ScalingStacks

6. Appendix: Regularization of qpsh functions [034I]

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. Appendix: Regularization of qpsh functions

It is well-known that every psh function φ\varphi can be locally regularized, i.e. one can find locally a sequence φj\varphi_{j} of smooth psh functions which decrease towards φ\varphi (see e.g. [15], chapter 1). Similarly one can always locally regularize ω\omega-psh functions. It is interesting to know whether one can also globally regularize ω\omega-psh functions.

When XX is a complex homogeneous manifold (i.e. when A​u​t​(X)Aut(X) acts transitively on XX), it is possible to approximate any ω\omega-psh function by a decreasing sequence of smooth ω\omega-psh functions (see [21], [25]). In general however there is a loss of positivity: it will be possible to approximate φ∈P​S​H​(X,ω)\varphi\in PSH(X,\omega) by a decreasing sequence of smooth functions φj\varphi_{j} but the curvature forms d​dc​φjdd^{c}\varphi_{j} will have to be more negative than −ω-\omega. How negative depends on the positivity of the cohomology class [ω][\omega].

Consider e.g. π:X→ℙ2\pi:X\rightarrow\mathbb{P}^{2} the blow up of ℙ2\mathbb{P}^{2} at point pp, E=π−1​(p)E=\pi^{-1}(p) the exceptional divisor and let ω=[E]\omega=[E] be the current of integration along EE. Then P​S​H​(X,ω)≃ℝPSH(X,\omega)\simeq\mathbb{R} (see Remark 1.5) so every psh function has logarithmic singularities along EE, hence is not smooth. Alternatively EE has self-intersection −1-1 so its cohomology class cannot be represented by smooth non-negative forms, not even by smooth forms with (very) small negativity.

Following Demailly’s fundamental work [10], [12], [16] (to cite a few) we show herebelow that regularization with no loss of positivity is possible when ω\omega is a Hodge form (i.e. a Kähler form with integer class). This yields a ”simple” regularization process when XX is projective. We would like to mention that Demailly has produced over the last twenty years much finer regularization results. We nevertheless think it is worth including a proof, since it is far less technical than Demailly’s more general results (although our proof heavily relies on his ideas). We thank P.Eyssidieux for his helpful contribution regarding that matter.

Theorem 6.1.

Let L→XL\rightarrow X be a positive holomorphic line bundle equipped with a smooth strictly positive metric hh, and set ω:=Θh>0\omega:=\Theta_{h}>0.

Then for every φ∈P​S​H​(X,ω)\varphi\in PSH(X,\omega), there exists a sequence φj∈P​S​H​(X,ω)∩𝒞∞​(X)\varphi_{j}\in PSH(X,\omega)\cap{\mathcal{C}}^{\infty}(X) such that φj\varphi_{j} decreases towards φ\varphi.

Proof.

Let φ∈P​S​H​(X,ω)\varphi\in PSH(X,\omega). We can assume w.l.o.g. that φ≤0\varphi\leq 0 on XX. Let ψ={ψα:=φ+hα∈PSH(𝒰α)}\psi=\{\psi_{\alpha}:=\varphi+h_{\alpha}\in PSH({\mathcal{U}}_{\alpha})\} denote the associated (singular) positive metric of LL on XX, where {𝒰α}\{{\mathcal{U}}_{\alpha}\} denotes an open cover of XX trivializing LL (see section 4).

Step 1. We consider the following Bergman spaces

ℋj,j0:={s∈Γ(X,Lj)/∫X|s|2e−2​hj,j0dVω<+∞},{\mathcal{H}}_{j,j_{0}}:=\left\{s\in\Gamma(X,L^{j})\,/\,\int_{X}|s|^{2}e^{-2h_{j,j_{0}}}dV_{\omega}<+\infty\right\},

where hj,j0=(j−j0)​ψ+j0​hh_{j,j_{0}}=(j-j_{0})\psi+j_{0}h, j0j_{0} a fixed large integer (to be specified later). Let σ1(j,j0),…,σsj(j,j0)\sigma_{1}^{(j,j_{0})},\ldots,\sigma_{s_{j}}^{(j,j_{0})} be an orthonormal basis of ℋj,j0{\mathcal{H}}_{j,j_{0}} and set

