Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Pacific Journal of Mathematics ON FINITE SUMS OF RECIPROCALS OF DISTINCT nTH POWERS RONALD L EWIS G RAHAM Vol. 14, No. 1 May 1964 ON FINITE SUMS OF RECIPROCALS OF DISTINCT nTR POWERS R. L. GRAHAM Introduction* It has long been known that every positive rational number can be represented as a finite sum of reciprocals of distinct positive integers (the first proof having been given by Leonardo Pisano [6] in 1202). It is the purpose of this paper to characterize {cf. Theorem 4) those rational numbers which can be written as finite sums of reciprocals of distinct nth. powers of integers, where n is an arbitrary (fixed) positive integer and "finite sum" denotes a sum with a finite number of summands. It will follow, for example, that p\q 1 is the finite sum of reciprocals of distinct squares if and only if Our starting point will be the following result: A. Let n be a positive integer and let Hn denote the sequence (l~ , 2~~n, 3~w, ••••). Then the rational number pjq is the n finite sum of distinct terms taken from H if and only if for all ε > 0, there is a finite sum s of distinct terms taken from Hn such that 0 ^ s — pjq < ε. THEOREM n Theorem A is an immediate consequence of a result of the author [2, Theorem 4] together with the fact that every sufficiently large integer is the sum of distinct nth. powers of positive integers (cf., [8], [7] or [3]). The main results* We begin with several definitions. Let S = (su s2, •••) denote a (possibly finite) sequence of real numbers. DEFINITION 1. P(S) is defined to be the set of all sums of the form Σ?=i εk8k- where εk = 0 or 1 and all but a finite number of the εk are 0. 2. Ac(S) is defined to be the set of all real numbers x such that for all ε > 0, there is an s e P(S) such that 0 ^ s — x < ε. Note that in this terminology Theorem A becomes: DEFINITION (1) P(Hn) - Ac(Hn) n Q Received May 13, 1963. 1 This result has also been obtained by P. Erdos (not published). 85 86 R. L. GRAHAM where Q denotes the set of rational numbers. 3. A term sn of S is said to be smoothly replaceable in S (abbreviated s.r. in S) if sn ^ Σ*U*»+* DEFINITION THEOREM 1. Let S = (slrs2, •••) be a sequence of real numbers such that: 2. There exists an integer r such that n Ξ> r implies that sn is smoothly replaceable in S. Then Ac(S) = U [π, π + <*) where P r _ x = P((su , sr_x)) (note that P o = {0}) and σ = Y^^rsk(where possibly σ is infinite). Proof. Let x e \JπePr_1 [π, π + σ) and assume that x ί Ac(S). Then x e [π, π + σ) for some π e Pr_i. A sum of the form π + Σ*U \ where T g ix < i2 < < ik will be called "minimal" if (2) π + Σ s ί f < x < π + Σ 8it ί=l * t=l (where a sum of the form Σ<*U is taken to be 0 for b < a). Note that since x £ Ac(S) ID P(S) then we never get equality in (2). Let M denote the set of minimal sums. Then M must contain infinitely many elements. For suppose I is a finite set. Let m denote the largest index of any sd which is used in any element of M and let p = π + Σ*=i sdk + s™ b e a n element of M which uses sm (where r ^ θΊ < 32 < < 3n < m and possibly n is zero). Thus we have n n oo π+ since sm is s.r. in S. Therefore t h e r e is a least d^l such t h a t x < p' = 7Γ + Σ l U s ifc + Σ*=i sm+f Hence p ' is a "minimal" sum which uses sm+d and m + d > m . This is a contradiction t o t h e definition of m and consequently Mmust be infinite. Now, let δ = inί{p -x:pe M}. Since x ί Ac(S) t h e n δ > 0. There exist p l f p 2 , e M such t h a t p Λ — x < <5 + S/2\ Since sn [ 0 then there exists c such t h a t n^c implies t h a t sn < δ/2. Also, there exists w such t h a t n^w implies t h a t pn uses an sk for some k^c (since only a finite number of p, can be formed from t h e sk with k < c). Therefore we can write pw = 7Γ + Σ?