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

Curves with stable or semistable normal bundle on Fano hypersurfaces

Ziv Ran
UC Math Dept.
Skye Surge Facility, Aberdeen-Inverness Road
Riverside CA 92521 US
ziv.ran @ ucr.edu
https://profiles.ucr.edu/app/home/profile/zivran
(Date: \DTMnow)
Abstract.

For every n3,g1n\geq 3,g\geq 1 and all large enough ee depending on n,gn,g, a general curve of genus gg, degree ee in a hypersurface of degree nn in n\mathbb{P}^{n}, or in n\mathbb{P}^{n} itself, has semistable normal bundle. If g,eg,e are large enough, the normal bundle is stable. Similar results hold for curves on general hypersurfaces of degree d<nd<n in n\mathbb{P}^{n} provided certain arithmetical conditions on d,ed,e hold. Previous results were restricted to the case of ambient space n\mathbb{P}^{n} where either the normal bundle has integer slope or n=3n=3 or g=1g=1 and e=n+1e=n+1.

Key words and phrases:
projective curves, stable bundles, normal bundle, hypersurfaces, degeneration methods
2010 Mathematics Subject Classification:
14n25, 14j45, 14m22
arxiv.org 2211.12661

0. Introduction

0.1. Set-up

A vector bundle EE on a curve CC is said to be semistable if every subbundle FEF\subset E has slope μ(F):=deg(F)/rk(F)μ(E)\mu(F):=\deg(F)/\text{rk}(F)\leq\mu(E). It is a natural question whether for a sufficiently general curve CC on a given projective variety XX, with given homology class or degree ee and genus g1g\geq 1, the normal bundle N=NC/XN=N_{C/X} is semistable. This intuitively means CC moves evenly in all directions.

0.2. Known results

Except for the case where μ(N)\mu(N) is an integer, existence results for curves with semistable normal bundle have been sporadic and restricted to the case of ambient space X=nX=\mathbb{P}^{n}, mostly n=3n=3. They are due to many mathematicians since the early 1980s, including Ballico-Ellia [2], Coskun-Larson-Vogt [3], Ein-Lazarsfeld [4], Ellia [6], Elligsrud-Hirschowitz [7], Ellingsrud-Laksov [8], Ghione-Sacchiero [9], Eisenbud-Van de Ven [5], Hartshorne [10], Hulek [11], Newstead [12], Sacchiero [15]. Notably, Ein and Lazarsfeld [4] have shown that an elliptic curve of (minimal) degree n+1n+1 in n\mathbb{P}^{n} has semistable normal bundle, and Coskun-Larson-Vogt [3] have shown recently that with few exceptions, Brill-Noether general curves in 3\mathbb{P}^{3} have stable normal bundle.

Along different lines, it was shown by Atanasov-Larson-Yang [1] that the normal bundle to a general nonspecial curve in n\mathbb{P}^{n} is interpolating or balanced in the sense that for a general effective divisor DD of any degree one has either H0(N(D))=0H^{0}(N(-D))=0 or H1(N(D))=0H^{1}(N(-D))=0. It is known (see lemma 1) that for any bundle with integral slope, interpolation implies semistability. Thus, [1] yields curves with semistable normal bundle in n\mathbb{P}^{n} with any genus gg and degree ee such that 2e+2g20modn12e+2g-2\equiv 0\mod n-1.

For curves on hypersurfaces XX of degree dnd\leq n in n\mathbb{P}^{n}, the results of [14] on interpolation again yield some curves whose normal bundle is balanced with integer slope, hence semistable. In the case of anticanonical hypersurfaces (d=nd=n) this includes curves on any genus gg and sufficiently large degree ee such that e+2g20modn2e+2g-2\equiv 0\mod n-2 (for d<nd<n there are additional congruence conditions). Apparently there are no known results where NN has non-intergal slope.

0.3. New results

The purpose of this paper is to expand the collection of curves with known stable or semistable normal bundle on hypersurfaces. To compactify the statement, we will adopt the following terminology. A (g,e)(g,e) curve is one of genus gg and degree ee; a curve is normally semistable (resp. stable) (in a given ambient space) if its normal bundle is semistable (resp. stable).

Our new results are as follows. First for semistability (see Theorem 5, Theorem 8, Theorem 11, Corollary 12, Example 13 for precise statements).

Theorem (Semistability).

(a) Given g1,e>>0,3n,dng\geq 1,e>>0,3\leq n,d\leq n, then

(i) a general (g,e)(g,e) curve in n\mathbb{P}^{n} is normally semistable;

(ii) a general hypersurface of degree nn in n\mathbb{P}^{n} contains a (g,e)(g,e) normally semistable curve;

(iii) if XX be a general hypersurface of degree d<nd<n in n\mathbb{P}^{n}, and g,e,n,dg,e,n,d satisfy a certain arithmetical condition, then XX contains a normally semistable curve; for example, the condition holds (for all g1,e>>0g\geq 1,e>>0)

- if d=n1d=n-1, whenever n0,2,3mod4n\equiv 0,2,3\mod 4;

- if d=n2d=n-2, whenever n1,3mod6n\equiv 1,3\mod 6.

For large enough genus and degree we also get some results on stability (see Theorem 17 and Theorem 18):

Theorem (Stability).

Notations as above,

(i) for gn,e3gng\geq n,e\geq 3gn, a general (g,e,)(g,e,) curve in n\mathbb{P}^{n} is normally stable;

(ii) for gng\geq n and e3n2ge\geq 3n^{2}g, a general hypersurface of degree nn in n\mathbb{P}^{n} contains a normally stable (g,e)(g,e) curve.

Remarks:

  1. (i)

    The normal bundles in question have in general non-integral slope.

  2. (ii)

    To my knowledge these are the first cases, beyond g=1g=1 or n=3n=3 or the case of integral slope, of genus-gg curves in n\mathbb{P}^{n} with known semistable normal bundle.

  3. (iii)

    Other than the integral slope case, there are no such results in the literature for hypersurfaces other than n\mathbb{P}^{n}.

  4. (iv)

    Our results will actually establish a property called cohomological semistablity which is a priori stronger than semistability.

  5. (v)

    For n=3n=3 Coskun-Larson-Vogt have proven stability of the normal bundle for a larger set of degrees and genera.

  6. (vi)

    We do not state a result on stability for curves on hypersurfaces of degree <n<n in n\mathbb{P}^{n}.

  7. (vii)

    There is no reason to expect the range of curve degrees that we obtain to be optimal: e.g. in the case of n\mathbb{P}^{n} one might optimistically expect (semi)stability for all Brill-Noeter general curves.

As for the geometric meaning of a curve having semistable normal bundle, note that a smooth subvariety containing the curve yields a subbundle of the normal bundle, so from the standard formula for the degree of the normal bundle and the definition of semistability, we have:

Lemma.

Let CXC\subset X be a smooth curve of genus gg with semistable normal bundle on a smooth nn-dimensional variety. Then for any subvariety YXY\subset X of dimension mm containing CC and smooth along it, we have

(n1)C.(KY)+(nm)(2g2)(m1)C.(KX).(n-1)C.(-K_{Y})+(n-m)(2g-2)\leq(m-1)C.(-K_{X}).

