ON THE CONVERGENCE TO0 OF mnξ mod 1
BASSAM FAYAD AND JEAN-PAUL THOUVENOT
ABSTRACT. We show that for any irrational numberαand a se- quence of integers{ml}l∈Nsuch that lim
l→∞|||mlα|||=0, there exists a continuous measureµon the circle such that lim
l→∞ Z
T|||mlθ|||dµ(θ) =0.
This implies that any rigidity sequence of any ergodic transfor- mation is a rigidity sequence for some weakly mixing dynamical system.
On the other hand, we show that for anyα ∈ R−Q, there exists a sequence of integers{ml}l∈N such that|||mlα||| → 0 and mlθ[1]is dense on the circle if and only ifθ∈/Qα+Q.
1. INTRODUCTION
We denoteT=R/Z. Forx∈ Rwe denote by|||x||| :=infn∈N|x− n|. We denote byx[1]the fractional part ofx.
In this note, we prove the following results.
Theorem 1. For anyα ∈ R−Q and a sequence of integers{ml}l∈N such that liml→∞|||mlα||| = 0, there exists a measure µ on T which has no atoms such that liml→∞R
T|||mlθ|||dµ(θ) =0.
Theorem 2. For any α ∈ R−Q, there exists a sequence of integers {ml}l∈N such that|||mlα||| → 0 and such thatmlθ[1]is dense in Tif and only ifθ ∈/Qα+Q.
Due to the Gaussian measure space construction (see [4], or for ex- ample Proposition 2.30 in [2]), Theorem 1 has a direct consequence on rigidity sequences of weakly mixing dynamical systems. The fol- lowing is in fact an equivalent statement to Theorem 1.
Corollary 1. For any α ∈ R−Qand a sequence of integers{ml}l∈N such that liml→∞|||mlα||| = 0, there exists a weak mixing dynam- ical system (T,M,m) such that {ml}l∈N is a rigidity sequence for (T,M,m).
Date: January 7, 2014.
Supported by ANR-11-BS01-0004.
1
A consequence of Corollary 1 is a positive answer to a question raised in [2], namely : a rigidity sequence of any ergodic transformation (on a probability space without atoms) with discrete spectrum is a rigidity sequence for some weakly mixing dynamical system. Indeed, Corollary 1 deals with the case of a pure point spectrum with an irrational rota- tion of the circle as a factor. The case of a purely rational spectrum was treated in [2], Proposition 3.27. In case the spectrum is purely rational, our proof of Theorem 1 given below applies with only one modification: instead of working with the orbit of 0 under the rota- tion Rα, (α ∈/ Q/Z), one considers the union of the orbits of 0 under the actions of all the finite groups which appear in the (necessarily dense) support of the spectral measure.
A completely different solution to the same question was given in [1], who proved directly Corollary 1 based on a sophisticated and involved cut and stack construction.
In contrast, our proof is much simpler and is based on the straight- forward characterization of rigidity as a spectral property, which re- duces the answer to the question to the construction of a continuous probability measure on the circle with a Fourier transform converg- ing to 1 along the rigidity subsequence as stated in Theorem 1. This possible approach to the question was discussed in detail in [2].
The second result, Theorem 2, asserts that it is not possible to ex- pect more than what is obtained in Theorem 1, namely that a strong convergence of |||mnθ||| to 1 on an uncountable set K is not possi- ble in general for a sequence{mn}n∈Nsuch that liml→∞|||mlα||| = 0, α ∈ R−Q. Constructing such a setKwas a possible strategy to prov- ing Corollary 1 (see for example Proposition 3.3 in [2]), and Theorem 2 shows that this approach cannot be adopted in general.
Given an increasing sequence of integers {mn}n∈N, the study of the accumulation points on the circle of the sequence {mnξ}, for ξ irrational, has a long history and a rich literature (see for example [3, 5] and references therein). Weyl [6] proved that for any increasing sequence{mn}n∈N, it holds that for almost everyξ, {mnξ} is dense on the circle. The set of irrationalsξsuch that{mnξ}is not dense inT is called the set of exceptional points for the sequence{mn}n∈N. Our result asserts the existence for anyα ∈R−Qof a sequence{mn}n∈N for which the set of exceptional points is reduced to Qα+Q. To our knowledge, no other examples of increasing sequences{mn}n∈N with a countable exceptional set are known in the literature.
ACKNOWELEDGMENT. The authors are grateful to the referee for suggesting improvements to the first version of the paper.
ON THE CONVERGENCE TO 0 OF nξ 1 3
2. PROOF OF THEOREM1
Fixα ∈ R−Qand a sequence of integers{ml}l∈Nsuch that
(?) lim
l→∞|||mlα||| =0.
For a probability measureµonTwe writeµn =|R
T|||mnθ|||dµ(θ)|. We will construct a sequence µp, p ≥ 0, of probability measures on Tof the form 21p ∑2i=p1δxi with xi = kiα such that there exists an increasing sequence{Np} for which
(1) For every p ≥ 1, for every j ∈ [0,p−1], for every n ∈ [Nj,Nj+1],µnp< 1
2j
(2) For every p0 ∈N∗, if we let ηp0 = 1
4 inf
1≤i<i0≤2p0|||kiα−ki0α|||
then for everyl ∈ Nand everyr ∈[1, 2p0],|||kl2p0+rα−krα||| <
ηp0
(3) µnp < 1
2p+1 forn≥ Np.
