Progress on Sum of Prime Factors Function

For any positive integer N let s(N) denote the sum of the prime 
factors of N.  For example, s(12)=2+2+3=7.  A previous note
described some investigations into this function, which led to a 
consideration of the set of integers {m1,m2,..,mk} for any given 
N such that  mj + s(mj) = N.  The article defined the function 
tau(N) in terms of this set of numbers as follows

                          k
          tau(N) =  N - SUM  s(m_j)  
                         j=1

It was noted that tau(N) is positive for nearly all N, and the 
average value of tau(N) increases as N increases.  In fact, for
positive values of tau(N) it seems that for any x there is an 
index k such that tau(N) > x  for all N > k.  

However, for a small number of integers N the value of tau(N) is 
*negative*.  The first such value is 239, for which tau(239)= -21.  
The article asked if tau(N)=0 for any integer N.  This is Problem 
#9 on the MathPages Most Wanted List, and the answer is still 
not known.  However, some progress has been made, thanks to some 
further numerical results computed by Jud Mccranie.

Jud checked integers up to 10^9 and found no cases with tau=0, but 
he did continue to find occassional integers with relatively small 
negative values of tau.  (He found about 2000 in all.) Interestingly,
he noted that only a specific set of values occur in the negative 
range.  In particular he found the following 15 integers with the 
tau(N) = -29:

        N         tau(N)   terms   factorization of N

      97547759      -29      5     97   1005647
     204393239      -29      5     2663   76753  
     223052759      -29      5     223052759  
     236437559      -29      5     5449   43391  
     341332559      -29      5     341332559  
     372043799      -29      5     41   113   131   613  
     476690039      -29      5     7   68098577  
     488336759      -29      5     23   21232033  
     492404039      -29      5     492404039  
     552351239      -29      5     11   311   161459  
     563508359      -29      5     31   18177689  
     633501359      -29      5     317   1998427  
     646689239      -29      5     7   92384177  
     677462759      -29      5     71   9541729  
     693707399      -29      5     7   11^2  819017

From this it's not hard to see what's happenning, and it's kind
of neat.  It depends on the existence of certain Egyptian fraction 
partitions of 1.  

Suppose we can find an integer N such that each of the following
five numbers is a prime:

             a = (N-2)/3   
             b = (N-3)/4    
             c = (N-4)/5   
             d = (N-5)/6
             e = (N-19)/20

It follows that m + s(m) = N  where m is any of the five numbers

          2a     3b     4c     5d     19e

If no other "m value" exists for this N, then tau(N)=-29.

What's so special about the set of numbers {2,3,4,5,19}?  Well, to 
see why this works, remember that according to the definition of 
tau(N) we have

     tau(N) = N - (2+a) - (3+b) - (4+c) - (5+d) - (19+e)

Substituting the above expressions for a,b,c,d,e and separating terms
we have

                      / 1   1   1   1    1 \
  tau(N) = N - 33 - N(  - + - + - + - + --  )
                      \ 3   4   5   6   20 /

                             / 2   3   4   5   19 \
                         +  (  - + - + - + - + --  )
                             \ 3   4   5   6   20 /

The sum of the unit fractions in the first parentheses is 1, and
the sum of the complement fractions in the second parentheses is 
4, giving the result tau(N) = -29  independent of N.  So it turns 
out the ancient Egyptian fractions play an important part in this 
phenomenon.  To make this work for some number besides -29 we would 
need to find another sum of distinct unit fractions that sum to 1, 
and whose denominators are all 1 greater than a prime (counting 4 
as a prime because s(4)=4).  

Let's say we have a set of k distinct primes {p1,p2,...pk} (again
allowing 4 to be considered as a prime) such that

         p1 + p2 + ... + pk  =  M

and
        1        1               1
       ----  +  ----  + ... +  ----   =   1
       p1+1     p2+1           pk+1

It follows that if N is any integer such that each of the numbers

        a1 = (N-p1)/(p1+1)
        a2 = (N-p2)/(p2+1)

            etc.

        ak = (N-pk)/(pk+1)

is a prime integer then tau(N) = k-1-M.  This pretty much rules
out finding an N *of this form* with tau(N)=0, because the sum of 
the k primes will clearly always be greater than k-1.  Still, I
find this to be a very interesting result.

This raises some interesting questions.  For any fixed set of 
primes {p1,p2,..,pk}  are there infinitely many N that give all 
prime values of a1,a2,..ak?  Also, are all negative values
of tau members of families of this type?   The answer to this
second question is no, because most of the integers with negative
tau are demonstrateably not of this form.  For example, the very
first occurrance of a negative tau is tau(239)=-21, which has the
set of "m values"

                158  =  (2)(79)
                177  =  (3)(59)
                188  =  (4)(47)
                203  =  (7)(29)
                209  =  (11)(19)

In general, for any N such that (N-2)/3, (N-3)/4, etc are all primes 
(and no other m-values exist) we have we have

                 /    N-2\     /    N-3\     /    N-4\
   tau(N) = N - ( 2 + --- ) - ( 3 + --- ) - ( 4 + --- )
                 \     3 /     \     4 /     \     5 /

                                   /    N-7\     /    N-11\
                                - ( 7 + --- ) - (11 + ---- )
                                   \     8 /     \     12 /

which reduces to

                   N + 1
        tau(N) =  -------  -  23
                    120

Notice that if N=2759 were such that (N-2)/3 etc were primes, then
it would give tau = 0.  It actually happens that (2759-2)/3 and
(2759-11)/12 are both primes, but the rest of the conditions are
not satisfied, so tau(2759) does not equal zero.  However, there
doesn't seem to be any a_priori reason why something like this
couldn't eventually happen for some N, giving a tau of zero.

Anyway, returning to the families of equal tau values, I've been 
able to identify the family for most of the persistent small 
negative values of tau that appeared in Jud's results.  These 
correspond to the following sets of "primes U 4"

                    pk                  tau(N)
             ----------------------   ---------
             2  3  4  5  19             -29
             2  3  4  5  29  59         -97
             2  3  4  7  19  23         -53
             2  3  5  7  11  23         -46
             2  3  5  7  13  23  83    -130


Return to MathPages Main Menu
Сайт управляется системой uCoz