ScalingStacks

6. Adapted volume forms [02F5]

Original official author HTML, exact retained edition. Historical TeX conversion verdicts remain unchanged. Cited-edition alignment and mathematical self-containment are not assessed.

Complete original source context · Original author HTML

6. Adapted volume forms

6.1. Monge-Ampère equations on normal Kähler spaces

Let Ω\Omega be a smooth Kähler metric on VV. A classical result of P. Lelong states that if UU is relatively compact in VV, then Ur​e​gU^{reg} is of finite volume with respect to the smooth volume form Ωr​e​gn\Omega_{reg}^{n}.

This has been generalized by E.Bedford and A.Taylor in [BT], where the authors study Monge-Ampère measures for locally bounded psh functions. Since these measures do not charge proper analytic subsets, we obtain:

Proposition 6.1.

Let Ω\Omega be a semi-Kähler current with Ll​o​c∞L_{loc}^{\infty} potentials on VV. The Monge-Ampère measure Ωr​e​gn\Omega_{reg}^{n} is well defined on Vr​e​gV_{reg} and satisfies ∫Ur​e​gΩr​e​gn<∞, for all relatively compact subset ​U⊂V.\int_{U^{reg}}\Omega_{reg}^{n}<\infty,\text{ for all relatively compact subset }U\subset V.

For any resolution π:X→V\pi:X\to V, the Monge-Ampère measure (π∗​Ω)n(\pi^{*}\Omega)^{n} is well defined on XX and satisfies π∗​(π∗​Ω)n=j∗​Ωr​e​gn\pi_{*}(\pi^{*}\Omega)^{n}=j_{*}\Omega_{reg}^{n}. Moreover if π¯:X¯→V\bar{\pi}:\bar{X}\to V is a resolution dominating π\pi (i.e. π¯=π∘ψ\bar{\pi}=\pi\circ\psi for some bimeromorphic proper holomorphic map ψ:X¯′→X′\psi:\bar{X}^{\prime}\to X^{\prime}), then ψ∗​(π¯∗​Ω)n=(π∗​Ω)n\psi_{*}(\bar{\pi}^{*}\Omega)^{n}=(\pi^{*}\Omega)^{n}.

The measure π∗​(π∗​Ω)n\pi_{*}(\pi^{*}\Omega)^{n} is thus well defined on VV and independent of the choice of resolution. We will call it the Monge-Ampère measure of Ω\Omega and denote it by Ωn\Omega^{n}. The mass of this measure only depends on the cohomology class of Ω\Omega, as follows again from [BT]:

Lemma 6.2.

Assume VV is compact. Let Ω1,Ω2\Omega_{1},\Omega_{2} two semi Kähler currents with Ll​o​c∞L_{loc}^{\infty} potentiel on VV. If they are cohomologous, i.e. Ω1=Ω2+d​dc​φ\Omega_{1}=\Omega_{2}+dd^{c}\varphi for some φ∈L∞​(X)\varphi\in L^{\infty}(X), then ∫VΩ1n=∫VΩ2n\int_{V}\Omega_{1}^{n}=\int_{V}\Omega_{2}^{n}.

We can now reformulate some of our previous results.

Theorem 6.3.

Let VV be a nn-dimensional compact normal Kähler space and Ω\Omega be a smooth Kähler form on VV. Then for every f∈Lp​(V,Ωn)f\in L^{p}(V,\Omega^{n}), p>1p>1, such that ∫Vf​Ωn=∫XΩn\int_{V}f\Omega^{n}=\int_{X}\Omega^{n}, there is a unique φ∈𝒞0​(V)\varphi\in{\mathcal{C}}^{0}(V) such that

(Ω+d​dc​φ)n=f​Ωn​ and ​supVφ=−1.(\Omega+dd^{c}\varphi)^{n}=f\Omega^{n}\text{ and }\sup_{V}\varphi=-1.
Proof.

