Criteria of ergodicity for
p
-adic dynamical systems in terms
of coordinate functions
Andrei Khrennikov
⇑, Ekaterina Yurova
International Center for Mathematical Modelling in Physics and Cognitive Sciences, Linnaeus University, Växjö S-35195, Sweden
a r t i c l e
i n f o
Article history:
Received 25 April 2013 Accepted 6 January 2014 Available online 3 February 2014
a b s t r a c t
This paper is devoted to the problem of ergodicity ofp-adic dynamical systems. We solved the problem of characterization of ergodicity and measure preserving for (discrete)p-adic dynamical systems for arbitrary primepfor iterations based on 1-Lipschitz functions. This problem was open since long time and only the casep¼2 was investigated in details. We formulated the criteria of ergodicity and measure preserving in terms of coordinate func-tions corresponding to digits in the canonical expansion ofp-adic numbers. (The coordinate representation can be useful, e.g., for applications to cryptography.) Moreover, by using this representation we can consider non-smoothp-adic transformations. The basic technical tools are van der Put series and usage of algebraic structure (permutations) induced by coordinate functions with partially frozen variables. We illustrate the basic theorems by presenting concrete classes of ergodic functions. As is well known,p-adic spaces have the fractal (although very special) structure. Hence, our study covers a large class of dynamical systems on fractals. Dynamical systems under investigation combine simplicity of the algebraic dynamical structure with very high complexity of behavior.
Ó2014 Elsevier Ltd. All rights reserved.
1. Introduction
Algebraic and arithmetic dynamics are actively devel-oped fields of general theory of dynamical systems. The bibliography collected by Franco Vivaldi [48] contains 216 articles and books; extended bibliography also can be found in books of Silverman [47] and Anashin and Khrennikov[6]. Such studies are based on combination of number theory and theory of dynamical systems. And, as it often happens in mathematics, combination can induce novel constructions interesting for both areas of research. By Ostrowsky theorem [46] the fields of real andp-adic numbers are the only completions of the field of rational numbersQ. This is one of (purely mathematical) motiva-tions to study dynamical systems inQp, see. e.g., [1–50]
(the complete list of reference would be very long; hence, we refer to[6,31,48]). We point to applications ofp-adic
dynamical system in physics (see[19]for a recent review), cognitive science and genetics[1,2,20,26,27,35–37,41].
This paper is devoted to the problem of ergodicity of
p-adic dynamical systems. We solved the problem of characterization of ergodicity and measure preserving for (discrete)p-adic dynamical systems for arbitrary primep
for iterations based on 1-Lipschitz functions. This problem has been open since long time[6]and only the casep¼2 has been investigated in details. As is well known, inp-adic analysis the cases of even and odd prime numbers differ essentially, see, e.g., [46], and transition from p¼2 to
p>2 typically requires elaboration of novel mathematical technique.
We remind that anyp-adic integer (an element of the ringZp) can be expanded into the series:
x¼x0þpx1þ þpkxkþ ;xj2 f0;1;. . .;p1g:
Let functionsdkðxÞ;k¼0;1;2;. . .arekth digits in a base-p expansion of the number x2Zp, i.e. dk:Zp! f0;1;. . .;
p1g;dkðxÞ ¼xk.
0960-0779/$ - see front matterÓ2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.chaos.2014.01.001
⇑Corresponding author.
E-mail address:[email protected](A. Khrennikov).
Contents lists available atScienceDirect
Chaos, Solitons & Fractals
Nonlinear Science, and Nonequilibrium and Complex Phenomena
Any mapf:Zp!Zp can be represented in the coordi-nate form:
fðxÞ ¼d0ðfðxÞÞ þpd1ðfðxÞÞ þ þpkd
kðfðxÞÞ þ : ð1:1Þ
Our aim is to find criteria of measure-preserving and ergodicity in terms of the coordinate functions dkðfðxÞÞ. We restrict our study to the class of 1-Lipschitz functions. The coordinate representation can be useful, e.g., for applications to cryptography. Moreover, by using this representation we can consider non-smooth p-adic transformations.
The basic technical tools are van der Put series and usage of algebraic structure (permutations) induced by coordinate functions with partially frozen variables. (In Section2we shall explain this technique in details.)
The representation of the function by the van der Put
seriesis actively used inp-adic analysis, see e.g.[40,46].
Marius van der Put introduced this series in his disserta-tion ‘‘Algèbres de foncdisserta-tions continues p-adiques’’ at Utrecht Universiteit in 1967, [42]. There are numerous results in studies of functions with zero derivatives, antiderivation[46]obtained using van der Put series. Later van der Put basis was adapted to the case of n-times continuously differentiable functions in one and several variables[18]. First results on applications of the van der Put series in theory of p-adic dynamical systems, the problems of ergodicity and measure preserving, were ob-tained in[7], see also[8,38,51,52]. In this paper we apply this technique to find criteria of ergodicity in terms of the coordinate representation.
In some Theorems (3.1 and 3.4) conditions of ergodicity are formulated in terms of integral sums for the Volkenborn integral, see, e.g.,[46]. One can expect to find formulations of ergodicity in terms of this integral (playing an important role in number theory). However, at the moment this is an open problem.
We illustrate the basic theorems by presenting concrete classes of ergodic functions which satisfy conditions of these theorems.
1.1. P-adic numbers
We recall some definitions related to thep-adic analysis and p-adic dynamical systems, as well as introduce the necessary notations.
For any prime numberpthep-adic normj jpis defined in the following way. For every nonzero integer n let
ordpðnÞ be the highest power of p which divides n, i.e.
n0ðmodpordpðnÞÞ;nX0ðmodpordpðnÞþ1Þ. Then we define jnjp¼pordpðnÞ;j0jp¼0. For rationals mn2Q we set
jn
mjp¼pordpðnÞþordpðmÞ.
The completion ofQwith respect to the p-adic metric
q
pðx;yÞ ¼ jxyjp is called the field of p-adic numbersQp.The metric
q
p satisfies the so-called strong triangle inequalityjxyjp6maxðjxjp;jyjpÞwhere equality holds ifjxjp–jyjp. The set fx2Qp:jxjp61g is called the set of p-adic integers. It is denoted by Zp. In some sense, the set ofp-adic integers is an analogue of the interval½0;1
for real numbers.
Hereinafter, we will consider only thep-adic integersZp.
Everyx2Zpcan be expanded in canonical form, namely in the form of a series which converges byp-adic norm:
x¼x0þpx1þ þpkxkþ ; xk2 f0;1;. . .;p1g; kP0:
If necessary, we can identify every p-adic integer with a sequence of digitsðx0;x1;. . .;xk;. . .Þ. As½xkwe denote the firstkelements of this sequence, i.e.
½xk¼ ðx0;x1;. . .;xk1Þ; kP1: ð1:2Þ
Let a2Zp and r be a positive integer. The set
BprðaÞ ¼ fx2Zp:jxaj
p6prg ¼aþprZp is a ball of radiuspr with centera.
1.2. P-adic functions
In this article will be considered functionsf:Zp!Zp, which satisfy the Lipschitz condition with constant 1 (1-Lipschitz functions). Recall that f:Zp!Zp is 1-Lips-chitz function if
jfðxÞ fðyÞjp6jxyjp; for allx;y2Zp:
This condition is equivalent to that from xyðmodpkÞ followsfðxÞ fðyÞðmodpkÞfor allkP1.
For allkP1 a 1-Lipschitz transformation f:Zp!Zp
the reduced mapping modulo pk is f
k1:Z=pkZ!Z=pkZ, z#fðzÞðmodpkÞ. Mappingf
k1 is a well defined (thefk1
does not depend on the choice of representative in the ball
zþpkZp).
The property of a function to be 1-Lipschitz is signifi-cant. For example, let f¼x2þx
2 be a 2-adic function. This
function is not 1-Lipschitz since jfð2Þ fð0Þj2¼1 but j20j2¼2
1
. Now let us define a function f1 as f1ðxðmod 2ÞÞ f1ðxÞðmod 2Þ. Thenf1ð0Þ ¼0 andf1ð2Þ ¼1.
However, 20ðmod 2Þ.
A 1-Lipschitz transformation f:Zp!Zp is called bijective modulopkif the reduced mappingf
k1is a
permu-tation onZ=pkZ. Andfis called transitive modulopkiff k1
is a permutation that is a cycle of lengthpk.
We introduce two representations ofp-adic functions which we use to describe ergodic properties, namely the van der Put series and the coordinate representation of 1-Lipschitzp-adic functions.
