Hostname: page-component-78c5997874-4rdpn Total loading time: 0 Render date: 2024-11-19T05:24:00.179Z Has data issue: false hasContentIssue false

EVERY ARITHMETIC PROGRESSION CONTAINS INFINITELY MANY b-NIVEN NUMBERS

Published online by Cambridge University Press:  31 July 2023

JOSHUA HARRINGTON
Affiliation:
Department of Mathematics, Cedar Crest College, 100 College Dr, Allentown, PA 18104, USA e-mail: joshua.harrington@cedarcrest.edu
MATTHEW LITMAN
Affiliation:
Department of Mathematics, University of California, Davis, 1 Shields Ave, Davis, CA 95616, USA e-mail: mclitman@ucdavis.edu
TONY W. H. WONG*
Affiliation:
Department of Mathematics, Kutztown University of Pennsylvania, 15200 Kutztown Rd, Kutztown, PA 19530, USA
Rights & Permissions [Opens in a new window]

Abstract

For an integer $b\geq 2$, a positive integer is called a b-Niven number if it is a multiple of the sum of the digits in its base-b representation. In this article, we show that every arithmetic progression contains infinitely many b-Niven numbers.

Type
Research Article
Copyright
© The Author(s), 2023. Published by Cambridge University Press on behalf of Australian Mathematical Publishing Association Inc.

1 Introduction

Let $\mathbb {N}$ denote the set of positive integers and let $b\geq 2$ be an integer. For all $n\in \mathbb {N}$ and $0\leq i\leq \lfloor \log _bn\rfloor $ , let $\nu _b(n,i)$ be nonnegative integers such that $\nu _b(n,i)\leq b-1$ and $n=\sum _{i=0}^{\lfloor \log _bn\rfloor }\nu _b(n,i)b^i$ . In other words, $\nu _b(n,i)$ is the $(i+1)$ st digit from the right in the base-b representation of n. Furthermore, define $s_b:\mathbb {N}\to \mathbb {N}$ by $s_b(n)=\sum _{i=0}^{\lfloor \log _bn\rfloor }\nu _b(n,i)$ . A positive integer n is b-Niven if $s_b(n)\mid n$ .

It was shown in 1993 by Cooper and Kennedy [Reference Cooper and Kennedy1] that there are no 21 consecutive $10$ -Niven numbers. Their result was generalised in 1994 by Grundman [Reference Grundman4], who showed that there are no $2b+1$ consecutive b-Niven numbers. In 1994, Wilson [Reference Wilson6] proved that for each b, there are infinitely many occurrences of $2b$ consecutive b-Niven numbers. These results were recently extended by Grundman et al. [Reference Grundman, Harrington and Wong5], who investigated the maximum lengths of arithmetic progressions of b-Niven numbers. Furthermore, an asymptotic estimate for the number of b-Niven numbers not exceeding x was found in 2003 by De Koninck et al. [Reference De Koninck, Doyon and Kátai2] and in 2008, they [Reference De Koninck, Doyon and Kátai3] showed that given any $r\in \{2,3,\ldots ,2b\}$ , there exists a constant $c=c(b,r)$ such that the number of r-tuples of consecutive b-Niven numbers not exceeding x is asymptotic to $cx/(\log x)^r$ as x tends to infinity.

In this article, we prove that every arithmetic progression contains infinitely many b-Niven numbers.

2 Main results

The following lemma is sometimes referred to as the ‘postage stamp theorem’, the ‘chicken McNugget theorem’ or the ‘Frobenius coin theorem’.

Lemma 2.1. Let u and v be integers with $uv\geq 0$ and $\gcd (u,v)=1$ . Then every integer w such that w shares the same sign with u and v and satisfies $|w|\geq (|u|-1)(|v|-1)$ can be written in the form $w=gu+hv$ for some nonnegative integers g and h.

The following two lemmas, which will be useful in our proof, are easy exercises in elementary number theory.

Lemma 2.2. If $d\mid b-1$ , then for all $u\in \mathbb {N}$ , we have $d\mid u$ if and only if $d\mid s_b(u)$ .

Lemma 2.3. For all integers n and $n'$ with $2\leq n'\leq n$ , $s_b(n')\leq (b-1)\lceil \log _b(n)\rceil $ .

