Introduction
This survey tells a short story. In three weeks of September 2026 a numerical obstruction for vector bundles on polarized surfaces was pushed, one step at a time, from projective geometry to local commutative algebra. The central steps are recorded in three preprints and an AI-generated manuscript; a further preprint, [24 ] , is a side branch towards higher dimension:
[1 ] 2 September: smooth polarized surfaces with no Ulrich bundle [1 ] ;
[24 ] 6 September (v3: 10 September): smooth polarized varieties of every dimension ≥ 2 \ge2 ≥ 2 with no Ulrich bundle [24 ] ; used in [2 §5] , not in the local arguments;
[2 ] 14 September: no arithmetically Cohen–Macaulay (aCM) bundles, hence section rings with no graded maximal Cohen–Macaulay (MCM) module [2 ] ;
[9 ] 21 September: a proposed local cohomology obstruction for arbitrary modules over the completed section ring, under K ≡ 4 H K \equiv4H K ≡ 4 H [9 ] ;
[26 ] 23 September (date of the manuscript): a complete local domain without a small Cohen–Macaulay module [26 ] , published on 6 October in OpenAI’s catalogue of mathematical manuscripts [27 ] .
Figure 1 places them on a timeline, and Figure 2 records the citations among them (details in § 11 ).
Figure 1. The five texts (September–October 2026).
Figure 2. Citations among the five texts (an arrow points from a text to one that uses or cites it). The central chain is drawn thick; [24 ] (dashed) is a side branch towards higher dimension. Dotted arrows lead from the texts on which § 10 builds to this paper’s higher-dimensional counterexamples to Hochster’s conjecture: the argument of [26 ] , the examples of [1 ] , the product construction of [2 ] , and the higher-dimensional Ulrich inequality of [24 ] , which § 10 recovers.
Three existence questions sit behind the story, and they form a tower (Figure 3 ). For a smooth projective surface Y Y Y with a polarization H H H , let R = R ( Y , H ) = ⨁ t ≥ 0 H 0 ( Y , O Y ( t H ) ) R = R(Y,H) = \bigoplus_{t\geq0} H^0(Y,\mathcal{O}_Y(tH)) R = R ( Y , H ) = ⨁ t ≥ 0 H 0 ( Y , O Y ( t H )) be the section ring and R ^ \widehat{R} R its completion at the vertex. The three questions ask for the existence of (R1) an Ulrich bundle on ( Y , H ) (Y,H) ( Y , H ) — the question of Eisenbud, Schreyer and Weyman [ESW];
Figure 3. The tower of existence questions.
(R2) an aCM bundle on ( Y , H ) (Y,H) ( Y , H ) , equivalently a graded MCM module over R R R — the question Eisenbud and Schreyer asked after [1 ] ;
(R3) a small Cohen–Macaulay module over R ^ \widehat{R} R — Hochster’s conjecture [17 , 18 , 20 ] .
Existence propagates downwards: (R1)⇒ \Rightarrow ⇒ (R2)⇒ \Rightarrow ⇒ (R3) (Lemma 2.5 ). Non-existence propagates upwards, so each step of the story excludes a further level ([24 ] extends the first level to higher dimension).
All obstructions compare the degree and the signature ,
d = H 2 , σ = σ ( Y ) = K Y 2 − 8 χ ( O Y ) = 1 3 ( K Y 2 − 2 c 2 ( Y ) ) . d = H^2,\qquad\sigma= \sigma(Y) = K_Y^2 - 8\chi(\mathcal{O}_Y) = \frac{1}{3}(K_Y^2 - 2c_2(Y)). d = H 2 , σ = σ ( Y ) = K Y 2 − 8 χ ( O Y ) = 3 1 ( K Y 2 − 2 c 2 ( Y )) .
Table 1 compares the numerical criteria, including Chen’s proposed one.
Table 1. The five results, and the extension of [26 ] proved in Section 10 (bpf = base-point-free).
Both local arguments take their examples from the complete-quadrangle family ( Y n , 3 A n − E n ) (Y_n,3A_n-E_n) ( Y n , 3 A n − E n ) (§ 4 ): Hirzebruch’s surfaces with the polarization 3 A n − E n 3A_n-E_n 3 A n − E n of [1 §7.2] , whose degree lies below the signature for n ≥ 6 n\geq6 n ≥ 6 . [26 ] takes the construction from [1 §7.2] and uses the prime n = 7 n=7 n = 7 , and [9 ] takes n = 6 n=6 n = 6 from [2 Rem. 4.2, Cor. 4.3] , where its section ring is shown to be quasi-Gorenstein. For both values of n n n , [2 ] had already shown that the section ring has no graded MCM module. The Hesse pairs, the main examples of [1 ] , are used by neither local argument; see Corollary 9.3 .
What this survey adds. Besides a unified exposition, the survey makes two observations explicit and proves one extension.
(a) Through a finite map f : Y → P 2 f:Y\to\mathbb{P}^2 f : Y → P 2 with f ∗ O ( 1 ) = O ( H ) f^*\mathcal{O}(1)=\mathcal{O}(H) f ∗ O ( 1 ) = O ( H ) , the three obstructions are one computation: a variance of the push-forward, decomposed along a filtration with semistable factors (Proposition 9.1 ). This identity keeps all terms in [26 (10), (11)] and contains [2 Prop. 2.6] . What changes from level to level is what we know about the push-forward: trivial, split, or the restriction to an exceptional plane of a sheaf that is trivial off it. For surfaces this explains the factor 2 between σ ≤ d \sigma\leq d σ ≤ d (graded) and σ ≤ 2 d \sigma\leq2d σ ≤ 2 d (local, [26 ] ); Chen’s argument fits the same computation, with a stronger lower bound for the variance (Remark 9.2).
(b) By [26 Thm. 1.2] (proved in § 10 ), the completed section rings of the Hesse pairs ( Y n , 4 A n − E n ) (Y_n,4A_n-E_n) ( Y n , 4 A n − E n ) , n ≥ 3 n \ge3 n ≥ 3 , of the dual Hesse pairs ( Y n , 6 A n − E n ) (Y_n,6A_n-E_n) ( Y n , 6 A n − E n ) , n ≥ 6 n \ge6 n ≥ 6 , and of the complete-quadrangle pairs ( Y n , 3 A n − E n ) (Y_n,3A_n-E_n) ( Y n , 3 A n − E n ) , n ≥ 7 n \ge7 n ≥ 7 , have no small CM module (Corollary 9.3 ); this includes the ball quotient Y 3 Y_3 Y 3 of [1 ] .
(c) We generalize the argument of [26 ] to every dimension m ≥ 2 m \ge2 m ≥ 2 (Theorem 10.3, with complete proof): for X X X smooth projective of dimension m m m and H H H ample and globally generated, a small CM module over R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) forces ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 ≤ ( m + 4 ) H m (K_X^2-2c_2(X))\cdot H^{m-2} \le(m+4)H^m ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 ≤ ( m + 4 ) H m , compared with the bound ( m + 1 ) H m (m+1)H^m ( m + 1 ) H m for Ulrich and aCM bundles. With X = Y n × P m − 2 X=Y_n \times\mathbb{P}^{m-2} X = Y n × P m − 2 , the products used in [2 Cor. 5.3] for the graded statement, this gives complete normal local domains with an isolated singularity, of every dimension ≥ 3 \ge3 ≥ 3 , without small CM modules (Corollaries 10.9 and 10.10). Adjoining power series variables, the local analogue of a remark in [2 §5] , gives such domains with non-isolated singularities (Corollary 10.2). In characteristic p p p the completed section rings of smooth varieties always have small CM modules, by Hartshorne–Hochster–Peskine–Szpiro and Shimomoto–Tavanfar, so the characteristic hypothesis in these results is essential even locally (Remark 10.11).
Organization. Sections 2–8 recall definitions and statements, with at most brief indications of proof: the three questions and the tower (§ 2 ), the Ulrich obstruction (§ 3 ), the examples (§ 4 ), the aCM obstruction (§ 5 ), the local setting (§ 6 ) and the two local arguments (§ 7 and § 8 ). Sections 9 and 10 contain the new material, with complete proofs: a single variance identity on P 2 \mathbb{P}^2 P 2 from which the three obstructions are read off (§ 9 ), and the extension of the local obstruction to every dimension (§ 10 ). Section 11 records provenance and status, and Section 12 open questions.
Conventions. We work over C \mathbb{C} C . Surfaces are smooth, connected and projective. For a coherent sheaf F \mathcal{F} F and a divisor H H H , F ( t ) = F ⊗ O ( t H ) \mathcal{F}(t)=\mathcal{F}\otimes\mathcal{O}(tH) F ( t ) = F ⊗ O ( t H ) . For a torsion-free sheaf G \mathcal{G} G of rank ρ \rho ρ on a polarized surface ( Y , H ) (Y,H) ( Y , H ) we use the slope μ H ( G ) = c 1 ( G ) ⋅ H / ρ \mu_H(\mathcal{G})=c_1(\mathcal{G})\cdot H/\rho μ H ( G ) = c 1 ( G ) ⋅ H / ρ and the normalized slope μ ( G ) = μ H ( G ) / H 2 \mu(\mathcal{G})=\mu_H(\mathcal{G})/H^2 μ ( G ) = μ H ( G ) / H 2 ; the discriminant is Δ ( G ) = 2 ρ c 2 − ( ρ − 1 ) c 1 2 \Delta(\mathcal{G})=2\rho c_2-(\rho-1)c_1^2 Δ ( G ) = 2 ρ c 2 − ( ρ − 1 ) c 1 2 . Bogomolov’s inequality says Δ ≥ 0 \Delta\ge0 Δ ≥ 0 for μ H \mu_H μ H -semistable torsion-free sheaves [16 Thm. 3.4.1] .
Three existence questions
This section states the three questions Q1–Q3 behind the levels (R1)–(R3) of the tower (Figure 3 ) and the implications between them.
Ulrich bundles
Let X ⊂ P N X \subset\mathbb{P}^N X ⊂ P N be a smooth projective variety of dimension m m m and H = O X ( 1 ) H=\mathcal{O}_X(1) H = O X ( 1 ) .
Definition 2.1. A vector bundle E \mathcal{E} E on X X X is Ulrich if H i ( X , E ( − p ) ) = 0 H^i(X,\mathcal{E}(-p))=0 H i ( X , E ( − p )) = 0 for all i i i and all 1 ≤ p ≤ m 1 \le p \le m 1 ≤ p ≤ m .
Proposition 2.2 ([ESW, Prop. 2.1, Cor. 2.2]). For a vector bundle E \mathcal{E} E on X X X the following are equivalent:
(i) E \mathcal{E} E is Ulrich;
(ii) for some (equivalently, every) finite linear projection π : X → P m \pi:X \to\mathbb{P}^m π : X → P m , the sheaf π ∗ E \pi_*\mathcal{E} π ∗ E is trivial, π ∗ E ≅ O P m ⊕ deg X \pi_*\mathcal{E} \cong\mathcal{O}_{\mathbb{P}^m}^{\oplus\deg X} π ∗ E ≅ O P m ⊕ d e g X ;
(iii) the module Γ ∗ ( E ) = ⨁ t H 0 ( X , E ( t ) ) \Gamma_*(\mathcal{E})=\bigoplus_t H^0(X,\mathcal{E}(t)) Γ ∗ ( E ) = ⨁ t H 0 ( X , E ( t )) is Cohen–Macaulay of dimension m + 1 m+1 m + 1 and has a linear minimal free resolution over S = C [ x 0 , … , x N ] S=\mathbb{C}[x_0,\ldots,x_N] S = C [ x 0 , … , x N ] , generated in degree 0.
On a surface, E \mathcal{E} E is Ulrich if and only if it is aCM (Definition 2.3 ) and χ ( E ( t ) ) = 1 2 rk E ⋅ H 2 ( t + 1 ) ( t + 2 ) \chi(\mathcal{E}(t))=\frac{1}{2}\operatorname{rk}\mathcal{E}\cdot H^2(t+1)(t+2) χ ( E ( t )) = 2 1 rk E ⋅ H 2 ( t + 1 ) ( t + 2 ) .
Question Q1 (Eisenbud–Schreyer–Weyman, 2003 [ESW, p. 543]). Does every embedded projective variety X ⊂ P N X \subset\mathbb{P}^N X ⊂ P N carry an Ulrich sheaf?
As in [1 §1] , following the restatement of Hanselka and Kummer [15 Problem 4.3] , the sheaf is required to have support X X X ; on smooth surfaces such sheaves are locally free [1 Lemma 9.1] . Since ( X , ℓ H ) (X,\ell H) ( X , ℓ H ) carries Ulrich bundles of rank two for ℓ ≫ 0 \ell\gg0 ℓ ≫ 0 , for every smooth surface X X X and ample H H H [CoHu, Thm. 1.2], Question Q1 concerns a fixed embedding. Following [4 ] , uc ( X , H ) \operatorname{uc}(X,H) uc ( X , H ) denotes the smallest rank of an Ulrich bundle, and uc = ∞ \operatorname{uc}=\infty uc = ∞ if there is none.
aCM bundles and graded MCM modules
Let Y Y Y be a smooth surface and H H H a divisor such that ∣ H ∣ |H| ∣ H ∣ is base-point-free and φ H \varphi_H φ H is finite; equivalently, H H H is ample and globally generated.
Definition 2.3. A vector bundle E \mathcal{E} E on Y Y Y is arithmetically Cohen–Macaulay (aCM) with respect to H H H if H 1 ( Y , E ( t ) ) = 0 H^1(Y,\mathcal{E}(t))=0 H 1 ( Y , E ( t )) = 0 for every t ∈ Z t\in\mathbb{Z} t ∈ Z .
Let R = R ( Y , H ) = ⨁ t ≥ 0 H 0 ( Y , O Y ( t H ) ) R=R(Y,H)=\bigoplus_{t\geq0}H^0(Y,\mathcal{O}_Y(tH)) R = R ( Y , H ) = ⨁ t ≥ 0 H 0 ( Y , O Y ( t H )) and m = R + \mathfrak{m}=R_+ m = R + . By [2 Lemma 1.2] , R R R is a finitely generated normal graded C \mathbb{C} C -domain of dimension 3 with Proj R = Y \operatorname{Proj}R=Y Proj R = Y , and its punctured spectrum is smooth.
Lemma 2.4 ([2 Lemma 3.2] ). The assignments M ↦ M ~ M\mapsto\widetilde{M} M ↦ M and E ↦ Γ ∗ ( E ) = ⨁ t ∈ Z H 0 ( Y , E ( t ) ) \mathcal{E}\mapsto\Gamma_*(\mathcal{E})=\bigoplus_{t\in\mathbb{Z}}H^0(Y,\mathcal{E}(t)) E ↦ Γ ∗ ( E ) = ⨁ t ∈ Z H 0 ( Y , E ( t )) are mutually inverse bijections between graded MCM R R R -modules and aCM bundles of positive rank on ( Y , H ) (Y,H) ( Y , H ) .
Question Q2 (Eisenbud–Schreyer, after [1 ] ; see [2 Acknowledgements] ). Do the polarized surfaces ( Y , H ) (Y,H) ( Y , H ) constructed in [1 ] carry an aCM bundle; equivalently (Lemma 2.4), do their section rings have a graded MCM module?
They also asked for the Boij–Söderberg cone of cohomology tables of aCM bundles on these surfaces. For H 2 < σ ( Y ) H^2<\sigma(Y) H 2 < σ ( Y ) , the cone spanned by the cohomology tables of vector bundles on ( Y , H ) (Y,H) ( Y , H ) meets the tables with vanishing h 1 h^1 h 1 -row only in 0 [2 Rem. 6.2] ; see item 4 of § 12 .
Small Cohen–Macaulay modules
Let ( R , m ) (R,\mathfrak{m}) ( R , m ) be a Noetherian local ring. A finitely generated R R R -module M ≠ 0 M\ne0 M = 0 with depth R M = dim R \operatorname{depth}_R M=\dim R depth R M = dim R is called a small (or maximal ) Cohen–Macaulay module.
Question Q3 (Hochster, 1973 [17 , 18 ] ). Does every complete Noetherian local domain have a small Cohen–Macaulay module?
Hochster’s small Cohen–Macaulay conjecture predicts a positive answer. It holds in dimension ≤ 2 \leq2 ≤ 2 . In dimension three and characteristic p > 0 p>0 p > 0 , every finitely generated N \mathbb{N} N -graded domain R R R with R 0 R_0 R 0 a perfect field has a graded MCM module (Hartshorne, Hochster, Peskine–Szpiro [19 , 28 ] ), and the conjecture is known for further classes, such as three-dimensional complete weakly F F F -split rings [29 ] . In any characteristic there are existence criteria in terms of deficiency modules [32 Thm. 3.2] , [Xie1, Xie2]. In characteristic zero the conjecture was open from dimension three on, even for localizations of affine domains at maximal ideals [6 §1.3] . The main motivation is Serre’s positivity of intersection multiplicities [18 ] .
The tower
Lemma 2.5. Let ( Y , H ) (Y,H) ( Y , H ) be as above and R ^ \widehat{R} R the m \mathfrak{m} m -adic completion of R = R ( Y , H ) R=R(Y,H) R = R ( Y , H ) .
(a) If H H H is very ample, every Ulrich bundle on ( Y , H ) (Y,H) ( Y , H ) is aCM.
(b) If M M M is a graded MCM R R R -module, then M ^ = M ⊗ R R ^ \widehat{M}=M\otimes_R\widehat{R} M = M ⊗ R R is a small CM R ^ \widehat{R} R -module.
Part (a) follows from Proposition 2.2 (ii), and (b) from the faithful flatness of R m → R ^ R_{\mathfrak{m}}\to\widehat{R} R m → R . Thus a complete local ring R ^ ( Y , H ) \widehat{R}(Y,H) R ( Y , H ) without small CM modules has no graded MCM module over R R R , no aCM bundle and no Ulrich bundle.
Degree, signature, and the Ulrich obstruction
For a smooth surface Y Y Y , Noether’s formula 12 χ ( O Y ) = K Y 2 + c 2 ( Y ) 12\chi(\mathcal{O}_Y)=K_Y^2+c_2(Y) 12 χ ( O Y ) = K Y 2 + c 2 ( Y ) gives
σ ( Y ) : = K Y 2 − 8 χ ( O Y ) = K Y 2 − 2 c 2 ( Y ) 3 . \sigma(Y):=K_Y^2-8\chi(\mathcal{O}_Y)=\frac{K_Y^2-2c_2(Y)}{3}. σ ( Y ) := K Y 2 − 8 χ ( O Y ) = 3 K Y 2 − 2 c 2 ( Y ) .
the topological signature of Y Y Y . If σ ( Y ) > 0 \sigma(Y)>0 σ ( Y ) > 0 and Y ≇ P 2 Y\not\cong\mathbb{P}^2 Y ≅ P 2 , then Y Y Y is of general type and, by the Bogomolov–Miyaoka–Yau inequality, σ ( Y ) ≤ χ ( O Y ) \sigma(Y)\leq\chi(\mathcal{O}_Y) σ ( Y ) ≤ χ ( O Y ) , with equality exactly for ball quotients; see [2 Rem. 1.6] .
In a 2017 lecture [3 ] Beauville observed that on a surface of Picard rank one Bogomolov’s inequality for an Ulrich bundle gives H 2 ≥ σ ( S ) H^2\geq\sigma(S) H 2 ≥ σ ( S ) , and noted that it seems hard to obtain a counterexample to Question Q1 this way. The Picard rank hypothesis is not needed.
Proposition 3.1 ([1 Prop. 2.1] ). Let ( S , H ) (S,H) ( S , H ) be a smooth surface with H H H very ample. If S S S carries an Ulrich bundle, then
H 2 ≥ K S 2 − 8 χ ( O S ) = σ ( S ) . H^2\geq K_S^2-8\chi(\mathcal{O}_S)=\sigma(S). H 2 ≥ K S 2 − 8 χ ( O S ) = σ ( S ) .
Proposition 3.2 ([24 Lemma 3.1] ). Let X X X be smooth of dimension m ≥ 2 m\geq2 m ≥ 2 and H H H very ample. If
( m + 1 ) H m < ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 , (m+1)H^m < (K_X^2-2c_2(X))\cdot H^{m-2}, ( m + 1 ) H m < ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 , then neither ( X , H ) (X,H) ( X , H ) nor any of its successive smooth hyperplane sections (down to dimension 2 2 2 ) carries an Ulrich bundle.
For m = 2 m=2 m = 2 this is Proposition 3.1. The difference of the two sides, the excess I ( X , H ) I(X,H) I ( X , H ) below, behaves simply under hypersurface sections.
Lemma 3.3 ([2 Rem. 5.7] ). Let X X X be smooth of dimension m ≥ 2 m\geq2 m ≥ 2 and H H H ample, and set I ( X , H ) = ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 − ( m + 1 ) H m I(X,H)=(K_X^2-2c_2(X))\cdot H^{m-2}-(m+1)H^m I ( X , H ) = ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 − ( m + 1 ) H m . If m ≥ 3 m\geq3 m ≥ 3 and X ′ ∈ ∣ a H ∣ X'\in\lvert aH\rvert X ′ ∈ ∣ a H ∣ is smooth and H ′ = H ∣ X ′ H'=H|_{X'} H ′ = H ∣ X ′ , then
I ( X ′ , H ′ ) = a [ I ( X , H ) − ( a 2 − 1 ) H m ] . I(X',H')=a\left[I(X,H)-(a^2-1)H^m\right]. I ( X ′ , H ′ ) = a [ I ( X , H ) − ( a 2 − 1 ) H m ] .
Hirzebruch–Kummer covers and the examples
The examples of [1 ] are Hirzebruch’s surfaces attached to line arrangements [14 BHH] . Let A = { L 1 , … , L k } \mathcal{A}=\{L_1,\ldots,L_k\} A = { L 1 , … , L k } be an arrangement of k ≥ 4 k\geq4 k ≥ 4 lines in P 2 \mathbb{P}^2 P 2 , not all concurrent, with defining forms ℓ i \ell_i ℓ i . Let t r t_r t r be the number of points lying on exactly r r r lines, and put
f 0 = ∑ r ≥ 2 t r , f 1 = ∑ r ≥ 2 r t r , T = ∑ r ≥ 3 t r . f_0=\sum_{r\geq2}t_r,\qquad f_1=\sum_{r\geq2}rt_r,\qquad T=\sum_{r\geq3}t_r. f 0 = r ≥ 2 ∑ t r , f 1 = r ≥ 2 ∑ r t r , T = r ≥ 3 ∑ t r .
The forms ℓ i \ell_i ℓ i define a linear embedding P 2 ↪ P k − 1 \mathbb{P}^2\hookrightarrow\mathbb{P}^{k-1} P 2 ↪ P k − 1 with image a plane Λ \Lambda Λ . For n ≥ 2 n\geq2 n ≥ 2 let
Φ n : P k − 1 ⟶ P k − 1 , [ z i ] ⟼ [ z i n ] , \Phi_n:\mathbb{P}^{k-1}\longrightarrow\mathbb{P}^{k-1},\quad[z_i]\longmapsto[z_i^n], Φ n : P k − 1 ⟶ P k − 1 , [ z i ] ⟼ [ z i n ] ,
and
X n = Φ n − 1 ( Λ ) , π n = Φ n ∣ X n : X n ⟶ Λ ≅ P 2 . X_n=\Phi_n^{-1}(\Lambda),\qquad\pi_n=\left.\Phi_n\right|_{X_n}:X_n\longrightarrow\Lambda\cong\mathbb{P}^2. X n = Φ n − 1 ( Λ ) , π n = Φ n ∣ X n : X n ⟶ Λ ≅ P 2 .
Proposition 4.1 ([1 Props. 3.2, 3.4, 3.6, Cors. 3.5, 3.7] ).
