1: \documentclass{article}

2: 

3: \usepackage{amsmath,amssymb,bm}

4: %\renewcommand{\atop}[2]{\genfrac{}{}{0pt}{}{#1}{#2}}

5: \usepackage{cite,url,color}

6: 

7: \usepackage[top=2.5cm,bottom=2.5cm,left=2cm,right=2cm]{geometry}

8: 

9: \usepackage{longtable}%,colortbl}

10: 

11: \newtheorem{definition}{Definition}

12: \newtheorem{notation}{Notation}

13: \newtheorem{theorem}{Theorem}

14: \newtheorem{lemma}{Lemma}

15: \newtheorem{corollary}{Corollary}

16: \newtheorem{conjecture}{Conjecture}

17: \newtheorem{example}{Example}

18: \newtheorem{remark}{Remark}

19: \newtheorem{problem}{Open Problem \#}

20: \newenvironment{proof}[1][Proof]{\par\noindent\textit{#1}: }{\hfill$\blacksquare$\vskip 0.5\baselineskip}

21: \newenvironment{solution}[1][Solution]{\par\noindent\textit{#1}: }{\hfill$\square$\vskip 0.5\baselineskip}

22: \newenvironment{explain}[1][Explain]{\par\noindent\textit{#1}: }{\hfill$\square$\vskip 0.5\baselineskip}

23: 

24: \newcommand\lcm{\mathrm{lcm}}

25: 

26: \newcommand\fproblem[1]{%

27: \begin{center}

28: \fbox{\parbox{0.95\textwidth}{%

29: \begin{problem}

30: #1

31: \end{problem}}}

32: \end{center}}

33: 

34: \usepackage[normalem]{ulem}

35: 

36: \usepackage{arydshln}

37: 

38: \begin{document}

39: 

40: \title{Permutation Polynomials modulo $m$}

41: \author{Shujun Li\\\url{http://www.hooklee.com}}

42: \date{\today}

43: 

44: \maketitle

45: 

46: \begin{abstract}

47: This paper mainly studies problems about so called ``permutation

48: polynomials modulo $m$", polynomials with integer

49: coefficients\footnote{In this paper, we always call them ``integer

50: polynomial" in short. Note that sometime another name ``integral

51: polynomial" is used \cite[Sec. 7.2]{HW:NumberTheory1979}. However,

52: we prefer to ``integer polynomial" to avoid confusion with the word

53: ``integral" as an adjective (see \cite{IntegerPolynomial}).} that

54: can induce bijections over $\mathbb{Z}_m=\{0,\cdots,m-1\}$. The

55: necessary and sufficient conditions of permutation polynomials are

56: given, and the number of all permutation polynomials of given degree

57: and the number induced bijections are estimated. A method is

58: proposed to determine all equivalent polynomials from the induced

59: polynomial function, which can be used to determine all equivalent

60: polynomials that induce a given bijection. A few problems have not

61: been solved yet in this paper and left for open study.

62: 

63: \textit{Note: After finishing the first draft, we noticed that some

64: results obtained in this paper can be proved in other ways (see

65: Remark \ref{remark:RepeatingWork}). In this case, this work gives

66: different and independent proofs of related results.}

67: \end{abstract}

68: 

69: \tableofcontents

70: 

71: \setlength{\arrayrulewidth}{0.6pt} \setlength{\doublerulesep}{0pt}

72: \renewcommand\arraystretch{1.25}

73: 

74: \section{Introduction}

75: 

76: Integer polynomials that can induce bijections over finite fields,

77: namely permutation polynomials, are firstly studied in algebra

78: community \cite{LauschNobauer:AlgebraPoly1973,

79: LidlMullen:PermPolyI:AMM1988, LidlMullen:PermPolyII:AMM1990,

80: RMT:DicksonPoly1993, Mullen:PermPolySurvey:FFaTA1995, SunWan:FF1987,

81: Lidl:FF1997}. Permutation polynomials have been used in cryptography

82: and coding \cite{PPinRSA-Crypt83, PP-Ciphers-EuroCrypt84,

83: MI-PP-PKC-IEEEISIT83, Cryptanalysis-PP-PKC-EuroCrypt84,

84: Cade-PP-PKC-SIAM-ALA85, BreakingCade-Crypto86,

85: ModifyingCade-Crypto86, SunWan:FF1987, Mullen:PP-NFSR:IEEETIT1989,

86: PP2Coding2005}. This paper studies permutation polynomials modulo an

87: integer, i.e., permutation polynomials over integer rings

88: \cite{Mullen:PolyFun-mod:AMH1984, RMT:DicksonPoly1993,

89: Rivest:PPmod2w:FFTA2001, Sun:PPmodCoding:IEEETIT2005}. Such

90: permutation polynomials have also been used in cryptography and

91: coding recently, such as in the RC6 block cipher \cite{RC6} a simple

92: permutation polynomial $f(x)=x(2x+1)$ modulo $2^d$ is used.

93: 

94: Assume $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial with integer

95: coefficients of degree $n\geq 1$ modulo $m$, where $a_n\not\equiv

96: 0\pmod m$. It is possible that $f(x)$ forms a bijection over

97: $\mathbb{Z}_m=\{0,\cdots,m-1\}$, i.e., $\forall

98: x_1,x_2\in\mathbb{Z}$ and $x_1\not\equiv x_2\pmod m$,

99: $f(x_1)\not\equiv f(x_2)\pmod m$. In other words, it is true that

100: $f(\mathbb{Z})=\mathbb{Z}_m$, or a complete system of residues

101: modulo $m$ is permuted by the polynomial $f$. The most common

102: permutation polynomial modulo $m$ is $f(x)=x$. In addition, as a

103: special case, Fermat's little theorem also gives one of the simplest

104: permutation polynomials modulo a prime $p$: $f(x)=x^p$, which

105: satisfies $\forall x\in\mathbb{Z}$, $f(x)\equiv x\pmod p$.

106: 

107: \iffalse In fact, permutation polynomials modulo $m=2^7$ have been

108: used in \cite{Chuang:BaseSwitchingImageEncryption:JEI97} to design

109: an image encryption scheme, though which is not secure following

110: results obtained in this paper.\fi

111: 

112: It is well-known that many problems on permutation polynomials over

113: finite fields are still open \cite{LidlMullen:PermPolyI:AMM1988,

114: LidlMullen:PermPolyII:AMM1990, Mullen:PermPolySurvey:FFaTA1995}.

115: Similarly, there are a few work on permutation polynomials modulo

116: integers, both in number theory \cite{HW:NumberTheory1979,

117: Eynden:NumberTheory1987, NZM:NumberTheory1991,

118: Rosen:NumberTheory1993, Rose:NumberTheory1994,

119: Pan:ConciseNumberTheory1998} and algebra communities. In this paper,

120: we try to find answers to the following questions on permutation

121: polynomials modulo $m$, where $m$ may be a prime, a prime power, or

122: a general composite.

123: \begin{itemize}

124: \item

125: What are necessary and sufficient conditions of permutation

126: polynomials modulo $m$?

127: 

128: \item

129: What is the number of distinct permutation polynomials of degree

130: $\leq n$ modulo $m$ and what is the number of distinct bijections

131: induced from these polynomials?

132: 

133: \item

134: Is there a practical way to enumerate all permutation polynomials of

135: degree $n$ modulo $m$ given a bijection (or a partial bijection)

136: over $\{0,\cdots,m-1\}$?

137: \end{itemize}

138: 

139: At present, the first two questions have been almost solved, but the

140: last one has not been solved at all when the degree of the

141: polynomials $n\geq p$ modulo $p^d$.

142: 

143: This paper is organized as follows. In next section, we first give

144: some preliminary definitions and lemma. Sec. \ref{section:composite}

145: discusses the case of composite moduli, and concludes that

146: permutation polynomials modulo a composite can be studied via

147: permutation polynomials modulo each prime power. In Sec.

148: \ref{section:prime}, we discuss the case of prime moduli and give

149: some limited results. Then, in Sec. \ref{section:PrimePowers}, the

150: general cases modulo $p^d$ ($d\geq 1$) are studied and some useful

151: results are obtained. Two open problems are raised in Sec.

152: \ref{section:PrimePowers} for future study.

153: 

154: \section{Preliminaries}

155: \label{section:preliminaries}

156: 

157: This section lists a number of definitions and notations used

158: throughout in this paper. Some preliminary lemmas are also given to

159: simplify the discussions in this paper. I try to keep the

160: definitions, notations and lemmas as simple as possible. Please feel

161: free to contact me if you have some idea of making them even

162: simpler, more elegant, more beautiful, and/or more rigorous in

163: mathematics.

164: 

165: \subsection{Some Simple Lemmas on Congruences}

166: 

167: The following lemmas will be extensively cited in this paper without

168: explicit citations.

169: 

170: \begin{lemma}\label{lemma:dvi-mod}

171: If $a\mid b$ and $a\mid m$, then $a\mid (b\bmod m)$.

172: \end{lemma}

173: \begin{proof}

174: From $a|b$, $\exists k_1\in\mathbb{Z}$, $b=ak_1$. From $a|m$,

175: $\exists k_2\in\mathbb{Z}$, $m=ak_2$. Assume $x=(b\bmod m)$, then

176: $\exists k_3\in\mathbb{Z}$, $x=mk_3+b=ak_2k_3+ak_1=a(k_1+k_2k_3)$.

177: So, $a\mid x$, which proves this lemma.

178: \end{proof}

179: 

180: \begin{lemma}\label{lemma:mod}

181: If $m\mid m'$, $(a\bmod m')\equiv a \pmod m$.

182: \end{lemma}

183: \begin{proof}

184: From $m|m'$, $\exists k_1\in\mathbb{Z}$, $m'=mk_1$. Assume $x=

185: a\bmod m'$, then $\exists k_2\in\mathbb{Z}$, $x=m'k_2+a=mk_1k_2+a$.

186: Then, $(a\bmod m')\equiv mk_1k_2+a\equiv a\pmod m$. This lemma is

187: proved.

188: \end{proof}

189: 

190: \begin{lemma}

191: If $a\equiv 0\pmod{m_1}$ and $b\equiv 0\pmod{m_2}$, then $ab\equiv

192: 0\pmod{m_1m_2}$.

193: \end{lemma}

194: 

195: \begin{lemma}[Theorem 2.2 in \cite{NZM:NumberTheory1991}]\label{lemma:polynomial-mod}

196: Assume $f(x)=a_nx^n+\cdots+a_1x+a_0$ is an integer polynomial. If

197: $x_1\equiv x_2\pmod m$, then $f(x_1)\equiv f(x_2)\pmod m$.

198: \end{lemma}

199: 

200: \begin{lemma}

201: Assume $\bm{A}$ is an $n\times n$ matrix, $\bm{X}$ is a vector of

202: $n$ unknown integers, and $\bm{B}$ is a vector of $n$ integers. If

203: $|\bm{A}|$ is relatively prime to $m$, i.e., $\gcd(|\bm{A}|,m)=1$,

204: then $\bm{A}\bm{X}\equiv\bm{B}\pmod m$ has a unique set of

205: incongruent solutions

206: $\bm{X}\equiv\overline{\Delta}(\mathrm{adj}(\bm{A}))\bm{B}\pmod m$,

207: where $\overline{\Delta}$ is an inverse of $\Delta=|\bm{A}|$ modulo

208: $m$ and $\mathrm{adj}(\bm{A})$ is the adjoint of $\bm{A}$.

209: \end{lemma}

210: \begin{proof}

211: This lemma is a direct result of Theorem 3.18 in

212: \cite{Rosen:NumberTheory1993} (see pages 151 and 152).

213: \end{proof}

214: 

215: \subsection[Polynomial Congruences Modulo $m$]%

216: {Polynomial Congruences Modulo $\bm{m}$}

217: 

218: The following definition is from Chap. VII of

219: \cite{HW:NumberTheory1979} and related concepts are slightly

220: extended.

221: 

222: \begin{definition}

223: Given two integer polynomials of degree $n$:

224: $f(x)=a_nx^n+\cdots+a_1x+a_0$ and $g(x)=b_nx^n+\cdots+b_1x+b_0$, if

225: $\forall i=0\sim n$, $a_i\equiv b_i\pmod m$, we say \uline{$f(x)$ is

226: congruent to $g(x)$ modulo $m$}, or $f(x)$ and $g(x)$ are

227: \uline{congruent (polynomials) modulo $m$}, which is denoted by

228: $f(x)\equiv g(x)\pmod m$. On the other hand, if $\exists

229: i\in\{1,\cdots,n\}$, such that $a_i\not\equiv b_i\pmod m$, we say

230: $f(x)$ and $g(x)$ are \uline{incongruent (polynomials) modulo $m$},

231: denoted by $f(x)\not\equiv g(x)\pmod m$.

232: \end{definition}

233: 

234: \begin{definition}

235: A \uline{polynomial congruence (residue) class modulo $m$} is a set

236: of all polynomials congruent to each other modulo $m$.

237: \end{definition}

238: 

239: \begin{definition}

240: A set of polynomials of degree $n$ modulo $m$ is a \uline{complete

241: system of polynomial residues of degree $n$ modulo $m$}, if for

242: every polynomial of degree $n$ modulo $m$ there is one and only one

243: congruent polynomial in this set.

244: \end{definition}

245: 

246: \begin{lemma}

247: The following set of polynomials is a complete system of polynomial

248: residues of degree $n$ modulo $m$:

249: \[

250: \mathbb{F}[x]=\left\{f(x)=a_nx^n+\cdots+a_1x+a_0\left|a_n\in\{1,\cdots,m-1\},a_{n-1},\cdots,a_0\in\{0,\cdots,m-1\}\right.\right\}.

251: \]

252: \end{lemma}

253: \begin{proof}

254: Assume $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial of degree $n$

255: modulo $m$. Choose $a_i^*=(a_i\bmod m)\in\{0,\cdots,m-1\}$ ($i=0\sim

256: n$), then $f^*(x)=a_n^*x^n+\cdots+a_1^*x+a_0^*\in\mathbb{F}$ is

257: congruent to $f(x)$. Assume that another polynomial

258: $g(x)=b_nx^n+\cdots+b_1x+b_0\in\mathbb{F}$ is also congruent to

259: $f(x)$. Then, $\forall i=0\sim n$, $b_i\equiv a_i^*\pmod m$. Since

260: $\{0,\cdots,m-1\}$ is a complete set of residues modulo $m$,

261: $b_i=a_i^*$. This means that $g(x)=f^*(x)$. This completes the proof

262: of this lemma.

263: \end{proof}

264: 

265: \begin{definition}

266: A set of polynomials of degree $\leq n$ modulo $m$ is a

267: \uline{complete system of polynomial residues of degree $\leq n$

268: modulo $m$}, if for every polynomial of degree $\leq n$ modulo $m$

269: there is one and only one congruence polynomial.

270: \end{definition}

271: 

272: \begin{lemma}

273: The following set of polynomials is a complete system of polynomial

274: residues of degree $n$ modulo $m$:

275: \[

276: \mathbb{F}[x]=\left\{f(x)=a_nx^n+\cdots+a_1x+a_0\left|a_n,a_{n-1},\cdots,a_0\in\{0,\cdots,m-1\}\right.\right\}.

277: \]

278: \end{lemma}

279: \begin{proof}

280: The proof is similar to the above lemma.

281: \end{proof}

282: 

283: \subsection[Polynomial Functions Modulo $m$]{Polynomial Functions

284: Modulo $\bm{m}$}

285: 

286: \begin{definition}

287: If a function over $\{0,\cdots,m-1\}$ can be represented by a

288: polynomial modulo $m$, we say this function is \uline{polynomial

289: modulo $m$}.

290: \end{definition}

291: 

292: \begin{lemma}\label{lemma:any-fun-poly-p}

293: Assume $p$ is a prime. Then, any function over $\{0,\cdots,p-1\}$ is

294: polynomial modulo $p$.

295: \end{lemma}

296: \begin{proof}

297: Assume $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial of degree

298: $n\geq p-1$ modulo $p$. Given a function

299: $F:\{0,\cdots,p-1\}\to\{0,\cdots,p-1\}$, one has the following

300: system of congruences:

301: \[

302: \left[\begin{matrix}%

303: 1 & 0 & 0^2 & \cdots & 0^{p-1}\\

304: 1 & 1 & 1^2 & \cdots & 1^{p-1}\\

305: 1 & 2 & 2^2 & \cdots & 2^{p-1}\\

306: \vdots & \vdots & \vdots & \ddots & \vdots\\

307: 1 & p-1 & (p-1)^2 & \cdots & (p-1)^{p-1}

308: \end{matrix}\right]

309: \left[\begin{matrix}%

310: a_0\\

311: a_1\\

312: a_2\\

313: \vdots\\

314: a_{p-1}

315: \end{matrix}\right]\equiv

316: \left[\begin{matrix}%

317: F(0)\\

318: F(1)-\sum_{i=p}^na_i\\

319: F(2)-\sum_{i=p}^n2^ia_i\\

320: \vdots\\

321: F(p-1)-\sum_{i=p}^n(p-1)^ia_i

322: \end{matrix}\right]\pmod{p^d}.

323: \]

324: Since the matrix at the left side is a Vondermonde matrix, one can

325: see its determinant is relatively prime to $p$. So, for each

326: combination of $a_p,\cdots,a_n$, there is a unique set of

327: incongruent solutions of $a_0,\cdots,a_{p-1}$. Thus this lemma is

328: proved.

329: \end{proof}

330: 

331: \subsection[Equivalent Polynomials Modulo $m$]{Equivalent Polynomials Modulo $\bm{m}$}

332: 

333: The concept of equivalent polynomial modulo $m$ is used to describe

334: incongruent but equivalent (for any integer) polynomials modulo $m$.

335: Note that some researchers call them ``residually congruent

336: polynomials modulo $m$" \cite{Kempner:PolyResidue:TAMS1921a,

337: Kempner:PolyResidue:TAMS1921b}.