Let π:X→V\pi:X\to V be a resolution of VV. We may define a semipositive big smooth form on XX by ω=π∗​Ω\omega=\pi^{*}\Omega. By Theorem 2.1 and Proposition 3.1 we can solve uniquely (ω+d​dc​φ¯)n=f∘π​ωn(\omega+dd^{c}\bar{\varphi})^{n}=f\circ\pi\omega^{n} where φ¯\bar{\varphi} is a continuous function on XX such that ω+d​dc​φ¯\omega+dd^{c}\bar{\varphi} is semipositive. Let FF be a fiber of π\pi and i:F→Xi:F\to X the inclusion map. FF is connected by Zariski’s main theorem. Furthermore i∗​ω+d​dc​i∗​φ¯i^{*}\omega+dd^{c}i^{*}\bar{\varphi} is semipositive on FF. Since i∗​ω=0i^{*}\omega=0, it follows that i∗​φi^{*}\varphi is a continuous psh function on FF. Hence i∗​φ¯i^{*}\bar{\varphi} is constant. This implies that φ¯=φ∘π\bar{\varphi}=\varphi\circ\pi where φ\varphi is a continuous function on VV. We do have (Ω+d​dc​φ)n=f​Ωn(\Omega+dd^{c}\varphi)^{n}=f\Omega^{n}. ∎

6.2. Adapted measures on log terminal Kähler spaces

Let VV be a nn-dimensional Gorenstein Kähler space and Ω\Omega be a smooth Kähler form on VV. Fix x∈Vx\in V and let α\alpha be a local generator of ωV\omega_{V} defined over an open subset x∈Ux\in U; then v=cn​α∧α¯v=c_{n}\alpha\wedge\bar{\alpha} is a positive definite volume form on Ur​e​gU^{reg}, for an appropriate choice of the constant cn=−1n​(−1)n⁡(n+1)2c_{n}=\sqrt{-1}^{n}(-1)^{\frac{n(n+1)}{2}}.

When VV is merely ℚ\mathbb{Q}-Gorenstein of finite index NN, we choose β\beta a local generator of ωV[N]\omega_{V}^{[N]} defined over an open subset x∈Ux\in U and we set

v=vβ=(−1N​n​(−1)N​n⁡(n+1)2​β∧β¯)1N.v=v_{\beta}=\left(\sqrt{-1}^{Nn}(-1)^{N\frac{n(n+1)}{2}}\beta\wedge\bar{\beta}\right)^{\frac{1}{N}}.

This is a positive definite volume form on Ur​e​gU^{reg}.

Our next observation is that log terminal singularities are the worst singularities we can allow in order to globally solve Monge-Ampère equations associated to volume forms on VV.

Lemma 6.4.

For every U1⊂⊂UU_{1}\subset\subset U, ∫U1r​e​gv<∞\int_{U_{1}^{reg}}v<\infty iff XX is log terminal.

If VV is log terminal, then the Radon measure μ=j∗​v\mu=j_{*}v satisfies μ=f​Ωn\mu=f\Omega^{n} with f∈L1+ε​(U1,Ωn)f\in L^{1+\varepsilon}(U_{1},\Omega^{n}) for some ε>0\varepsilon>0.

Proof.

Let π:X→V\pi:X\to V be a log resolution. Write KX≅π∗​KV+∑aE​EK_{X}\cong\pi^{*}K_{V}+\displaystyle\sum a_{E}E. Since e​x​c​(π)exc(\pi) has simple normal crossings, at every P∈E=e​x​c​(π)P\in E=exc(\pi) there are local coordinates (zi)i=1,…,n(z^{i})_{i=1,...,n} such that EE is described by the equation z1​…​zq=0z^{1}\ldots z^{q}=0. Let EjE_{j} be the divisor zj=0z_{j}=0. We have: π∗​v=∏j=1q|zj|2​aEj​d​λ\pi^{*}v=\prod_{j=1}^{q}|z^{j}|^{2a_{E_{j}}}d\lambda where d​λd\lambda is a Lebesgue measure on XX, hence the measure π∗​v\pi^{*}v has finite mass near PP iff ∀j,aEj>−1\forall j,a_{E_{j}}>-1. Thus ∫U1r​e​gv<∞\int_{U_{1}^{reg}}v<\infty iff ∀E,aE>−1\forall E,\ a_{E}>-1.