For positive integers m and r, let

$$ \begin{align*}\mathcal{S}_{m,r}=\{mx+r:x\in\mathbb{N}\}.\end{align*} $$

Proposition 2.4. Let $d=\gcd (s_b(m),s_b(r),b-1)$ . If $\gcd (s_b(m),s_b(r))=d$ , then $\mathcal {S}_{m,r}$ contains at least one b-Niven number.

Proof. Let $k_0(b,m,r)\in \mathbb {N}$ be such that for all integers $k\geq k_0$ ,

(2.1) $$ \begin{align} k\geq(b-1)\bigg\lceil\log_b\bigg(\frac{s_b(m)}{d}\cdot k+\frac{s_b(r)}{d}\bigg)\bigg\rceil+(b-2)\bigg((b-1)\bigg\lceil\log_b\bigg(\frac{s_b(m)}{d}\cdot k+\frac{s_b(r)}{d}\bigg)\bigg\rceil-1\bigg). \end{align} $$

Note that $k_0$ is well defined since $b,m,r$ are constants and the right-hand side of (2.1) is of order $O(\log k)$ . Using Dirichlet’s theorem on primes in arithmetic progressions, let $k\in \mathbb {N}$ be such that $k\geq \max \{k_0,b,m\}$ and

$$ \begin{align*}p=\frac{s_b(m)}{d}\cdot k+\frac{s_b(r)}{d}\end{align*} $$

is a prime. Since $p>k\geq \max \{b,m\}$ , we have $p\nmid bm$ . Furthermore, let $\tilde {x}$ be the smallest positive integer such that $\tilde {x}\equiv -m^{-1}r\text { (mod }p\text {)}$ . From Lemma 2.3 and (2.1),

$$ \begin{align*}k\geq s_b(\tilde{x})+(b-2)(s_b(p)-1).\end{align*} $$

By Lemma 2.1, there exist nonnegative integers g and h such that

$$ \begin{align*}k=s_b(\tilde{x})+g(b-1)+h\cdot s_b(p).\end{align*} $$

Let $\omega {\kern-1.3pt}\in{\kern-1.3pt} \mathbb {N}$ be a multiple of $p{\kern-1.3pt}-{\kern-1.3pt}1$ such that $b^\omega{\kern-1.3pt}>{\kern-1.3pt}\max \{m,r\}$ . Note that $b^\omega {\kern-1.3pt}\equiv{\kern-1.3pt} 1\text { (mod }p\text {)}$ by Fermat’s little theorem since $p\nmid b$ . We now define a function $\tau _b:\mathbb {N}\to \mathbb {N}$ as follows. For each fixed $n\in \mathbb {N}$ , let $\sigma _{-1}=0$ and $\sigma _i=\sum _{j=0}^i\nu _b(n,j)$ for $0\leq i\leq \lfloor \log _bn\rfloor $ . Then,

$$ \begin{align*}\tau_b(n)=\sum_{j=1}^{\sigma_{\lfloor\log_bn\rfloor}}b^{j\omega+\ell_j},\end{align*} $$

where $\ell _j=i$ for the unique $i\in \{0,1,2,\dotsc ,\lfloor \log _bn\rfloor \}$ satisfying $\sigma _{i-1}<j\leq \sigma _i$ . It is important to notice that the construction of $\tau _b(n)$ guarantees $s_b(\tau _b(n))=\sigma _{\lfloor \log _bn\rfloor }=s_b(n)$ and $\tau _b(n)\equiv \sum _{j=1}^{\sigma _{\lfloor \log _bn\rfloor }}b^{\ell _j}\equiv \sum _{i=0}^{\lfloor \log _bn\rfloor }\nu _b(n,i)b^i\equiv n\text { (mod }p\text {)}$ .

Let $x_0=\tau _b(\tilde {x})$ and, for each positive integer $t\leq g$ , let

$$ \begin{align*}x_t=x_{t-1}-b^{\lfloor\log_bx_{t-1}\rfloor}+\sum_{\iota=1}^bb^{\iota\omega+\lfloor\log_bx_{t-1}\rfloor-1}.\end{align*} $$