1.2.1. Van der Put series
The van der Put series are defined in the following way. Letf:Zp!Zpbe a continuous function. Then there exists a unique sequence ofp-adic coefficientsB0;B1;B2;. . .such
that
fðxÞ ¼X
1
m¼0
Bm
v
ðm;xÞ ð1:3Þfor all x2Zp. Here the characteristic function
v
ðm;xÞ is given byv
ðm;xÞ ¼ 1; if jxmjp6pn;0; otherwise;
book[46]for detailed presentation of theory of van der Put series). The numbernin the definition of
v
ðm;xÞhas a very natural meaning. It is just the number of digits in a base-pexpansion ofm2N0. Then
blogpmc ¼
ðthe number of digits in a base-pexpansion formÞ 1;
therefore n¼ blogpmc þ1 for all m2N0 and blogp0c ¼0 (recall thatb
a
cdenotes the integral part ofa
).The coefficientsBmare related to values of the functionf in the following way. Letm¼m0þ þmn2pn2þmn1 pn1;m
j2 f0;. . .;p1g;j¼0;1;. . .;n1 and mn1–0,
then
Bm¼ fðmÞ fðmmn1p
n1Þ; ifmPp;
fðmÞ; otherwise: (
We consider a simple example of the van der Put series for the functiongðxÞ ¼x2withp¼2. Then
gðxÞ ¼0
v
ðx;0Þ þ1v
ðx;1Þ þ4v
ðx;2Þ þ8v
ðx;3Þþ16
v
ðx;4Þ þ24v
ðx;5Þ þ32v
ðx;6Þ þ40v
ðx;7Þ þLetx¼5. The characteristic function
v
ðx;mÞis equal 1 form¼1 andm¼5, thengð5Þ ¼1
v
ð5;1Þ þ24v
ð5;5Þ ¼25. 1-Lipschitz functionsf:Zp!Zpin terms of the van der Put series were described in[46]. We follow Theorem 3.1[9]for convenience for further study. In this theorem, the functionfpresented via van der Put series(1.3)is 1-Lips-chitz if and only ifjBmjp6pblogpmcfor allmP0. Assuming
Bm¼pblogpmcbm, we find that the functionfis 1-Lipschitz if and only if it can be represented as
fðxÞ ¼X
1
m¼0
pblogpmcbm
v
ðm;xÞfor suitablebm2Zp;mP0.
1.2.2. Coordinate representation of 1-Lipschitz functions
Let the functionf:Zp!Zpbe represented in the coor-dinate form(1.1)(functiondkdefined onZp). According to Proposition 3.33 in [6],fis a 1-Lipschitz function if and only if for every kP1 the k-th coordinate function
dkðfðxÞÞ does not depend on dkþsðxÞ for all sP1, i.e.
dkðfðxþpkþ1ZpÞÞ ¼dkðfðxÞÞfor allx2 f0;1;. . .;pkþ11g. Taking into account Notation (1.2) for kP0, we consider the functions ofp-valued logic
u
k:f0;. . .;p1g f0;. . .;p1g |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}kþ1
! f0;. . .;p1g
and
u
k:½xkþ1#dkðfðxÞÞ.Using these functions ofp-valued logic, any 1-Lipschitz functionf:Zp!Zpcan be represented as:
fðxÞ ¼fðx0þ þpkxkþ Þ ¼
X1
k¼0
pk
u
kðx0;. . .;xkÞ
¼X
1
k¼0
pk
u
kð½xkþ1Þ ð1:4Þ
For example, in this notation, the functions fk;kP0 (defined for 1-Lipschitz functionf) can be represented as
fkðxÞ ¼Pki¼0pi
u
ið½xiþ1Þ.Hereinafter, we will also use the following method of defining the functions
u
kðx0;. . .;xkÞ;kP1 (as a function ofp-valued logic). The functionu
kðx0;. . .;xkÞcan be de-fined by its sub-functions obtained by fixing the first kvariablesðx0;. . .xk1Þ. Taking into account (1.2) the
sub-function of the sub-function
u
kðx0;. . .;xkÞ, obtained by fixing the variables x0¼a0;. . .;xk1¼ak1;ai2 f0;. . .;p1g, is denoted byu
k;½ak, where a¼a0þpa1þ. . .þp k1a
k1. In
this notations, the function
u
kðx0;. . .;xkÞ can be repre-sented asu
kðx0;. . .;xkÞ ¼X pk1
a¼0
I½akðx0;. . .;xk1Þ
u
k;½akðxkÞ; ð1:5ÞwhereI½ak is a characteristic function, i.e.
I½akðx0;. . .;xk1Þ ¼
1; if ðx0;. . .;xk1Þ ¼ ½ak;
0; otherwise:
:
In this notation, 1-Lipschitz function f:Zp!Zp
becomes:
fðxÞ ¼X
1
k¼0
pk
u
kðx0;. . .;xkÞ
¼
u
0þX1
k¼1
pkX pk1
a¼0
I½akð½xk1Þ
u
k;½akðxkÞ ð1:6ÞRelation of the form(1.6)we call the coordinate repre-sentation of 1-Lipschitz functionf. For example, letp¼3 and we need to find the value offmodulo 33given in the form(1.6)at the point 19. Then, using the notation(1.2)
fð19Þ
u
0ð1Þ þ3u
1;ð0Þð0Þ þ3 2u
2;ð1;0Þð2Þ ðmod 3 3Þ:
The need to use such a representation is due to the fact that the main results on ergodicity of 1-Lipschitz functions will be formulated in terms of functions
u
k;½ak.Any function
u
k can be given by a polynomial of the ring ofðkþ1Þ-variate polynomialsðZ=pZÞ½x0;. . .;xk, with coefficients from the residue ringZ=pZ, whose degree in each variable is at mostp1. In other words,u
kis defined as an element of a factor-ringðZ=pZÞ½x0;. . .;xkmodulo an ideal generated by all polynomialsxpi xi;i¼0;1;. . .;k. In particular, the function
u
kis represented in the form (expansion by the leading variablexk):u
kðx0;. . .;xkÞ ¼xpk1a
p1ðx0;. . .;xk1Þ þ. . .þ
a
0ðx0;. . .;xk1Þ;where
a
iðx0;. . .;xk1Þ 2 ðZ=pZÞ½x0;. . .;xk1;i¼0;1;. . .; p1. In particular,u
k;½ak¼x p1k
a
p1ð½akÞ þ þa
0ð½akÞis a polynomial fromðZ=pZÞ½xk(
a
ið½akÞÞ 2Z=pZ;i¼ ð0;1;. . .;p1).
polynomial where the leading coefficient depends onSfirst terms of the½ak, (S6k).
Thus, coordinate functions
u
k;½ak;u
0can be considered as the functions ofp-valued logic and as transformation of the ringZ=pZ. In each particular case we will determine the way of representation by setting the domain of defini-tion of these funcdefini-tions. For example, if we say thatu
0is gi-ven onf0;. . .;p1g, then it means that the functionu
0isregarded as a function ofp-valued logic.
Conditions on ergodicity are formulated in terms of compositions of functions
u
k;½ak. The composition of func-tions we denote as ‘‘’’ i.e.
u
k;½akðxkÞu
k;½bkðxkÞ ¼u
k;½akðu
k;½bkðxkÞÞ; a;b2 f0;1;. . .;p k1g:As will be shown inTheorem 2.1: 1-Lipschitz function preserves the measure as soon as all the functions
u
k;½ak;a2 f0;1;. . .;pk1g;kP0 are bijective on Z=pZ, i.e. define permutations on the set ofpelements. Therefore, in some cases, we use the terminology from the theory of groups of permutations. For example, if
u
k;½ak is transitive onZ=pZ, then it means that this function defines a transi-tive permutation on the set ofpelements, or of symmetric group of permutationsSp.
1.2.3. Some classes of 1-Lipschitz functions
This article will cover the following classes of 1-Lips-chitz functions. Letf:Zp!Zp be a 1-Lipschitz function.
The function f belongs to ‘0 if all the functions
u
kðx0;x1;. . .;xkÞ;kP1 (the functionu
0 is not underrestrictions) from(1.4)offdefined by polynomials over a fieldZ=pZof the form:
u
kðx0;x1;. . .;xkÞ ¼xkþa
kðx0;x1;. . .;xk1Þ; ð1:7Þwhere
a
kðx0;x1;. . .;xk1Þ 2 ðZ=pZÞ½x0;x1;. . .;xk1. In termsof the coordinate representation (1.6) the belonging to the class ‘0 determined by the conditions
u
k;½ak ¼xkþa
kð½akÞ;a2 f0;1;. . .;pk1g;kP0;a
kð½akÞ 2Z=pZ.LetSbe a positive integer. The functionfbelongs to‘S, if all the functions
u
ðx0;x1;. . .xkÞ;kP1 (the functionu
0is not under restrictions) are defined by polynomials overZ=pZof the form:
u
ðx0;x1;. . .xkÞ ¼xkAkðx0;x1;. . .;xSÞ þa
kðx0;x1;. . .;xk1Þ;ð1:8Þ
i.e. polynomialsAkdepend on no more thanSvariables. In terms of the coordinate representation(1.6)the condition of a functionfto belong to the class‘Sbecomes:
u
k;½ak¼xkAkð½akÞ þ
a
kð½akÞ; ifk<S; xkAkð½aSÞ þa
kð½akÞ; ifkPS;for alla2 f0;. . .;pk1gkP0 (hereA
kð½aSÞ 2Z=pZ). We also will use the definition of uniformly differentia-ble functions modulop(see, for example,[6, Defintion 3.27 and 3.28 on p.58, 60]).