338: 

339: \begin{definition}

340: Two integer polynomials $f(x)$ and $g(x)$ are \uline{equivalent

341: (polynomials) modulo $m$} if $\forall x\in\mathbb{Z}$, $f(x)\equiv

342: g(x)\pmod m$.

343: \end{definition}

344: Note that two equivalent polynomials modulo $m$ may not be congruent

345: modulo $p$, and may have distinct degrees. As a typical example,

346: when $p$ is a prime, $f(x)=x^p$ and $g(x)=x$ are equivalent

347: polynomials modulo $p$.

348: 

349: \begin{lemma}\label{lemma:equivalent-poly-degree1}

350: Two polynomials of degree 1 modulo $m$, $f(x)=a_1x+a_0$ and

351: $g(x)=b_1x+b_0$, are equivalent polynomials modulo $m$ if and only

352: if $f(x)\equiv g(x)\pmod m$, i.e., $a_1\equiv b_1\pmod m$ and

353: $a_0\equiv b_0\pmod m$.

354: \end{lemma}

355: \begin{proof}

356: The ``if" part is obvious from the definition of equivalent

357: polynomials modulo $m$, so we focus on the ``only if" part. Since

358: $f(x)$ and $g(x)$ are equivalent polynomials modulo $m$, then

359: $\forall x\in\{0,\cdots,m-1\}$,

360: $f(x)-g(x)=(a_1-b_1)x+(a_0-b_0)\equiv 0\pmod m$. Choosing $x\equiv

361: 0\pmod m$, one has $a_0\equiv b_0\pmod m$. Then, choosing $x\equiv

362: 1\pmod m$, one has $a_1\equiv b_1\pmod m$. Thus this lemma is

363: proved.

364: \end{proof}

365: 

366: \begin{lemma}\label{lemma:equivalent-poly-a0}

367: Two polynomials, $f(x)=a_{n_1}x^{n_1}+\cdots+a_0$ and

368: $g(x)=b_{n_2}x^{n_2}+\cdots+b_0$, are equivalent polynomials modulo

369: $m$, then $a_0\equiv b_0\pmod m$.

370: \end{lemma}

371: \begin{proof}

372: Choosing $x=0$, one has $f(x)-g(x)=a_0-b_0\equiv 0\pmod m$. This

373: lemma is proved.

374: \end{proof}

375: \begin{corollary}

376: Two polynomials, $f(x)=a_nx^n+\cdots+a_2x^2+a_0$ and

377: $g(x)=a_nx^n+\cdots+a_2x^2+b_0$, are equivalent polynomials modulo

378: $m$ if and only if $a_0\equiv b_0\pmod m$.

379: \end{corollary}

380: 

381: \begin{lemma}\label{lemma:equivalent-poly-pd}

382: Assume $p$ is a prime and $d\geq 1$. Two polynomials,

383: $f(x)=a_{p-1}x^{p-1}+\cdots+a_0$ and

384: $g(x)=b_{p-1}x^{p-1}+\cdots+b_0$, are equivalent polynomials modulo

385: $p^d$ if and only if $f(x)\equiv g(x)\pmod{p^d}$.

386: \end{lemma}

387: \begin{proof}

388: The ``if" part is obvious true, from the definition of equivalent

389: polynomials modulo $p^d$. So, we focus on the ``only if" part only.

390: From $f(x)-g(x)\equiv 0\pmod{p^d}$, choosing $x=0\sim p-1$, one can

391: get the following system of congruences in the matrix form

392: $\bm{A}\bm{X}_{a-b}\equiv \bm{B}\pmod{p^d}$:

393: \begin{equation}

394: \left[\begin{matrix}%

395: 1 & 0 & 0^2 & \cdots & 0^{p-1}\\

396: 1 & 1 & 1^2 & \cdots & 1^{p-1}\\

397: 1 & 2 & 2^2 & \cdots & 2^{p-1}\\

398: \vdots & \vdots & \vdots & \ddots & \vdots\\

399: 1 & p-1 & (p-1)^2 & \cdots & (p-1)^{p-1}

400: \end{matrix}\right]

401: \left[\begin{matrix}%

402: a_0-b_0\\

403: a_1-b_1\\

404: a_2-b_2\\

405: \vdots\\

406: a_{p-1}-b_{p-1}

407: \end{matrix}\right]\equiv

408: \left[\begin{matrix}%

409: f(0)-g(0)\\

410: f(1)-g(1)\\

411: f(2)-g(2)\\

412: \vdots\\

413: f(p-1)-g(p-1)

414: \end{matrix}\right]\equiv

415: \left[\begin{matrix}%

416: 0\\

417: 0\\

418: 0\\

419: \vdots\\

420: 0

421: \end{matrix}\right]\pmod{p^d}.\label{equation:equivalent-poly-pd}

422: \end{equation}

423: Since $\bm{A}$ is a Vandermonde sub-matrix, one can get

424: $|\bm{A}|=\prod_{0\leq i<j\leq p-1}(j-i)$

425: \cite[\S4.4]{Zhang:MatrixTheory1999}. From $p$ is a prime and

426: $1\leq(j-i)\leq p-1$, one has $\gcd(|\bm{A}|,p^d)=1$. Thus, the

427: above system of congruences has a unique set of incongruent

428: solutions. So, $\forall i=0\sim p-1$, one has $a_i\equiv

429: b_i\pmod{p^d}$. This completes the proof of this lemma.

430: \end{proof}

431: Note that in the above lemma $f(x)$ and $g(x)$ may be polynomials of

432: degree less than $p-1$ modulo $p^d$. In this case, the matrix at the

433: left side of the system of congruences may have a smaller size, but

434: its determinant is still relatively prime to $p^d$.

435: 

436: \begin{corollary}

437: Assume $p$ is a prime. Two polynomials, $f(x)=a_nx^n+\cdots+a_0$ and

438: $g(x)=b_nx^n+\cdots+b_0$, are equivalent polynomials modulo $p$ if

439: and only if $(f(x)\bmod(x^p-x))\equiv(g(x)\bmod(x^p-x))\pmod p$.

440: \end{corollary}

441: \begin{proof}

442: This corollary is a direct result of the above lemma and Fermat's

443: Little Theorem.

444: \end{proof}

445: 

446: \subsection[Permutation Polynomials Modulo $m$]

447: {Permutation Polynomials Modulo $\bm{m}$}

448: 

449: \begin{definition}

450: Assume $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial of degree

451: $n\geq 1$ modulo $m$, where $a_n\not\equiv 0\pmod m$. If

452: $f(x)=((a_nx^n+\cdots+a_1x+a_0)\bmod m)$ forms a bijection

453: $F:\{0,\cdots,m-1\}\to\{0,\cdots,m-1\}$, we say that $f(x)$ is a

454: \uline{permutation polynomial modulo $m$}, or $f(x)$ is

455: \uline{permutation modulo $m$}. The bijection $F$ is called the

456: \uline{induced bijection of the polynomial $f(x)$ modulo $m$}.

457: \end{definition}

458: 

459: \begin{definition}

460: If two permutation polynomials are equivalent modulo $m$, we say

461: they are \uline{equivalent permutation polynomials modulo $m$}. It

462: is obvious that equivalent permutation polynomials modulo $m$ induce

463: the same bijection over $\{0,\cdots,m-1\}$.

464: \end{definition}

465: 

466: \begin{lemma}\label{lemma:bijective}

467: A polynomial $f(x)$ is a permutation polynomial modulo $m$ if and

468: only if $g(x)=af(x)+b$ is a permutation polynomial modulo $m$, where

469: $\gcd(a,m)=1$ and $b\in\mathbb{Z}$.

470: \end{lemma}

471: \begin{proof}

472: This lemma is a direst result of Theorem 3.6 in

473: \cite{Rosen:NumberTheory1993} on a complete system of residues

474: modulo $m$.

475: \end{proof}

476: \begin{lemma}\label{lemma:bijective-poly-a0}

477: Two polynomials, $f(x)=a_{n_1}x^{n_1}+\cdots+a_1x+a_0$ and

478: $f(x)=b_{n_2}x^{n_2}+\cdots+b_1x+b_0$, is equivalent permutation

479: polynomials if and only if $f^*(x)=a_{n_1}x^{n_1}+\cdots+a_1x$ and

480: $g^*(x)=b_{n_2}x^{n_2}+\cdots+b_1x$ are equivalent permutation

481: polynomials modulo $m$ and $a_0\equiv b_0\pmod m$.

482: \end{lemma}

483: \begin{proof}

484: This lemma is a direct result of Lemmas

485: \ref{lemma:equivalent-poly-a0} and \ref{lemma:bijective}.

486: \end{proof}

487: From the above two lemmas, we can only study permutation polynomials

488: in the form $f(x)=a_nx^n+\cdots+a_1x$.

489: 

490: \begin{lemma}

491: If $f(x)\equiv g(x)\pmod m$ and $f(x)$ is a permutation polynomial

492: modulo $m$, then $g(x)$ is an equivalent permutation polynomial of

493: $f(x)$ modulo $m$.

494: \end{lemma}

495: \begin{proof}

496: Since $f(x)\equiv g(x)\pmod m$, $\forall a\in\mathbb{Z}$,

497: $f(a)\equiv g(a)\pmod m\Rightarrow f(a)\bmod m=g(a)\bmod m$, i.e.,

498: $F(a)=G(a)$. So $g(x)$ generates the same bijection as $f(x)$. This

499: completes the proof.

500: \end{proof}

501: 

502: \begin{theorem}\label{theorem:bijection-degree1}

503: The polynomial $f(x)=a_1x+a_0$ is a permutation polynomial modulo

504: $m$ if and only if $\gcd(a_1,m)=1$.

505: \end{theorem}

506: \begin{proof}

507: Assume $G=\langle 1\rangle$ is a cyclic group of order $m$. From

508: Theorem 3.24 in \cite{Gilbert:Algebra2005} (or Theorem 2 in \S 2.3

509: of \cite{Hu:Algebra1999}), $a_1=1^{a_1}$ is a generator of $G$ if

510: and only if $\gcd(a_1,m)=1$. Note that in group $G$ the binary

511: operator is defined as addition modulo $m$. It is obvious that $a_1$

512: is a generator of $G$ if and only if $f^*(x)=a_1x$ is a permutation

513: polynomial modulo $m$. Then from Lemma \ref{lemma:bijective}, this

514: theorem is proved.

515: \end{proof}

516: 

517: \begin{corollary}\label{corollary:bijective-poly-degree1}

518: The number of congruence classes of permutation polynomials

519: $f(x)=a_1x+a_0$ of degree 1 modulo $m$ is $\phi(m)m$. The number of

520: bijections induced from these permutation polynomials is also

521: $\phi(m)m$.

522: \end{corollary}

523: \begin{proof}

524: From Theorem \ref{theorem:bijection-degree1}, $a_1$ should satisfy

525: $\gcd(a_1,m)=1$, but $a_0$ can be any integer, so The number of

526: congruence classes of permutation polynomials of degree 1 modulo $m$

527: is $\phi(m)m$. From Lemma \ref{lemma:equivalent-poly-degree1}, the

528: $\phi(m)m$ permutation polynomials are not equivalent to each other,

529: so they induce $\phi(m)m$ distinct bijections.

530: \end{proof}

531: 

532: %\subsection{Sub-Bijection and Super-Bijection (New)}

533: 

534: \begin{definition}

535: Given a bijection $F:\mathbb{A}\to\mathbb{A}$. If for a set

536: $\mathbb{B}\subseteq\mathbb{A}$, $F(\mathbb{B})=\mathbb{B}$, then

537: $F_\mathbb{B}=\{(a,b)|a,b\in\mathbb{B}\}\subseteq F$ is a bijection

538: over $\mathbb{B}$, and we say the bijection

539: $F_\mathbb{B}:\mathbb{B}\to\mathbb{B}$ is a \uline{sub-bijection} of

540: $F$, and $F$ is a \uline{super-bijection} of $F_\mathbb{B}$.

541: \end{definition}

542: 

543: \begin{lemma}\label{lemma:sub-bijection}

544: Given a bijection $F:\mathbb{A}\to\mathbb{A}$. If

545: $F_\mathbb{B}:\mathbb{B}\to\mathbb{B}$ is a sub-bijection of $F$,

546: then

547: $F_{\mathbb{A}\backslash\mathbb{B}}:\mathbb{A}\backslash\mathbb{B}\to\mathbb{A}\backslash\mathbb{B}$

548: is also a sub-bijection of $F$.

549: \end{lemma}

550: \begin{proof}

551: Assume that $\exists a\in\mathbb{A}\backslash\mathbb{B}$,

552: $F(a)\in\mathbb{B}$. Since $F_\mathbb{B}$ is a bijection over

553: $\mathbb{B}$, then $F(a)$ has one and only one preimage in

554: $\mathbb{B}$. However, it is obvious that $a\not\in\mathbb{B}$ is

555: also the preimage of $F(a)$. We get a contradiction. So, $\forall

556: x\in\mathbb{A}\backslash\mathbb{B}$,

557: $F(x)\in\mathbb{A}\backslash\mathbb{B}$. This means that

558: $F_{\mathbb{A}\backslash\mathbb{B}}$ is a sub-bijection of $F$ over

559: $\mathbb{A}\backslash\mathbb{B}$.

560: \end{proof}

561: 

562: \subsection[Null Polynomials modulo $m$]{Null Polynomials modulo $\bm{m}$}

563: 

564: This concept was introduced in \cite{Li:NullPoly2005}, and also

565: studied by others without a special name

566: \cite{Kempner:PolyResidue:TAMS1921a, Kempner:PolyResidue:TAMS1921b}.

567: Here, we just give the definition and some simple lemmas on null

568: polynomials modulo $m$. For more advanced results, see

569: \cite{Li:NullPoly2005}.

570: 

571: \begin{definition}

572: A polynomial $f(x)$ of degree $n\geq 0$ modulo $m$ is a \uline{null

573: polynomial of degree $n$ modulo $m$}, if $\forall x\in\mathbb{Z}$,

574: $f(x)\equiv 0\pmod m$. Specially, $f(x)=0$ is a trivial null

575: polynomial of degree 0 modulo $m$.

576: \end{definition}

577: 

578: \begin{lemma}

579: If $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a null polynomial modulo $m$,

580: then $a_0\equiv 0\pmod m$.

581: \end{lemma}

582: 

583: \begin{lemma}

584: Given any null polynomial $f(x)$ modulo $m$, $af(x)$ will still be a

585: null polynomial modulo $m$, where $a$ is an arbitrary integer.

586: \end{lemma}

587: \begin{lemma}

588: A polynomial $f(x)$ is a null polynomial modulo $m$, if and only

589: $af(x)$ is a null polynomial modulo $m$, where $\gcd(a,m)=1$.

590: \end{lemma}

591: \begin{lemma}\label{lemma:null-poly-transitivity}

592: If $f(x)$ is a null polynomial modulo $m$ and $a\mid m$, then $f(x)$

593: is still a null polynomial modulo $a$.

594: \end{lemma}

595: The most frequently used form of the above lemma is as follows: if

596: $f(x)$ is a null polynomial modulo $p^d$, then $f(x)$ is still a

597: null polynomial modulo $p^i$ for any integer $i\leq d$.

598: 

599: \begin{lemma}\label{lemma:Equ-Poly-Null-Poly}

600: Two polynomials, $f_1(x)$ and $f_2(x)$, are equivalent polynomials

601: modulo $m$ if and only if $f_1(x)-f_2(x)$ is a null polynomial

602: modulo $m$.

603: \end{lemma}

604: 

605: \begin{definition}

606: Denote the least integer $n\geq 1$ such that there exists a null

607: polynomial of degree $n$ modulo $m$ by $\omega_0(m)$ and call it

608: \uline{the least null-polynomial degree modulo $m$}. Denote the

609: least integer $n\geq 1$ such that there exists a \textbf{monic} null

610: polynomial of degree $n$ modulo $m$ by $\omega_1(m)$ and call it

611: \uline{the least monic null-polynomial degree modulo $m$}. A (monic)

612: null polynomial of degree $\omega_0(m)$ or $\omega_1(m)$ is called

613: \uline{a least-degree (monic) null polynomial modulo $m$}. \iffalse

614: Sometime we use $\omega(m)$ to denote $\omega_1(m)$, since

615: $\omega_1(m)$ is much more important than $\omega_0(m)$.\fi

616: \end{definition}

617: 

618: \begin{lemma}

619: Every polynomial of degree $\geq\omega_1(m)$ modulo $m$ has one

620: equivalent polynomial of degree $\leq\omega_1(m)-1$ modulo $m$.

621: \end{lemma}

622: 

623: \subsection[Circular Shift of an Integer Set Modulo $m$ (New)]%

624: {Circular Shift of an Integer Set Modulo $\bm{m}$ (New)}

625: 

626: \begin{definition}

627: The \uline{$k$-th circular shift} of an integer set $\mathbb{A}$

628: \uline{modulo $m$} is defined by $\{x|x=(y+k)\bmod

629: m,y\in\mathbb{A}\}$ and denoted by $[(\mathbb{A}+k)\bmod m]$ in this

630: paper.

631: \end{definition}

632: 

633: \begin{lemma}\label{lemma:circular-shift}

634: If $\mathbb{A}_1,\cdots,\mathbb{A}_k$ is a partition of an integer

635: set $\mathbb{A}=\{0,\cdots,m-1\}$, then $\forall a\in\mathbb{Z}$,

636: $[(\mathbb{A}_1+a)\bmod m],\cdots,[(\mathbb{A}_k+a)\bmod m]$ is

637: still a partition of $\mathbb{A}$.

