By OpenAISourceGitHub

Manuscripts from openai/math

Fork
Clone
git clone https://openversions.com/git/index/openai-math.git
Cite
Version link
https://openversions.com/index/openai-math/commit/dfd8c46870b48511e697cbb43a3c8c91d375fa2f
Version identifier (commit hash)
dfd8c46870b48511e697cbb43a3c8c91d375fa2f

Fork to create your own copy with credit to the source. Clone to get a Git copy on your computer.

All papers

Claims

Findings

Logarithmic Kodaira dimension and whole-fiber variation

Let f:U→Vf : U \to V be a projective surjective morphism with connected fibers between smooth connected quasi-projective complex varieties. Assume κˉ(V)≥0\bar{\kappa}(V) \ge 0, and let FF be its geometric generic fiber. Then κˉ(U)≥κ(F)+max⁡{κˉ(V),Var(f)}\bar{\kappa}(U) \ge \kappa(F) + \max\{\bar{\kappa}(V), \mathrm{Var}(f)\}, proving Popa's Conjecture 3.8.

Let f:X→Yf : X \to Y be a surjective morphism with connected fibers between smooth connected projective complex varieties, with κ(Y)≥0\kappa(Y) \ge 0. For its geometric generic fiber FF, κ(X)≥κ(F)+max⁡{κ(Y),Var(f)}\kappa(X) \ge \kappa(F) + \max\{\kappa(Y), \mathrm{Var}(f)\}.

Let f:U→Vf : U \to V be a smooth projective surjection with connected fibers between smooth connected complex quasi-projective varieties such that every closed fiber is non-uniruled. Then κˉ(V)=−∞  ⟹  Var(f)<dim⁡V\bar{\kappa}(V) = -\infty \implies \mathrm{Var}(f) < \dim V, and κˉ(V)≥0  ⟹  Var(f)≤κˉ(V)\bar{\kappa}(V) \ge 0 \implies \mathrm{Var}(f) \le \bar{\kappa}(V). If VV is Campana-special, then Var(f)=0\mathrm{Var}(f) = 0; in particular, if κˉ(V)=0\bar{\kappa}(V) = 0, the family is birationally isotrivial.

Let π∘:P∘→C∘\pi^\circ : P^\circ \to C^\circ be a smooth projective morphism with connected fibers over a smooth connected complex algebraic curve, with very general fiber DD satisfying κ(D)=0\kappa(D) = 0 and h0(D,KD)=1h^0(D, K_D) = 1. If the entire top Hodge bundle Fn(Rnπ∗∘C⊗OC∘)F^n(R^n\pi^\circ_* \mathbf{C} \otimes \mathcal{O}_{C^\circ}) is a flat line with finite monodromy character, then there are a finite extension L/C(C∘)L/\mathbf{C}(C^\circ) and a smooth connected projective variety D0/CD_0/\mathbf{C} such that the generic fiber after extension to LL is birational to (D0)L(D_0)_L (equivariantly for any finite group action or finite fiberwise birational action).

Under the hypotheses of Theorem 8.1 (with g:W→Yg : W \to Y, h:W→Zh : W \to Z, B≥(g∗DY)redB \ge (g^* D_Y)_{\mathrm{red}}, κ(Y,KY+DY)≥0\kappa(Y, K_Y + D_Y) \ge 0, M=h∗AM = h^* A for a nef rational divisor AA on ZZ with KZ+aAK_Z + aA big for a≫1a \gg 1, and L(1)=KW+B+ML(1) = K_W + B + M big over YY), there exist smooth birational models with morphisms W→qU→jZW \xrightarrow{q} U \xrightarrow{j} Z, U→sT0U \xrightarrow{s} T_0, S→pT0S \xrightarrow{p} T_0, ρ:S→Y\rho : S \to Y such that κ(W,L(1))=dim⁡U=d+dim⁡T0\kappa(W, L(1)) = \dim U = d + \dim T_0 and κ(Y,KY+DY)≤dim⁡T0\kappa(Y, K_Y + D_Y) \le \dim T_0.

The models in Proposition 2.7 may be chosen together with smooth connected projective varieties S,T0S, T_0 and morphisms W→(S×T0Z)mainW \to (S \times_{T_0} Z)_{\mathrm{main}} (dominant and generically finite), ρ:S→Y\rho : S \to Y (birational), such that dim⁡(W/Y)=d\dim(W/Y) = d, dim⁡T0=p1\dim T_0 = p_1, dim⁡Z=d+dim⁡T0\dim Z = d + \dim T_0, C(T0)⊆C(Y)\mathbf{C}(T_0) \subseteq \mathbf{C}(Y), and κ(St,KSt+DS∣St)≥0\kappa(S_t, K_{S_t} + D_S|_{S_t}) \ge 0 for very general t∈T0t \in T_0.

For very general z∈Zz \in Z, M∣Wz∼Q0M|_{W_z} \sim_{\mathbf{Q}} 0 and κ(Wz,KWz+B∣Wz)=0\kappa(W_z, K_{W_z} + B|_{W_z}) = 0; consequently, the projective entire top line of the restricted root-cover variation is constant on the connected good locus of WzW_z, and its top-line character has finite image.

