Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
g-BERNOULLI AND ETJLERIAN NUMBERSBY
L. CARLITZ
1. Introduction. In a previous paper [2] the writer defined a set of
rational functions rjm of the indeterminate q by means of
(1.1) (qr,+ l)m = r,m(m> I), m - 1, m = 0,
and a set of polynomials
Vm(x) = Vm(x, q)
in g* by
(1.2) ym(x) = ([*] 4- qxv)m, Vm(0) = tj«,,
where [x] = (qx—l)/(q—l); also
(1.3) qxßm(x) = 7jm(x) 4- (q - l)vm+i(x), ßm(0) = ßm-
For 2=1, |3m reduces to the Bernoulli number Bm, ßm(x) reduces to the
Bernoulli polynomial Bm(x) ; nm however does not remain finite for m> 1.
In the present paper we first define polynomials Ama=Ame(q) by means of
r n ™ Fx + s - 11(1.4) [*]"=E^~ (m^l),
«_i L m J
where
l-x-l (g*- l)(g-*- !)•■• (g»-«H-i- 1)
lm]~ (q - l)(q2 - I) ■ ■ ■ (qm - I)
Alternatively if we define the rational function Hm = Hm(x, q) by means of
H0=l,H1 = l/(x-q),
(1.5) (qH+ 1)™ = xHm (m>l),
then we have
m
(1.6) Hm(x,q)=Am(x,q)/YL(x-q'),»=i
where
m
(1.7) Am(x, q) = Xi™!1-1 (m ^ 1),»-i
Presented to the Society, April 25, 1953; received by the editors February 5, 1953.
332
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
ç-BERNOULLI AND EULERIAN NUMBERS 333
and the coefficients are the same as those occurring in (1.4). For g=l, Ama
and Hm(x) reduce to well known functions; some of the properties of these
quantities are stated in §2 below. As Frobenius [3] showed, many of the prop-
erties of the Bernoulli and related numbers can be derived from properties
of Hm. We shall show that much the same is true in the case of the q analogues.
In [2] a theorem somewhat analogous to the Staudt-Clausen theorem
was obtained for ßm (with q an indeterminate). We now show that if p is an
odd prime and we put q = a, where the rational number a is integral (mod p),
then if a = l (mod p),
(1.8) pßm = - 1 (mod p)
provided p—l\m; otherwise ßm is integral (mod p). If a jé 1 (mod p) the situa-
tion is more complicated. In particular, if a is a primitive root (mod p2), then
ßm is integral (mod p) for p — l\m, while for p — \\m we have
1pßm = - — (mod p), (k = (a"'1 - l)/p).
k
In general the denominator of ßm may be divisible by arbitrarily high
powers of p (see Theorem 4 below).
Finally we derive some congruences of Rummer's type for Hm, etc. For
example if q = a is integral (mod p) while x is an indeterminate, then
Hm(Hw - l)r = 0 (mod pm, pre) (p'-Kp - 1) | w),
where after expansion of the left member IP is replaced by Hk. We also ob-
tain simple congruences for the numbers Ams defined in (1.4). The correspond-
ing results for r\m and ßm are more complicated.
2. Eulerian numbers. To facilitate comparison we quote the following
formulas from the papers of Frobenius [3] and Worpitzky [5].
m /% + s _ 1\(2.1) xm = J^AmA ) (m ^ 1),
«=i \ m /
(2.2) Am+i,s = (m + 2 — s)Am,,-! + sAm„
' (m + l\(2.3) Ame=Z(-l)r{ r )(s-r)»,
1 ^ / m V1(2.4) Fm= -_2(-l)—»( J Amr.
m + 1 r_i V - 1/
In the next place if we put
m
(2.5) Am = Am(x) = ^A^x'-1,Ê-i
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
334 L. CARLITZ [March
and let
Hm = Hm(x) = (x - l)-mFm(x),
then Hm satisfies
(2.6) (H +l)m = xHm (m^l), H0 = 1.
The connection between Hm and the Bernoulli numbers is furnished by
k-l Jfx kl-mç
(2.7) Z rrBm(-) =-mHm_i(ï),r=0 \ k / 1 - f
where f*—1, f^l. An immediate consequence of (2.7) is
(2.8) k~Bm(—\ - Bm= -«E -#«-i(f),\ a / o* r — i
where f runs through the Ath roots of unity distinct from 1.