=i sfc, where ATO ^ c. Hence ON FINITE SUMS OF RECIPROCALS OF DISTINCT ΛTH POWERS 87 which is a contradiction to the assumption that pw is "minimal/' Thus, we must have xeAc(S) and consequently (3) U [π,π + σ)c:Ac(S) . To show inclusion in the other direction let x e Ac(S) and suppose that x e U^€Pr_! ίπ> π + σ) Thus, either x < 0, x ^ Σ~=isfc> o r there exist π and TΓ' in P ^ such that π + σ ^ x < π' where no element of P r _ x is contained in the interval [π + σ, TΓ'). Since the first two possibilities imply that x $ Ac(S) (contradicting the hypothesis) then we may assume that the third possibility holds. Therefore there exists δ > 0 such that ( 4) x g π' - 3 . Let p be any element of P(S). Then p = ΣΓ=i S ^ + Σ2=i s % f° r m and w where I^i1<i2< < im ^ r - 1 < j \ < j 2 < s o m e < in . Thus for π* = ΣS=iβ<ί we have pe [TΓ*, TΓ* + σ). Consequently any element p of P(S) must fall into an interval [π*, TΓ* + o) for some TΓ* 6 Pr_i and therefore, if p exceeds x then it must exceed x by at least δ (since p ί [π + 0", π') and thus by (4) p > x e [π + (7, TΓ') implies p^ πf ^ x + δ). This contradicts the hypothesis that x e Ac(S) and hence we conclude that Ac(S)c \Jπepr_1[π>π + σ ) Thus, by (3) we have Ac(S) = U ^ f v - J ^ TΓ + σ) and the theorem is proved. THEOREM 2. Le£ S = (5 t , 52, •••) be a sequence of real numbers such that: 1. 8.10. 2. There exists an integer r such that n < r implies that sn is not s.r. in S while n^r implies that sn is s.r. in S. Then Ac(S) is the disjoint union of exactly 2r~1 half-open intervals each of length Σϊ=rsk. Proof. By Theorem 1 we have Ac(S) = U*e/>,._! [^ π + σ) where a n d π G = ΣΓ=r Sfc and P r _ x = P((sx, , s,^)). Let π = Σϊ=iβifc ' = s Σ*=i i* be any two formally distinct sums of the sn where 1 ^ ίi < < i u ^ T — 1 and 1 S ji < < U ^ ^ — 1 and we can assume without loss of generality that TΓ Ξg TΓ'. Then either there is a least m ^ 1 such that im Φ j m or we have ik = i^ for & = 1, 2, , v and 88 R. L. GRAHAM u > v. In the first case we have m—1 > Σ S JJC + Σ sim+k (since sim is not s.r. in S) Ξ> TΓ' + σ (since i m ^ im + 1) . In the second case we have π = Σ sik = Σ s;* + Σ *<* > Σβijfc + Σ 8 * β+1 +* (since «ίβ+1 is not s.r. in S) ^π'' +σ (since i β + i + l ^ iu + 1 ^ r) . Thus, in either case we see that π > πf + σ. Consequently, any two formally distinct sums in Pr_χ are separated by a distance of more than σ and hence, each element π of Pr-± gives rise to a half-open interval [TΓ, π + σ) which is disjoint from any other interval [TΓ', TΓ' + σ) for TΓ Φ τr'eP r _!. Therefore Ac(S) = \Jnepr-.x\.π9π + σ) is the disjoint union of exactly 2r~1 half-open intervals [TΓ, TΓ + σ), πe Pr-19 (since there are exactly 2r~~1 formally distinct sums of the form Σί=i ε Λ>εk — 0 or 1) where each interval is of length σ. This proves the theorem. We now need three additional lemmas in order to prove the main theorems. LEMMA 1. Let S = (slf s2, •) be a sequence of nonnegative real numbers and suppose that there exists an m such that n ^ m implies that sn^2sn+1. Then n^m implies that sn is s.r. in S (i.e., S &n = 2-Ak=l n+k) Proof. If Σ?=i sk — °° then the lemma is immediate. T h e n that Σ*=iβ*< °° n ^ m - = > sn+k ; ^ — Sn+k-!, fc oo 1 °° ' Σ S%+fc = "77 2-iSn+k-l Jfe=l 2 k=l β e s Therefore, sn ^ Σ?