ON THE SUMS XXo¿p AND JZi-o(-l)"kp DAVID ZEITLIN 1. Introduction. Let uk satisfy a linear, difference equation of order m with real, constant coefficients: um+k + aium+k-i + a2um+k-2 + • • • + amuk = 0, ßo = 1, am 9e 0, k = 0, 1, • • ■ . Using our formula, (2.1), for Et-o u^k, we will give a closed form for zZt-oukrk, as well as closed forms for XXo &" and E"=o ( — l)kkp, which are also expressible in closed form by means of Bernoulli and Euler polynomials of the first order, respectively. We will show that n -2p+lJZ(-l)kk" (1-2) =Ê2(-lr( P+L\)[(-l)^(n+l+s)p-sp] k=o «-o \s+k+l/ (P = 0,1, • • • ), " p *=o *=o /n+p+l-kXÇi, / p+l \ \s+k+l/ p + I / B-0 \ (p - 0, 1, - • . ). Closed forms for Et-o kpukrk, E?-o M*r*cos (&#)>and E"-o «*''* sm (&?) may also be obtained from (2.1), but the details are omitted. 2. A lemma. We now derive the following result: Lemma 1. Let uk satisfy (1.1). Then 1 + JZ of1 Proof (2.2) T,ukrk =¿Z\ X) ajUnJrm-k-i rn+m'k m-1 r- m-l-t -] ¡fc-oL y-o J of (2.1). Let Ait i = l, 2, • • • ,m,he real constants, »„+Í + Aivm+k-i + A2vm+k-2+ ■ ■ ■ + Amvk = 0, Defining Avk= Vk+i—vk,(2.2) can be rewritten and let Am ^ 0. as Received by the editors June 1962; and, in revised form, April 15, 1963. 642 License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use ON THE SUMSYl-o (-I)**» A?; 'm+4-2- 1 + Y Ai y+i (2.3) Aflm+*_i+ Y moreover, l + Y^i ¡—i by summing 643 *" AND^=0 -/ + i>k= 0; both sides of (2.3) for k = 0, 1, • • • , », we obtain m-2 r (»n+m— »to-i) + Eh+E^' y-oL (2.4) j+1 -1 »-i J (Vn+m-l-j ~ »m-2-j) î + XM,Y% = 0. i=l If we set Vk= ukrk, then Ai=a¡r\ m-2 J *=0 * = 1, 2, • • • , m. Since y+1 X Y a¿rn+m_1_ Un+m—1—> j=0 i=l TO—2 m—l—k Jt=0 (2.5) j-1 n—1 pm—1—A "1 ^ Y ajUn+m-k-i rn+m-k k=0 L j=0 J _ £ «„+m_tr«+»»-*) (2.4) simplifies to (2.1). 3. Application of Lemma 1. The generating function of uk is readily obtained from (2.1). Let R>0 and suppose 2^L0 M*r*converges for |r| <R. Then, for \r\ <R, unrn^>0 as »->°°. gives, as »—>=o, p (3.1) m "] oo Thus, for |r| <i?, (2.1) m—1 ["m— 1—& 1 + Y «S* Y »*rk= Y \ Y aium-i-k-i r"-1"*. L <-i J *=o k=oL y-o J Another proof of (3.1) (based on the additive property of generating functions), is given in Jordan [l, p. 27]. We note that the generating function of uk is a rational function in r. Recalling the well-known identity Ci ÍP + i\ l~o (s - 0,1, • - - , p) .(¿+l)!(-l)p+1 (s = p + l) it follows that Uk= kp, p = 0, 1, • • • , satisfies *+i (p + 1\ ,_o \ Y (-i)'í Thus, from (2.1), with ; . K+1+*-/= 0. / uk= kv, m = p + l, and License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use ]• 644 DAVID ZEITLIN [August j = 0, 1, • • • , p-\-lt we obtain n (-l)p(r- 1)"+1X) ¿^ V-k =E (3.3) p -E /A _|_ 1\ E(-l)^ . r v-k /j) _i_ j\ £(-!)'( k=o L y=o -1 )(n + p+l-k-j)p\ . )(p-k-j)p \ J «n+p+1—k r»-». / If we set r= —1 in (3.3), we obtain (1.2). Let/(r) denote the right-hand side of (3.3) so that 71 (-l)p(r- l)p+lJZkprk =f(r). k=0 In §4, it will be shown that [/(8,W]r=i = 0 for 5 = 0, 1, • • • , p. Thus, applying L'Hospital's rule (£ + 1) times to f(r)/(r —l)p+1 for r= 1, we obtain (1.3) from (3.3), noting that (n + p+l-kyp+» = (p+l)\( (n + p + 1 - k\ * ), \ p+ I / where x(-p+1)=x(x—l) • ■ • (x —p). Remarks. Nielsen [2, p. 27, (8)] defines for non-negative integers M, N, and P M,N (3.4) * (M BP ,(x) = JZ(-iy( + l\ . )(x+p-j)». Since p—A; P-* / s=o V + ¿k ++ 1/ <5 + E(-l)s * + 1 ; , , \ )(n+l + s)p = (-l)^E(-lV( (1.3) can be rewritten (3.5) E^ ¡fc-l . )(n+l + p-k-j)p, as = (-DpE( »-0 \ ^ # + 1 )^,P(» + 1) (#-1,2,...). / Noting that 5^x(x) = 0 (see (3.2)), we obtain from [2, p. 237, (5), with n and £ interchanged, x = n + l, m = p] License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use i964] tkp=tï(2n+i+S)-(n+1 (3.6) 645 ON THE SUMSZ*-o ** ANDX"-o (- D**" ti ir0La p +1 + S\' I B7(n+l) \ p + i ). (P = 1, 2, • • • ). Combining (3.5) and (3.6), we obtain r . " * (2n [i + (-Dp]2> *=1 (3.7) + 1 + s\ =E ,=0 \ P + 1 / +D (P - 1, 2, • • • ). Thus, from (3.7), we conclude (3-8) Z that ,/2» + 1 + 5 ,_o \ ' : ; ~)2£~0»+ 1)-0 p + 1 " .p jSf'Cn+l) ft=l s=0\ 4. Additional (#-1,3,5,- / p /2» + 1 + A (3.9) 2l>=Z P + 1 •), (# = 2,4,6,..-). / proofs. We will require the following result: Lemma 2. For s = 0, 1, • • ■ , p, and arbitrary p / #\ (4.1) PtP J_1 K(» » /x + # - k\ Ç* L( VZl ¿=o \t / «-o \ x, we have (p + 1\ )Z(-i)> \ J. /)(#-*-i)'-1 / y-o í = A"-s1p, wAere A-l»= E (-l)r( (4.2) 0 )(»-r+l)». Proof of (4.1). Let U(x, s) denote the left-hand side of (4.1). Since we have t( X )u(x,s) ,=o \p — s/ (4.3) ,=0 \ 1 / 4=0 \ P / j=0 \ J / ^É(>É(2*+0£c->)f+W<. <«o\l / k=o\ P I y=o v J License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use I 646 DAVIDZEITLIN [August Nielsen [2, p. 28, (10) and (14)] has shown that .... (4.4) , „ ™ (y + k - l\ (-D—y = i_o\ Z( k /m+l\ )E(-i)'( \ J. /)(*-i)" ) ,-_„ m (m 2: n). Setting m = p, n = p—i, and y —2x+l (4.5) JZ( in (4.4), we obtain from (4.3) )u(x,s)=iz(P)(-x)i(2x+l)p-i=(x+l)p. ,=o \P — s/ ¿_o\ t / Since the expansion, (x+l)p= Jz(x)a'Ip, ,_o \ 5 / known as Newton's interpolation formula, is unique (see [3, p. 27], [4, p. 17]), it follows that U(x,p-s) sA'li», and thus U(x, s) =Ap-"1¡\ This completes the proof of (4.1). In §3, we claimed that |/(*)(*)]¡e-i=/(*)(l) =0 for 5 = 0, 1, • • • , p. To show this, \etf(x)=h(x)—g(x), p r p-k where (see (3.3)) /í4-i\ 1 i+p+i-t (4.6)*(*)= E I E (-l)M . J (»+ # + 1- *-j)'\ xn (4.7)g(x)= èr2(-i)yfi+1)f>-*-i)*i*^. We now proceed to show that h(,)(I) = gU)(I) =s\Ap~'lp, • • • , p. Using the binomial theorem 5 = 0, 1, and noting that (n + p + 1 - k\ = s\( r \, (n + p+l-k)^ we obtain from (4.6) A<«>(1) = JZ{n+p+l-k)^JZ(-iy(P+V)(n+p+l-k-j)p k=0 (4.8) y=0 \ J / = s<.±(p)(»+1y±("+p+i-h) <_o\ î / k=o\ p-k /p+l\ y-o \ J s I • E (-1)' ( . Kp-h-f)*-' = í!A«~l>, / having applied (4.1) with x = n + l. License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use 1964] ON THE SUMSE*-o k" AND Et-o (-1)1*" 647 If we set x = 0 in (4.1), the only nonzero term occurs for i = 0. Thus, (4.9) ,?, IP —k\ tí /* + 1\ k-0\ \ E( )L(-1)M S / y=o . )(p-k-j)> J = A>~l> / (5 = 0, 1, • • •,/»); and from (4.7), we have g(>,(i)=s\±(p~ (4.10) k)Y(-w(p *=A s Ji~° + 1)(p-k-j)' Kj J - jIap-'I». Thus,/">(l)E=A<"(l)-g(,)(l)=0, 5 = 0, 1, •••,#, as claimed. Acknowledgment. I wish to thank Professor L. Carlitz for providing me with the outline of the proof for (4.1). References 1. C. Jordan, Calculus of finite differences, Chelsea, New York, 1960. 2. N. Nielsen, Traité élémentaire des nombres de Bernoulli, Gauthier-Villars, Paris, 1923. 3. L. M. Milne-Thomson, The calculus of finite differences, Macmillan, London, 1933. 4. N. E. Norlund, Differenzenrechnung, Honeywell, Minneapolis, Springer, Berlin, 1924. Minnesota1 1 The author is no longer employed at Honeywell. Reprint requests should be addressed to the author's home, 1650 Vincent Avenue North, Minneapolis, Minne- sota 55411. License or copyright restrictions may apply to redistribution; see http://www.ams.org/journal-terms-of-use