ScalingStacks

7.2. Metrics, heights and entropy [02XJ]

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 Β· Original author HTML

7.2. Metrics, heights and entropy

In this section we will consider some metrics arising from polytopes. We will use the notation of Β§4 and Β§5. In particular, we consider a split torus over the field of rational numbers 𝕋≃𝔾m,β„šn\mathbb{T}\simeq\mathbb{G}_{m,\mathbb{Q}}^{n} and we denote by N,M,Nℝ,MℝN,M,N_{\mathbb{R}},M_{\mathbb{R}} the lattices and dual spaces corresponding to 𝕋\mathbb{T}.

Let Ξ”βŠ‚Mℝ\Delta\subset M_{\mathbb{R}} be a lattice polytope of dimension nn. Let β„“i\ell_{i}, i=1,…,ri=1,\dots,r, be affine functions on MℝM_{\mathbb{R}} defined as β„“i​(x)=⟨ui,xβŸ©βˆ’Ξ»i\ell_{i}(x)=\langle u_{i},x\rangle-\lambda_{i} for some ui∈Nℝu_{i}\in N_{\mathbb{R}} and Ξ»iβˆˆβ„\lambda_{i}\in\mathbb{R} such that β„“iβ‰₯0\ell_{i}\geq 0 on Ξ”\Delta and let also ci>0c_{i}>0. Write β„“=(β„“1,…,β„“r)\ell=(\ell_{1},\dots,\ell_{r}) and c=(c1,…,cr)c=(c_{1},\dots,c_{r}). We consider the function ϑΔ,β„“,c:Δ→ℝ\vartheta_{\Delta,\ell,c}\colon\Delta\to\mathbb{R} defined, for xβˆˆΞ”x\in\Delta, by

(7.20) ϑΔ,β„“,c(x)=βˆ’βˆ‘i=1rciβ„“i(x)log(β„“i(x)).\vartheta_{\Delta,\ell,c}(x)=-\sum_{i=1}^{r}c_{i}\ell_{i}(x)\log(\ell_{i}(x)).

When Ξ”,β„“,c\Delta,\ell,c are clear from the context, we write for short Ο‘=ϑΔ,β„“,c\vartheta=\vartheta_{\Delta,\ell,c}.

Lemma 7.21.

Let notation be as above.

  1. (1)

    The function ϑΔ,β„“,c\vartheta_{\Delta,\ell,c} is concave.

  2. (2)

    If the family {ui}i\{u_{i}\}_{i} generates NℝN_{\mathbb{R}}, then ϑΔ,β„“,c\vartheta_{\Delta,\ell,c} is strictly concave.

  3. (3)

    If Ξ”=β‹‚i{x∈Mℝ|β„“i​(x)β‰₯0}\Delta=\bigcap_{i}\{x\in M_{\mathbb{R}}|\ell_{i}(x)\geq 0\}, then the restriction of ϑΔ,β„“,c\vartheta_{\Delta,\ell,c} to Ξ”βˆ˜\Delta^{\circ} is of Legendre type (Definition 3.51).

Proof.

Let 1≀i≀r1\leq i\leq r and consider the affine map β„“i:Δ→ℝβ‰₯0\ell_{i}\colon\Delta\to\mathbb{R}_{\geq 0}. We have that βˆ’z​log⁑(z)-z\log(z) is a strictly concave function on ℝβ‰₯0\mathbb{R}_{\geq 0} and βˆ’β„“i​log⁑(β„“i)=β„“iβˆ—β€‹(βˆ’z​log⁑(z))-\ell_{i}\log(\ell_{i})=\ell_{i}^{*}(-z\log(z)). Hence, each function βˆ’ci​ℓi​(x)​log⁑(β„“i​(x))-c_{i}\ell_{i}(x)\log(\ell_{i}(x)) is concave and so is Ο‘\vartheta, as stated in (1)

