and Stochastic Optimization, Laxenburg, 1990, to appear.