0.4. Method of proof

Our strategy is to first prove, in Theorem 1, semistability for genus 1 using a ’fish fang’ degeneration (compare [13], [14]) where the embedding CnC\subset\mathbb{P}^{n} degenerates to

C1p,qC2P1EP2C_{1}\cup_{p,q}C_{2}\subset P_{1}\cup_{E}P_{2}

where P1,P2P_{1},P_{2} are blowups of n\mathbb{P}^{n} in a suitable m\mathbb{P}^{m} resp n1m\mathbb{P}^{n-1-m} with common divisor E=m×n1mE=\mathbb{P}^{m}\times\mathbb{P}^{n-1-m} and where C1,C2C_{1},C_{2} are rational with C1E=C2E={p,q}C_{1}\cap E=C_{2}\cap E=\{p,q\}. In fact we will prove a more general result showing semistability of general modifications of the normal bundle. This result is then used to prove the main result for n\mathbb{P}^{n}, Theorem 5, by an induction on the genus, using another fang degeneration. The result for anticanonical hypersurfaces, Theorem 8 (slighly more general than the above statement), is proven using the n\mathbb{P}^{n} case plus a fan-quasi cone degeneration similar to the one used in [13]. Finally the result for lower-degree hypersurfaces, Theorem 11, is proven using a suitable fang as in [13], essentially putting the curve in a suitable projective bundle over a m\mathbb{P}^{m}, using semistablity of the horizontal part of the normal bundle by Theorem 5 and proving semistability of the vertical part by another degeneration argument.

Notations

The slope of a bundle EE, i.e. deg(E)/rk(E)\deg(E)/\text{rk}(E), is denoted μ(E)\mu(E). The remainder of aa divided by bb is denoted a%b{a}\text{\%}{b}. We denote deg(E)%rk(E){\deg(E)}\text{\%}{\text{rk}(E)} by ρ(E)\rho(E).

1. Preliminaries

A bundle AA on a smooth curve is said to be cohomologically semistable or c-semistable if for every irk(A)i\leq\text{rk}(A) and every line subbundle (or equivalently, rank-1 subsheaf) BiAB\subset\wedge^{i}A, one has

deg(B)μ(iA)=iμ(A)=ideg(A)/rk(A).\deg(B)\leq\mu(\wedge^{i}A)=i\mu(A)=i\deg(A)/\text{rk}(A).

By taking determinants, cohomological semistability implies semistability. AA is said to be balanced if for every tt and a general effective divisor of degree tt one has either H0(A(Dt))=0H^{0}(A(-D_{t}))=0 or H1(A(Dt))=0H^{1}(A(-D_{t}))=0. As mentioned above, balancedness implies semistability for bundles of integral slope:

Lemma 0.

If AA is balanced then we have for every subbundle BAB\subset A,

μ(B)μ(A)\mu(B)\leq\lceil\mu(A)\rceil

In particular, if AA is balanced and μ(A)\mu(A) is an integer then AA is semistable.

Proof.

Adding 1g1-g to both sides, it suffices to prove χ(B)/rk(B)χ(A)/rk(A)\chi(B)/\text{rk}(B)\leq\lceil\chi(A)/\text{rk}(A)\rceil. If tχ(A)/rk(A)t\geq\chi(A)/\text{rk}(A) then χ(A(Dt))0\chi(A(-D_{t}))\leq 0, hence H0(A(Dt))=0H^{0}(A(-D_{t}))=0. Therefore H0(B(Dt))=0H^{0}(B(-D_{t}))=0, hence χ(B(Dt))0\chi(B(-D_{t}))\leq 0, i.e. tχ(B)/rk(B)t\geq\chi(B)/\text{rk}(B). ∎

Note the Lemma does not prove that a balanced bundle with integer slope is cohomologically semistability.

Given a bundle EE on a curve, a general down modification of EE is the kernel of a torsion quotient EτE\to\tau such that the support of τ\tau consists of general points and the map EτE\to\tau is general. Specifying such a modification is equivalent to specifying some general points p1,,ptp_{1},...,p_{t} plus general subspaces ViE|pi,i=1,,tV_{i}\subset E|_{p_{i}},i=1,...,t. Note that general modifications make sense even if the curve is reducible, in which case it is assumed that each point pip_{i} us general in some component. It is easy to see that a general modification of a balanced bundle on 1\mathbb{P}^{1}

A bundle is said to be hyper-stable if its general down modification is stable. Ditto for semistable and for the c-versions.

A bundle EE admits a uniquely-determined increasing filtration called the Harder- Narasimhan filtration

(E)=(E0=(0)E1E2Ek=E)(E_{\bullet})=(E_{0}=(0)\subsetneq E_{1}\subsetneq E_{2}\subsetneq...\subsetneq E_{k}=E)

such that each Ei/Ei1E_{i}/E_{i-1} is semistable and μ(Ei/Ei1)>μ(Ei+1/Ei)\mu(E_{i}/E_{i-1})>\mu(E_{i+1}/E_{i}). In particular, E1E_{1} is called the first HN subbundle of EE and is characterized by having maximal slope, and maximal rank for its slope, among subsheaves of EE.

2. Genus 1 in Projective space

As mentioned above, Ein-Lazarsfeld [4] have proven c-semistability of the normal bundle of an elliptic normal curve, of degree n+1n+1 in n\mathbb{P}^{n}. Here we prove an analogous but logically independent result, showing semistability of the normal bundle of a general ellpitic curve of degree 2n22n-2 or 3n3\geq 3n-3 in n\mathbb{P}^{n}. In the case n=3n=3, this result follows from that of Coskun-Larson-Vogt [3].

First, as a matter of terminology, the bidegree of a bundle EE on a reducible curve C1C2C_{1}\cup C_{2} is by definition (deg(E|C1),deg(E|C2))(\deg(E|_{C_{1}}),\deg(E|_{C_{2}})).

Theorem 1.

The normal bundle of a general elliptic curve of degree e=2n2e=2n-2 or e3n3e\geq 3n-3 in n,n3\mathbb{P}^{n},n\geq 3 is c-hyper-semistable.

In view of Lemma Lemma, this implies:

Corollary 2.

Notations as above, any smooth mm-dimensional subvariety YY containing CC must have

C.(KY)(n+1)(m1)en1.C.(-K_{Y})\leq\frac{(n+1)(m-1)e}{n-1}.
Proof of Theorem.

We will do the case n=2mn=2m even, n4n\geq 4 as the case nn odd is similar and simpler (see comments at the end of the proof). Assume first that e=2n2e=2n-2. Consider a fang degeneration

P0=P1EP2P_{0}=P_{1}\cup_{E}P_{2}

where

P1=BmnE1=m×m1P_{1}=B_{\mathbb{P}^{m}}\mathbb{P}^{n}\supset E_{1}=\mathbb{P}^{m}\times\mathbb{P}^{m-1}
P2=Bm1nE2=m×m1P_{2}=B_{\mathbb{P}^{m-1}}\mathbb{P}^{n}\supset E_{2}=\mathbb{P}^{m}\times\mathbb{P}^{m-1}

in which E1P1,E2P2E_{1}\subset P_{1},E_{2}\subset P_{2} are exceptional divisors and P0P_{0} is constructed via isomorphisms E1EE2E_{1}\simeq E\simeq E_{2} which may be assume general. There is a standard smoothing of P0P_{0} to n\mathbb{P}^{n}. Consider curves

C1P1,C2P2C_{1}\subset P_{1},C_{2}\subset P_{2}

with each being a birational transform of a general rational curve of degree e1e_{1} resp. e2e_{2} meeting m\mathbb{P}^{m} resp. m1\mathbb{P}^{m-1} twice, such that

C1E=C2E={p,q}C_{1}\cap E=C_{2}\cap E=\{p,q\}

and such that

e1,e2{n}[2n1,).e_{1},e_{2}\in\{n\}\cup[2n-1,\infty).

Then C0=C1C2C_{0}=C_{1}\cup C_{2} is a nodal, lci ’fish’ curve in P0P_{0} and smooths out to an elliptic curve CC_{*} of degree e=e1+e22e=e_{1}+e_{2}-2 in n\mathbb{P}^{n} whose normal bundle is a deformation of NC0/P0=NC1/P1NC2/P2N_{C_{0}/P_{0}}=N_{C_{1}/P_{1}}\cup N_{C_{2}/P_{2}}. Note that every degree ee is in Theorem 1 is obtained. Now using Lemma 31 of [14], we may assume each CiC_{i} is balanced in Pi,i=1,2P_{i},i=1,2 , so that

NCk/Pk=ak𝒪(dk+1)(n1ak)𝒪(dk)N_{C_{k}/P_{k}}=a_{k}\mathcal{O}(d_{k}+1)\oplus(n-1-a_{k})\mathcal{O}(d_{k})

where

(2m1)d1+a1=(2m+1)e12m,(2m1)d2+a2=(2m+1)e22m2.(2m-1)d_{1}+a_{1}=(2m+1)e_{1}-2m,(2m-1)d_{2}+a_{2}=(2m+1)e_{2}-2m-2.