We also mention
m
(2.9) Hm=YJ(x- 1)-'A*0».r=0
3. Some preliminaries. We shall use the notation of [2]; see in particular
§2 of that paper. In addition the following remarks will be useful. Let/(w) be
a polynomial in qu of degree^m. Then the difference equation
(3.1) g(u + 1) - cg(u) = f(u) (c^qr)
has a unique polynomial solution g(u), as can easily be proved by comparison
of coefficients. To put the solution in more useful form we rewrite (3.1) as
(E — c)g(u) =/(«) and recall that
A = F - 1, A2 = (F - 1)(F - q), A* = (F - l)(E - q)(E - q2),
In the identity
1 1 2 — 2i 1 (z — Zi)(z — Zt) 1+-+ -7-r,-:-+ ■ ■ ■
t — Z I — Z\ t — Zt I — z2 (/ — Zi)(/ — z2) t — z3
(z — Zl) • • • (z - z„) 1
(/ — Zi) •■■(/ — z„) i — z
take t = c, z = E, zs = q3~1, so that we get
1 1 A A2
+ Z-7-,-T + 7-IT;-r,-- +C-l ÍC- l)ic -q) ic- l)ic - q)ic - q2)
A" 1+
(c - 1) • • • (c - g"-1) c - E
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] Î-BERNOULLI AND EULERIAN NUMBERS 335
Hence if we take n>m, we obtain the following formula for g(u):
" As/(«)
(3.2) giu) = Z-AJ—/- •s=o (c - l)(c - q) ■ ■ ■ (c - q*)
4. The number Ama. We suppose A^ defined by means of (1.4). Using
the identity
(?m+l _ l)(?z - 1) = iqm+ls _ 1)(ç*+» - 1) + qm+ls(qs _ 1)(çz+8-m-l _ ^
and multiplying both members of (1.4) by [x], we get
[*]-«- T,A„\[m + i-s]\ l\] + im+1-°U]\x+l~ HI1 Lm 4- 1J L m + 1 J)
= H\X + SI~ Hik + 2 - s]Am,B^ + q^->[s]Am,},,L«+1 J
which implies the recursion
(4.1) Am+i,a = [m+2 - sp4m,s_i 4- qm+1-°[s]Ama.
For g=l it is evident that (4.1) reduces to (2.2). As an immediate conse-
quence of (4.1) we infer that Am, is a polynomial in q with positive integral
coefficients.
It is easy to show that Am, is divisible by g(>»-«)("»-<+»/ä. Indeed if we put
(4.2) Ama = g(—.h-^-d/i^
then (4.1) becomes
(4.3) ^*+i,s = [m+2 - s]A*m.s-i + [s]aZs,
which proves the stated property. Moreover it follows easily from (4.3) that
(4.4) deg Ama = (s — l)(m - s).
Indeed assuming the truth of (4.4), we get
deg ([« -f 2 - sp4*.._i) = (m+l- s) + (s- 2)(m + 1 - s)
= is- l)(m +1- s),
■■ deg i[s]A*ms) - (j - 1) 4- (s - 1)(« - s) = is- l)(m + 1 - s),
so that
deg^4m+i,t = (s - l)(m + I — s),
which proves (4.4).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
336 L. CARLITZ
The symmetry properties
[March
(4.5)
and
(4.6)
A* - A*
A*Uq) = g<-l><—"iOg-1)
will be proved below.
Comparing coefficients of qmx on both sides of (1.4) we get
(4.7) Y,Ama = [«]! = [tn][m- l] • • • [l].
More generally if we expand both sides in powers of qx and equate coefficients
we get
(4.8) ¿¿mrg" = (W jgm'-,(«-1)/2[»»]! (Oáíá**).
The following table of .4^, l|s|má5, is easily computed by means of
(4.3).