Let f1f_{1} be the density of π∗​v\pi^{*}v with respect to d​λd\lambda. Since f1f_{1} is comparable to ∏j=1q|zj|2​aEj\prod_{j=1}^{q}|z^{j}|^{2a_{E_{j}}} near PP, it follows that f1f_{1} belongs actually to Lp​(X,d​λ)L^{p}(X,d\lambda) for some p>1p>1 when XX is log terminal.

Let D=1/fD=1/f be the density of Ωn\Omega^{n} with respect to vv. We will see here below that DD is bounded but it might have zeroes on EE, hence ff is unbounded in general. However we will show that f∘π∈Lα​(X,d​λ)f\circ\pi\in L^{\alpha}(X,d\lambda) for α>0\alpha>0 small enough, hence it follows from Hölder’s inequality (as in the proof of lemma 3.2) that

∫U1r​e​gf1+ε​Ωn=∫π−1​U1r​e​gfε​f1​𝑑λ<+∞\int_{U_{1}^{reg}}f^{1+\varepsilon}\Omega^{n}=\int_{\pi^{-1}U_{1}^{reg}}f^{\varepsilon}f_{1}d\lambda<+\infty

if ε>0\varepsilon>0 is small enough.

Fix x∈Vx\in V and let i:Ux→ℂmi:U_{x}\to\mathbb{C}^{m} be a local embedding of a neighborhood UxU_{x} of xx. We consider the (mn)\left(\begin{array}[]{c}m\\ n\end{array}\right) nn-forms on Uxr​e​gU_{x}^{reg} d​uI=d​ui1∧…​d​uindu^{I}=du^{i_{1}}\wedge\ldots du^{i_{n}}, where (ui)(u^{i}) is a set of affine coordinates on ℂm\mathbb{C}^{m}. Observe that Ωn\Omega^{n} is comparable to ∑Ivd​uI\sum_{I}v_{du^{I}} 1111 11 Note that the formula for vβv_{\beta} makes sense even if β\beta is not a local generator.. Since β\beta is a local generator at xx of ωV[N]\omega_{V}^{[N]}, we have (d​uI)N=fI​β(du^{I})^{N}=f_{I}\beta where fI∈𝒪V,xf_{I}\in\mathcal{O}_{V,x} is the germ of an holomorphic function at xx. Therefore Ωn\Omega^{n} is comparable to ∑I|fI|2N​v\sum_{I}|f_{I}|^{\frac{2}{N}}v, hence DD is comparable to [∑I|fI|2N]−1[\sum_{I}|f_{I}|^{\frac{2}{N}}]^{-1} near xx.

The functions (fI)(f_{I}) generate an ideal ℐx⊂𝒪V,x\mathcal{I}_{x}\subset\mathcal{O}_{V,x}. Actually, the construction can be globalized to provide a coherent ideal sheaf ℐ⊂𝒪V\mathcal{I}\subset\mathcal{O}_{V} cosupported on Vs​i​n​gV^{sing}.

We may assume [Hi] that π:X→V\pi:X\to V is a log resolution of (V,ℐ)(V,\mathcal{I}), namely a log resolution of VV with the additional property that the ideal sheaf π−1​ℐ.𝒪X\pi^{-1}\mathcal{I}.\mathcal{O}_{X} which is the ideal sheaf of 𝒪X\mathcal{O}_{X} generated by the family of holomorphic functions (π∗​fI)I(\pi^{*}f_{I})_{I}, satisfies π−1ℐ.𝒪X=𝒪X(−∑NbEE)⊂𝒪X\pi^{-1}\mathcal{I}.\mathcal{O}_{X}=\mathcal{O}_{X}(-\sum Nb_{E}E)\subset\mathcal{O}_{X} where N.bE∈ℕN.b_{E}\in\mathbb{N} is a positive multiplicity attached to any exceptional divisor of π\pi.