Now set a=a1+a2a=a_{1}+a_{2} and note that N0N_{0} has slope

μ=d1+d1+a2m1=(e1+e22)(2m+1)2m1.\mu=d_{1}+d_{1}+\frac{a}{2m-1}=\frac{(e_{1}+e_{2}-2)(2m+1)}{2m-1}.

Because NCk/PkN_{C_{k}/P_{k}} is balanced, it is easy to see that so is its general down modification.

/************

*********/

Note we have natural identifications

(1) NCi/Pi|p=TpEi,i=1,2\begin{split}N_{C_{i}/P_{i}}|_{p}=T_{p}E_{i},i=1,2\end{split}

and likewise for qq. The blowdown map P1nP_{1}\to\mathbb{P}^{n} contracts the vertical factor m1\mathbb{P}^{m-1} of E1E_{1} and because the upper subbundle a1𝒪(d1+1)NC1/P1a_{1}\mathcal{O}(d_{1}+1)\subset N_{C_{1}/P_{1}} maps isomorphically to its image in NC1/nN_{C_{1}/\mathbb{P}^{n}}, it follows that the fibre of the upper subbundle at pp is not contained in the vertical subspace Tpm1TpE1T_{p}\mathbb{P}^{m-1}\subset T_{p}E_{1}, and likewise at qq. Ditto for NC2/P2N_{C_{2}/P_{2}}.

Now I claim that with general choices, the upper subspaces of NC1/P1N_{C_{1}/P_{1}} and NC2/P2N_{C_{2}/P_{2}} at pp are in general position, and likewise for the exterior powers. To this end we use automorphisms. The automorphisms of n\mathbb{P}^{n} stabilizing m\mathbb{P}^{m} lift to automorphsms of P1P_{1} that send E1E_{1} to itself and are compatible with the projection E1mE_{1}\to\mathbb{P}^{m} (i.e. mapping a fibre to a fibre). Now the automorphism group of P1P_{1} fixing p,qp,q maps surjectively to the automorphism group of (E1/m,p,q)(E_{1}/\mathbb{P}^{m},p,q) and the latter acts transitively up to scalars on the ’nonvertical pairs’, i.e. pairs (vp,vq)TpE1TqE2(v_{p},v_{q})\in T_{p}E_{1}\oplus T_{q}E_{2} such that vpTpm,vqTqmv_{p}\not\in T_{p}\mathbb{P}^{m},v_{q}\not\in T_{q}\mathbb{P}^{m}. Such automorphisms of P1P_{1} move C1C_{1} through p,qp,q, compatibly with the isomorphism (1), hence also move the upper subsheaf a𝒪(d1+1)NC1/P1a\mathcal{O}(d_{1}+1)\subset N_{C_{1}/P_{1}} so its fibres at pp and qq are general subspaces. Ditto for C2C_{2}.

/**************************************** *******************************/

Now the following Lemma concludes the proof of Theorem 1.

Lemma 3.

Let C0=C1p,qC2C_{0}=C_{1}\cup_{p,q}C_{2} be a nodal curve with C1C21C_{1}\simeq C_{2}\simeq\mathbb{P}^{1}. Let N0N_{0} be a rank-rr vector bundle on C0C_{0} such that

Nk:=N0|Ckak𝒪(dk+1)(rak)𝒪(dk),k=1,2N_{k}:=N_{0}|_{C_{k}}\simeq a_{k}\mathcal{O}(d_{k}+1)\oplus(r-a_{k})\mathcal{O}(d_{k}),k=1,2

and such that, under the gluing maps N1|pN2|p,N1|qN2qN_{1}|_{p}\simeq N_{2}|_{p},N_{1}|_{q}\simeq N_{2}\simeq q, the upper subspaces a1𝒪C1(d1+1)|p,a2𝒪C2(d2+1)|pa_{1}\mathcal{O}_{C_{1}}(d_{1}+1)|_{p},a_{2}\mathcal{O}_{C_{2}}(d_{2}+1)|_{p} are in general position and likewise at qq. Then

(i) any line subbundle of iN\wedge^{i}N has degree at most iμ(N)=i(d1+d2+(a1+a2)/r)i\mu(N)=i(d_{1}+d_{2}+(a_{1}+a_{2})/r);

(ii) for any 1-parameter smoothing (C,N)(C,N) of (C0,N0)(C_{0},N_{0}) such that the total space of the curve family is smooth, NN is c-semistable.

Proof.

To begin with, the fact that (i) implies (ii) is standard: indeed if S/TS/T is a smooth surface with fibres CtC_{t} and special fibre C0C_{0}, and NN is a bundle on SS with Nt=N|CtN_{t}=N|_{C_{t}} and a general NtN_{t} is not semistable we may consider its ’first Harder-Narasimhan subbundle’ NtN^{\prime}_{t} (maximal slope μmax\mu_{\max}, maximal rank, say ii, among subbundles of slope μmax\mu_{\max}), which is a uniquely determined subbundle and determines a line subbundle of iNt\wedge^{i}N_{t} which by smoothness of SS extends to a line subbundle of iN\wedge^{i}N hence a line subbundle of iN0\wedge^{i}N_{0}, so (i) applies.

As for (i), we will assume μ:=μ(N0)=d1+d2+a/r\mu:=\mu(N_{0})=d_{1}+d_{2}+a/r is not an integer, i.e. 0<a<r0<a<r, as the case where μ\mu is an integer is simpler. We have

(2) iNk=(aki)𝒪(idk+i)(aki1)(rak)𝒪(idk+i1)(raki)𝒪(idk),k=1,2.\begin{split}\wedge^{i}N_{k}=\binom{a_{k}}{i}\mathcal{O}(id_{k}+i)\oplus\binom{a_{k}}{i-1}(r-a_{k})\mathcal{O}(id_{k}+i-1)\oplus...\\ \oplus\binom{r-a_{k}}{i}\mathcal{O}(id_{k}),k=1,2.\end{split}

It has slope

μ(iNk)=iμ(Nk)=i(dk+ak/r).\mu(\wedge^{i}N_{k})=i\mu(N_{k})=i(d_{k}+a_{k}/r).

/********* *********/

Set

dji(a,r)=(ai)+(ai1)(ra)++(aj)(raij).d^{i}_{j}(a,r)=\binom{a}{i}+\binom{a}{i-1}(r-a)+...+\binom{a}{j}\binom{r-a}{i-j}.

Note that if N(a,r)=a𝒪(1)(ra)𝒪N(a,r)=a\mathcal{O}(1)\oplus(r-a)\mathcal{O} on 1\mathbb{P}^{1} and Nji(a,r)=(iN(a,r))(j)N^{i}_{j}(a,r)=(\wedge^{i}N(a,r))(-j), then for any p1p\in\mathbb{P}^{1} we have

dji(a,r)=dim(im(H0(Nji(a,r))Nji(a,r)|pd^{i}_{j}(a,r)=\dim(\text{im}(H^{0}(N^{i}_{j}(a,r))\to N^{i}_{j}(a,r)|_{p}
=dim(im(H0(i(Nk(dk))(j))i(Nk(dk))(j)|p.=\dim(\text{im}(H^{0}(\wedge^{i}(N_{k}(-d_{k}))(-j))\to\wedge^{i}(N_{k}(-d_{k}))(-j)|_{p}.