12(5+1)
3g2+5g+3
4g3+9g2+9g+4
13g2+5g-|-3
6g4 + 16g3-f22g2+16g+6
14g3+9g2+9g+4
5. A formula for A^- It is easy to show that if fix) is a polynomial in
qx of degree ^m,
(5.1)
then
(5.2)
/(*),_o L w J
(*» ̂ 1).
cm0 = (_i)«g«(-n)/«/xo),
' r«4 n(5.3) Cm,m_r = £ (-1)' \f(r + 1 - j)g'C-»/*.
.-o L s J
Since
^ Ym + 1"1^(_l)T(.-i)/2 /(* + „+!_,) =o,—o L 5 J
we have in particular
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 337
Ï+1 Vm + n^(_1)T(S-i)/2 /(r+i-*)-o,a«o L î J
and (5.3) yields
(5.4) Cmr = ¿ (-1)—•«<•—>.<-•*->'» P " l~\sis - r),s-o L s A
which includes (5.2) also. Thus the coefficients in (5.1) are determined.
If we take/(x) = [x]m, then Cmr = Amr and we get after a little manipula-
tion
* A Ym + 11 r .(5.5) ¿C = g'(r-l>/t£ (_l)V(«-D/2 [r _ s]m.
s=0 L 5 J
for g=l, (5.5) reduces to (2.3).
Replacing q by q~l, (5.5) becomes
* Ä I"»» 4- l~l, ,ç(<-D('»-'-U„r(g-l) = ?(r-l)m+r(r-l)/2^ ( _ l).?.(.-l>/2+« [r _ 5]m>
«=0 L S J
where
j(j - 1) - (m + l)(m 4-2) (m + I - s)(m + 2 - s)e =-h ■-—
2 2
s(í 4- 1)4-(r — s + l)m = — (r — l)m.
Hence
Atr(q) = q^inm-r)Alr(q-i),
which is identical with (4.6).
In the next place we observe that exactly as in the proof of (6.2) of [2] we
have
Z ( ) [x]i+Ym~i)x-r- (qim+1)x - 1)
i-o \ i / i + l m + 1
™ rx 4- s - n= 2>-.g-H ^ , •«-i L m + 1 J
Divide both sides of this identity by [x] and then put x = 0. We find that
1 ■ • r m -y1fi»m-x——TE(-i)m-s-1?-(",-s)(m-s+1>/2 ! Ams.
[m+ 1J ,_i ]_s - 1J
Using (4.2) and (4.5) this becomes
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
338 L. CARLITZ [March
l » r m l"1
ß~ = r , ,, Z(-i)--1 . 4[w 4- l J «=i U — 1J
l ™ rw-]-1 *= T—7TE(-1)S ¿L,
[m + l\ a=i L s J
*
the first of which may be compared with (2.4).
6. The polynomial Am(x). The polynomial Am(x) = Am(x, q) is defined in
(1.7) formel; weput^40(x) = l. Put
(6.1) (t>m(x) = II (x - qs)s=0
and apply the Lagrange interpolation formula at the points x = q\
5 = 0, 1, • • • , m. Since
s—1 m
*'(?•) = ![(?•- îo n (?• - «o¿=.0 J=«+l
= (_l)"-«?»«-«(.-i)/2(c _ 1)»[j]![»í - s]!,
we get using (4.8)
<Pm(x)
(q - 1)m Zó " \ s / x — g*
As a first application of (6.2) consider
x-m-1g-m('n+1)/20m(x) » /m\ xgs
tf>„,(x) ™ /w\ 1(6.2) ¿m(*) = / 7 Z(-i)m->( )—
(g - l)m a=o \s / x -
m_i m(m+l)/2^ / ^ „ - v
ii^r-sr1)--b—^-^E (-D—( )■g-»(g-l)» ,_o \5/
»g-™(g - l)m i=0 \ 5 / X - g«
which gives
(6.3) x"-1g",(",_1,/24m(x-1, g-1) = AJix, q) im ^ 1).
Substituting from (1.7) in (6.3) we get
m m
qm(.m-l)liYáAm,(q-1)xm-' = Z^m.(î)*'_1.
«=1 «=1
which implies
(6.4) !"'"11"^(r1) = Am.^+iiq).
Hence by (4.2) and (4.6), (6.4) becomes
çm(m-l)/2-(»,-S)(m-S+l)/2-(,-l)(m-S)J4*s(ç-i;) = q*(,-l)l2A*m,m_e+iiq),
which is the same as (4.5).
