This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Bounds on the Torsion Subgroup Schemes of Néron–Severi Group Schemes

Hyuk Jun Kweon Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139-4307, USA [email protected] https://kweon7182.github.io/
Abstract.

Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth projective variety defined by homogeneous polynomials of degree d\leq d over an algebraically closed field. Let 𝐏𝐢𝐜X\operatorname{\mathbf{Pic}}X be the Picard scheme of XX. Let 𝐏𝐢𝐜0X\operatorname{\mathbf{Pic}}^{0}X be the identity component of 𝐏𝐢𝐜X\operatorname{\mathbf{Pic}}X. The Néron–Severi group scheme of XX is defined by 𝐍𝐒X=(𝐏𝐢𝐜X)/(𝐏𝐢𝐜0X)red\operatorname{\mathbf{NS}}X=(\operatorname{\mathbf{Pic}}X)/(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red}}. We give an explicit upper bound on the order of the finite group scheme (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} in terms of dd and rr. As a corollary, we give an upper bound on the order of the finite group πe´t1(X,x0)torab\pi^{1}_{\mathrm{\acute{e}t}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}. We also show that the torsion subgroup (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} of the Néron–Severi group of XX is generated by less than or equal to (degX1)(degX2)(\deg X-1)(\deg X-2) elements in various situations.

Key words and phrases:
Néron–Severi group, Castelnuovo–Mumford regularity, Gotzmann number
2010 Mathematics Subject Classification:
Primary 14C05; Secondary 14C20, 14C22
This research was supported in part by Samsung Scholarship and a grant from the Simons Foundation (#402472 to Bjorn Poonen, and #550033).

1. Introduction

In this paper, we work over an algebraically closed base field kk. Although chark\operatorname{char}k is arbitrary, we are mostly interested in the case where chark>0\operatorname{char}k>0. The Néron–Severi group NSX\operatorname{NS}X of a smooth projective variety XX is the group of divisors modulo algebraic equivalence. Thus, we have an exact sequence

0Pic0XPicXNSX0.0\rightarrow\operatorname{\mathrm{Pic}}^{0}X\rightarrow\operatorname{\mathrm{Pic}}X\rightarrow\operatorname{NS}X\rightarrow 0.

Néron [27, p. 145, Théorème 2] and Severi [31] proved that NSX\operatorname{NS}X is a finitely generated abelian group. Hence, its torsion subgroup (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is a finite abelian group. Poonen, Testa and van Luijk gave an algorithm for computing (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} [29, Theorem 8.32]. The author gave an explicit upper bound on the order of (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} [21, Theorem 4.12].

As in [32, 7.2], define the Néron–Severi group scheme 𝐍𝐒X\operatorname{\mathbf{NS}}X of XX by the exact sequence

0(𝐏𝐢𝐜0X)red𝐏𝐢𝐜X𝐍𝐒X0.0\rightarrow(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red}}\rightarrow\operatorname{\mathbf{Pic}}X\rightarrow\operatorname{\mathbf{NS}}X\rightarrow 0.

If chark=0\operatorname{char}k=0, then 𝐏𝐢𝐜0X\operatorname{\mathbf{Pic}}^{0}X is an abelian variety, so 𝐍𝐒X\operatorname{\mathbf{NS}}X is a disjoint union of copies of Speck\operatorname{Spec}k. However, if chark=p>0\operatorname{char}k=p>0, then 𝐏𝐢𝐜0X\operatorname{\mathbf{Pic}}^{0}X might not be reduced, and Igusa gave the first example [18]. Thus, the Néron–Severi group scheme may have additional infinitesimal pp-power torsion.

The torsion subgroup scheme (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} of 𝐍𝐒X\operatorname{\mathbf{NS}}X is a finite commutative group scheme. It is a birational invariant for smooth proper varieties due to [28, p. 92, Proposition 8] and [1, Proposition 3.4]. The first main goal of this paper is to give an explicit upper bound on the order of (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}. Let expabab\exp_{a}b\coloneqq a^{b}. {restatable*}thmNNSbound Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth connected projective variety defined by homogeneous polynomials of degree d\leq d. Then

#(𝐍𝐒X)torexp2exp2exp2expdexp2(2r+6log2r).\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+6\log_{2}r).

One motivation for studying (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} is its relationship with fundamental groups. Recall that if k=k=\mathbb{C}, then

π1(X,x0)torab\displaystyle\pi^{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} H1(X,)tor\displaystyle\simeq H_{1}(X,\mathbb{Z})_{\mathrm{tor}}
Hom(H2(X,)tor,/)\displaystyle\simeq\operatorname{Hom}\!\left(H^{2}(X,\mathbb{Z})_{\mathrm{tor}},\mathbb{Q}/\mathbb{Z}\right)
Hom((NSX)tor,/).\displaystyle\simeq\operatorname{Hom}((\operatorname{NS}X)_{\mathrm{tor}},\mathbb{Q}/\mathbb{Z}).

However, if chark=p>0\operatorname{char}k=p>0, then π1(X,x0)torab\pi^{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} is not determined by (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}}. Therefore, an upper bound on #(NSX)tor\#(\operatorname{NS}X)_{\mathrm{tor}} does not give an upper bound on #πe´t1(X,x0)torab\#\pi^{1}_{\mathrm{\acute{e}t}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}. Nevertheless, π1(X,x0)torab\pi^{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} is isomorphic to the group of kk-points of the Cartier dual of (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} [32, Proposition 69]. Hence, we give an upper bound on #πe´t1(X,x0)torab\#\pi^{1}_{\mathrm{\acute{e}t}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} as a corollary of Section 1. As far as the author knows, this is the first explicit upper bound on #πe´t1(X,x0)torab\#\pi^{1}_{\mathrm{\acute{e}t}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}.

{restatable*}

thmpiBound Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth connected projective variety defined by homogeneous polynomials of degree d\leq d with base point x0X(k)x_{0}\in X(k). Then

#π1e´t(X,x0)torabexp2exp2exp2expdexp2(2r+6log2r).\#\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+6\log_{2}r).

Let π1N(X,x0){\pi^{N}_{1}}(X,x_{0}) be the Nori’s fundamental group scheme [28] of (X,x0)(X,x_{0}). Then we also give a similar upper bound on #π1N(X,x0)torab\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}.

The second main goal of this paper is to give an upper bound on the number of generators of (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}}. If chark\ell\neq\operatorname{char}k, then (NSX)[](\operatorname{NS}X)[\ell^{\infty}] is generated by at most (degX1)(degX2)(\deg X-1)(\deg X-2) elements by [21, Corollary 6.4]. The main tool of this bound is the Lefschetz hyperplane theorem on étale fundamental groups [11, XII. Corollaire 3.5]. Similarly, we prove that the Lefschetz hyperplane theorem on 𝐏𝐢𝐜τX\operatorname{\mathbf{Pic}}^{\tau}X gives an upper bound on the number of generators of (NSX)[p](\operatorname{NS}X)[p^{\infty}]. However, the author does not know whether this is true in general. Nevertheless, Langer [22, Theorem 11.3] proved that if XX has a smooth lifting over W2(k)W_{2}(k), then the Lefschetz hyperplane theorem on 𝐏𝐢𝐜τX\operatorname{\mathbf{Pic}}^{\tau}X is true. This gives a bound on the number of generators of (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}}.

{restatable*}

thmNSGenLifting If XrX\hookrightarrow\mathbb{P}^{r} is the reduction of a smooth connected projective scheme 𝒳W2(k)r\mathcal{X}\hookrightarrow\mathbb{P}_{W_{2}(k)}^{r} over W2(k)W_{2}(k), then (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is generated by (degX1)(degX2)(\deg X-1)(\deg X-2) elements.

In Section 3, we prove that #(𝐍𝐒X)tordimkΓ(𝐇𝐢𝐥𝐛QX,𝒪𝐇𝐢𝐥𝐛QX)\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\dim_{k}\Gamma(\operatorname{\mathbf{Hilb}}_{Q}X,\mathcal{O}_{\operatorname{\mathbf{Hilb}}_{Q}X}) for some explicit Hilbert polynomial QQ. Section 4 bounds dimkΓ(Y,𝒪Y)\dim_{k}\Gamma(Y,\mathcal{O}_{Y}) for an arbitrary projective scheme YrY\hookrightarrow\mathbb{P}^{r} defined by homogeneous polynomials of degree at most dd. Section 5 gives an upper bound on #(𝐍𝐒X)tor\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}. Section 6 gives an upper bound on #π1e´t(X,x0)torab\#\pi_{1}^{\mathrm{\acute{e}t}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} and #π1N(X,x0)torab\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}. Section 7 discusses the Lefschetz-type theorem on 𝐏𝐢𝐜τX\operatorname{\mathbf{Pic}}^{\tau}X. Section 8 proves that (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is generated by less than or equal to (degX1)(degX2)(\deg X-1)(\deg X-2) elements if XX has a smooth lifting over W2(k)W_{2}(k).

2. Notation

Given a scheme XX over kk, let 𝒪X\mathcal{O}_{X} be the structure sheaf of XX. We sometimes denote Γ(X)=Γ(X,𝒪X)\Gamma(X)=\Gamma(X,\mathcal{O}_{X}). Let PicX\operatorname{\mathrm{Pic}}X be the Picard group of XX. If XX is integral and locally noetherian, then ClX\operatorname{\mathrm{Cl}}X denotes the Weil divisor class group of XX. If YY is a closed subscheme of XX, let YX\mathscr{I}_{Y\subset X} be the sheaf of ideal of YY on XX. If XX is a projective space and there is no confusion, then we may let Y=YX\mathscr{I}_{Y}=\mathscr{I}_{Y\subset X}. Given a kk-scheme TT and a kk-algebra AA, let X(T)=Hom(T,X)X(T)=\operatorname{Hom}(T,X) and X(A)=Hom(SpecA,X)X(A)=\operatorname{Hom}(\operatorname{Spec}A,X).

Suppose that XX is a projective scheme in r\mathbb{P}^{r}. Let 𝐇𝐢𝐥𝐛QX\operatorname{\mathbf{Hilb}}_{Q}X be the Hilbert scheme of XX corresponding to a Hilbert polynomial QQ. Let 𝐇𝐢𝐥𝐛PX\operatorname{\mathbf{Hilb}}^{P}X be the Hilbert scheme of XX parametrizing closed subschemes ZZ of XX such that Z\mathscr{I}_{Z} has Hilbert polynomial PP. In particular,

𝐇𝐢𝐥𝐛P(n)X=𝐇𝐢𝐥𝐛(n+rr)P(n)X.\operatorname{\mathbf{Hilb}}^{P(n)}X=\operatorname{\mathbf{Hilb}}_{\binom{n+r}{r}-P(n)}X.

Now, suppose that XX is smooth. Let 𝐂𝐃𝐢𝐯QX\operatorname{\mathbf{CDiv}}_{Q}X be the subscheme of 𝐇𝐢𝐥𝐛QX\operatorname{\mathbf{Hilb}}_{Q}X parametrizing divisors with Hilbert polynomial QQ. If DrD\subset\mathbb{P}^{r} is an effective divisor on XX, let HPD\operatorname{HP}_{D} be the Hilbert polynomial of DD as a subscheme.

Let 𝐏𝐢𝐜X\operatorname{\mathbf{Pic}}X be the Picard scheme of XX. Let 𝐏𝐢𝐜0X\operatorname{\mathbf{Pic}}^{0}X be the identity component of 𝐏𝐢𝐜0X\operatorname{\mathbf{Pic}}^{0}X. Let 𝐏𝐢𝐜τX\operatorname{\mathbf{Pic}}^{\tau}X be the disjoint union of the connected components of 𝐏𝐢𝐜X\operatorname{\mathbf{Pic}}X corresponding to (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}}. Let Pic0X=(𝐏𝐢𝐜0X)(k)\operatorname{\mathrm{Pic}}^{0}X=(\operatorname{\mathbf{Pic}}^{0}X)(k) and PicτX=(𝐏𝐢𝐜τX)(k)\operatorname{\mathrm{Pic}}^{\tau}X=(\operatorname{\mathbf{Pic}}^{\tau}X)(k). Let 𝐍𝐒X=𝐏𝐢𝐜X/(𝐏𝐢𝐜0X)red\operatorname{\mathbf{NS}}X=\operatorname{\mathbf{Pic}}X/(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red}}, and let NSX=PicX/Pic0X\operatorname{NS}X=\operatorname{\mathrm{Pic}}X/\operatorname{\mathrm{Pic}}^{0}X. Given x0X(k)x_{0}\in X(k), let π1e´t(X,x0)\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0}) and π1N(X,x0){\pi^{N}_{1}}(X,x_{0}) be the étale fundamental group and Nori’s fundamental group scheme [28] of (X,x0)(X,x_{0}), respectively. Given a vector space VV, let 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V) be the Grassmannian parametrizing nn-dimensional linear subspaces of VV.

Let SS be a separated kk-scheme of finite type, and let XX be a SS-scheme, with the structure map f:XSf\colon X\rightarrow S. Then given a point tSt\in S, let XtX_{t} be the fiber over tt. Let \mathscr{F} be a quasi-coherent sheaf on XX, then let t\mathscr{F}_{t} be the sheaf on XtX_{t} which is the fiber of \mathscr{F} over tt. Let g:TSg\colon T\rightarrow S be a morphism. Let p1:X×STXp_{1}\colon X\times_{S}T\rightarrow X and p2:X×STTp_{2}\colon X\times_{S}T\rightarrow T be the projections. Then as in as in [19, Definition 9.3.12], for an invertible sheaf \mathscr{L} on XX, let 𝐋𝐢𝐧𝐒𝐲𝐬/X/S\operatorname{\mathbf{LinSys}}_{\mathscr{L}/X/S} be the scheme given by

𝐋𝐢𝐧𝐒𝐲𝐬/X/S(T)={D|\displaystyle\operatorname{\mathbf{LinSys}}_{\mathscr{L}/X/S}(T)=\big{\{}D\,\big{|}\, DD is a relative effective divisor on XT/TX_{T}/T such that
𝒪XT(D)p1p2𝒩 for some invertible sheaf 𝒩 on T}.\displaystyle\text{$\mathcal{O}_{X_{T}}(D)\simeq p_{1}^{*}\mathscr{L}\otimes p_{2}^{*}\mathscr{N}$ for some invertible sheaf $\mathscr{N}$ on $T$}\big{\}}.

