2 April 2005

1991 Mathematics Subject Classification. Primary 11A25; Secondary 11B83.
<ph f="cmbx">Some properties of the pseudo-Smarandache function</ph>

Richard Pinch

2 Eldon Road, Cheltenham, Glos GL52 6TU, U.K. E-mail address : rgep@chalcedon.demon.co.uk

1 Introduction

We define the m   -th triangular number T ( m ) = m ( m + 1 ) 2   . Kashihara [2has defined the pseudo-Smarandache function Z ( n )   by Z ( n ) = min { m : n | T ( m ) } .   Charles Ashbacher [1has posed a number of questions relating to the pseudo-Smarandache function Z ( n )   . In this note we show that the ratio of consecutive values Z ( n ) / Z ( n 1 )   and Z ( n ) / Z ( n + 1 )   are unbounded; that Z ( 2 n ) / Z ( n )   is unbounded; and that n / Z ( n )   takes every integer value infinitely often. He notes that the series n 1 / Z ( n ) α   is divergent for α = 1   and asks whether it is convergent for α = 2   . He further suggests that the least value of α   for which the series converges “may never be known” . We resolve this problem by showing that the series converges for all α > 1   .

2 Some properties of the pseudo-Smarandache function

We record some elementary properties of the function Z   .
Lemma 1.
  • (1) If n T ( m )   then Z ( n ) m   . Z ( T ( m ) ) = m   .
  • (2) For all n   we have n < Z ( n )   .
  • (3) Z ( n ) 2 n 1   , and if n   is odd then Z ( n ) n 1   .
  • (4) If p   is an odd prime dividing n   then Z ( n ) p 1   .
  • (5) Z ( 2 k ) = 2 k + 1 1   .
  • (6) If p   is an odd prime then Z ( p k ) = p k 1   and Z ( 2 p k ) = p k 1   or p k   according as p k 1   or 3 m o d 4   .
We shall make use of Dirichlet's Theorem on primes in arithmetic progression in the following form.
Lemma 2. Let a , b   be coprime integers. Then the arithmetic progression a + b t   is prime for infinitely many values of t   .

3 Successive values of the pseudo-Smarandache function

Using properties (3) and (5), Ashbacher observed that | Z ( 2 k ) Z ( 2 k 1 ) | > 2 k   and so the difference between the conecutive values of Z   is unbounded. He asks about the ratio of consecutive values.
Theorem 1. For any given L > 0   there are infinitely many values of n   such that Z ( n + 1 ) / Z ( n ) > L   , and there are infinitely many values of n   such that Z ( n 1 ) / Z ( n ) > L   .
  • Proof. Choose k 3 m o d 4   , so that T ( k )   is even and k   divides T ( k )   . We consider the conditions k | m   and ( k + 1 ) | ( m + 1 )   . These are satisfied if m k m o d k ( k + 1 )   , that is, m = k + k ( k + 1 ) t   for some t   . We have m ( m + 1 ) = k ( 1 + ( k + 1 ) t ) ( k + 1 ) ( 1 + k t )   , so that if n = k ( k + 1 ) ( 1 + k t ) / 2   we have n | T ( m )   . Now consider n + 1 = T ( k ) + 1 + k T ( k ) t   . We have k | T ( k )   , so T ( k ) + 1   is coprime to both k   and T ( k )   . Thus the arithmetic progression T ( k ) + 1 + k T ( k ) t   has initial term coprime to its increment and by Dirichlet's Theorem contains infinitely many primes. We find that there are thus infinitely many values of t   for which n + 1   is prime and so Z ( n ) m = k + k ( k + 1 ) t   and Z ( n + 1 ) = n = T ( k ) ( 1 + k t )   . Hence Z ( n + 1 ) Z ( n ) n m = T ( k ) + k T ( k ) t k + 2 T ( k ) t > k 3 .   A similar argument holds if we consider the arithmetic progression T ( k ) 1 + k T ( k ) t   . We then find infinitely many values of t   for which n 1   is prime and Z ( n 1 ) Z ( n ) n 2 m = T ( k ) 2 + k T ( k ) t k + 2 T ( k ) t > k 4 .   The Theorem follows by taking k > 4 L   .
We note that this Theorem, combined with Lemma 1(2), gives another proof of the result that the difference of consecutive values is unbounded.

