DOI 10.1515/CRELLE.2011.139 angewandte Mathematik
(Walter de Gruyter BerlinNew York 2011
Explicit uniform estimation of rational points II. Hypersurface coverings
ByHuayi Chenat Paris
Abstract. We obtain an explicit uniform estimate for the number of rational points in a projective plane curve whose heights do not exceed the degree of the curve.
1. Introduction
This article is a continuation of [12]. LetK be a number field andX be a sub-variety ofPKn of dimensiond and of degreed. The purpose of this article is to establish the follow- ing explicit estimate (see Theorem 4.2):
Theorem A. Lete>0and D be an integer such that D>max
ðe1þ1Þ
2d1dðdþ1Þ þd2
;2ðndÞðd1Þ þdþ2
:
There is an explicitly computable constant C ¼Cðe;d;n;d;KÞsuch that,for any Bfee,the set S1ðX;BÞ of regular rational points of X with exponential heighteB is covered by not more than CBð1þeÞd
1
dðdþ1Þ hypersurfaces of degreeeD not containing X .
This theorem generalizes some results of Heath-Brown [16], Theorem 14, and Bro- berg [6], Theorem 1, in the sense that we estimate explicitly the degree and the number of the auxiliary hypersurfaces needed to cover the set of rational points with bounded height.
The strategy of Heath-Brown in the proof of [16], Theorem 14, consists of establish- ing that a family of rational points having the same reduction modulo a ‘‘large’’ prime number are contained in one hypersurface (not containing X) with ‘‘low’’ degree. This idea is inspired by results of Bombieri–Pila [1] and Pila [22], and has been developed later in [6], [7], [8], [9], [10], [14], [17], [18], [23], [24].
Suggested by Bost, we adapt the above idea into the framework of his slope method [2], [3], [4]. Note that Bogomolov has asked a similar question on the possibility of replac- ing the method of Heath-Brown by arguments in Arakelov geometry (see [13], Question 34). We consider the evaluation map from the space of homogeneous polynomials to the space of values of these polynomials on a family of rational points. If the rational points
in the family have the same reduction modulo some finite placepofK such that the norm ofpis big, then the (logarithmic) height of this evaluation map is very negative. Hence by the slope inequality, the evaluation map cannot be injective and thus we obtain a non-zero homogeneous polynomial whose image by the evaluation map vanishes. The desired hyper- surface is obtained as the zero locus of the homogeneous polynomial.
The flexibility of the geometric framework (see Theorem 3.1) permits us to develop several interesting variants. For example, instead of considering the reduction modulo a fi- nite placep, we treat the case where the family of rational points has the same reduction modulo some power ofp. In other words, we can take a finite placepwith relatively lower norm and consider a family of rational points whosep-adic distances are very small. Such a family is contained in a hypersurface of lower degree. This argument permits us to prove that the constantC figuring in Theorem A depends on the degree ofK overQbut not on the discriminant. Another variant consists of taking into account the local Hilbert–Samuel functions of the variety, which generalizes a result of Salberger [23], Theorem 3.2. This per- mits us to sharpen the constantC in Theorem A in the case whereX is a plane curve andB is small. As a consequence, we obtain the following result.
Theorem B. Assume that X is an integral plane curve of degreed. Then,for anye>0, one has
KSðX;dÞfK d2þe:
This gives an answer to a question of Heath-Brown [13], Question 27.
To obtain an explicit upper bound for the number and the degree of the auxiliary hy- persurfaces, we need several e¤ective estimates in algebraic geometry and in Arakelov ge- ometry, which shall be recalled in the second section. In the third section, we explain the conditions which ensure that a family of rational points lies in the same hypersurface of low degree. Finally, in the fourth section, we estimate the number of hypersurface needed to cover rational points; in the fifth section, we discuss the plane curve case.
We keep Notation 1–8 introduced in [12], §2. Remind thatK denotes a number field andOK denotes its integer ring. We shall also use the following notation.
Notation. 9. Denote by nANnf0gan integer and by E thetrivialHermitian vector bundle of ranknþ1. In other words,E ¼OlKðnþ1Þ, and for any embeddings:K !C, the canonical basis ofE is an orthonormal basis ofk ks. See Notation 4 for the notion of Her- mitian vector bundles.
10. Denote by L the universal quotient sheaf on POnK ¼PðEÞ, equipped with the Fubini–Study metrics.
11. Any point P¼ ðx0: . . . :xnÞAPnðKÞ gives rise to a uniqueOK-pointP APðEÞ.
The height of P (with respect to L) is by definition the slope (see Notation 6) ofPðLÞ, denoted byhðPÞ. Note that one has
hðPÞ ¼ 1
½K :Q
P
pASpmOK
log max
1eienjxijpþ1 2
P
s:K!C
logPn
j¼0
jxjjs2
:
See Notation 2 for the definition of the absolute values j jp and j js. Define HðPÞ:¼exp
½K :QhðPÞ
. Remind that here the logarithmic height functionhis absolute (i.e., invariant under finite field extensions ofK), while the exponential oneH is relative.
12. For any integerDf1, letEDbe theOK-moduleH0
PðEÞ;LnD
, equipped with the John metricsk ks;J associated to the sup-normk ks;sup. We remind that the sup-norm is defined as follows:
EsAEDn
OK;sC; ksks;sup:¼ sup
xAPsnðCÞ
ksðxÞks: The John normk ks;J is a Hermitian norm onEDn
OK;sCsuch that ksks;supeksks;Je ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
rkðEDÞ
p ksks;sup:
Denote byrðDÞthe rank ofED. One hasrðDÞ ¼ nþD D
.
13. Let X be an integral closed subscheme of PKn ¼PðEKÞ. Letd be the dimension ofX and d be the degree ofX. Recall that one has d¼deg
c1ðLKÞd ½X
. Denote by X the Zariski closure ofX inPðEÞ. The (relative) Arakelov height ofX is denoted byhLðXÞ.