Given a set SS and TT, let STS\amalg T be the disjoint union of SS and TT. Let #S\#S be the cardinality of SS. Let \mathbb{N} be the set of nonnegative integers, and <n\mathbb{N}_{<n} be the set of nonnegative integers strictly less than nn. We regard an nn-tuple 𝐚\mathbf{a} on SS as a function 𝐚:<nS\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow S. Thus, the iith component of 𝐚\mathbf{a} is denoted by 𝐚(i)\mathbf{a}(i).

Given a group GG, let GabG^{\mathrm{ab}} be the abelianization of GG. Suppose that GG is abelian. Then let GtorG_{\mathrm{tor}}, G[n]G[n] and G[p]G[p^{\infty}] be the torsion subgroup, nn-torsion subgroup and pp-power torsion subgroup of GG, respectively. For a finite abelian group GG, let GG^{\vee} equals Hom(G,/)\operatorname{Hom}(G,\mathbb{Q}/\mathbb{Z}).

Given a commutative group scheme GG, let GtorG_{\mathrm{tor}}, G[n]G[n] and G[p]G[p^{\infty}] be the torsion subgroup scheme, nn-torsion subgroup scheme, and pp-power torsion subgroup scheme of GG, respectively. Suppose that GG is finite. Then let #G\operatorname{{\#}}G be the order of GG, let GabG^{\mathrm{ab}} be the abelianization of GG, and let GG^{\vee} be the Cartier dual of GG.

Given a finite module MM, let Fitti(M)\operatorname{Fitt}_{i}(M) be the iith fitting ideal of MM [7, Proposition 20.4]. If MM is a graded module, its degree tt part is denoted as MtM_{t}. If MM is a finite k[x0,x1,,xr]k[x_{0},x_{1},\dots,x_{r}]-module, then let M~\widetilde{M} be the coherent sheaf on r\mathbb{P}^{r} associated to MM. Given a monomial order \preceq and an ideal Ik[x0,x1,,xr]I\subset k[x_{0},x_{1},\dots,x_{r}], let in(I)\operatorname{in}_{\preceq}(I) be the initial ideal of II with respect to \preceq. We write an m×nm\times n matrix as

(ai,j)i<m,j<n(a0,0a0,1a0,n1a1,0a1,1a1,n1am1,0am1,1am1,n1).(a_{i,j})_{i<m,j<n}\coloneqq\begin{pmatrix}a_{0,0}&a_{0,1}&\dots&a_{0,n-1}\\ a_{1,0}&a_{1,1}&\dots&a_{1,n-1}\\ \vdots&\vdots&\ddots&\vdots\\ a_{m-1,0}&a_{m-1,1}&\dots&a_{m-1,n-1}\end{pmatrix}.

Let AA and BB be are commutative rings, and f:ABf\colon A\rightarrow B be a homomorphism. Let M=(ai,j)i<m,j<nAm×nM=(a_{i,j})_{i<m,j<n}\in A^{m\times n}. Let imM\operatorname{im}M be the AA-module generated by the columns of MM, and let

f(M)(φ(ai,j))i<m,j<n.f(M)\coloneqq(\varphi(a_{i,j}))_{i<m,j<n}.

Given a linear map μ:AmAl\mu\colon A^{m}\rightarrow A^{l}, let μn(M)\mu^{\oplus n}(M) be the l×nl\times n matrix obtained by applying μ\mu to every column vector of MM.

3. Numerical Conditions

Let ı:Xr\imath\colon X\hookrightarrow\mathbb{P}^{r} be a closed embedding of a smooth connected projective variety XX. Let HXH\subset X be a hyperplane section of XX. The goal of this section is to describe an integer mm such that

(1) #(𝐍𝐒X)tordimkΓ(𝐇𝐢𝐥𝐛HPmHX).\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\dim_{k}\Gamma\left(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X\right).

For a positive integer mm, consider the diagram

𝐇𝐢𝐥𝐛HPmH𝐂𝐃𝐢𝐯HPmHX𝐏𝐢𝐜τX(𝐍𝐒X)tor,\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}\hookleftarrow\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}X\twoheadrightarrow(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}},

in which the middle morphism is yet to be defined. The natural embedding 𝐂𝐃𝐢𝐯HPmHX𝐇𝐢𝐥𝐛HPmHX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\hookrightarrow\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X is open and closed [20, Theorem 1.13]. The natural quotient 𝐏𝐢𝐜τX(𝐍𝐒X)tor\operatorname{\mathbf{Pic}}^{\tau}X\twoheadrightarrow(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} is faithfully flat [5, VIA, Théorème 3.2]. Suppose that there is a faithfully flat morphism 𝐂𝐃𝐢𝐯HPmHX𝐏𝐢𝐜τX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}X for some large mm. Then Γ((𝐍𝐒X)tor)Γ(𝐂𝐃𝐢𝐯HPmH)\Gamma((\operatorname{\mathbf{NS}}X)_{\mathrm{tor}})\hookrightarrow\Gamma(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}) is injective by [10, Corollaire 2.2.8], and Γ(𝐇𝐢𝐥𝐛HPmH)Γ(𝐂𝐃𝐢𝐯HPmH)\Gamma(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}})\twoheadrightarrow\Gamma(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}) is surjective, so we get (1). Such mm will be obtained as a byproduct of the construction of Picard schemes. The theorem and the proof below are a modification of the proof of [19, Theorem 9.4.8].

Theorem 3.1.

Let mm be an integer such that

(2) H1(X,(m))=0H^{1}(X,\mathscr{L}(m))=0

for every numerically zero invertible sheaf \mathscr{L}. Then the morphism given by

(𝐂𝐃𝐢𝐯HPmHX)(T)\displaystyle(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X)(T) (𝐏𝐢𝐜τX)(T)\displaystyle\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}X)(T)
D\displaystyle D 𝒪XT(DmHT)\displaystyle\mapsto\mathcal{O}_{X_{T}}(D-mH_{T})

is faithfully flat.

Proof.

Let 𝐏𝐢𝐜σX\operatorname{\mathbf{Pic}}^{\sigma}X be the union of the connected components of 𝐏𝐢𝐜X\operatorname{\mathbf{Pic}}X parametrizing invertible sheaves having the same Hilbert polynomial as 𝒪X(m)\mathcal{O}_{X}(m). Then for any kk-scheme TT,

(𝐏𝐢𝐜σX)(T)\displaystyle(\operatorname{\mathbf{Pic}}^{\sigma}X)(T) (𝐏𝐢𝐜τX)(T)\displaystyle\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}X)(T)
\displaystyle\mathscr{L} (m)\displaystyle\mapsto\mathscr{L}(-m)

gives an isomorphism. Because of the cycle map

(𝐂𝐃𝐢𝐯HPmHX)(T)\displaystyle(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X)(T) (𝐏𝐢𝐜σX)(T)\displaystyle\rightarrow(\operatorname{\mathbf{Pic}}^{\sigma}X)(T)
D\displaystyle D 𝒪XT(D),\displaystyle\mapsto\mathcal{O}_{X_{T}}(D),

𝐂𝐃𝐢𝐯HPmHX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X can be regarded as a (𝐏𝐢𝐜σX)(\operatorname{\mathbf{Pic}}^{\sigma}X)-scheme. The identity 𝐏𝐢𝐜σX𝐏𝐢𝐜σX\operatorname{\mathbf{Pic}}^{\sigma}X\rightarrow\operatorname{\mathbf{Pic}}^{\sigma}X gives a (𝐏𝐢𝐜σX)(\operatorname{\mathbf{Pic}}^{\sigma}X)-point of 𝐏𝐢𝐜σX\operatorname{\mathbf{Pic}}^{\sigma}X, and it is represented by an invertible sheaf \mathscr{L} on X𝐏𝐢𝐜σXX_{\operatorname{\mathbf{Pic}}^{\sigma}X}. Then 𝐂𝐃𝐢𝐯HPmHX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X is naturally isomorphic to 𝐋𝐢𝐧𝐒𝐲𝐬/X×𝐏𝐢𝐜σX/𝐏𝐢𝐜σX\operatorname{\mathbf{LinSys}}_{\mathscr{L}/X\times\operatorname{\mathbf{Pic}}^{\sigma}X/\operatorname{\mathbf{Pic}}^{\sigma}X} as (𝐏𝐢𝐜σX)(\operatorname{\mathbf{Pic}}^{\sigma}X)-schemes by definition. For every closed point t𝐏𝐢𝐜σXt\in\operatorname{\mathbf{Pic}}^{\sigma}X, t(m)\mathscr{L}_{t}(-m) is numerically zero, so

H1(X×{t},t)=0.H^{1}(X\times\{t\},\mathscr{L}_{t})=0.

Hence, [19, 9.3.10] implies that 𝐋𝐢𝐧𝐒𝐲𝐬/X×𝐏𝐢𝐜σX/𝐏𝐢𝐜σX(𝒬)\operatorname{\mathbf{LinSys}}_{\mathscr{L}/X\times\operatorname{\mathbf{Pic}}^{\sigma}X/\operatorname{\mathbf{Pic}}^{\sigma}X}\simeq\mathbb{P}(\mathcal{Q}) for some locally free sheaf 𝒬\mathcal{Q} on 𝐏𝐢𝐜σX\operatorname{\mathbf{Pic}}^{\sigma}X. Since (𝒬)𝐏𝐢𝐜σX\mathbb{P}(\mathcal{Q})\rightarrow\operatorname{\mathbf{Pic}}^{\sigma}X is faithfully flat, 𝐂𝐃𝐢𝐯HPmHX𝐏𝐢𝐜τX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}X is also faithfully flat. ∎

We now need to find an integer mm satisfying (2). Recall the definition of Castelnuovo–Mumford regularity.

Definition 3.2.

A coherent sheaf \mathscr{F} on r\mathbb{P}^{r} is mm-regular if and only if

Hi(r,(mi))=0H^{i}\left(\mathbb{P}^{r},\mathscr{F}(m-i)\right)=0

for every integer i>0i>0. The smallest such mm is called the Castelnuovo–Mumford regularity of \mathscr{F}.

Mumford [26, p. 99] showed that if \mathscr{F} is mm-regular, then it is also tt-regular for every tmt\geq m.

Definition 3.3.

Let PP be the Hilbert polynomial of some homogeneous ideal Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}]. The Gotzmann number φ(P)\varphi(P) of PP is defined as

φ(P)=inf{m|\displaystyle\varphi(P)=\inf\{m\,|\, Z is m-regular for every\displaystyle\mathscr{I}_{Z}\text{ is $m$-regular for every}
closed subvariety Zr with Hilbert polynomial P}.\displaystyle\text{closed subvariety $Z\subset\mathbb{P}^{r}$ with Hilbert polynomial $P$}\}.

If ZrZ\subset\mathbb{P}^{r} is a projective scheme, then let φ(Z)\varphi(Z) be the Gotzmann number of the Hilbert polynomial of Z\mathscr{I}_{Z}.

Given a homogeneous ideal Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}], Hoa gave an explicit finite upper bound on φ(HPI)\varphi(\operatorname{HP}_{I}) [16, Theorem 6.4(i)]. We will describe mm in terms of Gotzmann numbers of Hilbert polynomials.

Lemma 3.4.

Let \mathscr{L} be a numerically zero invertible sheaf on XX. Then ((d1)codimX)\mathscr{L}((d-1)\operatorname{codim}X) is generated by global sections.

Proof.

See [21, Lemma 3.5 (a)]. ∎

Lemma 3.5.

Let \mathscr{L} be a numerically zero invertible sheaf on XX. Then for every i1i\geq 1, n(d1)codimXn\geq(d-1)\operatorname{codim}X and mmax{φ(nH),φ(X)}m\geq\max\{\varphi(nH),\varphi(X)\}, we have

Hi(X,(mn))=0.H^{i}(X,\mathscr{L}(m-n))=0.
Proof.

There is an effective divisor ZZ on XX such that ZX=(n)\mathscr{I}_{Z\subset X}=\mathscr{L}(-n) by 3.4. Since Z\mathscr{I}_{Z} and X\mathscr{I}_{X} are mm-regular, Hi(r,Z(m))=0H^{i}(\mathbb{P}^{r},\mathscr{I}_{Z}(m))=0 and Hi(r,X(m))=0H^{i}(\mathbb{P}^{r},\mathscr{I}_{X}(m))=0 for all i1i\geq 1. Consider the exact sequence

0XZıZX0.0\rightarrow\mathscr{I}_{X}\rightarrow\mathscr{I}_{Z}\rightarrow\imath_{*}\mathscr{I}_{Z\subset X}\rightarrow 0.

Then its long exact sequence shows that

Hi(X,(mn))=Hi(r,ıZX(m))=0H^{i}(X,\mathscr{L}(m-n))=H^{i}(\mathbb{P}^{r},\imath_{*}\mathscr{I}_{Z\subset X}(m))=0

for every i1i\geq 1. ∎

We are ready to prove the main theorem of this section.

Theorem 3.6.

Assume that n(d1)codimXn\geq(d-1)\operatorname{codim}X and mmax{φ(nH),φ(X)}m\geq\max\{\varphi({nH}),\varphi(X)\}. Then

#(𝐍𝐒X)tordimkΓ(𝐇𝐢𝐥𝐛HPmHX).\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\dim_{k}\Gamma\left(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X\right).
Proof.

