You are on page 1of 14
‘two METWODe IM ELEMENTARY ANALYTIC WOXBER THEORY Separtaent of Mathenatics Athens, Georgia 30602 azerme., thle mer pes slates te tof be Setinates ror a variety of counting Functions ana for" SERS SS crutt e aoe ls Stent meteats iene teeta Daranster, replacing « finite sum with an infinite oup Ginny roniet oteema ar cranen., SULT EG eta fav deste MRoME hase Sie dele agro et ante SEP ake: Mg eases SPR Sy er Beet ehtlce nu etet ttle, he Se Sarah nae AE PER an factorizations" function, the maximal order for the number cities ean ts emret ae Bete oe te tr inhi. tere at 2a Ea ‘Supported in part by an NSP grant. fat ne er tes 8 s2,mwmopuerros, Te wmual definition of olenentery muaber Lueory 18 one of exclusion. rt does not use complex analysis, it does not deal vith algebraic number fields, it ia not probabilistic hmuaber theory, ete. Thus the title of this paper seens to be a contradiction in terms. wnat I mean by "elementary analytic number theory" is the use of elenentery techniques in areas whose greatest successes are dominated by analytic Rethods. For example, the estination of counting functions ie such an area. Its contral result, the prine munber theorem, is proved in its sharpest form vith analytic ethods, T would consider the various elenentary proofs of ‘the prine munber theoren as crouning achievements of elementary analytic number theory. Another achievenent can bbe found in Brun’s sieve and, in general, combinatorial sieve methods. ‘This paper vill discuss, in the context of = fev specific problens, variations on two elenentary theses. One of these thenes, sonetines referred to as *Rankin's ethod," can be used to obtain upper bounds, eonetines quite sharp ones, for certain counting functions. The other theme is concerned with lover bounds and has a distinctly combinatorial flavor. ‘The folloving problens vill be discusses 1, he distribution of smooth numbers, that ie, numbers which have only snail prine factors. ‘The maximal order of the function f(n) which counts the nunber of unordered factorizations of n . 3. The maxinal order af the function Win) whiel: counts ‘the munber of solutions ef #(a) =n, where vis Ruler‘e function, Me distribution of pesudoprines, Aithough this is primartiy a survey paper, fairly ‘complete proofs vill be given. It is possible that sone of the material presented here would be of use in a graduate nunber theory course. In ay opinion, to see the povor and breadth of elementary methods better desis role of analytic mathods in umber theory. For balance, the reader might algo coneult Tenenbaum (26), a survey paper shoving the role of analytic methods with some of these problens. $2, "ME DIBTRIBUTTON oF swooTH WoKDERS. Let #(x)y) denote the number of naturel numbers nsx , whose largest prime factor, P(n) , satisfies P(n) 5/3 « ‘Te function »(u) decays to 0 quite rapidly as use. Te de known (deBrusjn (3]) that (a) = exp(-G2 + 202))u tog w) as Ue ‘Thus from the vork of Hildebrand, Maier, and Tenenbaum, ve have the folloving veaker result Theorem 2.1, Suppose ¢ > 0 Ss arbitrarily email, but fixed. If y satiefies exp((log x) <¥ < exp( (Log 2)2°4) , then vouy) 2+ exp + aQ))u Log ay uniformy as x7 ©, where u= (log x)/log ¥ ‘This section will be devoted to an elementary proof of ‘Tooren 2.1. The arguaent presented pre-dates the finer results of (15], (16), (18). We begin with an upper bound argument. the idea vas ‘used in 1938 by Rankin (24) and was developed nore fully in dearuijn [4]. The upper bound implicit in Theorem 2.1 4s actually proved for a wider range of y than is indicated in the statenent. ‘The Key idea occurs in the very first step. If © > 0, 22) vay =D rsx DS wee Tape, ex Pay, psy Pinysy Wnere p denotes a prime. We thus replace a finite eux with an infinite sum vhich has an Euler product. our goal 4 is to estinate this product using prime nunber theory and Use choowe © optimaniy in (2.2). Note that if © 21/2 +¢, then 2.) TT a-pryt = exp|- E03 (a-p*) psy poy = ewlZ pres occ]. psy ‘me final um in (2.3) Ss easily estimated with the prine number theoren: if 0 ato) uniformly for mea, J €9(m) Putting this estimate in (2.6), we have sn eS a8). 