Contents lists available atScienceDirect
Journal of Number Theory
www.elsevier.com/locate/jnt
General Section
Small gaps between almost primes, the parity problem, and some conjectures of Erdős on consecutive integers II
DanielA. Goldston, Sidney W. Graham,Apoorva Panidapu, Janos Pintz1, Jordan Schettler∗, CemY. Yıldırım
a r t i c l e i n f o a bs t r a c t
Article history:
Received25April2020 Accepted28June2020 Availableonline16July2020 CommunicatedbyS.J.Miller
Keywords:
Almostprime Smallgaps Erdos Mirsky Divisor
Exponentpattern
Weshowthatforanypositiveintegern,thereissomefixed Asuchthatd(x)=d(x+n)=Ainfinitelyoftenwhered(x) denotesthenumberofdivisorsofx.Infact,weestablishthe stronger result that both x andx+nhave thesame fixed exponent pattern for infinitely many x. Here the exponent pattern of an integer x > 1 is the multiset of nonzero exponentswhichappearintheprimefactorizationofx.
©2020TheAuthor(s).PublishedbyElsevierInc.Thisisan openaccessarticleundertheCCBY-NC-NDlicense (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction
Thispaperisintendedasasequelto[GGPY11] writtenbyfourofthecoauthorshere.
Inthepaper,theyprovedastrongerform oftheErdős-Mirksyconjecturementionedin [EM52] whichstatesthatthere areinfinitelymanypositive integersxsuchthatd(x)= d(x+ 1) whered(x) denotesthenumberofdivisorsofx.Thisconjecturewasfirstproven byHeath-Brownin1984[HB84],butthemethoddidnotrevealthenatureofthesetof
* Correspondingauthor.
E-mailaddress:jordan.schettler@sjsu.edu(J. Schettler).
1 ResearchsupportedbytheNationalResearchDevelopmentandInnovationOffice,NKFIH,K119528.
https://doi.org/10.1016/j.jnt.2020.06.002
0022-314X/©2020TheAuthor(s).PublishedbyElsevierInc. ThisisanopenaccessarticleundertheCC BY-NC-NDlicense(http://creativecommons.org/licenses/by-nc-nd/4.0/).
valuesd(x) forsuchx.Inparticular,onecouldnotconcludethattherewasanyparticular valueAforwhichd(x)=d(x+ 1)=Ainfinitelyoften.In[GGPY11],theauthorsshowed that
d(x) =d(x+ 1) = 24 for infinitely many positive integersx. (0.1) Similarresultswereprovenfor otherrelatedarithmeticfunctionswhichcount numbers ofprimedivisors. Thegoalof thispaper isto establishresults foranarbitrary shiftn, i.e.,d(x)=d(x+n)=Ainfinitelyoftenforsomefixed A.
2. Notationandpreliminaries
Forour purposes,alinear form isan expression L(m)=am+b where a and b are integersand a> 0.We view L both as apolynomial and as a functionin m. We say L isreduced if gcd(a,b)= 1.If K(m)=cm+dis another linearform,then arelation betweenL andK is anequation oftheform |cL·L−cK·K|=nwhere cL, cK,n are allpositive integers.We call cL, cK the relationcoefficients and we call nthe relation value.Wedefine thedeterminant ofLandK asdet(L,K)=|ad−bc|.
Foraprimep,ak-tupleof linearforms L1,L2,. . . ,Lk iscalled p-admissible ifthere isanintegertp suchthat
L1(tp)L2(tp)· · ·Lk(tp)≡0 (modp)
Wesay thatak-tuple of linearforms is admissibleif itis p-admissiblefor everyprime p. Notethatak-tupleoflinearformsisadmissibleiffalltheformsarereducedandthe tupleisp-admissibleforeveryprimep≤k.
AnEr number isapositive integer thatisthe product ofr distinct primes.Several ofthecoauthorshereprovedthefollowing resultonE2-numbersinadmissibletriplesin [GGPY09].Later,FrankThorne[Tho08] obtainedageneralizationforEr-numberswith r≥3.
Theorem1.LetC beany constant.IfL1,L2,L3 isanadmissibletriple oflinearforms, then there are two among them, say Lj and Lk such that both Lj(x) and Lk(x) are E2-numberswithboth primefactors largerthan C forinfinitelymanyx.
The results obtained in this paper will use Theorem 1 above in combination with Theorem2below,aspecialcaseofwhichwasproveninthepreviouspaper[GGPY11].
Weprovideaproofhereofthegeneralversionsinceitcontainsimportantideasrelevant fortherestofthepaper.