638: \end{lemma}

639: \begin{proof}

640: From $\mathbb{A}_1,\cdots,\mathbb{A}_k$ is a partition of

641: $\mathbb{A}$, one has

642: $\bigcup_{i=1}^k\mathbb{A}_i=\{0,\cdots,m-1\}$. Then,

643: $\bigcup_{i=1}^k[(\mathbb{A}_i+a)\bmod

644: m]=\bigcup_{i=1}^k\{x|x=(y+a)\bmod

645: m,y\in\mathbb{A}_i\}=\left\{x|x=(y+a)\bmod

646: m,y\in\bigcup_{i=1}^k\mathbb{A}_i\right\}=\{x|x=(y+a)\bmod

647: m,y\in\mathbb{A}\}$. From $\mathbb{A}$ is a complete system of

648: residues modulo $m$, $\forall x_0\in\mathbb{A}$ and $\forall

649: a\in\mathbb{Z}$, $\exists y_0\in\mathbb{A}$, $x_0-a\equiv y_0\pmod

650: m\Leftrightarrow x_0\equiv y_0+a\pmod m$. Since $x_0\in\mathbb{A}$,

651: i.e., $0\leq x_0\leq m-1$, one has $x_0=(y_0+a)\bmod m$. This means

652: $\forall x_0\in\bigcup_{i=1}^k[(\mathbb{A}_i+a)\bmod

653: m]=\{x|x=(y+a)\bmod m,y\in\mathbb{A}\}$. So

654: $\bigcup_{i=1}^k[(\mathbb{A}_i+a)\bmod m]=\mathbb{A}$.

655: 

656: On the other hand, since $\mathbb{A}_1,\cdots,\mathbb{A}_k$ is a

657: partition of $\mathbb{A}$, $\forall y_1\in\mathbb{A}_i$ and $\forall

658: y_2\in\mathbb{A}_j$ ($i\neq j$), one has $y_1\neq y_2$. Considering

659: $\mathbb{A}$ is a complete system of residues modulo $m$, one

660: immediately gets $y_1\not\equiv y_2\pmod m

661: \Leftrightarrow(y_1+a)\not\equiv (y_2+a)\pmod m \Leftrightarrow

662: (y_1+a)\bmod m\neq (y_2+a)\bmod m$. This means

663: $\mathbb{A}_i\cap\mathbb{A}_j=\varnothing$.

664: 

665: The above two results proves this lemma.

666: \end{proof}

667: 

668: \subsection[Base-$p$ Resolution (New)]{Base-$\bm{p}$ Resolution (New)}

669: 

670: \begin{definition}

671: The \uline{base-$p$ resolution} of an integer $a$ is an integer

672: $i\geq 0$ such that $p^i\parallel a$, i.e., $p^i\mid a$ but

673: $p^{i+1}\nmid a$. Specially, define the base-$p$ resolution of 0 as

674: $+\infty$. When $p=2$, the base-$p$ resolution is also called the

675: \uline{binary resolution}.

676: \end{definition}

677: 

678: \begin{notation}

679: The set of all integers of base-$p$ resolution $i$ is denoted by

680: $\mathbb{Z}(i|_p)$. The set of all elements in $\mathbb{A}$ of

681: base-$p$ resolution $i$ is denoted by $\mathbb{A}(i|_p)$. The set of

682: all elements in $\mathbb{A}$ of base-$p$ resolution $i\geq a$ is

683: denoted by $\mathbb{A}(\geq a|_p)$; similarly, we can define

684: $\mathbb{A}(\leq a|_p)$ and $\mathbb{A}(\neq a|_p)$. The set of all

685: elements in $\mathbb{A}$ of base-$p$ resolution $a\leq i\leq b$ is

686: denoted by $\mathbb{A}(a\mapsto b|_p)$. The set of all elements in

687: $\mathbb{A}$ of base-$p$ resolution $i\in\{i_1,\cdots,i_k\}$ is

688: denoted by $\mathbb{A}(i_1,\cdots,i_k|_p)$. In the above notations,

689: the subscription ``$p$" denotes the base (radix) of the resolution.

690: \end{notation}

691: Apparently, $a\in\mathbb{Z}(\geq i|_p)\Leftrightarrow

692: p^i|a\Leftrightarrow a\equiv 0\pmod{p^i}$.

693: 

694: \begin{definition}

695: The \uline{base-$p$ multi-resolution partition} of the integer set

696: $\mathbb{Z}$ is a collection of the following sets:

697: $\{\mathbb{Z}(i|_p)\}_{i=0}^{+\infty}$. The \uline{base-$p$

698: multi-resolution partition} of an integer set $\mathbb{A}$ is a

699: collection of the following $(k_{\max}-k_{\min}+2)$ sets:

700: $\mathbb{A}(k_{\min}|_p),\cdots,\mathbb{A}(k_{\max}|_p)$ and

701: $\mathbb{A}(+\infty|_p)=\{0\}$, where $k_{\min}$ and $k_{\max}$

702: denote the minimal and the maximal base-$p$ resolution of all

703: non-zero integers in $\mathbb{A}$.

704: \end{definition}

705: Specially, the base-$p$ multi-resolution partition of

706: $\mathbb{A}=\{0,\cdots,p^d-1\}$ is a collection of the following

707: $d+1$ sets: $\mathbb{A}(0|_p),\cdots,\mathbb{A}(d-1|_p)$ and

708: $\mathbb{A}(+\infty|_p)=\{0\}$.

709: 

710: \iffalse The circular shifts of $\mathbb{Z}(i|_2)$ have some special

711: properties.

712: 

713: \begin{lemma}

714: Assume $m=2^d$ and $\mathbb{A}=\{0,\cdots,m-1\}$. If

715: $\mathbb{A}_1,\cdots,\mathbb{A}_k$ is a partition of

716: $\mathbb{A}(0|_2)$, then 1) $\forall a\in\mathbb{Z}(0|_2)$,

717: $[(\mathbb{A}_1+a)\bmod m],\cdots,[(\mathbb{A}_k+a)\bmod m]$ is a

718: partition of $\mathbb{A}(\geq 1|_2)$; 2) $\forall

719: a\in\mathbb{Z}(\geq 1|_2)$, $[(\mathbb{A}_1+a)\bmod

720: m],\cdots,[(\mathbb{A}_k+a)\bmod m]$ is still a partition of

721: $\mathbb{A}(0|_2)$.

722: \end{lemma}

723: \begin{lemma}

724: Assume $m=2^d$ and $\mathbb{A}=\{0,\cdots,m-1\}$. If

725: $\mathbb{A}_1,\cdots,\mathbb{A}_k$ is a partition of

726: $\mathbb{A}(\geq 1|_2)$, then 1) $\forall a\in\mathbb{Z}(0|_2)$,

727: $[(\mathbb{A}_1+a)\bmod m],\cdots,[(\mathbb{A}_k+a)\bmod m]$ is a

728: partition of $\mathbb{A}(0|_2)$; 2) $\forall a\in\mathbb{Z}(\geq

729: 1|_2)$, $[(\mathbb{A}_1+a)\bmod m],\cdots,[(\mathbb{A}_k+a)\bmod m]$

730: is still a partition of $\mathbb{A}(\geq 1|_2)$.

731: \end{lemma}

732: The proofs of the above two lemmas are simple, so they are omitted

733: here. A special case of the above two lemmas (when $k=1$) can be

734: described as the following lemma. The three lemmas will be

735: frequently cited in the following context without explicitly

736: quotations.

737: 

738: \begin{lemma}

739: Assume $m=2^d$ and $\mathbb{A}=\{0,\cdots,m-1\}$. When

740: $a\in\mathbb{Z}(0|_2)$, $[(\mathbb{A}(0|_2)+a)\bmod

741: m]=\mathbb{A}(\geq 1|_2)$ and $[(\mathbb{A}(\geq 1|_2)+a)\bmod

742: m]=\mathbb{A}(0|_2)$. When $a\in\mathbb{Z}(\geq 1|_2)$,

743: $[(\mathbb{A}(0|_2)+a)\bmod m]=\mathbb{A}(0|_2)$ and

744: $[(\mathbb{A}(\geq 1|_2)+a)\bmod m]=\mathbb{A}(\geq 1|_2)$.

745: \end{lemma}

746: \fi

747: 

748: \begin{definition}

749: Assume the base-$p$ resolution of an integer $a\neq 0$ is $i\geq 0$,

750: then the \uline{base-$p$ representation of $a$} is a sequence of $i$

751: integers $a_0,\cdots,a_i$, such that $a=\sum_{j=0}^ia_jp^j$ and

752: $a_j\in\{0,\cdots,p-1\}$. It is denoted by $a=(a_i\cdots a_0)_p$.

753: Specially, the base-$p$ representation of 0 is $(0)_p$. The $j$-th

754: integer in the base-$p$ representation of $a$ is called \uline{the

755: $j$-th base-$p$ digit} or \uline{the $j$-th digit of base $p$} or

756: \uline{the $j$-th digit} in short if the base is well defined in the

757: context.

758: \end{definition}

759: It is obvious that the base-$p$ resolution of an integer is unique

760: and $a_j=\lfloor a/p^j\rfloor\bmod p$.

761: 

762: \subsection{Determinants of Some Special Matrices}

763: 

764: \begin{lemma}\label{lemma:det-powers}

765: Assume $m\geq 1$. Given a $2m\times 2m$ matrix

766: $\bm{A}=\left[\begin{matrix}\bm{A}_1\\\bm{A}_2\end{matrix}\right]$,

767: where $\bm{A}_1=[X_j^{i-1}]_{1\leq j\leq m \atop 1\leq i\leq 2m}$

768: and $\bm{A}_2=[iX_j^{i-1}]_{1\leq j\leq m \atop 1\leq i\leq 2m}$,

769: i.e.,

770: \[

771: \bm{A}=\left[\begin{array}{cccc:ccc}%

772: 1 & X_1 & \cdots & X_1^{m-1} & X_1^m & \cdots & X_1^{2m-1}\\

773: 1 & X_2 & \cdots & X_2^{m-1} & X_2^m & \cdots & X_2^{2m-1}\\

774: \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots\\

775: 1 & X_m & \cdots & X_m^{m-1} & X_m^m & \cdots & X_m^{2m-1}\\

776: \hdashline

777: 1 & 2X_1 & \cdots & mX_1^{m-1} & (m+1)X_1^m & \cdots & 2mX_1^{2m-1}\\

778: 1 & 2X_2 & \cdots & mX_2^{m-1} & (m+1)X_2^m &

779: \cdots & 2mX_2^{2m-1}\\

780: \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots\\

781: 1 & 2X_m & \cdots & mX_m^{m-1} & (m+1)X_m^m & \cdots & 2mX_m^{2m-1}

782: \end{array}\right].

783: \]

784: Then, $|\bm{A}|=(-1)^{\frac{m(m-1)}{2}}\prod_{j=1}^mX_j\prod_{1\leq

785: i<j\leq m}(X_j-X_i)^4$.

786: \end{lemma}

787: \begin{proof}

788: A proof can be found in \cite{Li:Determinants2005}, or in

789: \cite{ADC1999} (as a special case of Theorem 20).

790: \end{proof}

791: \begin{corollary}\label{corollary:det-powers}

792: Assume $m\geq 1$. Given a $2m\times 2m$ matrix

793: $\bm{A}=\left[\begin{matrix}\bm{A}_1\\\bm{A}_2\end{matrix}\right]$,

794: where $\bm{A}_1=[X_i^{j+1}]_{1\leq i\leq m \atop 1\leq j\leq 2m}$

795: and $\bm{A}_2=[(j+1)X_i^j]_{1\leq i\leq m \atop 1\leq j\leq 2m}$.

796: Then,

797: $|\bm{A}|=(-1)^{\frac{m(m-1)}{2}}\prod_{i=1}^mX_i^4\prod_{1\leq

798: i<j\leq m}(X_j-X_i)^4$.

799: \end{corollary}

800: 

801: \begin{lemma}\label{lemma:det-binom-powers}

802: Assume $m\geq 1,n\geq l\geq 1$ and $\bm{A}$ is a block-wise

803: $ml\times ml$ matrix as follows:

804: \[

805: \bm{A}=\left[\begin{matrix}%

806: \bm{A}_1\\

807: \bm{A}_2\\

808: \vdots\\

809: \bm{A}_m

810: \end{matrix}\right],

811: \]

812: where for $i=1\sim m$,

813: \[

814: \bm{A}_i=\left[\binom{n+j-1}{k-1}X_i^{j-1}\right]_{1\leq

815: j\leq ml \atop 1\leq k\leq l}=\left[\begin{matrix}%

816: \binom{n}{0} & \binom{n+1}{0}X_i & \cdots &

817: \binom{n+(ml-1)}{0}X_i^{ml-1}\\

818: \binom{n}{1} & \binom{n+1}{1}X_i & \cdots &

819: \binom{n+(ml-1)}{1}X_i^{ml-1}\\

820: \vdots & \vdots & \ddots & \vdots\\

821: \binom{n}{l-1} & \binom{n+1}{l-1}X_i & \cdots &

822: \binom{n+(ml-1)}{l-1}X_i^{ml-1}

823: \end{matrix}\right]_{ml\times l}.

824: \]

825: Then, $|\bm{A}|=\prod_{i=1}^mX_i^{\frac{l(l-1)}{2}}\prod_{1\leq

826: i<j\leq m}(X_j-X_i)^{l^2}$.

827: \end{lemma}

828: \iffalse

829: \begin{corollary}\label{corollary:det-binom}

830: When $0\leq m\leq n-1$, the determinant of the $(m+1)\times (m+1)$

831: matrix $\bm{A}_{n,m}=\left[\binom{j}{i}\right]_{n\leq j\leq n+m

832: \atop 0\leq i\leq m}$ is always equal to 1.

833: \end{corollary}\fi

834: 

835: \section{Permutation Polynomials modulo $\bm{m=p_1^{d_1}\cdots p_r^{d_r}}$}

836: \label{section:composite}

837: 

838: The theorems given in this section says that we can focus our study

839: on permutation polynomials modulo prime and prime powers.

840: 

841: \begin{theorem}\label{theorem:bijective-poly-composite}

842: Assume $p_1$, $\cdots$, $p_r$ are $r$ distinct prime numbers and

843: $d_1$, $\cdots$, $d_r\geq 1$. A polynomial $f(x)$ is a permutation

844: polynomial modulo $m=\prod_{i=1}^rp_i^{d_i}$, if and only if

845: $\forall i=1\sim r$, $f(x)$ is a permutation polynomial modulo

846: $p_i^{d_i}$.

847: \end{theorem}

848: \begin{proof}

849: To simplify the following proof, $\forall i=1\sim r$, assume

850: $P_i=p_i^{d_i}$ and $\overline{P_i}=m/P_i$. Since $p_1,\cdots,p_r$

851: are all primes, it is obvious that $\gcd(P_i,\overline{P_i})=1$. In

852: addition, assume $\mathbb{M}=\{0,\cdots,m-1\}$, and $\forall i=1\sim

853: r$, $\mathbb{M}_i=\{0,\cdots,P_i-1\}$ and

854: $\mathbb{M}_i^*=\{0,\cdots,\overline{P_i}-1\}$.

855: 

856: First, $\forall i=1\sim r$, let us prove the ``only if" part.

857: $\forall a\in\mathbb{M}_i^*$, assume

858: $\mathbb{A}=\{x|x\in\mathbb{M},x\equiv

859: a\pmod{\overline{P_i}}\}=\{\overline{P_i}y+a|y\in\mathbb{M}_i\}$.

860: Since $f(x)$ is a permutation polynomial modulo

861: $m=P_i\overline{P_i}$, $\forall y_1,y_2\in\mathbb{A}$ and $y_1\neq

862: y_2$, one has $f(y_1)\not\equiv f(y_2)\pmod{P_i}$ or

863: $f(y_1)\not\equiv f(y_2)\pmod{\overline{P_i}}$, otherwise it

864: conflicts with part 3) of Theorem 2.3 in \cite{NZM:NumberTheory1991}

865: (i.e., Property IX in \S15 of \cite{Pan:ConciseNumberTheory1998}).

866: Since $f(y_1)\equiv f(y_2)\equiv a\pmod{\overline{P_i}}$, one

867: immediately knows $f(y_1)\not\equiv f(y_2)\pmod{P_i}$. From Lemma

868: \ref{theorem:bijection-degree1}, since $\gcd(\overline{P_i},P_i)=1$,

869: $g(y)=\overline{P_i}y+a$ is a permutation polynomial modulo $P_i$.

870: This means that $\mathbb{A}$ is a complete system of residues modulo

871: $P_i$. This leads to the result that $f(x)$ is a permutation

872: polynomial modulo $P_i$.

873: 

874: Next, we prove the ``if" part. Given $r$ integers as follows:

875: $a_1\in\mathbb{M}_1$, $\cdots$, $a_r\in\mathbb{M}_r$, construct the

876: system of $r$ simultaneous congruences, $i=1\sim r:f(x)\equiv

877: a_i\pmod{P_i}$. From the Chinese Remainder Theorem, there is exactly

878: one solution of $f(x)$ in each complete system of residues modulo

879: $m$. Since $f(x)$ is a permutation polynomial modulo each $P_i$, we

880: can construct $m=\prod_{i=1}^rp_i^{d_i}$ systems of $r$ simultaneous

881: congruences, and get $m$ distinct solutions of $f(x)$ in each

882: complete system of residues modulo $m$. Considering there are only

883: $m$ elements in each complete system of residues modulo $m$, one can

884: immediately deduce that $f(x)$ is also a permutation polynomial

885: modulo $m$.

886: \end{proof}

887: 

888: \begin{theorem}

889: Assume $p_1$, $\cdots$, $p_r$ are $r$ distinct prime numbers, $d_1$,

890: $\cdots$, $d_r\geq 1$ and $m=\prod_{i=1}^rp_i^{d_i}$. If $f_1(x)$,

891: $\cdots$, $f_r(x)$ are permutation polynomials modulo $p_1^{d_1}$,

892: $\cdots$, $p_r^{d_r}$, respectively, then there exists one and only

893: one permutation polynomial $f(x)$ modulo $m$ in each complete system

894: of polynomial residues modulo $m$, such that $f(x)\equiv

895: f_i(x)\pmod{p_i^{d_i}}$ holds for $i\in\{1,\cdots,r\}$.