For any pq1p\neq q\in\mathbb{P}^{1} we have

(3) dji(a,r)+dj1i(a,r)=dim(im(H0(Nji(a,r))Nji(a,r)|{p,q}=dim(im(H0(i(Nk(dk))(j))i(Nk(dk))(j)|{p,q}.\begin{split}d^{i}_{j}(a,r)+d^{i}_{j-1}(a,r)=\dim(\text{im}(H^{0}(N^{i}_{j}(a,r))\to N^{i}_{j}(a,r)|_{\{p,q\}}\\ =\dim(\text{im}(H^{0}(\wedge^{i}(N_{k}(-d_{k}))(-j))\to\wedge^{i}(N_{k}(-d_{k}))(-j)|_{\{p,q\}}.\end{split}
Lemma 4.

Assume a1+a2<ra_{1}+a_{2}<r, j1+j2>ia1+a2rj_{1}+j_{2}>i\frac{a_{1}+a_{2}}{r}. Then

dj1i(a1,r)+dj11i(a1,r)+dj2i(a2,r)+dj21i(a2,r)<2(ri).d^{i}_{j_{1}}(a_{1},r)+d^{i}_{j_{1}-1}(a_{1},r)+d^{i}_{j_{2}}(a_{2},r)+d^{i}_{j_{2}-1}(a_{2},r)<2\binom{r}{i}.
Proof.

We have (ij1)+(ij2)<i2ra1a2r(i-j_{1})+(i-j_{2})<i\frac{2r-a_{1}-a_{2}}{r} where the fraction is >1>1. The identity

(ri)=(ai)++(aj)(raij)+(rai)\binom{r}{i}=\binom{a}{i}+...+\binom{a}{j}\binom{r-a}{i-j}+...\binom{r-a}{i}

shows that

dji(a,r)+dij1i(ra,r)=(ri).d^{i}_{j}(a,r)+d^{i}_{i-j-1}(r-a,r)=\binom{r}{i}.

Therefore

dj2i(a2,r)<dj2i(ra1,r)<dij11i(ra1,r)d^{i}_{j_{2}}(a_{2},r)<d^{i}_{j_{2}}(r-a_{1},r)<d^{i}_{i-j_{1}-1}(r-a_{1},r)

Hence

dj2(a2,r)+dj11i(a1,r)<(ri)d_{j_{2}}(a_{2},r)+d^{i}_{j_{1}-1}(a_{1},r)<\binom{r}{i}

and similarly

dj1(a1,r)+dj21i(a2,r)<(ri).d_{j_{1}}(a_{1},r)+d^{i}_{j_{2}-1}(a_{2},r)<\binom{r}{i}.

Now consider a line subbundle AA of iN0\wedge^{i}N_{0}, with respective restrictions Ak,k=1,2A_{k},k=1,2, We may assume AkA_{k} has degree idk+jkid_{k}+j_{k}, hence corresponds to a section of i(N(dk))(jk)\wedge^{i}(N(-d_{k}))(-j_{k}), and A1A_{1} and A2A_{2} must match at pp and qq. By (3) and Lemma 4, and our general gluing hypothesis, we must have j1+j2ia1+a2rj_{1}+j_{2}\leq i\frac{a_{1}+a_{2}}{r}, hence AA has degree i(d1+d1)+j1+j2i(d1+d2)+(a1+a2)/r=μ(iN0)i(d_{1}+d_{1})+j_{1}+j_{2}\leq i(d_{1}+d_{2})+(a_{1}+a_{2})/r=\mu(\wedge^{i}N_{0}). This concludes the proof of Lemma 3, hence of Theorem 1. ∎

Now in case n=2m+1n=2m+1 odd the argument is the same using P0=P1P2P_{0}=P_{1}\cup P_{2} with Pi=Bmn,i=1,2P_{i}=B_{\mathbb{P}^{m}}\mathbb{P}^{n},i=1,2 glued along E1E2m×mE_{1}\simeq E_{2}\simeq\mathbb{P}^{m}\times\mathbb{P}^{m}. ∎

/*************** ********************/

3. Higher genus in Projective space

The following result extends Theorem 1 to higher genus, using an induction on the genus.

Theorem 5.

Let CC be a general curve of genus g1g\geq 1 and degree e(3g1)(n1)g+1e\geq(3g-1)(n-1)-g+1 in n\mathbb{P}^{n}. Then the normal bundle of CC in n\mathbb{P}^{n} c-hyper-semistable normal bundle;

Corollary 6.

Notations as above, if YnY\subset\mathbb{P}^{n} is a smooth pp-dimensional variety containing CC then

C.(KY)((p1)(n+1)e(np)(2g2))/(n1).C.(-K_{Y})\leq((p-1)(n+1)e-(n-p)(2g-2))/(n-1).
Proof of Theorem.

We use induction on gg, the case g=1g=1 being contained in Theorem 1 We use the same fang degeneration of n\mathbb{P}^{n} as in Theorem 1 but this time we put AA on P1P_{1}. Again we consider a curve

C0=C1C2P0=P1EP2.C_{0}=C_{1}\cup C_{2}\subset P_{0}=P_{1}\cup_{E}P_{2}.

For C1C_{1} we use the birational transform in P1P_{1} of a rational normal curve , with C1E={p,q}C_{1}\cap E=\{p,q\}. Thus we have a perfect normal bundle

NC1/P1=2m𝒪(2m+2).N_{C_{1}/P_{1}}=2m\mathcal{O}(2m+2).

For C2C_{2} we however use a disjoint union

C2=C2,1C2,2C_{2}=C_{2,1}\coprod C_{2,2}

of curves of respective genera 11, g1g-1 and respective degrees

e2,1=4m,e2,2=(6)mg+2,e_{2,1}=4m,e_{2,2}=(\ell-6)m-g+2,

(so that 66(g1)+2\ell-6\geq 6(g-1)+2), such that

C2,1E=p,C2,2E=q.C_{2,1}\cap E=p,C_{2,2}\cap E=q.

meeting the common divisor EE and C1C_{1} in pp resp. qq. Thus C0C_{0} has arithmetic genus gg and is smoothable to a curve of genus gg and degree e=2m+1+e2,1+e2,22e=2m+1+e_{2,1}+e_{2,2}-2 in n\mathbb{P}^{n}. By induction , the normal bundles NC2,i/P2N_{C_{2,i}}/P_{2} are semistable for i=1,2i=1,2, as are their general down modifications. Therefore by the elementary Lemma below so is, in a suitable sense NC0/P0N_{C_{0}/P_{0}} and likewise the smoothing NC/PN_{C/P}.

Lemma 7.

Let E0E_{0} be a vector bundle on a connected nodal curve C0C_{0} that is the union of smooth components, such that the restriction of E0E_{0} on each component is semistable (resp. c-semistable). Then for any smoothing (C,E)(C,E) of (C0,E0)(C_{0},E_{0}), we have

(i) EE is semistable (resp. c-semistable);

(ii) if moreover the restriction of E0E_{0} on at least one component of C0C_{0} is stable (resp. c-stable), EE is stable (resp. c-stable).

This completes the proof in case n=2m+1n=2m+1 odd. In case n=2mn=2m even we let P1=Bm1n,P2=BmnP_{1}=B_{\mathbb{P}^{m-1}}\mathbb{P}^{n},P_{2}=B_{\mathbb{P}^{m}}\mathbb{P}^{n} and proceed similarly. ∎

4. Fano hypersurfaces

The purpose of this section is to construct curves with semistable normal bundle on some general Fano hypersurfaces of dimension n3n\geq 3 in projective space. We begin with the case of anticanonical hypersurfaces (degree n+1n+1 in n+1\mathbb{P}^{n+1}).

Theorem 8.