After a finite extension of b(I)b(I), there is a normal projective variety V′V' over bb, finite over P0P_0, such that H0⇢I×bV′H_0 \dashrightarrow I \times_b V' birationally, every old B0B_0 component with a divisorial image on V′V' over the generic point of II has an image defined over bb, and the original generic xx-fiber J/KJ/K is geometrically birational to a smooth projective geometrically integral variety J0/kJ_0/k (k=b(V′)k = b(V')) with κ(J0,kˉ)=0\kappa(J_{0,\bar{k}}) = 0 and h0(J0,pKJ0)=1h^0(J_0, pK_{J_0}) = 1.

Let Λ\Lambda be an algebraic closure of L0=b(I)L_0 = b(I). There is a finitely generated field Fb/bF_b/b containing k=b(V′)k = b(V') and an isomorphism Frac(Λ⊗bFb)≃Frac(Λ⊗L0K(J))\mathrm{Frac}(\Lambda \otimes_b F_b) \simeq \mathrm{Frac}(\Lambda \otimes_{L_0} K(J)); in particular, Var(f)≤dim⁡T0\mathrm{Var}(f) \le \dim T_0.

Methods

Parameter field detection via contraction of a nonzero logarithmic pluriform on the base YY with tangent directions of the absolute logarithmic Iitaka base ZZ, controlling the differential rank of the Hilbert parameter map τ:Z⇢Hilb(Y)\tau : Z \dashrightarrow \mathrm{Hilb}(Y).

Source excerpts

“Lemma 2.8. For very general z ∈ Z, the form ξ in (2.2) induces nonzero sections of m(KWz + D0 W |Wz ) by contraction with suitable p1 tangent directions at z.” · Show in paper

“It follows that dτz has the same kernel as that normal map and therefore has rank p1. In characteristic zero, generic differential rank equals the dimension of the reduced image.” · Show in paper

Minimal-index ordinary cyclic root cover J~→J\widetilde{J} \to J of a Kodaira-dimension-zero fiber, producing a one-dimensional entire top-form space whose parabolic Hodge line extension MM is compared against actual divisorial orders tPt_P and all-valuation thresholds λP\lambda_P.

Source excerpts

“Lemma 3.1 (The root cover). Let p be the least positive integer for which H0(J, pKJ ) ̸= 0. A generator ωJ determines a geometrically connected cyclic cover of J.” · Show in paper

“Proposition 3.5 (The Hodge order). At every prime P of W, the parabolic order of the root form η is λP − 1. Consequently M ∼Q P P (λP − 1)P, T ∼Q B + M.” · Show in paper

Analytic birational constancy criterion over a curve: lifting a base direction to a meromorphic vector field vv annihilating a global top form Θ\Theta, proving every horizontal pole has positive asymptotic fixed order via L2/mL^{2/m} Bergman-weight estimates and Demailly–Hacon–Păun extension, and contracting poles on a relative log canonical model.

Source excerpts

“Lemma 4.3 (A meromorphic differential operator). The vector field v extends meromorphically to a neighborhood of each general fiber, with poles bounded by E.” · Show in paper

“Every horizontal pole divisor is contracted by χ. ... Thus ∂ is a regular derivation on the entire normal space.” · Show in paper

Recognizable markings on log canonical models: adding coordinate and sum hyperplanes (x0=0),(xj=0),(x0+xj=0)(x_0=0), (x_j=0), (x_0+x_j=0) with distinct small rational coefficients to force model-pair variation zero while recovering the finite normalization V′→P0V' \to P_0 and fixing old boundary images.

Source excerpts

“Choice of the markings. Choose a very ample line bundle L on P0 and a general basis x0, . . . , xN of H0(P0, L). ... We therefore use the divisors (6.7) (x0 = 0), (xj = 0), (x0 + xj = 0) (1 ≤ j ≤ N).” · Show in paper

Whole-fiber Galois descent using Hanamura's birational group structure to trivialize the abelian part of a descent cocycle, multiplicity-one moving-divisor thresholds to eliminate ramification, and Zariski–Nagata purity with invariance of finite étale covers.

Source excerpts

“Hanamura’s structure theorem for non-uniruled projective varieties in characteristic zero provides a suitable smooth projective birational model T of (J0)k¯ for which the reduced scheme of flat graphs of birational maps, B = Bir(T)red, is a group scheme locally of finite type over k¯.” · Show in paper

“Purity of the branch locus makes it finite étale. ... Finite étale covers are invariant under an extension of algebraically closed fields in characteristic zero.” · Show in paper

Numerical second Iitaka construction via exact divisorial section-transfer PU(c)P_U(c), degree-independent invariant volume lower bounds, and weak-positivity interpolation.

Source excerpts

“Lemma 8.5 (A degree bound for invariant growth). ... (8.11) lim sup n d! h0(H, nD) / nd ≥ v / e0.” · Show in paper

“PU (1) = (a − 1)/(a − c) PU (c) + (1 − c)/(a − c) PU (a). The two coefficients are positive, so PU (1) is big.” · Show in paper