(a) X n X_n X n is a normal integral complete intersection, cut out by k − 3 k-3 k − 3 hypersurfaces of degree n n n ; π n \pi_n π n is finite of degree n k − 3 n^{k-3} n k − 3 , with π n ∗ O ( 1 ) = O X n ( n ) \pi_n^*\mathcal{O}(1)=\mathcal{O}_{X_n}(n) π n ∗ O ( 1 ) = O X n ( n ) .
(b) X n X_n X n is smooth over points of P 2 \mathbb{P}^2 P 2 lying on at most two lines. Over an r r r -fold point p p p with r ≥ 3 r\geq3 r ≥ 3 it has n k − r − 1 n^{k-r-1} n k − r − 1 singular points, each with completed local ring isomorphic to that of the vertex of the affine cone over a smooth complete-intersection curve C n , r ⊂ P r − 1 C_{n,r}\subset\mathbb{P}^{r-1} C n , r ⊂ P r − 1 cut out by r − 2 r-2 r − 2 diagonal hypersurfaces of degree n n n ; deg C n , r = n r − 2 \deg C_{n,r}=n^{r-2} deg C n , r = n r − 2 and 2 g ( C n , r ) − 2 = n r − 2 ( ( r − 2 ) n − r ) 2g(C_{n,r})-2=n^{r-2}((r-2)n-r) 2 g ( C n , r ) − 2 = n r − 2 (( r − 2 ) n − r ) .
(c) Let Σ n = Sing ( X n ) r e d \Sigma_n=\operatorname{Sing}(X_n)_{\mathrm{red}} Σ n = Sing ( X n ) red be the reduced singular locus. Blowing up Σ n \Sigma_n Σ n gives a smooth surface Y n Y_n Y n ; the exceptional curves are copies of C n , r C_{n,r} C n , r , with self-intersection − n r − 2 -n^{r-2} − n r − 2 , and I Σ n O Y n = O Y n ( − E n ) \mathcal{I}_{\Sigma_n}\mathcal{O}_{Y_n}=\mathcal{O}_{Y_n}(-E_n) I Σ n O Y n = O Y n ( − E n ) for the reduced exceptional divisor E n E_n E n .
(d) With A n A_n A n the pull-back of O X n ( 1 ) \mathcal{O}_{X_n}(1) O X n ( 1 ) and E n , r E_{n,r} E n , r the exceptional curves over r r r -fold points,
A n 2 = n k − 3 , A n ⋅ E n = 0 , E n 2 = − T n k − 3 , A_n^2=n^{k-3},\qquad A_n\cdot E_n=0,\qquad E_n^2=-Tn^{k-3}, A n 2 = n k − 3 , A n ⋅ E n = 0 , E n 2 = − T n k − 3 ,
K Y n = ( ( k − 3 ) n − k ) A n − ∑ r ≥ 3 ( ( r − 2 ) ( n − 1 ) − 1 ) E n , r . K_{Y_n}=\bigl((k-3)n-k\bigr)A_n-\sum_{r\geq3}\bigl((r-2)(n-1)-1\bigr)E_{n,r}. K Y n = ( ( k − 3 ) n − k ) A n − r ≥ 3 ∑ ( ( r − 2 ) ( n − 1 ) − 1 ) E n , r .
(e) The signature is
σ ( Y n ) = n k − 3 3 ( Q A n 2 + R A ) , Q A = 3 − k + ∑ r ≥ 2 ( r − 2 ) t r , R A = k + 3 t 2 − f 1 − f 0 . \sigma(Y_n)=\frac{n^{k-3}}{3}\left(Q_A n^2+R_A\right),\qquad Q_A=3-k+\sum_{r\geq2}(r-2)t_r,\qquad R_A=k+3t_2-f_1-f_0. σ ( Y n ) = 3 n k − 3 ( Q A n 2 + R A ) , Q A = 3 − k + r ≥ 2 ∑ ( r − 2 ) t r , R A = k + 3 t 2 − f 1 − f 0 .
The three arrangements used below are the Hesse arrangement of the 12 lines through pairs of the 9 flexes of a smooth cubic (t 4 = 9 t_4=9 t 4 = 9 , t 2 = 12 t_2=12 t 2 = 12 , Q = 9 Q=9 Q = 9 , R = − 33 R=-33 R = − 33 ), its dual (k = 9 k=9 k = 9 , t 3 = 12 t_3=12 t 3 = 12 , Q = 6 Q=6 Q = 6 , R = − 39 R=-39 R = − 39 ), and the complete quadrangle of the six lines through four general points (t 3 = 4 t_3=4 t 3 = 4 , t 2 = 3 t_2=3 t 2 = 3 , Q = 1 Q=1 Q = 1 , R = − 10 R=-10 R = − 10 ), see Figure 4 . In all three, σ ( Y n ) \sigma(Y_n) σ ( Y n ) grows like n k − 1 n^{k-1} n k − 1 , while ( a A n − E n ) 2 = ( a 2 − T ) n k − 3 (aA_n-E_n)^2=(a^2-T)n^{k-3} ( a A n − E n ) 2 = ( a 2 − T ) n k − 3 grows like n k − 3 n^{k-3} n k − 3 .
Figure 4. The complete quadrangle: k = 6 k=6 k = 6 , four triple points (filled), three double points (open); Q = 1 Q=1 Q = 1 , R = − 10 R=-10 R = − 10 .
Lemma 4.2 ([1 Lemma 4.1] ). Let X X X be projective with O X ( 1 ) \mathcal{O}_X(1) O X ( 1 ) very ample, I ⊂ O X \mathcal{I}\subset\mathcal{O}_X I ⊂ O X a coherent ideal, ρ : Y = Bl I X → X \rho:Y=\operatorname{Bl}_{\mathcal{I}}X\to X ρ : Y = Bl I X → X , A = ρ ∗ O X ( 1 ) A=\rho^*\mathcal{O}_X(1) A = ρ ∗ O X ( 1 ) , and I O Y = O Y ( − E ) \mathcal{I}\mathcal{O}_Y=\mathcal{O}_Y(-E) I O Y = O Y ( − E ) . Suppose I ( ν ) \mathcal{I}(\nu) I ( ν ) is generated by global sections for some ν ≥ 1 \nu\geq1 ν ≥ 1 . Then the class ( ν + 1 ) A − E (\nu+1)A-E ( ν + 1 ) A − E is very ample on Y Y Y .
Proposition 4.3 ([1 Props. 5.1, 5.3] ). For the Hesse arrangement and every n ≥ 2 n\geq2 n ≥ 2 , the class H n = 4 A n − E n H_n=4A_n-E_n H n = 4 A n − E n is very ample on Y n Y_n Y n . The class 3 A n − E n 3A_n-E_n 3 A n − E n is nef with ( 3 A n − E n ) 2 = 0 (3A_n-E_n)^2=0 ( 3 A n − E n ) 2 = 0 , so 4 is the smallest integral coefficient a a a for which a A n − E n aA_n-E_n a A n − E n can be ample.
The proof applies Lemma 4.2 with ν = 3 \nu=3 ν = 3 : the four cubic monomials attached to the singular triangles of the Hesse pencil generate I Σ n ( 3 ) \mathcal{I}_{\Sigma_n}(3) I Σ n ( 3 ) , for every n n n .
Theorem 4.4 ([1 Thms. 1.1, 1.2, 5.6, §5.1] ). For every n ≥ 3 n\geq3 n ≥ 3 the Hesse pair ( Y n , H n = 4 A n − E n ) (Y_n,H_n=4A_n-E_n) ( Y n , H n = 4 A n − E n ) satisfies
H n 2 = 7 n 9 < ( 3 n 2 − 11 ) n 9 = σ ( Y n ) , H_n^2=7n^9<(3n^2-11)n^9=\sigma(Y_n), H n 2 = 7 n 9 < ( 3 n 2 − 11 ) n 9 = σ ( Y n ) , so it carries no Ulrich bundle of any rank: uc ( Y n , H n ) = ∞ \operatorname{uc}(Y_n,H_n)=\infty uc ( Y n , H n ) = ∞ . The surface Y 3 Y_3 Y 3 is a compact ball quotient, with K Y 3 = 15 A − 3 E K_{Y_3}=15A-3E K Y 3 = 15 A − 3 E ample, K 2 = 144 ⋅ 3 9 = 3 c 2 K^2=144\cdot3^9=3c_2 K 2 = 144 ⋅ 3 9 = 3 c 2 , χ ( O Y 3 ) = 16 ⋅ 3 9 \chi(\mathcal{O}_{Y_3})=16\cdot3^9 χ ( O Y 3 ) = 16 ⋅ 3 9 . The surfaces Y n Y_n Y n are pairwise non-isomorphic, since K Y n 2 = 9 n 9 ( n − 1 ) ( 5 n − 7 ) K_{Y_n}^2=9n^9(n-1)(5n-7) K Y n 2 = 9 n 9 ( n − 1 ) ( 5 n − 7 ) is strictly increasing.
The same mechanism works for other arrangements admitting low-degree generators of the singular ideal ([1 Thm. 1.3] ); Table 2 lists the three families treated in [1 §§5, 7, 8] . Lopez combined Proposition 3.2 with the Hesse pairs.
Table 2. The three families of [1 ] ; in each, H n 2 / σ ( Y n ) → 0 H_n^2/\sigma(Y_n)\to0 H n 2 / σ ( Y n ) → 0 .
Theorem 4.5 ([24 Thm. 1] ). Let m ≥ 3 m\geq3 m ≥ 3 , let ( Y n , H n ) (Y_n,H_n) ( Y n , H n ) be the Hesse pairs, Z Z Z a smooth projective ( m − 2 ) (m-2) ( m − 2 ) -dimensional variety with very ample M M M , X = Y n × Z X=Y_n\times Z X = Y n × Z , H = H n ⊠ M H=H_n\boxtimes M H = H n ⊠ M , and X ′ ∈ ∣ a H ∣ X'\in\lvert aH\rvert X ′ ∈ ∣ a H ∣ smooth with a ≥ 1 a\geq1 a ≥ 1 . For n ≫ 0 n\gg0 n ≫ 0 , the pairs ( X , H ) (X,H) ( X , H ) , ( X ′ , H ∣ X ′ ) (X',H|_{X'}) ( X ′ , H ∣ X ′ ) and all their successive hyperplane sections down to dimension 2 carry no Ulrich bundle; X ′ X' X ′ is not isomorphic to a nontrivial product.
No aCM bundles: the graded theorem
Throughout this section ( Y , H ) (Y,H) ( Y , H ) is a smooth surface with ∣ H ∣ |H| ∣ H ∣ base-point-free and φ H \varphi_H φ H finite; d = H 2 d=H^2 d = H 2 , κ = K Y ⋅ H \kappa=K_Y\cdot H κ = K Y ⋅ H .
Theorem 5.1 ([2 Thm. 1.1] ). If ( Y , H ) (Y,H) ( Y , H ) carries a nonzero aCM vector bundle, then H 2 ≥ σ ( Y ) H^2\geq\sigma(Y) H 2 ≥ σ ( Y ) . Equivalently, if H 2 < σ ( Y ) H^2<\sigma(Y) H 2 < σ ( Y ) there is no nonzero aCM bundle on ( Y , H ) (Y,H) ( Y , H ) , of any rank, semistable or not.
Theorem 5.2 ([2 Thm. 1.3] ). If H 2 < σ ( Y ) H^2<\sigma(Y) H 2 < σ ( Y ) , then R ( Y , H ) R(Y,H) R ( Y , H ) has no graded MCM module, depth m R = 2 \operatorname{depth}_m R=2 depth m R = 2 , and the vertex of Spec R \operatorname{Spec} R Spec R is an isolated non-Cohen–Macaulay singularity. If H H H is very ample, the homogeneous coordinate ring of Y ⊂ P N Y\subset\mathbb{P}^N Y ⊂ P N has no graded MCM module either.
The proof transports the problem to P 2 \mathbb{P}^2 P 2 , where aCM bundles split. Composing φ H \varphi_H φ H with a general linear projection gives a finite flat morphism π : Y → P 2 \pi:Y\to\mathbb{P}^2 π : Y → P 2 with π ∗ O ( 1 ) = O Y ( H ) \pi^*\mathcal{O}(1)=\mathcal{O}_Y(H) π ∗ O ( 1 ) = O Y ( H ) , deg π = d \deg\pi=d deg π = d , and H i ( Y , G ( t ) ) = H i ( P 2 , ( π ∗ G ) ( t ) ) H^i(Y,\mathcal{G}(t))=H^i(\mathbb{P}^2,(\pi_*\mathcal{G})(t)) H i ( Y , G ( t )) = H i ( P 2 , ( π ∗ G ) ( t )) for all i i i and t t t .
Lemma 5.3 ([2 Lemmas 2.1, 2.2] ). A vector bundle E \mathcal{E} E on Y Y Y is aCM if and only if π ∗ E ≅ ⨁ j = 1 N O P 2 ( a j ) \pi_*\mathcal{E}\cong\bigoplus_{j=1}^N\mathcal{O}_{\mathbb{P}^2}(a_j) π ∗ E ≅ ⨁ j = 1 N O P 2 ( a j ) , where N = d ⋅ rk E N=d\cdot\operatorname{rk}\mathcal{E} N = d ⋅ rk E . For a coherent sheaf G \mathcal{G} G of positive rank on Y Y Y , with h h h the line class of P 2 \mathbb{P}^2 P 2 ,
μ h ( π ∗ G ) = μ H ( G ) − κ / 2 d − 3 2 . \mu_h(\pi_*\mathcal{G})=\frac{\mu_H(\mathcal{G})-\kappa/2}{d}-\frac{3}{2}. μ h ( π ∗ G ) = d μ H ( G ) − κ /2 − 2 3 .
Let E \mathcal{E} E be an aCM bundle of rank r r r with Harder–Narasimhan factors Q i \mathcal{Q}_i Q i of ranks r i r_i r i , weights w i = r i / r w_i=r_i/r w i = r i / r and slopes μ H ( Q 1 ) > ⋯ > μ H ( Q s ) \mu_H(\mathcal{Q}_1)>\cdots>\mu_H(\mathcal{Q}_s) μ H ( Q 1 ) > ⋯ > μ H ( Q s ) . Then π ∗ E ≅ ⨁ j = 1 N O ( a j ) \pi_*\mathcal{E}\cong\bigoplus_{j=1}^N\mathcal{O}(a_j) π ∗ E ≅ ⨁ j = 1 N O ( a j ) has a filtration with successive quotients π ∗ Q i \pi_*\mathcal{Q}_i π ∗ Q i , of slopes b i = ( μ H ( Q i ) − κ / 2 ) / d − 3 / 2 b_i=(\mu_H(\mathcal{Q}_i)-\kappa/2)/d-3/2 b i = ( μ H ( Q i ) − κ /2 ) / d − 3/2 , b 1 > ⋯ > b s b_1>\cdots>b_s b 1 > ⋯ > b s . Put a ˉ = 1 N ∑ j a j = ∑ i w i b i \bar a=\frac{1}{N}\sum_j a_j=\sum_iw_ib_i a ˉ = N 1 ∑ j a j = ∑ i w i b i and
v a = 1 N ∑ j ( a j − a ˉ ) 2 , v b = ∑ i w i ( b i − a ˉ ) 2 . v_a=\frac{1}{N}\sum_j(a_j-\bar a)^2,\qquad v_b=\sum_iw_i(b_i-\bar a)^2. v a = N 1 j ∑ ( a j − a ˉ ) 2 , v b = i ∑ w i ( b i − a ˉ ) 2 .
Lemma 5.4 ([2 Lemma 2.4] ). Let V = ⨁ j = 1 N O P 2 ( a j ) V=\bigoplus_{j=1}^N\mathcal{O}_{\mathbb{P}^2}(a_j) V = ⨁ j = 1 N O P 2 ( a j ) , and let 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V s = V 0=V_0\subset V_1\subset\cdots\subset V_s=V 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V s = V be a filtration by subsheaves, with successive quotients of ranks n i > 0 n_i>0 n i > 0 and slopes b 1 ≥ b 2 ≥ ⋯ ≥ b s b_1\geq b_2\geq\cdots\geq b_s b 1 ≥ b 2 ≥ ⋯ ≥ b s . Then for every c ∈ R c\in\mathbb{R} c ∈ R
∑ i = 1 s n i N ( b i + c ) 2 ≤ 1 N ∑ j = 1 N ( a j + c ) 2 . \sum_{i=1}^s\frac{n_i}{N}(b_i+c)^2\leq\frac{1}{N}\sum_{j=1}^N(a_j+c)^2. i = 1 ∑ s N n i ( b i + c ) 2 ≤ N 1 j = 1 ∑ N ( a j + c ) 2 .
In particular v b ≤ v a v_b\leq v_a v b ≤ v a .
For a torsion-free sheaf F \mathcal{F} F of rank ρ \rho ρ on Y Y Y write ν = c 1 ( F ) / ρ ∈ NS ( Y ) Q \nu=c_1(\mathcal{F})/\rho\in\operatorname{NS}(Y)_{\mathbb{Q}} ν = c 1 ( F ) / ρ ∈ NS ( Y ) Q , μ H = ν ⋅ H \mu_H=\nu\cdot H μ H = ν ⋅ H , δ = Δ ( F ) / ρ 2 \delta=\Delta(\mathcal{F})/\rho^2 δ = Δ ( F ) / ρ 2 , β = ν − K Y / 2 \beta=\nu-K_Y/2 β = ν − K Y /2 , and α = β − β ⋅ H d H \alpha=\beta-\frac{\beta\cdot H}{d}H α = β − d β ⋅ H H , so that α ⋅ H = 0 \alpha\cdot H=0 α ⋅ H = 0 and α 2 ≤ 0 \alpha^2\leq0 α 2 ≤ 0 .
Lemma 5.5 ([2 (2)] ). For all t ∈ Z t\in\mathbb{Z} t ∈ Z ,
χ ( F ( t ) ) ρ = d 2 ( t + μ H − κ / 2 d ) 2 − σ 8 + α 2 − δ 2 . \frac{\chi(\mathcal{F}(t))}{\rho}
=\frac{d}{2}\left(t+\frac{\mu_H-\kappa/2}{d}\right)^2-\frac{\sigma}{8}+\frac{\alpha^2-\delta}{2}. ρ χ ( F ( t )) = 2 d ( t + d μ H − κ /2 ) 2 − 8 σ + 2 α 2 − δ .
Proposition 5.6 (Defect identity, [2 Prop. 2.6] ). Let E \mathcal{E} E be a nonzero aCM bundle with Harder–Narasimhan factors Q i Q_i Q i and weights w i w_i w i , and let v a , v b v_a,v_b v a , v b be as above. Then
d − σ 4 = d ( v a − v b ) + ∑ i w i ( δ ( Q i ) − α ( Q i ) 2 ) . \frac{d-\sigma}{4}=d\left(v_a-v_b\right)+\sum_i w_i\left(\delta(Q_i)-\alpha(Q_i)^2\right). 4 d − σ = d ( v a − v b ) + i ∑ w i ( δ ( Q i ) − α ( Q i ) 2 ) . Every term on the right is ≥ 0 \ge0 ≥ 0 : by Lemma 5.4, by Bogomolov’s inequality for the semistable factors, and by the Hodge index theorem. In particular d ≥ σ d\ge\sigma d ≥ σ , which proves Theorem 5.1.
Corollary 5.7 ([2 Cor. 1.4] ). For n ≥ 3 n\ge3 n ≥ 3 , the section ring R ( Y n , 4 A n − E n ) R(Y_n,4A_n-E_n) R ( Y n , 4 A n − E n ) of the Hesse pair is a three-dimensional normal graded C \mathbb{C} C -domain with an isolated non-Cohen–Macaulay singularity at the vertex, and it has no nonzero finitely generated graded MCM module. The same holds for the homogeneous coordinate ring of Y n ⊂ P N Y_n\subset\mathbb{P}^N Y n ⊂ P N .
Corollary 5.8 ([2 Cor. 4.3] ). For the complete quadrangle and n ≥ 6 n\ge6 n ≥ 6 , let H = 3 A n − E n H=3A_n-E_n H = 3 A n − E n on the quadrangle surface Y n Y_n Y n . Then K Y n ∼ ( n − 2 ) H K_{Y_n}\sim(n-2)H K Y n ∼ ( n − 2 ) H (linear equivalence), so R = R ( Y n , H ) R=R(Y_n,H) R = R ( Y n , H ) is quasi-Gorenstein with ω R ≅ R ( n − 2 ) \omega_R\cong R(n-2) ω R ≅ R ( n − 2 ) , and it has no graded MCM module. For n = 6 n=6 n = 6 : H 2 = 1080 < σ = 1872 H^2=1080<\sigma=1872 H 2 = 1080 < σ = 1872 , K 2 = 17280 K^2=17280 K 2 = 17280 , c 2 = 5832 c_2=5832 c 2 = 5832 , χ = 1926 \chi=1926 χ = 1926 , ω R ≅ R ( 4 ) \omega_R\cong R(4) ω R ≅ R ( 4 ) .
Proposition 5.9 ([2 Prop. 5.1] ). Let X X X be smooth of dimension m ≥ 2 m\ge2 m ≥ 2 with ∣ H ∣ |H| ∣ H ∣ base-point-free and ϕ H \phi_H ϕ H finite. If X X X carries a nonzero aCM bundle E \mathcal{E} E , that is, one with H i ( X , E ( t ) ) = 0 H^i(X,\mathcal{E}(t))=0 H i ( X , E ( t )) = 0 for 1 ≤ i ≤ m − 1 1\le i\le m-1 1 ≤ i ≤ m − 1 and all t t t , then
( m + 1 ) H m ≥ ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 . (m+1)H^m\ge(K_X^2-2c_2(X))\cdot H^{m-2}. ( m + 1 ) H m ≥ ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 . With X = Y n × P m − 2 X=Y_n\times\mathbb{P}^{m-2} X = Y n × P m − 2 and H = H n ⊠ O ( 1 ) H=H_n\boxtimes\mathcal{O}(1) H = H n ⊠ O ( 1 ) the inequality fails as soon as 7 ( m − 1 ) 2 < 3 n 2 − 11 7(m-1)^2<3n^2-11 7 ( m − 1 ) 2 < 3 n 2 − 11 , and the Segre products R ( Y n , H n ) # C [ s 0 , … , s m − 2 ] R(Y_n,H_n)\#\mathbb{C}[s_0,\ldots,s_{m-2}] R ( Y n , H n ) # C [ s 0 , … , s m − 2 ] are normal graded domains of every dimension m + 1 ≥ 3 m+1\ge3 m + 1 ≥ 3 without graded MCM modules [2 Cor. 5.3] .
From graded to local
[2 ] stops at level ( R 2 ) (R2) ( R 2 ) and leaves the local question open: Hochster’s conjecture concerns the complete local ring R ^ \widehat{R} R , whose finitely generated modules need not be graded. This section fixes the framework of the two local arguments. Let Y Y Y be a smooth surface and H H H an ample, globally generated divisor; put L = O Y ( H ) L=\mathcal{O}_Y(H) L = O Y ( H ) , R = R ( Y , H ) R=R(Y,H) R = R ( Y , H ) and R ^ \widehat{R} R its completion at the vertex. Three general sections s 0 , s 1 , s 2 s_0,s_1,s_2 s 0 , s 1 , s 2 of L L L define a finite morphism f : Y → P 2 f:Y\to\mathbb{P}^2 f : Y → P 2 with f ∗ O ( 1 ) = L f^*\mathcal{O}(1)=L f ∗ O ( 1 ) = L (the map π \pi π of § 5 ); with S = C [ x 0 , x 1 , x 2 ] → R S=\mathbb{C}[x_0,x_1,x_2]\to R S = C [ x 0 , x 1 , x 2 ] → R , x i ↦ s i x_i\mapsto s_i x i ↦ s i , and A = C [ [ x 0 , x 1 , x 2 ] ] A=\mathbb{C}[[x_0,x_1,x_2]] A = C [[ x 0 , x 1 , x 2 ]] , the ring R ^ = R ⊗ S A \widehat{R}=R\otimes_S A R = R ⊗ S A is finite over A A A .