Let XX be a general hypersurface of degree n+1n+1 in n+1,n3\mathbb{P}^{n+1},n\geq 3 and let g1,eg\geq 1,e be such that en((3g1)(n1)g+1)e\geq n((3g-1)(n-1)-g+1). Then XX contains a curve of genus gg and degree ee with c-hyper- semistable normal bundle.

Remark 9.

Note that for C,XC,X as above the slope

μ(NC/X)=e(n+1)+2g2n12(e+g1)n1mod\mu(N_{C/X})=\frac{e(n+1)+2g-2}{n-1}\equiv\frac{2(e+g-1)}{n-1}\mod\mathbb{Z}

is generally not an integer.

Corollary 10.

Notations as above, if the curve CC is contained in a smooth pp-dimensional subvariety YXY\subset X then

C.(KY)((p1)(n+1)e(np)(2g2))/(n1).C.(-K_{Y})\leq((p-1)(n+1)e-(n-p)(2g-2))/(n-1).
Proof of Theorem.

We will use the same fan- quasi cone degeneration as in [13]. Thus we take

P0=P1EP2P_{0}=P_{1}\cup_{E}P_{2}

with P1P_{1} the blowup pn n+1\mathbb{P}^{n+1} at a point pp, with exceptional divisor E=nE=\mathbb{P}^{n}, and P2=n+1P_{2}=\mathbb{P}^{n+1} containing EE as a hyperplane. In P0P_{0} we consider a Cartier divisor

X0=X1ZX2X_{0}=X_{1}\cup_{Z}X_{2}

with X1X_{1} the blowup of a quasi-cone X¯1\bar{X}_{1}, i.e. a hypersurface of degree n+1n+1 with multiplicity nn at pp, and X2X_{2} a general hyperurface of degree nn in P2=n+1P_{2}=\mathbb{P}^{n+1}. Then X0P0X_{0}\subset P_{0} smooths out to a hypersurface Xn+1X\subset\mathbb{P}^{n+1} of degree n+1n+1. As noted in [13], §4, X1X_{1} may be realized as the blowup of n\mathbb{P}^{n} in an (n,n+1)(n,n+1) complete intersection

Y=FnFn+1Y=F_{n}\cap F_{n+1}

where FnF_{n} corresponds to X1E=X2EX_{1}\cap E=X_{2}\cap E and Fn+1F_{n+1} (which is not unique) corresponds to a hyperplane section of X¯1\bar{X}_{1} and its proper transform is ZZ. Proceeding as in [13], write

e=(n+1)e1a,0an,e=(n+1)e_{1}-a,0\leq a\leq n,

and consider a curve C0X0C_{0}\subset X_{0} of the form

C0=C1C2.C_{0}=C_{1}\cup C_{2}.

Here C1X1C_{1}\subset X_{1} is the birational (=isomorphic) transform of a curve CC^{\prime} of genus gg and degree e1e_{1} in n\mathbb{P}^{n}, meeting YY in aa points p1,,pap_{1},...,p_{a} with general tangents, whose normal bundle NC/nN_{C^{\prime}/\mathbb{P}^{n}} is semistable and remains semistable after the general down modification at p1,,pap_{1},...,p_{a}, corresponding to the tangent spaces TpiYT_{p_{i}}Y. Such a curve exists by Theorem 5 and meets ZZ in CFn+1{p1,,pa}C^{\prime}\cap F_{n+1}\setminus\{p_{1},...,p_{a}\}. The latter modification coincides with NC1/P1N_{C_{1}/P_{1}}. And as in [13], C2X2C_{2}\subset X_{2} is a disjoint union of lines with trivial (hence semistable) normal bundle, meeting C1C_{1} in C1ZC_{1}\cap Z. Now we have

NC0/P0|Ci=NCi/Pi,i=1,2.N_{C_{0}/P_{0}}|_{C_{i}}=N_{C_{i}/P_{i}},i=1,2.

Therefore e.g. by Lemma 7 above or by an argument as in [13], a smoothing CXC\subset X of C0X0C_{0}\subset X_{0} to a curve on a hypersurface of degree n+1n+1 in n+1\mathbb{P}^{n+1} has semistable normal bundle.

The proof of c-hyper- semistability is the same, taking the modification centers to be general points on C1C_{1}. ∎

Next we take up the case of hypersurfaces of degree d<nd<n in n\mathbb{P}^{n}:

Theorem 11.

Let XX be a general hypersurface of degree dnd\leq n in n\mathbb{P}^{n}. Let 0<<e0e0<<e_{0}\leq e and gg be such that

(4) (d2)(nd+1)e=d(n2)e0+(nd)(2g2).\begin{split}(d-2)(n-d+1)e=d(n-2)e_{0}+(n-d)(2g-2).\end{split}

Then XX contains a smooth curve of degree ee and genus gg with c-semistable normal bundle.

Corollary 12.

Notations as above, assume either

  1. (i)

    ((d2)(nd+1),d(n2))=1((d-2)(n-d+1),d(n-2))=1; or

  2. (ii)

    dd is even and ((d/21)(nd+1),(d/2)(n2))=1((d/2-1)(n-d+1),(d/2)(n-2))=1; or

  3. (iii)

    dd is odd, nn is even and ((d2)((n+1d)/2,d(n2)/2)=1((d-2)((n+1-d)/2,d(n-2)/2)=1.

  4. (iv)

    d,nd,n both odd, ((d2)((n+1d)/2),d(n2))=1((d-2)((n+1-d)/2),d(n-2))=1

Then for any g1g\geq 1 and sufficiently large ee, XX contains a curve of genus gg, degree ee with c-semistable normal bundle.

Example 13.

For d=nd=n we recover a special case of the result of Theorem 8 with a shift of notation (namely, the case where a=0a=0 so the curve CC^{\prime} is disjoint from YY).

For d=n1d=n-1 the equation (4) is solvable for all g1g\geq 1 and all large ee provided ((n3),(n12))=1((n-3),\binom{n-1}{2})=1, i.e. either nn is even or n3mod4n\equiv 3\mod 4.

For d=n2d=n-2 the equation (4) is solvable for all g1g\geq 1 and all large ee provided nn is odd and n0,1mod3n\equiv 0,1\mod 3 or equivalently n1,3mod6n\equiv 1,3\mod 6.

Proof of Theorem.

We will use the fang setup as in [13], §6. Thus we set m=d1m=d-1 and consider a limiting form of n\mathbb{P}^{n} which is a fang of the form

Z0=Z1Z2Z_{0}=Z_{1}\cup Z_{2}

where

Z1=m(1,0nm),Z2=nm1(1,0m+1).Z_{1}=\mathbb{P}_{\mathbb{P}^{m}}(1,0^{n-m}),Z_{2}=\mathbb{P}_{\mathbb{P}^{n-m-1}}(1,0^{m+1}).

In Z0Z_{0} we consider a divisor which is a limiting form of a degree-dd hypersurface in n\mathbb{P}^{n} and has the form

X0=X1X2X_{0}=X_{1}\cup X_{2}

where X1=m(G)Z1X_{1}=\mathbb{P}_{\mathbb{P}^{m}}(G)\subset Z_{1} and X2Z2X_{2}\subset Z_{2} is fibred over nm1\mathbb{P}^{n-m-1} with general fibre a general hypersurface of degree mm in the m+1\mathbb{P}^{m+1} fibre of Z2Z_{2}. We recall that GG is a rank-nmn-m bundle over m\mathbb{P}^{m} which fits in an exact sequence

(5) 0𝒪(d+1)𝒪(1)(nm)𝒪G0.\begin{split}0\to\mathcal{O}(-d+1)\to\mathcal{O}(1)\oplus(n-m)\mathcal{O}\to G\to 0.\end{split}

Then in X0X_{0} we consider a connected lci curve of the form

C0=C1C2C_{0}=C_{1}\cup C_{2}

where C2X2C_{2}\subset X_{2} is a disjoint union of lines with trivial (hence c-semistable) normal bundle while C1X1C_{1}\subset X_{1} is a suitable isomorphic lift of n\mathbb{P}^{n}-degree ee of a smooth curve C+mC_{+}\subset\mathbb{P}^{m} of genus gg with c-semistable normal bundle. Then C0X0C_{0}\subset X_{0} deforms to a smooth curve CXC\subset X of degree ee and genus gg on a general hypersurface of degree dd and the normal bundle NC/XN_{C/X} will be c-semistable if NC0/X0N_{C_{0}/X_{0}} is, which in turn will be true provided NC1/X1N_{C_{1}/X_{1}} is c-semistable. Thus it would suffice to show that with suitable choices NC1/X1N_{C_{1}/X_{1}} may be assumed semistable. For convenience, let us call the n\mathbb{P}^{n} and m\mathbb{P}^{m} degrees of a curve C1X1C_{1}\subset X_{1} the upper and lower degrees, say e,e0e,e_{0}, and the pair (e,e0)(e,e_{0}) the bidegree.