896: \end{theorem}

897: \begin{proof}

898: Applying the Chinese remainder theorem on each coefficients of the

899: polynomials, one can immediately prove this theorem.

900: \end{proof}

901: 

902: \section{Permutation Polynomials modulo $\bm{p}$}

903: \label{section:prime}

904: 

905: It is natural to connect Fermat's Little Theorem with permutation

906: polynomials, since this theorem actually says that there always

907: exists a permutation polynomial $f(x)=x^p$ of degree $p$ modulo a

908: prime $p$ such that $\forall x\in\mathbb{Z}$, $f(x)\equiv x\pmod p$.

909: However, the original Fermat's Little Theorem say nothing about how

910: many permutation polynomials there are and how to calculate other

911: permutation polynomials (if any). We have an enhanced version to

912: answer this question.

913: 

914: \begin{theorem}\label{theorem:Fermat-enhanced}

915: Assume $p$ is a prime. There exist $(p-1)p!$ congruence classes of

916: permutation polynomials of degree $p$ modulo $p$. For each given

917: bijection over $\{0,\cdots,p-1\}$, there exist $p-1$ congruence

918: classes of permutation polynomial of degree $p$ modulo $p$.

919: Specially, there exists a permutation polynomial $f(x)=x^p$, such

920: that $\forall x\in\mathbb{Z}$, $f(x)\equiv x\pmod p$.

921: \end{theorem}

922: \begin{proof}

923: Assume $f(x)=a_px_p+a_{p-1}x^{p-1}+\cdots+a_1x+a_0$, where

924: $a_p\not\equiv 0\pmod p$. Choosing $x=0,\cdots,p-1$, respectively,

925: one can get the following $p$ congruences modulo $p$.

926: \begin{eqnarray*}

927: a_p\cdot 0^p+a_{p-1}\cdot 0^{p-1}+\cdots a_1\cdot 0+a_0 & \equiv &

928: f(0)\pmod p\\

929: a_p\cdot 1^p+a_{p-1}\cdot 1^{p-1}+\cdots a_1\cdot 1+a_0 & \equiv &

930: f(1)\pmod p\\

931: & \vdots\\

932: a_p\cdot (p-1)^p+a_{p-1}\cdot (p-1)^{p-1}+\cdots a_1\cdot(p-1)+a_0 &

933: \equiv & f(p-1)\pmod p

934: \end{eqnarray*}

935: Fixing $a_p$, rewrite the above system of congruences as the

936: following matrix form $\bm{A}\bm{X}_a\equiv \bm{B}\pmod p$.

937: \begin{equation}

938: \left[\begin{matrix}%

939: 1 & 0 & 0 & \cdots & 0\\

940: 1 & 1 & 1 & \cdots & 1\\

941: 1 & 2 & 2^2 & \cdots & 2^p\\

942: \vdots & \vdots & \vdots & \ddots & \vdots\\

943: 1 & p-1 & (p-1)^2 & \cdots & (p-1)^p

944: \end{matrix}\right]

945: \left[\begin{matrix}%

946: a_0\\

947: a_1\\

948: a_2\\

949: \vdots\\

950: a_{p-1}

951: \end{matrix}\right]\equiv

952: \left[\begin{matrix}%

953: f(0)\\

954: f(1)-a_p\\

955: f(2)-2^pa_p\\

956: \vdots\\

957: f(p-1)-(p-1)^pa^p

958: \end{matrix}\right]\pmod p\label{equation:Fermat-enhanced}

959: \end{equation}

960: Apparently, $\bm{A}$ is a Vandermonde matrix, so its determinant can

961: be calculated as $|\bm{A}|=\prod_{0\leq i<j\leq p-1}(j-i)$

962: \cite[\S4.4]{Zhang:MatrixTheory1999}. Since $p$ is a prime and

963: $0\leq(j-i)\leq p-1$, one has $\gcd(|\bm{A}|,p)=1$. Thus, the above

964: system of congruence has a unique (i.e., one and only one) solution

965: modulo $p$, for each combination of the values of

966: $f(0),\cdots,f(p-1)$ and $a_p$. For each possible value of $a_p$,

967: the number of all possible combinations of the values of

968: $f(0),\cdots,f(p-1)$ is $p!$. Since $a_p$ has $p-1$ congruence

969: classes modulo $p$, one immediately deduces that there exists

970: $(p-1)p!$ congruence classes of permutation polynomials of degree

971: $p$ modulo $p$. For each bijection over $\{0,\cdots,p-1\}$, i.e.,

972: for each combination of the values of $f(0),\cdots,f(p-1)$, there

973: are $p-1$ distinct congruence classes of permutation polynomials of

974: degree $p$ modulo $p$, each of which corresponds to one possible

975: value of $a_p$ modulo $p$.

976: 

977: When $a_p=1$, choosing $f(i)=i$ for $i=0\sim p-1$, one can get a

978: special solution: $a_0\equiv a_1\equiv \cdots\equiv a_{p-1}\equiv

979: 0\pmod p$. This leads to $f(x)=x^p\equiv x\pmod p$, which is the

980: permutation polynomial of degree $p$ modulo $p$ as mentioned in the

981: Fermat's little theorem.

982: \end{proof}

983: 

984: From the above theorem, one can get some more results on the number

985: of permutation polynomials and induced bijections modulo $p$.

986: 

987: \begin{notation}

988: Assume $p$ is a prime. Denote the number of distinct permutation

989: polynomials and the number of all distinct polynomials in a complete

990: system of polynomial resides of degree $\leq n$ modulo $p$ by

991: $N_{pp}(\leq n,p)$ and $N_p(\leq n,p)$ respectively. Here, the

992: subscript ``pp" means ``permutation polynomial" and ``p" denotes

993: ``polynomial". Similar functions will be defined later.

994: \end{notation}

995: 

996: \begin{corollary}\label{corollary:count-bijective-poly-p}

997: Assume $p$ is a prime. The following is true: when $n\geq p-1$,

998: $\dfrac{N_{pp}(\leq n,p)}{N_p(\leq n,p)}=\dfrac{(p-1)!}{p^{p-1}}$.

999: \end{corollary}

1000: \begin{proof}

1001: Recall the proof of Theorem \ref{theorem:Fermat-enhanced}, when

1002: $n\geq p-1$, changing the degree of the polynomial from $p$ to $n$

1003: and moving $a_p,\cdots,a_n$ to the right side, Eq.

1004: (\ref{equation:Fermat-enhanced}) has a unique set of incongruent

1005: solutions to the values of $f(0),\cdots,f(p-1),a_p,\cdots,a_n$.

1006: Since $f(0),\cdots,f(p-1)$ forms a complete permutation modulo $p$,

1007: one immediately has $N_{pp}(\leq n)/N_p(\leq

1008: n)=p!/p^p=(p-1)!/p^{p-1}$.

1009: \end{proof}

1010: 

1011: \begin{corollary}\label{corollary:count-bijection-p}

1012: Assume $p$ is a prime and $n\geq p-1$. The number of bijections

1013: induced from permutation polynomials of degree $\leq n$ modulo $p$

1014: is $p!$.

1015: \end{corollary}

1016: \begin{proof}

1017: This corollary can be proved in the same way as the above corollary,

1018: due to the fact that each permutation of

1019: $f(0),\cdots,f(p-1),a_p,\cdots,a_n$ corresponds to a unique set of

1020: incongruent solutions to $a_0,\cdots,a_{p-1}$.

1021: \end{proof}

1022: 

1023: \begin{corollary}[A special case of Lemma \ref{lemma:equivalent-poly-pd}]

1024: Assume $p$ is a prime. Two permutation polynomials of degree $\leq

1025: p-1$ modulo $p$, $f_1(x)$ and $f_2(x)$, are equivalent if and only

1026: if they are congruence polynomials modulo $p$, i.e., $f_1(x)\equiv

1027: f_2(x)\pmod p$.

1028: \end{corollary}

1029: \begin{proof}

1030: The ``if" part is obvious. Let us see the ``only if" part. From the

1031: above two corollaries, the number of permutation polynomials of

1032: degree $\leq p-1$ modulo $p$ and the number of bijections induced

1033: from these polynomials are both $p!$. This immediately leads to the

1034: fact that any two equivalent permutation polynomials are congruent

1035: polynomials, otherwise the number of bijections will be less than

1036: $p!$. Thus, this corollary is true.

1037: \end{proof}

1038: 

1039: \begin{corollary}

1040: Assume $p$ is a prime and $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a

1041: permutation polynomial of degree $n\geq p$ modulo $p$. Then $f(x)$

1042: has exactly $p^{n-p}$ equivalent polynomials of degree $\leq n$

1043: modulo $p$ (including itself).

1044: \end{corollary}

1045: \begin{proof}

1046: This corollary can be proved in a similar way to the above

1047: corollaries.

1048: \end{proof}

1049: 

1050: \section{Permutation Polynomials modulo $\bm{p^d}$ ($\bm{d\geq 1}$)}

1051: \label{section:PrimePowers}

1052: 

1053: \subsection{Hierarchy Theorem}

1054: 

1055: This theorem shows the hierarchical structure of the bijection

1056: induced from a permutation polynomial modulo $p^d$.

1057: 

1058: \begin{theorem}[Hierarchy Theorem]\label{theorem:hierarchy-pd}

1059: Assume $p$ is a prime and $f(x)=a_nx^n+\cdots+a_1x$ is a permutation

1060: polynomial of degree $n$ modulo $m=p^d$ and

1061: $\mathbb{A}=\{0,\cdots,p^d-1\}$. The following results are true.

1062: \begin{enumerate}

1063: \item

1064: The induced bijection $F:\mathbb{A}\to\mathbb{A}$ is composed of two

1065: sub-bijections, $F_1:\mathbb{A}(\geq 1|_p)\to\mathbb{A}(\geq 1|_p)$

1066: and $F_0:\mathbb{A}\backslash\mathbb{A}(\geq

1067: 1|_p)\to\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$.

1068: 

1069: \item

1070: The sub-bijection $F_0$ is composed of $p-1$ sub-bijections,

1071: $\forall i=1\sim p-1$, $F_{0,i}:[(\mathbb{A}(\geq 1|_p)+i)\bmod

1072: m]\to\left[\left(\mathbb{A}(\geq

1073: 1|_p)+\sum_{k=1}^ni^ka_k\right)\bmod m\right]$.

1074: 

1075: \item

1076: When $d\geq 2$, each of the $p$ sub-bijections, $F_1$ and

1077: $F_{0,1},\cdots,F_{0,p-1}$, corresponds to a permutation polynomial

1078: of degree $\leq d-1$ modulo $p^{d-1}$ in the following form:

1079: $f^*(z)=\sum_{i=d-1}^1b_ip^{i-1}z^i=b_{d-1}p^{d-2}z^{d-1}+\cdots+b_2pz+b_1z$.

1080: 

1081: \item

1082: When $d\geq 2$, $F_1$ has is composed of $d$ sub-bijections:

1083: $i=1\sim d-1$, $F_{1,i}:\mathbb{A}(i|_p)\to\mathbb{A}(i|_p)$, and

1084: $F_{1,d}:\mathbb{A}(+\infty|_p)\to\mathbb{A}(+\infty|_p)$. Each of

1085: $F_{0,1},\cdots,F_{0,p-1}$ is also composed of $d$ sub-bijections of

1086: this kind.

1087: 

1088: \item

1089: When $d\geq 1$, $\forall i,j\in\{0,\cdots,p-1\}$ and $i\neq j$,

1090: $\sum_{k=1}^na_k(j^k-i^k)=a_1(j-i)+a_2(j^2-i^2)+\cdots+a_n(j^n-i^n)\not\equiv

1091: 0\pmod p$.

1092: 

1093: \item

1094: When $d\geq 2$, $\forall i\in\{0,\cdots,p-1\}$,

1095: $\sum_{k=1}^nki^{k-1}a_k=a_1+2\cdot i^1\cdot a_2+\cdots+n\cdot

1096: i^{n-1}\cdot a_n\not\equiv 0\pmod p$.

1097: \end{enumerate}

1098: \end{theorem}

1099: \begin{proof}

1100: We prove all the results one by one. Note that $\mathbb{A}(\geq

1101: 1|_p)$, $[(\mathbb{A}(\geq 1|_p)+1)\bmod m],\cdots,[(\mathbb{A}(\geq

1102: 1|_p)+p-1)\bmod m]$ forms a partition of $\mathbb{A}$.

1103: 

1104: 1. $\forall x\in\mathbb{A}(\geq 1|_p)$, then $p\mid x$. Since $x\mid

1105: f(x)$, so $p\mid f(x)$. This means that $f(x)$ forms a sub-bijection

1106: $F_1:\mathbb{A}(\geq 1|_p)\to\mathbb{A}(\geq 1|_p)$. From Lemma

1107: \ref{lemma:sub-bijection}, there exists another sub-bijection

1108: $F_0:\mathbb{A}\backslash\mathbb{A}(\geq

1109: 1|_p)\to\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$.

1110: 

1111: 2 \& 5. $\forall i=1\sim p-1$ and $\forall x\in[(\mathbb{A}(\geq

1112: 1|_p)+i)\bmod p]$, one has $x\equiv i\mod p$. Assume $x=y+i$, where

1113: $y\in\mathbb{A}(\geq 1|_p)$, and one can get a new polynomial as

1114: follows:

1115: \[

1116: f_{0,i}^*(y)=f(y+i)=a_n(y+i)^n+\cdots+a_1(y+i)=f_{0,i}^{**}(y)+\sum_{k=1}^ni^ka_k,

1117: \]

1118: where

1119: $f_{0,i}^{**}(y)=\sum_{l=n}^1\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_ky^l\right)$.

1120: Applying the first result on $f_{0,i}^{**}(y)$, one knows it forms a

1121: sub-bijection over $\mathbb{A}(\geq 1|_p)$. So, $f(x)$ forms a

1122: sub-bijection $F_{0,i}:[(\mathbb{A}(\geq 1|_p)+i)\bmod

1123: p]\to\left[\left(\mathbb{A}(\geq

1124: 1|_p)+\sum_{k=1}^ni^ka_k\right)\bmod m\right]$. This proves the 2nd

1125: result of this theorem. Since $f(x)$ is a permutation polynomial

1126: modulo $p^d$, $\forall i\neq j$, the ranges of

1127: $F_{0,1},\cdots,F_{0,p-1}$ should form a partition of

1128: $\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$. This means that

1129: $\forall i,j\in\{0,\cdots,p-1\}$ and $i\neq j$,

1130: $\sum_{k=1}^ni^ka_k\not\equiv \sum_{k=1}^nj^ka_k\pmod p$. This leads

1131: to the 5th result of this theorem.

1132: 

1133: 3. For $F_1$, since $p\mid x$, let us assume $x=pz$, where

1134: $z\in\{0,\cdots,p^{d-1}-1\}$. Substitute $x=pz$ into $f(x)$, we have

1135: another polynomial

1136: $f_1(z)=f(pz)=a_n(pz)^n+\cdots+a_1(pz)=pf_1^*(z)$, where

1137: $f_1^*(z)=a_np^{n-1}z^n+\cdots+a_1z$. Apparently, over

1138: $\mathbb{A}(\geq |_p)$, $f(x)$ is uniquely determined by the

1139: polynomial $f_1^*(z)$ modulo $p^{d-1}$. This means that $f_1^*(z)$

1140: is a permutation polynomial modulo $p^{d-1}$. When $n\geq d$,

1141: $p^{d-1}\mid a_np^{n-1}z^n+\cdots+a_dp^{d-1}z^d$, so $f_1^*(z)\equiv

1142: a_{d-1}p^{d-2}z^{d-1}+\cdots+a_1z\pmod{p^{d-1}}$. As a result, the

1143: degree of $f_1^*(z)$ modulo $p^{d-1}$ is always not greater than

1144: $d-1$. For $F_{0,1},\cdots,F_{0,p-1}$, applying the same analysis on

1145: $f_{0,1}^{**}(y),\cdots,f_{0,p-1}^{**}(y)$, one can get a similar

1146: result.

1147: 

1148: 4 \& 6. When $x\in\mathbb{A}(+\infty|_p)=\{0\}$, $f(x)=f(0)=0$, so

1149: there exists a sub-bijection

1150: $F_{1,d}:\mathbb{A}(+\infty|_p)\to\mathbb{A}(+\infty|_p)$. $\forall

1151: i=1\sim d-1$ and $\forall x\in\mathbb{A}(i|_p)$, then $\exists

1152: k_1,k_2\in\mathbb{Z}$ and $k_2\not\equiv 0\pmod p$, such that

1153: $x=p^i(k_1p+k_2)$. Then,

1154: $x^2=p^{2i}(k_1p+k_2)^2=p^{i+1}p^{i-1}(k_1p+k_2)^2$, so $p^{i+1}\mid

1155: x^2$. As a result, $f(x)\equiv a_1x=a_1p^i(k_1p+k_2)\pmod{p^{i+1}}$.

1156: Assume $a_1\equiv 0\pmod p$, one has $f(x)\equiv 0\pmod{p^{i+1}}$.

1157: This means that $f(x)\in\mathbb{A}(\geq i+1|_p)\backslash\{0\}$.

1158: However, since $p\geq 2$, the cardinality of $\mathbb{A}(\geq

1159: i+1|_p)\backslash\{0\}$ is always smaller than the cardinality of

1160: $\mathbb{A}(i|_p)$, which conflicts with the fact that $f(x)$ is a

1161: permutation polynomial modulo $p^d$. So one immediately has

1162: $a_1\not\equiv 0\pmod p$ and $f(x)\in\mathbb{A}(i|_p)$, i.e., $f(x)$

1163: forms a sub-bijection over $\mathbb{A}(i|_p)$. For $i=1\sim p-1$,

1164: applying the same analysis for $f_{0,i}^{**}(y)$, we can get similar

1165: results: $\sum_{k=1}^nki^{k-1}a_k\not\equiv 0\pmod p$ and

1166: $f(x)_{0,i}^{**}(y)\in\mathbb{A}(i|_p)$. Thus the 4th and the 6th

1167: results have been proved.