Proposition 6.1 ([9 Prop. 2.1] , [26 §2] ). R ^ \widehat{R} R is a three-dimensional complete Noetherian normal local domain with residue field C \mathbb{C} C . There is a proper morphism p : W → Spec R ^ p:W\to\operatorname{Spec}\widehat{R} p : W → Spec R from a regular three-dimensional scheme W W W , obtained by base change from Tot ( L − 1 ) → Spec R \operatorname{Tot}(L^{-1})\to\operatorname{Spec}R Tot ( L − 1 ) → Spec R , such that
(a) the scheme-theoretic closed fibre is the zero section Y ⊂ W Y\subset W Y ⊂ W , an effective Cartier divisor, and O W ( − Y ) ∣ Y ≅ L \mathcal{O}_W(-Y)|_Y\cong L O W ( − Y ) ∣ Y ≅ L , i.e. the conormal bundle of Y Y Y is O Y ( H ) \mathcal{O}_Y(H) O Y ( H ) ;
(b) p p p restricts to an isomorphism U : = W ∖ Y ≅ Spec R ^ ∖ { m } U:=W\setminus Y\cong\operatorname{Spec}\widehat{R}\setminus\{\mathfrak{m}\} U := W ∖ Y ≅ Spec R ∖ { m } ;
(c) O W ( − Y ) \mathcal{O}_W(-Y) O W ( − Y ) is p p p -ample.
In the version of [26 ] one also keeps the base: Z = Bl 0 Spec A Z=\operatorname{Bl}_0\operatorname{Spec}A Z = Bl 0 Spec A with exceptional plane D ≅ P 2 D\cong\mathbb{P}^2 D ≅ P 2 , and a finite morphism g : W → Z g:W\to Z g : W → Z with g − 1 ( D ) = Y g^{-1}(D)=Y g − 1 ( D ) = Y , g ∣ Y = f g|_Y=f g ∣ Y = f , L Y = g ∗ I D L_Y=g^*I_D L Y = g ∗ I D and I D ∣ D ≅ O D ( 1 ) I_D|_D\cong\mathcal{O}_D(1) I D ∣ D ≅ O D ( 1 ) .
Lemma 6.2 ([9 Lemmas 2.4, 2.5, §6] ). Let M M M be a finite R ^ \widehat{R} R -module.
(a) If M M M is small Cohen–Macaulay, then M M M is torsion-free of full support and reflexive.
(b) If M M M is reflexive, then M M M is locally free on the punctured spectrum, H m 0 ( M ) = H m 1 ( M ) = 0 H^0_{\mathfrak{m}}(M)=H^1_{\mathfrak{m}}(M)=0 H m 0 ( M ) = H m 1 ( M ) = 0 , and
H m 2 ( M ) ≅ H 1 ( U , M ~ ∣ U ) . H^2_{\mathfrak{m}}(M)\cong H^1(U,\widetilde{M}|_U). H m 2 ( M ) ≅ H 1 ( U , M ∣ U ) .
By (a) and (b), R ^ \widehat{R} R has a small CM module if and only if there is a nonzero reflexive M M M with H 1 ( U , M ~ ∣ U ) = 0 H^1(U,\widetilde{M}|_U)=0 H 1 ( U , M ∣ U ) = 0 . A reflexive M M M is extended across Y Y Y by the reflexive sheaf F = ( p ∗ M ~ ) ∨ ∨ \mathcal{F}=(p_*\widetilde{M})^{\vee\vee} F = ( p ∗ M ) ∨∨ on W W W , with F ∣ U = M ~ ∣ U \mathcal{F}|_U=\widetilde{M}|_U F ∣ U = M ∣ U [9 Lemma 2.4] ; its restriction V = F ∣ Y V=\mathcal{F}|_Y V = F ∣ Y is torsion-free [9 Lemma 2.3] . The extension is not unique, and both local arguments use this freedom.
Lemma 6.3 ([9 Lemma 3.1] for F \mathcal{F} F reflexive; cf. the proof of [26 Lemma 3.1] for F \mathcal{F} F torsion-free). Let F \mathcal{F} F be a torsion-free coherent sheaf on W W W whose restriction V = F ∣ Y V=\mathcal{F}|_Y V = F ∣ Y is torsion-free, and let φ : V → Q \varphi:V\to Q φ : V → Q be a surjection onto a torsion-free sheaf with kernel K \mathcal{K} K . Put i : Y ↪ W i:Y\hookrightarrow W i : Y ↪ W and
F ′ = ker ( F → i ∗ V → i ∗ φ i ∗ Q ) . \mathcal{F}'=\ker\left(\mathcal{F}\to i_*V\xrightarrow{i_*\varphi}i_*Q\right). F ′ = ker ( F → i ∗ V i ∗ φ i ∗ Q ) . Then F ′ \mathcal{F}' F ′ is torsion-free (reflexive if F \mathcal{F} F is), F ′ ∣ U = F ∣ U \mathcal{F}'|_U=\mathcal{F}|_U F ′ ∣ U = F ∣ U , and its restriction to Y Y Y fits into
0 → Q ⊗ O Y ( H ) → F ′ ∣ Y → K → 0. 0\to Q\otimes\mathcal{O}_Y(H)\to\mathcal{F}'|_Y\to\mathcal{K}\to0. 0 → Q ⊗ O Y ( H ) → F ′ ∣ Y → K → 0.
Lemma 6.4 (Slope normalization). Let F \mathcal{F} F be as in Lemma 6.3 .
(a) After finitely many modifications along the last Harder–Narasimhan quotient, all normalized Harder–Narasimhan slopes μ \mu μ of F ∣ Y \mathcal{F}|_Y F ∣ Y lie in an interval of length at most 1 1 1 [26 Lemma 3.1] .
(b) If F \mathcal{F} F is reflexive, then after a twist by a multiple of Y Y Y and finitely many modifications, all normalized Harder–Narasimhan slopes of F ∣ Y \mathcal{F}|_Y F ∣ Y lie in the fixed window [ 3 2 , 5 2 ) [\frac{3}{2},\frac{5}{2}) [ 2 3 , 2 5 ) , and F \mathcal{F} F remains reflexive [9 Prop. 3.3] .
Chen's local cohomology obstruction
Chen presents the following as a proposed argument, circulated for independent verification, and states its consequence for Hochster’s conjecture conditionally.
Main claim ([9 Main claim 1.1] ). Let Y Y Y be a smooth surface and H H H an ample globally generated divisor with K Y ≡ 4 H K_Y\equiv4H K Y ≡ 4 H and
c ( Y , H ) : = 15 8 H 2 − χ ( O Y ) > 0. c(Y,H):=\frac{15}{8}H^2-\chi(\mathcal{O}_Y)>0. c ( Y , H ) := 8 15 H 2 − χ ( O Y ) > 0. Then, for each nonzero finitely generated reflexive R ^ \widehat{R} R -module M M M , the local cohomology module H m 2 ( M ) H^2_{\mathfrak{m}}(M) H m 2 ( M ) is of finite length, with
dim C H m 2 ( M ) ≥ c ( Y , H ) rk M . \dim_{\mathbb{C}}H^2_{\mathfrak{m}}(M)\ge c(Y,H)\operatorname{rk}M. dim C H m 2 ( M ) ≥ c ( Y , H ) rk M . In particular R ^ \widehat{R} R has no small Cohen–Macaulay module.
Under K ≡ 4 H K\equiv4H K ≡ 4 H one has 8 c ( Y , H ) = σ − d 8c(Y,H)=\sigma-d 8 c ( Y , H ) = σ − d , so c > 0 c>0 c > 0 is exactly the graded criterion d < σ d<\sigma d < σ of Theorem 5.1, now claimed for arbitrary modules. As [9 Rem. 1.3] notes, only the numerical equivalence is used; in Example 7.4 it holds linearly. The argument first uses Lemma 6.4 (b) to put all normalized Harder–Narasimhan slopes of V = F ∣ Y V=\mathcal{F}|_Y V = F ∣ Y in the window [ 3 2 , 5 2 ) [\frac{3}{2},\frac{5}{2}) [ 2 3 , 2 5 ) , centred at half the normalized slope K Y ⋅ H / H 2 = 4 K_Y\cdot H/H^2=4 K Y ⋅ H / H 2 = 4 of ω Y \omega_Y ω Y , and then combines the following three statements. Put T − ( V ) = ∑ j < 0 h 0 ( V ( j ) ) T_-(V)=\sum_{j<0}h^0(V(j)) T − ( V ) = ∑ j < 0 h 0 ( V ( j )) and T + ( V ) = ∑ j > 0 h 2 ( V ( j ) ) T_+(V)=\sum_{j>0}h^2(V(j)) T + ( V ) = ∑ j > 0 h 2 ( V ( j )) .
Proposition 7.1 ([9 Prop. 4.1] ). If K Y ≡ 4 H K_Y\equiv4H K Y ≡ 4 H and V V V is torsion-free of rank r r r with all Harder–Narasimhan slopes in [ 3 2 , 5 2 ) [\frac{3}{2},\frac{5}{2}) [ 2 3 , 2 5 ) , then χ ( V ) ≤ − c ( Y , H ) r \chi(V)\le-c(Y,H)r χ ( V ) ≤ − c ( Y , H ) r .
Lemma 7.2 ([9 Lemma 4.2, Prop. 4.3] ). If K Y ≡ 4 H K_Y\equiv4H K Y ≡ 4 H and all Harder–Narasimhan slopes of V V V lie in [ 3 2 , 5 2 ) [\frac{3}{2},\frac{5}{2}) [ 2 3 , 2 5 ) , then T − ( V ) ≤ h 0 ( V ) T_-(V)\le h^0(V) T − ( V ) ≤ h 0 ( V ) and T + ( V ) ≤ h 2 ( V ) T_+(V)\le h^2(V) T + ( V ) ≤ h 2 ( V ) .
Proposition 7.3 ([9 Prop. 5.1, Cor. 5.4] ). For a reflexive F \mathcal{F} F on W W W with V = F ∣ Y V=\mathcal{F}|_Y V = F ∣ Y and T ± ( V ) T_\pm(V) T ± ( V ) finite,
dim C H 1 ( U , F ∣ U ) ≥ h 1 ( V ) − T + ( V ) − T − ( V ) . \dim_{\mathbb{C}}H^1(U,\mathcal{F}|_U)\ge h^1(V)-T_+(V)-T_-(V). dim C H 1 ( U , F ∣ U ) ≥ h 1 ( V ) − T + ( V ) − T − ( V ) .
With Lemma 6.2 (b) they give dim C H m 2 ( M ) ≥ h 1 ( V ) − h 2 ( V ) − h 0 ( V ) = − χ ( V ) ≥ c ( Y , H ) rk M \dim_{\mathbb{C}}H^2_{\mathfrak{m}}(M)\ge h^1(V)-h^2(V)-h^0(V)=-\chi(V)\ge c(Y,H)\operatorname{rk}M dim C H m 2 ( M ) ≥ h 1 ( V ) − h 2 ( V ) − h 0 ( V ) = − χ ( V ) ≥ c ( Y , H ) rk M .
Example 7.4 ([9 Consequence 1.2, §7] ). Let X 6 ⊂ P 5 X_6 \subset\mathbb{P}^5 X 6 ⊂ P 5 be the complete intersection
u 3 6 − u 0 6 + u 1 6 = u 4 6 − u 0 6 + u 2 6 = u 5 6 − u 1 6 + u 2 6 = 0 , u_3^6-u_0^6+u_1^6=u_4^6-u_0^6+u_2^6=u_5^6-u_1^6+u_2^6=0, u 3 6 − u 0 6 + u 1 6 = u 4 6 − u 0 6 + u 2 6 = u 5 6 − u 1 6 + u 2 6 = 0 , the exponent-6 Kummer cover of the complete quadrangle. Its blow-up Y Y Y at the 144 singular points carries H = 3 A − E H=3A-E H = 3 A − E , ample and globally generated, with K Y ∼ 4 H K_Y\sim4H K Y ∼ 4 H , H 2 = 1080 H^2=1080 H 2 = 1080 , χ ( O Y ) = 1926 \chi(\mathcal{O}_Y)=1926 χ ( O Y ) = 1926 , so c ( Y , H ) = 99 c(Y,H)=99 c ( Y , H ) = 99 . This is the ring of Corollary 5.8 with n = 6 n=6 n = 6 , which [9 ] takes from [2 Rem. 4.2, Cor. 4.3] . Subject to the Main claim, R ^ \widehat{R} R — of depth 2, quasi-Gorenstein, with an isolated singularity — has no small CM module.
OpenAI's complete local domain
The manuscript [26 ] , dated 23 September 2026 and released on 6 October as family 195 of OpenAI’s catalogue [27 ] , claims a complete proof, under the numerical hypothesis σ ( Y ) > 2 H 2 \sigma(Y)>2H^2 σ ( Y ) > 2 H 2 , stronger than the graded one σ ( Y ) > H 2 \sigma(Y)>H^2 σ ( Y ) > H 2 . According to the release notes [27 ] , it is the output of an OpenAI model and has no human author.
Theorem 8.1 ([26 Thm. 1.1] ). There is a three-dimensional complete Noetherian normal local domain R ⊃ C R\supset\mathbb{C} R ⊃ C with residue field C \mathbb{C} C that has no nonzero finitely generated module of depth 3.
Theorem 8.2 ([26 Thm. 1.2] ). Let Y Y Y be a smooth projective surface and H H H an ample globally generated divisor. If R ^ ( Y , H ) \widehat{R}(Y,H) R ( Y , H ) has a small Cohen–Macaulay module, then
K Y 2 − 2 c 2 ( Y ) ≤ 6 H 2 , equivalently σ ( Y ) ≤ 2 H 2 . K_Y^2-2c_2(Y)\le6H^2,\qquad\text{equivalently}\qquad\sigma(Y)\le2H^2. K Y 2 − 2 c 2 ( Y ) ≤ 6 H 2 , equivalently σ ( Y ) ≤ 2 H 2 .
Corollary 8.3 ([26 Cor. 7.1] ). For the example of Proposition 8.6 with n = 7 n=7 n = 7 , the uncompleted local ring R ( Y , H ) m R(Y,H)_{\mathfrak m} R ( Y , H ) m — a three-dimensional local domain essentially of finite type over C \mathbb{C} C — also has no small CM module.
The proof uses the finite morphisms f : Y → P 2 f:Y\to\mathbb{P}^2 f : Y → P 2 and g : W → Z g:W\to Z g : W → Z of § 6 . Let M M M be a small CM module, choose an extension of M ‾ ∣ U \overline{M}|_U M ∣ U normalized by Lemma 6.4 (a), and let P P P be its restriction to Y Y Y . Since M M M is free over A A A by the Auslander–Buchsbaum formula, g ∗ g_* g ∗ of the extension is trivial off the exceptional plane D D D , and its restriction to D D D is f ∗ P f_*P f ∗ P . For a coherent sheaf F F F of rank N > 0 N>0 N > 0 on P 2 \mathbb{P}^2 P 2 set
v ( F ) = 2 ch 2 ( F ) N − ( c 1 ( F ) N ) 2 ; v(F)=\frac{2\operatorname{ch}_2(F)}{N}-\left(\frac{c_1(F)}{N}\right)^2; v ( F ) = N 2 ch 2 ( F ) − ( N c 1 ( F ) ) 2 ;
for F = ⨁ O ( a j ) F=\bigoplus\mathcal{O}(a_j) F = ⨁ O ( a j ) this is the variance of ( a j ) (a_j) ( a j ) , and v v v is invariant under twists.
Lemma 8.4 ([26 Lemma 4.1] ). Let Z Z Z be an integral Noetherian scheme containing an effective Cartier divisor D ≅ P 2 D\cong\mathbb{P}^2 D ≅ P 2 with I D ∣ D ≅ O D ( 1 ) \mathcal{I}_{D}|_D\cong\mathcal{O}_D(1) I D ∣ D ≅ O D ( 1 ) . If L \mathcal{L} L is a torsion-free coherent sheaf on Z Z Z with L ∣ Z ∖ D ≅ O N \mathcal{L}|_{Z\setminus D}\cong\mathcal{O}^{N} L ∣ Z ∖ D ≅ O N , then v ( L ∣ D ) ≥ 0 v(\mathcal{L}|_D)\ge0 v ( L ∣ D ) ≥ 0 .
Lemma 8.5 ([26 Lemma 5.1] ). Let f : Y → P 2 f:Y\to\mathbb{P}^2 f : Y → P 2 be finite with f ∗ O ( 1 ) = O ( H ) f^*\mathcal{O}(1)=\mathcal{O}(H) f ∗ O ( 1 ) = O ( H ) , d = H 2 d=H^2 d = H 2 . For every nonzero torsion-free μ H \mu_H μ H -semistable sheaf G G G on Y Y Y ,
v ( f ∗ G ) ≤ 1 4 K Y 2 − 2 c 2 ( Y ) 12 d = 1 4 − σ ( Y ) 4 d , c 1 ( f ∗ G ) rk f ∗ G = μ ( G ) − K Y ⋅ H 2 d − 3 2 . v(f_*G)\le\frac{1}{4}\frac{K_Y^2-2c_2(Y)}{12d}=\frac{1}{4}-\frac{\sigma(Y)}{4d},\qquad
\frac{c_1(f_*G)}{\operatorname{rk}f_*G}=\mu(G)-\frac{K_Y\cdot H}{2d}-\frac{3}{2}. v ( f ∗ G ) ≤ 4 1 12 d K Y 2 − 2 c 2 ( Y ) = 4 1 − 4 d σ ( Y ) , rk f ∗ G c 1 ( f ∗ G ) = μ ( G ) − 2 d K Y ⋅ H − 2 3 .
Theorem 8.2 follows by decomposing v ( f ∗ P ) v(f_*P) v ( f ∗ P ) along the Harder–Narasimhan filtration of P P P ; we carry this out in § 9 (Proposition 9.1 ), and in every dimension in § 10 .
Proposition 8.6 ([26 Prop. 6.1] ). For a prime n > 3 n>3 n > 3 , let π : Y → B \pi:Y\to B π : Y → B be the smooth ( Z / n ) 5 (\mathbb{Z}/n)^5 ( Z / n ) 5 -cover of the blow-up B B B of P 2 \mathbb{P}^2 P 2 at four general points, obtained by adjoining n n n -th roots of the ratios of the six lines of the complete quadrangle. There is an ample, globally generated divisor H H H on Y Y Y with n H ∼ π ∗ ( − K B ) nH\sim\pi^*(-K_B) n H ∼ π ∗ ( − K B ) , and
H 2 = 5 n 3 , K Y 2 − 2 c 2 ( Y ) = n 3 ( n 2 − 10 ) . H^2=5n^3,\qquad K_Y^2-2c_2(Y)=n^3(n^2-10). H 2 = 5 n 3 , K Y 2 − 2 c 2 ( Y ) = n 3 ( n 2 − 10 ) .
For prime n n n , this is the complete-quadrangle surface of Table 2 , as [26 §1] states. Its polarization is H = 3 J − E H = 3J - E H = 3 J − E [26 §6.4, (16)] , where the six sections of O Y ( J ) \mathcal{O}_Y(J) O Y ( J ) constructed there are the Kummer coordinates; hence J = A n J = A_n J = A n , E = E n E = E_n E = E n and H = 3 A n − E n H = 3A_n - E_n H = 3 A n − E n [1 §7.2] . For n = 7 n = 7 n = 7 : H 2 = 1715 H^2 = 1715 H 2 = 1715 and K 2 − 2 c 2 = 13377 > 10290 = 6 H 2 K^2 - 2c_2 = 13377 > 10290 = 6H^2 K 2 − 2 c 2 = 13377 > 10290 = 6 H 2 , so Theorem 8.2 gives Theorem 8.1 . Here K ≡ 5 H K \equiv5H K ≡ 5 H , so the Main claim of [9 ] does not apply.
Synthesis: one identity, three mechanisms
The three obstructions of § 3 , § 5 and § 8 can be read as a single computation on P 2 \mathbb{P}^2 P 2 . We isolate this computation and prove it in full. The identity of Proposition 9.1 is implicit in [26 (10), (11)] , where it is used with inequalities, and its aCM case is the defect identity [2 Prop. 2.6] ; what we add is that all three obstructions are read off from it. Here v v v is the variance of § 8 , and for a torsion-free sheaf G \mathcal{G} G of rank u u u on Y Y Y we use ν \nu ν , μ H = ν ⋅ H \mu_H = \nu\cdot H μ H = ν ⋅ H , δ \delta δ , β = ν − K Y / 2 \beta= \nu- K_Y/2 β = ν − K Y /2 and α = β − β ⋅ H d H \alpha= \beta- \frac{\beta\cdot H}{d}H α = β − d β ⋅ H H as in § 5 .
Proposition 9.1 (cf. [26 (10), (11)] , [2 Prop. 2.6] ). Let f : Y → P 2 f : Y \to\mathbb{P}^2 f : Y → P 2 be finite with f ∗ O ( 1 ) = O Y ( H ) f^*\mathcal{O}(1) = \mathcal{O}_Y(H) f ∗ O ( 1 ) = O Y ( H ) , d = H 2 d = H^2 d = H 2 . Let P P P be a torsion-free sheaf on Y Y Y with a filtration whose successive quotients G i G_i G i are torsion-free and μ H \mu_H μ H -semistable, with weights w i = rk G i / rk P w_i = \operatorname{rk}G_i/\operatorname{rk}P w i = rk G i / rk P and push-forward slopes b i = c 1 ( f ∗ G i ) / rk f ∗ G i b_i = c_1(f_*G_i)/\operatorname{rk}f_*G_i b i = c 1 ( f ∗ G i ) / rk f ∗ G i , and let v b = ∑ i w i ( b i − b ˉ ) 2 v_b = \sum_i w_i(b_i-\bar b)^2 v b = ∑ i w i ( b i − b ˉ ) 2 with b ˉ = ∑ i w i b i \bar b = \sum_i w_i b_i b ˉ = ∑ i w i b i . Then
v ( f ∗ P ) = ∑ i w i v ( f ∗ G i ) + v b , v ( f ∗ G i ) = 1 4 − σ 4 d − δ ( G i ) − α ( G i ) 2 d ≤ 1 4 − σ 4 d . v(f_*P) = \sum_i w_i v(f_*G_i) + v_b,\qquad v(f_*G_i) = \frac{1}{4} - \frac{\sigma}{4d} - \frac{\delta(G_i)-\alpha(G_i)^2}{d} \le\frac{1}{4} - \frac{\sigma}{4d}. v ( f ∗ P ) = i ∑ w i v ( f ∗ G i ) + v b , v ( f ∗ G i ) = 4 1 − 4 d σ − d δ ( G i ) − α ( G i ) 2 ≤ 4 1 − 4 d σ . with δ \delta δ , α \alpha α as in Lemma 5.5.