Our strategy for constructing C1C_{1} with c-semistable normal bundle is based on the following exact sequence

(6) 0Tv|C1NC1/X1NC+/m0\begin{split}0\to T_{v}|_{C_{1}}\to N_{C_{1}/X_{1}}\to N_{C_{+}/\mathbb{P}^{m}}\to 0\end{split}

combined with the following easy remark

Lemma 14.

Let

0E1EE200\to E_{1}\to E\to E_{2}\to 0

be an exact sequence of vector bundles on a curve such that two of E1E_{1} , E2E_{2} and EE are semistable with the same slope μ\mu. Then so is the third. In char. 0, ditto for c-semistability.

The c-semistable case uses the fact that tensor products of semistable bundles are semistable in char. 0.

Now given C+mC_{+}\subset\mathbb{P}^{m}, lifts of C+C_{+} to C1X1=(G)C_{1}\subset X_{1}=\mathbb{P}(G) correspond to invertible quotients

G|C+BG|_{C_{+}}\to B

and the n\mathbb{P}^{n}-degree of C1C_{1} coincides with deg(B)\deg(B). The argument of [14], proof of Theorem 41 shows that if e0e_{0} is large enough then such a lift C1X1C_{1}\subset X_{1} exists for any e:=deg(B)e0e:=\deg(B)\geq e_{0}. Then we have as in loc. cit. Tv=K(B)T_{v}=K^{*}(B) where KK is the kernel of the surjection G|C+BG|_{C_{+}}\to B, and Tv|C1T_{v}|_{C_{1}} has slope

μ(Tv|C1)=e+(ede0)/(nd).\mu(T_{v}|_{C_{1}})=e+(e-de_{0})/(n-d).

On the other hand we have

μ(NC+/m)=(m+1)e0+2g2m1=de0+2g2d2.\mu(N_{C_{+}/\mathbb{P}^{m}})=\frac{(m+1)e_{0}+2g-2}{m-1}=\frac{de_{0}+2g-2}{d-2}.

Equating these slopes per Lemma 14 leads to

(7) e=(n2)de0(nd)(2g2)(d2)(nd+1)\begin{split}e=\frac{(n-2)de_{0}-(n-d)(2g-2)}{(d-2)(n-d+1)}\end{split}

The condition for C1C_{1} to exist then is that the RHS of (7) be an integer e0\geq e_{0} and that Tv|C1T_{v}|_{C_{1}} is c-semistable for e0e_{0} large enough.

To show the latter we argue as in [14] by induction on the genus g1g\geq 1. As usual with restricted bundles, the hard case is the initial one g=1g=1. To prove this case we will use a suitable fish curve (the argument here is a bit subtle because semistability is false for the components and we will need to use Lemma 3 instead). Note in any event that for g=0g=0, it is proven in [13], Lemma 33, that for all e0e_{0} large enough (in fact e0me_{0}\geq m) and any ee0e\geq e_{0}, KK or equivalently a twist K(B)K^{*}(B) is balanced.

/***************** .********/

Now for g=1g=1, suppose given a bidgree (e,e0)(e,e_{0}) satisfying (7). In fact it sufffices to assume ee0>>0e\geq e_{0}>>0. We consider for C1C_{1} a reducible degenerate version, i.e. a nodal genus-1 curve of the form

C10=C11p,qC12C_{10}=C_{11}\cup_{p,q}C_{12}

where C11,C12C_{11},C_{12} have genus 0 and bidegree (ei,ei0)(e_{i},e_{i0}). Let D0=D1D2D_{0}=D_{1}\cup D_{2} be the projection of C10C_{10} to m\mathbb{P}^{m}. Dualizing the bundle GG in (5), we get an exact sequence on m\mathbb{P}^{m}:

(8) 0G(nm)𝒪𝒪(1)𝒪(m)0\begin{split}0\to G^{*}\to(n-m)\mathcal{O}\oplus\mathcal{O}(-1)\to\mathcal{O}(m)\to 0\end{split}

where the right map is (F0,F1,,Fnm)(F_{0},F_{1},...,F_{n-m}) with components homogeneous polynomials of degrees (m,,m,m+1)(m,...,m,m+1). We have

μ(G|Di)=(m+1)e0inm.\mu(G|_{D_{i}})=\frac{(m+1)e_{0i}}{n-m}.

Let ki=[μ(G|Di]k_{i}=[\mu(G|_{D_{i}}]. Then we have G|Di(ki)=ai𝒪(nmai)𝒪(1)G^{*}|_{D_{i}}(k_{i})=a_{i}\mathcal{O}\oplus(n-m-a_{i})\mathcal{O}(-1) so from (8) we get an exact sequence of sections on DiD_{i}

0H0(ai𝒪)H0((nmai)𝒪(ki)𝒪(ke0))H0(𝒪(me0i+ki))00\to H^{0}(a_{i}\mathcal{O})\to H^{0}((n-m-a_{i})\mathcal{O}(k_{i})\oplus\mathcal{O}(k-e_{0}))\to H^{0}(\mathcal{O}(me_{0i}+k_{i}))\to 0

where the right map is (F0,,Fnm)(F_{0},...,F_{n-m}) while the left map is the inclusion from the upper subbundle. Note that the images of B(ki)B^{*}(k_{i}) in the middle term (nm)𝒪(ki)𝒪(kie0i)(n-m)\mathcal{O}(k_{i})\oplus\mathcal{O}(k_{i}-e_{0i}) for i=1,2i=1,2 must coincide at p,qp,q since C1,C2C_{1},C_{2}, the lifts of D1,D2D_{1},D_{2}, must meet of p,qp,q. By choosing the polymomials FiF_{i} sufficiently general subject to this condition.

Now as G|Di,K|DiG|_{D_{i}},K|_{D_{i}} are balanced we can write K|Di=bi𝒪(αi+1)(nmbi)𝒪(αi)K|_{D_{i}}=b_{i}\mathcal{O}(\alpha_{i}+1)\oplus(n-m-b_{i})\mathcal{O}(\alpha_{i}) so the exact sequence defining KK resticted on DiD_{i} becomes

0bi𝒪(αi+1)(nmbi)𝒪(αi)ai𝒪(ki+1)(nmai)𝒪(ki)𝒪(ei)00\to b_{i}\mathcal{O}(\alpha_{i}+1)\oplus(n-m-b_{i})\mathcal{O}(\alpha_{i})\to a_{i}\mathcal{O}(k_{i}+1)\oplus(n-m-a_{i})\mathcal{O}(k_{i})\to\mathcal{O}(e_{i})\to 0

where the left map can be assumed general. By choosing eie_{i} large enough we can ensure that αi+1ki\alpha_{i}+1\leq k_{i} therefore we can choose the left maps general enough so their images for i=1,2i=1,2 are in general position relative to each other. By Lemma 3, this implies that K|D0K|_{D_{0}} and its smoothing are c-semistable. This completes the proof that Tv|C1T_{v}|_{C_{1}} is semistable for g=1g=1.

