Marc Del´eglise
CONTENTS Introduction
1. ExpressingA(x)as a Sum 2. Trivial Bounds forAk(x) 3. Lower Bound forA(x) 4. Upper Bounds forAk(x)
5. Mean Values off(n)rand Upper Bounds forAk(x) 6. Upper Bounds for the Euler Products V
k(r) 7. Numerical Results
8. Other Experimental Results References
We say that an integer n isabundantif the sum of the divisors of n is at least 2n. It has been known [Wall 1972] that the set of abundant numbers has a natural density A(2) and that 0.244<A(2)<0.291. We give the sharper bounds
0.2474<A(2)<0.2480.
INTRODUCTION
Letx be a positive real number, andn an integer.
Let(n) be the sum of the divisors ofn, and set f(n) = (n)
n ; A(x) =fn:f(n)xg: (0–1) A number inA(x) is called x-abundant, or simply abundantif x= 2.
Davenport proved thatA(x) has a natural den- sityA(x), and thatA(x) is a continuous function of x; see, for example, [Davenport 1933; Elliott 1979, Chapter 5; Tenenbaum 1995, III.1 and III.2].
Behrend [1933] proved that 0:241<A(2)<0:314, and Wall [1972] improved this to 0:244 < A(2)<
0:291. We prove here the following:
Theorem 0.1. The density A(2) of the set of abun- dant numbers satises
0:2474< A(2)<0:2480:
This answers a question asked by Henri Cohen: Is the proportion of abundant numbers more or less than a quarter? The method used is essentially that given by Behrend, the computer allowing us to do more computations. This method in fact gives the densityA(x) for every x.
Perhaps it could be worthwile to try an analytic method. Cohen, Deshouillers, Martinet showed in
c
A K Peters, Ltd.
1058-6458/1998 $0.50 per page Experimental Mathematics7:2, page 137
[Martinet et al. 1973] that the Mellin transform of A(x) is the function
g(s) = 1s
Y
p2
1 11ps 1
X
l0
p1l
1 1
pl+1
s
: Hence, by inversion, we have for every >1
A(x) = 12i
Z +i1
i1 x sg(s)ds;
but the computation of this integral seems to be dicult; takingx= 2 and = 2 we computed the sum between 2 10000i and 2 + 10000i, and got the approximate value 0:242. For large values of Im(s) the computation of g(s) is dicult.
1. EXPRESSINGA(x)AS A SUM
We denote by (pn)n1 the increasing sequence of primes. Let k be a xed integer. We consider the set
Ak(x) =fn:f(n)x; gcd(n; p1p2:::pk) = 1g:
(1–1)
This set has a density [Elliott 1979; Tenenbaum 1995], which will be denoted byAk(x).
Let n be an arbitrary integer. We denote by n1 the product of the prime factors of n among
fp1;p2;:::;pkg and we write n=n1n2. The func- tion f is multiplicative and f(n) = f(n1)f(n2) is greater than or equal to x if and only if f(n2) x=f(n1). This proves that A(x) is partitioned as follows:
A(x) = [
n1=p11:::pkkn1Ak
x f(n1)
: Considering the densities we have:
Proposition 1.1.
A(x) = X
n1=p11:::pkk
n11Ak
x f(n1)
; (1–2)
where the sum is taken over alln1 that are a prod- uct of primes belonging to fp1;p2;:::;pkg.
To see this, it is sucient to prove the following lemma.
Lemma 1.2. Let p be an integer greater than 1 and (A)0 a sequence of disjoint sets having densities d. Set A = S0pA. Then A has a density d(A) and
d(A) =X
0
p1d:
Proof. Write
A =
[
0rpA
[
[
>rpA
The second set in this union is formed of multiples of pr+1. Its upper density is bounded by 1=pr+1 and
X
0r
p1dd(A)d(A) X
0r
p1d+ 1pr+1 where d and d denote the lower and upper densi- ties. We letr !1and we get the result.
2. TRIVIAL BOUNDS FORAk(x)
Proposition 2.1. For every k 0 and every x > 0 we have
Ak(x)Fk (2–1)
and Ak(x) =Fk if x1; (2–2)
where Fk =Qki=1(1 1=pi).
Proof. Clear, since Ak(x) is formed only with in- tegers coprime with p1p2:::pk, and comprises all
these integers if x1.
