Abstract
The aim of the present article is twofold. We first give a survey on recent developments on the distribution of symbols in polynomial subsequences of the Thue–Morse sequence \(\mathbf {t}=(t(n))_{n\ge 0}\) by highlighting effective results. Secondly, we give explicit bounds on
for odd integers p, q, and on
where \(h_1, h_2\ge 1\), and \((\varepsilon _1, \varepsilon _2)\) is one of (0, 0), (0, 1), (1, 0), (1, 1).
Work supported by the ANR-FWF bilateral project MuDeRa “Multiplicativity: Determinism and Randomness” (France-Austria) and the joint project “Systèmes de numération : Propriétés arithmétiques, dynamiques et probabilistes” of the Université de Lorraine and the Conseil Régional de Lorraine.
Access provided by Autonomous University of Puebla. Download conference paper PDF
Similar content being viewed by others
Keywords
1 Introduction
The Thue–Morse sequence
can be defined via
where \(s_2(n)\) denotes the number of one bits in the binary expansion of n, or equivalently, the sum of digits of n in base 2. This sequence can be found in various fields of mathematics and computer science, such as combinatorics on words, number theory, harmonic analysis and differential geometry. We refer the reader to the survey articles of Allouche and Shallit [2], and of Mauduit [14] for a concise introduction to this sequence. As is well-known, Thue–Morse is 2-automatic and can be generated by the morphism \(0\mapsto 01\), \(1\mapsto 10\). It is also the prime example of an overlapfree sequence.
The overall distribution of the symbols 0 and 1 in Thue–Morse is trivial since the sequence consists exclusively of consecutive blocks of the forms 01 and 10, thus there are “as many 0’s as 1’s” in the sequence. The investigation of Thue–Morse along subsequences can be said to have started with an influential paper by Gelfond in 1967/68 [9]. He proved, via exponential sums techniques, that \(\mathbf {t}\) is uniformly distributed along arithmetic progressions, namely,
with an explicit error term. Gelfond’s result shows that there are “as many 0’s as 1’s” in the sequence also regarding arithmetic progressions. His result, however, gives no information on how long one actually has to wait to “see” the first, say, “1” along a specific arithmetic progression. Newman [19] showed that there is a weak preponderance of the 0’s over the 1’s in the sequence of the multiples of three. More precisely, he showed that
with \(c'_1 N^{\log _4 3} < C(N)< c'_2 N^{\log _4 3}\) for all \(N\ge 1\) and certain positive constants \(c'_1, c'_2\). For the multiples of three one has to wait for 7 terms to “see the first 1”, i.e.,
Morgenbesser, Shallit and Stoll [18] proved that for \(p\ge 1\),
and this becomes sharp for \(p=2^{2r}-1\) for \(r\ge 1\) (Note that \(3=2^{2\cdot 1}-1\) is exactly of that form). A huge literature is nowadays available for classes of arithmetic progressions where such Newman-type phenomena exist and many generalizations have been considered so far (see [3, 4, 8, 10, 12, 22] and the references given therein). Still, a full classification is not yet at our disposal.
Most of the results that hold true for Thue–Morse in the number-theoretic setting of (1) have been proven for the sum of digits function in base q, where q is an integer greater than or equal to 2, and where the reduction in (1) is done modulo an arbitrary integer \(m\ge 2\). We refrain here from the general statements and refer the interested readers to the original research papers.
Following the historical line, Gelfond [9] posed two challenging questions concerning the distribution of the sum of digits function along primes and along polynomial values instead of looking at linear subsequences. A third question was concerned with the simultaneous distribution when the sum of digits is taken to different bases; this question has been settled by Kim [13]. In recent years, this area of research gained much momentum due to an article by Mauduit and Rivat [15] who answered Gelfond’s question for primes with an explicit error term. In a second paper [16], they also answered Gelfond’s question for the sequence of squares. Their result implies that
In a very recent paper, Drmota, Mauduit and Rivat [6] showed that \(\mathbf {t}\) along squares gives indeed a normal sequence in base 2 meaning that each binary block appears with the expected frequency. This quantifies a result of Moshe [17] who answered a question posed by Allouche and Shallit [1, Problem10.12.7] about the complexity of Thue–Morse along polynomial extractions.