7. The functions Hm(x) and Hm(u, x). Using (6.2) and (1.6) we get
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 339
(7.1) (g - i)-#»(*) = (*-i)E(-i)"*-( )-•=o \ s / x — qs
We remark that (7.1) implies
00
(7.2) HJix) = (x - 1)Z x-'-^[r}mr=0
for |x| >|g*|, O^s^m. It is also evident that
m /m \ r / r \ 1
(l + gff)-= (x- 1)Z( )?'(3- l)"rZ(- D"( )-r-o \ r / s=o \ s / x — q3
™/w\ 1 JE, /m- s\= (*-i)E( )-Z(-i)-8( )g'(g- i)-rs_o \ 5 / x - g* r=s \r — s /
™/m\ 1 g* / -1 \"-'
" (*~ ̂ eíVí/x-g« (g-i)Ag-1/JE, fm\ qs
-(«-«—(*-l) E (-D — 1 )—!L<-0 \ 5 / X — g-
which implies
(7.3) (1 4-gF)1» = xF™ («w > 1).
We have therefore proved (1.5). Alternatively taking (7.3) as definition of
Hm one can work back to the earlier formulas obtained for Ama above.
For some purposes it is convenient to define Hmiu; x)=Hmiu; x, q), a
polynomial in qu. We put
JE, /m\ qeu(7.4) (g- l)"Fm(M;x) = (x- l)Z(-Dm-8( )—-.
.-o \ 5 / x — q3
so that Hmi0; x) —Hmix). It follows at once from (7.4) that
(7.5) ffm(l - u; x-\ g-1) = (-g)-Fm(M; x, g)
and that
(7.6) xHmiu; x) - Fm(w 4- 1; x) = (x - l)[u]m.
We have also
JE, /m\Z( \q'Hriu;x) = Hmiu + 1; x),
which becomes, using (7.6),
(7.7) (1 4- qHiu; *))- = xFm(«; x) - (x - !)[«]•
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
340 L. CARLITZ [March
For m = 0, (7.7) reduces to (7.3).
Clearly (7.6) implies
fc-i
(7.8) Z **"'[« + *]" = xkHmiu; x) - Hmiu + k; x),t-0
which includes (7.2) as a special case.
Since Hmiu; x) is a polynomial in qu of degree m, the remarks in §3 apply
to the difference equation (7.6). In particular, application of (3.2) leads to
(7.9)A A'[u\m ( • \
Hmiu; x) = Z -ry— (*.(*) = II (* - ?r) ),8=o yp,(x) \ r_i /
provided X7iqT, r = 0, l,y • • , »t. To simplify the right member of (7.9), we
used (2.6) and (3.1) of [2]; thus
A'[u]m = Z gr(r-1)/2<wH.Mr-ig'("_r+<'>
and (7.9) becomes after a little manipulation
m r Q$(u~ri-s)
(7.10) Hm(u; x) = Z <T(r-1)/2a»,.rZ , , N ['].[«]«•r=0 a~0 Ya(x)
If we let Gr(u) denote the inner sum it is clear from (3.2) that
xGr(u) — GT(U + 1) = [u\r.
In (7.10) put w = 0, then
flm.r[r]!(7.11) Fm(x) = Z ?r(r_1)/2
r=0 *r(*)
which for q= 1 reduces to (2.9).
Using (7.1) and (7.4) it is easy to verify the formula
m /m \(7.12) Hm(u; x) = ZÍ )?rttFr[«]—' = (guF + [«])«.
Next using (7.12), (7.11), and the explicit formula [2, (6.2)] for am,, we get
5 1 r r r 1(7.13) Fm(w;x) = Z -7VT E (-l)T(s_1>/2 [« + f - * 1",
r-0 IM«) «-0 L 5 J
which is useful later.
8. Connection with i)miu). Using the formula [2, (4.7)]
» /m\ s
(g - i) «^.(ii) = Z ( -1) --• ( ) -ft <r,
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 341
we find that
k-l
(gi-i)"-1Zr^(« + T;gM=Z(-l)m-s( )„ . , .r=o \ * / s=o \ 5 / rv -1
m-1 fm _ J\ ¡-„ktu+ku-l
^It-l)'-1"' J—-s=o \ 5 / q' - f g 1
where
r* -1, f^i.