3. LOWER BOUND FORA(x)
Let z be a arbitrary positive real parameter. If in (1{2) we just keep the integersn1=p11:::pkk z, we get a lower bound forA(x). Hence
A(x) nX1z
n1=p11:::pkk
n11Ak
x f(n1)
:
We still get a lower bound if we just keep thosen1 such that f(n1)x; hence
A(x) nX1z
n1=p11:::pkk f(n1)x
n11Ak
x f(n1)
:
By (2{2), all the Ak(x=f(n1)) are equal to Fk; hence
A(x)Fk nX1z n1=p11:::pkk
f(n1)x
n11: (3–1)
This lower bound is almost trivial and could have been shown slightly dierently. We choose an upper bound z and a set fp1;p2;:::;pkg of small primes. We compute all the integers m less than z, composed of prime factors fromfp1;p2;:::;pkg, and x-abundant. Every multiple of an abundant number being abundant, all the products of the numbers m thus obtained by some prime factors out of fp1;p2;:::;pkg are still abundant numbers.
The lower bound forA(x) is the density of this set, FkPm1=m.
4. UPPER BOUNDS FORAk(x)
As in the previous section, we introduce a real pos- itive parameter z and write
A(x) = nX1z
n1=p11:::pkk
n11Ak
x f(n1)
+ z<nX1
n1=p11:::pkk
n11Ak
x f(n1)
: In the second sum, each value of Ak is bounded from above byFk; thus the second sum is bounded from above by
Fk z<nX1
n1=p11:::pkk
n11=Fk
1Xn11 n1=p11:::pkk
n11 Fk nX1z n1=p11:::pkk
n11
= 1 Fk nX1z n1=p11:::pkk
n11;
so
A(x) nX1z
n1=p11:::pkk
n11Ak
x f(n1)
+1 Fk nX1z n1=p11:::pkk
n11: (4–1) It remains to bound the values of Ak that appear in the sum (4{1). If we just use the trivial upper bound Ak Fk we will get A(x) 1, so we need a nontrivial upper bound for Ak(x). This is the subject of the next section.
5. MEAN VALUES OFf(n)rAND UPPER BOUNDS FOR Ak(x)
Letfk be the multiplicative function that takes the value 1 forp withppk and the valuef(p) for p > pk. We x an integerrand we considerg=frk and the mean value of g computed on the rst n integers:
Mn= 1n
n
X
m=1g(m):
Let be the convolution product of g and the Mobius function:
(m) =X
djmm d
g(d): (5–1) The Mobius inversion formula gives
Mn = 1n
n
X
m=1g(n) = 1n
n
X
m=1
X
djm(d)
= 1n
n
X
d=1(d)hn d
i
n1
n
X
d=1(d)n d
1
X
d=1
(d)
d = k(r):
The function (d)=d is multiplicative, so k(r) is also equal to the value of the Euler product
k(r) =Y
p
1 +(p)
p +(p2)
p2 ++
: (5–2)
Using the denition equation (5{1) of, we have (p) =g(p) g(p 1)
=
1+1p++ 1p
r
1+1p++ 1p 1
r
whenp > pk and >0, otherwise (p) = 0.
We return to the sum nMn =Xn
m=1g(m):
LetBn be the number of integersmbetween 1 and nsuch thatfk(m)x, or equivalentlyg(m)xr. We collect the terms of this sum in two classes, rst those terms for which g(m) xr, that are bounded from below by xr, and the other terms, that are bounded from below by 1. We get
xrBn+n BnnMn nk(r);
dividing byn and lettingn!1we get Bk(x) k(r) 1
xr 1 ;
whereBk(x) is the density of the set of all m such that fk(m) x. This set is the disjoint union of the p11:::pkkAk(x), and we deduce the following upper bound, proved by Behrend [1933].
Proposition 5.1.For every integerr1and everyk, Ak(x)Fkk(r) 1
xr 1 : (5–3)
Table 1 gives the upper bounds for 95(r) 1 for r= 1;2;4;8;16;::: ;4096.
r 95(r) 1 r 95(r) 1
1 0.000284 64 0.0189
2 0.000568 128 0.0395
4 0.00114 256 0.0866
8 0.00228 512 0.213
16 0.00458 1024 0.726
32 0.00925 2048 12.3
4096 1:371017
TABLE 1. Upper bounds for 95(r) 1.