For statement (2), let x1,x2x_{1},x_{2} be two different points of Ξ”\Delta. The assumption that {ui}i\{u_{i}\}_{i} generates NℝN_{\mathbb{R}} implies that β„“i0​(x1)β‰ β„“i0​(x2)\ell_{i_{0}}(x_{1})\neq\ell_{i_{0}}(x_{2}) for some i0i_{0}. Hence, the affine map β„“i0\ell_{i_{0}} gives an injection of the segment x1​x2Β―{\overline{x_{1}x_{2}}} into ℝβ‰₯0\mathbb{R}_{\geq 0}. We deduce that βˆ’ci0​ℓi0​log⁑(β„“i0)-c_{i_{0}}\ell_{i_{0}}\log(\ell_{i_{0}}) is strictly concave on x1​x2Β―{\overline{x_{1}x_{2}}} and so is Ο‘\vartheta. Varying x1,x2x_{1},x_{2}, we deduce that Ο‘\vartheta is strictly concave on Ξ”\Delta.

For statement (3), it is clear that Ο‘|Ξ”βˆ˜\vartheta|_{\Delta^{\circ}} is differentiable. Moreover, the assumption that Ξ”\Delta is the intersection of the halfspaces defined by the β„“i\ell_{i}’s implies that the uiu_{i}’s generate NℝN_{\mathbb{R}} and so Ο‘\vartheta is strictly concave. The gradient of Ο‘\vartheta is given, for xβˆˆΞ”βˆ˜x\in\Delta^{\circ}, by

(7.22) βˆ‡Ο‘(x)=βˆ’βˆ‘i=1rciui(log(β„“i(x)+1).\nabla\vartheta(x)=-\sum_{i=1}^{r}c_{i}u_{i}(\log(\ell_{i}(x)+1).

Let βˆ₯β‹…βˆ₯\|\cdot\| be a fixed norm on MℝM_{\mathbb{R}} and (xj)jβ‰₯0(x_{j})_{j\geq 0} a sequence in Ξ”βˆ˜\Delta^{\circ} converging to a point in the border. Then there exists some i1i_{1} such β„“i1​(xj)β†’j0\ell_{i_{1}}(x_{j})\stackrel{{\scriptstyle j}}{{\to}}0. Thus, β€–βˆ‡Ο‘β€‹(x)β€–β†’j∞\|\nabla\vartheta(x)\|\stackrel{{\scriptstyle j}}{{\to}}\infty and the statement follows. ∎

Definition 7.23.

Let ΣΔ\Sigma_{\Delta} and ΨΔ\Psi_{\Delta} be the fan and the support function on NℝN_{\mathbb{R}} induced by Ξ”\Delta. Let (XΣΔ,DΨΔ)(X_{\Sigma_{\Delta}},D_{\Psi_{\Delta}}) be the associated polarized toric variety over β„š\mathbb{Q} and write L=π’ͺ⁑(DΨΔ)L={\mathcal{O}}(D_{\Psi_{\Delta}}). By Lemma 7.21(1), Ο‘\vartheta is a concave function on Ξ”\Delta. By Theorem 5.73, it corresponds to some approachable toric metric on L⁑(β„‚)L(\mathbb{C}). We denote this metric by βˆ₯β‹…βˆ₯Ξ”,β„“,c\|\cdot\|_{\Delta,\ell,c}. We write LΒ―{\overline{L}} for the line bundle LL equipped with the metric βˆ₯β‹…βˆ₯Ξ”,β„“,c\|\cdot\|_{\Delta,\ell,c} at the Archimedean place of β„š\mathbb{Q} and with the canonical metric at the non-Archimedean places. This is an example of an adelic toric metric.

Example 7.24.

Following the notation in Example 3.53, consider the standard simplex Ξ”n\Delta^{n} and the concave function Ο‘=12​Ρn\vartheta=\frac{1}{2}\varepsilon_{n} on Ξ”n\Delta^{n}. From examples 3.53 and 5.18(1), we deduce that the corresponding metric is the Fubini-Study metric of π’ͺ​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}}.