Recall that
hLðXÞ:¼ 1
½K :Qdegdegd
cc^1ðLÞdþ1 ½X :
14. For any integerDf1, letFDbe the saturation (inH0ðX;LjnXDÞ) of the image of the restriction map
hX;D:ED;K ¼H0
PðEK;LnKDÞ
!H0ðX;LjnXDÞ;
namely FD is the largest sub-OK-module of H0ðX;LjnXDÞ containing ImðhX;DÞ and such that FD;K ¼ImðhX;DÞK. We equip FD with the quotient metrics (from the metrics of ED) so that FD becomes a Hermitian vector bundle on SpecOK. Denote by r1ðDÞ the rank ofFD.
15. Let pbe a maximal ideal ofOK with residue field Fp. For any point in XðFpÞ, denote byOx the local ring ofX atxand by mxthe maximal ideal ofOx. Note thatOx is a local algebra overOK;p. Denote byHx:N!Nthe Hilbert–Samuel function ofOx=pOx
(which is the local ring ofXFp atx), namely, HxðkÞ ¼rkFp
ðmx=pOxÞk=ðmx=pOxÞkþ1 : Let
qxðmÞ
mf1 be the increasing sequence of non-negative integers such that the integer kAN appears exactly HxðkÞ times. Let QxðmÞ ¼qxð1Þ þ þqxðmÞ. Denote by mx the multiplicity of the local ringOx=pOx. Recall that one has
HxðkÞ ¼ mx
ðd1Þ!kd1þoðkd1Þ:
16. For any real number B>0, let SðX;BÞ be the subset of XðKÞ consisting of points P such that HðPÞeB (see Notation 11 for the definition of HðÞ). Denote by S1ðX;BÞ the subset of SðX;BÞ of regular points. Define NðX;BÞ and N1ðX;BÞto be the cardinality ofSðX;BÞandS1ðX;BÞrespectively.
17. For any maximal ideal pofOK, xAXðFpÞ and B>0, denote by SðX;B;xÞthe set of pointsPASðX;BÞwhose reduction modulo pisx. Define
S1ðX;B;pÞ ¼ S
xAXðFpÞ xregular
SðX;B;xÞ;
wherexregular means thatxis a regular point ofXFp, or equivalently,Ox=pOxis a regular local ring.
18. More generally, for any maximal ideal pand anyaANnf0g, denote byApðaÞ the Artinian local ringOK;p=paOK;p. For any pointhAXðApðaÞÞ, denote bySðX;B;hÞthe set of points inSðX;BÞwhose reduction modulopacoincides withh. We shall use the fact that
EaANnf0g; ExAXðFpÞ; SðX;B;xÞ ¼ S
hAXðApðaÞÞ x¼ðhmodpÞ
SðX;B;hÞ:
19. We introduce several constants as follows:
C1¼ ðdþ2Þmm^max
SdðE4Þ þ1
2ðdþ2Þlog rkðSdEÞ þd
2 log
ðdþ2ÞðndÞ þd
2ðdþ1Þlogðnþ1Þ;
C2¼r
2log rkðSdEÞ þ1
2log rkðLndEÞ þlog ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðndÞ!
p þ ðndÞlogd;
C3¼ ðndÞC1þC2:
Recall that the constantC1has been defined in [12], (21).1)With the notation of [12], The- orem 3.8, the constantC2is justC2ðE;nd;dÞ(see also Remark 3.9 loc. cit.). Finally, the constantC3appears in [12], Theorem 3.10. Recall that one hasC3fn;dd(see Theorem 3.10 loc. cit.).
20. By the e¤ective version of Chebotarev’s theorem (cf. [20], see also [25], Theorem 2) there exists an explicitly computable constantaðKÞsuch that, for any real numberxf1, there exists a finite place pASf such that NpA
x;aðKÞx
. This is an analogue of Ber- trand’s postulate for number fields.
2. Reminders
We recall in this section several results that we shall use in the sequel. They are either well known or described in [12].
1) SinceSdEis a direct sum of Hermitian line bundles, the quantity%ðdþ2Þ GdðEÞ
vanishes (see [12], §2.2).
Furthermore, whenEis trivial, one hasmm^maxðLndEÞ ¼0.
2.1. LetðPiÞiAI be a collection of distinct rational points ofX (see Notation 13) and Df1 be an integer. Assume that the evaluation map f :FD;K !L
iAI
PiLnDis an isomor- phism (see Notation 14). Then the equality
m^
mðFDÞ ¼ 1 r1ðDÞ
P
iAI
DhðPiÞ þhðLr1ðDÞfÞ
holds. In particular, one has m^ mðFDÞ
D esup
iAI
hðPiÞ þ 1
Dr1ðDÞhðLr1ðDÞfÞ;
ð1Þ
wherehðLr1ðDÞfÞis defined as hðLr1ðDÞfÞ ¼ 1
½K :Q P
p
logkLr1ðDÞfkpþ P
s:K!Q
logkLr1ðDÞfks
:
A slight variant of this argument shows that, ifðPiÞiAJ is a family of rational points ofX such that
sup
iAJ
hðPiÞ<mm^maxðFDÞ
D 1
2logðnþ1Þ;
ð2Þ
then there exists a hypersurface of degreeDinPKn not containingX which contains all ra- tional pointsPi. See [12], Proposition 2.12, for details.
2.2. For any integerDf1, one has the following estimates:
Dþdþ1 dþ1
Ddþdþ1 dþ1
er1ðDÞ:¼rkðFDÞed Dþd d
ð3Þ :
See [11] for the upper bound and [26] for the lower bound.
2.3. For any integerDf2ðndÞðd1Þ þdþ2, one has m^
mðFDÞ
D f d!
dð2dþ2Þdþ1hLðXÞ logðnþ1Þ 2d; ð4Þ
wherehLðXÞis the Arakelov height ofX. See [12], Theorem 4.8 and Remark 4.9, for details.
2.4. SinceFDis a quotient ofED, one has (see Notation 12) m^
mðFDÞfmm^minðEDÞf1
2Dlogðnþ1Þ:
ð5Þ
We refer to [12], Corollary 2.9, for the proof. Note that this bound is much less precise than (4). However, it works for any integerDf1.
2.5. For any integerDf1, one has m^
mðFDÞ D e 1
dhLðXÞ þ1
2logðnþ1Þ:
ð6Þ
See [12], Remark 4.11.