The quotient 𝐏𝐢𝐜τX(𝐍𝐒X)tor\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} is faithfully flat. 3.5 and 3.1 give a faithfully flat morphism 𝐂𝐃𝐢𝐯HPmHX𝐏𝐢𝐜τX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}X. As a result, their composition gives an injection

Γ((𝐍𝐒X)tor)Γ(𝐂𝐃𝐢𝐯HPmHX)\Gamma((\operatorname{\mathbf{NS}}X)_{\mathrm{tor}})\hookrightarrow\Gamma\left(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\right)

by [10, Corollaire 2.2.8]. Recall that 𝐂𝐃𝐢𝐯HPmHX\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X is an open and closed subscheme of 𝐇𝐢𝐥𝐛HPmHX\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X. Consequently,

#(𝐍𝐒X)tor\displaystyle\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} =dimkΓ(𝐍𝐒X)tor\displaystyle=\dim_{k}\Gamma(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}
dimkΓ(𝐂𝐃𝐢𝐯HPmHX)\displaystyle\leq\dim_{k}\Gamma\left(\operatorname{\mathbf{CDiv}}_{\operatorname{HP}_{mH}}X\right)
dimkΓ(𝐇𝐢𝐥𝐛HPmHX).\displaystyle\leq\dim_{k}\Gamma\left(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X\right).\qed

Therefore, if we bound dimkΓ(Y,𝒪Y)\dim_{k}\Gamma(Y,\mathcal{O}_{Y}) for a general projective scheme YY, and explicitly describe Hilbert schemes in projective spaces, then we can bound #(𝐍𝐒X)tor\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}.

4. The Dimension of Γ(Y,𝒪Y)\Gamma(Y,\mathcal{O}_{Y})

Let YrY\hookrightarrow\mathbb{P}^{r} be a projective scheme defined by a homogeneous ideal II generated by homogeneous polynomials of degree at most dd. The aim of the section is to give an upper bound on dimkΓ(Y,𝒪Y)\dim_{k}\Gamma(Y,\mathcal{O}_{Y}) in terms of rr and dd. First, we reduce to the case where II is a monomial ideal.

Lemma 4.1.

Let \preceq be a monomial order. Let Y0rY_{0}\hookrightarrow\mathbb{P}^{r} be a projective scheme defined by in(I)\operatorname{in}_{\preceq}(I). Then

dimkΓ(Y,𝒪Y)dimkΓ(Y0,𝒪Y0).\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq\dim_{k}\Gamma(Y_{0},\mathcal{O}_{Y_{0}}).
Proof.

By [15, Corollary 3.2.6] and [15, Theorem 3.1.2], there is a flat family 𝒴Speck[t]\mathcal{Y}\rightarrow\operatorname{Spec}k[t] such that

  1. (1)

    𝒴0Y0\mathcal{Y}_{0}\simeq Y_{0} and

  2. (2)

    𝒴tY\mathcal{Y}_{t}\simeq Y for every t0t\neq 0.

Hence, dimkΓ(Y,𝒪Y)dimkΓ(Y0,𝒪Y0)\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq\dim_{k}\Gamma(Y_{0},\mathcal{O}_{Y_{0}}) by the semicontinuity theorem [13, Theorem 12.8]. ∎

Theorem 4.2 (Dubé [6, Theorem 8.2]).

Let Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}] be an ideal generated by homogeneous polynomials of degree at most dd. Let \preceq be a monomial order. Then in(I)\operatorname{in}_{\preceq}(I) is generated by monomials of degree at most

2(d22+d)2r1.2\left(\frac{d^{2}}{2}+d\right)^{2^{r-1}}.

Hence, we will focus on the case where II is a monomial ideal. Let mm be a monomial and II be a monomial ideal of k[x0,,xr]k[x_{0},\dots,x_{r}]. Then by abusing notation, we denote

(m)I(m)I(m).\frac{(m)}{I}\coloneqq\frac{(m)}{I\cap(m)}.
Definition 4.3.

Let mk[x0,,xr]m\in k[x_{0},\dots,x_{r}] be a monomial, and let Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}] be a monomial ideal. Let M(m,I)M(m,I) be the set of monomials in (m)(I(x0d,,xrd))(m)\setminus(I\cup(x_{0}^{d},\dots,x_{r}^{d})). In particular,

#M(m,I)=dimk(m)I+(x0d,,xrd).\#M(m,I)=\dim_{k}\frac{(m)}{I+(x_{0}^{d},\dots,x_{r}^{d})}.
Definition 4.4.

Given a monomial mk[x0,,xr]m\in k[x_{0},\dots,x_{r}], let μ(m)+1\mu(m)+1 be the number of ii such that ximx_{i}\mid m.

Definition 4.5.

Let II be a monomial ideal with minimal monomial generators m0,m1,,mn1m_{0},m_{1},\dots,m_{n-1}. Then

μ(I)μ(m0)+μ(m1)++μ(mn1).\mu(I)\coloneqq\mu(m_{0})+\mu(m_{1})+\dots+\mu(m_{n-1}).

Suppose that II is generated by monomials of degree at most dd. We want to show that dimkΓ(Y,𝒪Y)dr\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq d^{r}. By induction on μ(I)\mu(I), we will show a slightly stronger proposition: for every monomial mk[x0,,xr]m\in k[x_{0},\dots,x_{r}] whose exponents are at most dd, we have

dimkΓ(r,(m)/I~)#M(m,I)d.\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)\leq\frac{\#M(m,I)}{d}.

Since Γ(Y,𝒪Y)=Γ(r,(1)/I~)\Gamma(Y,\mathcal{O}_{Y})=\Gamma(\mathbb{P}^{r},\widetilde{(1)/I}) and #M(m,I)dr+1\#M(m,I)\leq d^{r+1}, it follows that dimkΓ(Y,𝒪Y)dr\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq d^{r}.

Lemma 4.6.

For some nr+1n\leq r+1, let I=(x0b0,x1b1,,xn1bn1)I=(x_{0}^{b_{0}},x_{1}^{b_{1}},\dots,x_{n-1}^{b_{n-1}}) such that bi>0b_{i}>0 for every i<ni<n. Then

dimkΓ(Y,𝒪Y)={1if n<r,b0bn1if n=r, and0if n=r+1.\dim_{k}\Gamma(Y,\mathcal{O}_{Y})=\begin{cases}1&\text{if $n<r$,}\\ b_{0}\dots b_{n-1}&\text{if $n=r$, and}\\ 0&\text{if $n=r+1$.}\end{cases}
Proof.

Suppose that n<rn<r, and take a nonzero function fΓ(Y,𝒪Y)f\in\Gamma(Y,\mathcal{O}_{Y}). Take any ii and jj such that ni<jrn\leq i<j\leq r. In the open set xi0x_{i}\neq 0, we have f=g/xideggf=g/x_{i}^{\deg g} such that gk[x0,,xr]g\in k[x_{0},\dots,x_{r}] and xigx_{i}\nmid g. In the open set xj0x_{j}\neq 0, we have f=h/xjdeghf=h/x_{j}^{\deg h} such that hk[x0,,xr]h\in k[x_{0},\dots,x_{r}] and xjhx_{j}\nmid h. Therefore, in the open set xixj0x_{i}x_{j}\neq 0, we have

gxjdegh=hxidegg.gx_{j}^{\deg h}=hx_{i}^{\deg g}.

Therefore, degg=degh=0\deg g=\deg h=0. As a result, ff is constant meaning that dimkΓ(Y,𝒪Y)=1\dim_{k}\Gamma(Y,\mathcal{O}_{Y})=1.

If n=rn=r, then dimkΓ(Y,𝒪Y)=b0bn1\dim_{k}\Gamma(Y,\mathcal{O}_{Y})=b_{0}\dots b_{n-1} by Bézout’s theorem. If n=r+1n=r+1, then YY is empty, so dimkΓ(Y,𝒪Y)=0\dim_{k}\Gamma(Y,\mathcal{O}_{Y})=0. ∎

Lemma 4.7.

Let mm be a monomial and II be a monomial ideal such that μ(I)>0\mu(I)>0. Then there are monomials m0m_{0} and m1m_{1}, and monomial ideals I0I_{0} and I1I_{1} such that

M(m,I)\displaystyle M(m,I) =M(m0,I0)M(m1,I1),\displaystyle=M(m_{0},I_{0})\amalg M(m_{1},I_{1}),
dimkΓ(r,(m)/I~)\displaystyle\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right) dimkΓ(r,(m0)/I0~)+dimkΓ(r,(m1)/I1~),\displaystyle\leq\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m_{0})/I_{0}}\right)+\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m_{1})/I_{1}}\right),
μ(I0)\displaystyle\mu(I_{0}) <μ(I) and μ(I1)<μ(I).\displaystyle<\mu(I)\text{ and }\mu(I_{1})<\mu(I).
Proof.

Since μ(I)>0\mu(I)>0, there is a minimal monomial generator nn^{\prime} of II such that μ(n)>0\mu(n^{\prime})>0. Let JJ be the ideal generated by the other minimal monomial generators of II. Then I=(n)+JI=(n^{\prime})+J and nJn^{\prime}\not\in J. We may assume that n=xisnn^{\prime}=x_{i}^{s}n where s>0s>0 and xinx_{i}\nmid n. Grouping the monomials in M(m,I)M(m,I) according to whether the exponent of xix_{i} is s\geq s or <s<s yields

M(m,I)\displaystyle M(m,I) =M(lcm(xis,m),(n)+J)M(m,(xis)+J).\displaystyle=M\left(\operatorname{lcm}(x_{i}^{s},m),(n)+J\right)\amalg M\left(m,(x_{i}^{s})+J\right).

Moreover, the short exact sequence

0(lcm(xis,m))(n)+J(m)I(m)(xis)+J0.0\rightarrow\frac{\left(\operatorname{lcm}(x_{i}^{s},m)\right)}{(n)+J}\rightarrow\frac{(m)}{I}\rightarrow\frac{(m)}{(x_{i}^{s})+J}\rightarrow 0.

implies that

dimkΓ(r,(m)/I~)dimkΓ(r,(lcm(xis,m))(n)+J~)+dimkΓ(r,(m)(xis)+J~).\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)\leq\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{\frac{(\operatorname{lcm}(x_{i}^{s},m))}{(n)+J}}\right)+\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{\frac{(m)}{(x_{i}^{s})+J}}\right).

Finally, we have

μ((n)+J)μ(n)+μ(J)<μ(n)+μ(J)=μ(I)\mu((n)+J)\leq\mu(n)+\mu(J)<\mu(n^{\prime})+\mu(J)=\mu(I)

and similarly μ((xis)+J)<μ(I)\mu((x_{i}^{s})+J)<\mu(I). ∎

Lemma 4.8.

Let mk[x0,,xr]m\in k[x_{0},\dots,x_{r}] be a monomial whose exponents are at most dd. Let Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}] be an ideal generated by monomials of degree at most dd. Then

dimkΓ(r,(m)/I~)#M(m,I)d.\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)\leq\frac{\#M(m,I)}{d}.
Proof.

This will be shown by induction on μ(I)\mu(I). If μ(I)<0\mu(I)<0, then I=(1)I=(1), so

dimkΓ(r,(m)/I~)=0.\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)=0.

Suppose that μ(I)=0\mu(I)=0. Then by changing coordinates, we may assume that for some nr+1n\leq r+1,

I=(x0b0,x1b1,,xn1bn1) and m=x0a0x1a1xrarI=(x_{0}^{b_{0}},x_{1}^{b_{1}},\dots,x_{n-1}^{b_{n-1}})\text{ and }m=x_{0}^{a_{0}}x_{1}^{a_{1}}\dots x_{r}^{a_{r}}

such that 0<bid0<b_{i}\leq d for every i<ni<n, and 0ajd0\leq a_{j}\leq d for every jrj\leq r. If aibia_{i}\geq b_{i} for some i<ni<n, then mIm\in I, so (m)/I=0(m)/I=0. Thus, we may assume that ai<bida_{i}<b_{i}\leq d for every i<ni<n. Let ı:Yr\imath\colon Y\hookrightarrow\mathbb{P}^{r} be the projective scheme defined by II. Then (m)/I~\widetilde{(m)/I} is a subsheaf of ı𝒪Y\imath_{*}\mathcal{O}_{Y}. Hence, if n<rn<r, then 4.6 implies that

dimkΓ(r,(m)/I~)1.\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)\leq 1.

Suppose that n=rn=r. Since the open set defined by xr0x_{r}\neq 0 contains YY, multiplying by xrdegmx_{r}^{\deg m} gives an isomorphism

×xrdegm:(m)I~(m)I~(degm).\times x_{r}^{\deg m}\colon\widetilde{\frac{(m)}{I}}\rightarrow\widetilde{\frac{(m)}{I}}(\deg m).

Moreover, multiplication by m1m^{-1} gives an isomorphism

×m1:(m)I(degm)k[x0,,xr](x0b0a0,,xr1br1ar1).\times m^{-1}\colon{\frac{(m)}{I}}(\deg m)\rightarrow{\frac{k[x_{0},\dots,x_{r}]}{(x_{0}^{b_{0}-a_{0}},\dots,x_{r-1}^{b_{r-1}-a_{r-1}})}}.

Hence, by 4.6,

dimkΓ(r,(m)/I~)=(b0a0)(b1a1)(br1ar1).\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)=(b_{0}-a_{0})(b_{1}-a_{1})\dots(b_{r-1}-a_{r-1}).

Suppose that n=r+1n=r+1. Then since YY is empty,

dimkΓ(r,(m)/I~)=0.\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)=0.