A functionf:Zp!Zpis said to be uniformly differen-tiable modulo ps on Z
p if there exists positive integer
Nand@sfðxÞ 2Qpsuch that for anyk>Nthe congruence
fðxþpkhÞ fðxÞ þpkh@
sfðxÞðmodpkþsÞ holds
simulta-neously for allx;h2Zp. The smallest of theseNis denoted
by NsðfÞ. For example, functions defined by polynomials fromZp½xare uniformly differentiable modulop.
1.3. P-adic dynamics
Dynamical system theory study trajectories (orbits), i.e. sequences of iterations:
x0; x1¼fðx0Þ;. . .;xiþ1¼fðxiÞ ¼fðiþ1Þðx0Þ;. . .;
wherefðsÞðxÞ ¼fðfð. . .fðxÞÞÞ. . .Þ |fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl}
s .
We consider ap-adic autonomous dynamical system
hZp;
l
p;fi. The spaceZpis equipped with a natural proba-bility measure, namely, the Haar measure
l
p normalized so thatl
pðZpÞ ¼1. BallsBprðaÞof nonzero radii constitute the base of the correspondingr
-algebra of measurable subsets,l
pðBprðaÞÞ ¼pr.Note that the 1-Lipschitz functions f:Zp!Zp are continuous.
As usual, a measurable mappingf:Zp!Zp is called measure-preserving if
l
pðf1ðUÞÞ ¼l
pðUÞfor each measur-able subsetU Zp.A measure-preserving mapping f:Zp!Zp is called ergodic iff1ðUÞ ¼Uimplies either
l
pðUÞ ¼0 or
l
pðUÞ ¼1. In Section 2 we give a criterion of ergodicity ( Theo-rem 2.2) of 1-Lipschitz functionsf:Zp!Zpfor anyp–2. A description of ergodic 1-Lipschitz 2-adic functions is known (see, for example,[3,4,6]– in terms of Mahler’s ser-ies;[7,9,52]– in terms of van der Put’s series). To check the ergodicity offit is necessary for everykP0 to check the transitivity of all permutationsu
k;½ak;a2 f0;1;. . .;pk1g;
kP0 in especially order of their appearance in the result-ing product (as an ergodic function is measure-preservresult-ing, then the functions
u
k;½akare bijective onZ=pZ, i.e. are per-mutations on this set). The order of these perper-mutations in the resulting product is determined by the order of the elements fromZ=pkZin the sequence
s
k¼ 0;fk1ð0Þ;. . .;fðpk2Þ k1 ð0Þ;f
ðpk1Þ k1 ð0Þ
n o
:
The order of the elements in the sequence
s
k is signifi-cantly important. This is due to the fact that the symmetric groupSpis nonabelian. In some special cases, the permuta-tionsu
k;½ak;a2 f0;1;. . .;p
k1g are commuting. In this
case, the conditions of ergodicity can be formulated in a more compact way (as an example, see theCorollary 2.5). In theCorollary 2.6we answer the following question. Let fbe a measure-preserving 1-Lipschitz function. How much should one change such function to get an ergodic function? It turns out that such change, with a certain de-gree of conditionality, is surprisingly small.
It turned out that forkP1 of allpk permutations of
u
k;½ak;a2 f0;1;. . .;p
k1gone should in a special way pick
just one permutation, and the remainingpk1 permuta-tions can be chosen arbitrarily. This means, in particular, that usingTheorem 2.2andCorollary 2.6, we can induc-tively (by the parameterk) build any ergodic 1-Lipschitz functionf:Zp!Zp.
In Section3we give a criterion on ergodicity of 1-Lips-chitz functionsfof class‘0and ‘S(forp–2). In general, the functions of these classes are characterised by the fact that all functions
u
k;½ak;a2 f0;1;. . .;pk1g;kP0 in the
coordinate representation(1.6)are given by linear polyno-mials over a field Z=pZ. Ergodic functions of the class‘0
described inTheorem 3.1. Functions from‘0admit the
fol-lowing functional representation (seeProposition 3.1)
fðxÞ ¼fðx0þpx1þ Þ ¼
u
0ðx0Þ þ ðxx0Þ þpgðxÞ;where
u
0:Z=pZ!Z=pZ;gðxÞ:Zp!Zp is 1-Lipschitz function. The class ‘0 consists of ergodic functionsde-scribed in terms of the Mahler’s series (see, for example, Theorem 4.40 from [6]). Since permutations defined by the functions
u
k;½ak¼xkþa
kð½akÞ;a2 f0;1;. . .;pk1g
commute in the symmetric group Sp for allkP1, then the conditions of ergodicity are formulated in terms of
integral sums for the Volkenborn integral 1
pk Ppk1
i¼0 fðiÞ
, see, e.g.,[46].
Ergodicity criterion for the functions from the class‘Sis presented inTheorem 3.5(or in equivalent form in(3.6)). As a corollary from this theorem we formulate a criterion of ergodicity of uniformly differentiable modulop 1-Lips-chitz functions. Thus, open question 4.60 [6, p.132]are fully resolved. Note that 1-Lipschitz uniformly differentia-ble modulop2ergodic functions are described by V.
Ana-shin, see, for example, Theorem 4.55[6, p.126].
In Section3we also present new proofs of ergodicity for functions of the form fðxÞ ¼cþxþpðhðxþ1Þ hðxÞÞ, whereh:Zp!Zpis 1-Lipschitz function andc2Zp(such functions have been described in, for example,[6, Lemma 4.41, p. 112]). We generalize ergodicity criterion for func-tions of the form fðx0þpx1þ Þ ¼cþa0x0þ þ akpkxkþ , wherec;ak2Zp;kP0 to the case of an arbi-traryp(forp¼2 ergodic such functions are described in
[6, Theorem 9.20, p.286]).
In Section4we consider the casep¼3. The main char-acterization of the casep¼3 is that all the transformations of the field of residuesZ=3Zcan be set by linear polynomi-als onðZ=3ZÞ½x. As a corollary from the general criterion of ergodicity (Theorem 2.2) we describe ergodic 1-Lipschitz 3-adic functions using such characterization of the case
p¼3 (Theorem 4.2). InTheorem 4.4we give a 3-adic ana-logue ofCorollary 2.6. Forp¼3 conditions on construction of the ergodic function from measure-preserving function are simplified in comparison with the general case (for
p¼3 requires to determine in a special way the value of the function only at point 0).
In Section4we present the criteria on measure-preser-vation (Theorem 4.7) and ergodicity (Theorem 4.9) of 1-Lipschitz 3-adic functions with the use of the additive form of representation. Earlier additive representation was used to describe the measure-preserving 1-Lipschitz
p-adic functions (see[38]). In this form, measure-preserv-ing 1-Lipschitz function is the sum of an arbitrary 1-Lips-chitz function, and some special function which provides fulfiling the property of measure-preservation. It turned out that for p¼3 in this special function we can distin-guish a ‘‘constant’’ term (the form of representation of this
term can vary). InTheorems 4.7 and 4.9as a ‘‘constant’’ term is considered a functionx. As a result, the criteria on measure-preservation and ergodicity are obtained in a new form of representation. Using such a representation, theExamples 4.8 and 4.10illustrate how wide is the class of measure-preserving ergodic 1-Lipschitz 3-adic functions compared with the known classes of such functions (see, for example,[6]).
In Section5we give a criterion on ergodicity of 1-Lips-chitz 2-adic functions (Theorem 5.1), which is formulated in terms of integral sums for the Volkenborn integral, see, e.g., Schikhof[46].
2. General criteria on ergodicity for anyp.
In the section, we prove a criterion on ergodicity of 1-Lipschitz functionsf:Zp!Zp (Theorem 2.2) in terms of their coordinate representation(1.4). There is an example of a class of functions where conditions on ergodicity rep-resented in the ‘‘compact’’ form (Corollary 2.5). InCorollary 2.6we answer the following question ‘‘How much should one change measure-preserving 1-Lipschitz function to get an ergodic function?’’ In addition, we present a crite-rion on measure-preservation for 1-Lipschitz function rep-resented in the coordinate form (1.4). This theorem is needed for the proof of the ergodicity criterion.
