AN ANALOGUE OF THE VON STAUDT-CLAUSEN THEOREM

BY LEONARD ARLITZ
1. Introduction. Let GF(p n) denote fixed Galois field, and x an indeter-
minate over the field. The function
(- ) pn
(a.) (t)
=0 F
where
(1.2) [i] x pn X, F,
"’’"
[i][i- 1] [1] "(-’), F0 1,
is closely connected with the arithmetic of polynomials the GF(p"). In this
in
paper we study the coefficients in the reciprocal of (1.1), more precisely in t/.
In particular we shall be interested in proving an analogue of the yon Staudt-
Clausen theorem for these coefficients.
In order to define properly the coefficients in the reciprocal it is necessury to
define a "normalizing" factor (analogous to n in ordinary arithmetic). This
is done in the following wy. Let

(1.3) g(m) ... F ,

be the canonical expansion of m to the bse p"; then we put
FF; g(O) 1,
where F has the same significance as in (1.2). Thus for example
g(pne) = F., g(pn,__ 1)= (F F_)"-.

(.4) ,
We may now define the coefficients of the reciprocal by means of

= B

the summation obviously containing only terms in which m is a multiple of

p" 1. Clearly B0 1 and B is a rtional function of x. The analogy
between B and the ordinary Bernoulli numbers is brought out by the relation

(pn l lm)

_
E g(m)
where the summation is over all primary polynomials E, and

lira
[](Z i] :: [1]
Reeeived July 21, 1937.
See this Journal, vol. (1935), pp. 137-168. This paper will be cited as DJ.
DJ, p. 161, Theorem 9.3.
5O3
504 LEONARD CARLITZ

In the present paper we discuss some of the arithmetic properties of Bin. Our
principal result is the analogue of the von Staudt-Clausen theorem for the Ber-
noulli numbers. We find that
1
Bm G e
deg P---k
(P’ # 2),
where Gm is some polynomial (whose precise form is not determined), e is an
integer, not divisible by p, and the summation is over all irreducible polynomials
P of degree k; finally/ is a number depending on m whose existence depends
on a certain set of conditions (see (7.3) below); if the conditions are not satisfied
Bm G and is therefore a polynomial. When p 2 the result must be
modified slightly.
The method of proof depends on certain ideas due to A. Hurwitz. While
the proof is not particularly difficult, there are a number of details that make it
rather long. In particular it is necessary to prove certain lemmas on g(m)
which are of some interest in themselves.

2. Lemmas on g(m).

__
THEOREM 1. For ml m2 >- O, the quotient
g(ml +
g(ml)g(ms)
is integral (that is, a polynomial).
Let

=< , < p’);

(2.2)
ms 5"0 + ")’IPn ’sp’ (0
then
(2.3) m + m. (8o + "to) + + (/ + ’)’)P" (0 =< + < 2p").
If now we put
0 + 5’0 o "{" (oP n,
where i0 0 or 1, and 0 _-< a0 < p, we may define ti, ti recursively by
means of
+1
(2.4)
0
ti
"+ s ++
1 "’1
"’s
0ll

as + .p,

Mathematische Annalen, vol. 51 (1899), pp. 196-226 (= Mathematische Werke II,

Basel, 1933, pp. 342-373).
Throughout this paper the word "integral" will be used to denote a polynomial in x
with coefficients in GF(p’O.
 ANALOGUE OF VON STAUDT-CLAUSEN THEOREM 505

where each i 0 or 1, and 0 _-< al < p. Thus (2.3) becomes

mi -}- m2 a0 + alp + + pn8 +p n(s+l)

Hence by the definition of g(m) in (1.3),

g(ml + m:) F
Comparison with (2.2) leds at once to

g(m,)g(m,)
Since by (1.2) F+ [s + 1]FY ", the right member
s--1

Is + 1]tsF 8pn+"-8-" II F7
s--1
[s + 1]"F ’-t l-I F] --’,
by the last of (2.4). Again