In case Ξ”\Delta is the intersection of the halfspaces defined by the β„“i\ell_{i}’s, Lemma 7.21(3) shows that Ο‘|Ξ”βˆ˜\vartheta|_{\Delta^{\circ}} of Legendre type (Definition 3.51). By Theorem 3.52 and equation (7.22), the gradient of Ο‘\vartheta gives a homeomorphism between Ξ”βˆ˜\Delta^{\circ} and NℝN_{\mathbb{R}} and, for xβˆˆΞ”βˆ˜x\in\Delta^{\circ},

(7.25) Ο‘βˆ¨(βˆ‡Ο‘(x))=βˆ’βˆ‘i=1rciΞ»ilog(β„“i(x))+ci⟨ui,x⟩.\vartheta^{\vee}(\nabla\vartheta(x))=-\sum_{i=1}^{r}c_{i}\lambda_{i}\log(\ell_{i}(x))+c_{i}\langle u_{i},x\rangle.

This gives an explicit expression of the function ψβˆ₯β‹…βˆ₯Ξ”,β„“,c=Ο‘βˆ¨\psi_{\|\cdot\|_{\Delta,\ell,c}}=\vartheta^{\vee}, and a fortiori of the metric βˆ₯β‹…βˆ₯Ξ”,β„“,c{\|\cdot\|_{\Delta,\ell,c}}, in the coordinates of the polytope. Up to our knowledge, there is no simple expression for ψ\psi in linear coordinates of NℝN_{\mathbb{R}}, except for special cases like Fubini-Study.

Remark 7.26.

This kind of metrics are interesting when studying the KΓ€hler geometry of toric varieties. Given a Delzant polytope Ξ”βŠ‚Mℝ\Delta\subset M_{\mathbb{R}}, Guillemin has constructed a β€œcanonical” KΓ€hler structure on the associated symplectic toric variety [Gui95]. The corresponding symplectic potential is the function βˆ’Ο‘Ξ”,β„“,c-\vartheta_{\Delta,\ell,c}, for the case when rr is the number of facets of Ξ”\Delta, ci=1/2c_{i}=1/2 for all ii, and uiu_{i} is a primitive vector in NN and Ξ»i\lambda_{i} is an integer such that Ξ”={x∈Mℝ|⟨ui,x⟩β‰₯Ξ»i,i=1,…,r}\Delta=\{x\in M_{\mathbb{R}}|\langle u_{i},x\rangle\geq\lambda_{i},i=1,\dots,r\}, see [Gui95, Appendix 2, (3.9)].

In this case, the metric βˆ₯β‹…βˆ₯Ξ”,β„“,c\|\cdot\|_{\Delta,\ell,c} on the line bundle π’ͺ​(DΞ¨)an{\mathcal{O}}(D_{\Psi})^{{\text{\rm an}}} is smooth and positive and, as explained in Remark 5.74, its Chern form gives this canonical KΓ€hler form.

We obtain the following formula for the height of XΣΔX_{\Sigma_{\Delta}} with respect to the adelic metrized line bundle LΒ―\overline{L}, in terms of the coefficients Ck​(Ξ”,ui,V)C_{k}(\Delta,u_{i},V).

Proposition 7.27.

Let notation be as in Definition 7.23. Then hL¯⁑(XΣΔ)\operatorname{h}_{\overline{L}}(X_{\Sigma_{\Delta}}) equals

(n+1)!β€‹βˆ‘i=1rciβ€‹βˆ‘VβˆˆΞ”β‘(ui)βˆ‘k=0dim(V)Ck​(Ξ”,ui,V)​ℓi​(V)nβˆ’k+1(nβˆ’k+1)!​(βˆ‘j=2nβˆ’k+11jβˆ’log⁑(β„“i​(V))).{(n+1)!}\sum_{i=1}^{r}c_{i}\sum_{V\in\Delta(u_{i})}\sum_{k=0}^{\dim(V)}C_{k}(\Delta,u_{i},V)\frac{\ell_{i}(V)^{n-k+1}}{(n-k+1)!}\left(\sum_{j=2}^{n-k+1}\frac{1}{j}-\log(\ell_{i}(V))\right).