Theorem 2.1. Let f:Zp!Zpbe 1-Lipschitz function in the
coordinate representation (1.6). The function f preserves
measure if and only if all functions
u
0 andu
k;½ak; a2f0;1. . .;pk1g and kP1 is bijective (permutations) on
the setf0;. . .;p1g.
Proof. Let fðxÞ ¼P1m¼0Bm
v
ðm;xÞ ¼ P1m¼0pblogpmcbm
v
ðm;xÞ be the van der Put representation of the functionf anda2 f0;. . .;pk1g. Let us find values of the coefficientsbm. ThenbmðfÞ ¼BmðfÞ ¼fðmÞform2 f0;. . .;p1gand for
m¼aþpkx
k;a2 f0;. . .;pk1g;xk–0;kP1
baþpkx kðfÞ ¼
1
pkBaþpkx kðfÞ ¼
fðaþpkxkÞ fðaÞ pk
¼
u
kð½ak;xkÞu
kð½ak;0Þ þpðu
kþ1ð½ak;xk;0Þu
kþ1ð½ak;0;0Þ þ. . .ÞThus
bmðfÞ
u
0ðmÞðmodpÞ; m2 f0;. . .p1g;bmðfÞ ¼baþpkx
kðfÞ
u
kð½ak;xkÞu
kð½ak;0Þu
k;½akðxkÞu
k;½akð0ÞðmodpÞ; m¼aþpkxk; xk–0
By Theorem 2.1[38]the functionfpreserves the measure if and only if
1. b0ðfÞ;b1ðfÞ;. . .;bp1ðfÞ establish a complete set of
residues modulop;
2. baþpkðfÞ;. . .;baþpkðp1ÞðfÞare all nonzero residues modulop.
Now let us state a general criterion on ergodicity for a 1-Lipschitz function in terms of the coordinate functions
(1.6).
Theorem 2.2. Let f:Zp!Zpbe 1-Lipschitz function in the
coordinate representation(1.6),where all functions
u
0 andu
k;½ak, a2 f0;1;. . .;pk1g;kP1are permutations on the
setf0;. . .;p1g(i.e. f preserves the measure).
The function f is ergodic if and only if all permutations
u
0and
Fk¼
u
k; fðpk1Þk1 ð0Þ
k
u
k;fðpk2Þ k1 ð0Þ
k
u
k;½0k; kP1are transitive on thef0;. . .;p1g.
Proof. Let a¼a0þpa1þ. . .þpk1ak12 f0;1;. . .;pk1g.
And let us find the coordinate representation forfkðpk1ÞðaÞ,
kP1. Then we have:
fkðpk1Þ h i
k¼ ðF1;½a1ða0Þ;. . .;Fk1;½ak1ðak2Þ;Fk;½akðak1ÞÞ;
where
F0;½a1¼
u
0u
0u
0|fflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflffl} pk1
;
F1;½a1¼
u
2;fðpk11Þ 1 ð½a1Þu
2;f1ðpk12Þð½a1Þ
u
1;½a1;. . .
Fk1;½ak1¼
u
k1;fkðpk111Þð½ak1Þu
k1;fkðpk112Þð½ak1Þ
u
k1;½ak1;Fk;½ak¼
u
k;fðpk11Þ k1 ð½akÞu
k;fðpk12Þk1 ð½akÞ
u
k;½ak;ð2:1Þ
Assume thatfis ergodic. By Theorem 4.23[6, p.99]if the functionfis ergodic, then, for anysP0, the functionfsis transitive on the set of residues modulopsþ1and, in fact, ps is the minimal integer such thatfðpsÞ
s ðaÞ aðmodpsþ1Þ. The function fk:Z=pkþ1Z!Z=pkþ1Z induces a map on
Z=pkZ. This map coincides withfk1(fis 1-Lipschitz
func-tion by initial condifunc-tions).
It means that the permutationsF0;½a1;F1;½a1;. . .;Fk1;½ak1
from the coordinate representation of thefkðpk1Þ are iden-tical. Indeed, in this case each permutation is a degree of the identical permutation
, i.e.F0;½a1¼
;F1;½a1¼;. . .;Fk1;½ak1¼:As we know,fkðpkÞð½akÞ ½akðmodpkþ1Þandpk is minimal (transitivity offkis by Theorem 4.23[6, p.99]).
Then from the representation and
fkðpkÞ h i
k¼ f
ðpk1ÞðpÞ
k
k
¼
;;. . .;;Fk;½ak Fk;½ak |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}p 0
B @
1 C A
it follows that
Fk;½ak Fk;½ak |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
p
¼
ð2:2Þand, moreover,pis minimal natural number for holding the condition (2.2). Hence, the permutation Fk;½ak is
transitive. Since f0¼
u
0, then transitivity of the mapu
0follows from Theorem 4.23[6, p.99].
We now present the proof that transitivity all functions
u
0andu
k;½ak;a2 f0;1;. . .;pk1g;kP1 are sufficient for
ergodicity. We shall use induction with respect to the parameterk¼0;1;. . .to show transitivity of the functions
fk.
For k¼0 transitivity of f0 follows from the relation
f0¼
u
0and transitivity ofu
0follows from the condition of Theorem.Letfk1 be transitive on the set of residues modulopk. Sincefkinduces a map onZ=pk1, which coincides withfk1 (f is 1-Lipschitz function), then fkðpkÞ¼ ð;
;. . .;;Fk;½akÞ, where Fk;½ak is defined in (2.1) and is the identity permutation on the setf0;. . .;p1g. By condition of the Theorem, the mapFk;½ak is the transitive permutation on the setf0;. . .;p1g. Therefore,fkðpkÞ h i
k¼ f
ðpk1ÞðpÞ k
k
¼
;;. . .;;Fk;½ak. . .Fk;½ak |fflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflffl}p 0
B @
1 C A
¼ ð
;;. . .;Þand suchpkis minimal. It means thatf
kis transitive on the set of residues modulopk. Thus we proved that the func-tions fk;k¼0;1;. . . are transitive. Then by Proposition 4.35[6, p.105]the functionfis ergodic.
For the proof of the Theorem it is sufficient to show that the conditions of ergodicity criterion offdoes not depend on the choice of the parametera2 f0;. . .;pk1g;kP1. It
is enough to prove the following statement. Let
a–b2 f0;1;. . .;pk1g and kP1. If function f
k1 is transitive modulo pk (i.e. on f0;1;. . .;pk1g), then
permutations onf0;1;. . .;p1g
ga¼
u
k;fðpk1Þk1 ðaÞ
k
u
k;fðpk2Þ k1 ðaÞ
k
u
k;½ak;gb¼
u
k; fðpk1Þk1 ðbÞ
k
u
k;fðpk2Þ k1 ðbÞ
k
u
k;½bkhave the same cyclic structure. Indeed, fk1 is transitive
modulopk. Then there exist an integerr6pk11 such
thatfkðrÞ1ðaÞ ¼band, therefore,fkðtÞ1ðbÞ ¼fkðtþ1rÞðaÞ. Then
ga¼
u
k;fðpk1Þk1 ðaÞ
k
u
k;fðrþ1Þk1ðaÞ ½ k
u
k;fðrÞk1ðaÞ
½ k
u
k;½ak ð2:3Þ¼
u
k;fðpk11rÞk1 ðbÞ
k
u
k;fð2Þk1ðbÞ ½ k
u
k;½fk1ðbÞku
k;fðr1Þk1 ðaÞ ½ k
u
k;½ak ð2:4ÞConsistently making conjugation of the permutation
(2.3)by permutations
u
k;½ak;. . .;u
k;fðr1Þk1ðaÞ ½ k
, we obtain
u
k;½fðr1Þ k1ðaÞk
u
k;½akgau
k;½ak 1
u
k;½fðr1Þ k1 ðaÞk1
¼
u
k;½fðr1Þ k1 ðaÞk
u
k;½aku
k;½fðpk11rÞk k1 ðbÞk. . .
u
k;½fk1ðbÞk¼
u
k;½fðpk1Þk1 ðbÞk
u
k;½fðpkrÞ k1 ðbÞk
u
k;½fðpk1rÞk1 ðbÞk
u
k;½fk1ðbÞk¼gb:
Let G¼
u
k;fðr1Þk1 ðaÞ ½ k
u
k;½ak. Then GgaG
1
¼gb,
i.e., the permutationsga;gbare conjugate permutations. It is well known that cyclic structure of conjugate permuta-tions coincide.