Proceeding in this way we have finally

(2.5) g(m
g(ml)g(m)
+ m.)
Is + 1]8[s] 8-’... [1] .
Thus we have not only proved that (2.1) is integral but have derived the explicit
formula (2.5).
As an immediate corollary it is evident that for ml m O, the quotient
(2.6) g(m +... T m.)
g(mO g(m)
is integral.
If in (2.6) we take m m m, wesee that g(km)/g(m) is integral.
For later purposes it will be necessary to know that this quotient is divisible
by g(k). But we may prove without difficulty the following slightly stronger
result.
THEOREM 2. For m >= O, define (m) by means of
m= ao+ a +"" (0 <-_ o < p’),
(2.7)
=ao+a+"- +a..
Thus for m >- 1,/ >- 1. Then the quotient
g(km)
g(m)g,(k)
i8 ntegral.
Compare Bachmann, Niedere Zahlentheorie.
506 LEONARD CARLITZ

--
It will be convenient to take first the special case m so that
1. We shall require the
LEMMA. For all i, j >= O, the quotient
(2.8) F+j [i + j] [i j 1]P’ [i
(i, j)
Fn; F. F.
is integral.
This is easily proved by induction. From (2.8) follows

(i, j)
[i pn
j] Fi+i-1 ([j] + -o ,,,, +_
Ff F
pn-1
(i, j 1) + \ Ff: ] (i 1,j).

Since (i, O) 1 (j, 0) it is evident from this recursion formula that (i, j)
is integral for arbitrary non-negative i, j.
Returning to the general theorem, we now suppose m arbitrary but to begin
with again take k pn. From (2.7) and the definition of g(m) it follows
easily that
g(pnm) [S q- 1]""[s] "8-! [11 ".

Replace m by p’m and this becomes

[s q- 21 ’"[s q- 1 [2]-0;
gP’(pnm)
combining the last two equations wc have
g(P2’m) (Is q-2][8 q- 1]’) "... ([2][1]’) ".

-
g(pm)
Continuing in this way we see that

(2.9) g(pnim)
I {[i + j]"" [i 1]Pn(j-1)} ai

If we compare this with (2.8), it is clear that the right member is u multiple of
F] +’’’+" F,
so that the theorem is proved for the special case k pn’.
In the next place, for _-> 1, it follows that

(2.10) F’ g(pnim)
g,n (m)
 _
ANALOGUE OF VON STAUDT-CLAUSEN THEOREM 507

But by Theorem 1, g(p’m) lg(pnJm), so that (2.10) implies

g(p’m)
(.]1) Fi gflpn] (m)
thus the theorem is proved for k p’, where 0 < < p’.
Now take
(.2) ] opni pni (0 < ,< p";i j).
Then by (2.11) we have
g(p’im) g(p’m)

but by (2.12) and (1.3)

._"’
g(apni fjpnl)
(m) ’ (m)
Fi"Fj,
and by Theorem 1,
g(opnim)g(P nim) g{ (aP ni + p’)m},
so that
g(km)

for k as in (2.12). Proceeding in this way we see that Theorem 2 is true

generally.

3. H-series. If in the series

(3.1) S A’tm
g(m)’ ,=o

the coefficients Am are integral, we shall cM1 (3.1) an H-series. It follows at

once from the definition that if S and S’ are H-series, then AS + A’S’ is also
an H-series, where A nd A’ are any two polynomials in x alone. As for the
product of two H-series, if Am, A: are the coefficients of S and S’ respectively,
and C,,, denotes the general coefficient in SS’, we evidently have
Cm g(m___) A,A.
i+=m g(i)g(j)
By Theorem 1, the g-quotients on the right are integral so that Cm is integral.