ψj,j0:=12​j​log⁡[∑l=1sj|σl(j,j0)|2]=12​j​sups∈Bj,j0log⁡|s|2,\psi_{j,j_{0}}:=\frac{1}{2j}\log\left[\sum_{l=1}^{s_{j}}|\sigma_{l}^{(j,j_{0})}|^{2}\right]=\frac{1}{2j}\sup_{s\in B_{j,j_{0}}}\log|s|^{2},

where Bj,j0B_{j,j_{0}} denotes the unit ball of radius 1 centered at 0 in ℋj,j0{\mathcal{H}}_{j,j_{0}}. Clearly ψj,j0\psi_{j,j_{0}} defines a positive (singular) metric of LL on XX, equivalently φj,j0:=ψj,j0−h∈P​S​H​(X,ω)\varphi_{j,j_{0}}:=\psi_{j,j_{0}}-h\in PSH(X,\omega). If x∈𝒰αx\in{\mathcal{U}}_{\alpha} and s={sα}∈ℋj,j0s=\{s_{\alpha}\}\in{\mathcal{H}}_{j,j_{0}}, then |sα|2|s_{\alpha}|^{2} is subharmonic in 𝒰α{\mathcal{U}}_{\alpha} hence

|sα​(x)|2≤C1r2​n​∫B⁡(x,r)|sα​(x)|2≤C2r2​n​e2​supB⁡(x,r)hj,j0​∫X|s|2​e−2​hj,j0​d​Vω,|s_{\alpha}(x)|^{2}\leq\frac{C_{1}}{r^{2n}}\int_{B(x,r)}|s_{\alpha}(x)|^{2}\leq\frac{C_{2}}{r^{2n}}e^{2\sup_{B(x,r)}h_{j,j_{0}}}\int_{X}|s|^{2}e^{-2h_{j,j_{0}}}dV_{\omega},

where r>0r>0 is so small that B⁡(x,r)⊂𝒰αB(x,r)\subset{\mathcal{U}}_{\alpha}. We infer

(3) φj,j0​(x)≤(1−j0/j)​supB⁡(x,r)φ+C3−n​log⁡rj.\varphi_{j,j_{0}}(x)\leq(1-j_{0}/j)\sup_{B(x,r)}\varphi+\frac{C_{3}-n\log r}{j}.

There is also a reverse inequality which uses a deep extension result of Ohsawa-Takegoshi-Manivel (see [14]): there exists j0∈ℕj_{0}\in\mathbb{N} and C4>0C_{4}>0 large enough so that ∀x∈X,∀j∈ℕ\forall x\in X,\forall j\in\mathbb{N}, there exists s∈Γ⁡(X,Lj)s\in\Gamma(X,L^{j}) with

∫X|s|2​e−2​hj,j0​d​Vω≤C4​|s⁡(x)|2​e−2​hj,j0​(x).\int_{X}|s|^{2}e^{-2h_{j,j_{0}}}dV_{\omega}\leq C_{4}|s(x)|^{2}e^{-2h_{j,j_{0}}(x)}.

Choose ss so that the right hand side is equal to 11, hence s∈Bj,j0s\in B_{j,j_{0}}. Then

ψj,j0​(x)≥12​j​log⁡|s⁡(x)|2=(1−j0j)​ψ​(x)+jj0​h​(x)−log⁡C42​j.\psi_{j,j_{0}}(x)\geq\frac{1}{2j}\log|s(x)|^{2}=\left(1-\frac{j_{0}}{j}\right)\psi(x)+\frac{j}{j_{0}}h(x)-\frac{\log C_{4}}{2j}.

We infer

(4) φj,j0​(x)≥(1−j0j)​φ​(x)−log⁡C42​j≥φ⁡(x)−log⁡C42​j\varphi_{j,j_{0}}(x)\geq\left(1-\frac{j_{0}}{j}\right)\varphi(x)-\frac{\log C_{4}}{2j}\geq\varphi(x)-\frac{\log C_{4}}{2j}

since φ≤0\varphi\leq 0 on XX. It follows from (3) and (4) that φj→φ\varphi_{j}\rightarrow\varphi in L1​(X)L^{1}(X).

Step 2. We now show, following [16] that (φj,j0)j(\varphi_{j,j_{0}})_{j} is almost subadditive. Let s∈Γ⁡(X,Lj1+j2)s\in\Gamma(X,L^{j_{1}+j_{2}}) with

∫X|s|2​e−2​hj1+j2,j0​d​Vω≤1.\int_{X}|s|^{2}e^{-2h_{j_{1}+j_{2},j_{0}}}dV_{\omega}\leq 1.