In local coordinates near P∈XP\in X, π∗​D\pi^{*}D is comparable to ∏j|zj|2​bEj\prod_{j}|z_{j}|^{2b_{E_{j}}}, hence π∗​(f1​D−ε)​ is comparable to ​∏j|zj|2​(aEj−ε​bEj).\pi^{*}(f_{1}D^{-\varepsilon})\text{ is comparable to }\prod_{j}|z_{j}|^{2(a_{E_{j}}-\varepsilon b_{E_{j}})}. It follows that for every relatively compact subset U1⊂⊂U,f∈L1+ε​(U1,Ωn)U_{1}\subset\subset U,\ \ f\in L^{1+\varepsilon}(U_{1},\Omega^{n}) iff ∀E,π⁡(E)∩U≠∅⇒aE−ε​bE>−1\forall E,\ \ \pi(E)\cap U\not=\emptyset\Rightarrow a_{E}-\varepsilon b_{E}>-1. ∎

Definition 6.5.

Assume VV has only log terminal singularities. A positive definite adapted measure on VV is a positive Radon measure locally of the form ef.ve^{f}.v where ff is a bounded measurable function. A positive definite adapted measure has 𝒞0{\mathcal{C}}^{0}, 𝒞α{\mathcal{C}}^{\alpha}, 𝒞∞{\mathcal{C}}^{\infty} density if so is ff.

Remark 6.6.

It follows from lemma 6.4 that if VV is ℚ\mathbb{Q}-Gorenstein but has non log terminal singularities, vv is a volume form on Vr​e​gV^{reg} but does not extend to a measure on VV.

6.3. Adapted volume forms for klt Kähler pairs

We will be briefer since pairs are mainly of interest to MMP practitioners. The key definition for us will be:

Definition 6.7.

A pair (V,Δ)(V,\Delta) is klt iff KV+ΔK_{V}+\Delta is ℚ\mathbb{Q}-Cartier and if for any log-resolution π:X→V\pi:X\to V of (V,Δ)(V,\Delta), we have the numerical equivalence of Cartier divisors:

N⁡(KX+Δ′)≅π∗​N​(KV+Δ)+∑E​e​x​c.N​aE​EN(K_{X}+\Delta^{\prime})\cong\pi^{*}N(K_{V}+\Delta)+\sum_{E\ exc.}Na_{E}E

with aE>−1a_{E}>-1, Δ′\Delta^{\prime} the proper transform of Δ\Delta in XX (same multiplicities) and NN is an integer such that N⁡(KV+Δ)N(K_{V}+\Delta) is Cartier.

Thus a variety VV has only klt singularities iff (X,∅)(X,\emptyset) is klt.

Let β\beta be a local generator of 𝒪V​(N⁡(KV+Δ))\mathcal{O}_{V}(N(K_{V}+\Delta)). Then βVr​e​g\beta_{V^{reg}} can be viewed as a meromorphic N-canonical form with a pole of order N​diNd_{i} on EiE_{i} where Δ=∑idi​Ei\Delta=\sum_{i}d_{i}E_{i} is the decomposition of Δ\Delta into prime divisors. Thus vβVr​e​gv_{\beta_{V^{reg}}} defines a volume form with poles on Vr​e​gV^{reg}, namely vβVr​e​gv_{\beta_{V^{reg}}} is comparable to ∏i|si|−2​di​d​λ\prod_{i}|s_{i}|^{-2d_{i}}d\lambda, where sis_{i} denotes the canonical section of 𝒪⁡(Ei)\mathcal{O}(E_{i}). If vβVr​e​gv_{\beta_{V^{reg}}} is a finite measure then di<1d_{i}<1, but the converse is not true. We have the following staightforward extension of lemma 6.4:

Lemma 6.8.

Let j′:V−∪iEi→Vj^{\prime}:V-\cup_{i}E_{i}\to V be the canonical inclusion. j∗′​vβj^{\prime}_{*}v_{\beta} is a well defined Radon measure on VV iff (X,Δ)(X,\Delta) is klt.

The definition of an adapted measure for a klt pair is left to the reader.

Original mathematics by the credited authors. Source-backed reader collection; mathematical self-containment is not assessed.