We are still very far from understanding
where \(P(x)\in \mathbb {Z}[x]\) is a polynomial of degree \(\ge 3\). Drmota, Mauduit and Rivat [7] obtained an asymptotic formula for \(\#\{n<N: \; s_q(P(n))=\varepsilon \pmod m\}\) whenever q is sufficiently large in terms of the degree of P. The case of Thue–Morse is yet out of reach of current technology. The currently best result is due to the author [23], who showed that there exists a constant \(c=c(P)\) depending only on the polynomial P such that
This improves on a result of Dartyge and Tenenbaum [5] who had \(N^{2/(\deg P)!}\) for the lower bound. The method of proof for (2) is constructive and gives an explicit bound on the minimal non-trivial n such that \(t(n^h)=\varepsilon \) for fixed \(h\ge 1\). Since \(t(n^h)=1\) for all \(n=2^r\), and \(t(0^h)=0\), we restrict our attention to
From the proof of (2) follows that
Hence, there exists an absolute constant \(c_1>0\) such that
With some extra work, a similar result can be obtained for a general polynomial P(n) instead of \(n^h\), where there corresponding constant will depend on the coefficients of P.
The joint distribution of the binary digits of integer multiples has been studied by J. Schmid [20] and in the more general setting of polynomials by Steiner [21]. The asymptotic formulas do not imply effective bounds on the first n that realizes such a system and it is the aim of this paper to prove effective bounds in the case of two equations for integer multiples and for monomials.
Our first result is as follows.
Theorem 1
Let \(p>q\ge 1\) be odd integers. Then there exists an absolute constant \(c_2>0\) such that
Remark 1
Note that for \(p=2^r+1\) with \(r\ge 1\) we have \(t(pn)=0\) for all \(n<p-1\), so that there is no absolute bound for the minimal n.
There are examples that show that sometimes one has to “wait” quite some time to see all of the four possibilities for \((\varepsilon _1,\varepsilon _2)\) when the extraction is done along two monomial sequences. For instance, we have
for \(\varepsilon _1, \varepsilon _2=0,1\) but
The construction that we will use to prove Theorem 1 will not be useful to study the minimal n along polynomial subsequences since in this case we would need to keep track of the binary digits sum of various binomial coefficients. Instead, we will use ideas from work of Hare, Laishram and the author [11] to show the following result.
Theorem 2
Let \(h_1>h_2\ge 1\) be integers. Then there exists an absolute constant \(c_3>0\) such that
Remark 2
The method also allows to treat general monic polynomials \(P_1(x)\), \(P_2(x)\in \mathbb {Z}[x]\) of different degree \(h_1, h_2\) in place of \(x^{h_1}, x^{h_2}\). Even more generally, we can deal with non-monic polynomials \(P_1(x), P_2(x)\in \mathbb {Z}[x]\) provided \(h_1\) is odd. As we will see in the proof (compare with the remark after (14)), the latter condition relies on the fact that for odd h the congruence \(x^h\equiv a \pmod {16}\) admits a solution mod 16 for each odd a, while this is not true in general if h is even.
We write \(\log _2 \) for the logarithm to base 2. Moreover, for \(n=\sum _{j=0}^{\ell } n_j 2^j\) with \(n_j\in \{0,1\}\) and \(n_\ell \ne 0\) we write \((n_\ell ,n_{\ell -1},\cdots ,n_1,n_0)_2\) for its digital representation in base 2 and set \(\ell =\ell (n)\) for its length. To simplify our notation, we allow to fill up by a finite number of 0’s to the left, i.e., \((n_\ell n_{\ell -1}\cdots n_1 n_0)_2=(0n_\ell n_{\ell -1} \cdots n_1 n_0)_2=(0\cdots 0 n_\ell n_{\ell -1} \cdots n_1 n_0)_2\).
The paper is structured as follows. In Sect. 2 we prove Theorem 1 and in Sect. 3 we show Theorem 2.
2 Thue–Morse at Distinct Multiples
The proof of Theorem 1 is based on the following lemma.
