Brill--Noether Generality of Curves and K3 Surfaces

Lazarsfeld proved Brill–Noether generality of any smooth curve in the linear system $|H|$ where $(X,H)$ is a polarized K3 surface with $\mathrm{Pic}(X) = \mathbb{Z}\cdot H$. Mukai introduced the notion of Brill–Noether generality for quasi-polarized K3 surfaces. We prove Brill–Noether generality of any smooth curve in the linear system $|H|$ where $(X,H)$ is a Brill–Noether general quasi-polarized K3 surface.

💡 Research Summary

The paper extends Lazarsfeld’s classical result on Brill–Noether generality of curves on polarized K3 surfaces to the broader setting of Mukai’s Brill–Noether general quasi‑polarized K3 surfaces. Lazarsfeld proved that if a K3 surface X carries a primitive ample line bundle H with Pic X = ℤ·H, then every smooth curve C in the linear system |H| is Brill–Noether (BN) general, i.e. for any line bundle A on C the Brill–Noether number ρ(A)=g−h⁰(A)h¹(A) is non‑negative. This argument hinges on the Lazarsfeld bundle F_{A,C}, defined as the kernel of the evaluation map H⁰(A)⊗𝒪_X→i_*A, and on the fact that χ(F_{A,C}^∨⊗F_{A,C})=2−2ρ(A). If ρ(A)<0, F_{A,C} admits a non‑trivial endomorphism, which forces a decomposition H= D₁+D₂ into two effective non‑zero divisors, contradicting Pic X=ℤ·H.

Mukai relaxed the Picard‑rank condition by introducing the notion of a Brill–Noether general quasi‑polarized K3 surface (X,H): for any decomposition H=D₁+D₂ with D_i>0 one must have g(X,H)−h⁰(D₁)h⁰(D₂)≥0, where g(X,H)=½H²+1 is the genus of curves in |H|. This condition is Zariski‑open, unlike the Picard‑rank‑one hypothesis. The main theorem of the paper (Theorem 1.5) asserts that if (X,H) is Mukai‑BN‑general, then every smooth curve C∈|H| is BN‑general.

The proof proceeds through several new ingredients:

1. Reduction to globally generated bundles. Lemma 2.4 shows that any line bundle A with ρ(A)<0 can be replaced by a line bundle \bar A whose both \bar A and its adjoint \bar A^∨⊗ω_C are globally generated, preserving the negativity of ρ. This allows the authors to work with Lazarsfeld bundles associated to globally generated pairs (A,C).

2. Analysis of divisor decompositions. Section 3 studies possible decompositions H=D₁+…+D_n with h⁰(D_i)≥2. Proposition 3.8 provides necessary conditions for (X,H) to be BN‑general: if n≥5, or if n=4 with D₁²≥2, or if n=3 with certain large self‑intersection patterns, then (X,H) cannot be BN‑general. The proof uses elementary intersection theory on K3 surfaces (SD74) and the inequality χ(D₁+D₂)·χ(D₃)>g, which forces the existence of a BN‑locus of negative expected dimension.

3. Harmonic filtration of Lazarsfeld bundles. The authors introduce a new structure on F_{A,C}: a decreasing filtration F₁⊃F₂⊃…⊃F_n⊃0, called a harmonic filtration (Definition 4.5). Each graded piece G_i=F_i/F_{i+1} has a Mukai vector v(G_i)=(r_i,−D_i,s_i). Lemma 4.10 establishes key relations:

   • Σ r_i = h⁰(A), Σ s_i = h¹(A);
   • r_i s_i ≤ ½ D_i²+1;
   • r_i ≤ r_{i+1}+…+r_n, and s_n ≥1. These constraints tightly control the possible self‑intersections D_i² and the length n of the filtration.

4. Bounding the length of the filtration. Using Proposition 3.8 together with the inequalities from Lemma 4.10, the authors prove that n≤4. Moreover, for n=4 all D_i² must be zero; for n=3 the only admissible triples (D₁²,D₂²,D₃²) are (2,2,0), (4,2,0), (6,2,0) or (0,0,0). This mirrors the classification of Dynkin diagrams of types D_{m+3} and E_m (Remark 1.6).

5. Final assembly. Section 5 treats the three possible lengths separately:

   • n=2 (Proposition 5.2) is handled by a direct argument showing that any non‑trivial endomorphism of F_{A,C} would give a forbidden decomposition of H.
   • n=3 (Proposition 5.4) requires a delicate analysis of the three graded pieces; the allowed self‑intersection patterns are shown to be incompatible with the BN‑generality of (X,H) unless ρ(A)≥0.
   • n=4 (Proposition 5.5) uses the fact that all D_i²=0, which forces each D_i to be an elliptic divisor; the resulting configuration again contradicts the assumption that ρ(A)<0.

Since any hypothetical counterexample (a curve C∈|H| with a line bundle A violating BN‑generality) would produce a Lazarsfeld bundle with a harmonic filtration of length n≥2, the above analysis eliminates all possibilities. Consequently, for a Mukai‑BN‑general quasi‑polarized K3 surface (X,H), every smooth curve in |H| is BN‑general.

The result removes the restrictive Picard‑rank‑one hypothesis from Lazarsfeld’s theorem, providing a Zariski‑open condition that holds for a dense set of polarized K3 surfaces. The authors suggest applications to the study of projective models of K3 surfaces and Fano varieties, notably extending results of Bădescu–Kuznetsov–Mukai (BKM) from Picard rank 1 to the broader class of BN‑general K3 surfaces and consequently to BN‑general Fano varieties in Mukai’s sense. The paper thus bridges classical Brill–Noether theory, vector bundle techniques on K3 surfaces, and modern moduli considerations.