-
Timrefore the product of two H-series is itself un H-series, and generally for
the product of ny number of series.
Consider next the reciprocal of S. In general this is not an H-series. If
however A0 1--or any non-zero element of GF(p)--then the reciprocal is
also a,n H-series. Thus for A0 1, put S 1 S; then
1 1
1 -S-[-SI
Compare Hurwitz, loc. cit.
508 LEONARD CARLITZ

and it is clear from the above that the right member is an H-series. Finally it
follows that in this case SI/S is also an H-series.
Of special interest is the series
(.m ,s g(m)
m=l

which has no constant term. We have seen above that for the k-th power,

C is integral. We shall now

= =E o(m)’
show that C is a multiple of g(k), or what
amounts to the same thing, we prove
TnoM 3. If S is an H-series without constant term, then S/O(k) is also
an H-series.
Assume first k p". Then by (3.2),
phi
t""". (’).
E Ao(p"’m) O""’(m)
But by Theorem 2, the second fraction on the right is divisible by g(p") F.
If then we put
(.) ct"
E o(m)
we have
.,
F[C. Squaring both
c2r
o(m)’
sides of (3.3), we write

where C=
.+=
o(m)
o(e)o() CC.
Thus it is clear that F[C, in other words S "’/F is an H-series. Similarly
S /F In other words, S/g(k) is an H-series for k ap 0 < a < p
Suppose next that
(3.4) k ap ni + tip ", 0 < a, < p", i j.
Put
S pn E Ag(m)tm SlPn] lIra tm
E Ag(m)
so that
(3.5)
Then for
9(m) A,A.,.,,
+z: g(e-(f)
so that by (3.5)
 ANALOGUE OF VON STAUDT-CLAUSEN THEOREM 509

But by (3.4)
FTF g(apni + pnj) g(k),
so that the theorem is proved in this case also. It is now clear how the theorem
may be proved for general k.
As a corollary of some interest, we state: if an H-series without constant term
be substituted for in the H-series (3.1) and the result written as a series in t, then
this series is also an H-series.

4. Some theorems on Bin. If we define A(k) by means of

zo -1= ,=p.k-1 g(m)’

then we have the formula
Ak)
(4.1) B.-- pnkm--I nk
where
(4.2) L [k][k- 1]... [1], L0 1.
Now A () is integral and by Theorem 3 is a multiple of g(pn, 1). But by
(1.3) and (4.2),

(4.3)
g(p’- 1) (F F1) pn-1 &
Lk L_I [k]"
If we recall that [/] is the product of the irreducible polynomials whose degree
divides k, it is clear that if the left member of (4.3) be reduced to its lowest
terms, then the factors of the denominator are simple. Further, except for the
case p" 2 k, the irreducible factors are all of degree k. This proves the
following
THEOREM 4. If B N,/D,, where Nm and D are relatively prime, then
D, has only simple factors.
In the next place from the identity
(4.4) x(t) (xt)
follows

(4.5) xt
-t
x(t) b(xt) tP-(t) tv’-2(t) C,
(xt) (t) (xt)(t) (xt) x- v-(t) g(m) x"’
where, by the discussion of 3, the coefficients C, are clearly integral. Hence
by (1.4) and the last theorem, it follows that the product x(x 1)B is integral.
7DJ, p. 158, formula (8.11).
DJ, p. 150, formula (5.09).
510 LEONARD CARLITZ

But this result may be extended considerably. If G is an arbitrary poly-

nomial of degree s, say, then in place of (4.4) we have the more general formula

(4.6) G(t) (Gt) AV"J-l(t),

where the A are integral; indeed

Thus (4.5) becomes

Gt t__. Gk(t) (Gt) _,Ak"’
(Gt) (t) (Gt)(t) G a ibv"j-"
Now both numerator and denominator on the right are H-series, and it is easily
seen that if we replace by Gt the quotient becomes an H-series. Thus applying
Theorem 4, we have
TEOHEM 5. For G an arbitrary polynomial, the product G(G 1)Bin is
integral.
Assume now in the notation of Theorem 4 that P[D,, where P is an irre-
ducible polynomial of degree k. Since in the thec.rem just proved, G is quite
arbitrary we may take it equal to a primitive root (mod P). Now by the
theorem,
G -= 1 (mod P),
and therefore because of the nature of G, m must be a multiple of pn 1.
This proves
TIEOHEM 6. If P is an irreducible divisor of the denominator of B,,, then
nk
p 1 divides m.
If we return to the definition of A() and make use of (4.1) and (4.3), it is
now evident that for p" 1 not a divisor of m, A() is a multiple of P. There-
fore in determining the fractional part of Bm it is necessary to retain in the right
member of (4.1) only those terms for which pn
1 m. We may now state
the following
THEOREM 7. For P irreducible of degree k, we have the congruence

(4,7) pn,--I (2) (rood P)

"-1. g(m)
the summation extending over multiples of p’ 1 only. The formula (4.1)
reduces to
A)
(4.8)

DJ, p. 151, formula (5.11).