Theorem 2 (Adjoining Primes). Assume that Li = aim+bi for i = 1,. . . ,k gives an admissiblek-tuplewith relations |ci,jLi−cj,iLj|=ni,j.We can always“adjoin” prime factorstotherelationcoefficientswithoutchangingtherelationvalues:foreverychoiceof
positiveintegersr1,r2,. . .,rk suchthatgcd(ri,ai)= gcd(ri,det(Li,Lj))= gcd(ri,rj)= 1 whenever i = j, there is an admissible k-tuple of linear forms K1,K2,. . . ,Kk with relations |ci,jriKi−cj,irjKj|=ni,j.
Proof. LetxbeasolutionofthecongruencesLi(x)≡ri (modr2i) for1≤i≤k.Suchan xexistsbytheChineseRemainderTheoremsincegcd(ai,ri)= gcd(ri,rj)= 1.Thisxis uniquemodulor= (r1r2· · ·rk)2.Nowdefineanewk-tupleviaKi(m)=Li(rm+x)/ri. Byconstruction,wehave|ci,jriKi−cj,irjKj|=ni,j,soweonlyneedtocheck thatthis new k-tuple is admissible. Wewill show thatthe newk-tuple is p-admissiblefor every primep.Therearetwo cases.
Case 1:Suppose thatp|r. Sincegcd(ri,rj)= 1 for i=j,we havethatp|r for exactly oneindex.Now
K(0) =L(x)/r≡1 (modr)
so K(0)≡1≡0 (modp).WeclaimthatalsoKi(0)≡0 (modp) wheni=.Suppose, bywayofcontradiction,thatKi(0)≡0 (modp) forsomei=.ThenLi(x)≡0 (modp) sinceri≡0 (modp),butL(x)≡r≡0 (modp),so
det(L, Li) =|aib−abi|=|aiL(x)−aLi(x)| ≡0 (modp),
butthiscontradictstheassumptionthatgcd(r,det(L,Li))= 1.ThusK1(0)· · ·Kk(0)≡ 0 (modp).
Case 2: Now suppose p r. Since L1,. . . ,Lk is admissible, there is an integer tp such that L1(tp)· · ·Lk(tp) ≡ 0 (modp). Choose τp such that rτp+x ≡ tp (modp). Then Li(rτp+x)≡Li(tp)≡0 (modp) andri≡0 (modp) foralli,so
K1(τp)· · ·Kk(τp) =L1(rτp+x)
r1 · · ·Lk(rτp+x)
rk ≡0 (modp).
Letnbeapositiveintegerandwriteitsprimefactorizationasn=pk11pk22· · ·pkjj where the pi are distinct primes with ki >0.Then the exponent pattern of n is themultiset {k1,k2,. . . ,kj}where order doesnotmatter butrepetitionsare allowed.Thevalues of manyimportantarithmeticfunctionsdependonlyontheexponentpatternoftheinput;
suchfunctionsinclude:
d(x) = # of divisors ofx
Ω(x) = # of prime factors (counted with multiplicity) ofx ω(x) = # of distinct prime factors ofx
μ(x) = Möbius function = (−1)ω(x)ifnis squarefree, zero otherwise λ(x) = Liouville function = (−1)Ω(x)
Thus if both x and x+n have the same exponent pattern, then d(x) = d(x+n), Ω(x)= Ω(x+n), ω(x)=ω(x+n), etc. In establishingthe strong form of the Erdős- MirskyConjecture (0.1),theauthorsin[GGPY11] actually provedthefollowing result.
Theorem3.Thereareinfinitelymanypositiveintegersxsuchthatboth xandx+ 1have exponentpattern {2,1,1,1}.
Wewill show thatforany shift n,there are infinitelymany positive integersxsuch thatboth xand x+nhaveafixedsmallexponentpattern.Akeytool fordoingthisis containedinthenextremark.
Remark 4.Suppose we have an admissible triple of forms Li with relations |ci,jLi− cj,iLj| = n. For a given form Li in the triple, we call ci,j and ci,k where {i,j,k} = {1,2,3}thepairofrelationcoefficientsforLiinthetriple.Supposethesepairsofrelation coefficientsforeachforminthetriplehavematchingexponentpatterns,i.e.,ci,jandci,k
havethesameexponentpatternwithany choicesofi,j,ksuchthat{i,j,k}={1,2,3}. We then can choose pairwise coprime integers having any desired exponent pattern whicharerelativelyprimetoalllinearcoefficientsanddeterminants(sincedeterminants of distinct reduced forms are alwaysnonzero). In particular, we canadjoin integers to the relation coefficientsso thatthe new triple has theproperty thatall of its relation coefficientshaveanygivenexponentpattern Pwhichcontainstheexponentpatternsof everyci,j. Henceby Theorem 1, we would then getinfinitely many positive integersx suchthatbothxandx+nhaveexponentpatternP∪ {1,1}.TheproofsofTheorems5 and7belowwillrelyheavilyonthisidea.