00) gy seria Beene ny) Taptayye = 2f¥o()1,(40081/¢ x9 ¢m)+1) 210g 2)%0(") an = (ym /((ug(m) 1) 220g 2/0), where, for the last inequality, we used (2) ~ n(x) > 2/(2log 2) , which te valia for x (and hence 2) sufficiently large. The lover bound for 7(m)| used in (2.7) cones from the combinatorial counting principal for ‘the number of vays of choosing & not necessarily distinct thinge fron a t-elenent est: it is at least ye. We nave m2 el¥] , 20 that eatayst¥)) _ tog 28) 0 supe « MALAI og x _ Gy) 2 tog Tog ui msgtpt. since Loglog £ < logleg y = (log 8) + (2.8) implies there 1s sone constant Ki depending only on the choice of ¢ with 2 ((up(m) + 220g 21%) < epoRuy , for all m © A. ‘thus fron (2.7) we have (2.9) 9&2) > E exp (ru) uniformly for all mea. Putting (2.9) in (2.5) we have exp (-Keu) ze vouy) >> x ax emer Tee Bt morn) TE 2 sy (2.20) 2x > exp(-u(tog u + togiog w+ 0()9) + since Erecpey /P ~ Vio 8 - Znea ‘theoren. ote that (2.10) completes the proot of the theoren. ‘The combinatorial arguments used for the lover bound for ¥(x,y) Will be echoed in the folloving sections in various ways. the argunent just presented is a condensed version of the proof in (5]. ‘The lower bound for 1/m used in (2.10) cones from the multinomial 43, MYoMLY ractomanLe xoxmERa. Lat f(n) denote the number of unordered facterizations of nn into factors exceeding 1. By convention, ve let £@) =1. ten, for example, £(12) = 4, since 12 has tthe factorizations 12, 2:6, 3:4, 22:2. We say a natural number nis highly factorable it f(n) > £(m) for all» 0, en mos Dorm sx Seema. nex P(n) Steg x P(n)s2log x ‘This inequality is similar to (2.2), except that we cannot now replace the last sun with an Ruler product since the function £(n) is not multiplicative. Hovever, a generalization of a formula of MacMahon (17) giv something very sinila 0.2) So timmre= TE awry P(n) tog x Pimp s2iog x wi which can be easily seen by replacing (2-m"6)-1 with, +m +264... and multiplying out the product. 3 is easy to show that the right side of (3.2) i convergent for c>0 and so the left side is as well. Assuming © 2 1/2 +¢ , ve have from (3.1) and (3.2) senate rey cx TD P(m) 20g x wept 0 auch that Nin) 2-n© for infinitely many nm, vhere M(n) is the number of m with »(n) =n. Moreover, Erase conjectured that © can be taken arbitrarily close to 1. In this section ve shall discuss thie problen. NeG) = max(N(nys nS x) By an argunent very similar to the upper bound in Theorem 3.1, we have the following result which first appeared in (20). mmeorem 42. as xs #, HEC < x/EQ) OO) Prost. Lat n> Then we could prove We(x) = x/L(«)H42(2) follows from a weaker hypoth Note that Theoren 2.1 implies In fact this 185 we shall now ses. xe ¥¢x,00108 947? = sve + 0120) 05 29°77 restos 2 « 2. WIth M(x) defined an in nypottesis 4.2, Moo = rec + 20) G05 0? rogiog 0). Teoren 4.4. Assuming Hypothesis 4.9, we have N#(x) = ee eta ‘Broot. We have already seen in Theoren 4.1 that we have N#(x) < x/400)1*01) unconditionally. We now show the Feverse inequality. Let t= loglog x as in the proof of = (p prime: ps0” , P(p-a) < tog x) ‘Ten fron Hypothesie 4.2, ve have ep = oP Gteane tos & Thus > is entirely analogous to the set 9 constructed in the proof of Theorem 3.1. Let k= [(10g x)/ Giogieg *)2} . instead of choosing unordered k-tupll fron? , ve choose k-elenent sunsets. The number of ‘these subsets is (2}> ick @K(2-(at0(a)) tog 2) aay = apeon tte. Yor each k-elenent subset, ve multiply these prines together forming an integer eae Aithough Hypotheses 4.2 and 4.3 appear hopeless to prove at this time, we can salvage sonething. Definition 4.5. tet = denote the euprenum of the set of © (0/2) for which there is some cy > 0 with the property that the mumber of prines p< x with P(p-i) < x84 exceeds cqx/log x for all x22. In [7], Brdés used Brun’s method to shov that E> 0 and conjectured that == 1. Wooldridge (27) used selberg’s sieve to show E> 3-2/2 = .17187, In (20), Bonbieri’s theoren and sone results of Hooley aid Ivaniec ‘concerning the Brun-Titchnarsh theoren on average are used to show that > 5/9. dust this year, Friedlander [13] used his extension of Boubieri’s theoren with Bonbieri and Twaniec to show E21 - (2Ve)"! = .69673... . other relevant papers on the subject are Balog (1] and Fouvry and Grupp (12). Theorem 4.6. Ae x > 9 ve have we(x) 2 x2t0(2) , roof. Fix an arbitrary ¢ with 0<¢ <2 and let, B= (L-E4c)7l. Lat 9 be the oot of prines P< (log x)? with P(p-1) < log x. From the definition of EB, we have (4.4 IPL >> leg 2) /ogieg x - Let w= [(log *)/(@ loglog x)) and let 4 be the set of integers composed of u distinct primes fron 2. Then ean med satisties ms x and Plo(m)) < log x Moreover, from (4.4), = [2) > Since © maps £ toa set of size ¥(x,10g x) = x*(1) (see (2.6)), there is sone n counted by ¥(x,209 x) with x(0°U)/8%6@) pre-images under ¢ . Bat, so taarereay , @-ve=e-e, 0 that we(x) 2 xB-Ct=(2) | since € can be arbitrarily small ve have our theos ‘Theoren 4.6 vas first proved by Erdés in (7). 4.5. sum prerRrpurron oF pazuDoparies. Since Fermat ve have knovn that if nde prime and nia, than aR"l = 2 (nod n). thle congruence can cometines hhold when n is composite. For example, it holds for all no vhon a= 1, it holds for n= 241 ~ 11-21 when a= 2, and it holds fer n= 91 when a=3. If n isa ‘composite natural nunber and aM“ = 1 (mod n), then n is said to be a pseudoprine to the base a. Lat Pa(x) enote the number of base a peeudoprines n xola) , with x(x) the same as in sections 2 and 4. This proof As a bit technical and the theorem is probably not best possible. I conjecture that for every a with [al > 1 wo have P(x) = x/L(x) 202), It will probably be nore illuminating to work through the proof of & sinilar theorem that gives a stronger result for a more restrictive set of numbers. Sone composite numbers n , such as 561, 1105, and 1729, have ‘the property that they are "absolute peeddoprines,” that Ae, they are pecudoprines to avery base to which they are coprine. These munbers are also called Carmichael Let 4(n) denote the maxinal order for an element in th: multiplicative group (2/n)* . By the theoren on the primitive root, .(p8):» pt1(p - 1) for an odd prime p or for p= 2or4. also 1(28) = 2%? for az3. Further, from the chinese remainder theoren, it is easy to see that (nm) = lem (\(p8): pan) . Tt 4s also eaey to seo that a¥(9) © 1 (nod n) for all ac (8/n)*. thas Wwe have the folloving simple criterion: the composite integer n is a carmichael number if and only if dinpin = 2 In 1956, Erdis [9] proved that C(x) , the number of Carmichael numbers up to x , aatiaties 6(3) x x/L(H)* for sone © > 0 and x large. this vas improved to © 1+ e(1) in (23). We now give a simplifies proot. Moore 8.4. as x= 2, efx) ¢ x/E(e eC) , roof. Tf a is a natural number, we ask how many Carmichael numbers n x/1(x)? , namely n itself. But eince n> x/L(x) and P(n) < L(x) , the least such divisor 4 must also as xu - ‘mus by (5.3), tiaty oe Dra + gay, z ae where 5’ denotes a sum over 4 satisfying (5.5). we ‘thus have om store Et hy Bon “a ays Go Bees hf ee et nck(x? * a(ayam 4 We now treat the inner su in (5.6) by partial sunnatior Gn De 2B ope ab he a ee act Wo thus shall be interested in obtaining an upper bound for A(tym) , the mmber of ast with 1(d) =m. Momma 5.2. As te, A(tym) < t/E(t)¥6(2) unitormly forall a. Berore we prove the lemma, we show hov tne eheorem follows from it. Prom (5.7) and the Lemna, ee eee Big BE cig 9 Le Be sem uy) uniformly for all m. Putting this in (5.6) imediately gives c*(x) < x/L(x)##(2) , which as ve have noted, ie sufficient for the theoren. ‘To prove the Leama ve resort to the trick used in the previous sections. Note that ve may astune mst , for otherwise (tym) = 0. We have for any ¢> 0, Mem se Dees Fee 1a\-— pldep-tim =e TT Pele a-pet. Wo now have an expression that is essentially the sane the right side of (4.1), ¢o that the lemma follove from the rest of the proof of Theorem 4.1 Although ve st{11 do not