When going from the measure µp toµp+1 we will add2p masses at points selected nearbyx1, . . . ,x2p that are already chosen forµp.
Theorem 1 clearly follows from the above construction. Indeed, property (1) will imply that any weak limitµ∞ofµpsatisfiesµn∞ →0.
While by (2) we get that for each p0 the intervals (krα−ηp0,krα+ ηp0),r ∈ [1, 2p0], on the circle are disjoint and each have mass 1/2p0 for everyµp, p≥ p0, and hence forµ∞that has therefore no atoms.
Property (3) is not necessary in the proof of the theorem, but it is useful to fulfill the inductive hypotheses (1) and (2) of the construc- tion.
For p=0, we letk1 =0 andµ0is thus the Dirac measure at 0. We letN0 =0. Forp =1, we letk2 =1 soµ1is the average of the Dirac measures at 0 and atα. Observe that for any n,µn1 < 12 which fulfills (1) for p=1. We also chooseN1sufficiently large so thatµn1 < 212 for n ≥N1, the latter being possible due to(?).
We now assume selected ki fori ≤ 2p and Nl for l ≤ p such that (1) and (3) are satisfied up to p, and (2) is satisfied for every p0 ≤ p and every 0≤l ≤2p−p0 −1.
We choosek2p+1such thatk2p+1αis sufficiently close tok1αso that νp,1 = 1
2p+1
2p+1 i
∑
=1δk0
iα
where k0i = ki for i ≤ 2p and k02p+1 = k2p+1 while k02p+r = kr for r ∈ [2, 2p]satisfiesνnp,1 < 1
2j for everyn∈ [Nj,Nj+1]andj∈ [0,p−1]. Since for every n we have that |νnp,1−µnp| < 1
2p+1|||mnk2p+1α− mnk1α||| < 1
2p+1 we get by (3) that νnp,1 < 1
2p+1 + 1
2p+1 < 21p for ev- eryn ≥ Np. Next we choose, Np,1 > Np sufficiently large such that νp,1n < 2p+21 for n ≥ Np,1, which is possible by (?). In this way, we select inductivelyk2p+s, then Np,s fors =1, 2, . . . , 2p, and set
νp,s = 1 2p+1
2p+1 i
∑
=1δk0
iα
wherek0i =kifori≤2p+sandk02p+t = kt fort ∈ [s+1, 2p]. Choos- ing for each s, k2p+sα sufficiently close to ksα then Np,s sufficiently large we can insure that
• νnp,s < 1
2j for everyn∈ [Nj,Nj+1]andj ≤ p−1.
• νnp,s < 21p for everyn ≥Np
• νnp,s < 2p+21 for everyn ≥Np,s
The first point can be established inductively due to the fact that if k2p+sα is chosen very close toksα then the measuresνp,s−1 and νp,s
are very close.
The same argument gives the second point for Np ≤ n ≤ Np,s−1. As forn ≥ Np,s−1 we use the fact that |νnp,s−νnp,s−1| < 1
2p+1 and the fact thatνnp,s−1 < 2p+21 for everyn≥ Np,s−1to conclude thatνnp,s < 21p. For the third point we just choose Np,s sufficiently large and use(?). Finally, we let Np+1 =Np,2p andµp+1 =νp,2p and observe that the measureµp+1satisfies (1).
Also, since k2p+sα can be chosen arbitrarily close to ksα for s = 1, . . . , 2p, we get that for every p0 ≤ p+1, for every l = 2p−p0 + l0−1, l0 ≤ 2p−p0, |||kl2p0+rα||| ∼ |||kl02p0+rα||| ∼ |||krα||| from where (2) follows for p+1. The proof of Theorem 1 is thus complete.
3. PROOF OF THEOREM2 In all this sectionα∈ R−Qis fixed.
Definition1. For an interval I ⊂T, forε > 0, and integers N1 < N2, we say thatθ ∈ A(N1,N2,I,ε,α) if for everym ∈ [N1,N2)such that
|||mα||| <εwe have that{mθ} ∈/ I.
Lemma1. For everyl ≥2, there existsL(l) ∈ N, such that for every 0 < ε ≤ 2l12, for everyν > 0, N ∈ N, there exist K(ε) >0 and N0 =
ON THE CONVERGENCE TO 0 OF nξ 1 5
N0(l,ε,ν,N) ∈ Nsuch that θ ∈ A(N,N0,I,ε,α)for some interval I of size 1/l, implies that|||kα−sθ||| <νfor some|k| ≤K(ε)and some
|s| ≤ L(l).