Suppose furthermore that Ξ”βŠ‚β„n\Delta\subset\mathbb{R}^{n} is a simplex, r=n+1r=n+1 and that β„“i\ell_{i}, i=1,…,n+1i=1,\dots,n+1, are affine functions such that Ξ”=β‹‚i{x∈Mℝ|β„“i​(x)β‰₯0}\Delta=\bigcap_{i}\{x\in M_{\mathbb{R}}|\ell_{i}(x)\geq 0\}. Then

(7.28) hL¯⁑(XΣΔ)=n!​volM⁑(Ξ”)β€‹βˆ‘i=1n+1ci​ℓi​(Ξ½i)​(βˆ‘j=2n+11jβˆ’log⁑(β„“i​(Ξ½i))).\operatorname{h}_{\overline{L}}(X_{\Sigma_{\Delta}})=n!\operatorname{vol}_{M}(\Delta)\sum_{i=1}^{n+1}c_{i}\ell_{i}(\nu_{i})\bigg(\sum_{j=2}^{n+1}\frac{1}{j}-\log(\ell_{i}(\nu_{i}))\bigg).

where Ξ½i\nu_{i} is the unique vertex of Ξ”\Delta not contained in the facet defined by β„“i\ell_{i}.

Proof.

The first statement follows readily from Theorem 6.37 and Proposition 7.3 applied to the functions fi​(z)=(log⁑(zβˆ’Ξ»i)βˆ’βˆ‘j=2n+11j)​(zβˆ’Ξ»i)n+1/(n+1)!f_{i}(z)=\left(\log(z-\lambda_{i})-\sum_{j=2}^{n+1}\frac{1}{j}\right)(z-\lambda_{i})^{n+1}/(n+1)!. The second statement follows similarly from Proposition 7.15. ∎

Example 7.29.

Let π’ͺ⁑(1){\mathcal{O}}(1) be the universal line bundle of β„™n\mathbb{P}^{n}. The Fubini-Study metric of π’ͺ​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}} corresponds to the case of the standard simplex, β„“i​(x)=xi\ell_{i}(x)=x_{i}, i=1,…,ni=1,\dots,n and β„“n+1​(x)=1βˆ’βˆ‘i=1nxi\ell_{n+1}(x)=1-\sum_{i=1}^{n}x_{i} and the choice ci=1/2c_{i}=1/2 for all ii. Hence we recover from (7.28) the well known expression for the height of β„™n\mathbb{P}^{n} with respect to the Fubini-Study metric in [BGS94, Lemma 3.3.1]:

hπ’ͺ⁑(1)¯⁑(β„™n)=n+12β€‹βˆ‘j=2n+11j.\operatorname{h}_{{\overline{{\mathcal{O}}(1)}}}(\mathbb{P}^{n})=\frac{n+1}{2}\sum_{j=2}^{n+1}\frac{1}{j}.
Example 7.30.

In dimension 11, a polytope is an interval of the form Ξ”=[m0,m1]\Delta=[m_{0},m_{1}] for some miβˆˆβ„€m_{i}\in\mathbb{Z}. The corresponding roof function in (7.20) writes down, for x∈[m0,m1]x\in[m_{0},m_{1}], as

(7.31) Ο‘(x)=βˆ’βˆ‘i=1rciβ„“i(x)log(β„“i(x))\vartheta(x)=-\sum_{i=1}^{r}c_{i}\ell_{i}(x)\log(\ell_{i}(x))

for affine function β„“i=ui​xβˆ’Ξ»i\ell_{i}=u_{i}x-\lambda_{i} which take non negative values on the Ξ”\Delta and ci>0c_{i}>0