3. Shiftswhichareevenornotdivisibleby15
Theorem5. Letnbeapositiveintegerwith2|nor15n.Thenthereareinfinitelymany positiveintegers xsuchthatboth xandx+n haveexponent pattern{2,1,1,1,1}. Proof. Consider the following triple of linear forms:L1 = 2m+n,L2 = 3m+n, and L3= 5m+ 2n.Wehavetherelations
3L1−2L2=n 5L1−2L3=n 3L3−5L2=n
Now define gi = gcd(i,n) and reduce the linearforms: takeL1 =L1/g2, L2 =L2/g3, andL3=L3/g5.Then therelationsbecome
3·g2L1−2·g3L2=n 5·g2L1−2·g5L3=n
3·g5L3−5·g3L2=n
Case 1: Suppose n is even and write n = 2n2. Then g2 = 2, so L1 = m+n2, L2 = (3/g3)m+ 2(n2/g3),andL3= (5/g5)m+ 4(n2/g5).
Subcase 1a:Suppose2|n2.Then
L1(1)L2(1)L3(1)≡13≡0 (mod 2),
so the triple L1, L2, L3 is 2-admissible. Now we check this triple is also 3-admissible (and thereforeadmissible).
•If3n2,then
L1(0)L2(0)L3(0)≡n2(−n2)(n2/g5)≡0 (mod 3).
•If3|n2,theng3= 3,soL1≡m≡ ±L3 (mod 3).Nowchoosem0∈ {1,−1}suchthat L2(m0)≡0 (mod 3).Then
L1(m0)L2(m0)L3(m0)≡m0·L2(m0)·(±m0)≡0 (mod 3).
Here therelation coefficientsmatch inpairsfor agiven form inthetriple and allhave exponent patterns contained in {1,1}, so by appeal to Remark 4 we have a slightly stronger result, namely, there are infinitely many positive integers xsuch thatboth x and x+nhaveexponentpattern{1,1,1,1}.
Subcase 1b:Supposenow2n2.Let
K1=L1(4m+n2)/2 = 2m+n2
K2=L2(4m+n2) = 4· 3 g3
m+ 5· n2 g3
K3=L3(4m+n2) = 4· 5 g5
m+ 9· n2 g5
Ourrelationsthusbecome
22·3K1−2·g3K2=n 22·5K1−2·g5K3=n 3·g5K3−5·g3K2=n
Herethepairsofrelationcoefficientsforeachforminthetriplehavematchingexponent patterns.WewillcheckthatthetripleK1,K2,K3isadmissible.First,wenotethateach form isstillreduced:
K1= 2m+n2
isreducedsince2n2.
K2= 4· 3
g3m+ 5·n2
g3
isreducedsincetheconstanttermisoddand notdivisibleby3 ifg3= 1.
K3= 4· 5
g5m+ 9·n2
g5
isreducedsincetheconstanttermisoddand notdivisibleby5 ifg5= 1.
NextK1K2K3 ≡1 (mod 2), sothetriple isindeed2-admissible. Nowwecheck that thistripleis3-admissible.
•If3n2,theng3= 1,so
K1(−n2)K2(−n2)K3(−n2)≡(−n2)2(n2/g5)≡0 (mod 3)
•If 3| n2, then K1K3 ≡ ±m2 (mod 3). Choose m0 ∈ {1,−1}such thatK2(m0) ≡0 (mod 3).Then
K1(m0)K2(m0)K3(m0)≡ ±(m0)2K2(m0)≡0 (mod 3).
Heretherelationcoefficientsallhaveexponentpatternscontainedin{2,1,1},soadjoin- ingprimesagaingivesusthestatementofthetheorem.
Case2:Nowsupposenisodd,sog2= 1 fromnowon.OurrelationsforLibecome 3L1−2·g3L2=n
5L1−2·g5L3=n 3·g5L3−5·g3L2=n
Ifwelook atthismodulo2,wegetL1 ≡1,L2≡m+ 1,L3≡m.Thusthis tripleisnot 2-admissiblehere.However,we canrestrictm (mod 2) andreduceto get2-admissible.