Proof. The decomposition. The functor f ∗ f_* f ∗ is exact, so f ∗ P f_*P f ∗ P has a filtration with successive quotients f ∗ G i f_*G_i f ∗ G i , of ranks N i = w i N N_i = w_iN N i = w i N with N = rk f ∗ P N = \operatorname{rk}f_*P N = rk f ∗ P . Rank, c 1 c_1 c 1 and ch 2 \operatorname{ch}_2 ch 2 are additive, so with e i = ch 2 ( f ∗ G i ) / N i e_i = \operatorname{ch}_2(f_*G_i)/N_i e i = ch 2 ( f ∗ G i ) / N i one has b ˉ = c 1 ( f ∗ P ) / N \bar b = c_1(f_*P)/N b ˉ = c 1 ( f ∗ P ) / N and
v ( f ∗ P ) = ∑ i w i 2 e i − b ˉ 2 = ∑ i w i ( 2 e i − b i 2 ) + ( ∑ i w i b i 2 − b ˉ 2 ) = ∑ i w i v ( f ∗ G i ) + v b . v(f_*P) = \sum_i w_i 2e_i - \bar b^2 = \sum_i w_i(2e_i-b_i^2) + \left(\sum_i w_i b_i^2-\bar b^2\right) = \sum_i w_i v(f_*G_i)+v_b. v ( f ∗ P ) = i ∑ w i 2 e i − b ˉ 2 = i ∑ w i ( 2 e i − b i 2 ) + ( i ∑ w i b i 2 − b ˉ 2 ) = i ∑ w i v ( f ∗ G i ) + v b .
The variance of one push-forward. Let G \mathcal{G} G be torsion-free of rank u u u on Y Y Y and F = f ∗ G F = f_*\mathcal{G} F = f ∗ G , of rank u d ud u d ; put b = c 1 ( F ) / ( u d ) b = c_1(F)/(ud) b = c 1 ( F ) / ( u d ) and e = ch 2 ( F ) / ( u d ) e = \operatorname{ch}_2(F)/(ud) e = ch 2 ( F ) / ( u d ) . Since f f f is finite, χ ( F ( t ) ) = χ ( G ( t H ) ) \chi(F(t)) = \chi(\mathcal{G}(tH)) χ ( F ( t )) = χ ( G ( t H )) for all t t t . By Riemann–Roch on P 2 \mathbb{P}^2 P 2 ,
χ ( F ( t ) ) u d = t 2 2 + ( b + 3 2 ) t + 1 + 3 2 b + e . \frac{\chi(F(t))}{ud} = \frac{t^2}{2} + \left(b+\frac{3}{2}\right)t + 1+\frac{3}{2}b+e. u d χ ( F ( t )) = 2 t 2 + ( b + 2 3 ) t + 1 + 2 3 b + e .
By Riemann–Roch on Y Y Y ,
χ ( G ( t H ) ) u d = t 2 2 + μ H − κ / 2 d t + 1 d ( χ ( O Y ) + ν ⋅ ( ν − K Y ) 2 − δ 2 ) , κ = K Y ⋅ H . \frac{\chi(\mathcal{G}(tH))}{ud}
= \frac{t^2}{2}+\frac{\mu_H-\kappa/2}{d}t+\frac{1}{d}\left(\chi(\mathcal{O}_Y)+\frac{\nu\cdot(\nu-K_Y)}{2}-\frac{\delta}{2}\right),\qquad\kappa=K_Y\cdot H. u d χ ( G ( t H )) = 2 t 2 + d μ H − κ /2 t + d 1 ( χ ( O Y ) + 2 ν ⋅ ( ν − K Y ) − 2 δ ) , κ = K Y ⋅ H .
The linear terms give b + 3 2 = ( μ H − κ / 2 ) / d = β ⋅ H / d b+\frac{3}{2} = (\mu_H-\kappa/2)/d = \beta\cdot H/d b + 2 3 = ( μ H − κ /2 ) / d = β ⋅ H / d . The constant terms give
2 e = 2 d ( χ ( O Y ) + 1 2 ν ⋅ ( ν − K Y ) − δ 2 ) − 2 − 3 b , 2e = \frac{2}{d}\left(\chi(\mathcal{O}_Y)+\frac{1}{2}\nu\cdot(\nu-K_Y)-\frac{\delta}{2}\right)-2-3b, 2 e = d 2 ( χ ( O Y ) + 2 1 ν ⋅ ( ν − K Y ) − 2 δ ) − 2 − 3 b ,
hence
v ( F ) = 2 e − b 2 = 2 χ ( O Y ) + ν ⋅ ( ν − K Y ) − δ d − ( b + 3 2 ) 2 + 1 4 . v(F)=2e-b^2=\frac{2\chi(\mathcal{O}_Y)+\nu\cdot(\nu-K_Y)-\delta}{d}-\left(b+\frac{3}{2}\right)^2+\frac{1}{4}. v ( F ) = 2 e − b 2 = d 2 χ ( O Y ) + ν ⋅ ( ν − K Y ) − δ − ( b + 2 3 ) 2 + 4 1 .
Now ν ⋅ ( ν − K Y ) = β 2 − 1 4 K Y 2 \nu\cdot(\nu-K_Y)=\beta^2-\frac{1}{4}K_Y^2 ν ⋅ ( ν − K Y ) = β 2 − 4 1 K Y 2 and β 2 = α 2 + ( β ⋅ H ) 2 / d \beta^2=\alpha^2+(\beta\cdot H)^2/d β 2 = α 2 + ( β ⋅ H ) 2 / d , so the terms ( β ⋅ H ) 2 / d 2 (\beta\cdot H)^2/d^2 ( β ⋅ H ) 2 / d 2 cancel and
v ( F ) = 1 4 + 8 χ ( O Y ) − K Y 2 4 d + α 2 − δ d = 1 4 − σ 4 d − δ ( G ) − α ( G ) 2 d . v(F)=\frac{1}{4}+\frac{8\chi(\mathcal{O}_Y)-K_Y^2}{4d}+\frac{\alpha^2-\delta}{d}
=\frac{1}{4}-\frac{\sigma}{4d}-\frac{\delta(\mathcal{G})-\alpha(\mathcal{G})^2}{d}. v ( F ) = 4 1 + 4 d 8 χ ( O Y ) − K Y 2 + d α 2 − δ = 4 1 − 4 d σ − d δ ( G ) − α ( G ) 2 .
For G = G i \mathcal{G}=G_i G = G i semistable, δ ( G i ) ≥ 0 \delta(G_i)\ge0 δ ( G i ) ≥ 0 by Bogomolov’s inequality and α ( G i ) 2 ≤ 0 \alpha(G_i)^2\le0 α ( G i ) 2 ≤ 0 by the Hodge index theorem, since α ⋅ H = 0 \alpha\cdot H=0 α ⋅ H = 0 . □
The linear terms also show that b i − μ H ( G i ) / d b_i-\mu_H(G_i)/d b i − μ H ( G i ) / d does not depend on i i i : the push-forward slopes are the slopes of the G i G_i G i , shifted and rescaled by the same amounts (Lemmas 5.3 and 8.5).
Hence v ( f ∗ P ) ≤ 1 4 − σ 4 d + v b v(f_*P)\le\frac{1}{4}-\frac{\sigma}{4d}+v_b v ( f ∗ P ) ≤ 4 1 − 4 d σ + v b always. Each result combines this upper bound with a lower bound for v ( f ∗ P ) v(f_*P) v ( f ∗ P ) and an upper bound for v b v_b v b ; Table 3 lists them.
Table 3. One identity, three mechanisms.
Ulrich. For a linear projection, f ∗ E ≅ O N f_*\mathcal{E}\cong\mathcal{O}^{N} f ∗ E ≅ O N (Proposition 2.2 ), so v ( f ∗ E ) = v a = 0 v(f_*\mathcal{E})=v_a=0 v ( f ∗ E ) = v a = 0 . Lemma 5.4, applied to the push-forward of the Harder–Narasimhan filtration, gives v b ≤ v a = 0 v_b\leq v_a=0 v b ≤ v a = 0 , so all b i b_i b i coincide; since they are strictly decreasing, E \mathcal{E} E is μ H \mu_H μ H -semistable, which recovers [8 Thm. 2.9] . Hence 0 ≤ 1 4 − σ 4 d 0\leq\frac{1}{4}-\frac{\sigma}{4d} 0 ≤ 4 1 − 4 d σ . This is Proposition 3.1 ; (i) is the case v a = 0 v_a=0 v a = 0 of (ii).
aCM. f ∗ E f_*\mathcal{E} f ∗ E splits, v ( f ∗ E ) = v a v(f_*\mathcal{E})=v_a v ( f ∗ E ) = v a , and Lemma 5.4 applied to the push-forward of the Harder–Narasimhan filtration gives v b ≤ v a v_b\leq v_a v b ≤ v a . Hence 0 ≤ v a − v b ≤ 1 4 − σ 4 d 0\leq v_a-v_b\leq\frac{1}{4}-\frac{\sigma}{4d} 0 ≤ v a − v b ≤ 4 1 − 4 d σ . Keeping the equality of Proposition 9.1 , this is exactly the defect identity of Proposition 5.6 .
Local. Let M M M be a small CM module over R ^ ( Y , H ) \widehat{R}(Y,H) R ( Y , H ) and P P P the restriction to the exceptional surface of an extension normalized by Lemma 6.4 (a). We do not assume that f ∗ P f_*P f ∗ P splits; but f ∗ P f_*P f ∗ P is the restriction to the exceptional plane of a sheaf that is trivial off it, so v ( f ∗ P ) ≥ 0 v(f_*P)\geq0 v ( f ∗ P ) ≥ 0 by Lemma 8.4. The normalized slopes μ H ( G i ) / d \mu_H(G_i)/d μ H ( G i ) / d of the Harder–Narasimhan factors lie in an interval of length ≤ 1 \leq1 ≤ 1 , hence so do the b i b_i b i , and a weighted variance of points in such an interval is at most 1 4 \frac{1}{4} 4 1 : with γ \gamma γ the midpoint, v b ≤ ∑ i w i ( b i − γ ) 2 ≤ 1 4 v_b\leq\sum_i w_i(b_i-\gamma)^2\leq\frac{1}{4} v b ≤ ∑ i w i ( b i − γ ) 2 ≤ 4 1 . Hence 0 ≤ 1 2 − σ 4 d 0\leq\frac{1}{2}-\frac{\sigma}{4d} 0 ≤ 2 1 − 4 d σ , i.e. σ ≤ 2 d \sigma\leq2d σ ≤ 2 d . Given Lemmas 6.4 (a) and 8.4 and the identification of f ∗ P f_*P f ∗ P in § 8 , this proves Theorem 8.2 ; § 10 proves these ingredients as well, in arbitrary dimension.
For surfaces, the factor 2 between σ ≤ d \sigma\leq d σ ≤ d and σ ≤ 2 d \sigma\leq2d σ ≤ 2 d comes from replacing majorization, v b ≤ v a v_b\leq v_a v b ≤ v a , available when the push-forward splits, by the bound v b ≤ 1 4 v_b\leq\frac{1}{4} v b ≤ 4 1 for the variance of slopes in a window of width one. In dimension m m m the two bounds become ( m + 1 ) H m (m+1)H^m ( m + 1 ) H m and ( m + 4 ) H m (m+4)H^m ( m + 4 ) H m (§ 10.7 ). Chen’s argument (§ 7 ) fits the same scheme: measuring χ \chi χ instead of v v v gives a stronger lower bound for v ( f ∗ P ) v(f_*P) v ( f ∗ P ) , at the price of a window centred at half the normalized slope of ω Y \omega_Y ω Y , so that Serre duality, s ↦ λ − s s\mapsto\lambda-s s ↦ λ − s with λ = K Y ⋅ H / d \lambda=K_Y\cdot H/d λ = K Y ⋅ H / d , maps it into the closed window [ λ 2 − 1 2 , λ 2 + 1 2 ] \left[\frac{\lambda}{2}-\frac{1}{2},\frac{\lambda}{2}+\frac{1}{2}\right] [ 2 λ − 2 1 , 2 λ + 2 1 ] (s ↦ 4 − s s\mapsto4-s s ↦ 4 − s and [ 3 2 , 5 2 ] \left[\frac{3}{2},\frac{5}{2}\right] [ 2 3 , 2 5 ] when K ≡ 4 H K\equiv4H K ≡ 4 H ). In the subcanonical case K ≡ 4 H K\equiv4H K ≡ 4 H it would recover the graded bound σ ≤ d \sigma\leq d σ ≤ d locally.
Remark 9.2 (Chen’s mechanism as a lower bound for v v v ). For a coherent sheaf F F F of rank N > 0 N>0 N > 0 on P 2 \mathbb{P}^2 P 2 with b = c 1 ( F ) / N b=c_1(F)/N b = c 1 ( F ) / N , Riemann–Roch gives χ ( F ) / N = 1 + 3 2 b + 1 2 ( v ( F ) + b 2 ) \chi(F)/N=1+\frac{3}{2}b+\frac{1}{2}(v(F)+b^2) χ ( F ) / N = 1 + 2 3 b + 2 1 ( v ( F ) + b 2 ) , that is,
χ ( F ) N = 1 2 ( b + 3 2 ) 2 + v ( F ) 2 − 1 8 . \frac{\chi(F)}{N}=\frac{1}{2}\left(b+\frac{3}{2}\right)^2+\frac{v(F)}{2}-\frac{1}{8}. N χ ( F ) = 2 1 ( b + 2 3 ) 2 + 2 v ( F ) − 8 1 .
Let P P P be a torsion-free sheaf on Y Y Y with Harder–Narasimhan factors G i G_i G i , weights w i w_i w i , push-forward slopes b i b_i b i with mean b ˉ \bar b b ˉ , and put λ = K Y ⋅ H / d \lambda=K_Y\cdot H/d λ = K Y ⋅ H / d . Since ∑ i w i ( b i + 3 2 ) 2 = v b + ( b ˉ + 3 2 ) 2 \sum_i w_i\left(b_i+\frac{3}{2}\right)^2=v_b+\left(\bar b+\frac{3}{2}\right)^2 ∑ i w i ( b i + 2 3 ) 2 = v b + ( b ˉ + 2 3 ) 2 and b i + 3 2 = μ ( G i ) − λ 2 b_i+\frac{3}{2}=\mu(G_i)-\frac{\lambda}{2} b i + 2 3 = μ ( G i ) − 2 λ (Lemma 8.5), Proposition 9.1 and χ ( P ) = χ ( f ∗ P ) \chi(P)=\chi(f_*P) χ ( P ) = χ ( f ∗ P ) give
χ ( P ) rk P = d 2 ∑ i w i ( μ ( G i ) − λ 2 ) 2 − σ 8 − 1 2 ∑ i w i ( δ ( G i ) − α ( G i ) 2 ) . \frac{\chi(P)}{\operatorname{rk}P}=\frac{d}{2}\sum_i w_i\left(\mu(G_i)-\frac{\lambda}{2}\right)^2-\frac{\sigma}{8}-\frac{1}{2}\sum_i w_i\left(\delta(G_i)-\alpha(G_i)^2\right). rk P χ ( P ) = 2 d i ∑ w i ( μ ( G i ) − 2 λ ) 2 − 8 σ − 2 1 i ∑ w i ( δ ( G i ) − α ( G i ) 2 ) .
which is also Lemma 5.5 at t = 0 t=0 t = 0 , summed over the factors. If all μ ( G i ) \mu(G_i) μ ( G i ) lie in [ λ 2 − 1 2 , λ 2 + 1 2 ] \left[\frac{\lambda}{2}-\frac{1}{2},\frac{\lambda}{2}+\frac{1}{2}\right] [ 2 λ − 2 1 , 2 λ + 2 1 ] , then χ ( P ) ≤ d − σ 8 rk P \chi(P)\leq\frac{d-\sigma}{8}\operatorname{rk}P χ ( P ) ≤ 8 d − σ rk P ; for K Y ≡ 4 H K_Y\equiv4H K Y ≡ 4 H this is Proposition 7.1 , even for the closed window. In terms of v v v , the condition χ ( f ∗ P ) ≥ 0 \chi(f_*P)\geq0 χ ( f ∗ P ) ≥ 0 reads v ( f ∗ P ) ≥ 1 4 − ( b ˉ + 3 2 ) 2 v(f_*P)\geq\frac{1}{4}-(\bar b+\frac{3}{2})^2 v ( f ∗ P ) ≥ 4 1 − ( b ˉ + 2 3 ) 2 . Under K Y ≡ 4 H K_Y\equiv4H K Y ≡ 4 H , for a small CM module, the local input of [9 ] (Lemma 6.2(b), Lemma 7.2 and Proposition 7.3), if correct, gives χ ( P ) ≥ 0 \chi(P) \ge0 χ ( P ) ≥ 0 for the normalized extension. It thus replaces the lower bound v ( f ∗ P ) ≥ 0 v(f_*P) \ge0 v ( f ∗ P ) ≥ 0 of Lemma 8.4 by v ( f ∗ P ) ≥ 1 4 − ( b ˉ + 3 2 ) 2 v(f_*P) \ge\frac{1}{4}-(\bar b+\frac{3}{2})^2 v ( f ∗ P ) ≥ 4 1 − ( b ˉ + 2 3 ) 2 , and the bound v b ≤ 1 4 v_b \le\frac{1}{4} v b ≤ 4 1 by ∑ i w i ( b i + 3 2 ) 2 ≤ 1 4 \sum_i w_i(b_i+\frac{3}{2})^2 \le\frac{1}{4} ∑ i w i ( b i + 2 3 ) 2 ≤ 4 1 , which needs a window centred at λ 2 \frac{\lambda}{2} 2 λ (Lemma 6.4(b)) rather than an arbitrary window of width one (Lemma 6.4(a)). With Proposition 9.1 this gives
1 4 ≤ d − σ 4 d + ∑ i w i ( b i + 3 2 ) 2 ≤ d − σ 4 d + 1 4 , \frac{1}{4} \le\frac{d-\sigma}{4d}+\sum_i w_i\left(b_i+\frac{3}{2}\right)^2 \le\frac{d-\sigma}{4d}+\frac{1}{4}, 4 1 ≤ 4 d d − σ + i ∑ w i ( b i + 2 3 ) 2 ≤ 4 d d − σ + 4 1 ,
i.e. σ ≤ d \sigma\le d σ ≤ d . With the mean slope at the centre the two lower bounds differ by exactly 1 4 \frac{1}{4} 4 1 ; this is the difference between σ ≤ d \sigma\le d σ ≤ d and σ ≤ 2 d \sigma\le2d σ ≤ 2 d .
Comparison of the local criteria. Both local criteria lie inside the graded range σ > d \sigma> d σ > d , but they are incomparable with each other. The quadrangle ring with n = 6 n=6 n = 6 (d = 1080 < σ = 1872 < 2160 = 2 d d=1080 < \sigma=1872 < 2160=2d d = 1080 < σ = 1872 < 2160 = 2 d ) is covered locally only by the Main claim of [9 ] . Conversely, the criterion of [26 ] also applies to pairs outside the hypothesis K ≡ 4 H K \equiv4H K ≡ 4 H of [9 ] , such as the quadrangle with n = 7 n=7 n = 7 . The Hesse pairs are covered locally by Theorem 8.2 for every n ≥ 3 n \ge3 n ≥ 3 ; we make this explicit in Corollary 9.3 .
Corollary 9.3 (consequence of [26 Thm. 1.2] , proved in Theorem 10.3). For every n ≥ 3 n \ge3 n ≥ 3 , the completed section ring R ^ ( Y n , 4 A n − E n ) \widehat{R}(Y_n,4A_n-E_n) R ( Y n , 4 A n − E n ) of the Hesse pair has no small Cohen–Macaulay module. In particular this holds for the ball quotient Y 3 Y_3 Y 3 of Theorem 4.4. The same holds for the dual Hesse pairs with n ≥ 6 n \ge6 n ≥ 6 and for the complete-quadrangle pairs with n ≥ 7 n \ge7 n ≥ 7 .
Proof. By Proposition 4.3 , H n H_n H n is very ample, hence ample and globally generated; likewise 6 A n − E n 6A_n-E_n 6 A n − E n (dual Hesse) and 3 A n − E n 3A_n-E_n 3 A n − E n (quadrangle) are very ample for all n ≥ 2 n \ge2 n ≥ 2 [1 Lemmas 7.1, 7.3] . By Proposition 4.1 and Theorem 4.4,
K 2 − 2 c 2 = 3 σ = 3 ( 3 n 2 − 11 ) n 9 ≥ 48 n 9 > 42 n 9 = 6 H n 2 K^2-2c_2=3\sigma=3(3n^2-11)n^9\ge48n^9>42n^9=6H_n^2 K 2 − 2 c 2 = 3 σ = 3 ( 3 n 2 − 11 ) n 9 ≥ 48 n 9 > 42 n 9 = 6 H n 2
for n ≥ 3 n \ge3 n ≥ 3 , so Theorem 8.2 applies. For the dual Hesse pairs, 3 ( 2 n 2 − 13 ) > 6 ⋅ 24 3(2n^2-13)>6\cdot24 3 ( 2 n 2 − 13 ) > 6 ⋅ 24 if and only if n ≥ 6 n\ge6 n ≥ 6 ; for the quadrangle, n 2 − 10 > 30 n^2-10>30 n 2 − 10 > 30 if and only if n ≥ 7 n\ge7 n ≥ 7 .
Theorem 8.2 is the case m = 2 m=2 m = 2 of Theorem 10.3, proved in Section 10; so the corollary rests on that proof.
Thus the first counterexample to Question Q1, the Hesse ball quotient, also yields, through the argument of [26 ] , a counterexample to Hochster’s conjecture (Question Q3). The Hesse surfaces are ball quotients only for n = 3 n=3 n = 3 , and σ ( Y n ) / H n 2 = ( 3 n 2 − 11 ) / 7 → ∞ \sigma(Y_n)/H_n^2=(3n^2-11)/7\to\infty σ ( Y n ) / H n 2 = ( 3 n 2 − 11 ) /7 → ∞ , so the margin over 2 d 2d 2 d grows with n n n .
Higher dimension
The local rings of Sections 7–9 are three-dimensional. We consider two constructions in higher dimension. The first is formal and produces non-isolated singularities (Section 10.1). The second extends the argument of Section 8 to section rings of varieties of dimension m ≥ 2 m \ge2 m ≥ 2 , as [24 ] and [2 §5] extend the projective and graded obstructions, and produces isolated singularities in every dimension. This second route is not contained in the five texts. We give complete proofs; for m = 2 m = 2 m = 2 they supply all details of Theorem 8.2 .
Adjoining power series
The following is standard; its graded analogue is noted at the end of [2 §5] .
Proposition 10.1. Let ( B , n ) (B,\mathfrak n) ( B , n ) be a Noetherian local ring and B ′ = B [ [ x ] ] B' = B[[x]] B ′ = B [[ x ]] . If M M M is a small CM B ′ B' B ′ -module, then M / x M M/xM M / x M is a small CM B B B -module. Hence, if B B B has no small CM module, neither has B [ [ x 1 , … , x k ] ] B[[x_1,\ldots,x_k]] B [[ x 1 , … , x k ]] for any k ≥ 0 k \ge0 k ≥ 0 .
Proof. The ring B ′ B' B ′ is local with maximal ideal ( n , x ) (\mathfrak n,x) ( n , x ) , dim B ′ = dim B + 1 \dim B' = \dim B + 1 dim B ′ = dim B + 1 and B ′ / x B ′ = B B'/xB' = B B ′ / x B ′ = B , so x x x is part of a system of parameters of B ′ B' B ′ . Every system of parameters of B ′ B' B ′ is a regular sequence on the maximal Cohen–Macaulay module M M M [7 Thm. 2.1.2] , so x x x is M M M -regular. Hence M / x M M/xM M / x M is a finitely generated B B B -module, nonzero by Nakayama, with depth B ( M / x M ) = depth B ′ M − 1 = dim B \operatorname{depth}_B(M/xM) = \operatorname{depth}_{B'} M - 1 = \dim B depth B ( M / x M ) = depth B ′ M − 1 = dim B . Induct on k k k . □
Corollary 10.2. Let ( R ^ , m ^ ) (\widehat{R},\widehat{\mathfrak m}) ( R , m ) be a three-dimensional complete normal local domain with residue field C \mathbb{C} C and without small CM modules, for instance one of the rings of Corollary 9.3 . For every k ≥ 1 k \ge1 k ≥ 1 , B ′ = R ^ [ [ x 1 , … , x k ] ] B' = \widehat{R}[[x_1,\ldots,x_k]] B ′ = R [[ x 1 , … , x k ]] is a complete Noetherian normal local domain of dimension 3 + k 3 + k 3 + k , with residue field C \mathbb{C} C , without small CM modules. Its singular locus contains V ( m ^ B ′ ) V(\widehat{\mathfrak m}B') V ( m B ′ ) , which has dimension k k k ; in particular the singularity is not isolated.