Comparing with (7.4) we have therefore
(8.1) *]—1Er,u-(« + —» ?*) = -rJ-#m-i(¿«; fg-1),r=o \ * / i - r
and in particular for u = 0
(8.2) i-'Eri-T- ?*) = —^-F.-iifg"1),r-0 V k f 1 - f
which may be compared with (2.7).
Next using the multiplication formula (see [2, (4.12)]; note that a term
is missing in that formula)
*-1 / r \ k — [k](8.3) [ä]-1EM« + —> ?*) = n-(*«, ?) + (-!)" 7—7T-
_o \ . * / (g - l)m
together with (8.1) we get
¿[¿]m_V.(w4- y> gM - 7?m(Ä«, g) - (-1)'(g- l)"1
= —Z -Hm.i(ku;irl),q t*i i - f
and in particular for m = 0,
*- [k¡k[k]~-h,m(i-, A-vm-i-iy
(q- 1)»(8.5) W - - ■ -
= -^7—F^-iG-g-1).
By means of (1.3) it is easy to write down formulas like (8.1), • ■ • , (8.5) in-
volving ßm.
9. Multiplication formulas. For the polynomial Hm(u ; x) we have, using
(7.4),
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
342 L. CARLITZ [March
(g* - 1)- Z rrqrtSm(u + ■?-; fg-*', g*)r-0 \ k /
m /~- \ Qkau k—1
-<rr-'-i>£<->)"(j)ÎF^£r'î
™ (m\Hf-î^K-i)™-' )-
.=0 \ Í / f
kau k—1
fr*' - qks Í* "
Qkau J _ >—knk(a+t)
gk($+l) l _ !—l„s+i
Consequently if f*-1 = l, f^l, we get
í¿felmY! c-ror'Hmlu + — : ira-*". ok\ = -
(9.1)[¿]mZrrgr,Fm(W— ; fr*'. <?*) =-9-—Hm(ku; tq-', g),
r=0 \ * / f - q'
analogous to (8.3).
In the special case x= — g-1, the polynomial em(u) of [2, §8] satisfies
(9.2) e„(«) = Hm(u; -q~\ q);
in this case (9.1) becomes (f = — 1, t = l)
(9.3) [i]-¿ (-5)^^ + ^, çA = l±iim(^>ç)r-0 \ e ./ g 4-1
for & odd; note that (8.6) of [2] requires a slight correction.
10. Staudt-Clausen theorems for ßm. In [2] a theorem analogous to the
Staudt-Clausen theorem was proved for ßm with q indeterminate. Now on the
other hand we replace g by a rational number a which is assumed to be inte-
gral modulo a fixed prime p. We shall use the representation [2, (6.2)]
m
(lo.i) 0- = E(-i)'««..k]i/[i+i].8=0
where
0-«(«-n/2 < r 5 "1(10.2) <*„,, = --—-Z(-l)'g'(r-1>/2 [*-»-]-;
si r_o LrJ
the quantity am_, is a polynomial in g and has occurred in (7.10) and (7.11)
above.
Suppose first that a = l (mod p). Then from (10.1) or [2, §7] it is clear
that the 5th term in the right member of (10.1) is of the form u,
= N„(a)/F,+i(a), where F,+i(x) is the cyclotomic polynomial and N,(x) is a
polynomial with integral coefficients. If we recall that F*(l) =p when k=pe,
eèl, but Ffc(l) = l otherwise, it is clear that u„ is integral (mod p) except
possibly when s + l=pe; the same holds also for Fk(a). Now let 54-1 —p'.
Then by a simple computation it is seen that [s]\ is divisible by exactly p/,
where
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 343
f=(p<- l)/(p -l)-e,
while the denominator is divisible by exactly pe. Since (pe—l)/(p—l)^2e
for e^2, p^3, it follows that ua is integral in this case. If e=l, p~^3, we have
first p(a—l)/(ap — l) = l (mod p). As for the numerator of mp_i, it follows
readily from (10.2) and
[7] "CD <-")p-1 / P — l\
IP- l]!am,P_i^Z(-l)r( )r»r=0 \ r /
ÇÎ 7-1 (mod/-) (/>-l|m),
r»o I 0 (mod #) (¿> - l\ m).
that
We have therefore proved
Theorem 1. Letp^3, q = a = l (mod p). Then
7-1 (mod^) (P- l|«),(10.3) */3m = <
I 0 (mod #) ip - l\ m).