1168: \end{proof}

1169: 

1170: \subsection{Necessary and Sufficient Conditions for Permutation Polynomials}

1171: 

1172: \iffalse

1173: \begin{lemma}

1174: Assume $p$ is a prime. The polynomial $f(x)=a_nx^n+\cdots+a_1x$ is a

1175: permutation polynomial modulo $p$ if and only if $\forall

1176: i,j\in\{0,\cdots,p-1\}$ and $i\neq j$,

1177: $\sum_{k=1}^na_k(j^k-i^k)=a_1(j-i)+a_2(j^2-i^2)+\cdots+a_n(j^n-i^n)\not\equiv

1178: 0\pmod p$.

1179: \end{lemma}

1180: \begin{proof}

1181: Since $f(j)-f(i)=\sum_{k=1}^na_k(j^k-i^k)$, this lemma is true from

1182: the definition of a bijection.

1183: \end{proof}

1184: \fi

1185: 

1186: \begin{theorem}\label{theorem:bijective-poly-NSC-pd}

1187: Assume $p$ is a prime and $d\geq 2$. The polynomial

1188: $f(x)=a_nx^n+\cdots+a_1x$ is a permutation polynomial modulo $p^d$

1189: if and only if the following two conditions are true simultaneously:

1190: \begin{enumerate}

1191: \item

1192: $f(x)$ is a permutation polynomial modulo $p$, i.e., $\forall

1193: i,j\in\{0,\cdots,p-1\}$ and $i\neq j$,

1194: $f(j)-f(i)=\sum_{k=1}^na_k(j^k-i^k)=a_1(j-i)+a_2(j^2-i^2)+\cdots+a_n(j^n-i^n)\not\equiv

1195: 0\pmod p$.

1196: 

1197: \item

1198: $\forall i\in\{0,\cdots,p-1\}$,

1199: $\sum_{k=1}^nki^{k-1}a_k=a_1+2ia_2+\cdots+ni^{n-1}a_n\not\equiv

1200: 0\pmod p$.

1201: \end{enumerate}

1202: \end{theorem}

1203: \begin{proof}

1204: The ``only if" part of this theorem has been proved in Theorem

1205: \ref{theorem:hierarchy-pd}, so we only focus on the ``if" part. Let

1206: us use mathematical induction on $d$ to prove this part.

1207: 

1208: 1) When $d=2$, consider the $p$ sub-bijections, $F_1$,

1209: $F_{0,1},\cdots,F_{0,p-1}$, separately.

1210: 

1211: When $x\in\mathbb{A}(\geq 1|_p)$, assume $x=pz$, where

1212: $z\in\{0,\cdots,p-1\}$, so $f(x)=a_nx^n+\cdots+a_2x^2+a_1x\equiv

1213: p\cdot a_1z\pmod{p^2}$. Apparently, $f(x)$ is uniquely determined by

1214: the polynomial $f_1^*(z)=a_1z$ modulo $p$. Choosing $i=0$, the

1215: second necessary and sufficient condition becomes $a_1\not\equiv

1216: 0\pmod p$, which means $\gcd(a_1,p)=1$. Then, from Lemma

1217: \ref{theorem:bijection-degree1}, $f_1^*(z)=a_1z$ forms a bijection

1218: over $\{0,\cdots,p-1\}$. This means $f(x)$ forms a bijection $F_1$

1219: over $\mathbb{A}(\geq 1|_p)$.

1220: 

1221: $\forall i=1\sim p-1$ and $\forall x\in[(\mathbb{A}(\geq

1222: 1|_p)+i)\bmod p]$, assume $x=y+i$, where $y\in\mathbb{A}(\geq

1223: 1|_p)$. Substitute $x=y+i$ into $f(x)$, one has

1224: $f_{0,i}^*(y)=f_{0,i}^{**}(y)+\sum_{k=1}^ni^ka_k$, where

1225: $f_{0,i}^{**}(y)=\sum_{l=n}^1\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_ky^l\right)$.

1226: Assume $y=pz$, where $z\in\{0,\cdots,p-1\}$, due to the same reason

1227: in the case of $x\in\mathbb{A}(\geq 1|_p)$, $f_{0,i}^{**}(y)\equiv

1228: \sum_{k=1}^nki^{k-1}a_ky\equiv p\sum_{k=1}^nki^{k-1}a_kz\pmod{p^2}$.

1229: The second necessary and sufficient condition ensure that

1230: $\gcd\left(\sum_{k=1}^nki^{k-1}a_k,p\right)=1$, so

1231: $\sum_{k=1}^nki^{k-1}a_kz$ forms a bijection over $\{0,\cdots,p-1\}$

1232: and thus $f_{0,i}^{**}(y)$ forms a bijection over $\mathbb{A}(\geq

1233: 1|_p)$. This further leads to the fact that $f(x)$ forms a bijection

1234: $F_{0,i}:[(\mathbb{A}(\geq 1|_p)+i)\bmod

1235: p]\to\left[\left(\mathbb{A}(\geq

1236: 1|_p)+\sum_{k=1}^ni^ka_k\right)\bmod p\right]$.

1237: 

1238: In addition, the first necessary and sufficient condition ensures

1239: that the range of $F_{0,0},\cdots,F_{0,p-1}$ forms a partition of

1240: $\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$. This means that there

1241: exists a super-bijection $F_0$ over

1242: $\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$.

1243: 

1244: The above analyses show that $f(x)$ forms a bijection over

1245: $\mathbb{A}$.

1246: 

1247: 2) Assume the ``if" part is true for $2,\cdots,d-1$. Let us prove

1248: the case of $d\geq 3$. Similarly, let us consider the $p$

1249: sub-bijections, $F_1$, $F_{0,1},\cdots,F_{0,p-1}$, separately.

1250: 

1251: When $x\in\mathbb{A}(\geq 1|_p)$, assume $x=pz$, where

1252: $z\in\{0,\cdots,p^{d-1}-1\}$, so

1253: $f(x)=a_nx^n+\cdots+a_2x^2+a_1x\equiv

1254: p(a_np^{n-1}x^n+\cdots+a_2pz+a_1z)\pmod{p^d}$. Apparently, $f(x)$ is

1255: uniquely determined by the polynomial

1256: $f_1^*(z)=b_nz^n+\cdots+b_1z=a_np^{n-1}x^n+\cdots+a_2pz+a_1z$ modulo

1257: $p^{d-1}$. Since $b_i\equiv 0\pmod p$ when $i\geq 2$, one can easily

1258: verify that $f_1^*(z)$ satisfies the two necessary and sufficient

1259: conditions, so from the previous assumption, $f_1^*(z)$ is a

1260: permutation polynomial modulo $p^{d-1}$. This means that $f(x)$

1261: forms a bijection over $\mathbb{A}(\geq 1|_p)$.

1262: 

1263: $\forall i=1\sim p-1$ and $\forall x\in[(\mathbb{A}(\geq

1264: 1|_p)+i)\bmod p]$, assume $x=y+i$, where $y\in\mathbb{A}(\geq

1265: 1|_p)$. Substitute $x=y+i$ into $f(x)$, one has

1266: $f_{0,i}^*(y)=f_{0,i}^{**}(y)+\sum_{k=1}^ni^ka_k$, where

1267: $f_{0,i}^{**}(y)=\sum_{l=n}^1\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_ky^l\right)$.

1268: Assume $y=pz$, where $z\in\{0,\cdots,p-1\}$, one has

1269: $f_{0,i}^{**}(y)=\sum_{l=n}^1\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_kp^lz^l\right)

1270: =p\sum_{l=n}^1\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_kp^{l-1}z^l\right)$.

1271: Apparently, $f_{0,i}^{**}(y)$ is uniquely determined by the

1272: polynomial

1273: $f_{0,i}^{***}(z)=b_nz^n+\cdots+b_1z=\sum_{l=n}^2\left(\sum_{k=l}^n\binom{k}{l}i^{k-l}a_kp^{l-1}z^l\right)+

1274: \sum_{k=1}^nki^{k-1}a_kz$ modulo $p^{d-1}$. Similarly, since

1275: $b_k\equiv 0\pmod p$ when $k\geq 2$, $\forall

1276: i,j\in\{0,\cdots,p-1\}$ and $i\neq j$, one has

1277: $\sum_{k=1}^nki^{k-1}b_k\equiv b_1=\sum_{k=1}^nki^{k-1}a_k\not\equiv

1278: 0\pmod p$ and $\sum_{k=1}^nb_k(j^k-i^k)\equiv

1279: b_1(j-i)=(j-i)\sum_{k=1}^nki^{k-1}a_k\not\equiv 0\pmod p$, where

1280: note that $j-i\not\equiv 0\pmod p$. That is, the two necessary and

1281: sufficient conditions hold for $f_{0,i}^{***}(z)$, so from the

1282: previous assumption, $f_{0,i}^{***}(z)$ is a permutation polynomial

1283: modulo $p^{d-1}$, i.e., $f_{0,i}^{**}(y)$ forms a bijection over

1284: $\mathbb{A}(\geq 1|_p)$ and $f(x)$ forms a bijection

1285: $F_{0,i}:[(\mathbb{A}(\geq 1|_p)+i)\bmod

1286: p]\to\left[\left(\mathbb{A}(\geq

1287: 1|_p)+\sum_{k=1}^ni^ka_k\right)\bmod p\right]$.

1288: 

1289: In addition, the first necessary and sufficient condition ensures

1290: that the range of $F_{0,0},\cdots,F_{0,p-1}$ forms a partition of

1291: $\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$. This means that there

1292: exists a super-bijection $F_0$ over

1293: $\mathbb{A}\backslash\mathbb{A}(\geq 1|_p)$.

1294: 

1295: The above analyses show that $f(x)$ forms a bijection over

1296: $\mathbb{A}$. Thus this theorem is proved.

1297: \end{proof}

1298: 

1299: \begin{corollary}\label{corollary:bijective-poly-NSC-2d}

1300: The polynomial $f(x)=a_nx^n+\cdots+a_1x$ is a permutation polynomial

1301: modulo $2^d$ if and only if the following two conditions are true

1302: simultaneously: $a_1\equiv 1\pmod 2$, $a_2+a_4+\cdots\equiv

1303: a_3+a_5+\cdots\equiv 0\pmod 2$.

1304: \end{corollary}

1305: \begin{proof}

1306: From Theorem \ref{theorem:bijective-poly-NSC-pd}, choosing $p=2$,

1307: one has the following necessary and sufficient conditions:

1308: $\sum_{i=1}^na_i\not\equiv 0\pmod 2$, $a_1\not\equiv 0\pmod 2$ and

1309: $\sum_{i=1}^nia_i\not\equiv 0\pmod 2$. These conditions can be

1310: simplified to be: $a_1\equiv 1\pmod 2$, $\sum_{i=2}^na_i\equiv

1311: \sum_{i=2}^nia_i\equiv 0\pmod 2$. Removing even terms from

1312: $\sum_{i=2}^nia_i\equiv 0\pmod 2$, one has $a_3+a_5+\cdots\equiv

1313: 0\pmod 2$. Then, subtracting $a_3+a_5+\cdots$ from

1314: $\sum_{i=2}^na_i$, one has $a_2+a_4+\cdots\equiv 0\pmod 2$. This

1315: corollary is thus proved.

1316: \end{proof}

1317: 

1318: \begin{corollary}\label{corollary:bijective-poly-NSC-all-degrees}

1319: Assume $p$ is a prime and $d\geq 2$. If $f(x)$ is a permutation

1320: polynomial modulo $p^d$, then $\forall i\geq 1$, it is still a

1321: permutation polynomial modulo $p^i$.

1322: \end{corollary}

1323: \begin{proof}

1324: This corollary is a direct result of Theorem

1325: \ref{theorem:bijective-poly-NSC-pd}.

1326: \end{proof}

1327: 

1328: \iffalse

1329: \begin{corollary}\label{corollary:equivalent-special-poly-NSC}

1330: Assume $p$ is a prime and $d\geq 1$. A polynomial

1331: $f(x)=\sum_{i=n}^1a_ip^{i-1}x^i=a_np^{n-1}x^n+\cdots+a_2px^2+a_1x$

1332: is a permutation polynomial modulo $p^d$ if and only if

1333: $a_1\not\equiv 0\pmod p$.

1334: \end{corollary}

1335: \begin{proof}

1336: From Theorem \ref{theorem:bijective-poly-NSC-pd}, substitute the

1337: coefficients into the necessary and sufficient conditions. Since

1338: only the lowest coefficients are not congruent to 0 modulo $p$, one

1339: can get the following conditions: for $i=1\sim p-1$, $ia_1\not\equiv

1340: 0\pmod p$. This immediately leads to $a_1\not\equiv 0\pmod p$ and

1341: completes the proof of this corollary.

1342: \end{proof}

1343: The above corollary will be used later to estimate an upper bound of

1344: the number of bijections induced from all permutation polynomials

1345: modulo $p^d$.\fi

1346: 

1347: \begin{theorem}\label{theorem:bijective-poly-NSC-pd-degree2}

1348: Assume $p$ is a prime and $d\geq 1$. A binomial $f(x)=a_2x^2+a_1x$

1349: is a permutation polynomial modulo $p^d$ if and only if

1350: $a_1\not\equiv 0\pmod p$ and $a_2\equiv 0\pmod p$.

1351: \end{theorem}

1352: \begin{proof}

1353: The ``if" part can be easily verified by checking the necessary and

1354: sufficient conditions in Theorem

1355: \ref{theorem:bijective-poly-NSC-pd}. We focus on the ``only if"

1356: part.

1357: 

1358: When $p=2$, one can verify the result is true. Let us consider the

1359: case of $p>2$. From Theorem \ref{theorem:bijective-poly-NSC-pd},

1360: $a_1(1-(p-1))+a_2(1^2-(p-1)^2)\equiv 2a_1\not\equiv 0\pmod p$, which

1361: immediately leads to $a_1\not\equiv 0\pmod p$. Again, from Theorem

1362: \ref{theorem:bijective-poly-NSC-pd}, the following conditions hold:

1363: $\forall i=1\sim p-1$, $ia_1+i^2a_2\not\equiv 0\pmod p$. Since $p$

1364: is a prime, each integer in $\{1,\cdots,p-1\}$ has an inverse modulo

1365: $p$. Multiplying the inverse of $i$ at both sides of each condition,

1366: one gets $a_1+ia_2\not\equiv 0\pmod p$, so $a_2\not\equiv

1367: -a_1\bar{i}\pmod p$, where $\bar{i}$ is the inverse of $i$ modulo

1368: $p$. Since $\{i\}$ forms a reduced system of residues modulo $p$,

1369: $\{-a_1\bar{i}\}=\{-a_1,\cdots,-a_1(p-1)\}$ still forms a reduced

1370: system of residues modulo $p$. Thus $a_2\equiv 0\pmod p$.

1371: \end{proof}

1372: \begin{remark}

1373: Note that Theorem \ref{theorem:bijective-poly-NSC-pd-degree2}

1374: actually says that the first group of the necessary and sufficient

1375: conditions covers the second group of conditions when the degree is

1376: 1 or 2 modulo $p^d$.

1377: \end{remark}

1378: \begin{corollary}

1379: Assume $p$ is a prime and $d\geq 1$. If $f(x)$ is a permutation

1380: polynomial of degree 1 modulo $p^d$, then $\forall i\geq 1$, it is

1381: still a permutation polynomial modulo $p^i$.

1382: \end{corollary}

1383: \begin{proof}

1384: This corollary is a direct result of Theorem

1385: \ref{theorem:bijective-poly-NSC-pd-degree2}.

1386: \end{proof}

1387: 

1388: \fproblem{When $3\leq n\leq p-1$, is it possible to get further

1389: simplified necessary and sufficient conditions?}

1390: 

1391: \begin{remark}\label{remark:RepeatingWork}

1392: After finishing the first draft of this paper, we noticed Rivest's

1393: paper \cite{Rivest:PPmod2w:FFTA2001} and found Corollary

1394: \ref{corollary:bijective-poly-NSC-2d} was proved by the author in

1395: 2002. Through \cite{Rivest:PPmod2w:FFTA2001}, we further noticed

1396: Mullen's paper \cite{Mullen:PolyFun-mod:AMH1984} and realized that

1397: Theorem \ref{theorem:bijective-poly-NSC-pd} can also be derived from

1398: Theorem 123 in \cite{HW:NumberTheory1979}, where the second

1399: condition becomes that $f'(x)\not\equiv 0\pmod p$ holds for any

1400: integer $x$. In addition, recently we found yet another paper

1401: \cite{Sun:PPmodCoding:IEEETIT2005}, in which Theorem

1402: \ref{theorem:bijective-poly-NSC-pd-degree2} was also obtained in a

1403: similar way (Corollary 2.4). Furthermore, we also noticed Lemma 4.2

1404: in Chap. 4 of \cite{RMT:DicksonPoly1993} gives a more general form

1405: of Theorem \ref{theorem:bijective-poly-NSC-pd}. Considering the fact

1406: that our proof of Theorem \ref{theorem:bijective-poly-NSC-pd} is

1407: independent of Theorem 123 in \cite{HW:NumberTheory1979}, it can be

1408: considered as a different proof of this result.

1409: \end{remark}

1410: 

1411: \subsection{Counting Permutation Polynomials and Induced Permutations}

1412: 

1413: The case of $n\geq p-1$ modulo $p$ has been solved in Corollary

1414: \ref{corollary:count-bijective-poly-p}. This subsection discusses

1415: other cases modulo $p^d$ ($d\geq 1$).