2.6. There exists a Hermitian vector subbundleM ofSðd1ÞðndÞE such that (i) mm^minðMÞfðndÞhLðXÞ C3,
(ii) the subscheme of PðEÞ defined by vanishing of M contains the singular loci of fibres ofXbut not the generic point ofX,
where the constantC3is defined in Notation 19. This result has been proved in [12], Theo- rem 3.10. In particular, the singular locus of X is contained in a hypersurface of degree ðd1ÞðndÞnot containingX.
2.7. Suppose thatPAXðKÞis a regular point andP is theOK-point ofPðEÞextend- ingP. For any maximal idealpofOK, if the reduction ofP modulopis a singular point of XFp, we writeapðPÞ ¼1, else we writeapðPÞ ¼0. We have shown in [12], Proposition 2.11, that, for any real numberN0>0, the following inequality holds:
P
NpfN0
apðPÞeðndÞðd1ÞhðPÞ þ ðndÞhLðXÞ þC3
ðlogN0Þ=½K :Q : ð7Þ
In fact, it su‰ces to apply [12], Proposition 2.11, to the special caseI ¼M, whereM is as in §2.6.
2.8. The following estimates of binomial coe‰cients will be used:
ðNkþ1Þk
k! e N
k e
N ðk1Þ=2k
k! ; Nfkf1:
The second inequality comes from the comparison of the arithmetic and the geometric means:
NðN1Þ ðNkþ1Þe Nþ ðN1Þ þ þ ðNkþ1Þ k
k
:
3. Existence of the auxiliary hypersurface The purpose of this section is to establish the following theorem.
Theorem 3.1. Let S¼ ðpjÞjAJ be a finite family of maximal ideals of OK and ðajÞjAJ AðNnf0gÞJ. For eachpj,let hj be a point inXðApðaÞj Þ(see Notation18)whose reduc- tion modulopis denoted byxj. Assume thatðxjÞjAJ are distinct. Consider a familyðPiÞiAI of
rational points ofXK such that, for any iAI and any jAJ, the reduction of Pi modulopjaj coincides withhj. Assume that(see Notation11, 14and15)
sup
iAI
hðPiÞ<mmðF^ DÞ
D logr1ðDÞ
2D þ 1
½K :Q P
jAJ
Qxj r1ðDÞ
Dr1ðDÞ logNpajj: ð8Þ
Then there exists a section sAED;K which does not vanish identically on XK and such that PiAdivðsÞfor any iAI .
This theorem generalizes a result of Salberger [23], Theorem 3.2, in two aspects. On one hand, we treat projective varieties over a number field; on the other hand, we consider a family of thickenings of points over finite places.
The proof of Theorem 3.1 consists of adapting the idea of Bombieri–Pila and Heath- Brown in the framework of the slope method. Note that Broberg has generalized [16], The- orem 14, to the number field case, which corresponds to the case wherejJj ¼1 andaj¼1 here. However, his method is di¤erent from ours. In fact, the slope method permits us to avoid using Siegel’s lemma. Moreover, in (8), there appears only the degree of the number fieldK but not the discriminant.
The following subsections are devoted to the proof of Theorem 3.1 and to discuss several applications. We first estimate the heights of the determinants of some evaluation maps. This stage is quite similar to the determinant argument of Bombieri–Pila and Heath-Brown. Then we use the slope inequality to obtain the desired result. To apply the theorem, we need explicit estimates of the functionsQxj andr1ðDÞ, which we discuss in the end of this section.
3.1. Estimation of norms.
Lemma 3.2. Let A be a ring and M be an A-module.
(i) If N is a sub-A-module of M such that M=N is generated by q elements, then for any integer mfq,we haveLmM ¼ ðLmqNÞ5ðLqMÞ.
(ii) If M ¼M1IM2I IMiIMiþ1I is a decreasing sequence of sub-A- modules of M such that, for any if1, Mi=Miþ1 is isomorphic to a principal ideal of A, then for any integer rf1,we have
LrM¼M15M25 5Mr:
Proof. (ii) is a consequence of (i). To prove (i), by induction it su‰ces to establish the case wherem¼rþ1. SinceM=N is generated byqelements, we haveLrþ1ðM=NÞ ¼0 (see [5], Chapter III, §7, n3, Proposition 3). Furthermore, since the kernel of the canonical homomorphism of exterior algebrasLM!LðM=NÞis the ideal generated byN (loc. cit.), we obtain thatLrþ1MHN5ðLrMÞ. r
Lemma 3.3. Let k be a field equipped with a non-archimedean absolute value j j,U and V be two k-linear ultranormed spaces of finite rank and j:U !V be a k-linear
homomorphism. Let m be the rank of U. For any integer1eiem,let li ¼ inf
WHU codimW¼i1
kjjWk:
If i>m,letli ¼0. Then for any integer r>0,we have kLrjkeQr
i¼1
li: ð9Þ
Proof. Lete>0 be an arbitrary positive real number. We shall construct a decreas- ing filtration ofU,
U ¼U1lU2l lUm; ð10Þ
such that kjjUikeliþe. By definition, there exists a vector xmAU of norm 1 such that kjðxmÞkelmþe. Suppose that we have chosenUiþ1I IUmsuch thatkjjUjkeljþe for anyiþ1ejem. SinceUiþ1has codimensioniinU, the set of vectorsxAU of norm 1 withkjðxÞkeliþecan not be contained inUiþ1. Pick an elementxi AUnUiþ1of norm 1 withkjðxiÞkeliþe. Let Ui be the linear subspace generated byxi andUiþ1. Since the norm of U is ultrametric, one obtains kjjUikeliþe. By induction we can construct the filtration as announced. By Lemma 3.2, one obtains
kLrjkeQr
i¼1
ðliþeÞ:
Sincee>0 is arbitrary, the proposition is proved. r
3.2. A preliminary result on local homomorphisms. Letpbe a maximal ideal ofOK
and xbe anFp-point of X. Suppose given a familyðfiÞ1eiem of local homomorphisms of OK;p-algebras fromOx(see Notation 15) toOK;p. LetEbe a free sub-OK;p-module of finite type ofOxand let f be theOK;p-linear homomorphismðfijEÞ1eiem:E!OK;pm . As f1 is a homomorphism of OK;p-algebras, it is surjective. Let a be the kernel of f1. One has Ox=aGOK;p. Furthermore, sinceOxis a local ring of maximal ideal mx, one has mxIa.