/******** ****************/

Finally for g2g\geq 2 we use a 1-node reducible curve

C10=C11pC12C_{10}=C_{11}\cup_{p}C_{12}

with components of genera g1,g2g_{1},g_{2} adding up to gg and suitable degree distribution, to again conclude Tv|C10T_{v}|_{C_{10}} or equivalently K|D0K|_{D_{0}} is c-semistable and proceed as above.

5. Stability

Our strategy for constructing curves with stable normal bundle is to use the following easy remark

Lemma 15.

Let C0=C1C2C_{0}=C_{1}\cup C_{2} be a connected nodal curve. Let E0E_{0} be a bundle on C0C_{0} such that E|C1E|_{C_{1}} is stable and E|C2E|_{C_{2}} is semistable. Then E0E_{0} is stable as is a general smoothing of (E0,C0)(E_{0},C_{0}). Ditto for c-stability and semistability.

We want to use a curve in a fang C0=C1C2X0=X1X2C_{0}=C_{1}\subset C_{2}\subset X_{0}=X_{1}\cup X_{2} as above, and as we have already constructed many curves C2X2C_{2}\subset X_{2} of high genus and degree with semistable normal bundle, it would suffice to construct a ’base curve’ C1X1C_{1}\subset X_{1} of ’small’ genus and degree with stable normal bundle.

We begin with a remark on automorphisms of a product r×s\mathbb{P}^{r}\times\mathbb{P}^{s}. When r=sr=s, the dimension of the automorphism group is 4r(r+2)4r(r+2). When r<sr<s, the dimension is

r(r+2)+s(s+2)+(rs+r+s)+r=r2+s2+rs+4r+3s,r(r+2)+s(s+2)+(rs+r+s)+r=r^{2}+s^{2}+rs+4r+3s,

with the summands on the left corresponding respectively to maps rr,ss,rs,sr\mathbb{P}^{r}\to\mathbb{P}^{r},\mathbb{P}^{s}\to\mathbb{P}^{s},\mathbb{P}^{r}\to\mathbb{P}^{s},\mathbb{P}^{s}\to\mathbb{P}^{r} (the latter maps being necessarily constant). The group acts generally transitively on kk tuples (point, tangent direction ), i.e. acts with an open orbit on the set of kk tuples in the projectivized tangent bundle, i.e. (Tr×s)k\mathbb{P}(T_{\mathbb{P}^{r}\times\mathbb{P}^{s}})^{k}, provided k3r4srk\leq\frac{3r}{4s}r. If sr1s-r\leq 1 the bound is about (3/8)(r+s)(3/8)(r+s).

Now to construct our base curve C1C_{1} we use the following.

Lemma 16.

For n3n\geq 3 there exists g[2,n/2]g\in[2,n/2] and e2(g+3)ne\leq 2(g+3)n such that the normal bundle of a general nonspecial curve of genus gg and degree ee is c-hyper- stable.

Proof.

We first consider the case n=2m+1n=2m+1 odd. Consider a fang P0=P1EP2P_{0}=P_{1}\cup_{E}P_{2} as in the proof of Theorem 5, where the identification E1E2E_{1}\simeq E_{2} is chosen sufficiently general, and a nodal lci curve C0=C1C2P0C_{0}=C_{1}\cup C_{2}\subset P_{0} where

C=D+j=1kRjP,=1,2.C_{\ell}=D_{\ell}+\sum\limits_{j=1}^{k}R_{\ell j}\subset P_{\ell},\ell=1,2.

Here DD_{\ell}, the transverse component, is the inverse image of a general rational curve of degree ede_{\ell}^{d} disjoint from the exceptional divisor EE, while each ’bridge’ component RjR_{\ell j} is a general rational curve of degee ebe_{\ell}^{b} (independent of jj), meeting DD_{\ell} in 1 point qjq_{\ell j} and EE in 1 point pj=R1jE=R2jEp_{j}=R_{1j}\cap E=R_{2j}\cap E, and (p1,,pk)(p_{1},...,p_{k}) may be assumed general on EkE^{k}. Then C0P0C_{0}\subset P_{0} smooths out to a curve of genus g=k1g=k-1 and degree e=e1d+e2d+k((e1b+e2b1)e=e_{1}^{d}+e_{2}^{d}+k((e_{1}^{b}+e_{2}^{b}-1) in n\mathbb{P}^{n}.

/**************** ********/

We want to arrange degrees so that the bridges RjR_{\ell j} are perfect. Assume first that mm is odd. Then it suffices to choose bridge degrees

eb(m+1)/2mod2me_{\ell}^{b}\equiv(m+1)/2\mod 2m

(e.g. eb=(5m+1)/2e^{b}_{\ell}=(5m+1)/2). Then NC0|Rj=2m𝒪(sb)N_{C_{0}}|_{R_{\ell j}}=2m\mathcal{O}(s^{b}_{\ell}), sb=eb+(2eb1m)/(2m)s^{b}_{\ell}=e^{b}_{\ell}+(2e^{b}_{\ell}-1-m)/(2m). If mm is even we can use a fang P0P_{0} with double locus m1×m+1\mathbb{P}^{m-1}\times\mathbb{P}^{m+1} and construct perfect bridges R1j,R2jR_{1j},R_{2j} of respective degree e1bm/2mod2me^{b}_{1}\equiv m/2\mod 2m on X1=Bm+1nX_{1}=B_{\mathbb{P}^{m+1}}\mathbb{P}^{n} and e2bm/2+1mod2me^{b}_{2}\equiv m/2+1\mod 2m on X2=Bm1nX_{2}=B_{\mathbb{P}^{m-1}}\mathbb{P}^{n}.

Now let aa be the remainder

a=(2e+2g2sm)%(2m)=deg(N)%(n1).a={(2e+2g-2-sm)}\text{\%}{(2m)}={\deg(N^{\prime})}\text{\%}{(n-1)}.

Replacing the normal bundle NN by NN^{*}, we may assume ama\leq m. Now we choose e1e_{1}, e2e_{2} with respective remainders a1,a2a_{1},a_{2} such that a1+a2=aa_{1}+a_{2}=a. If aa is even we take a1=a2=a/2a_{1}=a_{2}=a/2 while if aa is odd we take a1=[a/2],a2=[a2]+1a_{1}=[a/2],a_{2}=[a_{2}]+1.

Now because NC|RjN_{C_{\ell}}|_{R_{\ell j}} is perfect, me may identify the upper subspace of NCN_{C_{\ell}} at pjp_{j}, considered as subspace of TpjET_{p_{j}}E, with the upper subspace of NC|DN_{C_{\ell}}|_{D_{\ell}} at qjq_{j}, and we claim that with general choices, the latter becomes a general subspace of TpjET_{p_{j}}E: to see this we can take DD_{\ell} of the form D=D+u=1aLuD_{\ell}=D^{\prime}_{\ell}+\sum\limits_{u=1}^{a_{\ell}}L_{\ell u} such that, setting C=D+RjC^{\prime}_{\ell}=D^{\prime}_{\ell}+\sum R_{\ell j}, NC|DN_{C^{\prime}_{\ell}}|_{D^{\prime}_{\ell}} is perfect and the LuL_{\ell u} are general 1-secant lines to DD^{\prime}_{\ell}. Then NC|DN_{C_{\ell}}|_{D^{\prime}_{\ell}} is an up modification of NC|DN_{C^{\prime}_{\ell}}|_{D^{\prime}_{\ell}} corresponding to these lines, so the upper subspace is general. Moreover as long as k(3/4)m=(3/8)n12k\leq(3/4)m=(3/8)\frac{n-1}{2}, the groups of automorphisms of E1,E2E_{1},E_{2} act generally transitively on the product of the projectivized tangent spaces to E1,E2E_{1},E_{2} at p1,,pkp_{1},...,p_{k}, so by choosing a general isomorphism ϕ:E1E2\phi:E_{1}\to E_{2} we may assume the upper subspaces for C1C_{1} and C2C_{2} at these points are in mutual general position.