Proof. B ′ B' B ′ is the completion of the excellent normal local ring R ^ [ x 1 , … , x k ] ( m ^ , x 1 , … , x k ) \widehat{R}[x_1,\ldots,x_k]_{(\widehat{\mathfrak m},x_1,\ldots,x_k)} R [ x 1 , … , x k ] ( m , x 1 , … , x k ) , hence a complete normal local domain [25 §32] . Proposition 10.1 excludes small CM modules. Let p = m ^ B ′ \mathfrak p = \widehat{\mathfrak m}B' p = m B ′ , a prime with B ′ / p = C [ [ x 1 , … , x k ] ] B'/\mathfrak p = \mathbb{C}[[x_1,\ldots,x_k]] B ′ / p = C [[ x 1 , … , x k ]] . The local homomorphism R ^ → B p ′ \widehat{R} \to B'_{\mathfrak p} R → B p ′ is flat, and its closed fibre is the fraction field of B ′ / p B'/\mathfrak p B ′ / p . If B p ′ B'_{\mathfrak p} B p ′ were regular, R ^ \widehat{R} R would be regular [25 Thm. 23.7] , and then R ^ \widehat{R} R itself would be a small CM module. So B p ′ B'_{\mathfrak p} B p ′ is singular, and so are all localizations at primes containing p \mathfrak p p . □
Setting and statement
For the rest of this section X X X is a smooth connected projective variety of dimension m ≥ 2 m \ge2 m ≥ 2 and H H H an ample, globally generated divisor on X X X . Put
d = H m , λ = K X ⋅ H m − 1 d , R = R ( X , H ) = ⨁ t ≥ 0 H 0 ( X , O X ( t H ) ) , m = R + , d = H^m,\qquad\lambda= \frac{K_X \cdot H^{m-1}}{d},\qquad R = R(X,H) = \bigoplus_{t\ge0} H^0(X,\mathcal O_X(tH)),\qquad\mathfrak m = R_+, d = H m , λ = d K X ⋅ H m − 1 , R = R ( X , H ) = t ≥ 0 ⨁ H 0 ( X , O X ( t H )) , m = R + ,
and let R ^ \widehat{R} R be the m \mathfrak m m -adic completion of R R R . For a torsion-free sheaf G \mathcal G G of rank u > 0 u > 0 u > 0 on X X X put
μ ( G ) = c 1 ( G ) ⋅ H m − 1 u d , ν ( G ) = c 1 ( G ) u , δ ( G ) = Δ ( G ) u 2 , \mu(\mathcal G) = \frac{c_1(\mathcal G)\cdot H^{m-1}}{u d},\qquad\nu(\mathcal G) = \frac{c_1(\mathcal G)}{u},\qquad\delta(\mathcal G) = \frac{\Delta(\mathcal G)}{u^2}, μ ( G ) = u d c 1 ( G ) ⋅ H m − 1 , ν ( G ) = u c 1 ( G ) , δ ( G ) = u 2 Δ ( G ) ,
where Δ ( G ) = 2 u c 2 ( G ) − ( u − 1 ) c 1 ( G ) 2 \Delta(\mathcal G) = 2u c_2(\mathcal G) - (u-1)c_1(\mathcal G)^2 Δ ( G ) = 2 u c 2 ( G ) − ( u − 1 ) c 1 ( G ) 2 . Harder–Narasimhan filtrations are taken with respect to the slope μ \mu μ [16 §1.6] . Recall the excess I ( X , H ) = ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 − ( m + 1 ) H m I(X,H) = (K_X^2 - 2c_2(X))\cdot H^{m-2} - (m+1)H^m I ( X , H ) = ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 − ( m + 1 ) H m (Lemma 3.3 ). By [Propositions 3.2 and 5.9], Ulrich bundles (for H H H very ample) and aCM bundles can exist only if I ( X , H ) ≤ 0 I(X,H) \le0 I ( X , H ) ≤ 0 .
Theorem 10.3. If R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) has a small Cohen–Macaulay module, then
( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 ≤ ( m + 4 ) H m , equivalently I ( X , H ) ≤ 3 H m . (K_X^2 - 2c_2(X))\cdot H^{m-2} \le(m+4)H^m,\qquad\text{equivalently}\qquad I(X,H) \le3H^m. ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 ≤ ( m + 4 ) H m , equivalently I ( X , H ) ≤ 3 H m .
For m = 2 m = 2 m = 2 one has K 2 − 2 c 2 = 3 σ K^2-2c_2 = 3\sigma K 2 − 2 c 2 = 3 σ , and this is Theorem 8.2 . The proof follows the argument of [26 ] (Sections 8 and 9) step by step. Several steps of [26 ] are written there for surfaces but do not depend on the dimension: the model (Proposition 10.4 , extending [CH, Prop. 2.1] and [26 §2] ), the variance lemma (Lemma 10.5 , extending [26 Lemma 4.1] ) and the modifications (Step 3, extending [26 Lemma 3.1] ). The dimension enters through the Riemann–Roch computation of Proposition 10.6 , in which the Todd class of P m \mathbb{P}^m P m produces the constant m + 1 12 \frac{m+1}{12} 12 m + 1 in place of 1 4 \frac{1}{4} 4 1 ; for a trivial push-forward it reduces to the identity in the proof of [24 Lemma 3.1] (Remark 10.7(iii)).
The model.
Proposition 10.4 (cf. [9 Prop. 2.1] , [26 §2] for m = 2 m=2 m = 2 ). Let s 0 , … , s m ∈ H 0 ( X , O X ( H ) ) s_0,\ldots,s_m\in H^0(X,\mathcal{O}_X(H)) s 0 , … , s m ∈ H 0 ( X , O X ( H )) be general sections, f : X → P m f:X\to\mathbb{P}^m f : X → P m the morphism they define, S = C [ x 0 , … , x m ] → R S=\mathbb{C}[x_0,\ldots,x_m]\to R S = C [ x 0 , … , x m ] → R the homomorphism x i ↦ s i x_i\mapsto s_i x i ↦ s i , and A = C [ [ x 0 , … , x m ] ] A=\mathbb{C}[[x_0,\ldots,x_m]] A = C [[ x 0 , … , x m ]] .
(a) f f f is finite and surjective of degree d d d , and f ∗ O ( 1 ) = O X ( H ) f^*\mathcal{O}(1)=\mathcal{O}_X(H) f ∗ O ( 1 ) = O X ( H ) .
(b) R R R is a normal domain, finite over S S S of generic rank d d d . Its completion is R ^ = R ⊗ S A \widehat{R}=R\otimes_S A R = R ⊗ S A , a complete normal local domain of dimension m + 1 m+1 m + 1 with residue field C \mathbb{C} C , finite over A A A of generic rank d d d .
(c) Let Z = Bl 0 Spec A Z=\operatorname{Bl}_0\operatorname{Spec} A Z = Bl 0 Spec A , with exceptional divisor D ≅ P m D\cong\mathbb{P}^m D ≅ P m , so that I D ∣ D ≅ O D ( 1 ) \mathcal{I}_D|_D\cong\mathcal{O}_D(1) I D ∣ D ≅ O D ( 1 ) . There are a regular integral scheme W W W , a finite morphism g : W → Z g:W\to Z g : W → Z and a proper morphism p : W → Spec R ^ p:W\to\operatorname{Spec}\widehat{R} p : W → Spec R with the following properties. The scheme-theoretic inverse image g − 1 ( D ) g^{-1}(D) g − 1 ( D ) is isomorphic to X X X , and under this identification X ⊂ W X\subset W X ⊂ W is an effective Cartier divisor, g ∣ X = f g|_X=f g ∣ X = f and I X ∣ X ≅ O X ( H ) \mathcal{I}_X|_X\cong\mathcal{O}_X(H) I X ∣ X ≅ O X ( H ) . Finally, p p p induces an isomorphism W ∖ X ≅ Spec R ^ ∖ { m ^ } W\setminus X\cong\operatorname{Spec}\widehat{R}\setminus\{\widehat{\mathfrak{m}}\} W ∖ X ≅ Spec R ∖ { m } lying over the isomorphism Z ∖ D ≅ Spec A ∖ { m A } Z\setminus D\cong\operatorname{Spec}A\setminus\{\mathfrak{m}_A\} Z ∖ D ≅ Spec A ∖ { m A } . In particular R ^ \widehat{R} R is regular away from the vertex: R ^ p \widehat{R}_{\mathfrak{p}} R p is regular for every prime p ≠ m ^ \mathfrak{p}\ne\widehat{\mathfrak{m}} p = m .
Proof. (a) Since ∣ H ∣ |H| ∣ H ∣ is base-point-free, m + 1 m+1 m + 1 general sections have no common zero. As f ∗ O ( 1 ) = O X ( H ) f^*\mathcal{O}(1)=\mathcal{O}_X(H) f ∗ O ( 1 ) = O X ( H ) is ample, f f f contracts no curve; a quasi-finite projective morphism is finite, and a finite morphism between varieties of the same dimension is surjective. Its degree is ∫ X f ∗ h m = H m \int_X f^*h^m=H^m ∫ X f ∗ h m = H m , where h h h is the hyperplane class.
(b) Let L = O X ( H ) L=\mathcal{O}_X(H) L = O X ( H ) , T = Spec X ⨁ t ≥ 0 L t T=\operatorname{Spec}_X\bigoplus_{t\ge0}L^t T = Spec X ⨁ t ≥ 0 L t the total space of L − 1 L^{-1} L − 1 , X 0 ⊂ T X_0\subset T X 0 ⊂ T its zero section, and T ∘ = T ∖ X 0 = Spec X ⨁ t ∈ Z L t T^\circ=T\setminus X_0=\operatorname{Spec}_X\bigoplus_{t\in\mathbb{Z}}L^t T ∘ = T ∖ X 0 = Spec X ⨁ t ∈ Z L t . For a homogeneous s ∈ R k s\in R_k s ∈ R k , k ≥ 1 k\ge1 k ≥ 1 , the open set X s = X s j X_s=X_{s^j} X s = X s j is affine, being the complement of a hyperplane section for an embedding by a very ample power L k j L^{kj} L k j , and Γ ( X s , L t ) = ( R s ) t \Gamma(X_s,L^t)=(R_s)_t Γ ( X s , L t ) = ( R s ) t [[13 Lemma II.5.14] ]; hence the preimage of D ( s ) ⊂ Spec R D(s)\subset\operatorname{Spec}R D ( s ) ⊂ Spec R in T T T is Spec R s \operatorname{Spec}R_s Spec R s . These sets cover T ∘ T^\circ T ∘ because L L L is ample, and X 0 X_0 X 0 maps to the vertex, so T ∘ ≅ Spec R ∖ V ( m ) T^\circ\cong\operatorname{Spec}R\setminus V(\mathfrak{m}) T ∘ ≅ Spec R ∖ V ( m ) . Since H 0 ( X , O X ( t H ) ) = 0 H^0(X,\mathcal{O}_X(tH))=0 H 0 ( X , O X ( t H )) = 0 for t < 0 t<0 t < 0 , R = Γ ( T ∘ , O ) R=\Gamma(T^\circ,\mathcal{O}) R = Γ ( T ∘ , O ) is the ring of functions of a smooth connected scheme, hence a normal domain. By the projection formula R = ⨁ t ≥ 0 H 0 ( P m , ( f ∗ O X ) ( t ) ) R=\bigoplus_{t\ge0}H^0(\mathbb{P}^m,(f_*\mathcal{O}_X)(t)) R = ⨁ t ≥ 0 H 0 ( P m , ( f ∗ O X ) ( t )) , a finitely generated graded S S S -module [[13 Ex. II.5.9] ] of rank rk f ∗ O X = d \operatorname{rk}f_*\mathcal{O}_X=d rk f ∗ O X = d . Since R / ( x 0 , … , x m ) R R/(x_0,\ldots,x_m)R R / ( x 0 , … , x m ) R is finite-dimensional, the ideal ( x 0 , … , x m ) R (x_0,\ldots,x_m)R ( x 0 , … , x m ) R is m \mathfrak{m} m -primary, and R ^ = R ⊗ S A \widehat{R}=R\otimes_S A R = R ⊗ S A . As R m R_{\mathfrak{m}} R m is excellent and normal, R ^ \widehat{R} R is normal [[25 §32] ], hence a domain, of dimension $m+1.
(c) Since ⨁ t L t = ⨁ t O ( t ) \bigoplus_tL^t=\bigoplus_t\mathcal{O}(t) ⨁ t L t = ⨁ t O ( t ) , we have T = X × P m T 0 T=X\times_{\mathbb{P}^m}T_0 T = X × P m T 0 with T 0 = Tot ( O P m ( − 1 ) ) = Bl 0 A m + 1 T_0=\operatorname{Tot}(\mathcal{O}_{\mathbb{P}^m}(-1))=\operatorname{Bl}_0\mathbb{A}^{m+1} T 0 = Tot ( O P m ( − 1 )) = Bl 0 A m + 1 . Put W = T ′ × A m + 1 Spec A W=T'\times_{\mathbb{A}^{m+1}}\operatorname{Spec}A W = T ′ × A m + 1 Spec A and Z = T 0 × A m + 1 Spec A = Bl 0 Spec A Z=T_0\times_{\mathbb{A}^{m+1}}\operatorname{Spec}A=\operatorname{Bl}_0\operatorname{Spec}A Z = T 0 × A m + 1 Spec A = Bl 0 Spec A (blowing up commutes with flat base change). Let g : W → Z g:W\to Z g : W → Z be the base change of the finite morphism T → T 0 T\to T_0 T → T 0 and p : W → Spec ( R ⊗ S A ) = Spec R ^ p:W\to\operatorname{Spec}(R\otimes_S A)=\operatorname{Spec}\widehat{R} p : W → Spec ( R ⊗ S A ) = Spec R that of T → Spec R T\to\operatorname{Spec}R T → Spec R . The morphism T → A m + 1 T\to\mathbb{A}^{m+1} T → A m + 1 is proper (finite followed by a blow-up), hence so is W → Spec A W\to\operatorname{Spec}A W → Spec A ; since Spec R ^ → Spec A \operatorname{Spec}\widehat{R}\to\operatorname{Spec}A Spec R → Spec A is finite, hence separated, p p p is proper as well. Since m A O Z = I D \mathfrak{m}_A\mathcal{O}_Z=\mathcal{I}_D m A O Z = I D , the scheme g − 1 ( D ) g^{-1}(D) g − 1 ( D ) is the fibre of W W W over the closed point, which equals the fibre of T T T over 0 0 0 , namely X × P m D 0 = X 0 ≅ X X\times_{\mathbb{P}^m}D_0=X_0\cong X X × P m D 0 = X 0 ≅ X , where D 0 ≅ P m D_0\cong\mathbb{P}^m D 0 ≅ P m is the zero section of T 0 T_0 T 0 . Thus I X = I D O W \mathcal{I}_X=\mathcal{I}_D\mathcal{O}_W I X = I D O W is locally principal. Its local generators are nonzerodivisors and W W W is regular along X X X : for a closed point x ∈ X ⊂ W x\in X\subset W x ∈ X ⊂ W with image x ′ ∈ X 0 ⊂ T x'\in X_0\subset T x ′ ∈ X 0 ⊂ T , the local homomorphism O T , x ′ → O W , x \mathcal{O}_{T,x'}\to\mathcal{O}_{W,x} O T , x ′ → O W , x is flat with closed fibre κ ( x ′ ) ⊗ C A / ( x 0 , … , x m ) A = C \kappa(x')\otimes_{\mathbb{C}}A/(x_0,\ldots,x_m)A=\mathbb{C} κ ( x ′ ) ⊗ C A / ( x 0 , … , x m ) A = C , so O W , x \mathcal{O}_{W,x} O W , x is regular of dimension m + 1 m+1 m + 1 [[25 Thm. 23.7] ] and flatness preserves the nonzerodivisor. Then I X / I X 2 = g ∗ ( I D / I D 2 ) \mathcal{I}_X/\mathcal{I}_X^2=g^*(\mathcal{I}_D/\mathcal{I}_D^2) I X / I X 2 = g ∗ ( I D / I D 2 ) , i.e. I X ∣ X ≅ f ∗ O D ( 1 ) = O X ( H ) \mathcal{I}_X|_X\cong f^*\mathcal{O}_{D}(1)=\mathcal{O}_X(H) I X ∣ X ≅ f ∗ O D ( 1 ) = O X ( H ) . Since W W W is proper over Spec A \operatorname{Spec}A Spec A , every closed point of W W W lies on X X X , so O W , x \mathcal{O}_{W,x} O W , x is regular at every closed point x x x . Every point y y y of the Noetherian scheme W W W specializes to a closed point x x x , and O W , y \mathcal{O}_{W,y} O W , y is a localization of O W , x \mathcal{O}_{W,x} O W , x , hence regular [[25 Thm. 19.3] ]; so W W W is regular. Every connected component of W W W is proper over Spec A \operatorname{Spec}A Spec A , so it contains a closed point and meets the connected X X X ; hence W W W is connected and, being regular, integral. Finally W ∖ X W \setminus X W ∖ X is the base change of T ∘ ≅ Spec R ∖ V ( m ) T^{\circ} \cong\operatorname{Spec} R \setminus V(\mathfrak m) T ∘ ≅ Spec R ∖ V ( m ) along Spec R ^ → Spec R \operatorname{Spec} \widehat{R} \to\operatorname{Spec} R Spec R → Spec R , and the preimage of V ( m ) V(\mathfrak m) V ( m ) in Spec R ^ \operatorname{Spec} \widehat{R} Spec R is { m ^ } \{\widehat{\mathfrak m}\} { m } . The same argument applied to T 0 T_0 T 0 gives Z ∖ D ≅ Spec A ∖ { m A } Z \setminus D \cong\operatorname{Spec} A \setminus\{\mathfrak m_A\} Z ∖ D ≅ Spec A ∖ { m A } , compatibly with g g g . □
For m = 2 m=2 m = 2 this is Proposition 6.1 together with the morphisms f f f and g g g of Section 6, with X X X here in the role of Y Y Y there.
The variance lemma on P m \mathbb{P}^m P m
In the rest of this section h h h denotes the hyperplane class of P m \mathbb{P}^m P m . For a coherent sheaf F F F of rank N > 0 N>0 N > 0 on P m \mathbb{P}^m P m put
deg F = c 1 ( F ) ⋅ h m − 1 , v ( F ) = 2 ch 2 ( F ) ⋅ h m − 2 N − ( deg F N ) 2 \deg F=c_1(F)\mathbin{\cdot}h^{m-1}, \qquad v(F)=\frac{2\operatorname{ch}_2(F)\mathbin{\cdot}h^{m-2}}{N}-\left(\frac{\deg F}{N}\right)^2 deg F = c 1 ( F ) ⋅ h m − 1 , v ( F ) = N 2 ch 2 ( F ) ⋅ h m − 2 − ( N deg F ) 2
For m = 2 m=2 m = 2 this is the v v v of Section 8. As there, v v v is invariant under twists, and v ( ⨁ j O ( a j ) ) v\left(\bigoplus_j \mathcal{O}(a_j)\right) v ( ⨁ j O ( a j ) ) is the variance of ( a j ) (a_j) ( a j ) .
Lemma 10.5 (cf. [26 Lemma 4.1] ). Let Z Z Z be an integral Noetherian scheme and D ⊂ Z D\subset Z D ⊂ Z an effective Cartier divisor with D ≅ P m D\cong\mathbb{P}^m D ≅ P m and I D ∣ D ≅ O D ( 1 ) \mathcal{I}_{D|D}\cong\mathcal{O}_D(1) I D ∣ D ≅ O D ( 1 ) . If L ′ \mathcal{L}' L ′ is a torsion-free coherent sheaf on Z Z Z with L ′ ∣ Z ∖ D ≅ O Z ∖ D N \mathcal{L}'|_{Z\setminus D}\cong\mathcal{O}_{Z\setminus D}^N L ′ ∣ Z ∖ D ≅ O Z ∖ D N , then v ( L ′ ∣ D ) ≥ 0 v(\mathcal{L}'|_D)\ge0 v ( L ′ ∣ D ) ≥ 0 .
Proof. Clearing denominators. Write I = I D \mathcal{I}=\mathcal{I}_D I = I D , L = O Z N \mathcal{L}=\mathcal{O}_Z^N L = O Z N and j : Z ∖ D ↪ Z j:Z\setminus D\hookrightarrow Z j : Z ∖ D ↪ Z . A torsion-free sheaf on the integral scheme Z Z Z has no nonzero local sections supported on D D D , so L ′ \mathcal{L}' L ′ and L \mathcal{L} L both embed in j ∗ j ∗ L ′ = j ∗ j ∗ L j_*j^*\mathcal{L}'=j_*j^*\mathcal{L} j ∗ j ∗ L ′ = j ∗ j ∗ L . This sheaf is the increasing union of the coherent subsheaves I − q L \mathcal{I}^{-q}\mathcal{L} I − q L , and also of the I − q L ′ \mathcal{I}^{-q}\mathcal{L}' I − q L ′ , q ≥ 0 q\ge0 q ≥ 0 : on an affine open set where D D D is defined by τ \tau τ , a section of j ∗ j ∗ L j_*j^*\mathcal{L} j ∗ j ∗ L becomes a section of L \mathcal{L} L after multiplication by a power of τ \tau τ . By Noetherianity, I q L ⊆ L ′ ⊆ I − q L \mathcal{I}^q\mathcal{L}\subseteq\mathcal{L}'\subseteq\mathcal{I}^{-q}\mathcal{L} I q L ⊆ L ′ ⊆ I − q L for some q q q . Replacing L ′ \mathcal{L}' L ′ by I q L ′ \mathcal{I}^q\mathcal{L}' I q L ′ replaces L ′ ∣ D \mathcal{L}'|_D L ′ ∣ D by its twist L ′ ∣ D ( q ) \mathcal{L}'|_D(q) L ′ ∣ D ( q ) and does not change v v v . So we may assume I c L ⊆ L ′ ⊆ L \mathcal{I}^c\mathcal{L}\subseteq\mathcal{L}'\subseteq\mathcal{L} I c L ⊆ L ′ ⊆ L with c = 2 q c=2q c = 2 q .
The graded pieces. Let Q = L / L ′ Q=\mathcal{L}/\mathcal{L}' Q = L / L ′ and Q j = I j Q / I j + 1 Q Q_j=\mathcal{I}^jQ/\mathcal{I}^{j+1}Q Q j = I j Q / I j + 1 Q for 0 ≤ j < c 0\le j<c 0 ≤ j < c . These are coherent O D \mathcal{O}_D O D -modules and quotients of I j L / I j + 1 L ≅ O D ( j ) N \mathcal{I}^j\mathcal{L}/\mathcal{I}^{j+1}\mathcal{L}\cong\mathcal{O}_D(j)^N I j L / I j + 1 L ≅ O D ( j ) N . Put n j = rk Q j ≤ N n_j=\operatorname{rk}Q_j\le N n j = rk Q j ≤ N . We claim deg Q j ≥ j n j \deg Q_j\ge jn_j deg Q j ≥ j n j . Indeed Q j ( − j ) Q_j(-j) Q j ( − j ) is globally generated, and a globally generated coherent sheaf G G G on P m \mathbb{P}^m P m has deg G ≥ 0 \deg G\ge0 deg G ≥ 0 . To see this, note that its torsion subsheaf contributes an effective divisor class to c 1 c_1 c 1 . If G / G t o r s G/G_{\mathrm{tors}} G / G tors has rank n > 0 n>0 n > 0 , the surjection from a trivial sheaf induces a nonzero map ⋀ n O ⊕ M → det ( G / G t o r s ) \bigwedge^n\mathcal{O}^{\oplus M}\to\det(G/G_{\mathrm{tors}}) ⋀ n O ⊕ M → det ( G / G tors ) , so the determinant has a nonzero section.