Moreover, since f1 is a local homomorphism, the equality aþpOx¼mx holds. For any integer jf0,aj=ajþ1is anOx=aGOK;p-module of finite type, and
Fpn
OK;pðaj=ajþ1ÞGða=pOxÞj=ða=pOxÞjþ1Gðmx=pOxÞj=ðmx=pOxÞjþ1:
By Nakayama’s lemma, the rank of aj=ajþ1 over OK;p is equal to the rank of ðmx=pOxÞj=ðmx=pOxÞjþ1overFp, that is,HxðjÞaccording to Notation 15. The filtration
Ox¼a0Ia1I IajIajþ1I ofOxinduces a filtration
F :E¼EXa0IEXa1I IEXajIEXajþ1I ð11Þ
ofEwhose j-th subquotientEXaj=EXajþ1is a freeOK;p-module of rankeHxðjÞ.
Assume that aANnf0gis such that the reductions of fi modulo pa are the same (in other words, the composed homomorphisms Ox!OK;p !OK;p=paOK;p
fi
are the same), then the restriction of f on EXaj has normeNpja. In fact, for any 1eiem, one has
fiðaÞHpaOK;p and hence fiðajÞHpajOK;p.
By Lemma 3.3, we obtain the following result.
Proposition 3.4. Let p be a maximal ideal of OK and xAXðFpÞ. Suppose that ðfiÞ1eiemis a family of localOK;p-linear homomorphisms fromOxtoOK;pwhose reductions modulo pa are the same, where aANnf0g. Let E be a free sub-OK;p-module of finite type ofOxand f ¼ ðfijEÞ1eiem. Then, for any integer rf1,one has
kLrfKkeNpQxðrÞa; ð12Þ
where Np is the degree ofFp over its characteristic field. See Notation 15 for the definition of Qx.
Proof. Consider the filtration (11) above. The restriction of f on EXaj has normeNpja, which implies that (see Notation 15 for the definition ofqx)
WHEinfK
codimW¼j1
kfKjWkeNpqxðjÞa;
where we have used the fact that rkðEXajÞ rkðEXajþ1ÞeHxðjÞ. The inequality (12) then follows from Lemma 3.3. r
3.3. Proof of Theorem 3.1. LetDf1 be an integer. LetFDandr1ðDÞ ¼rkFDbe as in Notation 14. Assume that the section predicted by the theorem does not exist. Then the evaluation map f :FD;K !L
iAI
PiLK is injective. By possibly replacingI by a subset, we may suppose that f is an isomorphism. For any embeddings:K !C, one has
1
r1ðDÞ logkLr1ðDÞfkselogkfkselog ffiffiffiffiffiffiffiffiffiffiffiffi r1ðDÞ
p ;
where the second inequality comes from the definition of metrics of John (see Notation 12).
Furthermore, f is induced by a homomorphism ofOK-modulesFD!L
iAI
PiLnD, where Pi denotes the OK-point of X extending Pi. Hence for any finite place p of K, one has logkLr1ðDÞfkpe0.
Let jAJ. For eachi AI, theOK-pointPi defines a local homomorphism fromOxj to OK;pj which isOK;pj-linear. By taking a local trivialization of Latxj, we identifyFDwith a
sub-OK;pj-module ofOxj. Proposition 3.4 then implies that logkLr1ðDÞfkpjeQxj
r1ðDÞ
logNpajj: We then obtain (see §2.1)
m^ mðFDÞ
D e sup
iAI
hðPiÞ þ 1
2D logr1ðDÞ 1
½K :Q P
jAJ
Qxj
r1ðDÞ
Dr1ðDÞ logNpajj;
which leads to a contradiction. Thus the evaluation homomorphismFD;K !L
iAI
PiLnD is not injective. In other words, there exists a homogeneous polynomial of degree D which is not identically zero onX but vanishes on eachPi.
3.4. Applications. Letpbe a maximal ideal ofOK and xbe a rational point of XFp. Recall (see Notation 15) thatOx denotes the local ring of Xat x,mx denotes its maximal ideal, and the local Hilbert–Samuel function ofxis defined as
HxðkÞ:¼rkFp
ðmx=pOxÞk=ðmx=pOxÞkþ1 :
In some particular cases, the local Hilbert–Samuel function of x can be explicitly estimated.
(i) If x is regular (i.e., the local ring Ox=pOx is regular), then one has HxðkÞ ¼ kþd1
d1
for anykf0.
(ii) Assume that the local ring Ox=pOx is one-dimensional and Cohen–Macaulay (that is, mx=pOx contains a non zero-divisor ofOx=pOx), then by [21], Theorem 1.9, one has HxðkÞemx for any integer kf0, where mx denotes the multiplicity of the local ring Ox=pOx. Moreover, ifkfmx1, then one hasHxðkÞ ¼mx (see [19], Theorem 2).
Proposition 3.5. Letpbe a maximal ideal ofOK,xbe a rational point ofXFp,and r be an integer,rf1.
(i) If theFp-pointxis regular,then(see Notation15for the definition of Qx)
QxðrÞ>ðd!Þ1d d
dþ1r1þ1d dþ3 2dþ2dr:
ð13Þ
(ii) If d ¼1andxis Cohen–Macaulay,then
QxðrÞf r2 2mx r
2mx: ð14Þ
Proof. LetUxbe the partial sum function ofHx. Namely, UxðkÞ:¼Hxð0Þ þ þHxðkÞ:
One has
Qx
UxðkÞ
¼Pk
j¼0
jHxðjÞ:
Moreover, ifrA
Uxðk1Þ;UxðkÞ
, then one hasQx
Uxðk1Þ
eQxðrÞeQx
UxðkÞ .