Because

#M(m,I)=(b0a0)(b1a1)(bn1an1)drn+1,\#M(m,I)=(b_{0}-a_{0})(b_{1}-a_{1})\dots(b_{n-1}-a_{n-1})d^{r-n+1},

we have Γ(r,(m)/I~)#M(m,I)/d\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right)\leq\#M(m,I)/{d} in every case.

Now, suppose that μ(I)>0\mu(I)>0 and assume the induction hypothesis. Then we have m0m_{0}, m1m_{1}, I0I_{0} and I1I_{1} as in 4.7. Thus,

Γ(r,(m)/I~)\displaystyle\Gamma\left(\mathbb{P}^{r},\widetilde{(m)/I}\right) Γ(r,(m0)/I0~)+Γ(r,(m0)/I0~)\displaystyle\leq\Gamma\left(\mathbb{P}^{r},\widetilde{(m_{0})/I_{0}}\right)+\Gamma\left(\mathbb{P}^{r},\widetilde{(m_{0})/I_{0}}\right)
#M(m0,I0)d+#M(m1,I1)d (by the induction hypothesis)\displaystyle\leq\frac{\#M(m_{0},I_{0})}{d}+\frac{\#M(m_{1},I_{1})}{d}\text{ (by the induction hypothesis)}
=#M(m,I)d.\displaystyle=\frac{\#M(m,I)}{d}.\qed
Corollary 4.9.

Let YrY\hookrightarrow\mathbb{P}^{r} be a projective scheme defined by monomials of degree at most dd. Then

dimkΓ(Y,𝒪Y)dr.\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq d^{r}.
Proof.

Let II be the ideal generated by the defining monomials of YY. Then

dimkΓ(Y,𝒪Y)\displaystyle\dim_{k}\Gamma(Y,\mathcal{O}_{Y}) =dimkΓ(r,(1)/I~)\displaystyle=\dim_{k}\Gamma\left(\mathbb{P}^{r},\widetilde{(1)/I}\right)
#M(1,I)d (by 4.8 with m=1)\displaystyle\leq\frac{\#M(1,I)}{d}\text{ (by \lx@cref{creftypecap~refnum}{lem:stronger monomial bound} with $m=1$)}
=dr.\displaystyle=d^{r}.\qed

Now, we are ready to prove the main theorem of this section.

Theorem 4.10.

Let YrY\hookrightarrow\mathbb{P}^{r} be a projective scheme defined by homogeneous polynomials of degree at most dd. Then

dimkΓ(Y,𝒪Y)2r(d22+d)r2r1\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq 2^{r}\left(\frac{d^{2}}{2}+d\right)^{r2^{r-1}}
Proof.

This follows from 4.1, 4.2 and 4.9. ∎

The artinian scheme SpecΓ(Y,𝒪Y)\operatorname{Spec}\Gamma(Y,\mathcal{O}_{Y}) satisfies the universal property: every morphism f:YGf\colon Y\rightarrow G to an artinian scheme GG uniquely factors through the natural morphism p:YSpecΓ(Y,𝒪Y)p\colon Y\rightarrow\operatorname{Spec}\Gamma(Y,\mathcal{O}_{Y}) [13, Exercise II.2.4]. Thus, YY and SpecΓ(Y,𝒪Y)\operatorname{Spec}\Gamma(Y,\mathcal{O}_{Y}) has the same number of connected components. Andreotti–Bézout inequality [4, Lemma 1.28] then implies that dimkΓ(Y,𝒪Y)reddr\dim_{k}\Gamma(Y,\mathcal{O}_{Y})_{\mathrm{red}}\leq d^{r}. Thus, we may expect that dimkΓ(Y,𝒪Y)dr\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq d^{r}, and this is true if YY is defined by monomials by 4.9.

Question 4.11.

Let YrY\hookrightarrow\mathbb{P}^{r} be a projective scheme defined by homogeneous polynomials of degree at most dd. Is

dimkΓ(Y,𝒪Y)dr?\dim_{k}\Gamma(Y,\mathcal{O}_{Y})\leq d^{r}?

One the other hand, Mayr and Meyer [25] constructed a family of ideals whose Castelnuovo–Mumford regularities are doubly exponential in rr. Thus, the doubly exponential upper bound on dimkΓ(Y,𝒪Y)\dim_{k}\Gamma(Y,\mathcal{O}_{Y}) might be unavoidable.

5. Hilbert Schemes

Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth connected projective variety defined by polynomials of degree at most dd. The aim of this section is to give an explicit upper bound on #(𝐍𝐒X)tor\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}. 3.6 implies that it suffices to give an upper bound on dimkΓ(𝐇𝐢𝐥𝐛PX)\dim_{k}\Gamma(\operatorname{\mathbf{Hilb}}^{P}X) for a polynomial PP. Gotzmann explicitly described 𝐇𝐢𝐥𝐛Pr\operatorname{\mathbf{Hilb}}^{P}\mathbb{P}^{r} as a closed subscheme of a Grassmannian [9][17, Proposition C.29]. Let Rnk[x0,,xr]nR_{n}\coloneqq k[x_{0},\dots,x_{r}]_{n} for nn\in\mathbb{N}.

Theorem 5.1 (Gotzmann).

Let tφ(P)t\geq\varphi(P) be an integer. Then there exists a closed immersion given by

(𝐇𝐢𝐥𝐛Pr)(A)\displaystyle\left(\operatorname{\mathbf{Hilb}}^{P}\mathbb{P}^{r}\right)(A) 𝐆𝐫(P(t),Rt)(A)\displaystyle\rightarrow\operatorname{\mathbf{Gr}}(P(t),R_{t})(A)
[Z]\displaystyle[Z] Γ(Ar,Z(t))\displaystyle\mapsto\Gamma(\mathbb{P}^{r}_{A},\mathscr{I}_{Z}(t))

for every kk-algebra AA.

Therefore, we have closed immersions

𝐇𝐢𝐥𝐛PX𝐇𝐢𝐥𝐛Pr𝐆𝐫(P(t),Rt)(P(t)Rt)\operatorname{\mathbf{Hilb}}^{P}X\hookrightarrow\operatorname{\mathbf{Hilb}}^{P}\mathbb{P}^{r}\hookrightarrow\operatorname{\mathbf{Gr}}(P(t),R_{t})\hookrightarrow\mathbb{P}\!\left({\textstyle\bigwedge^{P(t)}}R_{t}\right)

where 𝐇𝐢𝐥𝐛PX𝐇𝐢𝐥𝐛Pr\operatorname{\mathbf{Hilb}}^{P}X\hookrightarrow\operatorname{\mathbf{Hilb}}^{P}\mathbb{P}^{r} is the natural embedding and 𝐆𝐫(P(t),Rt)(P(t)Rt)\operatorname{\mathbf{Gr}}(P(t),R_{t})\hookrightarrow\mathbb{P}(\bigwedge^{P(t)}R_{t}) is the Plücker embedding. We will bound the degree of defining equations of 𝐇𝐢𝐥𝐛PX\operatorname{\mathbf{Hilb}}^{P}X in (P(t)Rt)\mathbb{P}(\bigwedge^{P(t)}R_{t}). Then 4.10 will gives an upper bound on dimkΓ(𝐇𝐢𝐥𝐛PX)\dim_{k}\Gamma(\operatorname{\mathbf{Hilb}}^{P}X). For simplicity, let

Q(n)dimRnP(n)=(n+rr)P(n).Q(n)\coloneqq\dim R_{n}-P(n)=\binom{n+r}{r}-P(n).

The theorem below is a refomulation of the work in [17, Appendix C].

Theorem 5.2.

Let tmax{φ(P),d}t\geq\max\{\varphi(P),d\} be an integer. Let AA be an kk-algebra, and let Sn=A[x0,,xr]nS_{n}=A[x_{0},\dots,x_{r}]_{n}. Then M𝐆𝐫(P(t),Rt)(A)M\in\operatorname{\mathbf{Gr}}(P(t),R_{t})(A) is in (𝐇𝐢𝐥𝐛PX)(A)(\operatorname{\mathbf{Hilb}}^{P}X)(A) if and only if

  1. (a)

    FittQ(t+1)1(St/(S1M))=0\operatorname{Fitt}_{Q(t+1)-1}(S_{t}/(S_{1}\cdot M))=0 and

  2. (b)

    Γ(Ar,XA(t))M\Gamma(\mathbb{P}^{r}_{A},\mathscr{I}_{X_{A}}(t))\subset M.

Proof.

Note that MStM\subset S_{t} is a projective AA-module of rank P(t)P(t). Nakayama’s lemma and [17, Proposition C.4] imply that St/(S1M)S_{t}/(S_{1}\cdot M) is locally generated by Q(t+1)Q(t+1) elements. Thus, [7, Proposition 20.6] implies that FittQ(t+1)(St/(S1M))=A\operatorname{Fitt}_{Q(t+1)}(S_{t}/(S_{1}\cdot M))=A. Hence, by [7, Proposition 20.8], MM satisfies (a) if and only if S1M𝐆𝐫(P(t+1),Rt+1)(A)S_{1}\cdot M\in\operatorname{\mathbf{Gr}}(P(t+1),R_{t+1})(A), which by [17, Proposition C.29] is equivalent to M(𝐇𝐢𝐥𝐛Pr)(A)M\in(\operatorname{\mathbf{Hilb}}^{P}\mathbb{P}^{r})(A).

Let ZZ be the AA-scheme defined by the polynomials in MM. Then MM satisfies (b) if and only if ZXAZ\subset X_{A}. Therefore, M𝐆𝐫(P(t),Rt)(A)M\in\operatorname{\mathbf{Gr}}(P(t),R_{t})(A) satisfies both (a) and (b) if and only if M(𝐇𝐢𝐥𝐛PX)(A)M\in(\operatorname{\mathbf{Hilb}}^{P}X)(A). ∎

For simplicity, we replace RtR_{t} by a kk-vector space VV, and P(t)P(t) by a nonnegative integer ndimVn\leq\dim V. Let e0,e1,,edimV1e_{0},e_{1},\dots,e_{\dim V-1} be an ordered basis of VV. Given 𝐚:<n<dimV\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}, let

z𝐚e𝐚(0)e𝐚(n1)nV.z_{\mathbf{a}}\coloneqq e_{\mathbf{a}(0)}\wedge\dots\wedge e_{\mathbf{a}(n-1)}\in{\textstyle\bigwedge^{n}}V.

Then

(nV)=ProjSym(nV)=Projk[{z𝐚|𝐚:<n<dimV}].\mathbb{P}\!\left({\textstyle\bigwedge^{n}}V\right)=\operatorname{Proj}\operatorname{Sym}\!\left({\textstyle\bigwedge^{n}}V\right)=\operatorname{Proj}k[\{z_{\mathbf{a}}\,|\,\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}\}].
Definition 5.3.

Given 𝐚:<n<dimV\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}, let

𝐚[ji](n){𝐚(n)if nj andiif n=j.\mathbf{a}[j\mapsto i](n)\coloneqq\begin{cases}\mathbf{a}(n)&\text{if $n\neq j$ and}\\ i&\text{if $n=j$.}\end{cases}
Definition 5.4.

Given 𝐚:<n<dimV\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}, let

K𝐚(z𝐚[ji])i<dimV,j<n.K_{\mathbf{a}}\coloneqq\left(z_{\mathbf{a}[j\mapsto i]}\right)_{i<\dim V,j<n}.

Let 𝐛0,𝐛1,,𝐛(dimV)n1\mathbf{b}_{0},\mathbf{b}_{1},\dots,\mathbf{b}_{(\dim V)^{n}-1} be all the functions <dimV<n\mathbb{N}_{<\dim V}\rightarrow\mathbb{N}_{<n}, and let

L(K𝐛0K𝐛1K𝐛(dimV)n1).L\coloneqq\left(\begin{matrix}K_{\mathbf{b}_{0}}&\vline&K_{\mathbf{b}_{1}}&\vline&\cdots&\vline&K_{\mathbf{b}_{(\dim V)^{n}-1}}\end{matrix}\right).

Because of the Pl̈ucker embedding, we can regard 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V) as a closed subscheme of (nV)\mathbb{P}(\bigwedge^{n}V). By abusing notation, we regard z𝐚z_{\mathbf{a}} as a global section of 𝒪𝐆𝐫(n,V)(1)\mathcal{O}_{\operatorname{\mathbf{Gr}}(n,V)}(1).

Lemma 5.5.

Let 𝒪𝐆𝐫(n,V)dimV\mathcal{B}\hookrightarrow\mathcal{O}_{\operatorname{\mathbf{Gr}}(n,V)}^{\dim V} be the universal vector bundle on 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V). Then imL\operatorname{im}L generates (1)\mathcal{B}(1).

Proof.

Let 𝐚:<n<dimV\mathbf{a}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V} be an injection. Let U𝐚𝐆𝐫(n,V)U_{\mathbf{a}}\subset\operatorname{\mathbf{Gr}}(n,V) be the affine open subscheme defined by z𝐚0z_{\mathbf{a}}\neq 0. Then it suffices to prove that the natural embedding ı:U𝐚𝐆𝐫(n,V)\imath\colon U_{\mathbf{a}}\hookrightarrow\operatorname{\mathbf{Gr}}(n,V) represents imz𝐚1L\operatorname{im}z_{\mathbf{a}}^{-1}L.