To complete the proof it suffices to choosea¼0 (for all
kP1) and setFk;½0k¼Fk. h
Remark 2.3. Theorem 2.2can be considered as generaliza-tion to the casep–2 of Theorem 4.39 (Folklore) from the book [6]. The latter was ‘‘known’’ by people working in cryptography, but without formal 2-adic presentation (and the authors of[6]do not know any rigorous mathe-matical proof before the one presented in this book).
Comment 2.4. To check a function on ergodicity by using conditions ofTheorem 2.2, one should check transitivity of the permutations Fk, k¼1;2;. . .. Each permutation is a product of permutations
u
k;½ak;a2 f0;. . .;p
k1g. The
order of their appearance in the resulting product (i.e. in
Fk) is defined by the sequence of residues modulopk
s
k¼ f0;fk1ð0Þ;. . .;fðpk2Þ k1 ð0Þ;f
ðpk1Þ k1 ð0Þg:
Note that the order of residues modulo pk in the sequence
s
k is significantly important. This is due to the fact that the symmetric groupSp (permutations onZ=pZ) is nonabelian. Therefore, in general by determiningFkwe cannot change the order of the permutationsu
k;½ak.If we suppose that the permutations
u
k;½ak2Sp;a2f0;1;. . .;pk1gcommute, then the conditions of
ergodic-ity offare simplified. In theCorollary 2.5we considered the case such that, for each k1 the permutations
u
k;½ak2Sp belong to the cyclic group generated by a permutation gk. In this case, allu
k;½ak2Sp commute,Fk does not depend on the order of elements in the sequences
k, and, moreover,Fk¼gakfor somea
depending only onk.Therefore, to verify the transitivity ofFk, there is no need to
build the sequence
s
k. This simplifies essentially theverification of ergodicity off.
Corollary 2.5. Let f:Zp!Zp be 1-Lipschitz function in the
coordinate representation(1.6).
Suppose that the subfunctions have the form:
u
k;½ak ¼g nkðaÞk ; a2 f0;. . .;p
k1g
; kP1; ð2:5Þ
where gk is a permutation onZ=pZand nkðaÞis a positive
integer (g0
k- identity permutation).
Then the function f is ergodic if and only if
1.
u
0;gkare transitive permutations;2. Ppak¼01nkðaÞX0 modpÞ; kP1.
Proof. Let f be an ergodic function. By Theorem 2.2
permutations
u
0 and Fk¼u
k ;fðpk1Þk1 ð0Þ
k
u
k;½fk1ð0Þku
k;½0k;kP1 are transitive onZ=pZ. Then
Fk¼
u
k;½fðpk1Þk1 ð0Þk
u
k;½fk1ð0Þku
k;½0k¼gnkðfkpk11ð0ÞÞ
k g
nkð0Þ
k ¼g
nkðpk1Þþþnkð0Þ
k ¼g
Rk k ;
whereRk¼nkðpk1Þ þ þnkð0Þ.
AsFk is transitive onZ=pZ, thenord Fk¼p.
Letord gk¼
a
. Thenp¼ord Fk¼ord gRk
k ¼
a
gcdða;RkÞ, i.e.p divides
a
. If the permutationgkhasr1cycles of the length 1,r2cycles of the length 2;. . .rpcycles of the lengthp, and r1þ2r2þ. . .þprp¼p, thena
¼ord gk¼lcmðijri–0Þ.As p is a prime number and p devides
a
, then the permutationgkconsists of one cycle of lengthp, i.e.a
¼pand gk is transitive permutation. From p¼gcdðaa;RkÞ;
a
¼p follows that gcdða
;RkÞ ¼gcdðp;RkÞ ¼1, i.e. Rk¼Ppk1 a¼0nk ðaÞX0ðmodpÞ.
And vice versa, let us find an order of each permutation
Fk,kP1. By one of the conditions of this corollary
u
0¼f0 is a transitive permutation and, in particular,ff0ðp1Þð0Þ;. . .;f0ð0Þ;0g ¼ f0;1;. . .;p1g. Then F1¼
u
1; ½f0ðp1Þð0Þ1u
1;½f0ð0Þ1u
1;½01 ¼gn1ðp1Þþþn1ð0Þ
1 ¼g
R1
1, whereR1 ¼n1ðp1Þ þ þn1ð0Þ. By initial conditions
g1 is transitive on Z=pZ, then ord g1¼p. As
R1X0ðmod pÞ, then gcdðp;R1Þ ¼1 and ord F1¼ ord
gR1
1 ¼
p
gcdðp;R1Þ¼ p. Thus F1 is transitive permutation on Z=pZ. Moreover, the functionf1is transitive onZ=p2Z.
Computation of the valuesord Fk;kP2 is performed in a similar way, taking into account that at each step we prove transitivity of the functionsfk1. Finally, we obtain that the permutationsFk;kP1 are transitive onZ=pZ. As
u
0is transitive onZ=pZ, then the functionfis ergodic byTheorem 2.2. h
In theCorollary 2.6, in particular, we answer the follow-ing question. Let f be a measure-preserving 1-Lipschitz function. How much should one change such function to get an ergodic function? It turns out that such change, with a certain degree of conditionality, is surprisingly small.
In representation(1.6)for allkP1 each ofpk1 per-mutations
u
k;½ak;a2 f1;2;. . .;pk1g(aX0ðmodpkÞ) can be chosen arbitrarily and only permutation
u
k;½0kis chosen in a special way.Corollary 2.6. Let f:Zp!Zpbe 1-Lipschitz function in the
coordinate representation(1.6),where the subfuctions of the
coordinate functions,
u
0 andu
k;½ak;a2 f0;1;. . .;p k1g;kP1are permutations on the setZ=pZand
u
0is transitive.Then it is always possible to construct permutations g1;g2;. . .;gk;. . . such that by setting
u
~k;½0k¼gk;k¼1;2;. . ., the corresponding function~f which is defined with the
aid of subfunctions
u
k;½ak; a–0, andu
~k;½0k is ergodic.Proof. Let us find permutationsg1;g2;. . .;gk;. . .by
induc-tion. By conditions of theorem, permutation
u
0¼f0 istransitive on Z=pZ, and, in particular, f0ðp1Þð0Þ;. . .;f0
n
ð0Þ;0g ¼ f0;1;. . .;p1g. By using notations from Theo-rem 2.2, we setG1¼
u
1;½fðp1Þ0 ð0Þ1
u
1;½f ðp2Þ0 ð0Þ1
u
1;½f0ð0Þ1.We also choose some transitive permutationH1onZ=pZ. Letg1¼G1
1 H1. Then set
u
~1;0¼g1, and we obtain that permutation Fe1¼u
1;½f0ðp1Þð0Þ1u
1;½fðp2Þ 0 ð0Þ1
u
1;½f0ð0Þ1~
u
1;½0k¼H1 is transitive onZ=pZ. Moreover, the function~
Now let permutations g1;g2;. . .;gk1;. . .have already been constructed, and the function~fk1 is transitive on
Z=pkZ. Let us construct a suitable permutationg k. Let Gk¼
u
k;½~fðkpk11Þð0Þ
k
u
k;½~f k1ð0ÞkandHkbe some transitive permutation onZ=pZ. And set
gk¼G
1
k Hk. Then, for
u
~k;½0k¼gk, we obtain that the permutatione Fk¼
u
k;½~fðkpk11Þð0Þ
k
u
k;½~f k1ð0Þk~
u
k;½0k¼Hkis transitive onZ=pZ. And the function~fk is transitive on
Z=pkþ1Z.
Thus we have shown existence of permutations
g1;g2;. . .;gk;. . . such that for
u
~k;½0k¼gk;kP1 permuta-tionseFkare transitive onZ=pZ. Asu
0is transitive onZ=pZ, then the function~f is ergodic byTheorem 2.2. hComment 2.7. Using the results ofCorollary 2.6and the criterion of Theorem 2.2 we can inductively construct any 1-Lipschitz ergodic function in the representation
(1.6)as follows. Permutation
u
0 we choose transitive on Z=pZ. Let the permutationsu
k1;½ak1;a2 f0;1;. . .; pk11g are constructed, i.e. all of the functionsu
s;s¼1;2;. . .;k1 are choosen. Let us constructu
k. Todo this, the permutations
u
k;½ak;a2 f1;. . .;p
k1g (i.e.
aX0ðmodpkÞÞwe choose arbitrarily and find
Gk¼
u
k;fðpk1Þk1 ð0Þ
k
u
k;½fk1ð0Þk(the function fk1 defined via
u
k1;½ak1;a2 f0;1;. . .; pk11g). Then we chooseu
k;½0k in such a way that Gk
u
k;½0k be a transitive permutation onZ=pZ. Aspis a prime number, then one can write this condition analyti-cally. Namely,u
k;½0k is a solution in permutations ðGkXÞp¼
, whereis the identical permutation.As a result, the function
u
kis constructed. The numberof suitable choice equals to ðp!Þk1ðp1Þ! (or 1
p of all
suitable functions
u
kfor construction ameasure-preserv-ing functions).
