VOL.

39, 1953

MA THEMA TICS: C. A. NICOL

963

of contact. If every matrix XoA + ,uoB has a multiple characteristic root it follows that C has a component which has to be counted double. Detailed proofs and extensions to fields of finite characteristic will appear elsewhere.
I Motzkin, T. S., and Taussky, Olga, "Pairs of Matrices with Property L," Trans. Amer. Math. Soc., 73, 108-114, (1952).

ON RESTRICTED PARTITIONS AND A GENERALIZATION OF THE EULER p NUMBER AND THE MOEBIUS FUNCTION
BY CHARLES A. NICOL
UNIVERSITY OF TEXAS
Communicated By H. S. Vandiver, June 23, 1953

Introduction.-In the present paper we shall treat the function

n-I

(z, x)

s=1

III (z-x8)

(1)

mainly from an arithmetic standpoint. Functions of this type have been studied extensively in the theory of partitions of positive integers' where z is replaced by 1 or -1 and the range of the product is infinite. For example, a famous result due to Euler for lxi < 1 may be written as
co

(1 -

X) (1

X2) (1

X3)

+ n= E

-X

(_j)nX(1/i)n(3n+1) (2)

The coefficients of the series admit the following combinatorial interpretation. If E(n) denotes the number of partitions of n into an even number of unequal parts and U(n) the number of partitions of n into an odd number of unequal parts then (2) may be stated in the following way.

E(n) = U(n) except when n = 2 2 k(3k + 1), when E(n) - U(n) = (-1)k The case where we have only a finite number of terms in (2) has received comparatively little attention. If in (1) we let z = 1 we have
n-1

F,,- (1, x)

s=1

II (1 - Xe).

(3)

The coefficient of xk for k < n will be E(k) - U(k). But the coefficient of xk for k > n is the number of partitions of k into an even number of unequal parts, none of which is larger than n, minus the number of partitions of k

964

MA THEMA TICS: C. A. NICOL

PROC. N. A. S.

into an odd number of unequal parts, none of which is larger than n. This then is an extension of the analogous problem arising in the use of the infinite product. Also if z is replaced by -1 we consider the following function.
n-i

(-) n-lFn-_ (-I xX) = I(1 + x8).

s-i

(4)

The coefficient of x" resulting from the expansion of this product is the number of partitions of k as a sum of distinct positive integers none of which is larger than n. This may also be stated as the number of solutions of the equation x1 + 2x2 + ... + (n - l)x-i= k, where for i = 1, 2, ..., (n - 1), xi is either zero or unity. The products (3) and (4) have been studied by Cauchy, T. Vahlen, von Sterneck, and others.2 In particular von Sterneck studied the case where the polynomial resulting from the expansion is reduced modulo a positive integer. Fundamental in this investigation will be the use of the number3

c1(k, n)

(n) (p(n/(k, n))_A(n/(k, n)),

(5)

where k and n are positive integers and (k, n) denotes the greatest common divisor of k and n. If n is a positive integer so(n) denotes as usual the number of positive integers less than n and prime to it. (s(1) = 1.) Also, for n a positive integer, p(n) is zero if n contains a repeated prime factor. Otherwise u2(n) is equal to (-I)T where Y is the number of distinct prime factors of n. (,u(1)-1.) Note that (5) reduces to ,A(n) when (k, n) = 1 and p(n) when (k, n) n. Although (5) appears more complicated than its constituents it will be shown that many of the principal theorems concerning it are hardly more complex than those involving the (p or IA number alone. The properties of the coefficients in the development of (1) are extensively used to obtain properties of (5) and vice versa. We now state a number of theorems without proof. We hope to publish the proofs elsewhere. In the following paragraphs the symbol [x] will denote the largest integer contained in the real number x. Also the symbol $(a, b) will denote the number defined in (5) where a and b are positive integers. We have if k, r, and n are positive integers that4
(r,n) - 1

F , E

exp (2rirk/n) = c1(k, n),

(6)

where the range of r is over all positive integers less than n and prime to it. (j2 = -1.)

VOL. 39, 1953

MA THEMA TICS: C. A. NICOL

965

",We also have

THIEOREM 1.
din
d

O otherwise.
(0 if (k, n) $ = (-n if (k, n)
n.
=

Similarly

l)d D(k, n/d) >Z(_ din

(Oif (k,n)

n, n even. n,n odd.

(8)

We may also prove THEOREM 2. Let a.(k) denote the sum of the divisors of k less than or equal to n. Then,
n

s=1

[n/s]P(k, s)
n

oan(k).

(9)

In case k = 1 this becomes the well-known relation"

E [n/s]A(s) = 1.
Also if k is replaced by n! we have another known result6
n

E [n/s] sp(s) s=l

THEOREM 3. If 5 n, then

n(n + 1)/2.

(d/ )

-1

(k, b/d)

0 (mod. n).
a

(10)
is a positive

COROLLARY. If now p denotes an odd prime and integer, then

pa) (pa/1) (mod. n).
THEOREM 4.

(lOa)
(11)

,
dln

R,(d)cI(k, n/d) = 0 (mod. n),

Rs(d) =
(d 9 ) (_1)a

where

and this sum is over all integral solutions at of the equation . < n/d. whereO < s< n, 0 . a <d, 0 <

(nld) a

+, = s

966

MA THEMA TICS: C. A. NICOL

PROC. N. A. S.