Without loss of generality, let 𝐚(i)=i\mathbf{a}(i)=i for every i<ni<n. By [12, p. 65], the affine coordinate ring of U𝐚U_{\mathbf{a}} is

Γ(U𝐚)=k[{xi,j|ni<dimV and j<n}],\Gamma(U_{\mathbf{a}})=k[\{x_{i,j}\,|\,n\leq i<\dim V\text{ and }j<n\}],

where xi,jx_{i,j} are indeterminate, and ı\imath represents imJ\operatorname{im}J such that

J=(1001xn,0xn,n1xdimV1,0xdimV1,n1).J=\begin{pmatrix}1&\cdots&0\\ \vdots&\ddots&\vdots\\ 0&\cdots&1\\ x_{n,0}&\cdots&x_{n,n-1}\\ \vdots&\ddots&\vdots\\ x_{\dim V-1,0}&\cdots&x_{\dim V-1,n-1}\\ \end{pmatrix}_{\textstyle.}

For simplicity, let (xi,j)i<n,j<n(x_{i,j})_{i<n,j<n} be the identity matrix, so J=(xi,j)i<dimV,j<nJ=(x_{i,j})_{i<\dim V,j<n}. Given 𝐛:<n<dimV\mathbf{b}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}, let J𝐛J_{\mathbf{b}} be the n×nn\times n matrix such that the iith row of J𝐛J_{\mathbf{b}} is the 𝐛(i)\mathbf{b}(i)th row of JJ. Then J𝐚J_{\mathbf{a}} is the identity matrix. By the definition of the Pl̈ucker embedding,

z𝐛z𝐚=detJ𝐛detJ𝐚=detJ𝐛Γ(U𝐚).\frac{z_{\mathbf{b}}}{z_{\mathbf{a}}}=\frac{\det J_{\mathbf{b}}}{\det J_{\mathbf{a}}}=\det J_{\mathbf{b}}\in\Gamma(U_{\mathbf{a}}).

In particular, z𝐚[ji]/z𝐚=xi,jz_{\mathbf{a}[j\mapsto i]}/z_{\mathbf{a}}=x_{i,j}, meaning that z𝐚1K𝐚=Jz_{\mathbf{a}}^{-1}K_{\mathbf{a}}=J. Thus, imJimz𝐚1L\operatorname{im}J\subset\operatorname{im}z_{\mathbf{a}}^{-1}L, so it suffices to prove that imz𝐚1K𝐛imJ\operatorname{im}z_{\mathbf{a}}^{-1}K_{\mathbf{b}}\subset\operatorname{im}J for every 𝐛:<n<dimV\mathbf{b}\colon\mathbb{N}_{<n}\rightarrow\mathbb{N}_{<\dim V}.

Given m<nm<n, let vmv_{m} and wmw_{m} be the mmth column vectors of J=z𝐚1K𝐚J=z_{\mathbf{a}}^{-1}K_{\mathbf{a}} and z𝐚1K𝐛z_{\mathbf{a}}^{-1}K_{\mathbf{b}}, respectively. Let Ci,jC_{i,j} be the (i,j)(i,j) cofactor of the matrix J𝐛J_{\mathbf{b}}. Then

wm=1z𝐚(z𝐛[m0]z𝐛[mdimkV1])=(detJ𝐛[m0]detJ𝐛[mdimV1])=j=0n1Cm,jvjimJw_{m}=\frac{1}{z_{\mathbf{a}}}\begin{pmatrix}z_{\mathbf{b}[m\mapsto 0]}\\ \vdots\\ z_{\mathbf{b}[m\mapsto\dim_{k}V-1]}\end{pmatrix}=\begin{pmatrix}\det J_{\mathbf{b}[m\mapsto 0]}\\ \vdots\\ \det J_{\mathbf{b}[m\mapsto\dim V-1]}\\ \end{pmatrix}=\sum_{j=0}^{n-1}C_{m,j}v_{j}\in\operatorname{im}J\qed
Lemma 5.6.

Let mm and qq be nonnegative intergers. Let WW be a kk-vector space and let u:VqWu\colon V^{q}\rightarrow W be a linear map. For every kk-algebra AA, let

𝐇(A)={M𝐆𝐫(n,V)(A)|FittdimWm(WA/uA(Mq))=0},\mathbf{H}(A)=\{M\in\operatorname{\mathbf{Gr}}(n,V)(A)\,|\,\operatorname{Fitt}_{\dim W-m}(W_{A}/u_{A}(M^{q}))=0\},

so 𝐇\mathbf{H} is a subfunctor of 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V). Then there is a closed scheme F𝐆𝐫(n,V)F\subset\operatorname{\mathbf{Gr}}(n,V) defined by homogeneous polynomials of degree mm such that 𝐇\mathbf{H} is represented by G𝐆𝐫(n,V)G\cap\operatorname{\mathbf{Gr}}(n,V).

Proof.

For i<qi<q, let uiu_{i} be the iith component of uu, so u=u0uq1u=u_{0}\oplus\dots\oplus u_{q-1}. Let

Λ\displaystyle\Lambda (u0n(dimV)n(L)u1n(dimV)n(L)uq1n(dimV)n(L)).\displaystyle\coloneqq\left(\begin{matrix}u_{0}^{\oplus n(\dim V)^{n}}(L)&\vline&u_{1}^{\oplus n(\dim V)^{n}}(L)&\vline&\cdots&\vline&u_{q-1}^{\oplus n(\dim V)^{n}}(L)\end{matrix}\right).

Let V(nV)V\hookrightarrow\mathbb{P}\left(\bigwedge^{n}V\right) be the closed subscheme defined by the m×mm\times m minors of Λ\Lambda. We will show that 𝐇\mathbf{H} is represented by F𝐆𝐫(n,V)F\cap\operatorname{\mathbf{Gr}}(n,V).

Take U𝐚U_{\mathbf{a}} as in the proof of 5.5. For a kk-algebra AA, take an AA-module MU𝐚(A)M\in U_{\mathbf{a}}(A). Then MM is represented by a kk-algebra morphism f:Γ(U𝐚)Af\colon\Gamma(U_{\mathbf{a}})\rightarrow A. Then 5.5 implies that M=imf(z𝐚1L)M=\operatorname{im}f(z_{\mathbf{a}}^{-1}L). Thus,

uA(Mq)\displaystyle u_{A}(M^{q}) =u0,A(imf(z𝐚1L))++uq1,A(imf(z𝐚1L))\displaystyle=u_{0,A}\!\left(\operatorname{im}f(z_{\mathbf{a}}^{-1}L)\right)+\dots+u_{q-1,A}\!\left(\operatorname{im}f(z_{\mathbf{a}}^{-1}L)\right)
=imf(u0n(dimV)n(z𝐚1L))++imf(uq1n(dimV)n(z𝐚1L))\displaystyle=\operatorname{im}f\left(u_{0}^{\oplus n(\dim V)^{n}}(z_{\mathbf{a}}^{-1}L)\right)+\dots+\operatorname{im}f\left(u_{q-1}^{\oplus n(\dim V)^{n}}(z_{\mathbf{a}}^{-1}L)\right)
=imf(z𝐚1Λ).\displaystyle=\operatorname{im}f(z_{\mathbf{a}}^{-1}\Lambda).

Hence, we have a free resolution

{\dots}Aqn(dimV)n{A^{qn(\dim V)^{n}}}WA{W_{A}}WAu(Mq){{\displaystyle\frac{W_{A}}{u(M^{q})}}}0.{0.}f(z𝐚1Λ)\scriptstyle{f(z_{\mathbf{a}}^{-1}\Lambda)}

Consequently, FittdimWm(WA/u(Ms))\operatorname{Fitt}_{\dim W-m}(W_{A}/u(M^{s})) is generated by m×mm\times m minors of f(z𝐚1Λ)f(z_{\mathbf{a}}^{-1}\Lambda). As a result, M𝐇(A)M\in\mathbf{H}(A) if and only if MF(A)M\in F(A). Since MM is arbitrary, this imples that 𝐇\mathbf{H} is represented by F𝐆𝐫(n,V)F\cap\operatorname{\mathbf{Gr}}(n,V). ∎

Lemma 5.7.

Let UU be a linear subspace of VV. For every kk-algebra AA, let

𝐇(A)={M𝐆𝐫(n,V)(A)|UAM},\mathbf{H}(A)=\{M\in\operatorname{\mathbf{Gr}}(n,V)(A)\,|\,U_{A}\subset M\},

so 𝐇\mathbf{H} is a subfunctor of 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V). Then 𝐇\mathbf{H} is represented by the intersection of a linear space and 𝐆𝐫(n,V)\operatorname{\mathbf{Gr}}(n,V) in (nV)\mathbb{P}\left(\bigwedge^{n}V\right).

Proof.

See [12, p. 66]. ∎

Hence, we can bound the degree of the defining equation of 𝐇𝐢𝐥𝐛PX(P(t)Rt)\operatorname{\mathbf{Hilb}}^{P}X\hookrightarrow\mathbb{P}(\bigwedge^{P(t)}R_{t}).

Theorem 5.8.

The closed embedding 𝐇𝐢𝐥𝐛PX(P(t)Rt)\operatorname{\mathbf{Hilb}}^{P}X\hookrightarrow\mathbb{P}(\bigwedge^{P(t)}R_{t}) is defined by homogeneous polynomials of degree at most P(t+1)+1P(t+1)+1.

Proof.

The Plücker relations are quadratic [12, p. 65]. Thus, it follows from 5.2, 5.6 with u(f0,f1,,fr)=x0f0+x1f1++xrfru(f_{0},f_{1},\dots,f_{r})=x_{0}f_{0}+x_{1}f_{1}+\dots+x_{r}f_{r} and 5.7. ∎

Therefore, an upper bound on Gotzmann numbers will give an explicit construction of the Hilbert scheme. Such a bound is given by Hoa [16, Theorem 6.4(i)].

Theorem 5.9 (Hoa).

Let Ik[x0,,xr]I\subset k[x_{0},\dots,x_{r}] be a nonzero ideal generated by homogeneous polynomials of degree at most d2d\geq 2. Let bb be the Krull dimension and c=r+1bc=r+1-b be the codimension of k[x0,,xr]/Ik[x_{0},\dots,x_{r}]/I. Then

φ(HPI)(32dc+d)b2b1.\varphi(\operatorname{HP}_{I})\leq\left(\frac{3}{2}d^{c}+d\right)^{b2^{b-1}}.

Now, we are ready to give an upper bound:

\NNSbound
Proof.

If XX is a projective space or dimX1\dim X\leq 1, then (𝐍𝐒X)tor(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} is trivial. Therefore, we may assume that d2d\geq 2, 2dimXr12\leq\dim X\leq r-1. Let n=drn=dr and m=(dr)r22r1m=(dr)^{r^{2}2^{r-1}}. Then n(d1)codimXn\geq(d-1)\operatorname{codim}X, and 5.9 with 2br2\leq b\leq r implies that

max{φ(nH),φ(X)}\displaystyle\max\{\varphi(nH),\varphi(X)\} (32(dr)r1+dr)r2r1(dr)r22r1=m.\displaystyle\leq\left(\frac{3}{2}(dr)^{r-1}+dr\right)^{r2^{r-1}}\leq(dr)^{r^{2}2^{r-1}}=m.

Therefore, 3.6 implies that

#(𝐍𝐒X)tordimkΓ(𝐇𝐢𝐥𝐛HPmHX).\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\dim_{k}\Gamma\left(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X\right).

Let t=(dr)r422r2t=(dr)^{r^{4}2^{2r-2}}. 5.9 implies that

max{φ(mH),d}\displaystyle\max\{\varphi(mH),d\} (32mr1+m)r2r1mr22r1=t.\displaystyle\leq\left(\frac{3}{2}m^{r-1}+m\right)^{r2^{r-1}}\leq m^{r^{2}2^{r-1}}=t.

Therefore, 5.8 implies that 𝐇𝐢𝐥𝐛HPmHX\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X is defined by polynomials of degree at most D=P(t+1)+1D=P(t+1)+1 in (P(t)Rt)\mathbb{P}(\bigwedge^{P(t)}R_{t}). Let N=dimkP(t)RtN=\dim_{k}\bigwedge^{P(t)}R_{t}, and let PP be the Hilbert polynomial of mH\mathscr{I}_{mH}. Because r3r\geq 3 and t61295drt\geq 6^{1295}dr,

D=P(t+1)+1(t+r+1r)+1tr.D=P(t+1)+1\leq\binom{t+r+1}{r}+1\leq t^{r}.

and

N=(dimRtP(t))2dimRt=2(t+rr)2tr/4.N=\binom{\dim R_{t}}{P(t)}\leq 2^{\dim R_{t}}=2^{\binom{t+r}{r}}\leq 2^{t^{r}/4}.

Therefore, 3.6 and 4.10 implies that

#(𝐍𝐒X)torΓ(𝐇𝐢𝐥𝐛HPmHX)2N(D2/2+D)N2N1(2D)N2N.\operatorname{{\#}}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\Gamma\left(\operatorname{\mathbf{Hilb}}_{\operatorname{HP}_{mH}}X\right)\leq 2^{N}(D^{2}/2+D)^{N2^{N-1}}\leq(2D)^{N2^{N}}.

Furthermore,

log2(2D)N2N\displaystyle\log_{2}(2D)^{N2^{N}} N2N(1+log2D)N2N(1+rlog2t)2N2N2tr/222tr\displaystyle\leq N2^{N}(1+\log_{2}D)\leq N\cdot 2^{N}\cdot(1+r\log_{2}t)\leq 2^{N}\cdot 2^{N}\cdot 2^{t^{r}/2}\leq 2^{2^{t^{r}}}
logdlog2log222tr\displaystyle\log_{d}\log_{2}\log_{2}2^{2^{t^{r}}} =rlogdt=r522r2(1+logdr)r622r2.\displaystyle=r\log_{d}t=r^{5}2^{2r-2}(1+\log_{d}r)\leq r^{6}2^{2r-2}.

As a resulut,

(3) #(𝐍𝐒X)torexp2exp2exp2expdexp2(2r+6log2r2).\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+6\log_{2}r-2).\qed

In the rest of this section, we drop the condition that XX is connected. Let Y0,Y1,,Yn1Y_{0},Y_{1},\dots,Y_{n-1} be the connected components of XX.