We may view ss as the restriction to the diagonal Δ\Delta of X×XX\times X of a section S∈Γ⁡(X×X,L1j1⊗L2j2)S\in\Gamma(X\times X,L_{1}^{j_{1}}\otimes L_{2}^{j_{2}}), where Li=πi∗​LL_{i}=\pi_{i}^{*}L and π:X×X→X\pi:X\times X\rightarrow X denotes the projection onto the it​hi^{th} factor, i=1,2i=1,2. Consider the Bergman spaces

ℋj1,j2,j0:={S∈Γ(X×X,L1j1⊗L2j2)/\displaystyle{\mathcal{H}}_{j_{1},j_{2},j_{0}}:=\left\{S\in\Gamma(X\times X,L_{1}^{j_{1}}\otimes L_{2}^{j_{2}})\,/\,\right.
∫X×X|S|2e−2​hj1,j0/2​(x)−2​hj2,j0/2​(y)dVω1(x)dVω2(y)<+∞},\displaystyle\left.\int_{X\times X}|S|^{2}e^{-2h_{j_{1},j_{0}/2}(x)-2h_{j_{2},j_{0}/2}(y)}dV_{\omega_{1}}(x)dV_{\omega_{2}}(y)<+\infty\right\},

where ωi=π∗​ω\omega_{i}=\pi^{*}\omega. It follows from the Ohsawa-Takegoshi-Manivel L2L^{2}-extension theorem [15] that there exists S∈Γ⁡(X×X,L1j1⊗L2j2)S\in\Gamma(X\times X,L_{1}^{j_{1}}\otimes L_{2}^{j_{2}}) such that S|Δ=sS_{|\Delta}=s and

∫X×X|S|2​e−2​hj1,j0/2−2​hj2,j0/2​d​Vω1​d​Vω2≤C5​∫X|s|2​e−2​hj1+j2,j0​d​Vω≤C5,\int_{X\times X}|S|^{2}e^{-2h_{j_{1},j_{0}/2}-2h_{j_{2},j_{0}/2}}dV_{\omega_{1}}dV_{\omega_{2}}\leq C_{5}\int_{X}|s|^{2}e^{-2h_{j_{1}+j_{2},j_{0}}}dV_{\omega}\leq C_{5},

where C5C_{5} only depends on the dimension n=dimℂXn=\dim_{\mathbb{C}}X. Observe that {σl1(j1,j0/2)​(x)⋅σl1(j1,j0/2)​(y)}l1,l2\{\sigma_{l_{1}}^{(j_{1},j_{0}/2)}(x)\cdot\sigma_{l_{1}}^{(j_{1},j_{0}/2)}(y)\}_{l_{1},l_{2}} forms an orthonormal basis of ℋj1,j2,j0{\mathcal{H}}_{j_{1},j_{2},j_{0}}, thus

S⁡(x,y)=∑l1,l2cl1,l2​σl1(j1,j0/2)​(x)​σl2(j2,j0/2)​(y)S(x,y)=\sum_{l_{1},l_{2}}c_{l_{1},l_{2}}\sigma_{l_{1}}^{(j_{1},j_{0}/2)}(x)\sigma_{l_{2}}^{(j_{2},j_{0}/2)}(y)

with ∑|cl1,l2|2≤C5\sum|c_{l_{1},l_{2}}|^{2}\leq C_{5}. It follows therefore from Cauchy-Schwarz inequality that

|s⁡(x)|2=|S⁡(x,x)|2≤C5​∑l1|σl1(j1,j0/2)​(x)|2​∑l2|σl2(j2,j0/2)​(y)|2,|s(x)|^{2}=|S(x,x)|^{2}\leq C_{5}\sum_{l_{1}}|\sigma_{l_{1}}^{(j_{1},j_{0}/2)}(x)|^{2}\sum_{l_{2}}|\sigma_{l_{2}}^{(j_{2},j_{0}/2)}(y)|^{2},

which yields

φj1+j2,j0≤log⁡C52​(j1+j2)+j1j1+j2​φj1,j0/2+j2j1+j2​φj2,j0/2.\varphi_{j_{1}+j_{2},j_{0}}\leq\frac{\log C_{5}}{2(j_{1}+j_{2})}+\frac{j_{1}}{j_{1}+j_{2}}\varphi_{j_{1},j_{0}/2}+\frac{j_{2}}{j_{1}+j_{2}}\varphi_{j_{2},j_{0}/2}.