(i) In the case where xis regular, one has UxðkÞ ¼Pk
j¼0
jþd1 d1
¼ kþd d
ð15Þ : Therefore Qx
UxðkÞ
¼Pk
j¼0
jHxðjÞ ¼Pk
j¼0
j jþd1 d1
¼Pk
j¼0
d jþd1 d
¼d kþd dþ1
:
Letrbe an integer in
Uxðk1Þ;UxðkÞ
. One has QxðrÞ ¼Qx
Uxðk1Þ þk
rUxðk1Þ ð16Þ
¼krþd kþd1 dþ1
k kþd1 d
¼kr kþd dþ1
¼krkþd
dþ1Uxðk1Þ> d
dþ1ðk1Þr;
where in the last inequality, we have used the estimateUxðk1Þ<r. Note that (see §2.8) reUxðkÞ ¼ kþd
d
e
kþ ðdþ1Þ=2d
d!
implies
kfðrd!Þd1 ðdþ1Þ=2:
ð17Þ
Combining with (16), we obtain that (13) holds.
(ii) Assume that d ¼1 and Ox=pOx contains a non zero-divisor, then one has 1eHxðkÞemx for any integer kf1. Let ðakÞkf1 be the increasing sequence of non- negative integers such that the integer 0 appears exactly one time, and other integers appear exactlymxtimes. Note that one hasqxðkÞfak for anykANnf0g. Hence
QxðrÞ ¼ Pr
k¼1
qxðkÞfPr
k¼1
ak ¼mx
2 AðAþ1Þ þ ðAþ1Þðr1mxAÞ
¼ ðAþ1Þðr1Þ mx
2 AðAþ1Þ ¼ ðAþ1Þðr1mxA=2Þ;
whereA¼ r1 mx
. Using the fact that r1
mx mx1
mx eAer1 mx ;
we obtain
QxðrÞf r
mx r1r1 2
f r2 2mx r
2mx: r
Remark 3.6. When d¼1, the estimate (13) is less precise than (14). The reason is that in the last inequality of (16), we have used the estimate Uxðk1Þ<r but not the more precise oneUxðk1Þer1.
Corollary 3.7. Let ðpjÞjAJ be a finite family of maximal ideals ofOK and e>0. For any jAJ, let ajANnf0g, xj AXðFpjÞ be a regular rational point ofXFpj and hj AXðApðajjÞÞ whose reduction modulopj is xj. If
P
jAJ
logNpajjfð1þeÞ
logBþ ½K :Qlogðnþ1Þ
dd1dþ1 d ; ð18Þ
then,for any integer D such that
D>ðe1þ1Þ
dd1ðdþ3Þ=2þd2 ð19Þ ;
there exists a hypersurface of degree D ofT PKn not containing X which contains
jAJ
SðX;B;hjÞ.
Proof. Assume that such hypersurface does not exist. By Theorem 3.1, one has logB
½K :Q fmmðF^ DÞ
D logr1ðDÞ
2D þP
jAJ
Qxj
r1ðDÞ Dr1ðDÞ
logNpajj
½K :Q: ð20Þ
Moreover, sincexj is regular, Proposition 3.5 shows that Qxj
r1ðDÞ
fðd!Þ1d d
dþ1r1ðDÞ1þ1d dþ3
2dþ2dr1ðDÞ:
Hence
Qxj r1ðDÞ
Dr1ðDÞ fðd!Þ1d d dþ1
r1ðDÞ1d
D ðdþ3Þd ð2dþ2ÞD: By a result of Sombra recalled in §2.2, one has (forDfd2)
r1ðDÞf Dþdþ1 dþ1
Ddþdþ1 dþ1
¼Pd
j¼1
Ddþdþ j d
fdðDdþ2Þd
d! :
Combining with (5) and the trivial estimater1ðDÞeðnþ1ÞD, (20) implies logB
½K :Q f1
2logðnþ1Þ 1
2logðnþ1Þ þ d1d d dþ1
Ddþ2
D ðdþ3Þd ð2dþ2ÞD
P
jAJ
logNpajj
½K :Q: Or equivalently
dd1 d
dþ1 P
jAJ
logNpajj
½K :Q logB
½K :Qlogðnþ1Þ
DeP
jAJ
logNpajj
½K :Q dd1 d
dþ1ðd2Þ þ dþ3 2dþ2d
:
By the hypothesis (18), the left side is not less than e
1þed1d d dþ1
P
jAJ
logNpajj
½K :QD;
which implies that
Deðe1þ1Þ
d1dðdþ3Þ=2þd2 :
This contradicts (19). r
Corollary 3.8. Assume thatX is Cohen–Macaulay and d¼1. LetðpjÞjAJ be a finite family of maximal ideals ofOK and e>0. For any j AJ, let ajANnf0g, xj AXðFpjÞ and hj AXðApðajjÞÞwhose reduction modulopj isxj. If
P
jAJ
logNpajj
mxj fð1þeÞ2 d
logBþ ½K :Qlogðnþ1Þ ð21Þ ;
then for any integer D such that
D>ð1þe1Þðd2þd1Þ;
ð22Þ
there exists a hypersurface of degree D ofPnnot containing X which contains T
jAJ
SðX;B;hjÞ.
Proof. The proof is quite similar to that of Corollary 3.7. By Proposition 3.5, one has the estimate
Qxj
r1ðDÞ
Dr1ðDÞ f r1ðDÞ 2mxjD 1
2mxjD:
Assume that the hypersurface does not exist. By Theorem 3.1, one has logB
½K :Qþlogðnþ1ÞfP
jAJ
logNpajj
½K :Q d
2mxjDdþ2
D 1
2mxjD
!