The polarized toric variety corresponding to Ξ”\Delta is β„™1\mathbb{P}^{1} together with the ample divisor m1​[(0:1)]βˆ’m0​[(1:0)]m_{1}[(0:1)]-m_{0}[(1:0)]. Write L=π’ͺβ„™1​(x1βˆ’x0)L={\mathcal{O}}_{\mathbb{P}^{1}}(x_{1}-x_{0}) for the associate line bundle and LΒ―{\overline{L}} for the adelic metrized line bundle corresponding to the function Ο‘\vartheta. The Legendre-Fenchel dual to βˆ’ci​ℓi​(x)​log⁑(β„“i​(x))-c_{i}\ell_{i}(x)\log(\ell_{i}(x)) is the function fi:ℝ→ℝf_{i}\colon\mathbb{R}\rightarrow\mathbb{R} defined, for vβˆˆβ„v\in\mathbb{R}, by

fi​(v)=Ξ»iui​vβˆ’ci​eβˆ’1βˆ’vci​ui.f_{i}(v)=\frac{\lambda_{i}}{u_{i}}v-c_{i}{\operatorname{e}}^{-1-\frac{v}{{c_{i}u_{i}}}}.

Therefore, the function ψ=Ο‘βˆ¨\psi=\vartheta^{\vee} is the sup-convolution of these function, namely ψ=f1βŠžβ‹―βŠžfm\psi=f_{1}\boxplus\dots\boxplus f_{m} For the height, a simple computation shows that

hL¯⁑(β„™1)=∫m0m1ϑ​d​x=βˆ‘i=1rci4​ui​[β„“i​(x)2​(1βˆ’2​log⁑(β„“i​(x)))]m0m1{\operatorname{h}_{\overline{L}}(\mathbb{P}^{1})}=\int_{m_{0}}^{m_{1}}\vartheta\,\text{\rm d}x=\sum_{i=1}^{r}\frac{c_{i}}{4u_{i}}\Big[\ell_{i}(x)^{2}\left(1-2\log(\ell_{i}(x))\right)\Big]^{m_{1}}_{m_{0}}

In some cases, the height of a toric variety with respect to the metrics constructed above has an interpretation in terms of the average entropy of some natural random processes. Let Ξ“\Gamma be an arbitrary polytope containing Ξ”\Delta. For a point x∈ri⁑(Ξ”)x\in\operatorname{ri}(\Delta), we consider the partition Ξ x\Pi_{x} of Ξ“\Gamma which consists of the cones Ξ·x,F\eta_{x,F} of vertex xx and base the relative interior of each proper face FF of Ξ“\Gamma.

We consider Ξ“\Gamma as a probability space endowed with the uniform probability distribution and Ξ²x\beta_{x} the random variable which, for a point yβˆˆΞ“y\in\Gamma, returns the base FF of the unique cone Ξ·x,F\eta_{x,F} it belongs to. Clearly, the probability that a given face FF is returned is the ratio of the volume of the cone based on FF to the volume of Ξ“\Gamma. We have voln⁑(Ξ·x,F)=nβˆ’1​dist⁑(x,F)​volnβˆ’1⁑(F){\operatorname{vol}}_{n}(\eta_{x,F})={n}^{-1}{\operatorname{dist}}(x,F){\operatorname{vol}}_{n-1}(F) where, as before, voln\operatorname{vol}_{n} and volnβˆ’1\operatorname{vol}_{n-1} denote the Lebesgue measure on ℝn\mathbb{R}^{n} and on LFL_{F}, respectively. Hence,

(7.32) P⁑(Ξ²x=F)={dist⁑(x,F)​volnβˆ’1⁑(F)n​voln​(Ξ“)Β if ​dim(F)=nβˆ’1,0Β if ​dim(F)≀nβˆ’2.P(\beta_{x}=F)=\begin{cases}\displaystyle\frac{{\operatorname{dist}}(x,F){\operatorname{vol}}_{n-1}(F)}{n{\operatorname{vol}}_{n}(\Gamma)}&\text{ if }\dim(F)=n-1,\\ 0&\text{ if }\dim(F)\leq n-2.\end{cases}

The entropy of the random variable Ξ²x\beta_{x} is

β„°(x)=βˆ’βˆ‘FP(Ξ²x=F)log(P(Ξ²x=F)),{\mathcal{E}}(x)=-\sum_{F}P(\beta_{x}=F)\log(P(\beta_{x}=F)),

where the sum is over the facets FF of Ξ“\Gamma.