=i »+fc> ΐ » w ^ ss r ln Assume = 1, 2, 3, 1 1 °° — ~X"Sw + -X" Σ Sn+k 2 2 A=l i n ^ 1 % n 2. Suppose that k-£(2r — I ) " ami &^ is s.r. in H {where Hn was defined to be the sequence (l~n, 2~n, ••*)). Then (k + 1)-* is also s.r. in H\ LEMMA ON FINITE SUMS OF RECIPROCALS OF DISTINCT ΛTH POWERS 89 Proof. k ^ (2lln - I ) - 1 = > — ^ 2lln - 1 k (5) Since by hypothesis, Σ~=*+i 3~n ^ A"*, then by ( 5 ) Σ t n ^ Ar - (k + l)~n ^ 2(fc + 1)-" ~(k + 1)-" = (k + 1)-" . j = k+2 w n Hence, (k + l)~ is s.r. in H and the lemma is proved. LEMMA 3. Suppose that k ^ (21/Λ - I)" 1 . Then k~n is s.r. in Hn. Proof. (2lln -I)-1 r^"k=>r^ 1 +r/ — ^2 r~n ^ 2(r + 1)— . Therefore, by Lemma 1, r~% is s.r. in Hn. n THEOREM 3. Let tn denote the largest integer k such that kr n n n n is not s.r. in H and let P denote P((l~ , 2~ , •••, t~ )). Then tn lln is the disjoint union of exactly 2 intervals. Moreover, tn < (2 — I)" and tn ~ n\ln 2 (where In 2 denotes loge 2). 1 Proof. With the exception of tn ~ n\ln 2, the theorem follows directly from the preceding results. The following argument, due to L. Shepp, shows that tn ~ n/ln 2. Consider the function fn(x) defined by Στ4τcr Λ *=i (x + kf for n = 2, 3, and x > 0. Since /.(*) = Σ ( i + -)~n Λ=l\ xn -1 X/ then fn(x) < 0 for sufficiently small ti > 0, fn(x) > 0 for sufficiently 90 R. L. GRAHAM large x, and fn(x) is continuous and monotone increasing for x > 0. Hence the equation fn(x) = 0 has a unique positive root xn and from the definition of tn it follows by (6) that 0 < xn - tn ^ 1. Thus, to show that tn ~ nβn 2, it suffices to show that xn ~ nβn 2. Now it is easily shown (cf , [4], p. 13) that for a > 0, (1 + a/n)~n is a decreasing function of n. Thus, f%{an) is a decreasing function of w and since f%(2a) < oo for a > 0 then limΛ {<*n) = lim Σ f 1 + — V * - 1 = Σ lim (ί i + - & VM = - l + Σ e~kl" = ( el/a> - 1 ) " 1 since the monotone convergence theorem (cf., [5]) allows us to interchange the sum and limit. Suppose now that for some e > 0, there exist nx<n2< such that xH > n^l/ln 2 + e). Then = (21/(1+εlw2) - I)- 1 - 1 > 0 which is a contradiction. Similarly, if for some ε, 0 < ε < lβn 2, there exist ^ < n2 < such that then 0 = BmAίag g M m ^ K l ^ " £ )) = (βd/ϊ a- )-1 _ l)-i _ 1 = (21/(1-8ϊ»2) - I)- 1 - 1 < 0 which is again impossible. Hence we have shown that for all s > 0, there exists an n0 such that n > n0 implies that ^ ln2 ε)^xn^ ) n( \ln2 or equivalently -ε ^ x » - 1 ϊ2 g e. ON FINITE SUMS OF RECIPROCALS OF DISTINCT nTR POWERS 91 Therefore, lim xjn = l / » 2 and the theorem is proved.2 The following table gives the values of tn for some small values of n. n [(21/κ - I)-1] 1 2 3 5 6 13 143 1442 K 0 1 2 4 5 12 1 2 3 4 5 10 100 1000 1 ? We may now combine Theorem 3 and Theorem A (cf. Eq. (1)) and express the result in ordinary terminology to give: THEOREM 4. Let n be a positive integer, let tn be the largest integer k such that k~n > 2?=i (& + J)~n and let P denote the set {Σ* =iεjj~~n: εJ — 0 or 1}. Then the rational number p\q can be written as a finite sum of reciprocals of distinct nth. powers of integers if and only if U πβP COROLLARY 1. p\q can expressed as the finite sum of reciprocals of distinct squares if and only if COROLLARY 2. p\q can be expressed as the finite sum of reciprocals of distinct cubes if and only if Ά e Γθ, ζ(8) - A ) u Γ l , ζ ( 3 ) - l ) U Γl, ζ(3) - ± ) U Γ», ζ(3)) q L w 8/ L8 / L ' 8/ L8 / 3 where ζ(3) = Σ*U &~ = 1.2020569REMARKS. In theory it should be possible to calculate directly from the relevant theorems (cf., [2], [3]) an explicit bound for the number of terms of Hn needed to represent p\q as an element