B G,
’
+ pnk--1 [m Lk

The statement
A,t"*
g(m) Z A’,.t" (mod P) is short for the infinite system of congru-
ences A,. A’ (mod P)
 ANALOGUE OF VON STAUDT-CLAUSEN THEOREM 511

where 11 Gm is integral, and the summation extends over such k for which p nk
divides m.

5. A lemma on Cpk_l. We now prove the following theorem, which is the

most important point in the proof of the main theorem concerning Bin.
THEOREM 8. For P irreducible of degree k,

(51) (mod P).

F
In other words, in forming the (pnk 1)-th power of (t) (mod P) we may
ignore those terms in p"i for which i is not a multiple of k. To prove (5.1),
we remark first that by Theorem 3,

Combining this with (4.7), we have the congruence

(k)
"* - 0 (mod P).

Z A
,"-1. g(u) --i =--- 0 (modP)
Picking out the coefficient of on the left, we get

(5.2) (-
We suppose hereafter that m 1 is a multiple of p" 1. But by Theorem 7,
()
A _,n 0 for p nk
lm--pni
that is, for i not multiple of k. Therefore (5.2) becomes
g(m)
(5.3) ,, (- 1) A (., O.
Fg(m- p")
Now if p"(m it is easily verified that the quotient of g(m) by g(m 1) 0
(rood P). Indeed if p""[ m, pn(s+l)m, the quotient L, as defined in (4.2).’
Hence for p" m, (5.3) becomes
(,) 1
(_ 1), o(m)
F ,o(m V
the summation extending over i > 0 only.
We next recall result proved elsewhere "lz
(5.5) A(),..... 0 for m > 1,
which we shall apply to (5.3) in order to show that A(2 0 (rood P) for
em
11 will generally denote a polynomial depending on the index m and not necessarily
the same in all formulas in which it occurs.
D J, p. 158, Theorem 8.3.
512 LEONARD CARLITZ

certain set of values of m. First we make a slight change in notation.

Assume now that

shows that A.(,
+
(5.4) replace m by m 1, so that
(5.6) pk 1 m,
(5.6) holds but in addition plm
is a sum of terms of the type
A (k).
but according to (5.5), this vanishes unless, for some i, m
that is, unless
(5.7) m pnki + (pnk
nk
pn

2). 1.
m

--
+ 2.
1.

1
Then clearly (5.4)

pnk
(i
pnk
In

> 0);
1,

In the same way if

p m+ 1, m-2, p
(so that pn> 2), then two applications of (5.4) lead to a linear homogeneous
expression for A(k) in terms of the type

-.
)
m_2_pnk i_pnk (i, j > 0),
which vanishes unless m -t- 2 pni pn" p 1, that is, unless
(5.8) m pn p’ + (p 3). 1 (p’ > 2).
Now in both (5.7) and (5.8) m is expressed as a sum of p’ 1 terms p".
We shall now show generally that unless
(5.9) m- pnki p, _.
r’- pnk --1,

A() 0 (mod P). For suppose

__ - _
(5.10)
P
nk
Im + i (i-- 1,... ,t),
nk

Apply (5.3) times nd A(,) is exhibited as a sum of terms

()
A, w=mt pnkQ p nkQ
Since by the second of (5.10) pnlw 1, follows from (5.5) that A( -= 0
unless w pk 1, that is, unless
m pnki _... pnkit (pnk__ t-- 1)" 1,
which is precisely the condition (5.9).
It is now easy to establish the congruence (5.1).
it is clear that u term involving occurs when
If we expand
"- directly,

(5.11) m pni + pni r=p nk --1.