4 Divisibility of the pseudo-Smarandache function

Theorem 2. For any integer k 2   , the equation n / Z ( n ) = k   has infinitely many solutions n   .
  • Proof. Fix an integer k 2   . Let p   be a prime 1 m o d 2 k   and put p + 1 = 2 k t   .
    Put n = T ( p ) / t = p ( p + 1 ) / 2 t = p k   . Then n | T ( p )   so that Z ( n ) p   . We have p | n   , so Z ( n ) p 1   : that is, Z ( n )   must be either p   or p 1   . Suppose, if possible, that it is the latter. In this case we have 2 n | p ( p + 1 )   and 2 n | ( p 1 ) p   , so 2 n   divides p ( p + 1 ) ( p 1 ) p = 2 p   : but this is impossible since k > 1   and so n > p   . We conclude that Z ( n ) = p   and n / Z ( n ) = k   as required. Further, for any given value of k   there are infinitely many prime values of p   satisfying the congruence condition and hence infinitely many values of n = T ( p )   such z / Z ( n ) = k   .

5 Another divisibility question

Theorem 3. The ratio Z ( 2 n ) / Z ( n )   is not bounded above.
  • Proof. Fix an integer k   . Let p 1 m o d 2 k   be prime and put n = T ( p )   . Then Z ( n ) = p   . Consider Z ( 2 n ) = m   . We have 2 k p | p ( p + 1 ) = 2 n   and this divides m ( m + 1 ) / 2   . We have m ε m o d p   and m δ m o d 2 k + 1   where each of ε , δ   can be either 0   or 1   .
    Let m = p t + ε   . Then m ε t δ m o d 2 k   : that is, t ε δ m o d 2 k   .
    This implies that either t = 1   or t 2 k 1   . Now if t = 1   then m p   and T ( m ) T ( p ) = n   , which is impossible since 2 n T ( m )   . Hence t 2 k 1   .
    Since Z ( 2 n ) / Z ( n ) = m / p > t / 2   , we see that the ratio Z ( 2 n ) / Z ( n )   can be made as large as desired.