1416: 

1417: \begin{notation}

1418: Assume $p$ is a prime and $d\geq 1$. Denote the number of

1419: permutation polynomials and the number of all polynomials in a

1420: complete system of polynomial resides of degree $\leq n$ modulo

1421: $p^d$ by $N_{pp}(\leq n,p^d)$ and $N_p(\leq n,p^d)$, respectively.

1422: \end{notation}

1423: 

1424: \begin{remark}

1425: Note that the number of permutation polynomials of degree $n$ modulo

1426: $p^d$ can be easily calculated to be $N_{pp}(\leq n,p^d)-N_{pp}(\leq

1427: n-1,p^d)$. So this paper only focuses on the number of permutation

1428: polynomials of degree $\leq n$ modulo $p^d$.

1429: \end{remark}

1430: 

1431: \begin{theorem}

1432: For any prime $p$ and $d\geq 1$, $\dfrac{N_{pp}(\leq

1433: 1,p^d)}{N_p(\leq 1,p^d)}=\dfrac{p-1}{p}$ and $\dfrac{N_{pp}(\leq

1434: 2,p^d)}{N_p(\leq 2,p^d)}=\dfrac{p-1}{p^2}$.

1435: \end{theorem}

1436: \begin{proof}

1437: This theorem is a direct result of Theorems

1438: \ref{theorem:bijection-degree1} and

1439: \ref{theorem:bijective-poly-NSC-pd-degree2}.

1440: \end{proof}

1441: 

1442: \begin{theorem}\label{theorem:count-bijective-poly-pd}

1443: For any prime $p$ and $d\geq 2$, $\dfrac{N_{pp}(\leq

1444: n,p^d)}{N_p(\leq n,p^d)}=\dfrac{(p-1)^p(p-1)!}{p^{2p-1}}$ when

1445: $n\geq 2p-1$.

1446: \end{theorem}

1447: \begin{proof}

1448: From Theorem \ref{theorem:bijective-poly-NSC-pd}, a bijective

1449: polynomial should satisfy the following conditions:

1450: \begin{itemize}

1451: \item

1452: $\binom{p}{2}=\frac{p(p-1)}{2}$ conditions: $\forall

1453: i,j\in\{0,\cdots,p-1\}$ and $i\neq j$,

1454: $\sum_{k=1}^na_k(i^k-j^k)=a_1(i-j)+\cdots+a_n(i^n-j^n)\not\equiv

1455: 0\pmod p$;

1456: 

1457: \item

1458: $p$ conditions: $\forall i\in\{0,\cdots,p-1\}$,

1459: $\sum_{k=1}^nki^{k-1}a_k=a_1+2ia_2+\cdots+ni^{n-1}a_n\not\equiv

1460: 0\pmod p$.

1461: \end{itemize}

1462: Among the above $\binom{p}{2}+p=\frac{p(p+1)}{2}$ conditions, choose

1463: the following $2p-1$ conditions:

1464: \begin{itemize}

1465: \item

1466: $p-1$ conditions: $\forall i\in\{1,\cdots,p-1\}$ and $j=0$,

1467: $\sum_{k=1}^na_ki^k=a_1i+\cdots+a_ni^n\equiv b_i\pmod p$, where

1468: $b_i\not\equiv 0\pmod p$;

1469: 

1470: \item

1471: $p$ conditions: $\forall i\in\{0,\cdots,p-1\}$,

1472: $\sum_{k=1}^nki^{k-1}a_k=a_1+2ia_2+\cdots+ni^{n-1}a_n\equiv

1473: b_{p+i}\pmod p$, where $b_{p+i}\not\equiv 0\pmod p$.

1474: \end{itemize}

1475: Rewrite the above $2p-1$ condition as a system of congruences:

1476: \[

1477: \left[\begin{matrix}%

1478: 1 & 1 & 1 & \cdots & 1\\

1479: 2 & 2^2 & 2^3 & \cdots & 2^n\\

1480: \vdots & \vdots & \vdots & \ddots & \vdots\\

1481: (p-1) & (p-1)^2 & (p-1)^3 & \cdots & (p-1)^n\\

1482: 1 & 0 & 0 & \cdots & 0\\

1483: 1 & 2 & 3 & \cdots & n\\

1484: 1 & 2\cdot 2 & 3\cdot 2^2 & \cdots & n\cdot 2^{n-1}\\

1485: \vdots & \vdots & \vdots & \ddots & \vdots\\

1486: 1 & 2\cdot(p-1) & 3\cdot(p-1)^2 & \cdots & n\cdot(p-1)^{n-1}

1487: \end{matrix}\right]

1488: \left[\begin{matrix}%

1489: a_1\\a_2\\a_3\\a_4\\\vdots\\a_{n-3}\\a_{n-2}\\a_{n-1}\\a_n

1490: \end{matrix}\right]\equiv

1491: \left[\begin{matrix}%

1492: b_1\\

1493: b_2\\

1494: \vdots\\

1495: b_{p-1}\\

1496: b_p\\

1497: b_{p+1}\\

1498: b_{p+2}\\

1499: \vdots\\

1500: b_{2p-1}

1501: \end{matrix}\right]\pmod p.

1502: \]

1503: If we only consider $a_1,\cdots,a_{2p-1}$ as unknown variables, the

1504: above system can be reduced to be the following system:\small

1505: \[

1506: \left[\begin{matrix}%

1507: 1 & 1 & 1 & \cdots & 1\\

1508: 2 & 2^2 & 2^3 & \cdots & 2^{2p-1}\\

1509: \vdots & \vdots & \vdots & \ddots & \vdots\\

1510: (p-1) & (p-1)^2 & (p-1)^3 & \cdots & (p-1)^{2p-1}\\

1511: 1 & 0 & 0 & \cdots & 0\\

1512: 1 & 2 & 3 & \cdots & 2p-1\\

1513: 1 & 2\cdot 2 & 3\cdot 2^2 & \cdots & (2p-1)\cdot 2^{2p-2}\\

1514: \vdots & \vdots & \vdots & \ddots & \vdots\\

1515: 1 & 2\cdot(p-1) & 3\cdot(p-1)^2 & \cdots & (2p-1)\cdot(p-1)^{2p-2}

1516: \end{matrix}\right]

1517: \left[\begin{matrix}%

1518: a_1\\a_2\\\vdots\\a_{p-1}\\a_p\\a_{p+1}\\a_{p+2}\\\vdots\\a_{2p-1}

1519: \end{matrix}\right]\equiv

1520: \left[\begin{matrix}%

1521: b_1-\sum_{k=2p}^na_i\\

1522: b_2-\sum_{k=2p}^na_i2^k\\

1523: \vdots\\

1524: b_{p-1}-\sum_{k=2p}^na_i(p-1)^k\\

1525: b_p\\

1526: b_{p+1}-\sum_{k=2p}^nka_i\\

1527: b_{p+2}-\sum_{k=2p}^nk2^{k-1}a_i\\

1528: \vdots\\

1529: b_{2p-1}-\sum_{k=2p}^nk(p-1)^{k-1}a_i

1530: \end{matrix}\right]\pmod p.

1531: \]\normalsize

1532: Denoting the above system by $\bm{A}\bm{X}\equiv\bm{B}\pmod p$, from

1533: Corollary \ref{corollary:det-powers}, one has

1534: \[

1535: |\bm{A}|=(-1)^{\frac{(p-1)(p-2)}{2}}\prod_{i=1}^{p-1}i^4\prod_{1\leq

1536: i<j\leq p-1}(j-i)^4.

1537: \]

1538: Since all factors of $|\bm{A}|$ are in $\{1,\cdots,p-1\}$ and $p$ is

1539: a prime, $\gcd(|\bm{A}|,p)=1$. Then, for each valid combination of

1540: $(b_1,\cdots,b_{2p-1},a_{2p},\cdots,a_n)$, the above system of

1541: congruences has a unique set of incongruent solutions.

1542: 

1543: Next, let us count the number of all valid combinations of

1544: $(b_1,\cdots,b_{2p-1},a_{2p},\cdots,a_n)$. It is obvious that

1545: $\{a_{2p},\cdots,a_n\}$ can be any value and $b_p,\cdots,b_{2p-1}$

1546: can be any nonzero value modulo $p$. However, $b_1,\cdots,b_{p-1}$

1547: are also constrained by the following conditions: $\forall

1548: i,j\in\{1,\cdots,p-1\}$ and $i\neq j$,

1549: $\sum_{k=1}^na_k(i^k-j^k)\equiv b_i-b_j\not\equiv 0\pmod p$. That

1550: is, $\{b_i\bmod p\}_{i=1}^{p-1}$ forms a complete permutation over

1551: $\{1,\cdots,p-1\}$, so the number of possible values of

1552: $(b_1,\cdots,b_{p-1})$ is $(p-1)!$ in total $p^{p-1}$ combinations

1553: of the $p-1$ values. Combining the above fact, one immediately gets

1554: $N_{pp}(\leq n,p^d)/N_p(\leq n,p^d)=(1-1/p)^p\cdot

1555: (p-1)!/p^{p-1}=(p-1)^p(p-1)!/p^{2p-1}$.

1556: 

1557: Thus this theorem is proved.

1558: \end{proof}

1559: 

1560: \begin{theorem}

1561: For any prime $p$ and $d\geq 2$, the following inequalities hold:

1562: \begin{enumerate}

1563: \item

1564: when $3\leq n\leq p$, $\dfrac{N_{pp}(\leq n,p^d)}{N_p(\leq

1565: n,p^d)}\leq\dfrac{(p-1)P(p-1,n-1)}{p^n}=\dfrac{(p-1)(n-1)!\binom{p-1}{n-1}}{p^n}$;

1566: 

1567: \item

1568: when $p+1\leq n\leq 2p-2$, $\dfrac{N_{pp}(\leq n,p^d)}{N_p(\leq

1569: n,p^d)}\leq\dfrac{(p-1)!}{p^{p-1}}\left(\dfrac{p-1}{p}\right)^n$.

1570: \end{enumerate}

1571: \end{theorem}

1572: \begin{proof}

1573: When $3\leq n\geq 2p-2$, the matrix in the proof of Theorem

1574: \ref{theorem:count-bijective-poly-pd} has at most $n$ free

1575: congruences and other $2p-1-n$ congruences are actually linear

1576: combinations of the $n$ free ones. This means that there exists an

1577: upper bound of $\frac{N_{pp}(\leq n,p^d)}{N_p(\leq n,p^d)}$. Note

1578: that the values of $b_1,\cdots,b_{p-1}$ should form a permutation

1579: over $\{1,\cdots,p-1\}$, so we consider the following two

1580: conditions, respectively.

1581: 

1582: 1) When $3\leq n\leq p$, $b_p$ has $(p-1)$ possible values and the

1583: $n-1$ left free variables have $P(p-1,n-1)=(n-1)!\binom{p-1}{n-1}$

1584: combinations, so $\dfrac{N_{pp}(\leq n,p^d)}{N_p(\leq

1585: n,p^d)}\leq\dfrac{(p-1)P(p-1,n-1)}{p^n}=\dfrac{(p-1)(n-1)!\binom{p-1}{n-1}}{p^n}$.

1586: 

1587: 2) When $p+1\leq n\leq 2p-2$, $p-1$ free variables form the

1588: permutation over $\{1,\cdots,p-1\}$ and other $n-(p-1)$ variables

1589: are totally free, so the number of possibilities of the $n$ free

1590: variables is $(p-1)!(p-1)^{n-(p-1)}$. Thus, $\dfrac{N_{pp}(\leq

1591: n,p^d)}{N_p(\leq

1592: n,p^d)}\leq\dfrac{(p-1)!(p-1)^{n-(p-1)}}{p^n}=\dfrac{(p-1)!}{p^{p-1}}\left(\dfrac{p-1}{p}\right)^n$.

1593: \end{proof}

1594: 

1595: \begin{theorem}

1596: For any prime $p$ and $3\leq n\leq p-2$, the following inequality

1597: hold:

1598: \[

1599: \dfrac{N_{pp}(\leq n,p)}{N_p(\leq

1600: n,p)}\leq\dfrac{P(p-1,n-1)}{p^n}=\dfrac{(n-1)!\binom{p-1}{n-1}}{p^n}.

1601: \]

1602: \end{theorem}

1603: \begin{proof}

1604: When $d=1$, the second group of necessary and sufficient conditions

1605: disappear. Then, following the similar idea of proving the above

1606: theorem, this theorem is proved.

1607: \end{proof}

1608: 

1609: \fproblem{When $3\leq n\leq 2p-2$, it is still possible to get a

1610: close form of the \textbf{exact} value of $\dfrac{N_{pp}(\leq

1611: n,p^d)}{N_p(\leq n,p^d)}$?}

1612: 

1613: \begin{example}

1614: When $d\geq 2$,

1615: \[

1616: \frac{N_{pp}(\leq n,2^d)}{N_p(\leq n,2^d)}=1/2^{\min(n,3)}=\begin{cases}%

1617: 1/2, & n=1,\\

1618: 1/2^2, & n=2,\\

1619: 1/2^3, & n\geq 3.

1620: \end{cases}

1621: \]

1622: \end{example}

1623: \begin{solution}

1624: When $n=1$, from Lemma \ref{theorem:bijection-degree1}, a polynomial

1625: is a permutation polynomial modulo $2^d$ if and only if

1626: $\gcd(a_1,2^d)=1$. So, $a_1\equiv 1\pmod 2$, which means

1627: $\frac{N_{pp}(\leq 1,2^d)}{N_p(\leq 1,2^d)}=1/2$.

1628: 

1629: When $n=2$, assume $f(x)=a_2x^2+a_1x$. From Theorem

1630: \ref{theorem:bijective-poly-NSC-pd-degree2}, the necessary and

1631: sufficient conditions are $a_1\equiv 1\pmod 2$ and $a_2\equiv 0\pmod

1632: 2$. Thus, $\frac{N_{pp}(\leq 2,2^d)}{N_p(\leq 2,2^d)}=1/2^2$.

1633: 

1634: When $n\geq 2p-1=3$, from Theorem

1635: \ref{theorem:count-bijective-poly-pd}, one has $\frac{N_{pp}(\leq

1636: n,2^d)}{N_p(\leq n,2^d)}=(2-1)^3\cdot(2-1)!/2^{2\cdot 2-1}=1/2^3$.

1637: 

1638: Computer experiments have been made to verify the above results.

1639: \end{solution}

1640: 

1641: \begin{example}

1642: When $d\geq 2$,

1643: \[

1644: \frac{N_{pp}(\leq n,3^d)}{N_p(\leq n,3^d)}=\begin{cases}%

1645: 2/3, & n=1,\\

1646: 2/3^2, & n=2,\\

1647: 4/3^n, & n=3,4,\\

1648: 16/3^5, & n\geq 5.

1649: \end{cases}

1650: \]

1651: \end{example}

1652: \begin{solution}

1653: When $n=1$, from Lemma \ref{theorem:bijection-degree1}, a polynomial

1654: is a permutation polynomial modulo $3^d$ if and only if

1655: $\gcd(a_1,3^d)=1$. So, $a_1\equiv 1,2\pmod 3$, which means

1656: $\frac{N_{pp}(\leq 1,3^d)}{N_p(\leq 1,3^d)}=2/3$.

1657: 

1658: When $n=2$, assume $f(x)=a_2x^2+a_1x$. From Theorem

1659: \ref{theorem:bijective-poly-NSC-pd-degree2}, the necessary and

1660: sufficient conditions are $a_1\not\equiv 0\pmod 3$ and $a_2\equiv

1661: 0\pmod 3$. This means that $\frac{N_{pp}(\leq 2,3^d)}{N_p(\leq

1662: 2,3^d)}=2/3^2$.

1663: 

1664: When $n=3$, from Theorem \ref{theorem:bijective-poly-NSC-pd}, the

1665: necessary and sufficient conditions are as follows:

1666: $a_1+a_2+a_3\not\equiv 0\pmod 3$, $2a_1+4a_2+8a_3\not\equiv 0\pmod

1667: 3$, $(2-1)a_1+(4-1)a_2+(8-1)a_3\not\equiv 0\pmod 3$, $a_1\not\equiv

1668: 0\pmod 3$, $a_1+2a_2+3a_3\not\equiv 0\pmod 3$ and

1669: $a_1+4a_2+12a_3\not\equiv 0\pmod 3$. These conditions can be further

1670: simplified as $a_1\not\equiv 0\pmod 3$, $a_2\equiv 0\pmod 3$ and

1671: $a_1+a_3\not\equiv 0\pmod 3$. So, the possible values of

1672: $(a_1,a_2,a_3)$ modulo 3 are $(1,0,0)$, $(1,0,1)$, $(2,0,0)$ and

1673: $(2,0,2)$. Thus, $\frac{N_{pp}(\leq 3,3^d)}{N_p(\leq 3,3^d)}=4/3^3$.

1674: In the same way, one can deduce the results when $n=4$.

1675: 

1676: When $n\geq 2p-1=5$, from Theorem

1677: \ref{theorem:count-bijective-poly-pd}, one has $\frac{N_{pp}(\leq

1678: n,3^d)}{N_p(\leq n,3^d)}=(3-1)^3\cdot(3-1)!/3^{2\cdot 3-1}=16/3^5$.

1679: 

1680: Computer experiments have been made to verify the above results.

1681: \end{solution}

1682: \iffalse Some experimental data on the number of permutation

1683: polynomials are listed in Table \ref{table:counting-power-p}. One

1684: can see that the theoretical results agree well with the

1685: experimental data.\fi

1686: 

1687: After getting the number of permutation polynomials of degree $\leq

1688: n$ modulo $p^d$, one can easily calculate the number of distinct

1689: permutations induced by the permutation polynomials of degree $\leq

1690: n$ modulo $p^d$, by using Lemma \ref{lemma:Equ-Poly-Null-Poly} and

1691: the results on null polynomials modulo $p^d$ given in

1692: \cite{Li:NullPoly2005}. We have the following theorem.