;
or equivalently D
P
jAJ
logNpajj
½K :Q d
2mxj logB
½K :Qlogðnþ1Þ
eP
jAJ
logNpajj
½K :Qmxj
dðd2Þ
2 þ1
2
:
By the assumption (21), one obtains D e
1þe P
jAJ
logNpajj
½K :Q d
2mxj eP
jAJ
logNpajj
½K :Qmxjd22dþ1
2 ;
Deð1þe1Þðd2þd1Þ:
The last formula leads to a contradiction. r
4. Covering rational points by hypersurfaces
In this section, we explain how to suitably cover S1ðX;BÞand SðX;BÞby hypersur- faces of low degree. If p is a maximal ideal ofOK and x is a singular rational point of XðFpÞ, there seems to be no general explicit estimate of the local Hilbert–Samuel func- tionQx.2) The idea of Heath-Brown is to consider only regular points. The di‰culty then comes from the fact that the reduction modulopof a regular pointPin XðKÞis not nec- essarily regular. Hence we need to estimate the ‘‘smallest’’ maximal idealpsuch thatPspe- cializes to a regular point modulop. This has been obtained in [16] and in [6] by using the Jacobian criterion. Here we prove that the singular loci of fibres of X are actually con- tained in a divisor whose degree and height are controlled.
Lemma 4.1. Let N0>0be a real number and r the integral part of the number ðndÞðd1ÞlogBþ
ðndÞhLðXÞ þC3
½K :Q logN0
þ1;
ð23Þ
where the constant C3 is defined in Notation19.Ifp1;. . .;pr are r distinct finite places of K such that NpifN0for any i,then
S1ðX;BÞ ¼ Sr
i¼1
S1ðX;B;piÞ:
Proof. With the notation of §2.7, ifPis a rational point inS1ðX;BÞwhich does not lie in anyS1ðX;B;piÞ, then one hasapiðPÞf1 for anyi¼1;. . .;r. Hence, by §2.7, one has
re P
NpfN0
apðPÞe ðndÞðd1ÞhðPÞ þ ðndÞhLðXÞ þC3
ðlogN0Þ=½K :Q ; which leads to a contradiction. r
Theorem 4.2. Lete>0be an arbitrary positive real number. Let D be an integer such that
D>max
ðe1þ1Þ
d1dðdþ3Þ=2þd2
;2ðndÞðd1Þ þdþ4
:
There exists an explicitly computable constant Cðe;d;n;d;KÞ such that, for any Bfee, the set S1ðX;BÞ is covered by not more than Cðe;d;n;d;KÞBð1þeÞd
1
dðdþ1Þ hypersurfaces of de- gree D not containing X .
Proof. In the first stage, we assume that hLðXÞeð2dþ2Þdþ1
d! d logB
½K :Qþ3
2logðnþ1Þ þ2d
:
2) In the case whereXis Cohen–Macaulay, there are explicit estimates (see for example [27]). However, they are far from optimal.
Let MANnf0g be the least common multiple3) of 1;2;. . .;½K :Q. Let N0Að0;þyÞ be such that
logN0¼ ð1þeÞdd1dþ1 dM
logBþ ½K :Qlogðnþ1Þ :
Letrbe the natural number as in Lemma 4.1. Note that one has
reA1logBþA2
logN0
þ1;
where
A1¼ ðndÞðd1Þ þð2dþ2Þdþ1
d! ðndÞd;
A2¼ ½K :Q
C3þð2dþ2Þdþ1 d! d 3
2 logðnþ1Þ þ2d
:
Recall that the constant C3 is defined in Notation 19. Since we have assumed that logBfe, the value of r is bounded from above by a constant A3 which depends only onM,e,n,d andd:
A3:¼M A1þe1A2
ð1þeÞd1dðdþ1Þ=dþ1:
By Bertrand’s postulate, there exist r distinct prime numbers p1;. . .;pr such that N0epie2iN0for anyiAf1;. . .;rg. We choose, for eachi, a maximal idealpiofOK lying over pi. By Lemma 4.1, one has S1ðX;BÞ ¼ Sr
i¼1
S1ðX;B;piÞ. Note that, for anyi, Npi is a power of piwhose exponent fi dividesM (since fie½K :Q). Letai ¼M=fi.
Letx be an arbitrary regularFpi-point ofXFpi. By Corollary 3.7, we obtain that, for anyhAXðApðaiiÞÞwhose reduction modulopi is x, SðX;B;hÞis contained in a hypersurface of degreeDnot containingX. Note that there exists at mostNpðaii1Þdpoints inXðApðaiiÞÞ(see Notation 15) whose reduction modulopi equalsxi and the cardinal ofXðFpiÞdoes not ex- ceedddNpd
i. HenceS1ðX;B;piÞis covered by at most
ddNpaiid ¼ddpiaifid¼ddpiMde2iMdddN0Md ð24Þ
hypersurfaces of degreeDnot containingX. Therefore, S1ðX;BÞis covered by at most
ddN0MdPr
i¼1
2iMdeddr2rMd
ðnþ1Þ½K:QBð1þeÞd
1 dðdþ1Þ
3) One has 2½K:QeMe½K:Qpð½K:QÞ. See [28], p. 30, for a proof.
such hypersurfaces. So the theorem is proved with the constant Cðe;d;n;d;KÞ ¼ddA32A3Mdðnþ1Þð1þeÞd
1
dðdþ1Þ½K:Q
ð25Þ :
Now we treat the case where logB
½K :Q< d!
dð2dþ2Þdþ1hLðXÞ 3
2 logðnþ1Þ 2d:
By §2.1, inequality (2) and §2.3, we obtain that the setSðX;BÞis contained in a hypersur- face of degreeDinPnwhich does not containX. The theorem is also true in this case. r Corollary 4.3. With the notation of Theorem4.2,assume that d ¼1. For any positive real number Bfee,one has
KSðX;BÞe
Cðe;d;n;d;KÞ þ1
dDBð1þeÞ2=d: ð26Þ
Proof. By Be´zout’s theorem, the intersection of each hypersurface in the conclusion of Theorem 4.2 and X contains at most dD rational points. Hence the corollary follows from Theorem 4.2 (see also §2.6). r
Remark 4.4. (i) Observe that one has A1fn;d d and A2fn;dd and hence A3fn;d;ed1þ1d. Therefore, one has
logCðe;d;n;d;KÞfn;d;K;ed1þd1:
Moreover, the constantCðe;d;n;d;KÞ does not depend on the discriminant ofK (but on the degree ofK overQ).
(ii) The original strategy of Heath-Brown corresponds essentially to the case where ai¼1 for any i. By taking a larger N0, his strategy also allows to obtain an explicit upper bound with the same exponent. However, the choice of maximal ideals forces us to use Bertrand’s postulate for the number fieldK where the discriminant ofK is inevitable, according to a counter-example of Heath-Brown that Browning has communicated to me.