For p = 2, the preceding argument shows that all terms in (10.1) are
integral (mod 2) except perhaps Mi and u3. Now
ffm.i 1Ml =
[2] a + 1
while
E(-l)vH)"PJ[3-r]'[4] (a + l)(a2 + 1)
Let 2"| (a+1), 2«+1>(a4-l); then (a24-a-f-l)2=T (mod 2«+x) and
¿(_l)rar(r-l)/2r ][3_f].
r-0 L r J
= (a2 4- a + l)"> - (a2 + a + l)(a + l)m + a(a2 + a + I)
0 (mod 2e+1) im even),
U 4- 1 (mod 2e+1) (m odd).t
Consequently u3 is integral (mod 2) for m even while for m odd 2w3= 1 (mod 2).
This yields the following supplement to Theorem 1.
Theorem 2. Let p = 2, q = a = l (mod 2); also let 2e|(a-fl), 2e+l\(a+l).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
344 L. CARLITZ [March
Then if e = l we have 2/3m = l (mod 2) for m even ¡± 2, 2/3i=-l (mod 2), while
ßm is integral (mod 2) for m odd ^3. If e>l then
(10.4) 2«p\» = 1 (mod 2)
for all m ̂ 1.
In particular it is evident from (10.4) that the denominator of ßm may be
divisible by arbitrarily high powers of 2.
In the next place we suppose g = a^l (mod p), p>2. It is now con-
venient to use [2, (5.3)]
JL /m \ s + 1(10.5) (g- l)-ßm = Z(-l)"-s( )t—"Tí-
*=0 \ 5 / [S + 1 J
We shall assume first that a is a primitive root (mod p2). Clearly in the right
member of (10.5) we need consider only those terms in which p— l|s-|-l.
Put a""1 = 1 +kp, p\k. Then
(Í)a(j.-l)r = 1 _|_ fkp _|_ { ^ ) tfpt _|_
a(p-l)r _ ! !
= * -f — (r - l)¿2/> -f • • • = * (mod p).rp 2
Thus (10.5) implies
(10.6) (a-l)"^„B(-iy»TE(> * ^(mod/.).* r>0 V(/> - 1) - 1/r>0 V(# - 1)
But it is known [4, p. 255] that
/ m \ 7-1 (rnod^) (P ~ l\m),
o<r(p-i)ém\r(p - 1) - 1/ (.O (mod p) (p - l\ m).
Hence (10.6) implies that p is integral when m^0 (mod p — l). This proves
Theorem 3. Let p^3, q = a, a primitive root (mod p2); then ßm is integral
(mod p) for p — l\m, while
1(10.7) pßm = - — (mod p) (p - 11 to),
w/zere k = (a"-1 — I)/p.
It is now clear how to handle the general situation. We may state
Theorem 4. Let p^3, q = a, where a belongs to the exponent e (mod p),
e>l. Put
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 345
(10.8) ae = 1 + plk ip\ k).
Then
e _ / m \(10.9) (a - l)">plßm =: — Z (-l)m-re[ ) (mod p).
k r>o \re — 1/
In particular if e = p — 1, then
i 0 (mod i) (p — l\m),
(10.10) (a - l)mplß ■ 1 .I-(mod p) (p — 11 m).I k
To prove (10.9) it is only necessary to observe that (10.8) implies
re re epl-= pl-m — (mod p).
a" - 1 (14- p'ky - 1 k
It is clear from (10.10) that the denominator of ßm may be divisible by
arbitrarily high powers of p. We also remark that theorems like Theorems 3
and 4 can be framed for rjm.