Consider now the function F,n-(z, x) defined in (1). If n > 1 and the product is expanded as a polynomial in x, we may write
n

E P,(z)x8, FP-1(z, x) = s=O

(12)

where ni = n(n - 1)/2 and P.(z) is a polynomial in z. Then we may define the polynomial B,(z) as
Mg

Bj(z)

E = k=O

Pkn+t(Z),

(13)

where n, = [(n - 1)/2 - t/n] and 1 < t < n. Then we may obtain THEOREM 5. If z is a number different from unity, then

l)d-W (t, n/d), (14) B,(z) = n(z 1- 1) dln (zn/d where 1 < t < n and Bt(z) is defined in (13). Denote the polynomial defined by the function F.-1(1, x) in (3) by
s-O

E Aox, where nj

n(n

1)/2.

-(15)

Also define the number C, by the relation

Mc

C: =E Ak*+:, where n, k=O

Then we obtain the following THEOREM 6.

[(n

1)/2 -t/n].

(15a)

Ct
where C, is defined in (15a). THEOREM 7.

n(t, n),

(16)

io(n)

=- E C2,
n

in

t=l

(17)

where Cs is defined in (15a). In view of theorem 6 we may write theorem 7 as

sp(n) = - Ej 12 (t, n).
n

lin

t=i

(18)

An observation of possible interest may be made concerning theorem 7 if it is noted that C, (o(n). Then (17) becomes a quadratic relation in (p(n). Employing the quadratic formula we find

VOL. 39, 1953

MA THEMA TICS: C. A. NICOL

967

(p(n)
n-1

(n ' V/n2 -4G(n))/2,

(19)

where G(n) = E C. Except in the case when so(n) = n/2 only one of the roots of (19) corresponds to sp(n). The significance of the remaining root has not been determined and would seem to be of interest. If x is replaced by exp(iO), where i2 = -1, we obtain the following THEOREM 8.

4(t, n)

{Fn_- (exp(i6)) E exp(-(kn + t)iO)} dO, (20)

n-I

where Fn_i(exp(iO)) = I (1 - exp(siO)) andlnf = [(n - 1)/2 - t/n]. $=1 Similarly for the numbers Ag defined in (15) we have

At

Jr 1Fn-l (exp(iO)) (exp(-tiO))) dO.

n
s=O

(2w

(21)

Furthermore the numbers A, defined in (15) have the following properties:

jA 1A

> n,

(22)

where ni = n (n - 1)/2. A rather unusual property of these numbers is: THEOREM 9. If di (n - 1), then A = 0,
(s, n-1) =d

(23)

where 0 < s < n(n - 1)/2. A by-product of these investigations is the following result: If p is an odd prime, the integral roots of the congruence
p-1

1 +

s= 1

E cb(s,p

- 1)xSO0 (mod.

P)

(24)

are the incongruent primitive roots modulo p. If we consider the function defined in (4) we may obtain the following: If n is an odd positive integer and t is an integer such that 1 < t < n, then

Bt(-1)
where B, (z) is defined -in (13).

= - E 2d 2n din

4(t, nld),

(25)

968

9.68MA THEMA TICS: C. A. NICOL

PROC. N. A. S.

Since B,(z) is a polynomial with integral coefficients the number B,( is an integer. In particular, if n is odd and t is less than n and prime to it, thern

1)

Bt(-1)

-E 2d ,(n/d). 2ndl-n

(26)

This number is a generalization of the Fermat Quotient, (2P-1 - 1)/p, where p denotes an odd prime. Acknowledgment.-The author is indebted to H. S. Vandiver for his many helpful suggestions and encouragement. I Dickson, L. E., History of the Theory of Numbers, Vol. 2, chapter 3, Carnegie Inst. of
Washington, Publication No. 256 (1920). Bachmann, P., "Niedere Zahlentheorie," Zweiter Teil, chapters 3-6, B. G. Teubner, Leipzig, 1910. 2 Vahlen, T., Bachmann, P., Ibid., 116, 167, 273. Von Sterneck, Sitzungsber. d. Wiener Akad., 111, 1567 (1902); 113, 326 (1904); 114,711(1905). Cauchy, Oeuvres D'Augustin Cauchy, 5 (series 1), 81-85, 135-152. Paris, GauthierVillars (1885). 3 Von Sterneck (P. Bachmann, Ibid.) introduced a function equivalent to cb(k, n). He used it to obtain results concerning partitions modulo a positive integer. Employing this function he obtained a special case of theorem 1. The number '(k, n) was used by R. Moller in the following result (Math. Monthly, 59, No. 4, 228 (April 1952)). If the numbers gd are all of the incongruent integers belonging to d modulo p, p being an odd prime and d a divisor of p - 1, then for any r, E gd =(r, d) (mod. p) 4 The sum E exp(2xrirk/n) is known as Ramanujan's sum (cf., Hardy, G. H.
(r, n) = 1

and Wright, E. M., Introduction to the Number Theory, Oxford, 1938. pp. 55,237). Another closed form for this sum was found previously by T. M. Apostal and D. R. Anderson and stated by them in an abstract in Bull. Am. Math. Soc., 58, No. 5, 559 (1952). The form they found in our notation is cJ(b) 3(a)/op(c) where a = (n, k): b = n/a, and c = (a, b). If ,u(b) . Owe have the relation p(n)/,o(b) = c,(a)/,i(c). Nagell, T., Introduction to Number Theory, Uppsala, 1951, p. 43. Review of Perez-Cacho, "The Function E(x) in the Theory of Numbers," Math. Reviews, 18, 913, (1952).