Derived restriction. Since 0 → I → O Z → O D → 0 0\to\mathcal{I}\to\mathcal{O}_Z\to\mathcal{O}_D\to0 0 → I → O Z → O D → 0 is a locally free resolution, Tor i O Z ( − , O D ) = 0 \operatorname{Tor}^{\mathcal{O}_Z}_i(-,\mathcal{O}_D)=0 Tor i O Z ( − , O D ) = 0 for i ≥ 2 i\ge2 i ≥ 2 , and [ F ] D : = [ F ⊗ O D ] − [ Tor 1 O Z ( F , O D ) ] ∈ K ( D ) [F]_D:=[F\otimes\mathcal{O}_D]-[\operatorname{Tor}^{\mathcal{O}_Z}_1(F,\mathcal{O}_D)]\in K(D) [ F ] D := [ F ⊗ O D ] − [ Tor 1 O Z ( F , O D )] ∈ K ( D ) is additive on short exact sequences. For F F F torsion-free a local equation of D D D is F F F -regular, so Tor 1 = 0 \operatorname{Tor}_1=0 Tor 1 = 0 . For an O D \mathcal{O}_D O D -module G G G the same resolution gives Tor 1 ( G , O D ) ≅ G ⊗ I ∣ D = G ( 1 ) \operatorname{Tor}_1(G,\mathcal{O}_D)\cong G\otimes\mathcal{I}|_D=G(1) Tor 1 ( G , O D ) ≅ G ⊗ I ∣ D = G ( 1 ) . Applying this to 0 → L ′ → L → Q → 0 0\to\mathcal{L}'\to\mathcal{L}\to Q\to0 0 → L ′ → L → Q → 0 and to the filtration of Q Q Q ,
[ L ′ ∣ D ] = [ O D N ] + ∑ j ( [ Q j ( 1 ) ] − [ Q j ] ) in K ( P m ) . [\mathcal{L}'|_D]=[\mathcal{O}_D^N]+\sum_j\bigl([Q_j(1)]-[Q_j]\bigr)\quad\text{in }K(\mathbb{P}^m). [ L ′ ∣ D ] = [ O D N ] + j ∑ ( [ Q j ( 1 )] − [ Q j ] ) in K ( P m ) .
Since ch ( Q j ( 1 ) ) − ch ( Q j ) = ch ( Q j ) ( e h − 1 ) = n j h + ( c 1 ( Q j ) ⋅ h + n j 2 h 2 ) + ⋯ \operatorname{ch}(Q_j(1))-\operatorname{ch}(Q_j)=\operatorname{ch}(Q_j)(e^h-1)=n_jh+\left(c_1(Q_j)\mathbin{\cdot}h+\frac{n_j}{2}h^2\right)+\cdots ch ( Q j ( 1 )) − ch ( Q j ) = ch ( Q j ) ( e h − 1 ) = n j h + ( c 1 ( Q j ) ⋅ h + 2 n j h 2 ) + ⋯ , the sheaf F = L ′ ∣ D F=\mathcal{L}'|_D F = L ′ ∣ D has rank N N N , deg F = ∑ j n j \deg F=\sum_j n_j deg F = ∑ j n j and
2 ch 2 ( F ) ⋅ h m − 2 = ∑ j ( 2 deg Q j + n j ) ≥ ∑ j ( 2 j + 1 ) n j . 2\operatorname{ch}_2(F)\mathbin{\cdot}h^{m-2}=\sum_j(2\deg Q_j+n_j)\ge\sum_j(2j+1)n_j. 2 ch 2 ( F ) ⋅ h m − 2 = j ∑ ( 2 deg Q j + n j ) ≥ j ∑ ( 2 j + 1 ) n j .
Conclusion. Therefore
N 2 v ( F ) ≥ N ∑ j ( 2 j + 1 ) n j − ( ∑ j n j ) 2 = ∑ j n j ( N − n j ) + 2 ∑ j > ℓ n j ( N − n ℓ ) ≥ 0. N^2v(F)\ge N\sum_j(2j+1)n_j-\left(\sum_j n_j\right)^2=\sum_j n_j(N-n_j)+2\sum_{j>\ell}n_j(N-n_\ell)\ge0. N 2 v ( F ) ≥ N j ∑ ( 2 j + 1 ) n j − ( j ∑ n j ) 2 = j ∑ n j ( N − n j ) + 2 j > ℓ ∑ n j ( N − n ℓ ) ≥ 0.
For m = 2 m=2 m = 2 this is Lemma 8.4; the proof uses the dimension of D D D only through the definition of v v v .
Riemann–Roch for push-forwards.
Proposition 10.6. Let f : X → P m f : X \to\mathbb{P}^{m} f : X → P m be finite with f ∗ O ( 1 ) = O X ( H ) f^{*}\mathcal{O}(1) = \mathcal{O}_{X}(H) f ∗ O ( 1 ) = O X ( H ) , and let G \mathcal{G} G be a torsion-free sheaf of rank u u u on X X X . Put β = ν ( G ) − 1 2 K X \beta= \nu(\mathcal{G}) - \frac{1}{2}K_{X} β = ν ( G ) − 2 1 K X and
θ ( G ) = ( β ⋅ H m − 1 ) 2 − ( β 2 ⋅ H m − 2 ) d d . \theta(\mathcal{G}) = \frac{(\beta\cdot H^{m-1})^{2} - (\beta^{2} \cdot H^{m-2})d}{d}. θ ( G ) = d ( β ⋅ H m − 1 ) 2 − ( β 2 ⋅ H m − 2 ) d . Then f ∗ G f_{*}\mathcal{G} f ∗ G has rank u d ud u d , and
b ( G ) : = deg f ∗ G u d = μ ( G ) − λ 2 − m + 1 2 , (1) b(\mathcal{G}) := \frac{\deg f_{*}\mathcal{G}}{ud} = \mu(\mathcal{G}) - \frac{\lambda}{2} - \frac{m+1}{2}, \tag*{(1)} b ( G ) := u d deg f ∗ G = μ ( G ) − 2 λ − 2 m + 1 , (1) v ( f ∗ G ) = m + 1 12 − ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 12 d − δ ( G ) ⋅ H m − 2 + θ ( G ) d . (2) v(f_{*}\mathcal{G}) = \frac{m+1}{12} - \frac{(K_{X}^{2} - 2c_{2}(X)) \cdot H^{m-2}}{12d} - \frac{\delta(\mathcal{G}) \cdot H^{m-2} + \theta(\mathcal{G})}{d}. \tag*{(2)} v ( f ∗ G ) = 12 m + 1 − 12 d ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 − d δ ( G ) ⋅ H m − 2 + θ ( G ) . (2) Moreover θ ( G ) ≥ 0 \theta(\mathcal{G}) \ge0 θ ( G ) ≥ 0 , and if G \mathcal{G} G is μ \mu μ -semistable then δ ( G ) ⋅ H m − 2 ≥ 0 \delta(\mathcal{G}) \cdot H^{m-2} \ge0 δ ( G ) ⋅ H m − 2 ≥ 0 , so that
v ( f ∗ G ) ≤ m + 1 12 − ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 12 d . v(f_{*}\mathcal{G}) \le\frac{m+1}{12} - \frac{(K_{X}^{2} - 2c_{2}(X)) \cdot H^{m-2}}{12d}. v ( f ∗ G ) ≤ 12 m + 1 − 12 d ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 .
Proof. Let F = f ∗ G F = f_{*}\mathcal{G} F = f ∗ G , of rank N = u d N = ud N = u d . Since f f f is finite, the projection formula gives H i ( P m , F ( t ) ) = H i ( X , G ( t H ) ) H^{i}(\mathbb{P}^{m}, F(t)) = H^{i}(X, \mathcal{G}(tH)) H i ( P m , F ( t )) = H i ( X , G ( t H )) , so χ ( F ( t ) ) = χ ( G ( t H ) ) \chi(F(t)) = \chi(\mathcal{G}(tH)) χ ( F ( t )) = χ ( G ( t H )) as polynomials in t t t . We compare the coefficients of t m − 1 t^{m-1} t m − 1 and t m − 2 t^{m-2} t m − 2 given by Hirzebruch–Riemann–Roch. The Todd classes are
td ( P m ) = ( h 1 − e − h ) m + 1 = 1 + m + 1 2 h + ( m + 1 ) ( 3 m + 2 ) 24 h 2 + ⋯ , \operatorname{td}(\mathbb{P}^{m}) = \left(\frac{h}{1-e^{-h}}\right)^{m+1} = 1 + \frac{m+1}{2}h + \frac{(m+1)(3m+2)}{24}h^{2} + \cdots, td ( P m ) = ( 1 − e − h h ) m + 1 = 1 + 2 m + 1 h + 24 ( m + 1 ) ( 3 m + 2 ) h 2 + ⋯ ,
td ( X ) = 1 − K X 2 + K X 2 + c 2 ( X ) 12 + ⋯ , \operatorname{td}(X) = 1 - \frac{K_{X}}{2} + \frac{K_{X}^{2} + c_{2}(X)}{12} + \cdots, td ( X ) = 1 − 2 K X + 12 K X 2 + c 2 ( X ) + ⋯ ,
where the h 2 h^{2} h 2 -coefficient is m + 1 12 + ( m + 1 2 ) 1 4 \frac{m+1}{12} + \binom{m+1}{2}\frac{1}{4} 12 m + 1 + ( 2 m + 1 ) 4 1 . In χ = ∫ ch ( ⋅ ) e t H td \chi= \int\operatorname{ch}(\cdot)e^{tH}\operatorname{td} χ = ∫ ch ( ⋅ ) e t H td , the coefficient of t k / k ! t^{k}/k! t k / k ! is ∫ H k ⋅ [ ch ( ⋅ ) td ] m − k \int H^{k}\cdot[\operatorname{ch}(\cdot)\operatorname{td}]_{m-k} ∫ H k ⋅ [ ch ( ⋅ ) td ] m − k . With b = deg F / N b = \deg F/N b = deg F / N and e = ch 2 ( F ) ⋅ h m − 2 / N e = \operatorname{ch}_{2}(F)\cdot h^{m-2}/N e = ch 2 ( F ) ⋅ h m − 2 / N we obtain, for k = m − 1 k=m-1 k = m − 1 and k = m − 2 k=m-2 k = m − 2 ,
N ( b + m + 1 2 ) = c 1 ( G ) ⋅ H m − 1 − u 2 K X ⋅ H m − 1 , N\left(b + \frac{m+1}{2}\right) = c_{1}(\mathcal{G})\cdot H^{m-1} - \frac{u}{2}K_{X}\cdot H^{m-1}, N ( b + 2 m + 1 ) = c 1 ( G ) ⋅ H m − 1 − 2 u K X ⋅ H m − 1 ,
N ( e + m + 1 2 b + ( m + 1 ) ( 3 m + 2 ) 24 ) = ( ch 2 ( G ) − 1 2 c 1 ( G ) ⋅ K X ) ⋅ H m − 2 + u 12 ( K X 2 + c 2 ( X ) ) ⋅ H m − 2 . N\left(e + \frac{m+1}{2}b + \frac{(m+1)(3m+2)}{24}\right)
= \left(\operatorname{ch}_{2}(\mathcal{G}) - \frac{1}{2}c_{1}(\mathcal{G})\cdot K_{X}\right)\cdot H^{m-2}
+ \frac{u}{12}(K_{X}^{2} + c_{2}(X))\cdot H^{m-2}. N ( e + 2 m + 1 b + 24 ( m + 1 ) ( 3 m + 2 ) ) = ( ch 2 ( G ) − 2 1 c 1 ( G ) ⋅ K X ) ⋅ H m − 2 + 12 u ( K X 2 + c 2 ( X )) ⋅ H m − 2 .
(For m = 2 m=2 m = 2 the second line compares constant terms.) The first line is (1) . In the second, write ch 2 ( G ) / u = 1 2 ( ν 2 − δ ) \operatorname{ch}_{2}(\mathcal{G})/u = \frac{1}{2}(\nu^{2} - \delta) ch 2 ( G ) / u = 2 1 ( ν 2 − δ ) , which is the definition of Δ \Delta Δ , and divide by N N N :
2 e = ( ν 2 − ν ⋅ K X − δ ) ⋅ H m − 2 d + ( K X 2 + c 2 ) ⋅ H m − 2 6 d − ( m + 1 ) b − ( m + 1 ) ( 3 m + 2 ) 12 . 2e = \frac{(\nu^{2} - \nu\cdot K_{X} - \delta)\cdot H^{m-2}}{d}
+ \frac{(K_{X}^{2} + c_{2})\cdot H^{m-2}}{6d}
- (m+1)b - \frac{(m+1)(3m+2)}{12}. 2 e = d ( ν 2 − ν ⋅ K X − δ ) ⋅ H m − 2 + 6 d ( K X 2 + c 2 ) ⋅ H m − 2 − ( m + 1 ) b − 12 ( m + 1 ) ( 3 m + 2 ) .
Now ν 2 − ν ⋅ K X = β 2 − 1 4 K X 2 \nu^{2} - \nu\cdot K_{X} = \beta^{2} - \frac{1}{4}K_{X}^{2} ν 2 − ν ⋅ K X = β 2 − 4 1 K X 2 , and by (1) b + m + 1 2 = β ⋅ H m − 1 d b + \frac{m+1}{2} = \frac{\beta\cdot H^{m-1}}{d} b + 2 m + 1 = d β ⋅ H m − 1 , so
b 2 + ( m + 1 ) b = ( β ⋅ H m − 1 d ) 2 − ( m + 1 ) 2 4 . b^{2} + (m+1)b = \left(\frac{\beta\cdot H^{m-1}}{d}\right)^{2} - \frac{(m+1)^{2}}{4}. b 2 + ( m + 1 ) b = ( d β ⋅ H m − 1 ) 2 − 4 ( m + 1 ) 2 .
Hence
v ( F ) = 2 e − b 2 = β 2 ⋅ H m − 2 d − ( β ⋅ H m − 1 ) 2 d 2 − δ ⋅ H m − 2 d + ( K X 2 + c 2 6 − K X 2 4 ) H m − 2 d + ( m + 1 ) 2 4 − ( m + 1 ) ( 3 m + 2 ) 12 . v(F) = 2e - b^{2}
= \frac{\beta^{2}\cdot H^{m-2}}{d}
- \frac{(\beta\cdot H^{m-1})^{2}}{d^{2}}
- \frac{\delta\cdot H^{m-2}}{d}
+ \left(\frac{K_{X}^{2}+c_{2}}{6} - \frac{K_{X}^{2}}{4}\right)\frac{H^{m-2}}{d}
+ \frac{(m+1)^{2}}{4} - \frac{(m+1)(3m+2)}{12}. v ( F ) = 2 e − b 2 = d β 2 ⋅ H m − 2 − d 2 ( β ⋅ H m − 1 ) 2 − d δ ⋅ H m − 2 + ( 6 K X 2 + c 2 − 4 K X 2 ) d H m − 2 + 4 ( m + 1 ) 2 − 12 ( m + 1 ) ( 3 m + 2 ) .
The first two terms are − θ / d -\theta/d − θ / d , the bracket is − 1 12 ( K X 2 − 2 c 2 ) -\frac{1}{12}(K_{X}^{2}-2c_{2}) − 12 1 ( K X 2 − 2 c 2 ) , and the constants add up to m + 1 12 \frac{m+1}{12} 12 m + 1 . This is (2) .
For θ ≥ 0 \theta\ge0 θ ≥ 0 : both sides of ( β ⋅ H m − 1 ) 2 ≥ ( β 2 ⋅ H m − 2 ) H m (\beta\cdot H^{m-1})^{2} \ge(\beta^{2}\cdot H^{m-2})H^{m} ( β ⋅ H m − 1 ) 2 ≥ ( β 2 ⋅ H m − 2 ) H m are homogeneous of degree 2 m − 2 2m-2 2 m − 2 in H H H , so we may replace H H H by a very ample multiple. Let X 2 X_{2} X 2 be the intersection of m − 2 m-2 m − 2 general members of ∣ H ∣ \lvert H\rvert ∣ H ∣ . It is smooth by Bertini’s theorem [13 Cor. III.10.9] and connected [13 Cor. III.7.9] . Restricting to X 2 X_2 X 2 , with β 2 = β ∣ X 2 \beta_2=\left.\beta\right|_{X_2} β 2 = β ∣ X 2 and H 2 = H ∣ X 2 H_2=\left.H\right|_{X_2} H 2 = H ∣ X 2 , gives β ⋅ H m − 1 = β 2 ⋅ H 2 \beta\cdot H^{m-1}=\beta_2\cdot H_2 β ⋅ H m − 1 = β 2 ⋅ H 2 , β 2 ⋅ H m − 2 = β 2 2 \beta^2\cdot H^{m-2}=\beta_2^2 β 2 ⋅ H m − 2 = β 2 2 and d = H 2 2 d=H_2^2 d = H 2 2 , and the Hodge index theorem on X 2 X_2 X 2 gives ( β 2 ⋅ H 2 ) 2 ≥ β 2 2 H 2 2 (\beta_2\cdot H_2)^2\geq\beta_2^2H_2^2 ( β 2 ⋅ H 2 ) 2 ≥ β 2 2 H 2 2 . Finally, Δ ( G ) ⋅ H m − 2 ≥ 0 \Delta(\mathcal{G})\cdot H^{m-2}\geq0 Δ ( G ) ⋅ H m − 2 ≥ 0 for μ \mu μ -semistable torsion-free sheaves is Bogomolov’s inequality [16 Thm. 7.3.1] . □ \square □
Remark 10.7. For m = 2 m=2 m = 2 , θ = − α 2 \theta=-\alpha^2 θ = − α 2 in the notation of Lemma 5.5, and (2) is the equality of Proposition 9.1 ; its inequality is Lemma 8.5. Three checks in dimension m m m :
for ( X , H ) = ( P m , O ( 1 ) ) (X,H)=(\mathbb{P}^m,\mathcal{O}(1)) ( X , H ) = ( P m , O ( 1 )) , f = id f=\operatorname{id} f = id and G = O \mathcal{G}=\mathcal{O} G = O one has K 2 − 2 c 2 = ( m + 1 ) h 2 K^2-2c_2=(m+1)h^2 K 2 − 2 c 2 = ( m + 1 ) h 2 , and both sides of (2) vanish;
for ( X , H ) = ( P m , O ( 2 ) ) (X,H)=(\mathbb{P}^m,\mathcal{O}(2)) ( X , H ) = ( P m , O ( 2 )) , f = ( x 0 2 : ⋯ : x m 2 ) f=(x_0^2:\cdots:x_m^2) f = ( x 0 2 : ⋯ : x m 2 ) and G = O \mathcal{G}=\mathcal{O} G = O , one has f ∗ O = ⨁ O ( − ∣ ϵ ∣ / 2 ) f_*\mathcal{O}=\bigoplus\mathcal{O}(-|\epsilon|/2) f ∗ O = ⨁ O ( − ∣ ϵ ∣/2 ) over the subsets ϵ ⊆ { 0 , … , m } \epsilon\subseteq\{0,\ldots,m\} ϵ ⊆ { 0 , … , m } of even cardinality, whose variance is m + 1 16 \frac{m+1}{16} 16 m + 1 for m ≥ 2 m\geq2 m ≥ 2 . This agrees with (2) , where δ = θ = 0 \delta=\theta=0 δ = θ = 0 , d = 2 m d=2^m d = 2 m and ( K 2 − 2 c 2 ) ⋅ H m − 2 = ( m + 1 ) 2 m − 2 (K^2-2c_2)\cdot H^{m-2}=(m+1)2^{m-2} ( K 2 − 2 c 2 ) ⋅ H m − 2 = ( m + 1 ) 2 m − 2 . For m = 2 m=2 m = 2 this is [2 Rem. 2.7] .
if f ∗ G f_*\mathcal{G} f ∗ G is trivial, then b ( G ) = 0 b(\mathcal{G})=0 b ( G ) = 0 , so β ⋅ H m − 1 = m + 1 2 d \beta\cdot H^{m-1}=\frac{m+1}{2}d β ⋅ H m − 1 = 2 m + 1 d by (1) , and θ ( G ) = − α ′ 2 ⋅ H m − 2 / u 2 \theta(\mathcal{G})=-\alpha'^2\cdot H^{m-2}/u^2 θ ( G ) = − α ′2 ⋅ H m − 2 / u 2 with α ′ = c 1 ( G ) − u 2 ( K X + ( m + 1 ) H ) \alpha'=c_1(\mathcal{G})-\frac{u}{2}(K_X+(m+1)H) α ′ = c 1 ( G ) − 2 u ( K X + ( m + 1 ) H ) ; multiplied by u 2 u^2 u 2 , (2) becomes Δ ( G ) ⋅ H m − 2 = α ′ 2 ⋅ H m − 2 + u 2 12 ( ( m + 1 ) H 2 − K X 2 + 2 c 2 ( X ) ) ⋅ H m − 2 \Delta(\mathcal{G})\cdot H^{m-2}=\alpha'^2\cdot H^{m-2}+\frac{u^2}{12}((m+1)H^2-K_X^2+2c_2(X))\cdot H^{m-2} Δ ( G ) ⋅ H m − 2 = α ′2 ⋅ H m − 2 + 12 u 2 (( m + 1 ) H 2 − K X 2 + 2 c 2 ( X )) ⋅ H m − 2 , the identity in the proof of [24 Lemma 3.1] .
Proof of Theorem 10.3
Let M M M be a small CM R ^ \widehat{R} R -module; we use the notation of Proposition 10.4 .
Step 1: freeness. The elements x 0 , … , x m x_0,\ldots,x_m x 0 , … , x m form a system of parameters of R ^ \widehat{R} R , hence an M M M -regular sequence. So depth A M = m + 1 \operatorname{depth}_A M=m+1 depth A M = m + 1 , and the Auslander–Buchsbaum formula over the regular local ring A A A gives M ≅ A N M\cong A^N M ≅ A N for some N ≥ 1 N\geq1 N ≥ 1 .
Step 2: extension. Every associated prime q \mathfrak{q} q of M M M satisfies dim R ^ / q ≥ depth M = dim R ^ \dim\widehat{R}/\mathfrak{q}\geq\operatorname{depth}M=\dim\widehat{R} dim R / q ≥ depth M = dim R [7 Prop. 1.2.13] , so q = 0 \mathfrak{q}=0 q = 0 : M M M is torsion-free, of some rank r ≥ 1 r\geq1 r ≥ 1 ; since R ^ \widehat{R} R has generic rank d d d over A A A , N = r d N=rd N = r d . For a prime p ≠ m ^ \mathfrak{p}\neq\widehat{\mathfrak{m}} p = m , M p M_{\mathfrak{p}} M p is Cohen–Macaulay [7 Thm. 2.1.3] with support Spec R ^ p \operatorname{Spec}\widehat{R}_{\mathfrak{p}} Spec R p , hence maximal Cohen–Macaulay over the regular local ring R ^ p \widehat{R}_{\mathfrak{p}} R p (Proposition 10.4 (c)), hence free. So M ~ \widetilde{M} M is locally free of rank r r r on U = Spec R ^ ∖ { m ^ } ≅ W ∖ X U=\operatorname{Spec}\widehat{R}\setminus\{\widehat{\mathfrak{m}}\}\cong W\setminus X U = Spec R ∖ { m } ≅ W ∖ X .