Note finally that φj,j0/2≤φj,j0\varphi_{j,j_{0}/2}\leq\varphi_{j,j_{0}} since φ=ψ−h≤0\varphi=\psi-h\leq 0, therefore φ^j:=φ2j,j0+2−j−2​log⁡C5\hat{\varphi}_{j}:=\varphi_{2^{j},j_{0}}+2^{-j-2}\log C_{5} is decreasing.

Step3. It remains to make φ^j\hat{\varphi}_{j} smooth. Indeed it has all the other required properties: it is decreasing and by Step 1 we have for all x∈Xx\in X,

(5) φ⁡(x)≤φ^j​(x)≤(1−j0​2−j)​supB⁡(x,r)φ+C6−n​log⁡r2j,\varphi(x)\leq\hat{\varphi}_{j}(x)\leq(1-j_{0}2^{-j})\sup_{B(x,r)}\varphi+\frac{C_{6}-n\log r}{2^{j}},

so that φ^j→φ\hat{\varphi}_{j}\rightarrow\varphi. Let σ1+s2j(2j),…,σNj(2j)∈Γ⁡(X,L2j)\sigma_{1+s_{2^{j}}}^{(2^{j})},\ldots,\sigma_{N_{j}}^{(2^{j})}\in\Gamma(X,L^{2^{j}}) be such that (σl(2j))l(\sigma_{l}^{(2^{j})})_{l} is a basis of Γ⁡(X,L2j)\Gamma(X,L^{2^{j}}) and set

φj:=12j+1​log⁡[∑l=1s2j|σl(2j)|2+εj​∑l=1+s2jNj|σl(2j)|2]+log⁡C52j+2−h.\varphi_{j}:=\frac{1}{2^{j+1}}\log\left[\sum_{l=1}^{s_{2^{j}}}|\sigma_{l}^{(2^{j})}|^{2}+\varepsilon_{j}\sum_{l=1+s_{2^{j}}}^{N_{j}}|\sigma_{l}^{(2^{j})}|^{2}\right]+\frac{\log C_{5}}{2^{j+2}}-h.

Clearly φj∈P​S​H​(X,ω)\varphi_{j}\in PSH(X,\omega). Moreover φj∈𝒞∞​(X)\varphi_{j}\in{\mathcal{C}}^{\infty}(X) because L2jL^{2^{j}} is very ample if jj is large enough (hence we can find, for every x∈Xx\in X, a holomorphic section of L2jL^{2^{j}} on XX which does not vanish at xx). Finally we can choose εj>0\varepsilon_{j}>0 that decrease so fast to zero that (φj)(\varphi_{j}) is still decreasing and converges to φ\varphi. ∎

Corollary 6.2.

Let ω\omega be a Kähler form on a projective algebraic manifold X. Then there exists A≥1A\geq 1 such that for every φ∈P​S​H​(X,ω)\varphi\in PSH(X,\omega), we can find φj∈P​S​H​(X,A​ω)∩𝒞∞​(X)\varphi_{j}\in PSH(X,A\omega)\cap{\mathcal{C}}^{\infty}(X) which decrease towards φ\varphi.

Proof.

Let φ∈P​S​H​(X,ω)\varphi\in PSH(X,\omega). Since XX is projective, we can find a Hodge form ω′\omega^{\prime}. Then C−1​ω′≤ω≤C​ω′C^{-1}\omega^{\prime}\leq\omega\leq C\omega^{\prime} for some constant C≥1C\geq 1. Since P​S​H​(X,ω)⊂P​S​H​(X,C​ω′)PSH(X,\omega)\subset PSH(X,C\omega^{\prime}), it follows from the previous theorem that we can find φj∈P​S​H​(X,C​ω′)∩𝒞∞​(X)\varphi_{j}\in PSH(X,C\omega^{\prime})\cap{\mathcal{C}}^{\infty}(X) that decrease towards φ\varphi. Now the result follows from P​S​H​(X,C​ω′)⊂P​S​H​(X,A​ω)PSH(X,C\omega^{\prime})\subset PSH(X,A\omega)with A=C2A=C^{2}. ∎

Remark 6.3.

When XX is merely Kähler, the above result still holds but the proof is far more intricate. We refer the reader to Demailly’s papers for a proof.

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