Proceeding in this way to select the functions
u
k;k1,then we construct a 1-Lipschitz ergodic function. It is clear that any 1-Lipschitz ergodic function can be constructed using this method.
3. Examples of classes of ergodic functions forp 6¼ 2
In this section, we prove a criterion on ergodicity of 1-Lipschitz functionsfof class‘0and‘S(Theorems 3.1 and
3.5). In general, classes‘0and‘S are characterized by the fact that the functions
u
k;½ak;a2 f0;1;. . .;pk1g;kP0
in the coordinate representation of(1.6)fdefined by linear polynomials over a fieldZ=pZ(for the class‘0these
poly-nomials are unitary, which allow us to represent functions from‘0explicitly, i.e. without using the coordinate
repre-sentation(1.6), seeLemma 3.1).
Examples of the application of the criteria on ergodicity for the classes‘0and‘Sare:
a new proof of the ergodicity for the functions of the
form cþxþpðhðxþ1Þ hðxÞÞ, where h:Zp!Zp is
1-Lipschitz function,cX0ðmodpÞ(Corollary 3.3);
ergodicity criterion (Theorem 3.4) for the functions of the form
fðx0þpx1þ Þ ¼cþa0x0þ þakpkxkþ ;
c;ak2Zp;kP0 forp–2 (the casep¼2 is considered in, for example,[6, Theorem 9.20 (part 2), p.286]);
description of uniformly differentiable modulo p 1-Lipschitz functions (see, for example, [6]), i.e. open question 4.60[6, p.132]are fully resolved.
3.1. Ergodic functions of the class‘0
In the subsection we prove a criterion on ergodicity of 1-Lipschitz functionsfof class‘0(Theorem 3.1). The
func-tions in this class are characterized by the fact that the sub-functions of the coordinate functions are of the form
(1.7), i.e. defined by linear unitary polynomials over a field
Z=pZ. The permutations that are defined by these polyno-mials commute in the symmetric group of permutations. This property allow us to formulate a criterion on ergodic-ity in terms of integral sums for the Volkenborn integral, see, e.g., Schikhof[46]. Functions from‘0admits an explicit
representation (Lemma 3.1). As an example of the use of
Theorem 3.1we give a new proof of(3.3)known criterion on ergodicity of functions of the form cþxþpðhðxþ1Þ hðxÞÞ (see [3,4]and[6, Lemma 4.41, p.112]). Note that these type of functions belong to the class ‘0. Another
example of application ofTheorem 3.1is a generalization of criterion on ergodicity for 2-adic functions (p¼2) of the form
fðx0þpx1þ þpkxkþ Þ ¼cþa0x0þ þakpkxkþ ;
whereak2Zp;kP1;c2Zpup to the casep–2.
The explicit form of the functions from the class‘0is
presented in the following statement.
Lemma 3.1. Let f:Zp!Zpbe 1-Lipschitz function. Function
f belongs to‘0if and only if f can be represented as
fðxÞ ¼fðx0þpx1þ þpkxkþ Þ
¼
u
0ðx0Þ þ ðxx0Þ þpgðxÞ;where g:Zp!Zp is 1-Lipschitz function and
u
0:Z=pZ!Z=pZ.
Proof. Let fðxÞ ¼P1
m¼0Bm
v
ðm;xÞ ¼P1m¼0pblogpmcbmv
ðm;xÞ be the van der Put representation of the 1-Lipschitz func-tion f2‘0. Let us find values of the coefficients bm formPp. Denote m¼m0þpm1þ þpkm k¼mþpkmk;
m2 f0;. . .;pk1g;m
k2 f1;. . .;p1g (i.e. mk–0). Then
(under notations from(1.4) and (1.6))
Bm¼Bmþpkm
k¼fðmþp
kmkÞ fðmÞ ¼pkð
u
kð½mk;mkÞ
u
kð½mk;0Þ þp kþ1ð. . .Þ;
i.e.
bm¼bmþpkx
k
u
kð½mk;mkÞu
kð½mk;0Þ mkþa
kð½mkÞa
kð½mkÞ mk ðmodpÞ:In other words, bm¼bmþpkmk¼mkþp
bmþpkmk¼mk þpbmfor suitablep-adic integerbm.
We set fðiÞ ¼Bi¼bi¼biðmodpÞ þpbi;bi2Zp;i¼
f0;. . .;p1g. Then the function
u
0:Z=pZ!Z=pZ fromrepresentation(1.6)takes the form
u
0ðxÞ ¼b0ðmodpÞv
ð0;xÞ þ þbp1ðmodpÞv
ðp1;xÞand
X1
m¼p
bmðmodpÞ pblogpmc
v
ðm;xÞ ¼X1
m¼p
qðmÞ
v
ðm;xÞ ¼xx0;whereqðmÞ ¼qðmþpkm
kÞ ¼pkmk;mk–0.
And let us consider the 1-Lipschitz function
gðxÞ ¼P1
m¼0bmpblogpmc
v
ðm;xÞ. Thus, fðxÞ ¼X1
m¼0
bmpblogpmc
v
ðm;xÞ¼X
1
m¼0
bmðmodpÞ pblogpmc
v
ðm;xÞþpX
1
m¼0
bmpblogpmc
v
ðm;xÞ¼X p1
m¼0
bmðmodpÞpblogpmc
v
ðm;xÞ þX1
m¼p
bmðmodpÞ
pblogpmc
v
ðm;xÞ þpgðxÞ ¼u
0ðx0Þ þ ðxx0Þ þpgðxÞ:Vice versa, let
fðxÞ ¼fðx0þpx1þ þpkxkþ Þ
¼
u
0ðx0Þ þ ðxx0Þ þpgðxÞandfðxÞ ¼P1m¼0bmpblogpmc
v
ðm;xÞ.Set m¼m0þpm1þ þpkmk¼mþpkmk;mk–0 we
obtain that formPp
bm¼bmþpkm k
1
pkðfðmþp
kmkÞ fðmÞÞ
1
pkð
u
0ðm0Þ þ ðmþpkmkm0Þ þpgðm þpkmkÞð
u
0ðm0Þ þ ðmm0Þ þpgðmÞÞÞ1 pkðp
kmkþp ðgðmþpkmkÞ gðmÞÞÞ
mkþ 1
pk1ðgðmþp
kmkÞ gðmÞÞ ðmodpÞ:
As the function g is 1-Lipschitz, then gðmþpkm kÞ
gðmÞð modpkÞand
1
pk1ðgðmþp
kmkÞ gðmÞÞ 0 modp:
ThenbmmkðmodpÞ.
On the other hand (under notations from(1.4)),
bm
u
kð½mk;mkÞu
kð½mk;0Þ modp:or
u
kð½mk;mkÞ ¼mkþu
kð½mk;0Þ ¼mkþa
kð½mkÞ:Thusf2‘0. h
Ergodicity criterion functions of the class‘0is presented
in the following theorem.
Theorem 3.1.Let p–2 and f:Zp!Zp be 1-Lipschitz function and f 2‘0, i.e.
fðxÞ ¼fðx0þpx1þ þpkxkþ Þ
¼
u
0ðx0Þ þ ðxx0Þ þpgðxÞ;where g:Zp!Zp is 1-Lipschitz function and
u
0:Z=pZ!Z=pZ. The function f is ergodic if and only if1.
u
0is a transitive (monocycle) permutation onZ=pZ;2. 2p2 þ1
pk Ppk1
a¼0fðaÞX0ðmodpÞ, or in the equivalent
form 1
pk1
Ppk1 a¼0gðaÞX2
p2
ðmodpÞ; kP1.
Proof. Note that the functionfpreserves measure as soon as
u
0is bijective onZ=pZ. Indeed,f2‘0implies that allfunctions
u
k;½ak;a2 f0;. . .;pk1g;kP1 in the coordinate
representation off(1.6)are bijective as linear polynomials (see(1.7)) over the field of residues modulop. Then by the criterion on measure-preserving functions in terms of coordinate functions, see Theorem 2.1 in[38], the function
f is measure-preserving and for all fixed a2 f0;1;. . .;
pk1gthe function
u
k;½akis a permutation on
Z=pZ;kP1. As inTheorem 2.2, let us consider the permutations
Fk¼
u
k;fðpk1Þk1 ð0Þ
k
u
k;fðpk2Þ k1 ð0Þ
k
u
k;½0k;wherekP1.