But by the result just proved the sum of such terms for fixed m will be 0
(mod P) unless m is of the form (5.9). But from this it follows almost imme-
 ANALOGUE OF YON STAUDT-CLAUSEN THEOREM 513

diately that each j in (5.11) is a multiple of k. Hence in forming it is

necessary to use only the terms in _pni but as remarked at the beginning of this
n
section, this is equivalent to (5.1). We have therefore established Theorem 8.
As we shall see in the next section it is now possible to evaluate ). In A
proving the von StaudtoClausen theorem for the ordinary Bernoulli numbers,
it is possible to make use of certain explicit formulas. Their analogues are not
available, and therefore some such method as used here seems necessary.
6. Further lemmas on r-1. We now consider

Rewrite (5.7) in the form

(6.2)
m a0 + ap + + ap (a 0),
p 1 0WW W a.
Then it is clear that (6.1) becomes
k(al+2a2+.
(pnk

the summation extending over all m for which (6.2) is solvable. Making use
of (5.1) we see that
g
(6.3) A (-1) ’+’’’+
0 F’ F’
where again m is of the form (6.2); for other m, A 0. All the congruences
are (mod P), where as above P is irreducible of degree k. To determine when
the multinomial coefficient in the right member of (6.3) is different from zero
(that is, not a multiple of p) we use a theorem of Dickson’s :
U the coecients
a in (6.2) are of the form
nk--1
(6.4) , ,iP (0 -< ,i <= P- 1),
then
(pn 1)!

is prime to p if and only if

(6.5) . p- 1 (j 0,..-,nk-- 1).
x Annals of Mathematics, (1), vol. 11 (1896-97), pp. 75-76; Quarterly Journal of Mathe-
matics, vol. 33 (1902), pp. 378-34.
514 LEONARD CARLITZ

If (6.5) is satisfied, then

(-- 1) ’
(6.6) (a0, "’, (rood p).

We may therefore assume that m satisfies both (6.2) and (6.5).

It is now easy to evaluate the g-quotient in (6.3). Put
(6.7) m fl0 + flip + flp + (0 =< fl, < p").
Comparison with (6.2) gives

(6.8) ai
k--1

i----0
k,+iP ni (i 0, ..., s).
Comparison with (6.4) gives
n--1
(6.9) Bk+i ’,0
Oi,n]+epe.

Now by (6.7), g(m) F 1F from which follows by use of (6.8),

g(m)
(Fl.. F],)(Fk+, F2__1)
F 1... F":
_= F,+k+ +’’" F+a++.......
But by (6.9) and (6.4)
n--1

Z ki--b] Z Z Oli,n]TePeO

so that the g-quotient

(6.10) (F F_) ’- (-1) -,
by Wilson’s Theorem and the fact that F is the product of the primary poly-
nomials of degree k. Therefore by (6.3) and (6.6) we conclude that for m
satisfying both (6.2) and (6.5), we have
(- 1)’
(6.11) A.( (- 1) -+("’+’’’+"’) (mod P);
H ai!
in all other cases, A () O.
Let m be fixed; we shall now show that at most one value of k can be found
for which (6.2) and (6.5) are simultaneously satisfied. For assume the relations
m "o .9[_ "1 pnl ._ "Yt penl

p nl 1 ’o ,
Bulletin of the American Mathematical Society, vol. 38 (1932), pp. 736-744; also DJ,
p. 140.
(6.12)
ANALOGUE OF VON STAUDT-CLAUSEN THEOREM

p--1
’
Then from (6.2) and (6.5) follows
nl--1

’0

,].
iO
5’’p (0 =< ’,
_ p
515

1),

nk--1
(p-- 1)nk;
on the other hand fom (6.12) follows
nl--1
(p- 1)nl.
io o
But clearly the a and coincide (except for numbering), and therefore
k l, as asserted above.
We remark, that there may be no value of k for which (6.2) and (6.5) hold.
In this event B has no fractional part.
7. The main theorem. We return to (4.1). In view of the last result in 6,
(4.1) becomes
1
(7.1) B., G.
+-- Ak)
provided ]c exists satisfying both (6.2) and (6.5); otherwise (7.1) is simply
(pn 2)