From this construction, $s_b(x_t)=s_b(x_{t-1})+b-1$ and

$$ \begin{align*}x_t\equiv x_{t-1}-b^{\lfloor\log_bx_{t-1}\rfloor}+b\cdot b^{\lfloor\log_bx_{t-1}\rfloor-1}\equiv x_{t-1}\text{ (mod }p\text{)}\end{align*} $$

for all $t\leq g$ . It follows that $s_b(x_g)=s_b(x_0)+g(b-1)=s_b(\tilde {x})+g(b-1)$ and ${x_g\equiv x_0\equiv \tilde {x}\text { (mod }p\text {)}}$ . Lastly, let $\alpha $ and $\beta $ be integers such that $b^{\alpha \omega }>x_g$ and $b^{\beta \omega }>\tau _b(p)$ . We define

$$ \begin{align*}x=x_g+\sum_{\iota=0}^{h-1}\tau_b(p)\cdot b^{(\iota\beta+\alpha)\omega}.\end{align*} $$

Now, $s_b(x)=s_b(x_g)+h\cdot s_b(\tau _b(p))=k$ , $x\equiv x_g+\sum _{\iota =0}^{h-1}p\cdot b^{(\iota \beta +\alpha )\omega }\equiv -m^{-1}r\text { (mod }p\text {)}$ and since every summand of x is a distinct power of b where the powers differ by at least $\omega $ , we have $s_b(mx+r)=s_b(m)\cdot s_b(x)+s_b(r)=s_b(m)\cdot k+s_b(r)=dp$ . Therefore, $mx+r$ is a b-Niven number based on the following observations:

  • $mx+r\equiv m(-m^{-1}r)+r\equiv 0\text { (mod }p\text {)}$ ;

  • $d\mid (mx+r)$ since $d\mid m$ and $d\mid r$ by Lemma 2.2;

  • $\gcd (p,d)=1$ since $p>b$ and $d\mid b-1$ .

Lemma 2.5. Let n be a nonnegative integer. For all nonnegative integers y, $s_b(yn)=ys_b(n)+z(b-1)$ for some integer z.

Proof. Note that for all nonnegative integers n, if $n=\sum _{i=0}^{\lfloor \log _bn\rfloor }\nu _b(n,i)b^i$ , then $s_b(n)=\sum _{i=0}^{\lfloor \log _bn\rfloor }\nu _b(n,i)\equiv n\text { (mod }b-1\text {)}$ . Hence, $s_b(yn)\equiv yn\equiv ys_b(n)\text { (mod }b-1\text {)}$ .

Proposition 2.6. Let $d=\gcd (s_b(m),s_b(r),b-1)$ . Then there exists a positive multiple $\overline {m}$ of m such that $\gcd (s_b(\overline {m}),s_b(r))=d$ .

Proof. Let $i_0$ be the smallest nonnegative integer such that $\nu _b(m,i_0)\neq 0$ . Then there exists a nonnegative integer $a\leq b-1$ such that $\nu _b(am,i_0)=\nu _b(a\cdot \nu _b(m,i_0),0)\geq {b}/{2}$ . Next, if $\nu _b(am,i_0+1)\neq b-1$ , then let $m'=am$ ; otherwise, let $m'=(b+1)am$ so that $\nu _b(m',i_0)=\nu _b(am,i_0)\geq {b}/{2}$ and

$$ \begin{align*}\nu_b(m',i_0+1)\equiv\nu_b(am,i_0+1)+\nu_b(am,i_0)\equiv b-1+\nu_b(am,i_0)\not\equiv {b-1}\, (\mathrm{mod}\ {b}).\end{align*} $$

Furthermore, define $m"$ to be a multiple of $m'$ such that the leading digit of $m"$ in base-b representation is at least ${b}/{2}$ , that is, $\nu _b(m",\lfloor \log _bm"\rfloor )\geq {b}/{2}$ . Let $m^*=b^2m"+m'$ . Then $m^*$ is a multiple of m such that $\nu _b(m^*,i_0)\geq {b}/{2}$ , $\nu _b(m^*,i_0+1)\neq b-1$ and $\nu _b(m^*,\lfloor \log _bm^*\rfloor )\geq {b}/{2}$ .