By the assumption the permutations
u
k;½akare given by linear polynomials of the formu
k;½akðxkÞ ¼xkþa
kð½akÞ. ThereforeFk¼xkþ
a
k fðp kÞ k1ð0Þh i
k
þ
a
k fðp k2Þ k1 ð0Þh i
k
þ þ
a
kð½0kÞ¼xkþX pk1
i¼0
a
k fkðiÞ1ð0Þh i
k
:
Let us show by induction with respect tokP0 that the functionsfkare transitive onZ=pkþ1Z. Fork¼0, the func-tionf0is transitive becausef0¼
u
0, whereu
0is transitiveonZ=pZby the assumption. Suppose now thatfk1is
tran-sitive onZ=pkZ. Then
Fk¼xkþX pk1
i¼0
a
k fkðiÞ1ð0Þh i
k
¼xkþX pk1
i¼0
a
kð½ikÞ:On the other hand, seta¼a0þpa1þ þpk1ak1(under
notations in(1.2)), then we obtain
X pk1
a¼0
fðaÞ X pk1
a¼0
pk
u
kða0;. . .;ak1;0Þ þ
Xk1
i¼0
pi
u
iða0;a1;. . .;aiÞ
!
X pk1
a¼0
pk
a
kð½akÞ þXk1
i¼0
pi
u
iðx0;x1;. . .;xiÞ
!
pkX pk1
a¼0
a
kð½akÞ þ X pk1a¼0
As fk1 is bijective on Z=pk1Z, then Pp k1 a¼0fk1ðaÞ ¼pkðp2k1Þ, i.e.
X pk1
a¼0
fðaÞ pkX pk1
a¼0
a
kð½akÞ þpkðpk1Þ
2 ðmodp
kþ1Þ;
or
1
pk X pk1
a¼0
fðaÞ X
pk1
a¼0
a
kð½akÞ þpk1
2
X pk1
a¼0
a
kð½akÞ 21
ðmodpÞ;
where 212p2ðmodpÞ.
Then by the second condition of this Theorem
X pk1
a¼0
a
kð½akÞ1
pk X pk1
a¼0
fðaÞ þ2p2
X0 ðmodpÞ:
And, therefore, the permutationFk is transitive onZ=pZ. This means that the functionfk is transitive onZ=pkZ. By Theorem 4.23 [6, p.99] from transitivity of fk on
Z=pkZ;kP0 follows that the functionfis ergodic. Vice versa, let the functionfbe ergodic function. Then
f0¼
u
0 and functionsFk;kP1 are transitive onZ=pZby Theorem 2.2. AsFk¼xkþX pk1
i¼0
a
k fkðiÞ1ð0Þh i
k
¼xkþX pk1
a¼0
a
kð½akÞ;then transitivity of Fk means that Ppk1
a¼0
a
kð½akÞX0ðmodpÞ. Then
1
pk X pk1
a¼0
fðaÞ X
pk1
a¼0
a
kð½akÞ þpk1
2
X pk1
a¼0
a
kð½akÞ 21
ðmodpÞ;
i.e. 2p2þ1
pk Ppk1
a¼0fðaÞX0ðmodpÞ;kP1.
To complete the proof, it remains to note that
2p2
þ1 pk
X pk1
a¼0
fðaÞ
2p2þ1 pk p
k1X
p1
½a1¼0
u
0ð½a1Þ þX pk1
a¼0
ða ½a1Þ þp
X pk1
a¼0
gðaÞ 0
@
1 A
2p2
þ p1
2 þ
pk11 2
þ 1 pk1
X pk1
x¼0
gðxÞ
2p2
þ 1 pk1
X pk1
a¼0
gðaÞ ðmodpÞ:
h
Comment 3.2.In theComment 2.4was stated that if for each k1 all permutations
u
k;½ak;a2 f0;. . .;p k11g
belong to a cyclic subgroup of the symmetric groupSp, then the conditions of ergodicity off are simplified. In Theo-rem 3.1we considered the case such that the cyclic group is generated by permutation onZ=pZ, which is given by a linear polynomialxþc;c–0 over the fieldZ=pZ. In this case, the ergodicity of f is determined by the value of
1
pk Ppk1
a¼0fðaÞ ðmodpÞ;k1.
Note that such sums appear in the definition of Volkenborn’s integral, see, for example, Definition 55.1
[46]. However, real coupling between theory of Volken-born integration and ergodicity of p-adic dynamical sys-tems has not yet been clarified.
As an example, the criterion ofTheorem 3.1we give a new proof of the known (for example, see Lemma 4.41, p.112 [6]) ergodicity criterion for functions of the form
fðxÞ ¼cþxþpðhðxþ1Þ hðxÞÞ, where h:Zp!Zp is 1-Lipschitz function. Note that the functions fromCorollary 3.3 belong to the class of functions considered in Theo-rem 3.1. In other words,Theorem 3.1generalize the results of Lemma 4.41,[6].
Corollary 3.3 (see Lemma 4.41, p.112[6]). Let f :Zp!Zp; fðxÞ ¼cþxþpðhðxþ1Þ hðxÞÞ, where h:Zp!Zp is
1-Lipschitz function and p–2. If cX0 modp, then f is an
ergodic function.
Proof. As
u
0ðxÞ f0ðxÞ fðxÞmodpÞ x0þcmodpÞ,wherex¼x0þpx1þ , then
u
0ðxÞis transitivepermuta-tion oncecX0ðmodpÞ. Ashis 1-Lipschitz function, then
hðpkÞ hð0Þ 0ðmodpkÞ. Then forkP2 1
pk1
X pk1
i¼0
ðhðiþ1Þ hðiÞÞ hðp kÞ hð0Þ
pk1
0X2p2
ðmodpÞ:
Thus the functionfis ergodic byTheorem 3.1. h
Another example of the use of the Theorem 3.1 is
Theorem 3.4. This generalizes the results of Theorem 9.20
[6, part 2, p.286]to the casep–2.
Theorem 3.4. Let f :Zp!Zpðp–2Þ be a 1-Lipschitz function of the form
fðx0þpx1þ þpkxkþ Þ ¼cþa0x0þ þakpkxkþ ;
where ak2Zp;kP1;c2Zp. The function f is ergodic if and
only if cX0ðmodpÞand ak1ðmodpÞ;kP1.
Proof. Let a function f:Zp!Zp be represented in the coordinate form. Then for suitable p-valued function
lkðx0;x1;. . .;xk1Þ
u
kðx0;x1;. . .;xkÞ 1pkðfðx0þ þp
kxkÞ fðx
0þ
þpk1xk
1ÞðmodpkÞÞðmodpÞ
akxkþlkðx0;x1;. . .;xk1ÞðmodpÞ:
Letfbe an ergodic function. ByTheorem 2.2 permuta-tions
u
0andFk¼
u
k;fðpk1Þk1 ð0Þ
k
u
k;½fk1ð0Þku
k;½0k; kP1are transitive on Z=pZ. As for suitable bk2Z=pZ;
k¼1;2;. . .
Fk¼ akxkþlk fðpk1Þ k1 ð0Þ
ðakxkþlkð0ÞÞ
¼apkkxkþbk;
then apk
k ak1ðmodpÞ. From transitivity of
u
0 cþa0x0ðmodpÞ follows that cX0ðmodpÞ and a01ðmodpÞ.
Vice versa, by assumptionak1ðmodpÞ;kP1. Then
u
kðx0;x1;. . .;xkÞ ¼xkþlkðx0;x1;. . .;xk1Þfor suitable p-valued function lkðx0;x1;. . .;xk1Þ. This, in
particular, means that the function fsatisfies the condi-tions ofTheorem 3.1.
Let us check the function f on ergodicity. As
cX0ðmodpÞ and a01ðmodpÞ, then
u
0cþx0ðmodpÞis transitive onZ=pZ. Note that,
1
pk X pk1
i¼0 fðiÞ 1
pk X
x0;...;xk1
ðcþa0x0þ þak1pk1xk1Þ
1
pk p
kcþa
0pk1
pðp1Þ
2 þ. . .þak1p2k2
pðp1Þ
2
cþa0
p1
2 ca02 p2
c2p2
ðmodpÞ:
Therefore, 2p2 þ1
pk Ppk1
i¼0 fðiÞ cX0ðmodpÞ. Then by Theorem 3.1, the functionfis ergodic. h
3.2. Ergodic functions of the classes‘S
In this subsection we prove the ergodicity criterion for functions from the class‘S, where the parameterSis a fixed positive integer (Theorem 3.5). These functions are charac-terized by the fact that their coordinate functions
u
k; kP0 are given by linear by the variablexk polynomi-als inðZ=pZÞ½x0;. . .;xk, in which the coefficients atxkare polynomials that depend on no more thanSvariables (see the representation (1.8)). In contrast to the general criterion of ergodicity (see Theorem 2.2), where it is required to calculate the values of all iterationsfðiÞð0Þ ðmodpkÞ;i2 f0;1;. . .;pk1g;kP1, to check ergodicity of functions from the class‘Swe need to calcu-late the iterations for thelimited number of values,namely,
fðiÞð0Þ ðmodpSÞ;i2 f0;1;. . .;pS1g, for everykPS. By the criterion of Theorem 3.5 the Corollary 3.7
describes the ergodic 1-Lipschitz uniformly differentiable function modulo p. The problem of description of these functions has been put in the open question 4.60 [6, p.132]. As a result, taking into account the criterion of ergodicity by V. Anashin for 1-Lipschitz uniformly differen-tiable function modulo p2 (see[3,4]or Theorem 4.55 [6, p.126]), a complete description of ergodic 1-Lipschitz uniformly differentiable function modulo pa for any
a
P1 was obtained.Ergodicity criterion for functions from‘Sis presented in the following theorem.