When x is very close to 1, almost every integer isx-abundant and the trivial upper bound (2{1) is better than the upper bound (5{3). Table 2 shows this phenomenon. It gives for some values of xthe best upper bound for Ak(x) obtained by formula (5{3) choosing the right value forr. The valuer= 0 on the rst line means that, for thisx= 1:0001, the trivial upper bound (2{1) is the better one.
x r A
95(x) x r A95(x) 1.0001 0 0.0897 1.005 2048 4:3510 5
1.001 1 0.0254 1.01 2048 1:6810 9 1.002 1024 0.0096 1.02 4096 9:2110 20
TABLE 2. Some upper bounds for A95(r) obtained using Table 1.
6. UPPER BOUNDS FOR THE EULER PRODUCTS V
k(r)
In this section we give some eective upper bounds used to get upper bounds for the Euler products k(r). In all this section we write
(p) =
1+ 1p++ 1p
r
1+ 1p++ 1p 1
r
=X
djp(d)
fp d
r
:
This is the function dened by (5{1) for k = 0.
We gave in [Deleglise and Nicolas 1994] a method to quickly compute a good approximate value of an Euler productQpg(1=p), whengis a holomorphic function around 0 whose rst Taylor series coe- cients are not too large. This method could have been used to get some very accurate values for the rst k(r). For very large values of r the accu- racy would not be so good. Since we just need an upper bound for each k(r), we will just use the trivial method: nd upper bounds for the partial products, and for the tails of the products.
Lemma 6.1. Let r be an integer 1 and p 2r.
Then
1 + 1p
r
1<1:3r p:
Proof. We have (1 + 1=p)r 1
r=p = exp rln(1 + 1=p) 1 r=p
< exp(r=p) 1 r=p
e1=2 1
1=2 <1:3:
Lemma 6.2. Let r an integer2 and p2r. Then
1 1 1=p
r 1
< 169 <1:78:
Proof. Letu= 1=p. Then y=
1 1 1=p
r 1
<1 11 =p
r
1
1 u
1=2u
; hence
ln(y) = 12uln
1 1 u
2ln
1 1 14
= ln 169; since the function (1=u)ln 1=(1 u)is increasing
for 0< u 21r 14.
Lemma 6.3. For every integer r and every prime p,
X
0
(p) p
1 + (1 + 1=p)r 1
p +r
1 1 1=p
r 1 1 p4 p2:
Proof. Set
Y = 1 + 1p++ 1p 1; X =Y + 1p: We get, for1,
(p)
p = 1p(Xr Yr)
= 1p2(Xr 1+Xr 2Y ++Yr 1)
pr2Xr 1 r p2
1 1 1=p
r 1
:
Using this upper bound for2 in the sum
X
0
(p) p
we get the conclusion.
Lemma 6.4. For every integer r 1 and every p max(2r;15) we have
X
0
(p)
p <1 + 1:31r p2:
Proof. The preceding three lemmas give, for every r2,
X
0
(p)
p 1 + 1:3 r
p2 + 1:78 r p4 p2
= 1 + r p2
1:3 + 1:78 1p2 1
1 + 1:31 r
p2 ifp15:
Forr= 1 this upper bound is still true, because
X
0
(p) p =
X
0
p12 = 1 + 1p2 p4:
Lemma 6.5.For every integer r with 1r10000 we have
Y
p>106
X
0
(p) p
1 + r 107:
Proof. Set
u= Y
p>106
X
0
(p) p
: Using Lemma 6.4 we get
ln(u)1:31r X
p>106
p12:
The sum of 1=p2 can be computed as explained in [Deleglise and Nicolas 1994, pp. 331{332], or it can be found in [Glaisher 1891]:
X
p
p12 = 0:452247420041::: :
Interval Interval Interval Interval
[1;101] 1 [1;106] 24799 [109;109 + 107) 2476049 [1014;1014+ 107) 2476150 [1;102] 24 [1;107] 2476741 [1010;1010+ 107) 2476372 [1015;1015+ 107) 2476212 [1;103] 249 [1;108] 24760673 [1011;1011+ 107) 2476154 [1016;1016+ 107) 2476247 [1;104] 2492 [1;109] 247610965 [1012;1012+ 107) 2476199 [1017;1017+ 107) 2476098 [1013;1013+ 107) 2476213 [1018;1018+ 107) 2476304
TABLE 3. Frequency of abundant numbers in dierent intervals.
Hence, subtractingPp1061=p2, we have
X
p>106
p12 = 0:0000000677:::
and ln(u)<0:9 r
107 <10 3; and nally
u=elnu<1 + r 107;
using the estimateet<1 +109t fort <0:001.
We get an upper bound for the Euler product (5{2), writing
Y
p>pk
X
0
(p) p
= Y
pk<p106
X
0
(p) p
Y
p>106
X
0
(p) p
: The rst product is bounded by Lemma 6.3 and the second by Lemma 6.5.
Table 1 gives the upper bounds for 95(r) 1 forr = 1;2;4;8;16;::: ;4096. These are the values used for bounding the values Ak that appear in formula (4{1).
7. NUMERICAL RESULTS
We have boundedA(2) using (3{1) and (4{1) with x= 2, k = 95 (which is the number of primes less than 500), and z = 1014. For the upper estimate each term
Ak
x
f(p11:::pkk)
in (4{1) is bounded using formula (5{3) with r = 1;2;4;8;::: ;4096 and the trivial bound (2{1); we keep the best result obtained. This requires the enumeration of all the p11:::pkk not greater than z, which is done by a backtracking procedure. The total number of thesenless than 1014whose prime factors are less than 500 is 23581230171.
The computation was performed on an HP900- 730 workstation, using about 100 hours of CPU time. It yields
0:2474< A(2)<0:2480; (7–1) in particular A(2) = 0:247:::.
8. OTHER EXPERIMENTAL RESULTS
We computed the number of abundant numbers less than N for N = 1;10;102;:::;109, and also the number of abundant numbers in the intervals [N; N+107) for N = 109;1010;:::;1018. The re- sults are given in Table 3 and seem to show that the next digit of A(2) is a 6.
We thank the referee for remarking to us that the number of abundant numbers given above in the intervals of size 107are compatible with a binomial law with parameters mp= 2476200 ands= 1365.
REFERENCES
[Behrend 1933] F. Behrend, \Uber Numeri abundantes, II", Sitzungsber. Preuss. Akad. Wiss. (1933), 280{
293.
[Davenport 1933] H. Davenport, \Uber Numeri abun- dantes", Sitzungsber. Preuss. Akad. Wiss. (1933), 830{837.
[Deleglise and Nicolas 1994] M. Deleglise and J.-L.
Nicolas, \Sur les entiers inferieurs a xayant plus de log(x) diviseurs", J. Theor. Nombres Bordeaux 6:2 (1994), 327{357.
[Elliott 1979] P. D. T. A. Elliott, Probabilistic num- ber theory, I: Mean-value theorems, Grundlehren der Mathematischen Wissenschaften, Springer, New York, 1979.
[Glaisher 1891] W. L. Glaisher, \On the sums of the inverse powers of the prime numbers", Quaterly Journal of Math.25(1891), 347{362.
[Martinet et al. 1973] J. Martinet, J. M. Deshouillers,
and H. Cohen, \La fonction somme des diviseurs", pp. Exp. No. 11 in Seminaire de Theorie des Nombres, 1972{1973 (Univ. Bordeaux I, Talence), Lab. Theorie des Nombres, Centre Nat. Recherche Sci., Talence, 1973.
[Tenenbaum 1995] G. Tenenbaum, Introduction a la theorie analytique et probabiliste des nombres, 2nd ed., Cours specialises 1, Soc. math. France, Paris, 1995.
[Wall 1972] C. R. Wall, \Density bounds for the sum of divisors function", pp. 283{287 in The theory of arithmetic functions (Kalamazoo, Mich., 1971), edited by A. A. Gioia and D. L. Goldsmith, Lecture Notes in Math.251, Springer, Berlin, 1972.
Marc Deleglise, Institut Girard Desargues (UPRES-A 5028), Departement de Mathematiques, Universite Lyon 1, 43 Bld du 11 Novembre 1918, 69622 Villeurbanne cedex, France (deleglis@desargues.univ-lyon1.fr)
Received January 21, 1997; accepted August 11, 1997