of P(Hn). However, since the theorems were not designed to minimize such a bound, but rather merely to show its existence, then understandably, this calculated bound would probably be many orders of 2 In fact, it can be shown that xn has the expansion n/ln2 — 1/2 + cιn~ι + ckn~k + OOt-*-1) for any k. 92 R. L. GRAHAM magnitude too large. Erdos and Stein [1] and, independently, van Albada and van Lint [9] have shown that if f(n) denotes the least 1 1 number of terms of H — (I" , 2~\ •) needed to represent the integer n y n as an element of P(£P) then f{n) ~ e ~ where 7 is Euler's constant. It should be pointed out that a more general form of Theorem A may be derived from [2] which can be used to prove results of the following type: COROLLARY A. The rational p\q with (p, q) = 1 can be expressed as a finite sum of reciprocals of distinct odd squares if and only if q is odd and pjq e [0, (π2/8) - 1) U [1, π2/8). B. The rational p/q with (p, q) = 1 can be expressed as a finite sum of reciprocals of distinct squares which are congruent to 4 modulo 5 if and only if (q, 5) = 1 and COROLLARY where a = 2(5 - vΊ>V/125 = Σ?=o((5A; + 2)~2 + (5k + 3)~2) = 0.43648It is not difficult to obtain representations of specific rationale as n elements of P(H ) (for small n), e.g., — = 2~2 + 3-2 + 4-2 + 5-2 + 6~2 + 15-2 + 18-2 + 36~2 + 60~2 + 180-2, Δ — = 2-2 + 4-2 + 10-2 + 12-2 + 20-2 + 30-2 + 60- 2 , 3 A = 2~3 + 5"3 + 10-3 + 15-3 + 16-3 + 74"3 + 111"3 + 185~3 + 240~3 2 3 3 + 296" + 444" + 1480- , etc.! REFERENCES 1. P. Erdδs and S. Stein, Sums of distinct unit fractions, Proc. Amer. Math. Soc, 14 (1963), 126-131. 2. R. Graham, On finite sums of unit fractions} Proc. London Math. Soc, (to appear). 3. , Complete sequences of polynomial values, Duke Math. Jour, (to appear). 4. P. P. Korovkin, Inequalities, Random House, New York (1961). 5. M. Loeve, Probability Theory, Van Nostrand, Princeton, (1960). 6. Leonardo Pisano, Scritti, vol. 1, B. Boncompagnί, Rome (1857) 77-83. 7. K. F. Roth and G. Szekeres, Some asymptotic formulae in the theory of partitions. Quart. Jour, of Math., 5 (1954), 241-259. 8. R. Sprague,' ϋber Zerlegungen in n te Potenzen mit lauter verschiedene Grundzahlen, Math. Zeit., 5 1 (1947-48), 466-468. 9. P. J. van Albada and J. H. van Lint, Reciprocal bases for the integers, Amer. Math. Monthly 7 0 (1963), 170-174. BELL TELEPHONE LABORATORIES, INC. PACIFIC JOURNAL OF MATHEMATICS EDITORS ROBERT OSSERMAN J. DUGUNDJI Stanford University Stanford, California University of Southern California Los Angeles 7, California M. G. ARSOVE LOWELL J. PAIGE University of Washington Seattle 5, Washington University of California Los Angeles 24, California ASSOCIATE EDITORS E. F. BECKENBACH B. H. NEUMANN F. WOLF K. YOSIDA SUPPORTING INSTITUTIONS UNIVERSITY OF BRITISH COLUMBIA CALIFORNIA INSTITUTE OF TECHNOLOGY UNIVERSITY OF CALIFORNIA MONTANA STATE UNIVERSITY UNIVERSITY OF NEVADA NEW MEXICO STATE UNIVERSITY OREGON STATE UNIVERSITY UNIVERSITY OF OREGON OSAKA UNIVERSITY UNIVERSITY OF SOUTHERN CALIFORNIA STANFORD UNIVERSITY UNIVERSITY OF TOKYO UNIVERSITY OF UTAH WASHINGTON STATE UNIVERSITY UNIVERSITY OF WASHINGTON * * * AMERICAN MATHEMATICAL SOCIETY CALIFORNIA RESEARCH CORPORATION SPACE TECHNOLOGY LABORATORIES NAVAL ORDNANCE TEST STATION Printed in Japan by International Academic Printing Co., Ltd., Tokyo, Japan Pacific Journal of Mathematics Vol. 14, No. 1 May, 1964 Richard Arens, Normal form for a Pfaffian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Charles Vernon Coffman, Non-linear differential equations on cones in Banach spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ralph DeMarr, Order convergence in linear topological spaces . . . . . . . . . . . . . . . . . Peter Larkin Duren, On the spectrum of a Toeplitz operator . . . . . . . . . . . . . . . . . . . . . Robert E. Edwards, Endomorphisms of function-spaces which leave stable all translation-invariant manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Erik Maurice Ellentuck, Infinite products of isols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . William James Firey, Some applications of means of convex bodies . . . . . . . . . . . . . . Haim Gaifman, Concerning measures on Boolean algebras . . . . . . . . . . . . . . . . . . . . . Richard Carl Gilbert, Extremal spectral functions of a symmetric operator . . . . . . . . Ronald Lewis Graham, On finite sums of reciprocals of distinct nth powers . . . . . . . Hwa Suk Hahn, On the relative growth of differences of partition functions . . . . . . . Isidore Isaac Hirschman, Jr., Extreme eigen values of Toeplitz forms associated with Jacobi polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chen-jung Hsu, Remarks on certain almost product spaces . . . . . . . . . . . . . . . . . . . . . George Seth Innis, Jr., Some reproducing kernels for the unit disk . . . . . . . . . . . . . . . . Ronald Jacobowitz, Multiplicativity of the local Hilbert symbol. . . . . . . . . . . . . . . . . . Paul Joseph Kelly, On some mappings related to graphs . . . . . . . . . . . . . . . . . . . . . . . . William A. Kirk, On curvature of a metric space at a point . . . . . . . . . . . . . . . . . . . . . G. J. Kurowski, On the convergence of semi-discrete analytic functions . . . . . . . . . . Richard George Laatsch, Extensions of subadditive functions . . . . . . . . . . . . . . . . . . . . V. Marić, On some properties of solutions of 1ψ + A(r 2 )X ∇ψ + C(r 2 )ψ = 0 . . . William H. Mills, Polynomials with minimal value sets . . . . . . . . . . . . . . . . . . . . . . . . . George James Minty, Jr., On the monotonicity of the gradient of a convex function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . George James Minty, Jr., On the solvability of nonlinear functional equations of ‘monotonic’ type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . J. B. Muskat, On the solvability of x e ≡ e (mod p) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Zeev Nehari, On an inequality of P. R. Bessack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Raymond Moos Redheffer and Ernst Gabor Straus, Degenerate elliptic equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Abraham Robinson, On generalized limits and linear functionals . . . . . . . . . . . . . . . . Bernard W. Roos, On a class of singular second order differential equations with a non linear parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Tôru Saitô, Ordered completely regular semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . Edward Silverman, A problem of least area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Robert C. Sine, Spectral decomposition of a class of operators . . . . . . . . . . . . . . . . . . Jonathan Dean Swift, Chains and graphs of Ostrom planes . . . . . . . . . . . . . . . . . . . . . John Griggs Thompson, 2-signalizers of finite groups . . . . . . . . . . . . . . . . . . . . . . . . . . Harold Widom, On the spectrum of a Toeplitz operator . . . . . . . . . . . . . . . . . . . . . . . . . 1 9 17 21 31 49 53 61 75 85 93 107 163 177 187 191 195 199 209 217 225 243 249 257 261 265 269 285 295 309 333 353 363 365