Theorem 3.5. Let p-adic (p–2) measure-preserving
1-Lipschitz function f:Zp!Zp be presented in the
coordi-nate form (1.4), where
u
kðx0;x1;. . .;xkÞ are such p-valued functions thatu
kðx0;. . .;xkÞ ¼xkAkðx0;. . .;xS1Þ þa
kðx0;. . .;xk1Þ; kPS for some fixed integer S.The function f is ergodic if and only if holds simultaneously
1. f is transitive modulo pS(i.e. f
S1is transitive onZ=pSZ);
2. QpS1
i¼0 Akð½iSÞ 1ðmodpÞ; kPS;
3. for kPSþ1
2p2X
pS1
i¼0
s
ðiÞk
X pS1
i¼0
s
ðiÞk pk
fðiþ1Þð0ÞðmodpSÞ pSk
X
pkS1
b¼0
f fðiÞð0ÞðmodpSÞ þpS
b
#
X0ðmodpÞ;
and for k¼S
0ðmodpÞXX
pS1
i¼0
s
ðiÞS pS½fðf
ðiÞð0ÞðmodpSÞÞ
fðiþ1Þð0ÞðmodpSÞ; where
s
ðiÞk ¼ QpS1
j¼iþ1Akð½fðjÞð0ÞSÞ;i¼0;1;. . .;pk2and
s
ðiÞ k ¼1for i¼pk1.
Proof. Sincefpreserves the measure, then byTheorem 2.1
functions
u
k;½xk andu
0are permutations onZ=pZ. Note that under the Notation(1.2)fork¼S;Sþ1;. . .Fk¼
u
k;fðpk1Þk1 ð0Þ
k
u
k;fðpk2Þ k1 ð0Þ
k
u
k;½0k¼ Ak fkðpk11Þð0Þ
h i
k
xkþ
a
k fðp k1Þ k1 ð0Þh i
k
ðAkð½0kÞ xkþ
a
kð½0kÞÞ ¼xk Y pk1i¼0
Akð½fðiÞ
k1ð0ÞkÞ !
þ
a
k ½fð pk11Þk1 ð0Þk
þ X pk12
i¼0
a
kð½fð iÞk1ð0ÞkÞ
Y pk1
j¼iþ1
Akð½fðjÞ
k1ð0ÞkÞ !!
¼xkQkþLk:
By induction on thek¼S;Sþ1;. . .we show that
Fk;0¼
u
k; fðpk1Þ k1 ð0Þ
k
u
k;½0kare transitive permutations onZ=pZ.
From transitivity of the function fmodulopS follows
that fSðpS1Þð0Þ 0ðmodpSÞ and fðpmþSÞ
Sþm1ð0Þ 0ðmodpSÞ;
m¼1;2;. . .. As by assumption p-valued functions Ak
depend only on variablesx0;x1;. . .;xS1, then any sequence
Akð½0kÞ;Akð½fk1ð0ÞkÞ;. . .;Ak fðp k1Þ k1 ð0Þ
h i
k
n o
;k¼S;Sþ1;. . .
has period of the lengthpS. Then from the second condition
of theorem follows that
Qk¼ Y pk1
i¼0
Ak fkðiÞ1ð0Þ
h i
k
¼Y pk1
i¼0
AkðiÞ ¼ Y pS1
i¼0
Akð½ikÞ !pkS
¼1
Lk¼
a
k fðpk1Þ k1 ð0Þ
h i
k
þX pk2
i¼0
a
k fðiÞ
k1ð0Þ
h i
k
Y
pk1
j¼iþ1
Ak fkðjÞ1ð0Þ
h i
k
!
¼ X pkS1
a¼0
a
k fðp S1þPSaÞk1 ð0Þ
h i
k
þX pS2
i¼0
Y pS1
j¼iþ1
Ak fðjÞ
k1ð0Þ
h i
k
! pXkS1
a¼0
a
k fð iþPSaÞ
k1 ð0Þ
h i
k
!
:
Let
s
ðiÞ k ¼QpS1
j¼iþ1Ak fkðjÞ1ð0Þ
h i
k
;i¼0;1;. . .;pk12;kPS
then
Lk¼ X pkS1
a¼0
a
k fðpS1þPSaÞ
k1 ð0Þ
h i
k
þX pS2
i¼0
s
ðiÞk X pkS1
a¼0
a
k fðiþP SaÞ k1 ð0Þh i
k
!
:
Thus,Fk¼xkþLk;kPS.
Suppose we have already shown thatFk1is transitive onZ=pZ, andfk1is transitive onZ=pkZ. Let us check that
Fk¼xkþLkis transitive onZ=pZ. It is enough to show that Lk–0. Asfis a 1-Lipschitz function and by the induction fk1is transitive onZ=pkZ, then fori2 f0;1;. . .;pS1g
fkðiþ1PSaÞð0Þðmodp
kÞ:
a
¼0;1;. . .;pkS1n o
¼ fkðiÞ1ð0Þðmodp
SÞ þpSb:b¼0;1;. . .;pkS1
n o
ð3:1Þ
andfkðiÞ1ð0Þ fSðiÞ1ð0ÞmodpSÞ.
Then
X pkS1
a¼0
a
k fðiþP SaÞ k1 ð0Þh i
k
¼ X pkS1
b¼0
a
k fSðiÞ1ð0ÞðmodpSÞ þpS
b
h i
k
¼ X pkS1
b¼0
a
kð½fðiÞð0ÞðmodpSÞ þpSbkÞand
Lk¼ X pkS1
a¼0
a
kð½fðp S1Þð0ÞðmodpSÞ þpS
a
kÞþX pS2
i¼0
s
ðiÞk X pkS1
a¼0
a
kð½fðiÞð0ÞðmodpSÞ þpSa
kÞ !:
As by assumptionAk;kPS depend only on variables
x0;x1;. . .;xS1, then
Ak fðjÞ
k1ð0Þ
h i
k
¼Ak fðjÞ
k1ð0Þðmodp
SÞ
h i
k
¼Ak fðjÞ
S1ð0Þ
h i
S
¼Akð½fðjÞð0Þ
SÞ:
And, therefore, fori2 f0;1;. . .;pk12g
s
ðiÞk ¼
Y pS1
j¼iþ1
Ak fSðjÞ1ð0Þ
h i
k
¼ Y pS1
j¼iþ1
Akð½fðjÞð0Þ
SÞ:
Let D2 f0;1;. . .;pS1g and b2 f0;1;. . .;pkS11g.
Fork>S, using coordinate representation of the function
fand relation(3.1)we obtain (under Notations(1.2))
X pkS1
b¼0
fð
D
þpSbÞ ¼ X pkS1b¼0
Xk1
i¼0
pi
u
iþpk
u
kð½D
S;½bkS;0Þ!
¼ X pkS1
b¼0
fð
D
þpSbÞðmodpkÞ
þpk X pkS1
b¼0
a
kð½D
þpSbkÞpkSfð
D
ÞðmodpSÞ þpSpkSðpkS1Þ2
þpk X pkS1
b¼0
a
kð½D
þpSbkÞðmodp kþ1Þ:Then, taking into account thatp–2 and that the element 2 is invertible inZ=pZ(21
2p2
ðmodpÞ), we obtain
X pkS1
b¼0
a
kð½D
þpSbkÞ 2 p21 pk p
kSðfð
D
ÞðmodpSÞÞX
pkS1
b¼0
fð
D
þpSbÞ !ðmodpÞ: ð3:2Þ
In the casek¼S,
fð
D
Þ ¼X S1i¼0
pi
u
iþpS
u
Sð½D
S;0Þfð
D
ÞðmodpSÞ þpSa
Sð½
D
ÞSÞðmodpSþ1
Þ;
or, respectively,
a
Sð½D