Todothis,wewrite
M1=L1(2m) = 4m+n M2=L2(2m) = 2· 3
g3
m+ n g3
M3=L3(2m)/2 = 5 g5
m+ n g5
.
ThetripleM1,M2,M3 hasreducedformsandis2-admissiblewith relations
3M1−2·g3M2=n 5M1−22·g5M3=n 2·3·g5M3−5·g3M2=n
Note, however, that the relation coefficients for M3 do not have matching exponent patterns.Wecanremedythis byrestrictingandreducingmodulo3.
Subcase 2a:Suppose3n,sog3= 1.Take
N1=M1(3m+n) = 12m+ 5n N2=M2(3m+n) = 18m+ 7n N3=M3(3m+n)/3 = 5
g5
m+ 2· n g5
Now wegetrelations
3N1−2N2=n 5N1−22·3·g5N3=n 2·32·g5N3−5N2=n
Alltheseformsarereducedandthetripleisstill2-admissiblesinceN1(1)N2(1)N3(1)≡ 13≡0 (mod 2).Infact,thetripleis3-admissible toosince
N1(0)N2(0)N3(0)≡(−n)(n)(−n/g5)≡0 (mod 3).
Heretherelationcoefficientsallhaveexponentpatternscontainedin{2,1,1},soadjoin- ing primes again gives us thestatement of the theorem. In fact, if we also have 5 n here, thentherelation coefficientsallhaveexponentpatterns containedin{2,1}sowe getinfinitelymanypositiveintegersxsuchthatxandx+nbothhaveexponentpattern {2,1,1,1}.
Subcase 2b: Suppose now 3| n,so 5n by ourassumption that 15 n. Westill must factorouta3 fromM3,butdoingso willforceus toalsofactorouta3 fromM1which thentellsustoalsofactorouta5 fromM1tomakeitspairofrelationcoefficientsinthe triplehavematchingexponentpatterns.Thuswewillrestrictmodulo15:writen= 3n3
and take
J1=M1(15m−4n)/15 = 4m−n J2=M2(15m−4n)/(g9/3) = 10· 9
g9
m−23· n g9
J3=M3(15m−4n)/3 = 25m−19n3
where, as indicated above, g9 = gcd(9,n) which is either 3 or 9 in this case. Here we haverelations
32·5J1−2·g9J2=n 3·52J1−22·3J3=n 2·32J3−5·g9J2=n
Alltheformsarereduced(since5n)andthetripleis2-admissiblesinceJ1(0)J2(0)J3(0)
≡13≡0 (mod 2).
Nowwecheckthatthistripleis3-admissible.
•If3n3,theng9= 3,so
J1(−n3)J2(−n3)J3(−n3)≡(−n3)(n3)2≡0 (mod 3).
•If3|n3,theng9= 9 soJ1J3≡m2 (mod 3).Choosem0∈ {1,−1}suchthatJ2(m0)≡ 0 (mod 3).Then
J1(m0)J2(m0)J3(m0)≡(m0)2J2(m0)≡0 (mod 3).
Here therelation coefficients allhaveexponent patterns containedin{2,1,1}(or even in{2,1}inthe casethat9|n), so adjoining primes againgives us thestatementof the theorem.
Remark6.Ifweassumethetwinprimeconjecture,thenforanypositiveintegern,there areprimespandp+ 2 suchthatneitherdivide15n.Inthiscase,wecanusethefollowing triple: L1 = 2m+n, L2 =pm+n(p−1)/2,L3 = (p+ 2)m+n(p+ 1)/2.Buildingoff this triple will show—asin Subcase 2a above—that there are infinitely many positive integers xsuch that xand x+n both haveexponent pattern {2,1,1,1}. We will not includethedetailsheresincewegiveanunconditionalproofofaresultfortheremaining casenotcoveredbyTheorem5.