Proof. For any ε > 0, consider an approximation by trigonometric polynomials of 2χε whereχεis the characteristic function of the sub- set ofT[0,ε]∪[1−ε, 1], namelyφε :T→Rsuch that
• φε(x)>1 for everyx ∈ [0,ε]∪[1−ε, 1]
• φε(x)>−ε3for everyx ∈T
• There existsK ∈Nsuch thatφε(x) =∑|k|≤Kφˆkei2πkx. Similarly, forl ≥2, letϕl :T→Rbe such that
• ϕl(y)>1 for everyy∈/[0,1l]
• |ϕl(y)| <l2for everyy∈ T
• There exists L∈ Nsuch that ϕl(y) = ∑0<|k|≤L ϕˆkei2πky. Note that the second requirement includes the fact thatR
ϕl(y)dy = 0.
Forψ :T2 →Rand (α,θ)∈ R2we define fork∈ N Sα,θk ψ(x,y) =
k−1 i
∑
=0ψ(x+iα,y+iθ). Fix I = [y0,y0+1l], for somey0∈ T,l ≥2.
Defineψε,l : T2 → Rbyψε,l(x,y) = φε(x)ϕl(y−y0). For N0 ∈ N sufficiently large there exist more thanε2N0integersi ∈[N,N0)such that |||iα||| < ε. If θ ∈ A(N,N0,I,ε,α) then Sα,θN0ψε,l(0, 0) > (ε2− l2ε3)N0 ≥ 12ε2N0.
On the other hand, we have that Sα,θN0ψε,l(x,y) =
∑
|k|≤K,0<|j|<L
φˆkϕˆj1−ei2πN0(kα+jθ)
1−ei2π(kα+jθ) ei2π(kx+jy) hence, if |||kα−jθ||| ≥ ν for every |k| ≤ K and every 0 < |j| ≤ L, thenSα,θN0ψε,l(x,y)is bounded independently ofN0which contradicts Sα,θN0ψε,l(0, 0)> 12ε2N0.
Proof of Theorem 2. Define for n ≥ 1, the sequenceln = n+1 and Ln := L(ln) given by Lemma 1. Let εn = 2(n+11)2 and Kn = K(εn) given by Lemma 1. Letνn = 1n inf0<|k|≤(n+1)K
n+1|||kα|||. TakeN0 = 0 and apply Lemma 1 withl =l1,ε =ε1, N = N0andν =ν1. Denote then N1 = N0(l1,ε1,ν1,N0). We then apply inductively Lemma 1
withl = ln, ε =εn, N = Nn and ν =νn and choose Nn+1arbitrarily large such that Nn+1 ≥N0(ln,εn,νn,Nn).
We define an increasing sequence ml by taking successively for everyi, all the integersm∈ [Ni,Ni+1)such that|||mα||| <εi(choosing Nn+1to be sufficiently large in our inductive construction guarantees that the sequencemnis not empty).
Suppose nowθis such thatmnθ[1]is not dense on the circle. Then, there existskand an intervalIof sizelksuch that for everyn,mnθ[1]∈/ I. In other words,θ ∈ A(Nn,Nn+1,I,εn,α) for everyn ≥ n0, n0 suf- ficiently large. LetL =Lk. By Lemma 1 we get that|||knα−lθ||| <νn for some|kn| ≤ Kn and some 0<|l| ≤ L. Hence|||k0nα−L!θ||| <L!νn
for some|k0n| ≤ L!Kn. It follows that|||(k0n+1−k0n)α||| <2L!νn. From the definition ofνn this implies that k0n+1 = k0n for sufficiently large n, say n ≥ n1. Since νn → 0, we get that|||k0n1α−L!θ||| = 0, which givesθ ∈ Qα+Q. Conversely, {mnθ} forθ ∈ Qα+Qis clearly not dense on the circle and Theorem 2 is proved.
REFERENCES
[1] T. Adams,Tower multiplexing and slow weak mixing, arXiv:1301.0791.
[2] V. Bergelson, A. Del Junco, M. Lemanczyk, J. Rosenblatt, Rigidity and non- recurrence along sequences, Ergodic Theory and Dynamical Systems, First View Article (2013), 1-39.
[3] Y. Bugeaud, On sequences (a(n)ξ),n > 0, converging modulo 1, Proc. Amer.
Math. Soc. 137 (2009), 2609–2612.
[4] I. P. Cornfeld, S. V. Fomin, and Ya. G. Sinai, Ergodic Theory, Springer-Verlag, Berlin-Heidelberg-New York, 1982.
[5] A. Dubickas,On the limit points of(anξ)∞n=1mod 1 for slowly increasing integer sequences(an)∞n=1, Proceedings of the American Mathematical Society, 137 (2) (2009), 449–456.
[6] H. Weyl,Uber die Gleichverteilung von Zahlen mod. Eins, Math. Ann. 77 (1916), 313–352.
IMJ-PRG, CNRS, UP7D-CAMPUSGRANDMOULIN, B ˆATIMENTSOPHIEGER-
MAIN, CASE7012, 75205 PARISCEDEX13,BASSAM@MATH.JUSSIEU.FR
LPMA, UNIVERSITE´ PIERRE ET MARIE CURIE, 4 PL JUSSIEU, 75252 PARIS,
JEAN-PAUL.THOUVENOT@UPMC.FR