#A3 INTEGERS 21 (2021) A METHOD FOR GENERATING PERMUTATION POLYNOMIALS MODULO pn
Rajesh P. Singh Dept. of Mathematics, Central University of South Bihar, Gaya, Bihar, India [email protected]
Received: 6/10/20, Revised: 11/6/20, Accepted: 12/16/20, Published: 1/4/21
Abstract Using p-adic representation of integers, we obtain sufficient conditions for a polyno- mial over the residue class ring Zpn to be a permutation polynomial of Zpn , where p is a prime number and n a positive integer.
1. Introduction
Let R be a finite commutative ring. A polynomial f(x) ∈ R[x] is called a permuta- tion polynomial of R if the function x 7→ f(x) is a bijection of R. A natural question d is the following: given a polynomial f(x) = a0 + a1x + ··· + adx ∈ R[x], what are necessary and sufficient conditions on the coefficients a0, a1, . . . , ad for f(x) to be a permutation of R? This problem has not yet been solved. Permutation polynomials have several applications in combinatorics, coding theory, and cryptography, mostly when R is a finite field and the ring of residue classes of integers, (see, [1], [10], [11], [12], [13]). We denote Zm as the ring of non-negative integers less than m, under addition m ki and multiplication modulo m. It is known that for any m = Πi=1 pi , where pi’s are distinct prime numbers, f(x) is a permutation polynomial over Zm if and only if f(x) is also permutation polynomial over Z ki , for each i (see [12]). In 2001, pi n Rivest [8] considered the ring Zm, where m = 2 , n ≥ 1, and proved that the d polynomial f(x) = a0 + a1x + ··· + adx with integral coefficients is a permutation n polynomial modulo 2 if and only if a1 is odd, and both a2 + a4 + a6 + ··· and a3 + a5 + a7 + ··· are even. This is a special case of a more general characterization given by N¨obauer(1953)[7]:the polynomial f is a permutation polynomial of Zpn 0 if and only if f is a permutation polynomial of Zp and f (x) 6= 0 for all x ∈ Zp. Thereby N¨obauer’scharacterization reduced the problem to characterizing permutation polynomials over the finite field Zp. In 1983, Mullen and Stevens [6] used N¨obauer’scharacterization to count the number of permutation polynomial functions over the residue class ring Zpn , where p is a prime number and n a positive INTEGERS: 21 (2021) 2 integer. Recently G¨orcs¨os,Horv´athand M´esz´aros[2] have generalized these results about permutation polynomials over residue class ring Zpn to finite commutative unital local rings. Permutation polynomials with some additional properties are useful to construct public-key cryptosystems. Suppose we have a permutation polynomial p(x) for which computing the inverse is hard without some additional knowledge about p(x), but some structural information about p(x) paves the way to make it easy, also computationally. Such a polynomial can be effectively used for cryptosystems. In [11], the authors used two permutation polynomials f and g over finite fields F2m and imposed a relation f(s(x)) = g(t(y)) between plaintext variable x and cipher- text variable y, where s and t are secret invertible linear maps of F2m , to construct a multivariate public key cryptosystem. In [3], Khachatrian and Kyureghyan used linearized permutation polynomials over finite fields to propose a public-key cryp- tosystem. Permutation polynomials over residue class ring Zm are used in the design of RC6 block cipher [9], and in coding theory to construct a class of deterministic interleavers for turbo codes, [12], [13]. In this paper, we obtain certain sufficient conditions for a polynomial over the residue class ring Zpn to be a permutation polynomial. We show that every poly- nomial f(x) over the ring Zpn can be expressed as an n-tuple of multivariate trian- gular polynomials over Zp, that is, f(x) can be expressed as (f1, f2, . . . , fn), where fi = fi(x1, x2, . . . , xi). Using this representation, we obtain the desired sufficient conditions.
2. The Main Result
n n Let p be a prime and n > 1. A mapping F : Zp → Zp is triangular if F (x1, . . . , xn) = i (f1(x1), f2(x1, x2), . . . , fi(x1, . . . , xi), . . . , fn(x1, . . . , xn)), where fi : Zp → Zp are arbitrary functions, mapping (x1, x2, ··· , xi) 7→ fi(x1, x2, ··· , xi), for i = 1, 2. ··· , n. In 2002 [4], Klimov et. al. construct some classes of invertible transformations over n-bit words which can mix arithmatic and boolean operations (not, xor, and or). Motivated by this work, we show that every polynomial over finite ring Zpn can be expressed as a n-tuple of multivariate triangular polynomials over Zp. Note that by using p-adic representation of integers, any element x ∈ Zpn can be uniquely Pn i−1 n expressed as x = i=1 xip , where xi ∈ Zp. Let θ : Zpn 7→ Zp be a map de- fined as θ(x) = (x1, x2, . . . , xn). It is easy to see that the map θ is a bijection. n Since there is a one-one correspondence between Zpn and Zp , we can identify x by n n-tuple (x1, x2, . . . , xn) over Zp . Let Zpn [x] be the ring of polynomials over Zpn ; let R = Zp[x1, x2, . . . , xn] be the ring of multivariate polynomials in the variables Pn i−1 Pn i−1 x1, x2, . . . , xn. Now using the expansion f(x) = f( i=1 xip ) = i=1 fip , where f(x) ∈ Zpn [x], and fi ∈ R, 1 ≤ i ≤ n, we have an induced mapping INTEGERS: 21 (2021) 3
n Zpn [x] → R , which we also denote by θ given by θ(f(x)) = (f1, f2, . . . , fn), each fi is a multivariate polynomial over Zp in variables x1, x2, . . . , xn. To present our results systematically we need some lemmas.
Lemma 1. Let x, y ∈ Zpn and xi, yi, (x + y)i and (xy)i respectively denote the i-th n coordinates in the n-tuple representation of x, y, x + y and xy over Zp . Then
(i) (x + y)1 = (x1 + y1) mod p.
(ii) for i ≥ 2, (x + y)i = (xi + yi + αi) mod p, where αi is a function of the coordinates x1, x2, . . . , xi−1, y1, y2, . . . , yi−1.
(iii) for i ≥ 2, (xy)i = (xiy1 + x1yi + βi) mod p, where βi is a function of the coordinates x1, x2, . . . , xi−1, y1, y2, . . . , yi−1.
m m m m−1 (iv) for any integer m ≥ 2, (x )1 = (x1) mod p and (x )i = (mxi(x1) + γi) mod p, if i ≥ 2, where γi is a function of the coordinates x1, x2, . . . , xi−1.
Pn i−1 Pn i−1 Proof. We have x = i=1 xip and y = (y1, y2, . . . , yn) = i=1 yip . The first assertion in trivial. In (ii) αi is the carry over from the previous coordinates in the sum, so depends only on the coordinates x1, x2, . . . , xi−1, y1, y2, . . . , yi−1. For (iii) we note that xy = x1y1 + (x1y2 + y1x2)p + ··· + (xiy1 + yix1 + xi−1y2 + ··· + i−1 n−1 x2yi−1)p +···+(xny1 +xn−1y2 +···+x1yn)p from which it follows that (xy)i is the sum modulo p of the coefficient of pi−1 and the carry over from the previous coordinates. Finally, the second part of (iv) follows easily from (iii) by induction on m.
Pn i−1 Lemma 2. Let x = i=1 xip , k ≥ 1, 1 ≤ r ≤ p − 1. Then for i ≥ 2