know if there are infinitely many Carmichael numbers, probably Theorem 5.1 is close to beat possible. An elaharatian of = henrtatic argiment given by Erdos in [9], suggests that C(x) 2 x/L(x)1*=(2) ‘This argunent is very similar to the proof of Theorem 4.47 ve now sketch it. Let PY(n) denote the largest prine power factor of n. If M’(x) is the number of primes pcx with Pr(p - 1) ¢ ofted 1972 ‘/(209 x)¥/2 , we conjecture, analogously to ypothesis 2, ent te.8) won) = avec + 6029) G09 0) ¥7este9 0 Tt 4s not hard to show fron Theorem 2.1 that the number of nsx with Prin) < 22°97) c109 43/2 gatioties the ‘sane estinate, so that (5.8) te perhaps a reasonable conjecture. ‘As in the two previous sections, let ¢ = loglog x . tet A= A(%) denote the least comon multiple of the integers up to (Log x)/loglog x (=et/2) . Tf a is the set Of products of (10g x)/(logleg x)?} “distinct prines P with tog xs psec and p~- aia, then from (5.9) {and the argunent in section 4 (see (4.3)) we have la] exept), Wote that every member m of 4 is composite, mc x, (ua) = 1, and ACm IA We now make a second heuristic assumption. we conjecture that the menbers of 4 are approximately uniformly distributed anong the residue classes nod A that are coprine to A. If 20, ve would expect about [AIA embers m of X to satiety m=1 (nod a), But AS exp(o(iog x/logieg x) = 1(x)2() , 0 we are conjecturing there are at least x/1(x)20(2) menbors m of 4 with m=1 (med A). But since mca implies (m)IA , each such m is a carnichael munber. Tt is possible to shov there are infinitely many Peewdoprines to the base a for any fixed a. the best Feouit of this sort is in (22}: for any fixed a> 0, B/ (BE) eq), Pao) > exp (tog x) where Ee given in Derinition 4.5. Thus, for example, using Priedlander’s result that E21 - (2/e)"), we have Pa(s) > exp((log 2)85/207) for x 2 x9 - Again, the conjecture {s that for each a 4 0,11 we have aCe) = x/bG) OC), Better results can be proved if we average over a. In [21], it is shown enat ws) tea dS Pato) 5 x/uQ42O) ‘Te upper bound proof is similar to the proof of Theoren 5.1 and the lover bound proof is similar to the proof of Theoren 4.6. Assuning Hypothesis 4.2 it is possible to show that ‘the upper Bound is sharp. Finally ve note that the upper bound in (5.9) actually has a "real world" application, as noted in [2]. Wanely, if you would 1ike to quickly find a random prine number y ', Nederl. Akad. Wetensch, Proc Ser. A 54(1951), 50-60 11, 69(1966), 239-247. E.R. Canfield, P. Erdis, and c. Ponerance, ‘on a problem of Oppenheim concerning "Factorisatio Nunerorun*’, J, Munbex Theory 37(1983), 1-28. .V. Ennola, ‘On numbers with small prine divisors’, ‘Ann. Acad. Sci. Fenn. Sex. AT (1969) Wo. 440, 16 pp y’, J. lumber Theory 22(1986), 269-207. ‘A Wildebrand and ¢. Tenenbaum, ‘on integers free of large prine factors’, Trans. Aner. Math, Soc, 296 (1986), 265-290. B.A, MacMahon, "The enuneration of the partitions of multipartite numbers’, Proc, cambridge Philos, Soc. 22(1925), 951-963, 4H, Maier, ‘on integers tree of large prine factors’, lunpublished manuscript. A, Oppenheim, ‘on an arithnetic function’, J. London Math. Sos, 2(2926), 205-211; 11 (1927), 123-130. 2. c. Ponerance, ‘Popular values of Euler’s function’, Mathenatixa 27(1980), 24-29. cc. Ponerance, ‘On the distribution of psoudoprines’, wath. comp. 37(1981), 587-593. c. Pomerence, ‘A new lover bound for the pseudoprine counting function’, ILlinois J. Math. 26(1982), 4-9. c. Ponerance, J.t. Selfridge, and 5.8. wagstaft, or. ‘me peeudoprines to 251097, Math, Cop. 35(2980), R.A, Rankin, ‘The difference between consecutive prime numbers’, J. London Wath. Soc, 13(1938), 242-247. W. Specht, ‘Zahlenfolgen mit endiich vielen Printeilern’, Bayer. Akad, Wiss. Math, = Natur. Abt. S.cB, 1940(2949), 149-169. G. Tenenbaum, ‘La méthode du col en théorie analytique des nombres’, Siminaixe de Théorie des Yonbres de Paris 986-87 (Birknbuser), «12-442. K.R, Wooleidge, ‘Values taken many tines by Buler’s phi-function’, eros, Aner. Wath, Soc. 76(1979),

You might also like