Now let NN^{\prime} denote the down modification of NC0/P0N_{C_{0}/P_{0}} in [s/2][s/2] general points on D1D_{1} and s[s/2]s-[s/2] general points on D2D_{2}. and write its slope on C1C_{1} as slope

μ1=μ(N|C1)=[μ1]+a1/(n1),\mu_{1}=\mu(N^{\prime}|_{C_{1}})=[\mu_{1}]+a_{1}/(n-1),

and similarly for μ2,a2\mu_{2},a_{2}. We assume a1,a20a_{1},a_{2}\neq 0 and consider a line subbundle A=A1A2A=A_{1}\cup A_{2} of iN\wedge^{i}N^{\prime} of degree [μ1]+[μ2]\geq[\mu_{1}]+[\mu_{2}]. We may assume A1A_{1} has degree [μ1][\mu_{1}] while A2A_{2} has degree [μ2][\mu_{2}]. The family of pairs (A1,A2)(A_{1},A_{2}) has dimension at most

ia(n1)(n1i)(n1)4(n1i),\frac{ia}{(n-1)}\binom{n-1}{i}\leq\frac{(n-1)}{4}\binom{n-1}{i},

while the condition to match A1A_{1} and A2A_{2} at each of the pjp_{j} is a total of k((n1i)1)k(\binom{n-1}{i}-1) conditions. To be precise, letting b=ia(n1)(n1i)1,=1,2b_{\ell}=\frac{ia_{\ell}}{(n-1)}\binom{n-1}{i}-1,\ell=1,2, we have a restriction map

b1×b2j=1kN|pj\mathbb{P}^{b_{1}}\times\mathbb{P}^{b_{2}}\to\prod\limits_{j=1}^{k}N^{\prime}|_{p_{j}}

whose image may be assumed in general position with respect to the graph of ϕ\phi, hence disjoint from it as soon as k=g+1>n/4k=g+1>n/4. Therefore AA cannot exist in that case.

This completes the proff in case nn is odd, and the case n=2mn=2m even is handled similarly, using a fang X1X2X_{1}\cup X_{2} with X1=BmnX_{1}=B_{\mathbb{P}^{m}}\mathbb{P}^{n} and X2=Bm1nX_{2}=B_{\mathbb{P}^{m-1}}\mathbb{P}^{n}.

Now we can deduce the general case for existence of stable normal bundles, as outlined above:

Theorem 17.

For any n3n\geq 3 there exists g0ng_{0}\leq n such that for all gg0g\geq g_{0}, e3nge\geq 3ng, the normal bundle of a general curve of genus gg and degree ee in n\mathbb{P}^{n} is c-hyper-stable.

Proof.

Let the initial genus g0g_{0} be as in Lemma 16. For g>g0g>g_{0} we use induction on gg and a connected nodal fang curve of the form

C0=C1pC2P1EP2C_{0}=C_{1}\cup_{p}C_{2}\subset P_{1}\cup_{E}P_{2}

with E=m×mE=\mathbb{P}^{m}\times\mathbb{P}^{m} (if n=2m+1n=2m+1) or E=m×m1E=\mathbb{P}^{m}\times\mathbb{P}^{m-1} (if n=2mn=2m), and with C1P1C_{1}\subset P_{1} the proper transform of a curve of genus gg, degree e13nge_{1}\geq 3ng meeting m\mathbb{P}^{m} in 1 general point, and C2C_{2} te proper transform of a general curve of genus 1 meeting m\mathbb{P}^{m} or m1\mathbb{P}^{m-1} in 1 general point. By Lemma 16 and Theorem 1 respectively, the general down modification of normal bundle NC1/X1N_{C_{1}/X_{1}} is stable, while that of NC2/X2N_{C_{2}/X_{2}} is semistable. Hence a general down modification of NC0/P0N_{C_{0}/P_{0}} is stable. ∎

Next we consider the case of anticanonical hypersurfaces.

Theorem 18.

Let XX be a general hypersurface of degree nn in n,n4\mathbb{P}^{n},n\geq 4. Then for all gng\geq n and e3n2ge\geq 3n^{2}g, XX contains a curve of genus gg and degree ee whose normal bundle is c-hyper-stable.

Proof.

The proof mimics exactly that of Theorem 5, where we take for C1C_{1} the proper transform of a curve with stable normal bundle as in Theorem 17. The latter result implies that the normal bundle NC1/X1N_{C_{1}/X_{1}}, being a general down modification of NC/nN_{C^{\prime}/\mathbb{P}^{n}}, is c-hyper-stable, therefore so is NC0/X0N_{C_{0}/X_{0}} and its smoothing. ∎

We don’t have a stability results for curves on hypersurfaces of degree <n<n, because the analogue of Lemma 14 fails, e.g. an extension of stable bundles of the same degree is not stable.

/***********

References

  • [1] A. Atanasov, E. Larson, and D. Yang, Interpolation for normal bundles of general curves, Mem. AMS (2016), arxiv 1509.01724v3.
  • [2] E. Ballico and Ph. Ellia, Some more examples of curves in 3\mathbb{P}^{3} with stable normal bundle, J. Reine Angew. Math. 350 (1984), 87–93.
  • [3] I. Coskun, E. Larson, and I. Vogt, Stability of normal bundles of space curves, Algebra and Number Theory 16 (2022), 919–953, arxiv.math 2003.02964.
  • [4] L. Ein and R. Lazarsfeld, Stability and restrictions of Picard bundles, with an application to the normal bundles of elliptic curves, Complex projective geometry (Trieste 1989/ Bergen 1989) (G. Ellingsrud, C. Peskine, G. Sacchiero, and S. A. Stromme, eds.), London Math. Society Lecture Note series, vol. 179, Cambridge University Press, 1992, pp. 149–156.
  • [5] D. Eisenbud and A. Van de Ven, On the normal bundle of smooth rational space curves, Math. Ann. 256 (1981), 453–463.
  • [6] P. Ellia, Exemples de courbes de 3\mathbb{P}^{3} a fibré normal semi-stable, Math. Ann 264 (1983), 389–396.
  • [7] G. Ellingsrud and A. Hirschowitz, Sur le fibré normal des courbes gauches, C. R. Acad. Sci. Paris Sér I 299 (1984), no. 7, 245–248.
  • [8] G. Ellingsrud and D. Laksov, The normal bundle of elliptic space curves of degree 5, 18th Scand. Congress of Math. Proc. (E. Balslev, ed.), Birkhauser, 1980, pp. 258–287.
  • [9] F. Ghione and G. Sacchiero, Normal bundles of rational curves in 3\mathbb{P}^{3}, Manuscripta Math. 33 (1980), 111–128.
  • [10] R. Hartshorne, Classification of algebraic space curves, Proc. Symp. Pure Math 46 (1985), 145–164.
  • [11] K. Hulek, Projective geometry of elliptic curves, Algebraic Geometry- Open Problems, Lecture Notes in Math., vol. 997, Springer-Verlag, 1983, pp. 228–266.
  • [12] P. E. Newstead, A space curve whose normal bundle is stable, J. London Math. Soc. (2) 28 (1983), 428–434.
  • [13] Z. Ran, Balanced curves and minimal rational connectedness on Fano hypersurfaces, Internat. Math. Research Notices (2022), arxiv.math:2008.01235.
  • [14] by same author, Interpolation of curves on Fano hypersurfces, Communications in Comtemporary Math. (2023), arxiv.math 2201.09793.
  • [15] G. Sacchiero, Exemple de courbes de 3\mathbb{P}^{3} de fibré normal stable, Comm. Algebra 11 (2007), no. 18, 2115–2121.