Lemma 1
Let p, q be odd positive integers with \(p>q\ge 1\) and let \((\varepsilon _1,\varepsilon _2)\) be one of (0, 0), (0, 1), (1, 0), (1, 1). Then we have
where
Proof
Recall that for \(1\le b<2^k\) and \(a,k\ge 1\), we have
In the sequel we will make frequent use of these splitting formulas. We first deal with the two cases when \((\varepsilon _1,\varepsilon _2)\) is one of (0, 0), (1, 1). If \(2^k>p>q\) then \(s_2(p(2^k-1))=s_2(q(2^k-1))=k\). Moreover, since \(k\equiv 0\) or 1 mod 2 and
we get that \(C_{0,0}(p,q),C_{1,1}(p,q)\le 4p\). Finding explicit bounds for \(C_{0,1}(p,q)\) and \(C_{1,0}(p,q)\) is more involved. To begin with, we first claim that there exists \(n_1\ge 1\) with the following two properties:
-
(a)
\(\ell (pn_1)>\ell (qn_1)\),
-
(b)
\(pn_1\equiv 1 \mathrm mod4\).
As for (a), we need to find two integers \(a, n_1\) such that \(2^a\le pn_1\) and \(2^a> qn_1\). This is equivalent to
For odd k, n either \(kn\equiv 1\) (mod 4) or \(k(n+2)\equiv 1\) (mod 4), so provided \(2^a\left( \frac{1}{q}-\frac{1}{p}\right) \ge 4\), we can find an odd \(n_1\) that satisfies both (a) and (b). By taking a to be the unique integer with
we get an \(n_1\) with
Now, define \(n_2=2^{\ell (pn_1)}+1\). Since both p and \(n_1\) are odd we have \(n_2\le pn_1\). Then
since there is exactly one carry propagation from the most significant digit of \(pn_1\) to the digit at digit place \(\ell (pn_1)\) of \(pn_1 2^{\ell (pn_1)}\) which stops immediately after one step because of property (b). On the other hand, (a) implies that
because the terms \(qn_1\) and \(qn_1 2^{\ell (pn_1)}\) do not interfere and there is therefore no carry propagation while adding these two terms. We therefore can set
to get that \(C_{1,0}(p,q)\le 2^6\cdot \frac{p^3}{(p-q)^2}\). For \((\varepsilon _1,\varepsilon _2)=(0,1)\), take m to be the unique odd integer with
and put
Then by (7),
A similar calculation shows by (8) that
We set \(n=n_1n_2n_3\) and get
This completes the proof. \(\square \)
Proof
(Theorem 1 ). This follows directly from Lemma 1 and
\(\square \)
3 Thue–Morse at Two Polynomials
This section is devoted to the proof of a technical result which implies Theorem 2. Considering the extractions of Thue–Morse along \(n^{h_1}\) and \(n^{h_2}\), it is simple to get two and not too difficult to get three out of the four possibilities for \((\varepsilon _1,\varepsilon _2)\). However, to ensure that we see all of the four possibilities, we need a rather subtle construction. The difficulty is similar to that one to get \(C_{0,0}(p,q)\) in the proof of Theorem 1. The idea is to shift two specific blocks against each other while all other terms in the expansions are non-interfering. Via this procedure, we will be able to keep track of the number of carry propagations. In the proof of Theorem 1 we have used the blocks \(pn_1\) and \(qn_1\). In the following, we will make use of \(u_1=9=(1001)_2\) and \(u_2=1=(0001)_2\). Then
and mod 2 we get the sequence (0, 1, 1, 0). These particular expansions and additions will be of great importance in our argument (compare with (16)–(19)).