Let E = ( p ∗ M ~ ) ∨ ∨ \mathcal{E}=(p^*\widetilde{M})^{\vee\vee} E = ( p ∗ M ) ∨∨ . It is torsion-free and E ∣ W ∖ X ≅ M ~ ∣ U \mathcal{E}|_{W\setminus X}\cong\widetilde{M}|_U E ∣ W ∖ X ≅ M ∣ U . For x ∈ W x\in W x ∈ W one has depth E x ≥ min ( 2 , dim O W , x ) \operatorname{depth}\mathcal{E}_x\geq\min(2,\dim\mathcal{O}_{W,x}) depth E x ≥ min ( 2 , dim O W , x ) , since E x = Hom ( E x ∨ , O W , x ) \mathcal{E}_x=\operatorname{Hom}(\mathcal{E}_x^\vee,\mathcal{O}_{W,x}) E x = Hom ( E x ∨ , O W , x ) and Hom into a module of depth ≥ 2 \geq2 ≥ 2 has depth ≥ 2 \geq2 ≥ 2 [7 Ex. 1.4.19] . If x ∈ X x\in X x ∈ X is not the generic point of X X X , then dim O W , x ≥ 2 \dim\mathcal{O}_{W,x}\geq2 dim O W , x ≥ 2 and a local equation of X X X is E x \mathcal{E}_x E x -regular, so depth ( E ∣ X ) x ≥ 1 \operatorname{depth}(\mathcal{E}|_X)_x\geq1 depth ( E ∣ X ) x ≥ 1 . Hence E ∣ X \mathcal{E}|_X E ∣ X is torsion-free. At the generic point of X X X the local ring of W W W is a discrete valuation ring, so E \mathcal{E} E is free of rank r r r there and E ∣ X \mathcal{E}|_X E ∣ X has rank r r r .
Step 3: modifications ([9 Lemma 3.1] , [26 Lemma 3.1] ). Let i : X ↪ W i:X\hookrightarrow W i : X ↪ W be the inclusion, F \mathcal{F} F a torsion-free sheaf on W W W with F ∣ X \mathcal{F}|_X F ∣ X torsion-free, and F ∣ X → Q \mathcal{F}|_X\to\mathcal{Q} F ∣ X → Q a torsion-free quotient with kernel K \mathcal{K} K . The composite F → i ∗ ( F ∣ X ) → i ∗ Q \mathcal{F}\to i_*(\mathcal{F}|_X)\to i_*\mathcal{Q} F → i ∗ ( F ∣ X ) → i ∗ Q is surjective; let F ′ \mathcal{F}' F ′ be its kernel. It is torsion-free and agrees with F \mathcal{F} F off X X X . Tensor 0 → F ′ → F → i ∗ Q → 0 0\to\mathcal{F}'\to\mathcal{F}\to i_*\mathcal{Q}\to0 0 → F ′ → F → i ∗ Q → 0 with O X \mathcal{O}_X O X . A local equation of X X X is a nonzerodivisor on the torsion-free sheaf F \mathcal{F} F on the integral scheme W W W , so Tor 1 O W ( F , O X ) = 0 \operatorname{Tor}_1^{\mathcal{O}_W}(\mathcal{F},\mathcal{O}_X)=0 Tor 1 O W ( F , O X ) = 0 , while the resolution 0 → I X → O W → O X → 0 0\to\mathcal{I}_X\to\mathcal{O}_W\to\mathcal{O}_X\to0 0 → I X → O W → O X → 0 gives Tor 1 O W ( i ∗ Q , O X ) ≅ Q ⊗ I X ∣ X = Q ( H ) \operatorname{Tor}_1^{\mathcal{O}_W}(i_*\mathcal{Q},\mathcal{O}_X)\cong\mathcal{Q}\otimes\mathcal{I}_X|_X=\mathcal{Q}(H) Tor 1 O W ( i ∗ Q , O X ) ≅ Q ⊗ I X ∣ X = Q ( H ) . The long exact sequence is 0 → Q ( H ) → F ′ ∣ X → F ∣ X → Q → 0 0\to\mathcal{Q}(H)\to\mathcal{F}'|_X\to\mathcal{F}|_X\to\mathcal{Q}\to0 0 → Q ( H ) → F ′ ∣ X → F ∣ X → Q → 0 , that is,
0 → Q ( H ) → F ′ ∣ X → K → 0. 0\to\mathcal{Q}(H)\to\mathcal{F}'|_X\to\mathcal{K}\to0. 0 → Q ( H ) → F ′ ∣ X → K → 0.
So F ′ ∣ X \mathcal{F}'|_X F ′ ∣ X is torsion-free of the same rank, and μ ( Q ( H ) ) = μ ( Q ) + 1 \mu(\mathcal{Q}(H))=\mu(\mathcal{Q})+1 μ ( Q ( H )) = μ ( Q ) + 1 . Now suppose μ max ( P ) − μ min ( P ) > 1 \mu_{\max}(P)-\mu_{\min}(P)>1 μ m a x ( P ) − μ m i n ( P ) > 1 for P = F ∣ X P=\mathcal{F}|_X P = F ∣ X , and modify along the first Harder–Narasimhan quotient G G G of P P P , of slope μ min ( P ) \mu_{\min}(P) μ m i n ( P ) , with kernel B B B . The new restriction P ′ P' P ′ is an extension of B B B by the semistable sheaf G ( H ) G(H) G ( H ) of slope μ min ( P ) + 1 < μ max ( P ) \mu_{\min}(P)+1<\mu_{\max}(P) μ m i n ( P ) + 1 < μ m a x ( P ) , so μ max ( P ′ ) ≤ μ max ( P ) \mu_{\max}(P')\leq\mu_{\max}(P) μ m a x ( P ′ ) ≤ μ m a x ( P ) , while μ ( P ′ ) = μ ( P ) + r K / r P ≥ μ ( P ) + 1 / r \mu(P')=\mu(P)+r_K/r_P\geq\mu(P)+1/r μ ( P ′ ) = μ ( P ) + r K / r P ≥ μ ( P ) + 1/ r . Since μ ( P ′ ) ≤ μ max ( P ′ ) \mu(P')\leq\mu_{\max}(P') μ ( P ′ ) ≤ μ m a x ( P ′ ) stays bounded by the initial μ max \mu_{\max} μ m a x , the process stops after finitely many steps. Starting from E \mathcal{E} E of Step 2, we obtain a torsion-free sheaf on W W W with E ∣ W ∖ X ≅ M ~ ∣ U \mathcal{E}|_{W\setminus X}\cong\widetilde{M}|_U E ∣ W ∖ X ≅ M ∣ U , whose restriction P = E ∣ X P=\mathcal{E}|_X P = E ∣ X is torsion-free of rank r r r , with Harder–Narasimhan factors G 1 , … , G s \mathcal{G}_1,\ldots,\mathcal{G}_s G 1 , … , G s satisfying μ ( G 1 ) − μ ( G s ) ≤ 1 \mu(\mathcal{G}_1)-\mu(\mathcal{G}_s)\leq1 μ ( G 1 ) − μ ( G s ) ≤ 1 . For m = 2 m=2 m = 2 this is Lemma 6.4(a).
Step 4: v ( f ∗ P ) ≥ 0 v(f_*P) \ge0 v ( f ∗ P ) ≥ 0 . The sheaf L ′ = g ∗ E \mathcal{L}' = g_*\mathcal{E} L ′ = g ∗ E on Z Z Z is coherent, and torsion-free because g g g is finite and dominant and W W W is integral. Over Z ∖ D ≅ Spec A ∖ { m A } Z \setminus D \cong\operatorname{Spec} A \setminus\{\mathfrak{m}_A\} Z ∖ D ≅ Spec A ∖ { m A } , the morphism g g g is identified with Spec R ^ ∖ { m ^ } → Spec A ∖ { m A } \operatorname{Spec} \widehat{R} \setminus\{\widehat{\mathfrak{m}}\} \to\operatorname{Spec} A \setminus\{\mathfrak{m}_A\} Spec R ∖ { m } → Spec A ∖ { m A } (Proposition 10.4(c)), and E ∣ W ∖ X ≅ M ^ ∣ U \mathcal{E}|_{W \setminus X} \cong\widehat{M}|_U E ∣ W ∖ X ≅ M ∣ U is unchanged by Step 3; so L ′ \mathcal{L}' L ′ is there the sheaf associated with the A A A -module M ≅ A N M \cong A^N M ≅ A N , hence trivial. Since g g g is affine and g ∗ O D = O X g_*\mathcal{O}_D = \mathcal{O}_X g ∗ O D = O X , L ′ ∣ D = g ∗ ( E ∣ X ) = f ∗ P \mathcal{L}'|_D = g_*(\mathcal{E}|_X) = f_*P L ′ ∣ D = g ∗ ( E ∣ X ) = f ∗ P . Lemma 10.5 gives v ( f ∗ P ) ≥ 0 v(f_*P) \ge0 v ( f ∗ P ) ≥ 0 .
Step 5: upper bound. The functor f ∗ f_* f ∗ is exact, so f ∗ P f_*P f ∗ P has a filtration with successive quotients f ∗ G i f_*\mathcal{G}_i f ∗ G i , with weights w i = rk G i / r w_i = \operatorname{rk}\mathcal{G}_i/r w i = rk G i / r . With b i = b ( G i ) b_i = b(\mathcal{G}_i) b i = b ( G i ) and b ˉ = ∑ i w i b i = deg f ∗ P / N \bar{b} = \sum_i w_i b_i = \deg f_*P/N b ˉ = ∑ i w i b i = deg f ∗ P / N , additivity of rank and ch \operatorname{ch} ch gives, as in the proof of Proposition 9.1,
v ( f ∗ P ) = ∑ i w i v ( f ∗ G i ) + ∑ i w i ( b i − b ˉ ) 2 . v(f_*P) = \sum_i w_i v(f_*\mathcal{G}_i) + \sum_i w_i(b_i-\bar{b})^2. v ( f ∗ P ) = i ∑ w i v ( f ∗ G i ) + i ∑ w i ( b i − b ˉ ) 2 .
By (1) , b i − μ ( G i ) b_i-\mu(\mathcal{G}_i) b i − μ ( G i ) does not depend on i i i , so the b i b_i b i lie in an interval of length ≤ 1 \le1 ≤ 1 ; with γ \gamma γ its midpoint, ∑ i w i ( b i − b ˉ ) 2 ≤ ∑ i w i ( b i − γ ) 2 ≤ 1 4 \sum_i w_i(b_i-\bar{b})^2 \le\sum_i w_i(b_i-\gamma)^2 \le\frac{1}{4} ∑ i w i ( b i − b ˉ ) 2 ≤ ∑ i w i ( b i − γ ) 2 ≤ 4 1 . By Proposition 10.6 applied to the semistable G i \mathcal{G}_i G i , and by Step 4,
0 ≤ v ( f ∗ P ) ≤ m + 1 12 − ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 12 d + 1 4 , 0 \le v(f_*P) \le\frac{m+1}{12} - \frac{(K_X^2-2c_2(X))\cdot H^{m-2}}{12d}+\frac{1}{4}, 0 ≤ v ( f ∗ P ) ≤ 12 m + 1 − 12 d ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 + 4 1 ,
that is, ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 ≤ ( m + 4 ) d (K_X^2-2c_2(X))\cdot H^{m-2} \le(m+4)d ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 ≤ ( m + 4 ) d .
If R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) has no small CM module, then neither has R ( X , H ) m R(X,H)_m R ( X , H ) m : by faithful flatness of completion, an M M M -regular sequence of length m + 1 m+1 m + 1 over R m R_m R m stays regular on M ^ \widehat{M} M (as in Corollary 8.3).
One identity in every dimension
Let f : X → P m f:X\to\mathbb{P}^m f : X → P m be finite with f ∗ O ( 1 ) = O X ( H ) f^*\mathcal{O}(1)=\mathcal{O}_X(H) f ∗ O ( 1 ) = O X ( H ) , and let P P P be a torsion-free sheaf on X X X with Harder–Narasimhan factors G i \mathcal{G}_i G i , weights w i w_i w i , push-forward slopes b i b_i b i with weighted mean b ˉ \bar{b} b ˉ , and v b = ∑ i w i ( b i − b ˉ ) 2 v_b=\sum_i w_i(b_i-\bar{b})^2 v b = ∑ i w i ( b i − b ˉ ) 2 . Proposition 10.6 and the decomposition of Step 5 give
− I ( X , H ) 12 = d ( v ( f ∗ P ) − v b ) + ∑ i w i ( δ ( G i ) ⋅ H m − 2 + θ ( G i ) ) . (3) -\frac{I(X,H)}{12}=d\bigl(v(f_*P)-v_b\bigr)+\sum_i w_i\bigl(\delta(\mathcal{G}_i)\cdot H^{m-2}+\theta(\mathcal{G}_i)\bigr). \tag*{(3)} − 12 I ( X , H ) = d ( v ( f ∗ P ) − v b ) + i ∑ w i ( δ ( G i ) ⋅ H m − 2 + θ ( G i ) ) . (3)
and the last sum is ≥ 0 \ge0 ≥ 0 . For m = 2 m=2 m = 2 , − I / 12 = ( d − σ ) / 4 -I/12=(d-\sigma)/4 − I /12 = ( d − σ ) /4 and (3) is the identity behind Table 3 . For the aCM case we need majorization on P m \mathbb{P}^m P m .
Lemma 10.8. Let V = ⨁ j = 1 N O P m ( a j ) V=\bigoplus_{j=1}^{N}\mathcal{O}_{\mathbb{P}^m}(a_j) V = ⨁ j = 1 N O P m ( a j ) with a 1 ≥ ⋯ ≥ a N a_1\ge\cdots\ge a_N a 1 ≥ ⋯ ≥ a N , and let 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V s = V 0=V_0\subset V_1\subset\cdots\subset V_s=V 0 = V 0 ⊂ V 1 ⊂ ⋯ ⊂ V s = V be a filtration by subsheaves whose successive quotients have ranks n i > 0 n_i>0 n i > 0 and slopes b 1 ≥ ⋯ ≥ b s b_1\ge\cdots\ge b_s b 1 ≥ ⋯ ≥ b s (slope = deg / rk =\deg/\operatorname{rk} = deg / rk ). Then ∑ i n i ( b i + c ) 2 ≤ ∑ j ( a j + c ) 2 \sum_i n_i(b_i+c)^2\le\sum_j(a_j+c)^2 ∑ i n i ( b i + c ) 2 ≤ ∑ j ( a j + c ) 2 for every c ∈ R c\in\mathbb{R} c ∈ R . In particular, with a ˉ = 1 N ∑ j a j \bar{a}=\frac{1}{N}\sum_j a_j a ˉ = N 1 ∑ j a j , v a = 1 N ∑ j ( a j − a ˉ ) 2 v_a=\frac{1}{N}\sum_j(a_j-\bar{a})^2 v a = N 1 ∑ j ( a j − a ˉ ) 2 and v b = ∑ i n i N ( b i − a ˉ ) 2 v_b=\sum_i\frac{n_i}{N}(b_i-\bar{a})^2 v b = ∑ i N n i ( b i − a ˉ ) 2 , one has v b ≤ v a v_b\le v_a v b ≤ v a .
Proof. Step 1. If S ⊂ V S\subset V S ⊂ V is a subsheaf of rank p p p , then deg S ≤ a 1 + ⋯ + a p \deg S\le a_1+\cdots+a_p deg S ≤ a 1 + ⋯ + a p . Indeed det S = ( ⋀ p S ) ∨ ∨ \det S=(\bigwedge^p S)^{\vee\vee} det S = ( ⋀ p S ) ∨∨ is invertible, and ⋀ p S → ⋀ p V = ⨁ J O ( a J ) \bigwedge^p S\to\bigwedge^p V=\bigoplus_J\mathcal{O}(a_J) ⋀ p S → ⋀ p V = ⨁ J O ( a J ) , ∣ J ∣ = p |J|=p ∣ J ∣ = p , is nonzero; so some component det S → O ( a J ) \det S\to\mathcal{O}(a_J) det S → O ( a J ) is nonzero on the open set where S S S is locally free, whose complement has codimension ≥ 2 \ge2 ≥ 2 since S ⊂ V S\subset V S ⊂ V is torsion-free on the smooth variety P m \mathbb{P}^m P m , and extends across it. Hence deg S ≤ a J ≤ a 1 + ⋯ + a p \deg S\le a_J\le a_1+\cdots+a_p deg S ≤ a J ≤ a 1 + ⋯ + a p .
Step 2. Let x 1 ≥ ⋯ ≥ x N x_1\ge\cdots\ge x_N x 1 ≥ ⋯ ≥ x N be the sequence in which b i b_i b i is repeated n i n_i n i times, and Σ x ( p ) = ∑ j ≤ p x j \Sigma_x(p)=\sum_{j\le p}x_j Σ x ( p ) = ∑ j ≤ p x j , Σ a ( p ) = ∑ j ≤ p a j \Sigma_a(p)=\sum_{j\le p}a_j Σ a ( p ) = ∑ j ≤ p a j . By Step 1, Σ x ( p ) = deg V i ≤ Σ a ( p ) \Sigma_x(p)=\deg V_i\le\Sigma_a(p) Σ x ( p ) = deg V i ≤ Σ a ( p ) for p = n 1 + ⋯ + n i p=n_1+\cdots+n_i p = n 1 + ⋯ + n i . Between two such values Σ x \Sigma_x Σ x is affine and Σ a \Sigma_a Σ a is concave, so Σ x ≤ Σ a \Sigma_x\le\Sigma_a Σ x ≤ Σ a everywhere, with equality at p = N p=N p = N (Shatz [Shal]).
Step 3. Put S p = Σ a ( p ) − Σ x ( p ) ≥ 0 S_p=\Sigma_a(p)-\Sigma_x(p)\ge0 S p = Σ a ( p ) − Σ x ( p ) ≥ 0 , so S 0 = S N = 0 S_0=S_N=0 S 0 = S N = 0 , and y p = a p + x p + 2 c y_p=a_p+x_p+2c y p = a p + x p + 2 c , a non-increasing sequence. Abel summation gives
∑ j [ ( a j + c ) 2 − ( x j + c ) 2 ] = ∑ j ( a j − x j ) y j = ∑ p < N S p ( y p − y p + 1 ) ≥ 0 , \sum_j\left[(a_j+c)^2-(x_j+c)^2\right]=\sum_j(a_j-x_j)y_j=\sum_{p<N}S_p(y_p-y_{p+1})\ge0, j ∑ [ ( a j + c ) 2 − ( x j + c ) 2 ] = j ∑ ( a j − x j ) y j = p < N ∑ S p ( y p − y p + 1 ) ≥ 0 ,
which is Karamata’s inequality [22 ] ; compare [23 Lemma 1.5] . For m = 2 m=2 m = 2 this is Lemma 5.4. All three obstructions are read off from (3) in every dimension.
(i) Ulrich (H H H very ample, f f f a linear projection). f ∗ E f_*\mathcal{E} f ∗ E is trivial by Proposition 2.2 , so v = v a = 0 v=v_a=0 v = v a = 0 , and Lemma 10.8 gives v b = 0 v_b=0 v b = 0 : E \mathcal{E} E is μ \mu μ -semistable (cf. [8 Thm. 2.9] ). Hence I ≤ 0 I\leq0 I ≤ 0 , which is the first assertion of Proposition 3.2 .
(ii) aCM (H i ( X , E ( t ) ) = 0 H^i(X,\mathcal{E}(t))=0 H i ( X , E ( t )) = 0 for 0 < i < m 0<i<m 0 < i < m and all t t t ). f f f is flat by miracle flatness [25 Thm. 23.1] , so f ∗ E f_*\mathcal{E} f ∗ E is locally free, has no intermediate cohomology and splits by Horrocks’ criterion [21 ] , so v = v a v=v_a v = v a . The push-forward slopes of the Harder–Narasimhan factors are decreasing by (1) , so Lemma 10.8 gives v b ≤ v a v_b\leq v_a v b ≤ v a . Hence I ≤ 0 I\leq0 I ≤ 0 , which is Proposition 5.9 , now without restricting the bundle to a surface section (the Hodge index theorem is still applied on one, in Proposition 10.6 ). For m = 2 m=2 m = 2 , (3) is the defect identity of Proposition 5.6 .
(iii) Local. v ≥ 0 v\geq0 v ≥ 0 by Lemma 10.5 and v b ≤ 1 4 v_b\leq\frac{1}{4} v b ≤ 4 1 after modifications. Hence I ≤ 3 d I\leq3d I ≤ 3 d , which is Theorem 10.3.
The gap between the graded and the local bound is always 3 H m 3H^m 3 H m , the same 1 4 \frac{1}{4} 4 1 as for surfaces. Relative to ( m + 1 ) H m (m+1)H^m ( m + 1 ) H m , however, it shrinks: the ratio of the two bounds on ( K 2 − 2 c 2 ) ⋅ H m − 2 (K^2-2c_2)\cdot H^{m-2} ( K 2 − 2 c 2 ) ⋅ H m − 2 is m + 4 m + 1 \frac{m+4}{m+1} m + 1 m + 4 .
Isolated singularities in every dimension.
Corollary 10.9 (cf. [2 Cor. 5.3, Rem. 5.5] for the graded part). Let Y Y Y be a smooth projective surface, H Y H_Y H Y an ample, globally generated divisor, d Y = H Y 2 d_Y=H_Y^2 d Y = H Y 2 and σ = σ ( Y ) \sigma=\sigma(Y) σ = σ ( Y ) . For m ≥ 2 m\geq2 m ≥ 2 let X = Y × P m − 2 X=Y\times\mathbb{P}^{m-2} X = Y × P m − 2 and H = H Y ⊠ O ( 1 ) H=H_Y\boxtimes\mathcal{O}(1) H = H Y ⊠ O ( 1 ) . Then
I ( X , H ) = 3 ( σ − ( m − 1 ) 2 d Y ) , H m = ( m 2 ) d Y . I(X,H)=3\bigl(\sigma-(m-1)^2d_Y\bigr),\qquad H^m=\binom{m}{2}d_Y. I ( X , H ) = 3 ( σ − ( m − 1 ) 2 d Y ) , H m = ( 2 m ) d Y . Consequently ( X , H ) (X,H) ( X , H ) carries no aCM bundle if σ > ( m − 1 ) 2 d Y \sigma>(m-1)^2d_Y σ > ( m − 1 ) 2 d Y (Proposition 5.9 ). If
σ > ( m − 1 ) ( 3 m − 2 ) 2 d Y , \sigma>\frac{(m-1)(3m-2)}{2}d_Y, σ > 2 ( m − 1 ) ( 3 m − 2 ) d Y , then the completion R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) of the Segre product R ( Y , H Y ) # C [ s 0 , … , s m − 2 ] R(Y,H_Y)\mathbin{\#}\mathbb{C}[s_0,\ldots,s_{m-2}] R ( Y , H Y ) # C [ s 0 , … , s m − 2 ] is a complete normal local domain of dimension m + 1 m+1 m + 1 , with residue field C \mathbb{C} C and an isolated singularity, that has no small CM module.