Let $x,y,z$ be integers such that $xs_b(r)+ys_b(m)+z(b-1)=d$ . Define $y^*$ such that $m^*=y^*m$ and let $z^*$ be an integer such that $s_b(m^*)=y^*s_b(m)+z^*(b-1)$ by Lemma 2.5. Letting $m^{**}=(b^{\lfloor \log _bm^*\rfloor -i_0}+1)m^*$ , we see that $\nu _b(m^{**}, \lfloor \log _bm^*\rfloor )=\nu _b(m^*,\lfloor \log _bm^*\rfloor )+\nu _b(m^*,i_0)-b$ and $\nu _b(m^{**},\lfloor \log _bm^*\rfloor +1)=\nu _b(m^*, i_0+1)+1\leq b-1$ . Hence, $s_b(m^{**})=2s_b(m^*)-(b-1)=2y^*s_b(m)+(2z^*-1)(b-1)$ . By Lemma 2.1, there exist nonnegative integers g and h such that $gz^*+h(2z^*-1)\equiv z\text { (mod }s_b(r)\text {)}$ . Let j be a nonnegative integer such that $gy^*+h(2y^*)+j\equiv y\text { (mod }s_b(r)\text {)}$ . Consider

$$ \begin{align*} \overline{m}=&\sum_{\iota=0}^{g-1}m^*b^{\iota(\lfloor\log_bm^*\rfloor+1)}+\sum_{\iota=0}^{h-1}m^{**}b^{\iota(\lfloor\log_bm^{**}\rfloor+1)+g(\lfloor\log_bm^*\rfloor+1)}\\ &+\sum_{\iota=0}^{j-1}mb^{\iota(\lfloor\log_bm\rfloor+1)+g(\lfloor\log_bm^*\rfloor+1)+h(\lfloor\log_bm^{**}\rfloor+1)}. \end{align*} $$

By construction, $\overline {m}$ is a multiple of m and

$$ \begin{align*} s_b(\overline{m})&=gs_b(m^*)+hs_b(m^{**})+js_b(m)\\ &=g(y^*s_b(m)+z^*(b-1))+h(2y^*s_b(m)+(2z^*-1)(b-1))+js_b(m)\\ &=(gy^*+h(2y^*)+j)s_b(m)+(gz^*+h(2z^*-1))(b-1)\\ &\equiv ys_b(m)+z(b-1)\equiv d\text{ (mod }s_b(r)\text{)}. \end{align*} $$

Note that $d\mid s_b(\overline {m})$ since $d\mid s_b(m)$ and $d\mid b-1$ . Therefore, $\gcd (s_b(\overline {m}),s_b(r))=d$ .

Combining Propositions 2.4 and 2.6, we obtain the following theorem.

Theorem 2.7. Let m and r be positive integers. The arithmetic progression $\mathcal {S}_{m,r}$ contains infinitely many b-Niven numbers.

Proof. By Proposition 2.6, there exists a multiple $\overline {m}$ of m such that

$$ \begin{align*}\gcd(s_b(\overline{m}),s_b(r),b-1)=\gcd(s_b(\overline{m}),s_b(r)). \end{align*} $$

Hence, by Proposition 2.4, $\mathcal {S}_{\overline {m},r}$ , and thus $\mathcal {S}_{m,r}$ , contains at least one b-Niven number since $\mathcal {S}_{\overline {m},r}$ is a subset of $\mathcal {S}_{m,r}$ . Let this b-Niven number be $\eta m+r$ for some nonnegative integer $\eta $ . Applying the same argument on the arithmetic progression, $\mathcal {S}_{m,(\eta +1)m+r}$ yields another b-Niven number and our proof is complete by induction.

References

Cooper, C. and Kennedy, R. E., ‘On consecutive Niven numbers’, Fibonacci Quart. 31 (1993), 146151.CrossRefGoogle Scholar
De Koninck, J.-M., Doyon, N. and Kátai, I., ‘On the counting function for the Niven numbers’, Acta Arith. 106 (2003), 265275.CrossRefGoogle Scholar
De Koninck, J.-M., Doyon, N. and Kátai, I., ‘Counting the number of twin Niven numbers’, Ramanujan J. 17 (2008), 89105.CrossRefGoogle Scholar
Grundman, H., ‘Sequences of consecutive $n$ -Niven numbers’, Fibonacci Quart. 32 (1994), 174175.CrossRefGoogle Scholar
Grundman, H., Harrington, J. and Wong, T. W. H., ‘Arithmetic progressions of $b$ -Niven numbers’, Rocky Mountain J. Math., to appear.Google Scholar
Wilson, B., ‘Construction of $2\ast n$ consecutive $n$ -Niven numbers’, Fibonacci Quart. 35 (1997), 122128.Google Scholar