Lemma 2
Let \(h_1\), \(h_2\) be positive integers with \(h_1>h_2\ge 1\) and let \((\varepsilon _1,\varepsilon _2)\) be one of (0, 0), (0, 1), (1, 0), (1, 1). Then we have
where
Proof
Let \(h\ge 1\) and put
It is straightforward to check that for all \(h\ge 1\), we have
Obviously, we have \(b\ge M\). Moreover,
and therefore \(a\ge b\). Furthermore, for \(h\ge 2\), we have \(aM>b^2\) since by
we have
Let
and write \(T(x)^h=\sum _{i=0}^{5h} \alpha _i x^i\). Obviously, \(\alpha _0>0\) and \(\alpha _1<0\). We claim that for \(h\ge 1\) we have \(\alpha _i>0\) for \(2\le i\le 5h\). To see this we write
with
Since \(a\ge b\ge M\) the coefficient of \(x^i\) in the first term in (11) is \(\ge M^h\). On the other hand,
and
which proves the claim. Next, we need a bound on the size of \(\alpha _i\). The coefficients \(\alpha _i\), \(0\le i\le 5h-2\), are bounded by the corresponding coefficients in the expansion of \((ax^5+bx^4+M(x^3+x^2+x+1))^h\). Since \(aM>b^2\) and \(M\le 16 a^{1-1/h}\), each of these coefficients is bounded by
and therefore
Moreover, we have
and
which is true for all \(a\ge 1\) and \(h\ge 1\). Note that both the bound in (12) and the coefficient \(\alpha _{5h-1}\) are increasing functions in h. From now on, suppose that \(h\ge 2\). We further claim that
which will give us the wanted overlap for the digital blocks of \(\alpha _{5h-1}\) and \(\alpha _{5h}\). By (14) the binary expansion of \(\alpha _{5h-1}\) is \((1001\cdots )_2\) and interferes with the digital block coming from \(\alpha _{5h}= a^h\) which is \((\cdots 0001)_2\) since \(a\equiv 1 \pmod {16}\). To prove (14), we show a stronger inequality that in turn implies (14), namely,
Passing to logarithms, this is equivalent to
with
We rewrite
which on the one hand shows that
and on the other hand by \(-x/(1-x)<\log (1-x)<0\) for \(0<x<1\) that
Finally, we easily check that for all \(h\ge 2\) and \(l\ge 5\) we have
which finishes the proof of (15) and thus of (14).
After this technical preliminaries we proceed to the evaluation of the sum of digits. First note that for all \(h\ge 1\) by construction none of \(n=T(2^k)\), \(T(2^{k+1})\), \(T(2^{k+2})\), \(T(2^{k+3})\) is a power of two and therefore \(n\in \mathcal {A}\) in these four cases. Let \(h=h_1\ge 2\) and define a, b, M, l, k according to (9). To begin with, by (12)–(14) and the splitting formulas (4), we calculate
where
Note that the summand k in (16) comes from formula (4) due to the negative coefficient \(\alpha _1\), and the \(-1\) comes from the addition of \((\cdots 0001)_2\) and \((1001)_2\) (the ending and starting blocks corresponding to \(\alpha _{5h_1}\) and \(\alpha _{5h_1-1}\)) which gives rise to exactly one carry. A similar calculation shows that
In (17) we add \((1001)_2\) to \((\cdots 00010)_2\) which gives no carry. The same happens for the addition of \((1001)_2\) to \((\cdots 000100)_2\) in (18). Finally, in (19) we again have exactly one carry in the addition of \((1001)_2\) to \((\cdots 0001000)_2\). If we look mod 2 this shows that
is either (0, 0, 1, 1) or (1, 1, 0, 0). For \(h_2<h_1\) with \(h_2\ge 1\) all coefficients are non-interfering. To see this, consider the coefficients of \(T(x)^{h_2}=\sum _{i=0}^{5h_2} \alpha '_i x^i\). By (12), they are clearly bounded in modulus by \(a^{h_1-\frac{1}{h_1}}\cdot 16 \cdot 8^{h_1}<2^k\) for \(i=0,1,2,\ldots ,5h_2-2\). Also, by (10) and \(h_1\ge 2\),
and thus we don’t have carry propagations in the addition of terms in the expansion of \(T(2^k)^{h_2}\). Similarly, we show that
is either (0, 1, 0, 1) or (1, 0, 1, 0). This yields in any case that
which are the four desired values. Finally,
which completes the proof. \(\square \)
Proof
(Theorem 2 ). This follows from Lemma 2 and
for some suitable positive constant \(c_3\). \(\square \)
References