When p'\a, it is evident from (10.5) that
(10.11) ßmM Z (~l)g(m)(s+ 1) =0(modp°) (m>l).«=0 \ 5 /
11. Congruences. The formula (7.11) together with (10.2) makes it pos-
sible to derive certain congruences satisfied by Hmix, q). We observe, to
begin with, that if q = a is integral (mod p) then um— [s\\am,a satisfies, for 5
fixed and ip — l)pe~l\w,
(11.1) umiuw - l)r = 0 (mod pm, pre),
where after expansion of the left member, un is replaced by un. To prove
(11.1) we need only remark that
«">(«<» - l)r = a-3<-*-l»2J^ (-l)'a'-('-1)/2rS][5 - r]mi[s - r]» - l)\r-o L r J
If we look on x in (7.11) as an indeterminate and apply (11.1), we can
assert that
(11.2) HmiH<° - iy = 0 (mod pm, p").
We interpret this congruence in the following manner. The left member of
(11.2) is a rational function of x such that the coefficient of each term in the
numerator =0 (mod pm, pr). We may call (11.2) Kummer's congruence for
Hm. Using (7.13) we can prove like results for Hmiu; x), where u is now an
integer.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
346 L. CARLITZ [March
In view of (1.6) the result (11.2) can be restated in terms of Am(x):
r / T \ m+rw
(11.3) Z(-l)r_s( )Am+Ux) II (* - «0 - 0 (mod pm, p").«=0 V 5 / i=m+aw+l
We may state
Theorem 5. Let q — abe integral (mod p), x an indeterminate, and r^l;
then (11.3) holds.
In the next place (11.3) implies congruences for the Am,a of (1.7). (For the
case g = l, compare [l].) Since
m+rw (r— a)w p(y _ s)w~~\
(X - a*) = [ (-l)(.r-a)w-iai(i+l)ll+i(.m+,w) ^' \ xi¡
i=m+aw+l ¿=0 L I A
examination of the coefficient of x*-1 in (11.3) implies
¿* / r\ ^ Vi* - s)wlNE(-l)r_,( )Z (-l)(r-s)"'-^»1+»u,,it-iaiCi+1)/2+i(ra+>w)
(11.4) J=,o \s / i L i J
h 0 (mod />"*, />re)-
In order to obtain simpler results we consider some special values of the
parameters. In the first place we take r = l, so that (11.4) becomes
(11.5) Am+W,k- Z(-l)tt-i^,*-;ai(i+1)/2+i"TW] = 0 (mod p", p°).
(11.6)
Now suppose first that a = l (mod p). It is necessary to examine
(of» - 1) • • • (a«"-^1 - 1)
[:]- (a - 1) ••• (a* - 1)
We assume from now on that p>2. If we put a = l+plh, p\h, then as in the
proof of Theorem 4 we find that
[w "1 fw \il snd O
are divisible by exactly the same power of p. But if i<p>, it is clear from the
identity
(w \ w /w — 1\
*/ TV*- i/that
(w \1 = 0 (mod p*-t) (j g c).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 347
Consequently if p'~l^k<pi and j<e, we see that (11.5) implies
(11.7) Am+w.k = Amk (mod pm, p—f).
This proves
Theorem 6. Let p¡t3, q = a = l (mod p), p'~1(p — l)\w, and p'~xSk<p',
where j <e. Then (11.7) holds.
When a^l (mod p) let a belong to the exponent t (mod p). Then it is
clear from (11.6) that we need only consider those factors in the right member
with exponents divisible by t. Thus if io is the greatest integer &i/t we need
only examine
rawith a replaced by a'. The preceding discussion therefore applies and we ob-
tain the following theorem which includes Theorem 6.
Theorem 7. Letp^3, let q = a belong to the exponent t (mod p),pe~1(p—1)\ w
and k<tp', where j <e. Then (11.7) holds.
The case k = w is not covered by the theorem. We find for example that
if w = t=p— 1 (so that a is a primitive root (mod />)), then
Lm+p—l,k
Am,k (mod p) (k < p - 1),
Am.p-i + ( — jAm.o (mod p) (k = p — I),
where (a/p) is Legendre's symbol.
