ScalingStacks

Proof. [02YJ]

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

Proof.

By Corollary 5.17, we have Ψ=rec⁡(ψ∞)\Psi=\operatorname{rec}(\psi_{\infty}). By equation (3.49), we have rec⁡(ψ∞)=limλ→∞λ−1​ψ∞​(λ⁡(u,v))\operatorname{rec}(\psi_{\infty})=\lim_{\lambda\to\infty}\lambda^{-1}\psi_{\infty}(\lambda(u,v)). Statement (2) follows readily from this and from the expression for ψ∞\psi_{\infty} in Lemma 8.17.

The function Ψ\Psi is strictly concave on Σ\Sigma, because 𝒪ℙ⁡(E)​(1){\mathcal{O}}_{\mathbb{P}(E)}(1) is an ample line bundle. Hence Σ=Π⁡(Ψ)\Sigma=\Pi(\Psi) and this is the fan described in statement (1).

Let (e1∨,…,en∨,f1∨,…,fr∨)(e_{1}^{\vee},\dots,e_{n}^{\vee},f_{1}^{\vee},\dots,f_{r}^{\vee}) be the dual basis of MM induced by the basis of NN. By Proposition 3.64 and statement (2), we have

Δ=conv⁡(0,(a0​ek∨)1≤k≤n,(fℓ∨)1≤ℓ≤r,(aℓ​ek∨+fℓ∨)1≤ℓ≤r1≤k≤n).\Delta=\operatorname{conv}\bigg(0,(a_{0}e^{\vee}_{k})_{1\leq k\leq n},(f^{\vee}_{\ell})_{1\leq\ell\leq r},(a_{\ell}e^{\vee}_{k}+f^{\vee}_{\ell})_{\stackrel{{\scriptstyle 1\leq k\leq n}}{{\scriptscriptstyle 1\leq\ell\leq r}}}\bigg).

Statement (3) follows readily from this.

For the first part of statement (4), it suffices to compute the Legendre-Fenchel dual of ψ∞\psi_{\infty} at a point (x,y)(x,y) in the interior of the polytope. Lemma 8.17 shows that ψ∞\psi_{\infty} is strictly concave. Hence, by Theorem 3.52(3), ∇ψ∞\nabla\psi_{\infty} is a homeomorphism between NℝN_{\mathbb{R}} and Δ∘\Delta^{\circ}. Thus, there exist a unique (u,v)∈Nℝ(u,v)\in N_{\mathbb{R}} such that, for i=1,…,ni=1,\dots,n and j=1,…,rj=1,\dots,r,

xi=∂ψ∞∂ui​(u,v),yj=∂ψ∞∂vj​(u,v).x_{i}=\frac{\partial\psi_{\infty}}{\partial u_{i}}(u,v),\quad y_{j}=\frac{\partial\psi_{\infty}}{\partial v_{j}}(u,v).

We use the conventions x0=L⁡(y)−∑i=1nxix_{0}=L(y)-\sum_{i=1}^{n}x_{i}, y0=1−∑j=1ryjy_{0}=1-\sum_{j=1}^{r}y_{j}, and u0=v0=0u_{0}=v_{0}=0 as before, and also η=∑i=0ne−2​ui\eta=\sum_{i=0}^{n}\operatorname{e}^{-2u_{i}} and ψ=ψ∞\psi=\psi_{\infty}, so that −2​ψ=log⁡(∑j=0re−2​vj⁡ηaj)-2\psi=\log\big(\sum_{j=0}^{r}\operatorname{e}^{-2v_{j}}\eta^{a_{j}}\big). Computing the gradient of ψ\psi, we obtain, for i=1,…,ni=1,\dots,n and j=1,…,rj=1,\dots,r,

xi​e−2​ψ=(∑j=0raj​ηaj−1​e−2​vj)​e−2​ui,yj​e−2​ψ=ηaj​e−2​vj.x_{i}\operatorname{e}^{-2\psi}=\Big(\sum_{j=0}^{r}a_{j}\eta^{a_{j}-1}\operatorname{e}^{-2v_{j}}\Big)\operatorname{e}^{-2u_{i}},\quad y_{j}\operatorname{e}^{-2\psi}=\eta^{a_{j}}\operatorname{e}^{-2v_{j}}.

Combining these expressions, we obtain, for i=0,…,ni=0,\dots,n and j=0,…,rj=0,\dots,r,

xiL⁡(y)=e−2​uiη,yj=e−2​vj+2​ψηaj.\frac{x_{i}}{L(y)}=\frac{\operatorname{e}^{-2u_{i}}}{\eta},\quad y_{j}=\frac{\operatorname{e}^{-2v_{j}+2\psi}}{\eta^{a_{j}}}.

From the case i=0i=0 we deduce η=L⁡(y)/x0\eta=L(y)/x_{0} and from the case j=0j=0 it results 2​ψ=log⁡(y0)+a0​log⁡(x0/L⁡(y))2\psi=\log(y_{0})+a_{0}\log(x_{0}/L(y)). From this, one can verify

ui=12​log⁡(x0xi),vj=12​log⁡(y0yj)+a0−aj2​log⁡(x0L⁡(y)).u_{i}=\frac{1}{2}\log\Big(\frac{x_{0}}{x_{i}}\Big),\quad v_{j}=\frac{1}{2}\log\Big(\frac{y_{0}}{y_{j}}\Big)+\frac{a_{0}-a_{j}}{2}\log\Big(\frac{x_{0}}{L(y)}\Big).

From Theorem 3.52(4), we have ψ∨​(x,y)=⟨x,u⟩+⟨y,v⟩−ψ⁡(u,v)\psi^{\vee}(x,y)=\langle x,u\rangle+\langle y,v\rangle-\psi(u,v). Inserting the expressions above for ψ\psi, uiu_{i} and vjv_{j} in terms of x,yx,y, we obtain the stated formula.

For v≠∞v\neq\infty, we have ψv=Ψ\psi_{v}=\Psi. The last statement follows from Example 3.16. ∎

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