1693: 

1694: \begin{theorem}

1695: Assume $p$ is a prime, $d\geq 1$ and $N_{np}(\leq n,p^d)$ denotes

1696: the number of null polynomials of degree $\leq n$ modulo $p^d$.

1697: Then, the number of distinct permutations induced from polynomials

1698: of degree $\leq n$ modulo $p^d$ is $N_{pp}(\leq n,p^d)/N_{np}(\leq

1699: n,p^d)$.

1700: \end{theorem}

1701: \begin{proof}

1702: It is obvious since each polynomial has $N_{np}(\leq n,p^d)$

1703: equivalent polynomials from Lemma \ref{lemma:Equ-Poly-Null-Poly}.

1704: \end{proof}

1705: 

1706: \begin{remark}

1707: Note that Corollary 4.1 of \cite{KO:CountingPF:DMJ1968} gives a

1708: different proof of Theorem \ref{theorem:count-bijective-poly-pd}.

1709: However, \cite{KO:CountingPF:DMJ1968} mainly focuses on the total

1710: number of non-equivalent polynomial functions of arbitrary degree

1711: modulo $p^d$ and does not study the case when the degree is also

1712: given. So, the results given in this paper are more complete.

1713: \end{remark}

1714: 

1715: \subsection{Determining (Permutation) Polynomials from Induced Bijection}

1716: 

1717: In this subsection, we study the problem of determining all

1718: equivalent polynomials when the induced polynomial function is (or

1719: partially) known. Note that the following results are also valid for

1720: polynomials that induce any polynomial functions (maybe not

1721: bijections).

1722: 

1723: Following Lemma \ref{lemma:Equ-Poly-Null-Poly} and the results

1724: obtained in \cite{Li:NullPoly2005}, once we get one permutation

1725: polynomials inducing the given bijection, we can determine all

1726: equivalent permutation polynomials. So, it is sufficient to derive

1727: only one equivalent polynomial as a seed.

1728: 

1729: \begin{theorem}\label{theorem:solving-poly-small-n}

1730: Assume $p$ is a prime, $d\geq 1$ and $f(x)=a_nx^n+\cdots+a_1x+a_0$

1731: is a polynomial of degree $n\leq p-1$ modulo $p^d$. Given

1732: $x_0,\cdots,x_n\in\mathbb{Z}$, if $\forall i,j\in\{0,\cdots,n\}$ and

1733: $i\neq j$, $x_i\not\equiv x_j\pmod p$, then $f(x)$ can be uniquely

1734: determined by solving the following system of congruence:

1735: \begin{equation}

1736: \left[\begin{matrix}%

1737: 1 & x_0 & x_0^2 & \cdots & x_0^n\\

1738: 1 & x_1 & x_1^2 & \cdots & x_1^n\\

1739: \vdots & \vdots & \vdots & \ddots & \vdots\\

1740: 1 & x_n & x_n^2 & \cdots & x_n^n

1741: \end{matrix}\right]%

1742: \left[\begin{matrix}%

1743: a_0\\a_1\\\vdots\\a_n

1744: \end{matrix}\right]\equiv

1745: \left[\begin{matrix}%

1746: f(x_0)\\f(x_1)\\\vdots\\f(x_n)

1747: \end{matrix}\right]\pmod{p^d}.\label{equation:solve-PP-nLessp}

1748: \end{equation}

1749: \end{theorem}

1750: \begin{proof}

1751: Denote the system of congruences by

1752: $\bm{A}_x\bm{X}_a\equiv\bm{B}_f\pmod{p^d}$. Since $\bm{A}_x$ is a

1753: Vondermonde matrix, $|\bm{A}_x|=\prod_{0\leq i<j\leq n}(x_j-x_i)$.

1754: From $x_i\not\equiv x_j\pmod p$, $\gcd(x_j-x_i,p)=1$, so

1755: $\gcd(|\bm{A}_x|,p^d)=1$. Thus, the system of congruences has a

1756: unique set of incongruent solutions and this theorem is proved.

1757: \end{proof}

1758: \begin{remark}

1759: When $n=p-1$, it is obvious that $x_0,\cdots,x_n$ form a complete

1760: system of residues modulo $p$. When $n<p-1$, $x_0,\cdots,x_n$ form

1761: an incomplete system of residues modulo $p$. The simplest choice of

1762: the $n+1$ values is: $\{x_i=i\}_{i=0}^n=\{0,\cdots,n\}$.

1763: \end{remark}

1764: \begin{corollary}\label{corollary:solving-poly-small-n}

1765: Assume $p$ is a prime, $d\geq 1$ and $f(x)=a_nx^n+\cdots+a_1x$ is a

1766: polynomial of degree $n\leq p-1$ modulo $p^d$. Given

1767: $x_1,\cdots,x_n\not\equiv 0\pmod p$, if $\forall

1768: i,j\in\{1,\cdots,n\}$ and $i\neq j$, $x_i\not\equiv x_j\pmod p$,

1769: then $f(x)$ can be uniquely determined by solving the following

1770: system of congruence:

1771: \begin{equation}

1772: \left[\begin{matrix}%

1773: x_1 & x_1^2 & \cdots & x_1^n\\

1774: x_2 & x_2^2 & \cdots & x_2^n\\

1775: \vdots & \vdots & \ddots & \vdots\\

1776: x_n & x_n^2 & \cdots & x_n^n

1777: \end{matrix}\right]%

1778: \left[\begin{matrix}%

1779: a_1\\a_2\\\vdots\\a_n

1780: \end{matrix}\right]\equiv

1781: \left[\begin{matrix}%

1782: f(x_1)\\f(x_2)\\\vdots\\f(x_n)

1783: \end{matrix}\right]\pmod{p^d}.

1784: \end{equation}

1785: \end{corollary}

1786: \begin{proof}

1787: This corollary is a special case of Theorem

1788: \ref{theorem:solving-poly-small-n}. For $i=1\sim n$, factoring out

1789: $x_i$ from row $i$ of the matrix, one immediately has

1790: $|\bm{A}|=\prod_{i=1}^nx_i\prod_{1\leq i<j\leq n}(x_j-x_i)$. From

1791: the conditions of $\{x_i\}_{i=1}^n$, $|\bm{A}|$ is relatively prime

1792: to $p$ and the system of congruences has a unique set of incongruent

1793: solutions, thus this corollary is proved.

1794: \end{proof}

1795: 

1796: When $n\geq p$ or the value of $n$ is unknown, the above method

1797: cannot be directly used to determine polynomials that induce the

1798: given polynomial function. If we can find a way to reduce the degree

1799: of polynomials, then the above method can be employed to determine

1800: the coefficients. In the following, we give a way to achieve this

1801: task.

1802: 

1803: \begin{lemma}\label{lemma:determine-PP-dLessp}

1804: Assume $p$ is a prime and $2\leq d\leq p$. If

1805: $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial of degree $n\leq pd-1$

1806: modulo $p^d$, then all its equivalent polynomials $\leq pd-1$ modulo

1807: $p^d$ can be determined from the induced polynomial function over

1808: $\{0,\cdots,p^d-1\}$.

1809: \end{lemma}

1810: \begin{proof}

1811: Choosing $x=py_1+b_0$, i.e., $y_1=\lfloor x/y\rfloor$ and

1812: $b_0=(x\bmod p)\in\{0,\cdots,p-1\}$, we have $p$ sub-polynomials:

1813: \[

1814: f_{b_0}(y_1)=\begin{cases}%

1815: \sum_{k=n}^0p^ka_ky_1^k, & \mbox{when }b_0=0,\\

1816: \sum_{k=n}^0\left(\sum_{l=n}^ka_l\binom{l}{k}p^ky_1^kb_0^{l-k}\right),

1817: & \mbox{when }b_0\in\{1,\cdots,p-1\}.

1818: \end{cases}

1819: \]

1820: Apparently, they have a uniform form:

1821: $f_{b_0}(y_1)=\sum_{k=n}^0p^ka_k^{(b_0)}y_1^k$, where

1822: $a_k^{(0)}=a_k$ and

1823: $a_k^{(b_0)}=\sum_{l=n}^ka_l\binom{l}{k}b_0^{l-k}$ when

1824: $b_0\in\{1,\cdots,p-1\}$. Note that $f_{b_0}(y_1)\equiv

1825: f_{b_0}^*(y_1)=\sum_{k=d-1}^0p^ka_k^{(b_0)}y_1^k\pmod{p^d}$, so we

1826: can focus on $f_{b_0}^*(y_1)$ only. Since $d\leq p$, each

1827: sub-polynomial $f_{b_0}^*(y_1)$ is of degree less than $p$ modulo

1828: $p^d$, so all the coefficients can be uniquely solved modulo $p^d$,

1829: i.e. the value of each $a_k^{(b_0)}$, can be uniquely solved modulo

1830: $p^{d-k}$. It is obvious that the $p^k$ distinct valid values of

1831: $a_k^{(b_0)}$ modulo $p^d$ are equivalent for $f_{b_0}(y_1)$ and so

1832: equivalent for $f(x)$.

1833: 

1834: For each set of the valid values of all coefficients modulo $p^d$,

1835: one has a system of congruences in the form

1836: $\bm{A}\bm{X}_a\equiv\bm{B}\pmod{p^d}$:

1837: \begin{equation}

1838: \left[\begin{matrix}%

1839: \bm{A}_0\\\bm{A}_1\\\vdots\\\bm{A}_{p-1}

1840: \end{matrix}\right]

1841: \left[\begin{matrix}%

1842: a_0\\a_1\\\vdots\\a_{pd-1}

1843: \end{matrix}\right]\equiv

1844: \left[\begin{matrix}%

1845: \bm{B}_0\\\bm{B}_1\\\vdots\\\bm{B}_{p-1}

1846: \end{matrix}\right]\pmod{p^d},\label{equation:solve-PP-pd}

1847: \end{equation}

1848: where

1849: \[

1850: \bm{A}_0=\left[\begin{matrix}%

1851: \bm{I}_{d\times d} & \bm{0}_{d \times d(p-1)}\end{matrix}\right]

1852: =\left[\begin{array}{ccccc:ccccc}%

1853: 1 & 0 & 0 & \cdots & 0 & 0 & 0 & 0 & \cdots & 0\\

1854: 0 & 1 & 0 & \cdots & 0 & 0 & 0 & 0 & \cdots & 0\\

1855: 0 & 0 & 1 & \cdots & 0 & 0 & 0 & 0 & \cdots & 0\\

1856: \vdots & \vdots & \vdots & \ddots & \vdots & \vdots & \vdots &

1857: \vdots & \ddots & \vdots\\

1858: 0 & 0 & 0 & \cdots & 1 & 0 & 0 & 0 & \cdots & 0

1859: \end{array}\right]_{d\times pd},

1860: \]

1861: for $b_0=1\sim p-1$,

1862: \[

1863: \bm{A}_{b_0}=\left[\begin{matrix}%

1864: \bm{A}_{b_0}^{(L)} & \bm{A}_{b_0}^{(R)}\end{matrix}\right]=

1865: \left[\begin{array}{ccccc:ccc}%

1866: 1 & b_0 & b_0^2 & \cdots & b_0^{d-1} & b_0^d & \cdots & b_0^{pd}\\

1867: 0 & 1 & 2b_0 & \cdots & (d-1)b_0^{d-2} & db_0^{d-1} & \cdots & (pd-1)b_0^{pd-1}\\

1868: 0 & 0 & \binom{2}{2} & \cdots & \binom{d-1}{2}b_0^{d-3} &

1869: \binom{d}{2}b_0^{d-2} & \cdots & \binom{pd-1}{2}b_0^{pd-2}\\

1870: \vdots & \vdots & \vdots & \ddots & \vdots & \vdots & \ddots & \vdots\\

1871: 0 & 0 & 0 & \cdots & \binom{d-1}{d-1} & \binom{d}{d-1}b_0 & \cdots &

1872: \binom{pd-1}{d-1}b_0^{pd-(d-1)}

1873: \end{array}\right]_{d\times pd},

1874: \]

1875: and for $b_0=0\sim p-1$,

1876: $\bm{B}_{b_0}=\left[\begin{matrix}a_0^{(b_0)} & a_1^{(b_0)} &

1877: a_{d-1}^{(b_0)}\end{matrix}\right]^T$. From Lemma

1878: \ref{lemma:det-binom-powers}, one can see $|\bm{A}|$ is relatively

1879: prime to $p$. Thus, for each valid set of the values of the

1880: coefficients $\left\{a_k^{(b_0)}\right\}_{0\leq k\leq p-1 \atop

1881: 0\leq b_0\leq p-1}$, the above system of congruences has a unique

1882: set of solutions modulo $p^d$. One can easily verify that each set

1883: of solutions corresponds to an equivalent of the polynomial $f(x)$.

1884: Thus this lemma is proved.

1885: \end{proof}

1886: \begin{remark}

1887: In fact, in the proof of the above lemma, we can also calculate the

1888: number of equivalent polynomials of $f(x)$ of degree $\leq pd-1$

1889: modulo $p^d$. It is $p^{(1+\cdots+(d-1))p}=p^{\frac{d(d-1)p}{2}}$.

1890: From Lemma \ref{lemma:Equ-Poly-Null-Poly}, this number should be

1891: equal to the number of null polynomials of degree $\leq pd-1$ modulo

1892: $p^d$. Clearly, $p^{\frac{d(d-1)p}{2}}$ agrees with the results

1893: (Lemma 34 and Theorem 43) obtained in \cite{Li:NullPoly2005}.

1894: \end{remark}

1895: 

1896: \begin{corollary}

1897: Assume $p$ is a prime and $2\leq d\leq p$. If

1898: $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial modulo $p^d$, then all

1899: its equivalent polynomials modulo $p^d$ an be determined from the

1900: induced polynomial function over $\{0,\cdots,p^d-1\}$.

1901: \end{corollary}

1902: \begin{proof}

1903: When $n\geq pd$, one can move $a_{pd},\cdots,a_n$ to the right side

1904: of the matrix in the proof of the above lemma. Then,

1905: $a_{pd},\cdots,a_n$ become free variables, so each set of their

1906: values corresponds to $p^{\frac{d(d-1)p}{2}}$ equivalent polynomials

1907: modulo $p^d$. That is, in total we have

1908: $p^{d(n-pd+1)\frac{d(d-1)p}{2}}$ equivalent polynomials. Thus this

1909: corollary is also true.

1910: \end{proof}

1911: 

1912: \begin{theorem}

1913: Assume $p$ is a prime and $d\geq 2$. If

1914: $f(x)=a_nx^n+\cdots+a_1x+a_0$ is a polynomial modulo $p^d$, then all

1915: its equivalent polynomials modulo $p^d$ an be determined from the

1916: induced polynomial function over $\{0,\cdots,p^d-1\}$.

1917: \end{theorem}

1918: \begin{proof}

1919: We use induction on $d$ to prove this theorem. The case of $2\leq

1920: d\leq p$ has been proved above. Let us prove the case of $d>p$ under

1921: the assumption that this theorem is true for any integer less than

1922: $d$.

1923: 

1924: Using the same way in the proof of Lemma

1925: \ref{lemma:determine-PP-dLessp}, we can get $p$ sub-polynomials

1926: $f_{b_0}^*(y_1)$, which uniquely determine the induced bijection.

1927: Since the degree of $f_{b_0}^*(y_1)$ modulo $p^d$ is not less than

1928: $p$, the coefficients cannot be uniquely solved, let us try to

1929: further decompose each sub-polynomial in the same way.

1930: 

1931: At first, note that the value of $a_0^{(b_0)}$ modulo $p^d$ can

1932: always be uniquely solved by choosing $y_1\equiv 0\pmod{p^d}$. Then,

1933: for $b_0\in\{1,\cdots,p-1\}$, subtracting $a_0^{(b_0)}$ from

1934: $f_{b_0}^*(y_1)$, one has

1935: $f_{b_0}^*(y_1)-a_0^{(b_0)}=p\left(\sum_{k=d-1}^1p^{k-1}a_k^{(b_0)}y_1^k\right)=pf_{b_0}^{**}(y_1)$.

1936: For $b_0=0$, one can make the similar operation to get

1937: $f_{0}(x)-a_0^{(0)}=p\left(\sum_{k=d-1}^1p^{k-1}a_ky_1^k\right)=pf_0^{**}(y_1)$.

1938: Apparently, $f_{b_0}(x)$ is uniquely determined by $a_0^{(b_0)}$

1939: modulo $p^d$ and $f_{b_0}^{**}(y_1)$ modulo $p^{d-1}$. Applying the

1940: hypothesis on $f_{b_0}^{**}(y_1)$, all equivalent polynomials of

1941: $f_{b_0}^{**}(y_1)$ can be determined modulo $p^{d-1}$. Then, with

1942: each valid\footnote{Note that not all equivalent polynomials of

1943: $f_{b_0}^{**}(y_1)$ are valid, due to the existence of some power of

1944: $p$ in each coefficient.} equivalent polynomial of

1945: $f_{b_0}^{**}(y_1)$ and the value of $a_0^{(b_0)}$, one can further

1946: uniquely determine all coefficients of $f(x)$ modulo $p^d$ in the

1947: same way given in the proof of Lemma

1948: \ref{lemma:determine-PP-dLessp}. Thus this theorem is proved.

1949: \end{proof}

1950: 

1951: The above theorem tells us that all equivalent polynomials that

1952: induce a given polynomial function modulo $p^d$ can be determined

1953: via a recursive manner. Considering the complexity of solving Eq.

1954: (\ref{equation:solve-PP-nLessp}) is $O(p^3)$ and the complexity of

1955: solving Eq. (\ref{equation:solve-PP-pd}) is $O((pd)^3)$, the total

1956: complexity of deriving one equivalent polynomial via the recursive

1957: procedure is

1958: \begin{equation}

1959: O\left((pd)^3+(p(d-1))^3p+\cdots+\left(p\left(d-(d-p))\right)^3p^{d-p}\right)+p^3p^{d-p+1}\right)=O\left(p^{d-p+6}\right).