Returning to (11.2) we can also consider the case in which x is put equal
to an integer (mod p), provided the resulting denominators are not divisible
by p. Now the least common denominator is evidently
m+rw
^m+rw(x) = II (X - O").s-1
It will therefore suffice to assume that x^as (mod p) for any s. We may
therefore state
Theorem 8. Let a and x be rational numbers that are integral (mod p) and
suppose that xféa3 (mod p) for any s. Let
Pe~1(P — 1) | a> and r ^ 1.
Then
(11.8) Hm(x)(Hm(x) - iy m 0 (mod pm, p").
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
348 L. CARLITZ [March
In particular the theorem may be applied with slight changes to em(u)
= em(u, a) defined in (9.2); we have explicitly [2, (8.18)]
» (-l)'a'tm(u) = ¿^
tZ (a+l)(a2+l) ■■■ («•+14-1)
• ¿(-lFa'C-D^r5 \[u + s - r]'",r-0 L r J
which is included in (7.13). If u is an integer we have
em(u)(ew(u) - l)r = 0 (mod pm, pre)
provided p = 3 (mod 4) and a is a quadratic residue (mod p). For in this case
— 1 is a nonresidue (mod p) and therefore — 1 fea3 for any s.
12. Congruences involving r¡m and ßm. Let
If.. / r \ k- [k])(12.1) um = um,k,r = — <k[k "-V T' ?M -»)»»- (-l)m 7-77-7
m { \k / (q - l)m)
so that by (8.5)
1 f^1(12.2) um = — Z-Fm_!(rg-i),
q 1*1 1 — f
where f runs through the £th roots of unity distinct from 1. As for the de-
nominators in the right member of (12.2), note that
II («' - f) = a = «,(*_1) + • • ■ 4- aa 4- 1,r„, a» - 1
which is prime to p for all 5 provided p\k. We may therefore state
Theorem 9. If a is integral (mod p) and p\k, then
(12.3) com(ùj"" - l)r = 0 (mod ¿"lr-1, ¿>re),
wAere 03m is defined by (12.1) and pe~1(p — 1) | w.
As for ßm we have
k[k]m-Yßm(^> CM -¿
(12.4)(« + l)(g - 1) f« w f^1
=-Z--Hm(Çq-1) + — Z =-: F^-iífg-1)g t*i 1 - f Ç r*i 1 - f
analogous to (8.5). In much the same way as above (12.4) implies
Theorem 10. If a is integral (mod p) and p\k then
(12.5) Q,m(W - l)r = 0 (mod pn~l, />(^1)e),
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1954] g-BERNOULLI AND EULERIAN NUMBERS 349
where 17m stands for the left member of (12.4) and p"~1(p — l) | w.
Unfortunately we seem unable to obtain simpler congruences for ßm and
13. Combinatorial interpretation of am3T. Put
.-i ma — / . amsroq
(»-l)(m-«)
Z a»r=0
(r„ = sis - l)/2 4- r).
The following combinatorial interpretation of the coefficients amar was kindly
suggested by J. Riordan. The number amtr is the number of permutations of m
things requiring 5 readings and such that r = r2 + 2r3+ • • ■ +is— l)re, where
rk is the number of elements read on the ¿th reading. The following numerical
illustration for m = 4 was also supplied by Riordan.
Permutation
1234
124314234123
13241342312431423412
213423142341
143241324312
21342413243142134231
321432413421
4321
Reading
1234
123 4
12 34
1 234
12 3 4
l|23|4
1 2 34
12 3 4
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
350 L. CARLITZ
References
1. L. Carlitz and J. Riordan, Congruences for the Eulerian numbers, Duke Math. J. vol.
20 (1953) pp. 339-344.2. L. Carlitz, q-Bernoulli numbers and polynomials, Duke Math. J. vol. 15 (1948) pp. 987-
1000.3. G. Frobenius, Über die Bernoullischen Zahlen und die Eulerschen Polynome, Preuss.
Akad. Wiss. Sitzungsber. (1910) pp. 809-847.4. N. Nielsen, Sur le théorème de v. Staudt et de Th. Clausen relatif aux nombres de
Bernoulli, Annali di Matemática (3) vol. 22 (1914) pp. 249-261.5. J. Worpitzky, Studient über die Bernoullischen und Eulerschen Zahlen, J. Reine Angew.
Math. vol. 94 (1883) pp. 202-232.
Duke University,
Durham, N. C.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use