5. The case of a plane curve
In this section, we assume thatX is an integral plane curve (that is,d ¼1 andn¼2).
Note that the model X of X is Cohen–Macaulay since it is a subscheme of PO2K defined by one homogeneous equation. We obtain, for ‘‘small’’ value ofB, an explicit estimate of KSðX;BÞ.
Theorem 5.1. Assume that d ¼1and n¼2. Let D¼ b2ðd2þd1Þc þ1. Then,for any real number BAðe;ed2Þ,one has
KSðX;BÞeC4ðK;BÞdD;
ð27Þ
where
C4ðK;BÞ ¼ ð ffiffiffiffiffiffiffiffiffiffiffi logB
p þ1ÞaðKÞ2 ffiffiffiffiffiffiffiffi
logB
p þ2exp 8logBþ ½K :Qlog 2 ffiffiffiffiffiffiffiffiffiffiffi logB p
" #
þ ðlogBÞ ffiffiffiffiffiffiffiffi
logB
p þ1 d1
d ffiffiffiffiffiffiffiffiffiffiffi logB p
! ffiffiffiffiffiffiffiffiplogB þ1
;
aðKÞbeing the constant introduced in Notation20.
Proof. LetN0Að0;þyÞbe such that
logN0¼4logBþ ½K :Qlog 2 ffiffiffiffiffiffiffiffiffiffiffi logB
p :
Let r¼ d ffiffiffiffiffiffiffiffiffiffiffi logB
p e. Choose a family ðpiÞi¼1r of distinct maximal ideals ofOK such that N0eNpieaðKÞiN0, where aðKÞ is the constant of Bertrand’s postulate introduced in Notation 20. For anyðxiÞi¼1r A Qr
i¼1
XðFpiÞ, let
S
X;B;ðxiÞi¼1r :¼ Tr
i¼1
SðX;B;xiÞ:
Note that one has
SðX;BÞ ¼ Sr
i¼1
S
xAXðFpiÞ mxed= ffiffiffiffiffiffiffiffi
plogB
SðX;B;xÞ
W S
ðxiÞi¼1r AU
r i¼1
XðFpiÞ mxi>d= ffiffiffiffiffiffiffiffi
logB
p S
X;B;ðxiÞi¼1r ð28Þ :
LetpAfp1;. . .;prg. Assume thatxis anFp-point ofXFp whose multiplicitymxsatis- fiesmxed= ffiffiffiffiffiffiffiffiffiffiffi
logB
p . One has
logNp
mx f logN0
d= ffiffiffiffiffiffiffiffiffiffiffi logB
p ¼4
dðlogBþ ½K :Qlog 2Þ:
By Corollary 3.8 (the case wheree¼1 andjJj ¼1), there exists a hypersurface of degreeD not containingX which containsSðX;B;xÞ. Note that the cardinal of the set
Sr
i¼1
fxAXðFpiÞ jmxed= ffiffiffiffiffiffiffiffiffiffiffi logB
p g
does not exceed
Pr
i¼1
KP2ðFpiÞePr
i¼1
Np2
ieraðKÞ2rN02:
ð29Þ
LetiAf1;. . .;rg. By Be´zout’s theorem (see [15], 5–22, p. 115), one has P
xAXðFpiÞ
mxðmx1Þedðd1Þ:
ð30Þ
Hence
KfxAXðFpiÞ jmx>d= ffiffiffiffiffiffiffiffiffiffiffi logB
p geðlogBÞ d1 d ffiffiffiffiffiffiffiffiffiffiffi
logB
p ;
which implies that the number ofr-tubesðxiÞi¼1r A Qr
i¼1
XðFpiÞwithmxifd= ffiffiffiffiffiffiffiffiffiffiffi logB
p does not
exceed
ðlogBÞr d1 d ffiffiffiffiffiffiffiffiffiffiffi
logB p
!r
ð31Þ :
Note that the inequality (30) also implies that mxed for any xAXðFpiÞ. Therefore, if ðxiÞi¼1r is an element in Qr
i¼1
XðFpiÞ, then one has
Pr
i¼1
logNpi
mxi frlogN0
d f4
dðlogBþ ½K :Qlog 2Þ;
where the second inequality comes from the estimate rf ffiffiffiffiffiffiffiffiffiffiffi logB
p . Still by Corollary 3.8, one obtains thatS
X;B;ðxiÞi¼1r
is contained in a hypersurface of degree D not contain- ingX.
By (28), (29) and (31), the set SðX;BÞ is contained in a family of hypersurfaces of degree D not containing X, and the number of the hypersurfaces in the family does not exceed
raðKÞ2rN02þ ðlogBÞr d1 d ffiffiffiffiffiffiffiffiffiffiffi
logB p
!r
eC4ðK;BÞ:
By Be´zout’s theorem the intersection of each hypersurface withX contains at most dDra- tional points. Therefore, we obtain
KSðX;BÞeC4ðK;BÞdD: r
Remark 5.2. The logarithmic of the first summand ofC4ðK;dÞis 8logdþ ½K :Qlog 2
ffiffiffiffiffiffiffiffiffiffi logd
p þ ð2 ffiffiffiffiffiffiffiffiffiffi logd
p þ2ÞlogaðKÞ þlogð ffiffiffiffiffiffiffiffiffiffi logd
p þ1Þf ffiffiffiffiffiffiffiffiffiffi logd
p ðd!yÞ;
while the logarithmic of the second summand is ð ffiffiffiffiffiffiffiffiffiffi
logd p þ1Þ
0
@log logdþlog d1 d ffiffiffiffiffiffiffiffiffiffi
logd p
!1
Af ffiffiffiffiffiffiffiffiffiffi logd
p log logd ðd!yÞ:
Hence there exists a constantMK which only depends onK such that C4ðK;dÞeM
ffiffiffiffiffiffiffi plogd
log logdþ ffiffiffiffiffiffiffi
logd
p
K fK de
for anye>0.