Proof. Since T X = pr 1 ∗ T Y ⊕ pr 2 ∗ T P m − 2 T_X=\operatorname{pr}_1^*T_Y\oplus\operatorname{pr}_2^*T_{\mathbb{P}^{m-2}} T X = pr 1 ∗ T Y ⊕ pr 2 ∗ T P m − 2 and K 2 − 2 c 2 = 2 ch 2 ( T ) K^2-2c_2=2\operatorname{ch}_2(T) K 2 − 2 c 2 = 2 ch 2 ( T ) is additive,
K X 2 − 2 c 2 ( X ) = pr 1 ∗ ( K Y 2 − 2 c 2 ( Y ) ) + ( m − 1 ) pr 2 ∗ h 2 , K_X^2-2c_2(X)=\operatorname{pr}_1^*(K_Y^2-2c_2(Y))+(m-1)\operatorname{pr}_2^*h^2, K X 2 − 2 c 2 ( X ) = pr 1 ∗ ( K Y 2 − 2 c 2 ( Y )) + ( m − 1 ) pr 2 ∗ h 2 ,
because 2 ch 2 ( T P k ) = ( ( k + 1 ) 2 − ( k + 1 ) k ) h 2 = ( k + 1 ) h 2 2\operatorname{ch}_2(T_{\mathbb{P}^k})=((k+1)^2-(k+1)k)h^2=(k+1)h^2 2 ch 2 ( T P k ) = (( k + 1 ) 2 − ( k + 1 ) k ) h 2 = ( k + 1 ) h 2 . Expand
H m − 2 = ∑ j ( m − 2 j ) H Y j h m − 2 − j . H^{m-2}=\sum_j\binom{m-2}{j}H_Y^j h^{m-2-j}. H m − 2 = j ∑ ( j m − 2 ) H Y j h m − 2 − j .
Only j = 0 j=0 j = 0 pairs nontrivially with a 0 0 0 -cycle pulled back from Y Y Y , and only j = 2 j=2 j = 2 with h 2 h^2 h 2 . Hence ( K X 2 − 2 c 2 ( X ) ) ⋅ H m − 2 = 3 σ + ( m − 1 ) ( m − 2 2 ) d Y (K_X^2-2c_2(X))\cdot H^{m-2}=3\sigma+(m-1)\binom{m-2}{2}d_Y ( K X 2 − 2 c 2 ( X )) ⋅ H m − 2 = 3 σ + ( m − 1 ) ( 2 m − 2 ) d Y and H m = ( m 2 ) d Y H^m=\binom{m}{2}d_Y H m = ( 2 m ) d Y . Since ( m − 1 ) ( m − 2 2 ) − ( m + 1 ) ( m 2 ) = − 3 ( m − 1 ) 2 (m-1)\binom{m-2}{2}-(m+1)\binom{m}{2}=-3(m-1)^2 ( m − 1 ) ( 2 m − 2 ) − ( m + 1 ) ( 2 m ) = − 3 ( m − 1 ) 2 , this gives I ( X , H ) I(X,H) I ( X , H ) . The condition I > 3 H m I>3H^m I > 3 H m reads σ > ( ( m − 1 ) 2 + ( m 2 ) ) d Y = ( m − 1 ) ( 3 m − 2 ) 2 d Y \sigma>((m-1)^2+\binom{m}{2})d_Y=\frac{(m-1)(3m-2)}{2}d_Y σ > (( m − 1 ) 2 + ( 2 m ) ) d Y = 2 ( m − 1 ) ( 3 m − 2 ) d Y . The divisor H H H is ample and globally generated, and R ( X , H ) R(X,H) R ( X , H ) is the Segre product by the Künneth formula. Theorem 10.3 and Proposition 10.4 (c) conclude. □
For m = 2 m=2 m = 2 the two conditions are σ > d Y \sigma>d_Y σ > d Y and σ > 2 d Y \sigma>2d_Y σ > 2 d Y . Table 5 applies them to the families of Table 2 , using that H n H_n H n is ample and globally generated (indeed very ample) for all $n\geq 2 (Proposition 4.3 and [1 Lemmas 7.1, 7.3] ).
Table 5. For X = Y n × P m − 2 X=Y_n\times\mathbb{P}^{m-2} X = Y n × P m − 2 and H = H n ⊠ O ( 1 ) H=H_n\boxtimes\mathcal{O}(1) H = H n ⊠ O ( 1 ) , with σ = σ ( Y n ) \sigma=\sigma(Y_n) σ = σ ( Y n ) and d Y = H n 2 d_Y=H_n^2 d Y = H n 2 : the smallest exponent n n n with σ / d Y > ( m − 1 ) 2 \sigma/d_Y>(m-1)^2 σ / d Y > ( m − 1 ) 2 , so that ( X , H ) (X,H) ( X , H ) has no aCM bundle (Proposition 5.9 ) / the smallest n n n with σ / d Y > ( m − 1 ) ( 3 m − 2 ) / 2 \sigma/d_Y>(m-1)(3m-2)/2 σ / d Y > ( m − 1 ) ( 3 m − 2 ) /2 , so that R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) , of dimension m + 1 m+1 m + 1 , has no small CM module (Corollary 10.9). Both criteria are only sufficient, and since σ / d Y \sigma/d_Y σ / d Y increases with n n n they then hold for all larger n n n . The column m = 2 m=2 m = 2 is Table 4 .
Table 4. The three families against the three criteria. On the quadrangle family K ∼ ( n − 2 ) H K\sim(n-2)H K ∼ ( n − 2 ) H ; on the Hesse and dual Hesse families K K K is not proportional to H H H . The sources contain the Hesse and quadrangle entries of the graded column [2 Cors. 1.4, 4.3] , the quadrangle at n = 6 n=6 n = 6 in the [9 ] column [9 Consequence 1.2] , the quadrangle at n = 7 n=7 n = 7 in the [26 ] column, and the non-proportionality of K K K and H H H on the Hesse family [2 Rem. 4.4] ; the other entries apply the criteria here.
Corollary 10.10. For every m ≥ 2 m\geq2 m ≥ 2 there is a complete Noetherian normal local domain of dimension m + 1 m+1 m + 1 , with residue field C \mathbb{C} C and an isolated singularity, that has no small Cohen–Macaulay module. For example, take R ^ ( Y n × P m − 2 , H n ⊠ O ( 1 ) ) \widehat{R}(Y_n\times\mathbb{P}^{m-2},H_n\boxtimes\mathcal{O}(1)) R ( Y n × P m − 2 , H n ⊠ O ( 1 )) for the Hesse surfaces Y n Y_n Y n , with n n n such that 3 n 2 − 11 > 7 2 ( m − 1 ) ( 3 m − 2 ) 3n^2-11>\frac{7}{2}(m-1)(3m-2) 3 n 2 − 11 > 2 7 ( m − 1 ) ( 3 m − 2 ) .
Proof. Corollary 10.9 with σ / d Y = ( 3 n 2 − 11 ) / 7 \sigma/d_Y=(3n^2-11)/7 σ / d Y = ( 3 n 2 − 11 ) /7 (Theorem 4.4).
The ball quotient Y 3 Y_3 Y 3 works here only for m = 2 m=2 m = 2 , since σ / d Y = 16 7 < 7 \sigma/d_Y=\frac{16}{7}<7 σ / d Y = 7 16 < 7 .
Unlike the rings of Corollary 10.2, these have isolated singularities. For them a small CM module amounts to a nonzero vector bundle V \mathcal{V} V on the regular punctured spectrum U U U with H i ( U , V ) = 0 H^i(U,\mathcal{V})=0 H i ( U , V ) = 0 for 1 ≤ i ≤ m − 1 1\le i\le m-1 1 ≤ i ≤ m − 1 (take M = Γ ( U , V ) M=\Gamma(U,\mathcal{V}) M = Γ ( U , V ) ). This module is finitely generated: it is the module of global sections of the reflexive hull of any coherent extension of V \mathcal{V} V to Spec R ^ \operatorname{Spec}\widehat{R} Spec R , since R ^ \widehat{R} R is normal and { m ^ } \{\widehat{\mathfrak{m}}\} { m } has codimension ≥ 2 \ge2 ≥ 2 . This is the analogue of Lemma 6.2 , using that maximal Cohen–Macaulay modules over the regular local rings of U U U are free.
Remark 10.11 (Contrast with characteristic p p p ). Let k k k be an algebraically closed field of characteristic p > 0 p>0 p > 0 , X X X a smooth connected projective k k k -variety of dimension m ≥ 2 m\ge2 m ≥ 2 and H H H an ample Cartier divisor. Then R ( X , H ) R(X,H) R ( X , H ) has a graded MCM module, and therefore its completion has a small CM module, by Lemma 2.5 (b), whose proof (faithful flatness of completion) uses neither the dimension nor the characteristic. For m = 2 m=2 m = 2 this is the theorem of Hartshorne, Hochster and Peskine–Szpiro [19 Thm. 1, Cor. 2] ; in particular, by Lemma 2.5 (b), the completion of every three-dimensional graded domain as in [19 Cor. 2] has a small CM module. For m ≥ 3 m\ge3 m ≥ 3 it follows from [32 Cor. 4.5(i)] , which characterizes the existence of a graded MCM module over R ( X , H ) R(X,H) R ( X , H ) by the existence of a coherent sheaf on X X X with MCM stalks, a role played here by O X \mathcal{O}_X O X . Alternatively, for every m ≥ 2 m\ge2 m ≥ 2 : since X X X is Cohen–Macaulay, Serre vanishing and Serre duality show that B = { s ∈ Z : H i ( X , O X ( s H ) ) ≠ 0 for some 0 < i < m } B=\{s\in\mathbb{Z}:H^i(X,\mathcal{O}_X(sH))\ne0\text{ for some }0<i<m\} B = { s ∈ Z : H i ( X , O X ( sH )) = 0 for some 0 < i < m } is finite, say B ⊂ [ − c , c ] B\subset[-c,c] B ⊂ [ − c , c ] . Let F F F be the absolute Frobenius of X X X , choose e e e with p e > 2 c + 1 p^e>2c+1 p e > 2 c + 1 , and put G = F ∗ e O X ( ( c + 1 ) H ) \mathcal{G}=F^e_*\mathcal{O}_X((c+1)H) G = F ∗ e O X (( c + 1 ) H ) , a coherent sheaf since k k k is perfect. As F e ∗ O X ( H ) = O X ( p e H ) F^{e*}\mathcal{O}_X(H)=\mathcal{O}_X(p^eH) F e ∗ O X ( H ) = O X ( p e H ) , the projection formula gives G ( t ) = F ∗ e O X ( ( c + 1 + p e t ) H ) \mathcal{G}(t)=F^e_*\mathcal{O}_X((c+1+p^et)H) G ( t ) = F ∗ e O X (( c + 1 + p e t ) H ) , and c + 1 + p e t ∉ [ − c , c ] c+1+p^et\notin[-c,c] c + 1 + p e t ∈ / [ − c , c ] for every t ∈ Z t\in\mathbb{Z} t ∈ Z . Hence H i ( X , G ( t ) ) = 0 H^i(X,\mathcal{G}(t))=0 H i ( X , G ( t )) = 0 for 0 < i < m 0<i<m 0 < i < m and all t t t . Put q = p e q=p^e q = p e . Since c + 1 + q t < 0 c+1+qt<0 c + 1 + q t < 0 for t < 0 t<0 t < 0 ,
Γ ∗ ( G ) = ⨁ t ≥ 0 H 0 ( X , O X ( ( c + 1 + q t ) H ) ) . \Gamma_*(\mathcal{G})=\bigoplus_{t\ge0}H^0(X,\mathcal{O}_X((c+1+qt)H)). Γ ∗ ( G ) = t ≥ 0 ⨁ H 0 ( X , O X (( c + 1 + q t ) H )) .
where r ∈ R ( X , H ) ℓ r\in R(X,H)_\ell r ∈ R ( X , H ) ℓ acts by s ↦ r ℓ s s\mapsto r^\ell s s ↦ r ℓ s , of degree t ↦ t + ℓ t\mapsto t+\ell t ↦ t + ℓ (projection formula). Thus Γ ∗ ( G ) \Gamma_*(\mathcal{G}) Γ ∗ ( G ) is the part of F ∗ e R ( X , H ) F^e_*R(X,H) F ∗ e R ( X , H ) in degrees ≡ c + 1 ( m o d q ) \equiv c+1\pmod q ≡ c + 1 ( mod q ) , regraded by t t t , hence finitely generated as k k k is perfect. It is a nonzero graded MCM R ( X , H ) R(X,H) R ( X , H ) -module: H m 0 H^0_m H m 0 and H m m H^m_m H m m vanish as for any module of sections, and H m i + 1 ( Γ ∗ ( G ) ) ≅ ⨁ t ∈ Z H i ( X , G ( t ) ) = 0 H^{i+1}_m(\Gamma_*(\mathcal{G}))\cong\bigoplus_{t\in\mathbb{Z}}H^i(X,\mathcal{G}(t))=0 H m i + 1 ( Γ ∗ ( G )) ≅ ⨁ t ∈ Z H i ( X , G ( t )) = 0 for 1 ≤ i ≤ m − 1 1\le i\le m-1 1 ≤ i ≤ m − 1 .
Over C \mathbb{C} C both statements fail, even after completion: for m = 2 m=2 m = 2 by Corollary 9.3 , and for every m ≥ 2 m\ge2 m ≥ 2 by Corollary 10.9, since Y n × P m − 2 Y_n\times\mathbb{P}^{m-2} Y n × P m − 2 is smooth. Thus [2 ] shows that in characteristic zero Hochster’s conjecture has no graded refinement, and Corollary 10.10 shows that the local consequence of the graded theorem of characteristic p p p fails as well, in every dimension. Likewise, the characteristic hypothesis of [32 Cor. 4.5(i)] is essential not only for its graded statement, as shown in [2 Cor. 5.4] with the threefolds Y n × P 1 Y_n\times\mathbb{P}^1 Y n × P 1 , n ≥ 4 n\ge4 n ≥ 4 , but also for its local consequence, with Y n × P 1 Y_n\times\mathbb{P}^1 Y n × P 1 , n ≥ 5 n\ge5 n ≥ 5 . The case n = 4 n=4 n = 4 lies in the gap of item 5 of § 12 . As in [2 ] , the step that has no analogue in characteristic p p p is Bogomolov’s inequality, used in Proposition 10.6 .
Provenance and status
Figure 2 in the introduction records the citations. [24 ] starts from the examples of [1 ] ; [2 ] extends Lopez’s inequality to aCM bundles; [9 ] takes its polarized surface from [2 Rem. 4.2, Cor. 4.3] and identifies its hypothesis c > 0 c > 0 c > 0 with the graded criterion d < σ d < \sigma d < σ of [2 ] (see [9 Rem. 1.3] ), while its own argument does not use the graded theorem. [26 ] attributes the surface inequality without Picard-rank hypothesis to [1 Prop. 2.1, Rem. 2.3] , takes the complete-quadrangle construction and polarization from [1 §7.2] , compares with the graded results of [2 ] , and acknowledges that its existential claim overlaps with [9 ] .
Status. To the author’s knowledge, none of the five texts has yet appeared in a refereed publication. [1 ] , [24 ] and [2 ] present their projective and graded statements with complete proofs. [9 ] presents its Main claim as a proposed argument circulated for verification, and states its consequence for Hochster’s conjecture conditionally; § 7 states its main steps without claiming independent verification, except that Proposition 7.1 follows from Lemma 5.5 (Remark 9.2). [26 ] claims a complete proof. Its argument is short — § 2 –§ 5 of the manuscript, summarized in § 6 and § 8 above — and uses only standard tools. § 10 gives a complete proof of its main theorem, in every dimension. § 9 and § 10 represent the new part of this survey.
Open questions
The sharp local obstruction. Does the existence of a small CM module over R ^ ( Y , H ) \widehat{R}(Y,H) R ( Y , H ) force H 2 ≥ σ ( Y ) H^2 \ge\sigma(Y) H 2 ≥ σ ( Y ) , as in the graded case? The argument of Lemma 6.4 (b), a twist followed by modifications, places the normalized Harder–Narasimhan slopes of the restriction P P P of a reflexive extension (Lemma 6.2 (a)) in [ λ 2 − 1 2 , λ 2 + 1 2 ) \left[\frac{\lambda}{2}-\frac{1}{2},\frac{\lambda}{2}+\frac{1}{2}\right) [ 2 λ − 2 1 , 2 λ + 2 1 ) , λ = K Y ⋅ H / H 2 \lambda= K_Y \cdot H/H^2 λ = K Y ⋅ H / H 2 , for every ( Y , H ) (Y,H) ( Y , H ) . By Remark 9.2, it would then suffice that χ ( P ) ≥ 0 \chi(P) \ge0 χ ( P ) ≥ 0 for such an extension; this is the missing input, and [9 ] claims it when K ≡ 4 H K \equiv4H K ≡ 4 H . The quadrangle ring with n = 6 n = 6 n = 6 lies in the gap d < σ < 2 d d < \sigma< 2d d < σ < 2 d .
Beauville’s refinement. Is there a smooth surface S S S with ρ ( S ) = 1 \rho(S) = 1 ρ ( S ) = 1 and a very ample H H H with H 2 < σ ( S ) H^2 < \sigma(S) H 2 < σ ( S ) ?
Thresholds. The Veronese subrings R ( Y n , H n ) ( ℓ ) R(Y_n,H_n)^{(\ell)} R ( Y n , H n ) ( ℓ ) have no graded MCM module for 7 ℓ 2 < 3 n 2 − 11 7\ell^2 < 3n^2 - 11 7 ℓ 2 < 3 n 2 − 11 and many for ℓ ≥ ℓ 0 \ell\ge\ell_0 ℓ ≥ ℓ 0 [2 Rem. 6.1] . What happens in between, graded and locally?
The Boij–Söderberg cone: the second question of Eisenbud and Schreyer. Eisenbud and Schreyer asked for the Boij–Söderberg cone of cohomology tables of aCM bundles on the surfaces of [1 ] (see [2 Acknowledgements] ). For H 2 < σ ( Y ) H^2 < \sigma(Y) H 2 < σ ( Y ) there are no aCM bundles (Theorem 5.1), so this cone is { 0 } \{0\} { 0 } , and their question becomes one about all vector bundles. By [30 Thm. 5.3] , the cone of nonnegative rational combinations of cohomology tables of coherent sheaves on X ⊂ P N X \subset\mathbb{P}^N X ⊂ P N coincides with that of P dim X \mathbb{P}^{\dim X} P d i m X if and only if X X X carries an Ulrich sheaf; on P 2 \mathbb{P}^2 P 2 the tables of vector bundles are positive combinations of tables of supernatural bundles [11 ] . For H 2 < σ ( Y ) H^2 < \sigma(Y) H 2 < σ ( Y ) , the cone C v b ( Y , H ) C_{\mathrm{vb}}(Y,H) C vb ( Y , H ) of nonnegative rational combinations of the tables of vector bundles on ( Y , H ) (Y,H) ( Y , H ) contains no nonzero table with vanishing h 1 h^1 h 1 -row [2 Rem. 6.2] ; in particular it misses the supernatural rays of P 2 \mathbb{P}^2 P 2 with consecutive roots, those of the line bundles O ( a ) \mathcal{O}(a) O ( a ) . What is C v b ( Y , H ) C_{\mathrm{vb}}(Y,H) C vb ( Y , H ) ? For a coherent sheaf F F F of rank N > 0 N > 0 N > 0 on P 2 \mathbb{P}^2 P 2 with b = c 1 ( F ) / N b = c_1(F)/N b = c 1 ( F ) / N , Remark 9.2 applied to F ( t ) F(t) F ( t ) shows that the roots of χ ( F ( t ) ) \chi(F(t)) χ ( F ( t )) are − b − 3 2 ± 1 2 1 − 4 ν ( F ) -b - \frac{3}{2} \pm\frac{1}{2}\sqrt{1 - 4\nu(F)} − b − 2 3 ± 2 1 1 − 4 ν ( F ) ; they are real, at distance 1 − 4 ν ( F ) \sqrt{1 - 4\nu(F)} 1 − 4 ν ( F ) , exactly when ν ( F ) ≤ 1 4 \nu(F) \le\frac{1}{4} ν ( F ) ≤ 4 1 . For a μ H \mu_H μ H -semistable bundle E E E on Y Y Y , Proposition 9.1 gives ν ( f ∗ E ) ≤ 1 4 − σ 4 d \nu(f_*E) \le\frac{1}{4} - \frac{\sigma}{4d} ν ( f ∗ E ) ≤ 4 1 − 4 d σ , so the Hilbert polynomial of $E
has real roots at distance at least σ / d > 1 \sqrt{\sigma/d}>1 σ / d > 1 . Is C v b ( Y , H ) C_{\mathrm{vb}}(Y,H) C vb ( Y , H ) cut out from the cone of P 2 \mathbb{P}^{2} P 2 by conditions of this kind, once unstable bundles (the term v b v_{\mathrm{b}} v b ) are taken into account?
For ( Y , ℓ H ) (Y,\ell H) ( Y , ℓ H ) the bound σ / ( ℓ 2 d ) \sqrt{\sigma/(\ell^{2}d)} σ / ( ℓ 2 d ) drops below 1 exactly when ℓ 2 H 2 > σ \ell^{2}H^{2}>\sigma ℓ 2 H 2 > σ , consistent with the existence of Ulrich bundles for ℓ ≫ 0 \ell\gg0 ℓ ≫ 0 [CoHu]; how does the cone fill up as ℓ \ell ℓ grows?
Higher dimension. Theorem 10.3 gives I ( X , H ) ≤ 3 H m I(X,H)\leq3H^{m} I ( X , H ) ≤ 3 H m for small CM modules over R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) , while aCM bundles require I ( X , H ) ≤ 0 I(X,H)\leq0 I ( X , H ) ≤ 0 . For the products of Corollary 10.9 this leaves the range ( m − 1 ) 2 < σ ( Y ) / H Y 2 ≤ ( m − 1 ) ( 3 m − 2 ) / 2 (m-1)^{2}<\sigma(Y)/H_{Y}^{2}\leq(m-1)(3m-2)/2 ( m − 1 ) 2 < σ ( Y ) / H Y 2 ≤ ( m − 1 ) ( 3 m − 2 ) /2 open, e.g. the Hesse pairs with ( m , n ) = ( 3 , 4 ) (m,n)=(3,4) ( m , n ) = ( 3 , 4 ) . Is I ( X , H ) ≤ 0 I(X,H)\leq0 I ( X , H ) ≤ 0 necessary locally, as item 1 asks for surfaces? Does Chen’s argument extend beyond surfaces?
Mixed characteristic. After spreading out, the reductions modulo primes of good reduction of the rings above have graded MCM modules, hence their completions have small CM modules (Remark 10.11); Bogomolov’s inequality is the step that fails. Are there analogous counterexamples in mixed characteristic, where the conjecture was originally motivated by Serre’s positivity of intersection multiplicities?
A characterization. For which smooth projective surfaces Y Y Y and ample, globally generated H H H does R ^ ( Y , H ) \widehat{R}(Y,H) R ( Y , H ) admit a small CM module? By Theorem 8.2, σ ≤ 2 H 2 \sigma\leq2H^{2} σ ≤ 2 H 2 is necessary, and by Coskun–Huizenga existence holds for ( Y , ℓ H ) (Y,\ell H) ( Y , ℓ H ) with ℓ ≫ 0 \ell\gg0 ℓ ≫ 0 .
Non-graded examples. Are there complete local domains without small CM modules that are not completions of graded rings, in particular not of the form R ^ ( X , H ) \widehat{R}(X,H) R ( X , H ) ?
Statements and declarations
Use of generative AI. ChatGPT and Claude were used for exploratory computations, literature search and editorial assistance, including drafting of the proofs in Sections 9 and 10. The author is solely responsible for the final text.
Simon Stoilow Institute of Mathematics of the Romanian Academy, 21 Calea Griviței, 010702 Bucharest, Romania
Email address: cristian.anghel@imar.ro