6 Convergence of a series

Ashbacher observes that the series n 1 / Z ( n ) α   diverges for α = 1   and asks whether it converges for α = 2   .
In this section we prove convergence for all α > 1   .
Lemma 3. log n m = 1 n 1 m 1 + log n ;   1 2 ( log n ) 2 0.257 m = 1 n log m m 1 2 ( log n ) 2 + 0.110 for n 4 .  
  • Proof. For the first part, we have 1 / m 1 / t 1 / ( m 1 )   for t [ m 1 , m ]   .
    Integrating, 1 m m 1 m 1 t d t 1 m 1 .   Summing, 2 n 1 m 1 n 1 t d t 2 n 1 m 1 ,   that is, 1 n 1 m 1 + log n and log n 1 n 1 1 m .   The result follows.
    For the second part, we similarly have log m / m log t / t log ( m 1 ) / ( m 1 )   for t [ m 1 , m ]   when m 4   , since log x / x   is monotonic decreasing for x > e   .
    Integrating, log m m m 1 m log t t d t log ( m 1 ) m 1 .   Summing, 4 n log m m 3 n log t t d t 4 n log ( m 1 ) m 1 ,   that is,
    1 n log m m log 2 2 log 3 3
    1 2 ( log n ) 2 1 2 ( log 3 ) 2
    1 n log m m log n n log 2 2 .
    We approximate the numerical values log 2 2 + log 3 3 1 2 ( log 3 ) 2 < 0.110   and log 2 2 1 2 ( log 3 ) 2 > 0.257 .   to obtain the result.
Lemma 4. Let d ( m )   be the function which counts the divisors of m   .
For n 2   we have m = 1 n d ( m ) / m < 7 ( log n ) 2 .  
  • Proof. We verify the assertion numerically for n 6   . Now assume that n 8 > e 2   . We have
    m = 1 n d ( m ) m = m = 1 n d e = m 1 m = d n d e n 1 d e
    = d n 1 d e < n / d 1 e d n 1 d ( 1 + log ( n / d ) )
    ( 1 + log n ) 2 1 2 ( log n ) 2 + 0.257
    = 1.257 + 2 log n + 1 2 ( log n ) 2
    < 4 3 ( log n 2 ) 2 + 2 log n ( log n 2 ) + 1 2 ( log n ) 2
    < 2 ( log n ) 2 .
Lemma 5. Fix an integer t 5   . Let e t > Y > e ( t 1 ) / 2   . The number of integers n   with e t 1 < n e t   such that Z ( n ) Y   is at most 196 Y t 2   .
  • Proof. Consider such an n   with m = Z ( n ) Y   . Now n | m ( m + 1 )   , say k 1 n 1 = m   and k 2 n 2 = m + 1   , with n = n 1 n 2   . Thus k = k 1 k 2 = m ( m + 1 ) / n   and k 1 n 1 Y   . The value of k   is bounded below by 2 and above by m ( m + 1 ) / n 2 Y 2 / e t 1 = K   , say. Given a pair ( k 1 , k 2 )   , the possible values of n 1   are bounded above by Y / k 1   and must satisfy the congruence condition k 1 n 1 + 1 0   modulo k 2   : there are therefore at most Y / k 1 k 2 + 1   such values.
    Since Y / k Y / K = e t 1 / 2 Y > 1 / 2 e   , we have Y / k + 1 < ( 2 e + 1 ) Y / k < 7 Y / k   .
    Given values for k 1 , k 2   and n 1   , the value of n 2   is fixed as n 2 = ( k 1 n 1 + 1 ) / k 2   .
    There are thus at most k K d ( k )   possible pairs ( k 1 , k 2 )   and hence at most k K 7 Y d ( k ) / k   possible quadruples ( k 1 , k 2 , n 1 , n 2 )   . We have K > 2   so that the previous Lemma applies and we can deduce that the number of values of n   satisfying the given conditions is at most 49 Y ( log K ) 2   . Now K = 2 Y 2 / e t 1 < 2 e t + 1   so log K < t + 1 + log 2 < 2 t   . This establishes the claimed upper bound of 196 Y t 2   .
Theorem 4. Fix 1 2 < β < 1   and an integer t 5   . The number of integers n   with e t 1 < n e t   such that Z ( n ) < n β   is at most 196 t 2 e β t   .
  • Proof. We apply the previous result with Y = e β t   . The conditions of β   ensure that the previous lemma is applicable and the upper bound on the number of such n   is 196 e β t t 2   as claimed.
Theorem 5. The series n = 1 1 Z ( n ) α   is convergent for any α > 2   .
  • Proof. We note that if α > 2   then 1 / Z ( n ) α < 1 / n α / 2   and the series is convergent.
    So we may assume 2 < α 2   . Fix β   with 1 / α < β < α / 2   . We have 1 2 < β < 1 2 < α / 2   .
    We split the positive integers n > e 4   into two classes A   and B   . We let class A   be the union of the A t   where, for positive integer t 5   we put into class A t   those integers n   such that e t 1 < n e t   for integer t   and Z ( n ) n β   . All values of n   with Z ( n ) > n β   we put into class B   . We consider the sum of 1 / Z ( n ) α   over each of the two classes. Since all terms are positive, it is sufficient to prove that each series separately is convergent.
    Firstly we observe that for n B   , we have 1 / Z ( n ) α < 1 / n α β   and since α β > 1   the series summed over the class B   is convergent.
    Consider the elements n   of A t   : so for such n   we have e t 1 < n e t   and Z ( n ) < n β   . By the previous result, the number of values of n   satisfying these conditions is at most 196 t 2 e β t   . For n A t   , we have Z ( n ) n   , so 1 / Z ( n ) α 1 / n α / 2 < 1 / e α ( t 1 ) / 2   . Hence the sum of the subseries n A t 1 / Z ( n ) α   is at most 196 e α / 2 t 2 e ( β α / 2 ) t   . Since β < α / 2   for α > 2   , the sum over all t   of these terms is finite.
    We conclude that n = 1 1 / Z ( n ) α   is convergent for α > 2  
Theorem 6. The series n = 1 1 Z ( n ) α   is convergent for any α > 1   .
  • Proof. We fix β 0 = 1 > β 1 > > β r = 1 2   with β j < α β j + 1   for 0 j r 1   .
    We define a partition of the integers e t 1 < n < e t   into classes B t   and C t ( j )   , 1 j r 1   . Into B t   place those n   with Z ( n ) > n β 1   . Into C t ( j )   place those n   with n β j + 1 < Z ( n ) < n β j   . Since β r = 1 2   we see that every n   with e t 1 < n < e t   is placed into one of the classes.
    The number of elements in C t ( j )   is at most 196 t 2 e β j t   and so n C t ( j ) 1 Z ( n ) α < 196 t 2 e β j t e β j + 1 α ( t 1 ) = 196 e β j + 1 α t 2 e ( β j α β j + 1 ) t .   For each j   we have β j < α β j + 1   so each sum over t   converges.
    The sum over the union of the B t   is bounded above by n 1 n α β 1 ,   which is convergent since α β 1 > β 0 = 1   .
    We conclude that n = 1 1 / Z ( n ) α   is convergent.
References

  1. Charles Ashbacher, Pluckings from the tree of Smarandache sequences and functions, American Research Press, 1998, http://www.gallup.unm.edu/   smarandache/Ashbacher-pluckings.pdf.
  2. K. Kashihara, Comments and topics on Smarandache notions and problems, Erhus University Press, Vail, AZ, USA, 1996.

2 Eldon Road, Cheltenham, Glos GL52 6TU, U.K. E-mail address : rgep@chalcedon.demon.co.uk