1960: \label{equation:solve-PP-complexity1}

1961: \end{equation}

1962: In fact, this complexity can be further reduced, due to the

1963: existence of a power of $p$ in each coefficient (except

1964: $a_0^{(b_0)}$) of each sub-polynomial $f_{b_0}^*(y_1)$, which will

1965: make more coefficients disappear as the value of $d$ decreases. In

1966: the following, let us study what will happen when the sub-polynomial

1967: and its derivatives are further decomposed.

1968: 

1969: At first, let us see the decomposition of the $p$ sub-polynomials

1970: $\left\{f_{b_0}^*(y_1)=\sum_{k=d-1}^1p^{k-1}y_1^k\right\}_{b_0=0}^{p-1}$.

1971: Similarly, choosing $y_1=py_2+b_1$, i.e., $y_2=\lfloor

1972: y_1/p\rfloor=\lfloor x/p^2\rfloor$ and $b_1=(y_1\bmod

1973: p)\in\{0,\cdots,p-1\}$, we have $p^2$ sub-polynomials as follows:

1974: $\forall b_0,b_1\in\{0,\cdots,p-1\}$,

1975: \[

1976: f_{b_1,b_0}^*(y_2)=\begin{cases}%

1977: \sum_{k=d-1}^1p^{2k-1}a_k^{(b_0)}y_2^k, & \mbox{when }b_1=0,\\

1978: \sum_{k=d-1}^1\left(\sum_{l=d-1}^kp^{l-1}a_l^{(b_0)}\binom{l}{k}b_1^{l-k}p^ky_2^k\right)+\sum_{l=d-1}^1p^{l-1}a_l^{(b_0)},

1979: & \mbox{when }b_1\in\{1,\cdots,p-1\}.

1980: \end{cases}

1981: \]

1982: The above polynomial can be rewritten in the following form:

1983: \[

1984: f_{b_1,b_0}^*(y_2)=\begin{cases}%

1985: \sum_{k=d-1}^1p^{2k-1}a_k^{(b_1,b_0)}y_2^k, & \mbox{when }b_1=0,\\

1986: \sum_{k=d-1}^1p^{2k-1}a_k^{(b_1,b_0)}y_2^k+a_0^{(b_1,b_0)}, &

1987: \mbox{when }b_1\in\{1,\cdots,p-1\},

1988: \end{cases}

1989: \]

1990: where $a_k^{(0,b_0)}=a_k^{(b_0)}$ and

1991: $a_k^{(b_1,b_0)}=\sum_{l=d-1}^kp^{l-k}a_l^{(b_0)}\binom{l}{k}b_1^{l-k}$

1992: for $1\leq k\leq d-1$ and $1\leq b_1\leq p-1$. Then, solving

1993: $a_0^{(b_1,b_0)}$ and subtracting it from the involved polynomial,

1994: one can get $p^2$ polynomials modulo $p^{d-2}$ as follows:

1995: \[

1996: \left\{f_{b_1,b_0}^{**}(y_2)=\sum_{k=d-2}^1p^{2(k-1)}a_k^{(b_1,b_0)}y_2^k\right\}_{0\leq

1997: b_0,b_1\leq p-1}.

1998: \]

1999: 

2000: Repeat the above procedure for $i$ times, where $1\leq i\leq d-2$,

2001: one can get $p^{i+1}$ polynomials modulo $p^{d-i}$: $\forall

2002: b_0,\cdots,b_i\in\{0,\cdots,p-1\}$,

2003: \[

2004: f_{b_i,\cdots,b_0}^*(y_{i+1})=\begin{cases}%

2005: \sum_{k=d-i}^1p^{i(k-1)+1}a_k^{(b_i,\cdots,b_0)}y_{i+1}^k, & \mbox{when }b_i=0,\\

2006: \sum_{k=d-i}^1p^{i(k-1)+1}a_k^{(b_i,\cdots,b_0)}y_{i+1}^k+a_0^{(b_i,\cdots,b_0)},

2007: & \mbox{when }b_i\in\{1,\cdots,p-1\},

2008: \end{cases}

2009: \]

2010: and $p^{i+1}$ polynomials modulo $p^{d-i-1}$

2011: \[

2012: \left\{f_{b_i,\cdots,b_0}^{**}(y_{i+1})=\sum_{k=d-i-1}^1p^{i(k-1)}a_k^{(b_i,\cdots,b_0)}y_{i+1}^k\right\}_{0\leq

2013: b_0,\cdots,b_i\leq p-1},

2014: \]

2015: where $y_{i+1}=\lfloor y_i/p\rfloor=\lfloor x/p^{i+1}\rfloor$ and

2016: $b_i=(y_i\bmod p)\in\{0,\cdots,p-1\}$. Observing the above

2017: polynomials, one can see that some higher coefficients disappear

2018: modulo $p^{d-i}$ or modulo $p^{d-i-1}$ due to the existence of

2019: powers of $p$. Assume $p^{i(k-1)+1}<p^{d-i}$ or

2020: $p^{i(k-1)}<p^{d-i-1}$, one has $k<\frac{d-1}{i}$, so

2021: $a_k^{(b_i,\cdots,b_0)}$ is valid only when

2022: $k\leq\left\lceil\frac{d-1}{i}\right\rceil-1$. When

2023: $i\geq\left\lceil\frac{d-1}{p}\right\rceil$, one has

2024: $i\geq\frac{d-1}{p}\Rightarrow\frac{d-1}{i}\leq

2025: p\Rightarrow\left\lceil\frac{d-1}{i}\right\rceil\leq p\Rightarrow

2026: \left\lceil\frac{d-1}{i}\right\rceil-1\leq p-1$, so the coefficients

2027: of $f_{b_i,\cdots,b_0}^*(y_{i+1})$ can be uniquely determined. This

2028: means that the complexity of deriving one equivalent polynomial via

2029: the procedure is reduced to be

2030: \[

2031: O\left((pd)^3+(p(d-1))^3p+\cdots+

2032: \left(p\left(d-\left\lceil\frac{d-1}{p}\right\rceil\right)\right)^3p^{\left\lceil\frac{d-1}{p}\right\rceil}

2033: +p^3p^{\left\lceil\frac{d-1}{p}\right\rceil+1}\right)\approx

2034: O\left(d^3p^{\left\lceil\frac{d-1}{p}\right\rceil+3}\right),

2035: \]

2036: which is much smaller than Eq. (\ref{equation:solve-PP-complexity1})

2037: when $d\gg p$.

2038: 

2039: \begin{remark}

2040: From Theorem \ref{theorem:solving-poly-small-n}, solving the

2041: coefficients of a polynomial of degree $\leq p$ needs only $p$

2042: input-output values of the polynomial function. This means that it

2043: is still possible to determine the polynomials if the induced

2044: polynomial function is partially known. For the polynomials of

2045: arbitrary degree modulo $p^d$, only

2046: $p^{\left\lceil\frac{d-1}{p}\right\rceil+1}$ input-output values are

2047: needed. Note that the needed input-output values should satisfy some

2048: certain distribution modulo $p^d$, so the number of required

2049: input-output values will be larger if the values are observed in a

2050: random process\footnote{For example, if they are collected in a

2051: known-plaintext attack to an encryption procedure based on a

2052: permutation polynomials.}.

2053: \end{remark}

2054: 

2055: \iffalse

2056: \section{Conclusions}

2057: 

2058: This paper studies some typical problems about permutation

2059: polynomials modulo $m$. Some problems have not been solved and open

2060: for future research, especially about enumerating equivalent

2061: polynomials of a given permutation polynomial modulo $p^d$ and

2062: counting the number of bijections induced from permutation

2063: polynomials modulo $p^d$. \fi

2064: 

2065: %\section*{Acknowledgments}

2066: 

2067: \bibliographystyle{unsrt}

2068: \bibliography{BPC}

2069: 

2070: \iffalse

2071: \section*{Appendix}

2072: 

2073: In the appendix, we give some experimental data on the number of

2074: permutation polynomials modulo $p^d$.

2075: 

2076: \begin{longtable}[c]{c|c|c|c|c|c}

2077: \caption{The number of all permutation polynomials of degree $\leq

2078: n$

2079: modulo $2^d$ (exhaustively enumerated via a computer program).}\label{table:counting-power2}\\

2080: 

2081: \hline\hline \multicolumn{1}{c|}{$m=2^d$} & \multicolumn{1}{c|}{$n$}

2082: & \multicolumn{1}{c|}{$N_b$} & \multicolumn{1}{c|}{$N_b'$}

2083: & \multicolumn{1}{c|}{$N=m^n=2^{nd}$} & \multicolumn{1}{c}{$N_b/N$}\\

2084: \hline\hline

2085: \endfirsthead

2086: 

2087: \multicolumn{6}{c}%

2088: {{\bfseries \tablename\ \thetable{} -- continued from previous page}} \\

2089: \hline\hline \multicolumn{1}{c|}{$m=2^d$} & \multicolumn{1}{c|}{$n$}

2090: & \multicolumn{1}{c|}{$N_b$} & \multicolumn{1}{c|}{$N_b'$}

2091: & \multicolumn{1}{c|}{$N=m^n=2^{nd}$} & \multicolumn{1}{c}{$N_b/N$}\\

2092: \hline

2093: \endhead

2094: 

2095: \hline\multicolumn{6}{r}{{Continued on next page}} \\

2096: \hline\hline

2097: \endfoot

2098: 

2099: \hline\hline

2100: \endlastfoot

2101: 

2102: $2^1=2$ & 1 & 1 & 1 & 2 & 1/2\\

2103: \cline{2-6} & 2 & 2 & 2 & $2^2$ & $1/2$\\

2104: \cline{2-6} & 3 & 4 & 4 & $2^3$ & $1/2$\\

2105: \cline{2-6} & 4 & 8 & 8 & $2^4$ & $1/2$\\

2106: \cline{2-6} & 5 & 16 & 16 & $2^5$ & $1/2$\\

2107: \hline $2^2=4$ & 1 & 2 & 2 & 4 & $1/2$\\

2108: \cline{2-6} & 2 & 4 & 4 & $4^2$ & $1/2^2$\\

2109: \cline{2-6} & 3 & 8 & 8 & $4^3$ & $1/2^3$\\

2110: \cline{2-6} & 4 & 32 & 32 & $4^4$ & $1/2^3$\\

2111: \cline{2-6} & 5 & 128 & 128 & $4^5$ & $1/2^3$\\

2112: \hline $2^3=8$ & 1 & 4 & 4 & 8 & $1/2$\\

2113: \cline{2-6} & 2 & 16 & 16 & $8^2$ & $1/2^2$\\

2114: \cline{2-6} & 3 & 64 & 64 & $8^3$ & $1/2^3$\\

2115: \cline{2-6} & 4 & 512 & 512 & $8^4$ & $1/2^3$\\

2116: \cline{2-6} & 5 & 4096 & 4096 & $8^5$ & $1/2^3$\\

2117: \hline $2^4=16$ & 1 & 8 & 8 & 16 & $1/2$\\

2118: \cline{2-6} & 2 & 64 & 64 & $16^2$ & $1/2^2$\\

2119: \cline{2-6} & 3 & 512 & 512 & $16^3$ & $1/2^3$\\

2120: \cline{2-6} & 4 & 8192 & 8192 & $16^4$ & $1/2^3$\\

2121: \cline{2-6} & 5 & 131072 & 131072 & $16^5$ & $1/2^3$\\

2122: \hline $2^5=32$ & 1 & 16 & 16 & 32 & $1/2$\\

2123: \cline{2-6} & 2 & 256 & 256 & $32^2$ & $1/2^2$\\

2124: \cline{2-6} & 3 & 4096 & 4096 & $32^3$ & $1/2^3$\\

2125: \cline{2-6} & 4 & 131072 & 131072 & $32^4$ & $1/2^3$\\

2126: \cline{2-6} & 5 & 4194304 & 4194304 & $32^5$ & $1/2^3$\\

2127: \hline $2^6=64$ & 1 & 32 & 32 & 64 & $1/2$\\

2128: \cline{2-6} & 2 & 1024 & 1024 & $64^2$ & $1/2^2$\\

2129: \cline{2-6} & 3 & 32768 & 32768 & $64^3$ & $1/2^3$\\

2130: \cline{2-6} & 4 & 2097152 & 2097152 & $64^4$ & $1/2^3$\\

2131: \cline{2-6} & 5 & 134217728 & 134217728 & $64^5$ & $1/2^3$\\

2132: \hline $2^7=128$ & 1 & 64 & 64 & 128 & $1/2$\\

2133: \cline{2-6} & 2 & 4096 & 4096 & $128^2$ & $1/2^2$\\

2134: \cline{2-6} & 3 & 262144 & 262144 & $128^3$ & $1/2^3$\\

2135: \cline{2-6} & 4 & 33554432 & 33554432 & $128^4$ & $1/2^3$

2136: \end{longtable}

2137: 

2138: \begin{longtable}[c]{c|c|c|c|c}

2139: \caption{The number of real permutation polynomials of degree $\leq

2140: n$ modulo $p^d$): experimental data.}\label{table:counting-power-p}\\

2141: 

2142: \hline \multicolumn{1}{c|}{$q$} & \multicolumn{1}{c|}{$n$} &

2143: \multicolumn{1}{c|}{$N_b$}

2144: & \multicolumn{1}{c|}{$N=q^n$} & \multicolumn{1}{c}{$N_b/N$}\\

2145: \hline\hline

2146: \endfirsthead

2147: 

2148: \multicolumn{5}{c}%

2149: {{\bfseries \tablename\ \thetable{} -- continued from previous page}} \\

2150: \hline \multicolumn{1}{c|}{$q$} & \multicolumn{1}{c|}{$n$} &

2151: \multicolumn{1}{c|}{$N_b$}

2152: & \multicolumn{1}{c|}{$N=q^n$} & \multicolumn{1}{c}{$N_b/N$}\\

2153: \hline\hline

2154: \endhead

2155: 

2156: \hline \multicolumn{5}{r}{{Continued on next page}} \\ \hline

2157: \endfoot

2158: 

2159: \hline

2160: \endlastfoot

2161: 

2162: \multicolumn{5}{l}{$q=3^\alpha$ ($\alpha\geq 1$)}\\

2163: \hline 3 & 1 & 2 & 3 & $2/3$\\

2164: \cline{2-5} & $\geq 2$ & $2\times 3^{n-2}$ & $3^n$ & $2/3^2$\\

2165: \hline $3^2=9$ & 1 & 6 & 9 & $2/3$\\

2166: \cline{2-5} & 2 & 18 & $9^2$ & $2/3^2$\\

2167: \cline{2-5} & 3 & 108 & $9^3$ & $4/3^3$\\

2168: \cline{2-5} & 4 & 324 & $9^4$ & $4/3^4$\\

2169: \cline{2-5} & 5 & 3888 & $9^5$ & $16/3^5$\\

2170: \cline{2-5} & 6 & 34992 & $9^6$ & $16/3^5$\\

2171: \hline $3^3=27$ & 1 & 18 & 27 & $2/3$\\

2172: \cline{2-5} & 2 & 162 & $27^2$ & $2/3^2$\\

2173: \cline{2-5} & 3 & 2916 & $27^3$ & $4/3^3$\\

2174: \cline{2-5} & 4 & 26244 & $27^4$ & $4/3^4$\\

2175: \cline{2-5} & 5 & 944784 & $27^5$ & $16/3^5$\\

2176: \hline $3^4=81$ & 1 & 54 & 81 & $2/3$\\

2177: \cline{2-5} & 2 & 1458 & $81^2$ & $2/3^2$\\

2178: \cline{2-5} & 3 & 78732 & $81^3$ & $4/3^3$\\

2179: \cline{2-5} & 4 & 2125764 & $81^4$ & $4/3^4$\\

2180: \cline{2-5} & 5 & 229582512 & $81^5$ & $16/3^5$\\

2181: \hline $3^5=243$ & 1 & 162 & 243 & $2/3$\\

2182: \cline{2-5} & 2 & 13122 & $243^2$ & $2/3^2$\\

2183: \cline{2-5} & 3 & 2125764 & $243^3$ & $4/3^3$\\

2184: \cline{2-5} & 4 & 172186884 & $243^4$ & $4/3^4$\\

2185: \hdashline random test & 5 & (276632) & (4194304) & $\approx 16/3^5$\\

2186: \iffalse

2187: \hline\hline\multicolumn{5}{l}{$q=5^\alpha$ ($\alpha\geq 1$)}\\

2188: \hline 5 & 1 & 4 & 5 & $4/5$\\

2189: \cline{2-5} & 2 & 4 & $5^2$ & $4/5^2$\\

2190: \cline{2-5} & 3 & 24 & $5^3$ & $24/5^3$\\

2191: \cline{2-5} & $\geq 4$ & $24\times 5^{n-4}$ & $5^n$ & $24/5^4$\\

2192: \hline\hline\multicolumn{5}{l}{$q=7^\alpha$ ($\alpha\geq 1$)}\\

2193: \hline 7 & 1 & 6 & 7 & $6/7$\\

2194: \cline{2-5} & 2 & 6 & $7^2$ & $6/7^2$\\

2195: \cline{2-5} & 3 & 6 & $7^3$ & $6/7^3$\\

2196: \cline{2-5} & 4 & 90 & $7^4$ & $90/7^4$\\

2197: \cline{2-5} & 5 & 720 & $7^5$ & $720/7^5$\\

2198: \cline{2-5} & $\geq 6$ & $720\times 7^{n-6}$ & $7^n$ &

2199: $720/7^6$\\\fi\hline

2200: \end{longtable}

2201: \fi

2202: 

2203: \end{document}

2204: