On certain arithmetical functions

Transactions of the Cambridge Philosophical Society, XXII, No.9, 1916, 159 – 184

1. Let σs (n) denote the sum of the sth powers of the divisors of n (including 1 and n), and
let
σs (0) = 12 ζ(−s),
where ζ(s) is the Riemann Zeta-function. Further let
X
(n) = σr (0)σs (n) + σr (1)σs (n − 1) + · · · + σr (n)σs (0). (1)
r,s

In this paper I prove that

X Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)
(n) = σr+s+1 (n)
r,s
Γ(r + s + 2) ζ(r + s + 2)

ζ(1 − r) + ζ(1 − s) 2
+ nσr+s−1 (n) + O{n 3 (r+s+1) }, (2)
r+s
whenever r and s are positive odd integers. I also prove that there is no error term on the
right-hand side of (2) in the following nine cases: r = 1, s = 1; r = 1, s = 3; r = 1, s =
5; r = 1, sP= 7; r = 1, s = 11, r = 3, s = 3; r = 3, s = 5; r = 3, , s = 9; r = 5, s = 7. That
is to say r,s (n) has a finite expression in terms of σr+s+1 (n) and σr+s−1 (n) in these nine
cases; but for other values of r and s it involves other arithmetical functions as well.
It appears probable, from the empirical results I obtain in §§ 18-23, that the error term on
the right-hand side of (2) is of the form
1
O{n 2 (r+s+1+ǫ) }, (3)

where ǫ is any positive number, and not of the form

1
o{n 2 (r+s+1) }. (4)

But all I can prove rigorously is (i) that the error is of the form
2
O{n 3 (r+s+1) }

in all cases, (ii) that it is of the form

2 3
O{n 3 (r+s+ 4 ) } (5)

if r + s is of the form 6m, (iii) that it is of the form

2 1
O{n 3 (r+s+ 2 ) } (6)
On certain arithmetical functions 175

if r + s is of the form 6m + 4, and (iv) that it is not of the form

1
o{n 2 (r+s) }. (7)

It follows from (2) that, if r and s are positive odd integers, then
X Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)
(n) ∼ σr+s+1 (n). (8)
r,s
Γ(r + s + 2) ζ(r + s + 2)

It seems very likely that (8) is true for all positive values of r and s, but this I am at present
unable to prove.
P
2. If r,s (n)/σr+s+1 (n) tends to a limit, then the limit must be

Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)

.
Γ(r + s + 2) ζ(r + s + 2)

For then
P P P P
r,s (n) r,s (1) + r,s (2) + ··· + r,s (n)
lim = lim
n→∞ σr+s+1 (n) n→∞ σr+s+1 (1) + σr+s+1 (2) + · · · + σr+s+1 (n)
2
P P P
r,s (0) + r,s (1)x + r,s (2)x + · · ·
= lim
x→1 σr+s+1 (0) + σr+s+1 (1)x + σr+s+1 (2)x2 + · · ·
Sr Ss
= lim ,
x→1 Sr+s+1

where

1r x 2r x2 3r x3
Sr = 12 ζ(−r) + + + + ··· (9)
1 − x 1 − x2 1 − x3
Now it is known that, if r > 0, then

Γ(r + 1)ζ(r + 1)
Sr ∼ , (10)
(1 − x)r+1

as x → 1 ∗ . Hence we obtain the result stated.

3. It is easy to see that

σr (1) + σr (2) + σr (3) + · · · + σr (n)

= u1 + u2 + u3 + u4 + · · · + un ,
∗
Knopp, Dissertation (Berlin, 1907), p.34.
176 Paper 18

where h n ir
ut = 1r + 2r + 3r + · · · + .
t
From this it is easy to deduce that
nr+1 ∗
σr (1) + σr (2) + · · · + σr (n) ∼ ζ(r + 1) (11)
r+1
and
Γ(r + 1)Γ(s + 1)
σr (1)(n − 1)s + σr (2)(n − 2)s + · · · + σr (n − 1)1s ∼ ζ(r + 1)nr+s+1 ,
Γ(r + s + 2)
provided r > 0, s ≥ 0. Now
σs (n) > ns ,
and
σs (n) < ns (1−s + 2−s + 3−s + · · ·) = ns ζ(s).
From these inequalities and (1) it follows that
P
r,s (n) Γ(r + 1)Γ(s + 1)
lim r+s+1 ≥ ζ(r + 1), (12)
n Γ(r + s + 2)
if r > 0 and s ≥ 0; and
P
r,s (n)
Γ(r + 1)Γ(s + 1)
lim r+s+1
≤ ζ(r + 1)ζ(s), (13)
n Γ(r + s + 2)

if r > 0 and s > 1. Thus n−r−s−1 r,s (n) oscillates between limits included in the interval
P

Γ(r + 1)Γ(s + 1) Γ(r + 1)Γ(s + 1)

ζ(r + 1), ζ(r + 1)ζ(s).
Γ(r + s + 2) Γ(r + s + 2)

On the other hand n−r−s−1 σr+s+1 (n) oscillates between 1 and ζ(r + s + 1), assuming values
as near as we please to either of
Pthese limits. The formula (8) shews that the actual limits
of indetermination of n−r−s−1 r,s (n) are

Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)

,
Γ(r + s + 2) ζ(r + s + 2)
Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)ζ(r + s + 1)
. (14)
Γ(r + s + 2) ζ(r + s + 2)
Naturally
ζ(r + 1)ζ(s + 1) ζ(r + 1)ζ(s + 1)ζ(r + s + 1)
ζ(r + 1) < < < ζ(r + 1)ζ(s) .†
ζ(r + s + 2) ζ(r + s + 2)
∗
(10) follows from this as an immediate corollary.
On certain arithmetical functions 177

What is remarkable about the formula (8) is that it shews the asymptotic equality of two
functions neither of which itself increases in a regular manner.

4. It is easy to see that, if n is a positive integer, then

cot 21 θ sin nθ = 1 + 2 cos θ + 2 cos 2θ + · · · + 2 cos(n − 1)θ + cos nθ.

Suppose now that

2
x sin θ x2 sin 2θ x3 sin 3θ

1 1
4 cot 2θ + + + + ···
1−x 1 − x2 1 − x3
= ( 14 cot 12 θ)2 + C0 + C1 cos θ + C2 cos 2θ + C3 cos 3θ + · · · ,

where Cn is independent of θ. Then we have

x2 x3
 
1 x
C0 = 2 + + + ···
1 − x 1 − x2 1 − x3
2 2 3 2
( 2   )
x x x
+ 21 + + + ···
1−x 1 − x2 1 − x3

x2 x3
 
1 x
= 2 + + + ···
(1 − x)2 (1 − x2 )2 (1 − x3 )2
2x2 3x3
 
x
= 21 + + + · · · . (15)
1 − x 1 − x2 1 − x3

Again

1 xn xn+1 xn+2 xn+3

Cn = 21 + + + + ···
− xn 1 − xn+1 1 − xn+2 1 − xn+3
x xn+1 x2 xn+2 x3 xn+3
+ · + · + · + ···
1 − x 1 − xn+1 1 − x2 1 − xn+2 1 − x3 1 − xn+3
xn−1 x2 xn−2 xn−1
 
1 x x
− 2 · + · + ··· + · .
1 − x 1 − xn−1 1 − x2 1 − xn−2 1 − xn−1 1 − x

Hence
xn+1 x2 xn+2
   
Cn x
(1 − xn ) = 1
2 + − + − + ···
xn 1 − x 1 − xn+1 1 − x2 1 − xn+2
xn−1 x2 xn−2
   
1 x
− 2 1+ + + 1+ +
1 − x 1 − xn−1 1 − x2 1 − xn−2
†
For example when r = 1 and s = 9 this inequality becomes 1.64493 . . . < 1.64616 . . . < 1.64697 . . . <
1.64823 . . ..
178 Paper 18

xn−1
 
x
+··· + 1+ +
1 − xn−1 1 − x
1 n
= − .
1 − xn 2
That is to say
xn nxn
Cn = − . (16)
(1 − xn )2 2(1 − xn )
It follows that
2
x sin θ x2 sin 2θ x3 sin 3θ

1
4 cot 12 θ + + + + ···
1−x 1 − x2 1 − x3

1 x cos θ x2 cos 2θ

2 x3 cos 3θ
= 4 cot 21 θ
+ + + + ···
(1 − x)2 (1 − x2 )2 (1 − x3 )2
2x2 3x3
 
1 x
+2 (1 − cos θ) + (1 − cos 2θ) + (1 − cos 3θ) + · · · . (17)
1−x 1 − x2 1 − x3
Similarly, using the equation
cot2 12 θ(1 − cos nθ) =
(2n − 1) + 4(n − 1) cos θ + 4(n − 2) cos 2θ + · · · + 4 cos(n − 1)θ + cos nθ,

we can shew that

2x2

x
1
8 cot2 21 θ + 1
+
(1 − cos θ) +
12 (1 − cos 2θ)+
1−x 1 − x2
2
3x3 1 2
(1 − cos 3θ) + · · · = 18 cot2 12 θ + 12


1−x 3
 3
23 x2 33 x3

1 1 x
+ 12 (5 + cos θ) + (5 + cos 2θ) + (5 + cos 3θ) + · · · . (18)
1−x 1 − x2 1 − x3

For example, putting θ = 32 π and θ = 21 π in (17), we obtain

2
x2 x4 x5

1 x
+6 − + − + ···
1 − x 1 − x2 1 − x4 1 − x5
2x2 4x4 5x5
 
1 1 x
= 36 +3 + + + + ··· , (19)
1 − x 1 − x2 1 − x4 1 − x5
where 1, 2, 4, 5, . . . are the natural numbers without the multiples of 3; and
2
x3 x5 x7

1 x
4 + 1 − x − 1 − x3 + 1 − x5 − 1 − x7 + · · ·
On certain arithmetical functions 179

2x2 3x3 5x5

 
1 1 x
= 16 + 2 + + + + ··· , (20)
1 − x 1 − x2 1 − x3 1 − x5
where 1, 2, 3, 5, . . . are the natural numbers without the multiples of 4.

5. It follows from (18) that

2
θ2 θ4 θ6

1
+ S3 − S5 + S7 − · · ·
2θ 2 2! 4! 6!
 2
θ4 θ6

1 1 1 θ
= 4 + 2 S3 − 12 S5 − S7 + S9 − · · · , (21)
4θ 2! 4! 6!
where Sr is the same as in (9). Equating the coefficients of θ n in both sides in (21), we
obtain
   
(n − 2)(n + 5) n n
Sn+3 = S3 Sn−1 + S5 Sn−8 +
12(n + 1)(n + 2) 2 4
   
n n
S7 Sn−5 + · · · + Sn−1 S3 , (22)
6 n−2
where  
n n!
= ,
r r!(n − r)!
if n is an even integer greater than 2.
Let us now suppose that
m=∞
X n=∞
X
Φr,s (x) = mr ns xmn , (23)
m=1 n=1

so that
Φr,s (x) = Φs,r (x),
and

1s x 2s x2 3s x3
Φ0,s (x) = + + + · · · = Ss − 12 ζ(−s),


1 − x 1 − x2 1 − x3


. (24)
1s x 2s x2 3s x3 
Φ1,s (x) = + + + ···


(1 − x)2 (1 − x2 )2 (1 − x3 )2


Further let
2x2 3x3
  
x
P = −24S1 = 1 − 24 + + + · · · ∗,


1 − x 1 − x2 1 − x3




 3 
23 x2 33 x3
 
1 x

Q = 240S3 = 1 + 240 + + + · · · , . (25)
1 − x 1 − x2 1 − x3 

 5 
25 x2 35 x3
 
1 x


R = −540S5 = 1 − 504 + + + ···


1 − x 1 − x2 1 − x3

180 Paper 18

The putting n = 4, 6, 8, . . . in (22) we obtain the results contained in the following table.

TABLE 1

1. 1 − 24Φ0,1 (x) = P.
2. 1 + 240Φ0,3 (x) = Q.
3. 1 − 504Φ0,5 (x) = R.
4. 1 + 480Φ0,7 (x) = Q2 .
5. 1 − 264Φ0,9 (x) = QR.
6. 691 + 65520Φ0,11 (x) = 441Q3 + 250R2 .
7. 1 − 24Φ0,13 (x) = Q2 R.
8. 3617 + 16320Φ0,15 (x) = 1617Q4 + 2000QR2 .
9. 43867 − 28728Φ0,17 (x) = 38367Q3 R + 5500R3 .
10. 174611 + 13200Φ0,19 (x) = 53361Q5 + 121250Q2 R2 .
11. 77683 − 552Φ0,21 (x) = 57183Q4 R + 20500QR3 .
12. 236364091 + 131040Φ0,23 (x) = 49679091Q6 + 176400000Q3 R2
+10285000R4 .
13. 657931 − 24Φ0,25 (x) = 392931Q5 R + 265000Q2 R3 .
14. 3392780147 + 6960Φ0,27 (x) = 489693897Q7 + 2507636250Q4 R2
+395450000QR4 .
15. 1723168255201 − 171864Φ0,29 (x) = 815806500201Q6 R
+881340705000Q3 R3 + 26021050000R5 .
16. 7709321041217 + 32640Φ0,31 (x) = 764412173217Q8
+5323905468000Q5 R2 + 1621003400000Q2 R4 .
In general
X
1
2 ζ(−s) + Φ0,s (x) = Km,n Qm Rn , (26)

where Km,n is a constant and m and n are positive integers (including zero) satisfying the
equation
4m + 6n = s + 1.
This is easily proved by induction, using (22).
∗
If x = q 2 , then in the notation of elliptic functions
 2  
12ηω 2K 3E 2
P = = + k − 2 ,
π2 π K
4
12g2 ω 4

2K
Q = = (1 − k2 + k4 ),
π4 π
6
216g3 ω 6

2K
R = = (1 + k2 )(1 − 2k2 )(1 − 12 k2 ).
π6 π
On certain arithmetical functions 181

6. Again from (17) we have

2
θ3 θ5

1 θ
+ S1 − S3 + S5 − · · ·
2θ 1! 3! 5!
1 θ2 θ4 θ6
= 2
+ S1 − Φ1,2 (x) + Φ1,4 (x) − Φ1,6 (x) + · · ·
4θ 2! 4! 6!
 2 4 θ6

1 θ θ
+2 S3 − S5 + S7 − · · · . (27)
2! 4! 6!
Equating the coefficients of θ n in both sides in (27) we obtain
   
n+3 n n
Sn+1 − Φ1,n (x) = S1 Sn−1 + S3 Sn−3 +
2(n + 1) 1 3
   
n n
S5 Sn−5 + · · · + Sn−1 S1 , (28)
5 n−1
if n is a positive even integer. From this we deduce the results contained in Table II.

TABLE II
1. 288Φ1,2 (x) = Q − P 2 .
2. 720Φ1,4 (x) = P Q − R.
3. 1008Φ1,6 (x) = Q2 − P R.
4. 720Φ1,8 (x) = Q(P Q − R).
5. 1584Φ1,10 (x) = 3Q3 + 2R2 − 5P QR.
6. 65520Φ1,12 (x) = P (441Q3 + 250R2 ) − 691Q2 R.
7. 144Φ1,14 (x) = Q(3Q3 + 4R2 − 7P QR).
In general
X
Φ1,s (x) = Kl,m,n P l Qm Rn , (29)

where l ≤ 2 and 2l + 4m + 6n = s + 2. This is easily proved by induction, using (28).

7. We have
dP P2 − Q

x = − 24Φ1,2 (x) = , 

dx 12




dQ PQ − R

x = 240Φ1,4 (x) = , (30)
dx 3 


dR P R − Q2


x = − 504Φ1,6 (x) =


dx 2
Suppose now that r < s and that r + s is even. Then
d r
 
Φr,s (x) = x Φ0,s−r (x), (31)
dx
182 Paper 18

and Φ0,s−r (x) is a polynomial in Q and R. Also

dP dQ dR
x ,x ,x
dx dx dx
are polynomials in P, Q and R. Hence Φr,s (x) is a polynomial in P, Q and R. Thus we
deduce the results contained in Table III.

TABLE III
1. 1728Φ2,3 (x) = 3P Q − 2R − P 3 .
2. 1728Φ2,5 (x) = P 2 Q − 2P R + Q2 .
3. 1728Φ2,7 (x) = 2P Q2 − P 2 R − QR.
4. 8640Φ2,9 (x) = 9P 2 Q2 − 18P QR + 5Q3 + 4R2 .
5. 1728Φ2,11 (x) = 6P Q3 − 5P 2 QR + 4P R2 − 5Q2 R.
6. 6912Φ3,4 (x) = 6P 2 Q − 8P R + 3Q2 − P 4 .
7. 3456Φ3,6 (x) = P 3 Q − 3P 2 R + 3P Q2 − QR.
8. 5184Φ3,8 (x) = 6P 2 Q2 − 2P 3 R − 6P QR + Q3 + R2 .
9. 20736Φ4,5 (x) = 15P Q2 − 20P 2 R + 10P 3 Q − 4QR − P 5 .
10. 41472Φ4,7 (x) = 7(P 4 Q − 4P 3 R + 6P 2 Q2 − 4P QR) + 3Q3 + 4R2 .
In general
X
Φr,s (x) = Kl,m,n P l Qm Rn , (32)

where l − 1 does not exceed the smaller of r and s and

2l + 4m + 6n = r + s + 1.

The results contained in these three tables are of course really results in the theory of
elliptic functions. For example Q and R are substantially the invariants g2 and g3 , and the
formulæ of Table I are equivalent to the formulæ which express the coefficients in the series
1 g2 u2 g3 u4 g22 u6 3g2 g3 u8
℘(u) = + + + + + ···
u2 20 28 1200 6160
in terms of g2 and g3 . The elementary proof of these formulæ given in the preceding sections
seems to be of some interest in itself.

8. In what follows we shall require to know the form of Φ1,s (x) more precisely than is
shewn by the formula (29).
We have
X
1
2 ζ(−s) + Φ0,s (x) = Km,n Qm Rn , (33)

where s is an odd integer greater than 1 and 4m + 6n = s + 1. Also

d m n m n 
x (Qm Rn ) = + P Qm R n − Qm−1 Rn+1 + Qm+2 Rn−1 . (34)
dx 3 2 3 3
On certain arithmetical functions 183

Differentiating (33) and using (34) we obtain

X
Φ1,s+1 (x) = 1
12 (s + 1)P { 12 ζ(−s) + Φ0,s (x)} + Km,n Qm Rn , (35)

where s is an odd integer greater than 1 and 4m + 6n = s + 3. But when s = 1 we have

Q − P2
Φ1,2 (x) = . (36)
288

9. Suppose now that

Fr,s (x) = { 21 ζ(−r) + Φ0,r (x)}{ 12 ζ(−s) + Φ0,s (x)}
ζ(1 − r) + ζ(1 − s) Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)
− Φ1,r+s (x) −
r+s Γ(r + s + 2) ζ(r + s + 2)
×{ 12 ζ(−r − s − 1) + Φ0,r+s+1 (x)}. (37)

Then it follows from (33), (35) and (36) that, if r and s are positive odd integers,
X
Fr,s (x) = Km,n Qm Rn , (38)

where
4m + 6n = r + s + 2.
But it is easy to see, from the functional equation satisfied by ζ(s), viz.

(2π)−s Γ(s)ζ(s) cos 12 πs = 21 ζ(1 − s), (39)

that

Fr,s (0) = 0. (40)

Hence Q3 − R2 is a factor of the right-hand side in (38), that is to say

X
Fr,s (x) = (Q3 − R2 ) Km,n Qm Rn , (41)

where
4m + 6n = r + s − 10.

10. It is easy to deduce from (30) that

d
x log(Q3 − R2 ) = P. (42)
dx
But it is obvious that
d
P =x log[x{(1 − x)(1 − x2 )(1 − x3 ) · · ·}24 ]; (43)
dx
184 Paper 18

and the coefficient of x in Q3 − R2 = 1728. Hence

Q3 − R2 = 1728x{(1 − x)(1 − x2 )(1 − x3 ) · · ·}24 . (44)

But it is known that

{(1 − x)(1 − x2 )(1 − x3 )(1 − x4 ) · · ·}3

= 1 − 3x + 5x3 − 7x6 + 9x10 − · · · (45)

Hence

Q3 − R2 = 1728x(1 − 3x + 5x3 − 7x6 + · · ·)8 . (46)

The coefficient of xν−1 in 1−3x+5x3 −· · · is numerically less than (8ν), and the coefficient
p

of xν in Q3 − R2 is therefore numerically less than that of xν in

1728x{ (8ν)(1 + x + x3 + x6 + · · ·)}8 .

p

But
13 x 23 x2 33 x3
x(1 + x + x3 + x6 + · · ·)8 = 2
+ 4
+ + ···, (47)
1−x 1−x 1 − x6
and the coefficient of xν in the right-hand side is positive and less than
 
1 1 1
ν3 + + + · · · .
13 33 53

Hence the coefficient of xν in Q3 − R2 is of the form

ν 4 O(ν 3 ) = O(ν 7 ).

That is to say
X
Q3 − R 2 = O(ν 7 )xν . (48)

Differentiating (48) and using (42) we obtain

X
P (Q3 − R2 ) = O(ν 8 )xν . (49)

Differentiating this again with respect to x we have

X
A(P 2 − Q)(Q3 − R2 ) + BQ(Q3 − R2 ) = O(ν 9 )xν ,

where A and B are constants. But

12 x 22 x2
 
2
P − Q = −288Φ1,2 (x) = −288 + + ··· ,
(1 − x)2 (1 − x2 )2
On certain arithmetical functions 185

and the coefficient of xν in the right-hand side is a constant multiple of νσ1 (ν). Hence
X X
(P 2 − Q)(Q3 − R2 ) = Oνσ1 (ν)xν O(ν 7 )xν
X
= O(ν 8 ){σ1 (1) + σ1 (2) + · · ·
X
+σ1 (ν)}xν = O(ν 10 )xν ,

and so
X
Q(Q3 − R2 ) = O(ν 10 )xν . (50)

Differentiating this again with respect to x and using arguments similar to those used above,
we deduce
X
R(Q3 − R2 ) = O(ν 12 )xν . (51)

Suppose now that m and n are any two positive integers including zero, and that m + n is
not zero. Then

Qm Rn (Q3 − R2 ) = Q(Q3 − R2 )Qm−1 Rn

X X X
= O(ν 10 )xν { O(ν 3 )xν }m−1 { O(ν 5 )xν }n
X X X
= O(ν 10 )xν O(ν 4m−5 )xν O(ν 6n−1 )xν

X
= O(ν 4m+6n+6 )xν ,

If m is not zero, Similarly we can shew that

Qm Rn (Q3 − R2 ) = R(Q3 − R2 )Qm Rn−1

X
= O(ν 4m+6n+6 )xν ,

if n is not zero. Therefore in any case

X
(Q3 − R2 )Qm Rn = O(ν 4m+6n+6 )xν . (52)

11. Now let r and s be any two positive odd integers including zero. Then, when r + s is
equal to 2,4,6,8 or 12, there are no values of m and n satisfying the relation

4m + 6n = r + s − 10

in (41); consequently in these cases

Fr,s (x) = 0. (53)

186 Paper 18

When r + s = 10, m and n must both be zero, and this result does not apply; but it follows
from (41) and (48) that

X
Fr,s (x) = O(ν 7 )xν . (54)

And when r + s ≥ 14 it follows from (52) that

X
Fr,s (x) = O(ν r+s−4 )xν . (55)

Equating the coefficients of xν in both sides in (53), (54) and (55) we obtain

X Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)

(n) = σr+s+1 (n)
r,s
Γ(r + s + 2) ζ(r + s + 2)

ζ(1 − r) + ζ(1 − s)
+ nσr+s−1 (n) + Er,s (n), (56)
r+s

where

Er,s (n) = 0, r + s = 2, 4, 6, 8, 12;

Er,s (n) = O(n7 ), r + s = 10;
Er,s (n) = O(nr+s−4 ), r + s ≥ 14.

Since σr+s+1 (n) is of order nr+s+1 , it follows that in all cases

X Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)

(n) ∼ σr+s+1 (n). (57)
r,s
Γ(r + s + 2) ζ(r + s + 2)

P
The following table gives the values of r,s (n) when r + s = 2, 4, 6, 8, 12.

TABLE IV
P 5σ3 (n)−6nσ1 (n)
1. 1,1 (n) = 12 .
P 7σ5 (n)−10nσ3 (n)
2. 1,3 (n) = 80 .
On certain arithmetical functions 187

P σ7 (n)
3. 3,3 (n) = 120 .

P 10σ7 (n)−21nσ5 (n)

4. 1,5 (n) = 252 .
P 11σ9 (n)
5. 3,5 (n) = 5040 .

P 11σ9 (n)−30nσ7 (n)

6. 1,7 (n) = 480 .
P σ13 (n)
7. 5,7 (n) = 10080 .

P σ13 (n)
8. 3,9 (n) = 2640 .

P 691σ13 (n)−2730nσ11 (n)

9. 1,11 (n) = 65520 .

12. In this connection it may be interesting to note that

σ1 (1)σ3 (n) + σ1 (3)σ3 (n − 1) + σ1 (5)σ3 (n − 2) + · · ·

1
+σ1 (2n + 1)σ3 (0) = 240 σ5 (2n + 1). (58)

This formula may be deduced from the identity

15 x 35 x2 55 x3
+ + + ···
1 − x 1 − x3 1 − x5
3x2 5x3
 
x
=Q + + + ··· , (59)
1 − x 1 − x3 1 − x5

which can be proved by means of the theory of elliptic functions or by elementary methods.

13. More precise results concerning the order of Er,s (n) can be deduced from the theory
of elliptic functions. Let
x = q2.
Then we have
Q = φ8 (q){1 − (kk′ )2 }




R = φ12 (q)(k′2 − k2 ){1 + 12 (kk′ )2 } , (60)

= φ12 (q){1 + 12 (kk′ )2 } {1 − (2kk′ )2 }
p 


where φ(q) = 1 + 2q + 2q 4 + 2q 9 + · · ·
But, if
1
f (q) = q 24 (1 − q)(1 − q 2 )(1 − q 3 ) · · · ,
188 Paper 18

then we know that

1 1 1

2 6 f (q)= k 12 k′ 3 φ(q) 



1 1 
6 ′
2 f (−q) = (kk ) φ(q) 
12

1 1
(61)
2 3 f (q 2 ) = (kk′ ) 6 φ(q) 



2 1 1


2 3 f (q 4 ) ′
= k 3 k 12 φ(q)


It follows from (41), (60) and (61) that, if r + s is of the form 4m + 2, but not equal to 2
or to 6, then
1
(r+s−6)
f 4(r+s−4) (−q) 4 X f 24n (q 2 )
Fr,s (q 2 ) = 2(r+s−10) 2 Kn , (62)
f (q ) 1
f 24n (−q)

and if r + s is of the form 4m, but not equal to 4, 8 or 12, then

1
(r+s−8)
f 4(r+s−6) (−q) 4 X
f 24n (q 2 )
Fr,s (q 2 ) = 2(r+s−10) 2 {f 8 (q) − 16f 8 (q 4 )} Kn , (63)
f (q ) f 24n (−q)
1

when Kn depends on r and s only. Hence it is easy to see that in all cases Fr,s (q 2 ) can be
expressed as
b  c  d
f 5 (−q) f 5 (q 2 ) f 5 (q) 3
X 
3 a
Ka,b,c,d,e,h,k {f (−q)} f (q)
f 2 (q 2 ) f 2 (−q) f 2 (q 2 )
e
f 5 (q 4 ) 3 4

× 2 2
f (q ) f h (−q)f k (q 2 ), (64)
f (q )

where a, b, c, d, e, h, k are zero or positive integers such that

a + b + c + 2(d + e) = [ 32 (r + s + 2)],
h + k = 2(r + s + 2) − 3[ 32 (r + s + 2)],

and [x] denotes as usual the greatest integer in x. But

12 52 72 112

f (q) = q 24 − q 24 − q 24 + q 24 + ··· 




12 32 52 72

f 3 (q) = q − 3q + 5q − 7q + ···

8 8 8 8 

12 52 72 112
, (65)
f 5 (q)
f 2 (q 2 ) = q 24 − 5q 24 + 7q 24 − 11q + ··· 
24 




f 5 (q 2 ) 12 22 42 52


f 2 (−q) = q 3 − 2q 3 + 4q 3 − 5q 3 + ··· 
On certain arithmetical functions 189

where 1, 2, 4, 5, · · · are the natural numbers without the multiples of 3, and 1, 5, 7, 11, · · ·
are the natural odd numbers without the multiples of 3.
Hence it is easy to see that
1
n− 2 (a+b+c)−d−e Er,s (n)
is not of higher order than the coefficient of q 2n in
1 1 1 1 1 2 1 1 1
φa (q 8 )φb (q 24 )φc (q 3 ){φ(q 24 )φ(q 8 )}d {φ(q 3 )φ(q 2 )}e φh (q 24 )φk (q 12 ),

or the coefficient of q 48n in

φa+d (q 3 )φb+d+h (q)φc (q 8 )φe (q 16 )φe (q 12 )φk (q 2 ).

But the coefficient of q ν in φ2 (q 2 ) cannot exceed that of q ν in φ2 (q), since

φ2 (q) + φ2 (−q) = 2φ2 (q 2 ); (66)

and it is evident that the coefficient of q ν in φ(q 4λ ) cannot exceed that of q ν in φ(q λ ). Hence
it follows that
1 2
n− 2 [ 3 (r+s+2)] Er,s (n)
is not of higher order than the coefficient of q 48n in

φA (q)φB (q 3 )φC (q 2 ),

where A, B, C are zero or positive integers such that

A + B + C = 2(r + s + 2) − 2[ 23 (r + s + 2)],

and C is 0 or 1.
Now, if r + s ≥ 14, we have
A + B + C ≥ 12,
and so
A + B ≥ 11.
Therefore one at least of A and B is greater than 5. But
∞
X
6
φ (q) = O(ν 2 )q ν .∗ (67)
0

Hence it is easily deduced that

X 1
φA (q)φB (q 3 )φC (q 2 ) = O{ν 2 (A+B+C)−1 }q ν . (68)
∗
See §§24–25.
190 Paper 18

It follows that
1 2
Er,s (n) = O{nr+s− 2 [ 3 (r+s−1)] }, (69)

If r + s ≥ 14. We have already shewn in § 11 that, if r + s = 10, then

Er,s (n) = O(n7 ). (70)

This agrees with (69). Thus we see that in all cases

2
Er,s (n) = O{n 3 (r+s+1) }; (71)

and that, if r + s is of the form 6m, then

2 3
Er,s (n) = O{n 3 (r+s+ 4 ) }, (72)

and if of the form 6m + 4, then

2 1
Er,s (n) = O{n 3 (r+s+ 2 ) }. (73)

1
14. I shall now prove that the order of Er,s (n) is not less than that of n 2 (r+s) . In order to
prove this result I shall follow the method used by Messrs Hardy and Littlewood in their
paper “Some problems of Diophantine approximation” (II) ∗ .
Let
q = eπiτ , q ′ = eπiT ,
where
c + dτ
T = ,
a + bτ
and
ad − bc = 1.
Also let
v
V = .
a + bτ
Then we have
√ √
ω veπibvV ϑ1 (v, τ ) = V ϑ1 (V, T ), (74)

where ω is an eighth root of unity and

1
2n 2n
ϑ1 (v, τ ) = 2 sin πv · q 4 Π∞
1 (1 − q )(1 − 2q cos 2πv + q 4n ). (75)
∗
Acta Mathematica, Vol. XXXVII, pp. 193 – 238.
On certain arithmetical functions 191

From (75) we have

∞ 2n
X q (1 + 2 cos 2nπv)
log ϑ1 (v, τ ) = log(2, sin πv) + 41 log q − . (76)
1
n(1 − q 2n )

It follows from (74) and (76) that

∞ 2n
X q (1 + 2 cos 2nπv)
log sin πv + 12 log v + 14 log q + log ω −
n(1 − q 2n )
1
∞ ′ 2n
1
X q (1 + 2 cos 2nπV )
= log sin πV + 2 log V + 14 log q ′ − πibvV − . (77)
1
n(1 − q ′ 2n )

Equating the coefficients of v 8+1 on the two sides of (77), we obtain

1s q 2 2s q 4 3s q 6
 
s+1 1
(a + bτ ) 2 ζ(−s) + 1 − q 2 + 1 − q 4 + 1 − q 6 + · · ·

1s q ′2 2s q ′4 3s q ′6
= 21 ζ(−s) + + + + ···, (78)
1 − q ′2 1 − q ′4 1 − q ′6

provided that s is an odd integer greater than 1. If, in particular,we put s = 3 and s = 5
in (78) we obtain
 3 2
23 q 4 33 q 6
 
4 1 q
(a + bτ ) 1 + 240 + + + ···
1 − q2 1 − q4 1 − q6
 3 ′2
23 q ′4 33 q ′6
 
1 q
= 1 + 240 + + + ··· , (79)
1 − q ′2 1 − q ′4 1 − q ′6

and
 5 2
25 q 4 35 q 6
 
6 1 q
(a + bτ ) 1 − 504 + + + ···
1 − q2 1 − q4 1 − q6
 5 ′2
25 q ′4 35 q ′6
 
1 q
= 1 − 504 + + + · · · . (80)
1 − q ′2 1 − q ′4 1 − q ′6

It follows from (38), (79) and (80) that

(a + bτ )r+s+2 Fr,s (q 2 ) = Fr,s (q ′2 ). (81)

It can easily be seen from (56) and (37) that

∞
X
Fr,s (x) = Er,s (n)xn . (82)
1
192 Paper 18

Hence
∞
X ∞
X
(a + bτ )r+s+2 Er,s (n)q 2n = Er,s (n)q ′2n . (83)
1 1

It is important to observe that

ζ(−r) + ζ(−s) ζ(1 − r) + ζ(1 − s)
Er,s (1) = −
2 r+s
Γ(r + 1)Γ(s + 1) ζ(r + 1)ζ(s + 1)
− 6= 0, (84)
Γ(r + s + 2) ζ(r + s + 2)
if r + s is not equal to 2,4,6,8 or 12. This is easily proved by the help of the equation (39).

15. Now let

τ = u + iy, t = e−πy (u > 0, y > 0, 0 < t < 1),
so that
q = eπiu−πy = teπiu ;
and let us suppose that pn /qn is a convergent to
1 1 1
u= ,
a1 + a2 + a3 + · · ·
so that
ηn = pn−1 qn − pn qn−1 = ±1.
Further, let us suppose that
a = pn , b = −qn ,
c = ηn pn−1 , d = −ηn qn−1 ,
so that
ad − bc = ηn2 = 1.
Furthermore, let
′
y = 1/(qn qn+1 ),
where
′
qn+1 = a′n+1 qn + qn−1 ,
and a′n+1 is the complete quotient corresponding to an+1 .
Then we have
√
| ± 1 − i| 2
|a + bτ | = |pn − qn u − iqn y| = ′ = ′ , (85)
qn+1 qn+1

and
|q ′ | = e−πλ ,
On certain arithmetical functions 193

where
   
c + dτ d 1
λ = I(T ) = I +I −
a + bτ b b(a + bτ )
′
qn+1
y
= ′ = , (86)
(1/qn+1 )2 + qn2 y 2 2qn

and I(T ) is the imaginary part of T . It follows from (83), (85) and (86) that
∞ ′ r+s+2 X∞
qn+1
X 
| Er,s (n)q 2n | = √ | Er,s (n)q ′2n |
1
2 1
′ r+s+2
qn+1

≥ √ {|Er,s (1)|e−2πλ − |Er,s (2)e−4πλ − |Er,s (3)|e−6πλ − · · ·}. (87)
2
We can choose a number λ0 , depending only on r and s, such that

|Er,s (1)|e−2πλ > 2{|Er,s (2)|e−4πλ + |Er,s (3)|e−6πλ + · · ·}

for λ ≥ λ0 . Let us suppose λ0 > 10. Let us also suppose that the continued fraction for u
satisfies the condition
′
4λ0 qn > qn+1 > 2λ0 qn (88)

for an infinity of values of n. Then

∞   ′ r+s+2
X 
2n  qn+1
E (n)q ≥ 1
|E (1)| √ e−4πλ0 > K(qn+1
′
)r+s+2 , (89)

 r,s  2 r,s

1
 2

where K depends on r and s only. Also

′
qn qn+1 = 1/y,
s 
′ 1 π K
qn+1 >√ = >p .
y log(1/t) (1 − t)
It follows that, if u is an irrational number such that the condition (88) is satisfied for an
infinity of values of n, then
 
X∞  1
Er,s (n)q 2n  > K(1 − t)− 2 (r+s+2) (90)
 

 
1

for an infinity of values of t tending to unity. But if we had

1
Er,s (n) = o{n 2 (r+s) }
194 Paper 18

then we should have

∞
X 1
| Er,s (n)q 2n | = o{(1 − t)− 2 (r+s+2) },
1
P
which contradicts (90). It follows that the error term in r,s (n) is not of the form
1
o{n 2 (r+s) }. (91)

The arithmetical function τ (n).

16. We have seen that

Er,s (n) = 0,
P
if r + s is equal to 2,4,6,8, or 12. In these P
cases r,s (n) has a finite expression in terms of
σr+s+1 (n) and σr+s−1 (n). In other cases r,s (n) involves other arithmetical functions as
well. The simplest of these is the function τ (n) defined by
∞
X
τ (n)xn = x{(1 − x)(1 − x2 )(1 − x3 ) · · ·}24 . (92)
1

These cases arise when r + s has one of the values 10, 14, 16, 18, 20 or 24.
Suppose that r + s has one of these values. Then
1728 ∞ n
P
1 Er,s (n)x
(Q3 − R2 )Er,s (1)
is, by (41) and (82), equal to the corresponding one of the functions

1, Q, R, Q2 , QR, Q2 R.

In other words
∞
X ∞
X
n
Er,s (n)x = Er,s (1) τ (n)xn
1 1
∞
( )
2 X xn
1+ nr+s−11 . (93)
ζ(11 − r − s) 1 − xn
1

We thus deduce the formulæ

Er,s (n) = Er,s (1)τ (n), (94)

if r + s = 10; and

σr+s−11 (0)Er,s (n) = Er,s (1){σr+s−11 (0)τ (n)

On certain arithmetical functions 195

+σr+s−11 (1)τ (n − 1) + · · · + σr+s−11 (n − 1)τ (1)}, (95)

if r + s is equal to 14, 16, 18, 20 or 24. It follows from (94) and (95) that, if r + s = r ′ + s′ ,
then

Er,s (n)Er′ ,s′ (1) = Er,s (1)Er′ ,s′ (n), (96)

and in general

Er,s (m)Er′ ,s′ (n) = Er,s (n)Er′ ,s′ (m), (97)

when r + s has one of the values in question. The different cases in which r + s has the
same value are therefore not fundamentally distinct.

17. The values of τ (n) may be calculated as follows: differentiating (92) logarithmically
with respect to x, we obtain

∞
X ∞
X
n
nτ (n)x = P τ (n)xn . (98)
1 1

Equating the coefficients of xn in both sides in (98),we have

24
τ (n) = {σ1 (1)τ (n − 1) + σ1 (2)τ (n − 2) + · · · + σ1 (n − 1)τ (1)}. (99)
1−n

If, instead of starting with (92), we start with

∞
X
τ (n)xn = x(1 − 3x + 5x3 − 7x6 + · · ·)8 ,
1

we can shew that

(n − 1)τ (n) − 3(n − 10)τ (n − 1) + 5(n − 28)τ (n − 3) − 7

(n − 55)τ (n − 6) + · · · to [ 12 {1 + (8n − 7)}] terms = 0,
p
(100)

where the rth term of the sequence 0,1,3,6, . . . is 12 r(r − 1), and the rth term of the
sequence 1,10,28,55, . . . is 1 + 29 r(r − 1). We thus obtain the values of τ (n) in the following
table.

TABLE V
196 Paper 18

n τ (n) n τ (n)
1 +1 16 +987136
2 −24 17 −6905934
3 +252 18 +2727432
4 −1472 19 +10661420
5 +4830 20 −7109760
6 −6048 21 −4219488
7 −16744 22 −12830688
8 +84480 23 +18643272
9 −113643 24 +21288960
10 −115920 25 −25499225
11 +534612 26 +13865712
12 −370944 27 −73279080
13 −577738 28 +24647168
14 +401856 29 +128406630
15 +1217160 30 −29211840

18. Let us consider more particularly the case in which r + s = 10. The order of Er,s (n)
is then the same as that of τ (n). The determination of this order is a problem interesting
in itself. We have proved that Er,s (n), and therefore τ (n), is of the form O(n7 ) and not of
11
the form o(n5 ). There is reason for supposing that τ (n) is of the form O(n 2 +ǫ ) and not of
11
the form o(n 2 ). For it appears that
∞
X τ (n) Y 1
= . (101)
nt p
1− τ (p)p−t + p11−2t
1

This assertion is equivalent to the assertion that, if

n = pa11 pa22 pa33 · · · par r ,

where p1 , p2 , . . . , pr are the prime divisors of n, then

11 sin(1 + a1 )θp1 sin(1 + a2 )θp2 sin(1 + ar )θpr
n− 2 τ (n) = ··· , (102)
sin θp1 sin θp2 sin θpr
where
11
cos θp = 12 p− 2 τ (p).
It would follow that, if n and n′ are prime to each other, we must have

τ (nn′ ) = τ (n)τ (n′ ). (103)

Let us suppose that (102) is true, and also that (as appears to be highly probable)

{2τ (p)}2 ≤ p11 , (104)

On certain arithmetical functions 197

so that θp is real. Then it follows from (102) that

11
n− 2 |τ (n)| ≤ (1 + a1 )(1 + a2 ) · · · (1 + ar ),

that is to say
11
|τ (n)| ≤ n 2 d(n), (105)

where d(n) denotes the number of divisors of n.

Now let us suppose that n = pa , so that
11 sin(1 + a)θp
n− 2 τ (n) = .
sin θp
Then we can choose a as large as we please and such that
 
 sin(1 + a)θp 
  ≥ 1.
 sin θp 

Hence
11
|τ (n)| ≥ n 2 (106)

for an infinity of values of n.

19. It should be observed that precisely similar questions arise with regard to the arith-
metical function Ψ(n) defined by
∞
X
Ψ(n)xn = f a1 (xc1 )f a2 (xc2 ) · · · f ar (xcr ), (107)
0

where
1
f (x) = x 24 (1 − x)(1 − x2 )(1 − x3 ) · · · ,
the a’s and c’s are integers, the latter being positive,
1
24 (a1 c1 − a2 c2 + · · · + ar cr )

is equal to 0 or 1, and  
a1 a2 ar
l + + ··· + ,
c1 c2 cr
where l is the least common multiple of c1 , c2 , . . . , cr , is equal to 0 or to a divisor of 24.
The arithmetical functions χ(n), P (n), χ4 (n), Ω(n) and Θ(n), studied by Dr. Glaisher in
the Quarterly Journal, Vols. XXXVI-XXXVIII, are of this type. Thus
∞
X
χ(n)xn = f 6 (x4 ),
1
198 Paper 18

∞
X
P (n)xn = f 4 (x2 )f 4 (x4 ),
1
∞
X
χ4 (n)xn = f 4 (x)f 2 (x2 )f 4 (x4 ),
1
∞
X
Ω(n)xn = f 12 (x2 ),
1
∞
X
Θ(n)xn = f 8 (x)f 8 (x2 ).
1

20. The results (101) and (104) may be written as

∞
X Er,s (n) Y 1
= Er,s (1) , (108)
nt p
1 − 2cp p−t + pr+s+1−2t
1

where
c2p ≤ pr+s+1 ,
and
2cp Er,s (1) = Er,s (p).
It seems probable that the result (108) is true not only for r + s = 10 but also when r + s
is equal to 14, 16, 18, 20 or 24, and that
 
 Er,s (n)  1
(r+s+1)
 Er,s (1)  ≤ n 2 d(n) (109)
 

for all values of n, and

 
 Er,s (n)  1
(r+s+1)
 Er,s (1)  ≥ n 2 (110)
 

for an infinity of values of n. If this be so, then

1 1
Er,s (n) = O{n 2 (r+s+1+ǫ) }, Er,s (n) 6= o{n 2 (r+s+1) }. (111)

And it seems very likely that these equations hold generally, whenever r and s are positive
odd integers.

21. It is of some interest to see what confirmation of these conjectures can be found from
a study of the coefficients in the expansion of
∞
X
24/α 48/α 72/α a
x{(1 − x )(1 − x )(1 − x ) · · ·} = Ψα (n)xn ,
1
On certain arithmetical functions 199

where α is a divisor of 24. When α = 1 and α = 3 we know the actual value of Ψα (n). For
we have
∞
2 2 2 2 2 2
X
Ψ1 (n)xn = x1 − x5 − x7 + x11 + x13 − x17 − · · · , (112)
1

where 1, 5, 7, 11, . . . the natural odd numbers without the multiples of 3; and
∞
2 2 2 2
X
Ψ3 (n)xn = x1 − 3x3 + 5x5 − 7x7 + · · · (113)
1

The corresponding Dirichlet’s series are

∞
X Ψ1 (n) 1
= , (114)
ns (1 + 5−2s )(1 + 7−2s )(1 − 11−2s )(1 − 13−2s ) · · ·
1

where 5, 7, 11, 13, . . . are the primes greater than 3, those of the form 12n ± 5 having the
plus sign and those of the form 12n ± 1 the minus sign; and
∞
X Ψ3 (n) 1
= (115)
ns (1 + 31−2s )(1 − 51−2s )(1 + 71−2s )(1 + 111−2s ) · · ·
1

where 3, 5, 7, 11, . . . are the odd primes, those of the form 4n − 1 having the plus sign and
those of the form 4n + 1 the minus sign.
It is easy to see that
√
|Ψ1 (n)| ≤ 1, |Ψ3 (n)| ≤ n (116)

for all values of n, and

√
|Ψ1 (n)| = 1, |Ψ3 (n)| = n (117)

for an infinity of values of n.

The next simplest case is that in which α = 2. In this case it appears that
∞
X Ψ2 (n)
= Π1 Π2 , (118)
1
ns

where
1
Π1 = ,
(1 + 5−2s )(1 − 7−2s )(1 − 11−2s )(1 + 17−2s ) · · ·
5, 7, 11, . . . being the primes of the forms 12n − 1 and 12n ± 5, those of the form 12n + 5
having the plus sign and the rest the minus sign; and
1
Π2 = ,
(1 + 13−s )2 (1 − 37−s )2 (1 − 61−s )2 (1 + 73−s )2 · · ·
200 Paper 18

13, 37, 61, . . . being the primes of the form 12n + 1, those of the form m2 + (6n − 3)2
having the plus sign and those of the form m2 + (6n)2 the minus sign.
This is equivalent to the assertion that if

n = (5a5 · 7a7 · 11a11 · 17a17 · · ·)2 13a13 · 37a37 · 61a61 · 73a73 · · · ,

where ap is zero or a positive integer, then

Ψ2 (n) = (−1)a5 +a13 +a17 +a29 +a41 +··· (1 + a13 )(1 + a37 )(1 + a61 ) · · · , (119)

where 5, 13, 17, 29, . . . are the primes of the form 4n + 1, excluding those of the form
m2 + (6n)2 ; and that otherwise

Ψ2 (n) = 0. (120)

It follows that

|Ψ2 (n)| ≤ d(n) (121)

for all values of n, and

|Ψ2 (n)| ≥ 1 (122)

for an infinity of values of n. These results are easily proved to be actually true.

22. I have investigated also the cases in which α has one of the values 4, 6, 8 or 12. Thus
for example, when α = 6, I find
∞
X Ψ6 (n)
= Π1 Π2 , ∗ (123)
1
ns

where
1
Π1 = ,
(1 − 32−2s )(1 − 72−2s )(1 − 112−2s ) · · ·
3, 7, 11, . . . being the primes of the form 4n − 1; and
1
Π2 = ,
(1 − 2c5 · 5−s + 52−2s )(1 − 2c13 · 13−s + 132−2s ) · · ·

5, 13, 17, . . . being the primes of the form 4n + 1, and cp = u2 − (2v)2 , where u and v are the
unique pair of positive integers for which p = u2 + (2v)2 . This is equivalent to the assertion
that if
n = (3a3 · 7a7 · 11a11 · · ·)2 · 5a5 · 13a13 · 17a17 · · · ,
∗
Ψ6 (n) is Dr. Glaisher’s λ(n).
On certain arithmetical functions 201

then
Ψ6 (n) sin(1 + a5 )θ5 sin(1 + a13 )θ13 sin(1 + a17 )θ17
= · · ···, (124)
n sin θ5 sin θ13 sin θ17
where
u
tan 12 θp =
(0 < θp < π),
2v
and that otherwise Ψ6 (n) = 0. From these results it would follow that

|Ψ6 (n)| ≤ nd(n) (125)

for all values of n, and

|Ψ6 (n)| ≥ n (126)

for an infinity of values of n. What can actually be proved to be true is that

|Ψ6 (n) < 2nd(n)

for all values of n, and

|Ψ6 (n)| ≥ n
for an infinity of values of n.

23. In the case in which α = 4 I find that, if

n = (5a5 · 11a11 · 17a17 · · ·)2 · 7a7 · 13a13 · 19a19 · · · ,

where 5, 11, 17, . . . are the primes of the form 6m − 1 and 7, 13, 19, . . . are those of the
form 6m + 1, then
Ψ4 (n) sin(1 + a7 )θ7 sin(1 + a13 )θ13
√ = (−1)a5 +a11 +a17 +··· · ···, (127)
n sin θ7 sin θ13
where √
u 3
tan θp = (0 < θp < π),
1 ± 3v
and u and v are the unique pair of positive integers for which p = 3u2 + (1 ± 3v)2 ; and that
Ψ4 (n) = 0 for other values.
In the case in which α = 8 I find that, if

n = (2a2 · 5a5 · 11a11 · · ·)2 · 7a7 · 13a13 · 19a19 · · · ,

where 2, 5, 11, . . . are the primes of the form 3m − 1 and 7, 13, 19, . . . are those of the
form 6m + 1, then
Ψ8 (n) sin 3(1 + a7 )θ7 sin 3(1 + a13 )θ13
√ = (−1)a2 +a5 +a11 +··· · ···, (128)
n n sin 3θ7 sin 3θ13
202 Paper 18

where θp is the same as in (127); and that Ψ8 (n) = 0 for other values.
The case in which α = 12 will be considered in § 28.
In short, such evidence as I have been able to find, while not conclusive, points to the truth
of the results conjectured in § 18.

24. Analysis similar to that of the preceding sections may be applied to some interesting
arithmetical functions of a different kind. Let
∞
X
φs (q) = 1 + 2 rs (n)q n , (129)
1

where
φ(q) = 1 + 2q + 2q 4 + 2q 9 + · · · ,
so that rs (n) is the number of representations of n as the sum of s squares. Further let
∞
q3 q5
 
X
n q
δ2 (n)q = 2 − + − ···
1 − q 1 − q3 1 − q5
1

q2 q3
 
q
= 2 + + + ··· ; (130)
1 + q2 1 + q4 1 + q6

∞
1s−1 q 2s−1 q 2 3s−1 q 3
X  
s n
(2 − 1)Bs δ2s (n)q = s + + + ··· , (131)
1+q 1 − q2 1 + q3
1

when s is a multiple of 4;
∞
1s−1 q 2s−1 q 2 3s−1 q 3
X  
s n
(2 − 1)Bs δ2s (n)q = s + + + ··· , (132)
1
1−q 1 + q2 1 − q3

when s + 2 is a multiple of 4;
∞
1s−1 q 2s−1 q 2 3s−1 q 3
X  
n s
Es δ2s (n)q = 2 + + + ···
1 + q2 1 + q4 1 + q6
1
 s−1 s−1 3
5s−1 q 5

1 q3 q
+2 + − ··· , (133)
1 − q 1 − q3 1 − q5
when s − 1 is a multiple of 4;
∞
1s−1 q 2s−1 q 2 3s−1 q 3
X  
n s
Es δ2s (n)q = 2 + + + ···
1
1 + q2 1 + q4 1 + q6
 s−1
1 q 3s−1 q 3 5s−1 q 5

−2 − + − ··· , (134)
1−q 1 − q3 1 − q5
On certain arithmetical functions 203

when s + 1 is a multiple of 4. In these formulæ

B2 = 61 , B4 = 1
30 , B6 = 1
42 , B8 = 1
30 , B10 = 5
66 , . . .

are Bernoulli’s numbers, and

E1 = 1, E3 = 1, E5 = 5, E7 = 61, E9 = 1385, . . .

are Euler’s numbers. Then δ2s (n) is in all cases an arithmetical function depending on the
real divisors of n; thus, for example, when s + 2 is a multiple of 4, we have

(2s − 1)Bs δ2s (n) = s{σs−1 (n) − 2s σs−1 ( 14 n)}, (135)

where σs (x) should be considered as equal to zero if x is not an integer.

Now let

r2s (n) = δ2s (n) + e2s (n). (136)

Then I can prove (see § 26) that

e2s (n) = 0 (137)

if s = 1, 2, 3, 4 and that
1 2
e2s (n) = O(ns−1− 2 [ 3 s]+ǫ) (138)

for all positive integral values of s. But it is easy to see that, if s ≥ 3, then

Hns−1 < δ2s (n) < Kns−1 , (139)

where H and K are positive constants. It follows that

r2s (n) ∼ δ2s (n) (140)

for all positive integral values of s.

It appears probable, from the empirical results I obtain at the end of this paper, that
1
e2s (n) = O{n 2 (s−1)+ǫ } (141)

for all positive integral values of s; and that

1
e2s (n) 6= o{n 2 (s−1) } (142)

if s ≥ 5. But all that I can actually prove is that

1 2
e2s (n) = O(ns−1− 2 [ 3 s] ) (143)
204 Paper 18

if s ≥ 9 and that
1
e2s (n) 6= o(n 2 s−1 ) (144)

if s ≥ 5.

25. Let
∞
X ∞
X
f2s (q) = e2s (n)q n = {r2s (n) − δ2s (n)}q n . (145)
1 1

Then it can be shewn by the theory of elliptic functions that

X
f2s (q) = φ2s (q) Kn (kk′ )2n , (146)
1
1≤n≤ 4 (s−1)

that is to say that

f 4s (−q) X f 24n (q 2 )
f2s (q) = Kn , (147)
f 2s (q 2 ) 1
f 24n (−q)
1≤n≤ 4 (s−1)

where φ(q) and f (q) are the same as in § 13. We thus obtain the results contained in the
following table.

TABLE VI

1. f2 (q) = 0, f4 (q) = 0, f6 (q) = 0, f8 (q) = 0.

f 14 (q 2 )
2. 5f10 (q) = 16 f 4 (−q) , f12 (q) = 8f 12 (q 2 ).
3. 61f14 (q) = 728f 4 (−q)f 10 (q 2 ), 17f16 (q) = 256f 8 (−q)f 8 (q 2 ).
30 2
4. 1385f18 (q) = 24416f 12 (−q)f 6 (q 2 ) − 256 ff12 (−q)
(q )
.
28 2
5. 31f20 (q) = 616f 16 (−q)f 4 (q 2 ) − 128 ff 8 (−q)
(q )
.
26 2
6. 50521f22 (q) = 1103272f 20 (−q)f 2 (q 2 ) − 821888 ff 4 (−q)
(q )
.
7. 691f24 (q) = 16576f 24 (−q) − 32768f 24 (q 2 ).
It follows from the last formula of Table VI that
691
64 e24 (n) = (−1)n−1 259τ (n) − 512τ ( 12 n), (148)

where τ (n) is the same as in § 16, and τ (x) should be considered as equal to zero if x is
not an integer.
On certain arithmetical functions 205

Results equivalent to 1,2,3,4 of Table VI were given by Dr. Glaisher in the Quarterly
Journal, Vol. XXXVIII. The arithmetical functions called by him

χ4 (n), Ω(n), W (n), Θ(n), U (n)

are the coefficients of q n in

f 14 (q 2 ) 12 2 4
, f (q ), f (−q)f 10 (q 2 ), f 8 (q)f 8 (q 2 ), f 12 (−q)f 6 (q 2 ).
f 4 (−q)

He gave reduction formulæ for these functions and observed how the functions which I call
e10 (n), e12 (n) and e16 (n) can be defined by means of the complex divisors of n. It is very
likely that τ (n) is also capable of such a definition.

26. Now let us consider the order of e2s (n). It is easy to see from (147) that f2s (q) can be
expressed in the form
b  c
f 5 (−q) f 5 (q 2 )
X 
3 a
Ka,b,c,h,k {f (−q)} f h (−q)f k (q 2 ), (149)
f 2 (q 2 ) f 2 (−q)

where a, b, c, h, k are zero or positive integers, such that

a + b + c = [ 23 s], h + k = 2s − 3[ 32 s].

Proceeding as in § 13 we can easily shew that

1 2
n− 2 [ 3 s] e2s (n)

cannot be of higher order than the coefficient of q 24n in

φA (q)φB (q 3 )φC (q 2 ), (150)

where C is 0 or 1 and
A + B + C = 2s − 2[ 32 s].
Now, if s ≥ 5, A + B + C ≥ 4; and so A + B ≥ 3. Hence one at least of A and B is greater
than 1. But we know that X
φ2 (q) = O(ν ǫ )q ν .

It follows that the coefficient of q 24n in (150) is of order not exceeding

1
n 2 (A+B+C)−1+ǫ .

Thus
1 2
e2s (n) = O(ns−1− 2 [ 3 s]+ǫ ) (151)
206 Paper 18

for all positive integral values of s.

27. When s ≥ 9 we can obtain a slightly more precise result.

If s ≥ 16 we have A + B + C ≥ 12; and so A + B ≥ 11. Hence one at least of A and B is
greater than 5. But X
φ6 (q) = O(ν 2 )q ν .

It follows that the coefficient of q 24n in (150) is of order not exceeding

1
n 2 (A+B+C)−1 ,

or that
1 2
e2s (n) = O(ns−1− 2 [ 3 s] ), (152)

if s ≥ 16. We can easily shew that (152) is true when 9 ≤ s ≤ 16 considering all the cases
separately, using the identities.

f 12 (−q)f 6 (q 2 ) = {f 3 (−q)}4 , {f 3 (q 2 )}2 ,

 5 2 6
f 30 (q 2 ) f (q )
= ,
f 12 (−q) f 2 (−q)
 5 4  5 2 2
f (−q) f (q )
f 16 (−q)f 4 (q 2 ) = f 2 (q 2 ),
f 2 (q 2 ) f 2 (−q)
 5 2 4
f 28 (q 2 ) f (q )
8
= {f 3 (q 2 )}2 f 2 (q 2 ), · · · ,
f (−q) f 2 (−q)

and proceeding as in the previous two sections.

The argument of §§ 14-15 may also be applied to the function e2s (n). We find that
1
e2s (n) 6= o(n 2 s−1 ). (153)

I leave the proof to the reader.

28. There is reason to suppose that

1

e2s (n) = O{n 2 (s−1+ǫ) } 
1
, (154)
e2s (n) 6= o{n 2 (s−1) } 

if s ≥ 5. I find, for example, that

∞
X e10 (n) e10 (1)
= Π1 Π2 , (155)
ns 1 + 22−s
1
On certain arithmetical functions 207

where
1
Π1 = ,
(1 − 34−2s )(1 − 74−2s )(1 − 114−2s ) · · ·
3, 7, 11, . . . being the primes of the form 4n − 1, and
1
Π2 = ,
(1 − 2c5 · 5−s + 54−2s )(1 − 2c13 · 13−s + 134−2s ) · · ·
5, 13, 17, . . . being the primes of the form 4n + 1, and

cp = u2 − (4v)2 ,

where u and v are the unique pair of positive integers satisfying the equation

u2 + (4v)2 = p2 .

The equation (155) is equivalent to the assertion that, if

n = (3a3 · 7a7 · 11a11 · · ·)2 · 2a2 · 5a5 · 13a13 · · · ,

where ap is zero or a positive integer, then

e10 (n) sin 4(1 + a5 )θ5 sin 4(1 + a13 )θ13

2
= (−1)a2 · ···, (156)
n e10 (1) sin 4θ5 sin 4θ13
where
u
tan θp =(0 < θp < 12 π),
v
u and v being integers satisfying the equation u2 + v 2 = p; and e10 (n) = 0 otherwise. If
this is true then we should have
 
 e10 (n)  2
 e10 (1)  ≤ n d(n) (157)
 

for all values of n, and

 
 e10 (n)  2
 e10 (1)  ≥ n (158)
 

for an infinity of values of n. In this case we can prove that, if n is the square of a prime
of the form 4m − 1, then  
 e10 (n)  2
 e10 (1)  = n .
 

Similarly I find that

∞ Y 
X e12 (n) 1
= e12 (1) , (159)
1
ns p
1 + 2cp · p−s + p5−2s
208 Paper 18

p being an odd prime and c2p ≤ p5 . From this it would follow that
 
 e12 (n)  5
 e12 (1)  ≤ n 2 d(n) (160)
 

for all values of n, and

 
 e12 (n)  5
 e12 (1)  ≥ n 2 (161)
 

for an infinity of values of n.

Finally I find that
∞  
X e16 (n) e16 (1) Y 1
= , (162)
1
ns 1 + 23−s p 1 + 2cp · p−s + p7−2s

p being an odd prime and c2p ≤ p7 . From this it would follow that
 
 e16 (n)  7
 e16 (1)  ≤ n 2 d(n) (163)
 

for all values of n, and

 
 e16 (n)  7
 e16 (1)  ≥ n 2 (164)
 

for an infinity of values of n.

In the case in which 2s = 24 we have
691
64 e24 (n) = (−1)n−1 259τ (n) − 512τ ( 21 n).

I have already stated the reasons for supposing that

11
|τ (n)| ≤ n 2 d(n)

for all values of n, and

11
|τ (n)| ≥ n 2
for an infinity of values of n.