Bm= Gin, so that B is integral and nothing further need be said. Assuming
then that a/ exists, we make use of (4.3). If we exclude for the moment the
case pn 2, it foIlows that the irreducible divisors of the denominator of B
are all of degree k. Now it is easy to show (see the remark immediately following
(4.3)) that
[] E P’
,
where the summation extends over all irreducible P of degree a divisor of k,
and P’ denotes the derivative of P. By (4.3) we have
A (k) A(2 P
(7.2) G L_I dog P= --,
L P
the summation now extending over irreducibles of degree k only. But 15
(- 1)-ILk_ P’ (mod P),
so that

1
(7.2) becomes
A (2)
L
DJ, p. 166, formulu (11.10).
e (-1)-1A(,) .
1
 LEONARD CARLITZ

Finally, making use of (6.11) and substituting in (7.1), we have

TEOREM 9 (p" 2). For given m, the system

(7.3)

is either (i) inconsistent, or (ii) consistent for a single value of k, in which case the
as, ai are uniquely determined. In case (i) B, is integral; in case (ii) we have

(7.4)

Here G, is integral and the summation is over all irreducible P of degree k.

It remains to consider the excluded case p" 2. We may no longer conclude
from (4.3) that the denominator of B contains only irreducibles of degree k;
polynomials of the first degree may also occur. To decide when this happens
we examine

’= +:+-ff+... t’+F-++""
Clearly A( is different from zero only when m is of the form 2"
)
+2 > 2.
Since in this case
IF, Fo for a#,
g(m)
[F,+ for a #,
it,is evident that for a #,

while for a #,
1
A) g(m) [a + 1][a].
Thus for a #, it is clear that A is s multiple of L.
For a #, there are several possibilities. If a > # > 0, then A() is divisible
by (x 4- x)*; but if a > # 0, then A( is divisible by x 4- x only and the
quotient An/(x 4- x) is congruent to I (rood x 4- x). Again for a # (mod 2),
A divisible by 4- 4- 1, while for a yd fl (rood 2), A( 1 (mod
is x x
x -F x 4- 1). We now note that if the system (7.3) is satisfied for p" 2 k,
then it follows at once that ra 4 4- 2.4;; in other words, this is the case
a # (rood 2). Also it is easily seen that any other value of k is inconsistent
 ANALOGUE OF YON STAUDT-CLAUSEN THEOREM 517

with m 2" -F 2 (c > ). Hence we have the following supplement to

Theorem 9"
ToM 10 (p 2). If the system (7.3) is consistent for ] 2, then

_- ;
(7.5)
deg P==k

if k 2, then for m even,

(7.6) B, G. -9 x2 x
-+. I
while for m odd,

(7.7) B G 1
+ x4+x G 1+
1_
+ x x+l 1
x2+x+l
If (7.3) is consistent for no value of lc, then for m even, B, G, while for m odd,
B,=G,nulq 1
x x+l"
The following remark may be useful in testing (7.3). We assume m fixed;
pick a k such that pnk 1 Im. Then (because of the condition 0 < a < pnk)
the a are uniquely determined. If their sum is not p
1, we go no further.
If however the sum is equal to p’ 1, we use the equation a a.p" to
determine the a. (because of the condition 0 < a. < p, the determination is
unique). It is then only necessary to check the system of equations
p-- 1 a..
A partial check on Theorems 9 and 10 is furnished by the case m p 1,
for here a simple explicit formula is available for B=

(7.8)

-
L L
For p" 2 k, this reduces to
1 1 1 1
Ba
[2] x x -b 1 x: -b x -F 1’
which agrees with (7.7). In all other cases the irreducible divisors of the
denominator of B,, are of degree/c, and it is evident that (7.8) is in agreement
with (7.5). For this value of m, it is clear that s 0, a0 p’ 1, a p 1,
’)
A(,n g(m).
DUKE UNIVERSITY.
DJ, p. 159, formula (9.02).