4. Shiftswhichareoddanddivisible by15
Theorem 7. Let n be a positive integer with 2 n and 15|n. Then there are infinitely manypositiveintegersxsuchboth xandx+nhave exponentpattern {3,2,1,1,1,1,1}. Proof. Byconsideringtheadmissibletriplem,m+4,m+10,wefindthatforanyconstant C there areinfinitelymanypairsof E2 numberseachhavingprimefactorsbigger than C and which areadistance of either4,6, or 10 apart.Inparticular, there are oddE2 numbersq1,q2suchthatgcd(qi,n)= 1 fori= 1,2 andq2=q1+ 2j wherej ∈ {2,3,5}. Thuswemaywriteq1=p1,1p1,2 andq2=p2,1p2,2where p1,1,p1,2,p2,1,andp2,2 areall
distinct primes,noneofwhichdivide2n.There areintegersa,bwith aevenandb odd suchthat−aq2+bq1= 1.Writea= 2a2 anddefinethetripleoflinearforms
L1=q1m+a2n L2= 2q2m+bn L3= 4·j
gm+ (b−a)n g
where g = 1 if j = 2 and g = j otherwise. Now we check that this triple is admissi- ble. Weonlyneedto checkfor2-admissibleand3-admissiblesinceeachform isreduced by construction. The triple is 2-admissible since L1·L2·L3 ≡ L1·1·1 (mod 2). To check thetriple is 3-admissible, choosem0 ∈ {1,−1} withL3(m0)≡0 (mod 3). Then L1(m0)L2(m0)L3(m0)≡(q1m0)(−q2m0)L3(m0)≡0 (mod 3). Moreover,thetriplesat- isfiestherelations
q1L2−2q2L1=n
gq1L3−22jL1=n (7.1)
gq2L3−2jL2=n
However, the pairs of relation coefficients for L1, L2 do not have matching exponent patterns inthetriple, sowewill needtoadjoin primesusing Theorem2. Wewillbreak up theproofinto casesdepending onthevalueof j,butinboth casesweneed tonote thatthepairwisedeterminantsarerelativelyprimetotheintegerswewantto adjoin:
det(L1, L2) =q1bn−2a2nq2=n det(L1, L3) =q1(b−a)n
g −4a2n· j g = n
g det(L2, L3) = 2q2(b−a)n
g −4bn· j
g = 2·n g Case 1:Supposej= 2,sog= 1.
We apply Theorem 2directlywith r1 =p22,1p2,2, r2 =p1,1, andr3 = 1,so we get a newadmissibletripleofforms Ki whichsatisfiesthefollowing relations:
|p21,1p1,2K2−2p32,1p22,2K1|=n
|q1K3−23p22,1p2,2K1|=n
|q2K3−22p1,1K2|=n.
Here the relation coefficients of K1 both have exponent pattern {3,2,1}, the relation coefficients of K2 both have exponent pattern {2,1}, and the relation coefficients of K3 both have exponentpattern {1,1}. Thus by another applicationof Theorem2 via
Remark 4 we can arrange an admissible triple with common relation value n and all relation coefficients having exponent pattern {3,2,1,1,1} (or even {3,2,1,1} in this case).
Case 2: Suppose j = 2, so g = j. We apply Theorem 2 directly with r1 = p2,1, and r2=r3= 1,so weget anew admissibletripleofforms Ki whichsatisfiesthefollowing relations:
|q1K2−2p22,1p2,2K1|=n
|jq1K3−22jp2,1K1|=n
|jq2K3−2jK2|=n.
Here the relation coefficients of K1 both haveexponent pattern {2,1,1}, the relation coefficientsofK2both haveexponentpattern{1,1},and therelation coefficientsof K3 bothhaveexponentpattern{1,1,1}.ThusbyappealtoTheorem2viaRemark4wecan arrange anadmissible triplewith common relation valuen and allrelation coefficients havingexponentpattern{3,2,1,1,1}(or even{2,1,1,1}inthiscase).
Therefore, in either case, there are infinitely many pairs of positive integers both havingexponentpattern{3,2,1,1,1,1,1}whichareadistance ofnapart.
References
[EM52]PaulErdős,LeonMirsky,Thedistributionofvaluesofthedivisorfunctiond(n),Proc.Lond.
Math.Soc.2 (3)(1952)257–271.
[GGPY09]D.A.Goldston,S.W.Graham,J.Pintz,C.Y.Yıldırım,Smallgapsbetweenproductsoftwo primes,Proc.Lond.Math.Soc.98 (3)(2009)741–774.
[GGPY11]D.A. Goldston,S.W.Graham,J.Pintz,C.Y.Yıldırım,Smallgapsbetweenalmostprimes, theparityproblem,andsomeconjecturesofErdősonconsecutiveintegers,Int.Math.Res.
Not.2011 (7)(2011)1439–1450.
[HB84]D.R. Heath-Brown, The divisor function at consecutive integers, Mathematika 31(1984) 141–149.
[Tho08]Frank Thorne, Boundedgaps betweenproductsof primeswithapplications toideal class groupsandellipticcurves,Int.Math.Res.Not.2008 (5)(2008)156.