For each facet FF of Ξ“\Gamma we let uFβˆˆβ„nu_{F}\in\mathbb{R}^{n} be the inner normal vector to FF of Euclidean norm (nβˆ’1)!​volnβˆ’1⁑(F)(n-1)!\operatorname{vol}_{n-1}(F) and Ξ»F=ΨΓ​(uF)βˆˆβ„\lambda_{F}=\Psi_{\Gamma}(u_{F})\in\mathbb{R} and consider the affine form β„“F\ell_{F} defined as β„“F​(x)=⟨uF,xβŸ©βˆ’Ξ»F\ell_{F}(x)=\langle u_{F},x\rangle-\lambda_{F}. Hence, Ξ“={x∈Mℝ|β„“F​(x)β‰₯0}\Gamma=\{x\in M_{\mathbb{R}}|\ell_{F}(x)\geq 0\}. Let also cF=cc_{F}=c for some constant c>0c>0. By the Minkowski condition, βˆ‘FuF=0\sum_{F}u_{F}=0. Hence βˆ‘Fβ„“F=βˆ’βˆ‘FΞ»F\sum_{F}\ell_{F}=-\sum_{F}\lambda_{F}.

Remark 7.33.

Suppose that Ξ“\Gamma is a lattice polytope and let FF be a facet of Ξ“\Gamma. Recall that M⁑(F)M(F) is the lattice LF∩ML_{F}\cap M and let M​(F)β€²M(F)^{\prime} be the sublattice of M⁑(F)M(F) generated by the differences of the lattice points in FF. Then the vector uFu_{F} can be alternatively defined as [M(F):M(F)β€²][M(F):M(F)^{\prime}] times the primitive inner normal vector to the facet FF.

The concave function Ο‘=βˆ’βˆ‘Fcβ„“F(x)log(β„“F(x))\vartheta=-\sum_{F}{c\,\ell_{F}(x)}\log({\ell_{F}(x)}) belongs to the class of functions considered in Definition 7.23. Thus, we obtain a line bundle with an adelic toric metric LΒ―{\overline{L}} on XΞ”X_{\Delta}. For short, we write X=XΞ”X=X_{\Delta}. The following result shows that the average entropy of the random variable Ξ²x\beta_{x} with respect to the uniform distribution on Ξ”\Delta can be expressed in terms of the height of the toric variety XX with respect to LΒ―{\overline{L}}.

Proposition 7.34.

With the above notation,

1voln⁑(Ξ”)β€‹βˆ«Ξ”β„°β‘(x)​d​voln=1n!​voln​(Ξ“)​(hL¯⁑(X)c⁑(n+1)​degL​(X)βˆ’log⁑(n!​voln⁑(Ξ“))​(βˆ‘FΞ»F))\frac{1}{\operatorname{vol}_{n}(\Delta)}\int_{\Delta}{\mathcal{E}}(x)\,\text{\rm d}\operatorname{vol}_{n}=\frac{1}{n!\operatorname{vol}_{n}(\Gamma)}\bigg(\frac{\operatorname{h}_{{\overline{L}}}(X)}{c(n+1)\deg_{L}(X)}-{\log(n!\operatorname{vol}_{n}(\Gamma))}\Big(\sum_{F}\lambda_{F}\Big)\bigg)

where the sum is over the facets FF of Ξ“\Gamma. In particular, if Ξ“=Ξ”\Gamma=\Delta,