Allouche, J.-P., Shallit, J.: Automatic Sequences: Theory, Applications Generalizations. Cambridge University Press, Cambridge (2003)
Allouche, J.-P., Shallit, J.: The ubiquitous Prouhet-Thue-Morse sequence. In: Ding, C., Helleseth, T., Niederreiter, H. (eds.) Sequences and Their Applications. Discrete Mathematics and Theoretical Computer Science, pp. 1–16. Springer, London (1999)
Boreico, I., El-Baz, D., Stoll, T.: On a conjecture of Dekking: the sum of digits of even numbers. J. Théor. Nombres Bordeaux 26, 17–24 (2014)
Coquet, J.: A summation formula related to the binary digits. Invent. Math. 73, 107–115 (1983)
Dartyge, C., Tenenbaum, G.: Congruences de sommes de chiffres de valeurs polynomiales. Bull. London Math. Soc. 38(1), 61–69 (2006)
Drmota M., Mauduit C., Rivat J.: The Thue-Morse sequence along squares is normal, manuscript. http://www.dmg.tuwien.ac.at/drmota/alongsquares.pdf
Drmota, M., Mauduit, C., Rivat, J.: The sum of digits function of polynomial sequences. J. London Math. Soc. 84, 81–102 (2011)
Drmota, M., Skałba, M.: Rarified sums of the Thue-Morse sequence. Trans. Am. Math. Soc. 352, 609–642 (2000)
Gelfond, A.O.: Sur les nombres qui ont des propriétés additives et multiplicatives données. Acta Arith. 13, pp. 259–265 (1967/1968)
Goldstein, S., Kelly, K.A., Speer, E.R.: The fractal structure of rarefied sums of the Thue-Morse sequence. J. Number Theor. 42, 1–19 (1992)
Hare, K.G., Laishram, S., Stoll, T.: Stolarsky’s conjecture and the sum of digits of polynomial values. Proc. Am. Math. Soc. 139, 39–49 (2011)
Hofer, R.: Coquet-type formulas for the rarefied weighted Thue-Morse sequence. Discrete Math. 311, 1724–1734 (2011)
Kim, D.-H.: On the joint distribution of \(q\)-additive functions in residue classes. J. Number Theor. 74, 307–336 (1999)
Mauduit, C.: Multiplicative properties of the Thue-Morse sequence. Period. Math. Hungar. 43, 137–153 (2001)
Mauduit, C., Rivat, J.: Sur un problème de Gelfond: la somme des chiffres des nombres premiers. Ann. Math. 171, 1591–1646 (2010)
Mauduit, C., Rivat, J.: La somme des chiffres des carrés. Acta Math. 203, 107–148 (2009)
Moshe, Y.: On the subword complexity of Thue-Morse polynomial extractions. Theor. Comput. Sci. 389, 318–329 (2007)
Morgenbesser, J., Shallit, J., Stoll, T.: Thue-Morse at multiples of an integer. J. Number Theor. 131, 1498–1512 (2011)
Newman, D.J.: On the number of binary digits in a multiple of three. Proc. Am. Math. Soc. 21, 719–721 (1969)
Schmid, J.: The joint distribution of the binary digits of integer multiples. Acta Arith. 43, 391–415 (1984)
Steiner W.: On the joint distribution of \(q\)-additive functions on polynomial sequences. In: Proceedings of the Conference Dedicated to the 90th Anniversary of Boris Vladimirovich Gnedenko (Kyiv, 2002), Theory Stochastic Process, 8, pp. 336–357 (2002)
Stoll T.: Multi-parametric extensions of Newman’s phenomenon. Integers (electronic), 5, A14, p. 14 (2005)
Stoll, T.: The sum of digits of polynomial values in arithmetic progressions. Funct. Approx. Comment. Math. 47, 233–239 (2012). part 2
Acknowledgements
I am pleased to thank Jeff Shallit for bringing to my attention the question on bounding Thue–Morse on two multiples. I also thank him and E. Rowland for several discussions.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2015 Springer International Publishing Switzerland
About this paper
Cite this paper
Stoll, T. (2015). Thue–Morse Along Two Polynomial Subsequences. In: Manea, F., Nowotka, D. (eds) Combinatorics on Words. WORDS 2015. Lecture Notes in Computer Science(), vol 9304. Springer, Cham. https://doi.org/10.1007/978-3-319-23660-5_5
Download citation
DOI: https://doi.org/10.1007/978-3-319-23660-5_5
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-23659-9
Online ISBN: 978-3-319-23660-5
eBook Packages: Computer ScienceComputer Science (R0)