Theorem 5.10.

Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth projective variety defined by homogeneous polynomials of degree d\leq d. Then

#(𝐍𝐒X)torexp2exp2exp2expdexp2(2r+7log2r).\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+7\log_{2}r).
Proof.

Since

𝐏𝐢𝐜X=𝐏𝐢𝐜Y0×𝐏𝐢𝐜Y1×𝐏𝐢𝐜Yn1,\operatorname{\mathbf{Pic}}X=\operatorname{\mathbf{Pic}}Y_{0}\times\operatorname{\mathbf{Pic}}Y_{1}\dots\times\operatorname{\mathbf{Pic}}Y_{n-1},

we have

(𝐍𝐒X)tor=(𝐍𝐒Y0)tor×(𝐍𝐒Y1)tor×(𝐍𝐒Yn1)tor.(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}=(\operatorname{\mathbf{NS}}Y_{0})_{\mathrm{tor}}\times(\operatorname{\mathbf{NS}}Y_{1})_{\mathrm{tor}}\dots\times(\operatorname{\mathbf{NS}}Y_{n-1})_{\mathrm{tor}}.

The Andreotti–Bézout inequality [4, Lemma 1.28] implies that ndrn\leq d^{r} and degYidr\deg Y_{i}\leq d^{r} for every ii. Moreover, every YiY_{i} is defined by homogeneous polynomials of degree at most degYi\deg Y_{i} by [14, Proposition 3]. Without loss of generality, we may assume that

#(𝐍𝐒Y0)tor=max{#(𝐍𝐒Yi)tor| 0i<n}.\#(\operatorname{\mathbf{NS}}Y_{0})_{\mathrm{tor}}=\max\{\#(\operatorname{\mathbf{NS}}Y_{i})_{\mathrm{tor}}\,|\,0\leq i<n\}.

Then

#(𝐍𝐒X)tor\displaystyle\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}} (#(𝐍𝐒Y0)tor)dr\displaystyle\leq\left(\#(\operatorname{\mathbf{NS}}Y_{0})_{\mathrm{tor}}\right)^{d^{r}}
(exp2exp2exp2expdrexp2(2r+6log2r2))dr (by (3))\displaystyle\leq\left(\exp_{2}\exp_{2}\exp_{2}\exp_{d^{r}}\exp_{2}(2r+6\log_{2}r-2)\right)^{d^{r}}\text{ (by (\ref{eqn:better NNS bound}))}
exp2exp2exp2((dr)r622r2dr)\displaystyle\leq\exp_{2}\exp_{2}\exp_{2}\left((d^{r})^{r^{6}2^{2r-2}}{d^{r}}\right)
exp2exp2exp2expd(r722r)\displaystyle\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\left(r^{7}2^{2r}\right)
=exp2exp2exp2expdexp2(2r+7log2r).\displaystyle=\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+7\log_{2}r).\qed

6. Application to Fundamental groups

Let XX be a smooth connected projective variety with base point x0X(k)x_{0}\in X(k). If k=k=\mathbb{C}, then the torsion abelian group (PicX)tor(\operatorname{\mathrm{Pic}}X)_{\mathrm{tor}} is the dual of the profinite abelian group π1e´t(X,x0)ab\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}. More explicitly,

π1e´t(X,x0)ab=Hom((PicτX)tor,/)=limn>0(PicτX)[n].\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}=\operatorname{Hom}((\operatorname{\mathrm{Pic}}^{\tau}X)_{\mathrm{tor}},\mathbb{Q}/\mathbb{Z})=\varprojlim_{n>0}(\operatorname{\mathrm{Pic}}^{\tau}X)[n]^{\vee}.

This can be generalized to algebraically closed fields kk of arbitrary characteristic by using 𝐏𝐢𝐜τX\operatorname{\mathbf{Pic}}^{\tau}X and Nori’s fundamental group scheme π1N(X,x0){\pi^{N}_{1}}(X,x_{0}). Before proceeding, note that π1e´t(X,x0)ab=π1N(X,x0)ab(k)\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}={\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}(k) due to [8, Lemma 3.1].

Theorem 6.1 (Antei [1, Proposition 3.4]).

The commutative torsion group scheme (𝐏𝐢𝐜τX)tor(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}} is the Cartier dual of the commutative profinite group scheme π1N(X,x0)ab{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}. More precisely,

π1N(X,x0)ab=limn>0(𝐏𝐢𝐜τX)[n].{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}=\varprojlim_{n>0}(\operatorname{\mathbf{Pic}}^{\tau}X)[n]^{\vee}.

If k=k=\mathbb{C}, then π1e´t(X,x0)torab(NSX)tor\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}\simeq(\operatorname{NS}X)_{\mathrm{tor}}^{\vee}. We will show that π1N(X,x0)torab(𝐍𝐒X)tor{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}\simeq(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}^{\vee} for general base fields. Then Section 1 gives an upper bound on #π1N(X,x0)torab\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}.

Lemma 6.2.

Let AA be a proper commutative group scheme. Let A0A^{0} be the identity component of AA. Then for every large and divisible mm, we have mAtor=Ared,tor0mA_{\mathrm{tor}}=A^{0}_{\mathrm{red},\mathrm{tor}}.

Proof.

Notice that Ared0A^{0}_{\mathrm{red}} is an abelian variety and A/Ared0A/A^{0}_{\mathrm{red}} is a finite group scheme. Suppose that mm divides #(A/Ared0)\#(A/A^{0}_{\mathrm{red}}). Since multiplication by m is surjective on the abelian variety Ared0A^{0}_{\mathrm{red}}, we get mA=Ared0mA=A^{0}_{\mathrm{red}}. As a result, mA=Ared0mA=A^{0}_{\mathrm{red}}, meaning that mAtor=Ared,tor0mA_{\mathrm{tor}}=A^{0}_{\mathrm{red},\mathrm{tor}}. ∎

Lemma 6.3.

Let AA be a commutative group scheme of finite type. Let BB be a subgroup scheme of AA. Suppose that BB is an abelian variety. Then for every positive integer nn, the natural morphism

φ:A[n]B[n]AB[n]\varphi\colon\frac{A[n]}{B[n]}\rightarrow\frac{A}{B}[n]

is an isomorphism.

Proof.

By the snake lemma,

0{0}B{B}A{A}A/B{A/B}0{0}0{0}B{B}A{A}A/B{A/B}0{0}×n\scriptstyle{\times n}×n\scriptstyle{\times n}×n\scriptstyle{\times n}

gives the exact sequence

0B[n]A[n]AB[n]BnB.0\longrightarrow B[n]\longrightarrow A[n]\longrightarrow\frac{A}{B}[n]\longrightarrow\frac{B}{nB}.

Since BB is an abelian variety, we have B/nBB/nB = 0. ∎

Theorem 6.4.

Let XX be a smooth connected projective variety with base point x0X(k)x_{0}\in X(k). Then

π1N(X,x0)torab\displaystyle{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} (𝐍𝐒X)tor.\displaystyle\simeq(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}^{\vee}.
Proof.

Keep in mind that limits commute with kernels and colimits commute with cokernels. Keep in mind that the Cartier duality is a contravariant equivalence. 6.1 implies that

π1N(X,x0)torab\displaystyle{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}} =limm>0(limn>0(𝐏𝐢𝐜τX)[n])[m]\displaystyle=\varinjlim_{m>0}\left(\varprojlim_{n>0}(\operatorname{\mathbf{Pic}}^{\tau}X)[n]^{\vee}\right)[m]
=limm>0limn>0((𝐏𝐢𝐜τX)[n][m])\displaystyle=\varinjlim_{m>0}\varprojlim_{n>0}\left((\operatorname{\mathbf{Pic}}^{\tau}X)[n]^{\vee}[m]\right)
=limm>0limn>0((𝐏𝐢𝐜τX)[n]m(𝐏𝐢𝐜τX)[n])\displaystyle=\varinjlim_{m>0}\varprojlim_{n>0}\left(\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}{m(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}\right)^{\vee}
=limm>0(limn>0(𝐏𝐢𝐜τX)[n]m(𝐏𝐢𝐜τX)[n])\displaystyle=\varinjlim_{m>0}\left(\varinjlim_{n>0}\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}{m(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}\right)^{\vee}
=limm>0(limn>0(𝐏𝐢𝐜τX)[n]limn>0m(𝐏𝐢𝐜τX)[n])\displaystyle=\varinjlim_{m>0}\left(\frac{\varinjlim_{n>0}(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}{\varinjlim_{n>0}m(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}\right)^{\vee}
=limm>0((𝐏𝐢𝐜τX)torm(𝐏𝐢𝐜τX)tor)\displaystyle=\varinjlim_{m>0}\left(\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}}}{m(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}}}\right)^{\vee}
=(limm>0(𝐏𝐢𝐜τX)torm(𝐏𝐢𝐜τX)tor)\displaystyle=\left(\varprojlim_{m>0}\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}}}{m(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}}}\right)^{\vee}
=((𝐏𝐢𝐜τX)tor(𝐏𝐢𝐜0X)red,tor) (by 6.2)\displaystyle=\left(\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)_{\mathrm{tor}}}{(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red},\mathrm{tor}}}\right)^{\vee}\text{ (by \lx@cref{creftypecap~refnum}{lem:group lem 0})}
=(limn>0(𝐏𝐢𝐜τX)[n]limn>0(𝐏𝐢𝐜0X)red[n])\displaystyle=\left(\frac{\varinjlim_{n>0}(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}{\varinjlim_{n>0}(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red}}[n]}\right)^{\vee}
=(limn>0(𝐏𝐢𝐜τX)[n](𝐏𝐢𝐜0X)red[n])\displaystyle=\left(\varinjlim_{n>0}\frac{(\operatorname{\mathbf{Pic}}^{\tau}X)[n]}{(\operatorname{\mathbf{Pic}}^{0}X)_{\mathrm{red}}[n]}\right)^{\vee}
=(limn>0(𝐍𝐒X)tor[n]) (by 6.3)\displaystyle=\left(\varinjlim_{n>0}(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}[n]\right)^{\vee}\text{ (by \lx@cref{creftypecap~refnum}{lem:group lem 1})}
=(𝐍𝐒X)tor.\displaystyle=(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}^{\vee}.\qed

Therefore, we obtain an upper bound on #π1N(X,x0)torab\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}.

Theorem 6.5.

Let XrX\hookrightarrow\mathbb{P}^{r} be a smooth connected projective variety defined by homogeneous polynomials of degree d\leq d with base point x0X(k)x_{0}\in X(k). Then

#π1N(X,x0)torabexp2exp2exp2expdexp2(2r+6log2r).\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}\leq\exp_{2}\exp_{2}\exp_{2}\exp_{d}\exp_{2}(2r+6\log_{2}r).
Proof.

6.4 implies that #π1N(X,x0)torab=#(𝐍𝐒X)tor\#{\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}=\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}. Therefore, the bound follows from Section 1. ∎

Suppose that chark=p>0\operatorname{char}k=p>0 and p\ell\neq p is a prime number. Since π1e´t(X,x0)ab=π1N(X,x0)ab(k)\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}={\pi^{N}_{1}}(X,x_{0})^{\mathrm{ab}}(k), 6.4 implies that π1e´t(X,x0)ab[]=(NSX)[]\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}[\ell^{\infty}]=(\operatorname{NS}X)[\ell^{\infty}]^{\vee}. However, this is not true for pp-power torsions. Hence, a bound on #(NSX)tor\#(\operatorname{NS}X)_{\mathrm{tor}} such as [21, Theorem 4.12] does not give a bound on #π1e´t(X,x0)torab\#\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}.

\piBound
Proof.

By [32, Proposition 69], we have π1e´t(X,x0)torab=(𝐍𝐒X)tor(k)\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}=(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}^{\vee}(k). Therefore,

#π1e´t(X,x0)torab=#(𝐍𝐒X)tor(k)#(𝐍𝐒X)tor,\#\pi^{\mathrm{\acute{e}t}}_{1}(X,x_{0})^{\mathrm{ab}}_{\mathrm{tor}}=\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}}^{\vee}(k)\leq\#(\operatorname{\mathbf{NS}}X)_{\mathrm{tor}},

and the bound follows from Section 1. ∎

7. Lefschetz Hyperplane Theorem

Throughout the section, XrX\hookrightarrow\mathbb{P}^{r} is a smooth connected projective variety satisfying dimX2\dim X\geq 2, and HH is a smooth hyperplane section of XX. In this section, we discuss the Lefschetz hyperplane theorem for PicτX\operatorname{\mathrm{Pic}}^{\tau}X.

If k=k=\mathbb{C}, the exponential sequence gives a short exact sequence

0H1(X,)H1(X,)PicτXH2(X,)tor0.0\longrightarrow\frac{H^{1}(X,\mathbb{Z})\otimes_{\mathbb{Z}}\mathbb{R}}{H^{1}(X,\mathbb{Z})}\longrightarrow\operatorname{\mathrm{Pic}}^{\tau}X\longrightarrow H^{2}(X,\mathbb{Z})_{\mathrm{tor}}\longrightarrow 0.

With the natural analytic topology, the Pontryagin duality gives an exact sequence

0H1(X,)torHomcont(PicτX,/)H1(X,)H1(X,)tor0.0\longrightarrow H_{1}(X,\mathbb{Z})_{\mathrm{tor}}\longrightarrow\operatorname{Hom}_{\mathrm{cont}}(\operatorname{\mathrm{Pic}}^{\tau}X,\mathbb{R}/\mathbb{Z})\longrightarrow\frac{H_{1}(X,\mathbb{Z})}{H_{1}(X,\mathbb{Z})_{\mathrm{tor}}}\longrightarrow 0.