1voln⁑(Ξ”)β€‹βˆ«Ξ”β„°β‘(x)​d​voln=hL¯⁑(X)c⁑(n+1)​degL​(X)2βˆ’log⁑(degL⁑(X))degL⁑(X)​(βˆ‘FΞ»F).\frac{1}{\operatorname{vol}_{n}(\Delta)}\int_{\Delta}{\mathcal{E}}(x)\,\text{\rm d}\operatorname{vol}_{n}=\frac{\operatorname{h}_{{\overline{L}}}(X)}{c(n+1)\deg_{L}(X)^{2}}-\frac{\log(\deg_{L}(X))}{\deg_{L}(X)}\Big(\sum_{F}\lambda_{F}\Big).
Proof.

For x∈ri⁑(Ξ”)x\in\operatorname{ri}(\Delta) and FF a facet of Ξ“\Gamma, we deduce from equation (7.32) that P⁑(Ξ²x=F)=β„“F​(x)/(n!​voln⁑(Ξ“))P(\beta_{x}=F)=\ell_{F}(x)/(n!\operatorname{vol}_{n}(\Gamma)). Hence,

ℰ⁑(x)\displaystyle{\mathcal{E}}(x) =βˆ’βˆ‘Fβ„“F​(x)n!​voln​(Ξ“)log(β„“F​(x)n!​voln​(Ξ“))\displaystyle=-\sum_{F}\frac{\ell_{F}(x)}{n!{\operatorname{vol}_{n}}(\Gamma)}\log\Big(\frac{\ell_{F}(x)}{n!{\operatorname{vol}_{n}}(\Gamma)}\Big)
=1n!​voln​(Ξ“)(βˆ’βˆ‘Fβ„“F(x)log(β„“F(x))βˆ’log(n!voln(Ξ“))(βˆ‘FΞ»F))\displaystyle=\frac{1}{n!{\operatorname{vol}_{n}}(\Gamma)}\bigg(-\sum_{F}{\ell_{F}(x)}\log({\ell_{F}(x)})-\log({n!{\operatorname{vol}_{n}}(\Gamma)})\Big(\sum_{F}\lambda_{F}\Big)\bigg)
=1n!​voln​(Ξ“)​(ϑ⁑(x)cβˆ’log⁑(n!​voln⁑(Ξ“))​(βˆ‘FΞ»F)).\displaystyle=\frac{1}{n!{\operatorname{vol}_{n}}(\Gamma)}\bigg(\frac{\vartheta(x)}{c}-\log({n!{\operatorname{vol}_{n}}(\Gamma)})\Big(\sum_{F}\lambda_{F}\Big)\bigg).

The result then follows from Theorem 6.37. ∎

Example 7.35.

The Fubini-Study metric of π’ͺ​(1)an{\mathcal{O}}(1)^{{\text{\rm an}}} corresponds to the case when Ξ“\Gamma and Ξ”\Delta are the standard simplex Ξ”n\Delta^{n} and c=1/2c=1/2. In that case, the average entropy of the random variable Ξ²x\beta_{x} is

1n!β€‹βˆ«Ξ”nℰ⁑(x)​d​voln=2​hπ’ͺ⁑(1)¯​(β„™n)(n+1)=βˆ‘j=2n+11j.\frac{1}{n!}\int_{\Delta^{n}}{\mathcal{E}}(x)\,\text{\rm d}\operatorname{vol}_{n}=\frac{2\operatorname{h}_{{\overline{{\mathcal{O}}(1)}}}(\mathbb{P}^{n})}{(n+1)}=\sum_{j=2}^{n+1}\frac{1}{j}.
Remark 7.36.

In case Ξ”\Delta is a Delzant polytope whose facets have lattice volume 1, Ξ“=Ξ”\Gamma=\Delta, and c=1/2c=1/2, the roof function Ο‘\vartheta coincides with the symplectic potential of Guillemin canonical KΓ€hler metric, see Remark 7.26.

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