Corollary 5.2. Assume that X is an integral plane curve of degree d. Then, for any e>0,one has
KSðX;dÞfK d2þe:
Acknowledgment. Inspired by a talk of P. Salberger at Go¨ttingen University, J.-B.
Bost suggested me to use the slope method to study the density of rational points and shared with me his personal notes. S. David and P. Philippon kindly explained to me their work on the lower bound of arithmetic Hilbert–Samuel function. During the preparation of this work, I benefited from discussions with D. Bertrand, J.-B. Bost, R. De la Brete`che, T. Browning, A. Chambert-Loir, M. Chardin, D. R. Heath-Brown, H. A. Helfgott, H.
Randriambololona, P. Salberger, C. Soule´ and B. Teissier and received many helpful sug- gestions. I would like to express my deep gratitude to them all. Part of this work has been written during my visit to theInstitut des Hautes E´ tudes Scientifiques. I would like to thank the institute for hospitalities. Finally I am grateful to the referees for their valuable re- marks.
References
[1] E. BombieriandJ. Pila, The number of integral points on arcs and ovals, Duke Math. J.59(1989), no. 2, 337–357.
[2] J.-B. Bost, Pe´riodes et isoge´nies des varie´te´s abe´liennes sur les corps de nombres (d’apre`s D. Masser et G.
Wu¨stholz), Se´minaire Bourbaki, Vol. 1994/1995, Aste´risque237(1996), Exp. No. 795, 4, 115–161.
[3] J.-B. Bost, Algebraic leaves of algebraic foliations over number fields, Publ. Math. Inst. Hautes E´ t. Sci.93 (2001), 161–221.
[4] J.-B. Bost, Evaluation maps, slopes, and algebraicity criteria, Proc. Internat. Congr. Math. (Madrid 2006) II, Euro. Math. Soc. (2007), 537–562.
[5] N. Bourbaki, E´ le´ments de mathe´matique, Alge`bre, Chapitres 1 a` 3, Hermann, Paris 1970.
[6] N. Broberg, A note on a paper by R. Heath-Brown: ‘‘The density of rational points on curves and surfaces’’
[Ann. Math. (2)155(2002), no. 2, 553–595], J. reine angew. Math.571(2004), 159–178.
[7] T. D. BrowningandD. R. Heath-Brown, Counting rational points on hypersurfaces, J. reine angew. Math.
584(2005), 83–115.
[8] T. D. BrowningandD. R. Heath-Brown, The density of rational points on non-singular hypersurfaces, I, Bull.
Lond. Math. Soc.38(2006), no. 3, 401–410.
[9] T. D. BrowningandD. R. Heath-Brown, The density of rational points on non-singular hypersurfaces, II, Proc. Lond. Math. Soc. (3)93(2006), no. 2, 273–303.
[10] T. D. Browning,D. R. Heath-BrownandP. Salberger, Counting rational points on algebraic varieties, Duke Math. J.132(2006), no. 3, 545–578.
[11] M. Chardin, Une majoration de la fonction de Hilbert et ses conse´quences pour l’interpolation alge´brique, Bull. Soc. Math. France117(1989), no. 3, 305–318.
[12] H. Chen, Explicit uniform estimation of rational points I, Estimation of heights, J. reine angew. Math., to appear.
[13] Colliot-The´le`neet al., Report on the Workshop ‘‘Rational and integral points on higher dimensional vari- eties’’, 2002.
[14] J. EllenbergandA. Venkatesh, On uniform bounds for rational points on nonrational curves, Int. Math. Res.
Not. (2005), no. 35, 2163–2181.
[15] W. Fulton, Algebraic curves, An introduction to algebraic geometry, Notes written with the collaboration of Richard Weiss, Math. Lect. Notes Ser., W. A. Benjamin, Inc., New York–Amsterdam 1969.
[16] D. R. Heath-Brown, The density of rational points on curves and surfaces, Ann. Math. (2)155(2002), no. 2, 553–595.
[17] H. A. Helfgott, Power-free values, large deviations and integer points on irrational curves, J. Th. Nombres Bordeaux19(2007), no. 2, 433–472.
[18] H. A. HelfgottandA. Venkatesh, Integral points on elliptic curves and 3-torsion in class groups, J. Amer.
Math. Soc.19(2006), no. 3, 527–550.
[19] D. Kirby, The reduction number of a one-dimensional local ring, J. Lond. Math. Soc. (2)10(1975), no. 4, 471–481.
[20] J. C. LagariasandA. M. Odlyzko, E¤ective versions of the Chebotarev density theorem, in: Algebraic num- ber fields: L-functions and Galois properties (Proc. Sympos., Durham 1975), Academic Press, London (1977), 409–464.
[21] J. Lipman, Stable ideals and Arf rings, Amer. J. Math.93(1971), 649–685.
[22] J. Pila, Density of integral and rational points on varieties, Columbia University Number Theory Seminar (New York 1992) Aste´risque228(1995), 4, 183–187.
[23] P. Salberger, On the density of rational and integral points on algebraic varieties, J. reine angew. Math.606 (2007), 123–147.
[24] P. Salberger, Rational points of bounded height on projective surfaces, Math. Z.258(2008), 805–826.
[25] J.-P. Serre, Quelques applications du the´ore`me de densite´ de Chebotarev, Publ. Math. Inst. Hautes E´ t. Sci.
54(1981), 323–401.
[26] M. Sombra, Bounds for the Hilbert function of polynomial ideals and for the degrees in the Nullstellensatz, Algorithms for algebra (Eindhoven 1996), J. Pure Appl. Algebra117/118(1997), 565–599.
[27] V. SrinivasandV. Trivedi, On the Hilbert function of a Cohen–Macaulay local ring, J. Alg. Geom.6(1997), no. 4, 733–751.
[28] G. Tenenbaum, Introduction a` la the´orie analytique et probabiliste des nombres, 3rd ed., Collection E´ chelles, E´ ditions Belin, Paris 2008.
Institut de Mathe´matiques de Jussieu, Universite´ Paris Diderot—Paris 7, France e-mail: chenhuayi@math.jussieu.fr
Eingegangen 12. November 2009, in revidierter Fassung 17. Januar 2011