The Lefschetz hyperplane theorem implies that the map H1(H,)H1(X,)H_{1}(H,\mathbb{Z})\rightarrow H_{1}(X,\mathbb{Z}) is surjective. Therefore, the lemma below implies that PicτXPicτH\operatorname{\mathrm{Pic}}^{\tau}X\rightarrow\operatorname{\mathrm{Pic}}^{\tau}H is injective.

Lemma 7.1.

Let f:ABf\colon A\rightarrow B be a surjective morphism between finitely generated abelian groups. Consider the morphism between exact sequences

0{0}Ator{A_{\mathrm{tor}}}A{A^{\prime}}A/Ator{A/A_{\mathrm{tor}}}0{0}0{0}Btor{B_{\mathrm{tor}}}B{B^{\prime}}B/Btor{B/B_{\mathrm{tor}}}0{0}f\scriptstyle{f_{*}}g\scriptstyle{g}f\scriptstyle{f_{*}}

where the outer vertical morphisms are induced by ff. Then g:ABg\colon A^{\prime}\rightarrow B^{\prime} is surjective.

Proof.

This follows from the snake lemma. ∎

We expect a similar Lefschetz-type theorem for any algebraically closed field kk. Over a general base field, we have the Lefschetz hyperplane theorem for étale fundamental groups [11, XII. Corollaire 3.5]. Hence, for a prime number chark\ell\neq\operatorname{char}k, (PicX)[](PicH)[](\operatorname{\mathrm{Pic}}X)[\ell^{\infty}]\rightarrow(\operatorname{\mathrm{Pic}}H)[\ell^{\infty}] is injective by 6.1. However, étale fundamental groups lack the information of (PicX)[p](\operatorname{\mathrm{Pic}}X)[p^{\infty}].

Theorem 7.2.

Let XX be a smooth connected projective variety, and let HH be a smooth hyperplane section of XX. If dimX2\dim X\geq 2, then the kernel of

r:𝐏𝐢𝐜τX𝐏𝐢𝐜τHr\colon\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H

is a finite commutative group scheme with a connected Cartier dual.

Proof.

Let chark\ell\neq\operatorname{char}k be a prime number. The connected component (kerr)red0(\ker r)^{0}_{\mathrm{red}} of (kerr)red(\ker r)_{\mathrm{red}} is an abelian variety without \ell-torsion points. Therefore, (kerr)red0=0(\ker r)^{0}_{\mathrm{red}}=0 and kerr\ker r is a finite commutative group scheme.

Take a base point xHx\in H, and let M=#(kerr)M=\#(\ker r). Then

(kerr)\displaystyle(\ker r)^{\vee} =ker(𝐏𝐢𝐜τX𝐏𝐢𝐜τH)\displaystyle=\ker(\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H)^{\vee}
=ker((𝐏𝐢𝐜X)[M](𝐏𝐢𝐜H)[M])\displaystyle=\ker\big{(}(\operatorname{\mathbf{Pic}}X)[M]\rightarrow(\operatorname{\mathbf{Pic}}H)[M]\big{)}^{\vee}
=coker((𝐏𝐢𝐜H)[M]𝐏𝐢𝐜X)[M])\displaystyle=\operatorname{coker}\left((\operatorname{\mathbf{Pic}}H)[M]^{\vee}\rightarrow\operatorname{\mathbf{Pic}}X)[M]^{\vee}\right)
=coker(π1N(H,x)abMπ1N(H,x)abπ1N(X,x)abMπ1N(X,x)ab) (by 6.1).\displaystyle=\operatorname{coker}\left(\frac{{\pi^{N}_{1}}(H,x)^{\mathrm{ab}}}{M{\pi^{N}_{1}}(H,x)^{\mathrm{ab}}}\rightarrow\frac{{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}{M{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}\right)\text{ (by \lx@cref{creftypecap~refnum}{thm:pi pic duality})}.

We have

π1N(X,x)abMπ1N(X,x)ab(k)=π1e´t(X,x)abMπ1e´t(X,x)ab\frac{{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}{M{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}(k)=\frac{\pi^{\mathrm{\acute{e}t}}_{1}(X,x)^{\mathrm{ab}}}{M\pi^{\mathrm{\acute{e}t}}_{1}(X,x)^{\mathrm{ab}}}

by [8, Lemma 3.1]. Moreover,

π1e´t(H,x)abMπ1e´t(H,x)abπ1e´t(X,x)abMπ1e´t(X,x)ab\frac{\pi^{\mathrm{\acute{e}t}}_{1}(H,x)^{\mathrm{ab}}}{M\pi^{\mathrm{\acute{e}t}}_{1}(H,x)^{\mathrm{ab}}}\rightarrow\frac{{\pi}^{\mathrm{\acute{e}t}}_{1}(X,x)^{\mathrm{ab}}}{M{\pi}^{\mathrm{\acute{e}t}}_{1}(X,x)^{\mathrm{ab}}}

is surjective by the Lefschetz hyperplane theorem for étale fundamental groups [11, XII. Corollaire 3.5]. Thus, (kerr)(k)=0(\ker r)^{\vee}(k)=0, meaning that (kerr)(\ker r)^{\vee} is connected. ∎

One may want to show that r:𝐏𝐢𝐜τX𝐏𝐢𝐜τHr\colon\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H is injective. Unfortunately, the Lefschetz hyperplane theorem for Nori’s fundamental group scheme is no longer true [2, Remark 2.4]. Nonetheless, we still have the Lefschetz hyperplane theorem in some cases.

Theorem 7.3.

Let dd be a sufficiently large integer. Then for any smooth hyperplane section HH^{\prime} of the dd-uple embedding of XrX\hookrightarrow\mathbb{P}^{r}, the natural map

r:𝐏𝐢𝐜τX𝐏𝐢𝐜τHr\colon\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H^{\prime}

is a closed embedding.

Proof.

Let M=#(kerr)M=\#(\ker r). If d is sufficiently large, then the natural map π1N(H,x)π1N(X,x){\pi^{N}_{1}}(H^{\prime},x)\rightarrow{\pi^{N}_{1}}(X,x) is faithfully flat by [3, Theorem 1.1]. In this case,

(kerr)=coker(π1N(H,x)abMπ1N(H,x)abπ1N(X,x)abMπ1N(X,x)ab)=0(\ker r)^{\vee}=\operatorname{coker}\left(\frac{{\pi^{N}_{1}}(H,x)^{\mathrm{ab}}}{M{\pi^{N}_{1}}(H,x)^{\mathrm{ab}}}\rightarrow\frac{{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}{M{\pi^{N}_{1}}(X,x)^{\mathrm{ab}}}\right)=0

as in the proof of 7.2. ∎

Theorem 7.4 (Langer [22, Theorem 11.3]).

Suppose that XX has a lifting to a smooth projective scheme over W2(k)W_{2}(k). Then

r:𝐏𝐢𝐜τX𝐏𝐢𝐜τHr\colon\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H

is a closed embedding.

The author does not know whether or not PicτXPicτH\operatorname{\mathrm{Pic}}^{\tau}X\rightarrow\operatorname{\mathrm{Pic}}^{\tau}H is injective in general. However, the author conjectures that 𝐏𝐢𝐜τX𝐏𝐢𝐜τH\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H can fail to be a closed embedding if chark=p>0\operatorname{char}k=p>0, because of the failure of the Kodaira vanishing theorem. The argument below is a reformulation of the work in [3, Section 2] and [22, Example 10.1].

Definition 7.5.

Let 𝛂pn\boldsymbol{\alpha}_{p^{n}} be the kernel of the pnp^{n}-power Frobenius endomorphism on 𝔾a\mathbb{G}_{a}.

Theorem 7.6.

Let DD be a smooth ample effective divisor on XX. Then the natural map

r(𝜶p):(𝐏𝐢𝐜τX)(𝜶p)(𝐏𝐢𝐜τD)(𝜶p)r(\boldsymbol{\alpha}_{p})\colon(\operatorname{\mathbf{Pic}}^{\tau}X)(\boldsymbol{\alpha}_{p})\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}D)(\boldsymbol{\alpha}_{p})

is injective if and only if H1(X,𝒪X(D))=0H^{1}(X,\mathcal{O}_{X}(-D))=0.

Proof.

Suppose that H1(X,𝒪X(D))0H^{1}(X,\mathcal{O}_{X}(-D))\neq 0. Because dimX2\dim X\geq 2, there is a large integer nn such that H1(X,𝒪X(pnD))=0H^{1}(X,\mathcal{O}_{X}(-p^{n}D))=0. Let FX:XXF_{X}\colon X\rightarrow X be the absolute Frobenius morphism. Then (FXn)𝒪X(D)=𝒪X(pnD)(F_{X}^{n})^{*}\mathcal{O}_{X}(-D)=\mathcal{O}_{X}(-p^{n}D). Thus, we have the diagram with exact rows and columns as below.

0{0}0{0}Hfppf1(X,𝜶pn){H^{1}_{\mathrm{fppf}}(X,\boldsymbol{\alpha}_{p^{n}})}Hfppf1(D,𝜶pn){H^{1}_{\mathrm{fppf}}(D,\boldsymbol{\alpha}_{p^{n}})}0{0}H1(X,𝒪X(D)){H^{1}(X,\mathcal{O}_{X}(-D))}H1(X,𝒪X){H^{1}(X,\mathcal{O}_{X})}H1(D,𝒪D){H^{1}(D,\mathcal{O}_{D})}0{0}H1(X,𝒪X(pnD)){H^{1}(X,\mathcal{O}_{X}(-p^{n}D))}H1(X,𝒪X){H^{1}(X,\mathcal{O}_{X})}h\scriptstyle{h}i\scriptstyle{i}f\scriptstyle{f}(FXn)\scriptstyle{(F_{X}^{n})^{*}}g\scriptstyle{g}(FXn)\scriptstyle{(F_{X}^{n})^{*}}

Take a nonzero sH1(X,𝒪X(D))s\in H^{1}(X,\mathcal{O}_{X}(-D)). Then (FXn)(f(s))=0(F_{X}^{n})^{*}(f(s))=0, because H1(X,𝒪X(pnD))=0H^{1}(X,\mathcal{O}_{X}(-p^{n}D))=0. Thus, there is a nonzero tHfppf1(X,𝜶pn)t\in H^{1}_{\mathrm{fppf}}(X,\boldsymbol{\alpha}_{p^{n}}) such that f(s)=i(t)f(s)=i(t). On the other hand, h(t)=0h(t)=0, since g(f(s))=0g(f(s))=0. Therefore, hh is not injective. By [1, Proposition 3.2], the natural morphism

r(𝜶pn):(𝐏𝐢𝐜τX)(𝜶pn)(𝐏𝐢𝐜τD)(𝜶pn)r(\boldsymbol{\alpha}_{p^{n}})\colon(\operatorname{\mathbf{Pic}}^{\tau}X)(\boldsymbol{\alpha}_{p^{n}})\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}D)(\boldsymbol{\alpha}_{p^{n}})

is isomorphic to hh, so is also not injective. Let φ:𝜶pn𝐏𝐢𝐜τX\varphi\colon\boldsymbol{\alpha}_{p^{n}}\rightarrow\operatorname{\mathbf{Pic}}^{\tau}X be a nonzero element of kerr(𝜶pn)\ker r(\boldsymbol{\alpha}_{p^{n}}). Then the image of φ\varphi is 𝜶pm\boldsymbol{\alpha}_{p^{m}} for some m>0m>0. Furthermore, 𝜶pm\boldsymbol{\alpha}_{p^{m}} has a subgroup isomorphic to 𝜶p\boldsymbol{\alpha}_{p}. Thus, we have the commutative diagram below.

𝜶pn{\boldsymbol{\alpha}_{p^{n}}}𝜶p{\boldsymbol{\alpha}_{p}}𝜶pm{\boldsymbol{\alpha}_{p^{m}}}𝐏𝐢𝐜τX{\operatorname{\mathbf{Pic}}^{\tau}X}𝐏𝐢𝐜τD{\operatorname{\mathbf{Pic}}^{\tau}D}φ\scriptstyle{\varphi}

The second row gives a nonzero element in the kernel of

r(𝜶p):(𝐏𝐢𝐜τX)(𝜶p)(𝐏𝐢𝐜τD)(𝜶p).r(\boldsymbol{\alpha}_{p})\colon(\operatorname{\mathbf{Pic}}^{\tau}X)(\boldsymbol{\alpha}_{p})\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}D)(\boldsymbol{\alpha}_{p}).

Now, suppose that H1(X,𝒪X(D))=0H^{1}(X,\mathcal{O}_{X}(-D))=0. Then H1(X,𝒪X)H1(D,𝒪D)H^{1}(X,\mathcal{O}_{X})\rightarrow H^{1}(D,\mathcal{O}_{D}) is injective, and so is Hfppf1(X,𝜶p)Hfppf1(D,𝜶p)H^{1}_{\mathrm{fppf}}(X,\boldsymbol{\alpha}_{p})\rightarrow H^{1}_{\mathrm{fppf}}(D,\boldsymbol{\alpha}_{p}). By [1, Proposition 3.2],

r(𝜶p):(𝐏𝐢𝐜τX)(𝜶p)(𝐏𝐢𝐜τD)(𝜶p)r(\boldsymbol{\alpha}_{p})\colon(\operatorname{\mathbf{Pic}}^{\tau}X)(\boldsymbol{\alpha}_{p})\rightarrow(\operatorname{\mathbf{Pic}}^{\tau}D)(\boldsymbol{\alpha}_{p})

is also injective. ∎

Raynaud [30] gave an ample effective divisor DD such that H1(X,𝒪X(D))0H^{1}(X,\mathcal{O}_{X}(-D))\neq 0. Lauritzen [23][24] showed that the Kodaira vanishing theorem can fail even if DD is very ample. However, the author does not know any example of a very ample divisor DD such that H1(X,𝒪X(D))0H^{1}(X,\mathcal{O}_{X}(-D))\neq 0.

Thus we still do not know the answers to the following.

Question 7.7.

Let XX be a smooth connected projective variety of dimX2\dim X\geq 2, and let HH be a smooth hyperplane section. Is the map PicτXPicτH\operatorname{\mathrm{Pic}}^{\tau}X\rightarrow\operatorname{\mathrm{Pic}}^{\tau}H always injective? Is the map 𝐏𝐢𝐜τX𝐏𝐢𝐜τH\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Pic}}^{\tau}H always a closed embedding?

8. The Number of Generators of (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}}

Let dimXr\dim X\hookrightarrow\mathbb{P}^{r} be a smooth connected projective variety such that dimX1\dim X\geq 1. The aim of this section is to prove that (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is generated by at most (degX1)(degX2)(\deg X-1)(\deg X-2) elements, if XX has a lifting over W2(k)W_{2}(k).

In the rest of the section, let CC be the intersection of XX with dimX1\dim X-1 general hyperplanes in r\mathbb{P}^{r}. Then CC is a smooth connected curve, and the genus g(C)g(C) of CC is at most (degX1)(degX2)/2(\deg X-1)(\deg X-2)/2. Since 𝐏𝐢𝐜τC=𝐉𝐚𝐜C\operatorname{\mathbf{Pic}}^{\tau}C=\operatorname{\mathbf{Jac}}C, we have a natural map

r:𝐏𝐢𝐜τX𝐉𝐚𝐜C.r\colon\operatorname{\mathbf{Pic}}^{\tau}X\rightarrow\operatorname{\mathbf{Jac}}C.
Theorem 8.1.

Let CC be a general curve section. Then kerr\ker r is a finite commutative group scheme with a connected dual.

Proof.

By 7.2 applied repeatedly, rr is a composition of homomorphisms with finite connected kernel, so kerr\ker r also is finite and connected. ∎

Lemma 8.2.

If XrX\hookrightarrow\mathbb{P}^{r} is the reduction of a smooth connected projective scheme 𝒳W2(k)r\mathcal{X}\hookrightarrow\mathbb{P}_{W_{2}(k)}^{r} over W2(k)W_{2}(k), then kerr=0\ker r=0.

Proof.

Let VrV\hookrightarrow\mathbb{P}^{r} be a general hyperplane, and let H=XVH=X\cap V. Let 𝒱W2(k)r\mathcal{V}\hookrightarrow\mathbb{P}_{W_{2}(k)}^{r} be a lifting of VV. Then 𝒱𝒳\mathcal{V}\cap\mathcal{X} is a smooth projective scheme, and HH is its reduction. Therefore, if dimX1\dim X\geq 1, then the hypothesis on XX is inherited by a general hyperplane section HH. We may now apply 7.4 iteratively to obtain the result. ∎

Theorem 8.3.

If kerr\ker r is connected, then (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is generated by (degX1)(degX2)(\deg X-1)(\deg X-2) elements.

Proof.

Let N=#(NSX)torN=\#(\operatorname{NS}X)_{\mathrm{tor}}. Then (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is a quotient group of (PicX)[N](\operatorname{\mathrm{Pic}}X)[N]. Since kerr\ker r is connected,

PicτX(𝐉𝐚𝐜C)(k)\operatorname{\mathrm{Pic}}^{\tau}X\rightarrow(\operatorname{\mathbf{Jac}}C)(k)

is injective. Thus,

(PicτX)[N](𝐉𝐚𝐜C)(k)[N](\operatorname{\mathrm{Pic}}^{\tau}X)[N]\rightarrow(\operatorname{\mathbf{Jac}}C)(k)[N]

is also injective.

The group (𝐉𝐚𝐜C)(k)[N](\operatorname{\mathbf{Jac}}C)(k)[N] is generated by 2g(C)2g(C) elements, and 2g(C)(degX1)(degX2)2g(C)\leq(\deg X-1)(\deg X-2). Thus, its subquotient (NSX)tor(\operatorname{NS}X)_{\mathrm{tor}} is also generated by (degX1)(degX2)\leq(\deg X-1)(\deg X-2) elements. ∎

\NSGenLifting
Proof.

This follows from 8.2 and 8.3. ∎

Furthermore, the pp-power torsion subgroups have smaller upper bounds.

Theorem 8.4.

Suppose that chark=p>0\operatorname{char}k=p>0. Then (𝐍𝐒X)[p](k)(\operatorname{\mathbf{NS}}X)[p^{\infty}]^{\vee}(k) is generated by at most (degX1)(degX2)/2(\deg X-1)(\deg X-2)/2 elements. If kerr\ker r is connected, then (NSX)[p](\operatorname{NS}X)[p^{\infty}] is also generated by at most (degX1)(degX2)/2(\deg X-1)(\deg X-2)/2 elements.

Proof.

Take a sufficiently large integer NN. Let (𝐉𝐚𝐜C)(\operatorname{\mathbf{Jac}}C)^{\vee} be the dual abelian variety of (𝐉𝐚𝐜C)(\operatorname{\mathbf{Jac}}C). Since (𝐉𝐚𝐜C)[pN]=(𝐉𝐚𝐜C)[pN](\operatorname{\mathbf{Jac}}C)^{\vee}[p^{N}]=(\operatorname{\mathbf{Jac}}C)[p^{N}]^{\vee},

(𝐉𝐚𝐜C)[pN](k)(𝐏𝐢𝐜X)[pN](k)(\operatorname{\mathbf{Jac}}C)^{\vee}[p^{N}](k)\rightarrow(\operatorname{\mathbf{Pic}}X)[p^{N}]^{\vee}(k)

is surjective by 8.1. Because NN is large, (𝐍𝐒X)[p](\operatorname{\mathbf{NS}}X)[p^{\infty}]^{\vee} is a subgroup scheme of (𝐏𝐢𝐜X)[pN](\operatorname{\mathbf{Pic}}X)[p^{N}]^{\vee}. Since (𝐉𝐚𝐜C)(k)[pN](\operatorname{\mathbf{Jac}}C)^{\vee}(k)[p^{N}] is generated by (degX1)(degX2)/2\leq(\deg X-1)(\deg X-2)/2 elements, so is (𝐍𝐒X)[p](k)(\operatorname{\mathbf{NS}}X)[p^{\infty}]^{\vee}(k).

Now, suppose that kerr\ker r is connected. Then we have an injection

(PicτX)[pN](𝐉𝐚𝐜C)(k)[pN].(\operatorname{\mathrm{Pic}}^{\tau}X)[p^{N}]\rightarrow(\operatorname{\mathbf{Jac}}C)(k)[p^{N}].

Because NN is large, (NSX)[p](\operatorname{NS}X)[p^{\infty}] is a quotient of (PicτX)[pN](\operatorname{\mathrm{Pic}}^{\tau}X)[p^{N}]. Since (𝐉𝐚𝐜C)(k)[pN](\operatorname{\mathbf{Jac}}C)(k)[p^{N}] is generated by (degX1)(degX2)/2\leq(\deg X-1)(\deg X-2)/2 elements, so is (NSX)[p](\operatorname{NS}X)[p^{\infty}]. ∎

Acknowledgement

The author thanks his advisor Bjorn Poonen for his careful advice. The author thanks János Kollár for answering questions regarding Lefschetz-type theorems. The author thanks Barry Mazur for suggesting Nori’s fundamental group schemes. The author also thanks Chenyang Xu, Steven Kleiman and Davesh Maulik for many helpful conversations.

References

  • [1] Marco Antei. On the abelian fundamental group scheme of a family of varieties. Israel J. Math., 186:427–446, 2011.
  • [2] Indranil Biswas and Yogish I. Holla. Comparison of fundamental group schemes of a projective variety and an ample hypersurface. J. Algebraic Geom., 16(3):547–597, 2007.
  • [3] Indranil Biswas and Yogish I. Holla. Comparison of fundamental group schemes of a projective variety and an ample hypersurface. J. Algebraic Geom., 16(3):547–597, 2007.
  • [4] Fabrizio Catanese. Chow varieties, Hilbert schemes and moduli spaces of surfaces of general type. J. Algebraic Geom., 1(4):561–595, 1992.
  • [5] M. Demazure, A. Grothendieck, and M. Artin. Schémas en groupes (SGA 3): Propriétés générales des schémas en groupes. Documents mathématiques. Société mathematique de France, 2011.
  • [6] Thomas W. Dubé. The structure of polynomial ideals and Gröbner bases. SIAM J. Comput., 19(4):750–775, 1990.
  • [7] David Eisenbud. Commutative Algebra: with a view toward algebraic geometry, volume 150. Springer Science & Business Media, 2013.
  • [8] Hélène Esnault, Phùng Hô Hai, and Xiaotao Sun. On Nori’s fundamental group scheme. In Geometry and dynamics of groups and spaces, volume 265 of Progr. Math., pages 377–398. Birkhäuser, Basel, 2008.
  • [9] Gerd Gotzmann. Eine Bedingung für die Flachheit und das Hilbertpolynom eines graduierten Ringes. Math. Z., 158(1):61–70, 1978.
  • [10] Alexander Grothendieck. Éléments de géométrie algébrique: IV. étude locale des schémas et des morphismes de schémas, seconde partie. Publications Mathématiques de l’IHÉS, 24:5–231, 1965.
  • [11] Alexander Grothendieck. Cohomologie locale des faisceaux cohérents et théorèmes de Lefschetz locaux et globaux (SGA(SGA 2)2). North-Holland Publishing Co., Amsterdam; Masson & Cie, Éditeur, Paris, 1968. Augmenté d’un exposé par Michèle Raynaud, Séminaire de Géométrie Algébrique du Bois-Marie, 1962, Advanced Studies in Pure Mathematics, Vol. 2.
  • [12] Joe Harris. Algebraic geometry, volume 133 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1992. A first course.
  • [13] Robin Hartshorne. Algebraic geometry. Springer-Verlag, New York-Heidelberg, 1977. Graduate Texts in Mathematics, No. 52.
  • [14] Joos Heintz. Definability and fast quantifier elimination in algebraically closed fields. Theoret. Comput. Sci., 24(3):239–277, 1983.
  • [15] Jürgen Herzog and Takayuki Hibi. Monomial ideals, volume 260 of Graduate Texts in Mathematics. Springer-Verlag London, Ltd., London, 2011.
  • [16] Lê Tuân Hoa. Finiteness of Hilbert functions and bounds for Castelnuovo-Mumford regularity of initial ideals. Trans. Amer. Math. Soc., 360(9):4519–4540, 2008.
  • [17] Anthony Iarrobino and Vassil Kanev. Power sums, Gorenstein algebras, and determinantal loci, volume 1721 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1999. Appendix C by Iarrobino and Steven L. Kleiman.
  • [18] Jun-ichi Igusa. On some problems in abstract algebraic geometry. Proc. Nat. Acad. Sci. U.S.A., 41:964–967, 1955.
  • [19] Steven L. Kleiman. The Picard scheme. In Fundamental algebraic geometry, volume 123 of Math. Surveys Monogr., pages 235–321. Amer. Math. Soc., Providence, RI, 2005.
  • [20] János Kollár. Rational curves on algebraic varieties, volume 32 of Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. Folge. A Series of Modern Surveys in Mathematics [Results in Mathematics and Related Areas. 3rd Series. A Series of Modern Surveys in Mathematics]. Springer-Verlag, Berlin, 1996.
  • [21] Hyuk Jun Kweon. Bounds on the torsion subgroups of néron-severi groups, 2019. https://arxiv.org/pdf/1902.02753.pdf.
  • [22] Adrian Langer. On the S-fundamental group scheme. Ann. Inst. Fourier (Grenoble), 61(5):2077–2119 (2012), 2011.
  • [23] N. Lauritzen and A. P. Rao. Elementary counterexamples to Kodaira vanishing in prime characteristic. Proc. Indian Acad. Sci. Math. Sci., 107(1):21–25, 1997.
  • [24] Niels Lauritzen. Embeddings of homogeneous spaces in prime characteristics. Amer. J. Math., 118(2):377–387, 1996.
  • [25] Ernst W. Mayr and Albert R. Meyer. The complexity of the word problems for commutative semigroups and polynomial ideals. Adv. in Math., 46(3):305–329, 1982.
  • [26] David Mumford. Lectures on curves on an algebraic surface. With a section by G. M. Bergman. Annals of Mathematics Studies, No. 59. Princeton University Press, Princeton, N.J., 1966.
  • [27] André Néron. Problèmes arithmétiques et géométriques rattachés à la notion de rang d’une courbe algébrique dans un corps. Bull. Soc. Math. France, 80:101–166, 1952.
  • [28] Madhav V. Nori. The fundamental group-scheme. Proc. Indian Acad. Sci. Math. Sci., 91(2):73–122, 1982.
  • [29] Bjorn Poonen, Damiano Testa, and Ronald van Luijk. Computing Néron-Severi groups and cycle class groups. Compos. Math., 151(4):713–734, 2015.
  • [30] M. Raynaud. Contre-exemple au “vanishing theorem” en caractéristique p>0p>0. In C. P. Ramanujam—a tribute, volume 8 of Tata Inst. Fund. Res. Studies in Math., pages 273–278. Springer, Berlin-New York, 1978.
  • [31] Francesco Severi. La base per le varietà algebriche di dimensione qualunque contenuta in una data, e la teoria generate delle corrispondénze fra i punti di due superficie algebriche. Mem. R. Accad. Italia, 5:239–83, 1933.
  • [32] Jakob Stix. Rational points and arithmetic of fundamental groups, volume 2054 of Lecture Notes in Mathematics. Springer, Heidelberg, 2013. Evidence for the section conjecture.