Balanced Vector Optimization

1 Balanced Vector Optimization

1.1 Introduction

This blueprint documents the Lean 4 formalization of a theorem about symmetric log-concave functions on weak compositions. The main result states that such functions are maximized on balanced vectors and minimized on concentrated vectors.

The primary application is to descendant integrals on moduli spaces of curves:

\[ \langle \tau _{e_1} \cdots \tau _{e_n} \rangle _g = \int _{\overline{\mathcal{M}}_{g,n}} \psi _1^{e_1} \cdots \psi _n^{e_n} \]

1.2 Definitions

Definition 1 Weak Composition
#

A weak composition of \(d\) into \(n\) parts is an \(n\)-tuple \(e = (e_1, \ldots , e_n) \in \mathbb {Z}^n\) such that:

  • \(\sum _{i=1}^n e_i = d\)

  • \(e_i \geq 0\) for all \(i\)

We denote the set of such tuples by \(E(n,d)\).

Definition 2 Balanced Vector
#

A weak composition \(e \in E(n,d)\) is balanced if \(|e_i - e_j| \leq 1\) for all \(i, j \in \{ 1, \ldots , n\} \).

Definition 3 Concentrated Vector
#

A weak composition \(e \in E(n,d)\) is concentrated if there exists \(k\) such that \(e = d \cdot \delta _k\), i.e., all mass is on a single index.

Definition 4 Symmetric Function
#

A function \(D : E(n,d) \to \mathbb {Q}\) is \(S_n\)-symmetric if \(D(e \circ \sigma ^{-1}) = D(e)\) for all permutations \(\sigma \in S_n\).

Definition 5 Log-Concave Function
#

A function \(D : E(n,d) \to \mathbb {Q}\) is log-concave if for all \(e \in E(n,d)\) and indices \(i \neq j\) with \(e_i \geq 1\) and \(e_j \geq 1\):

\[ D(e)^2 \geq D(e - \delta _i + \delta _j) \cdot D(e + \delta _i - \delta _j) \]

(The conditions \(e_i \geq 1\) and \(e_j \geq 1\) ensure both modified vectors remain in \(E(n,d)\).)

Definition 6 Symmetric Log-Concave Function

A symmetric log-concave function on \(E(n,d)\) is a function \(D : E(n,d) \to \mathbb {Q}\) satisfying:

  1. \(S_n\)-symmetry

  2. Log-concavity

  3. Strict positivity: \(D(e) {\gt} 0\) for all \(e\)

1.3 Main Result

Theorem 7 Main Theorem

Let \(D : E(n,d) \to \mathbb {Q}\) be a function satisfying:

  1. \(S_n\)-symmetry: \(D(e \circ \sigma ^{-1}) = D(e)\) for all permutations \(\sigma \)

  2. Log-concavity: \(D(e)^2 \geq D(e - \delta _i + \delta _j) \cdot D(e + \delta _i - \delta _j)\) for all \(i \neq j\) with \(e_i, e_j \geq 1\)

  3. Strict positivity: \(D(e) {\gt} 0\) for all \(e\)

Then:

  • Maximum: There exists a balanced vector \(b \in E(n,d)\) such that \(D(e) \leq D(b)\) for all \(e \in E(n,d)\).

  • Minimum: There exists a concentrated vector \(c \in E(n,d)\) such that \(D(c) \leq D(e)\) for all \(e \in E(n,d)\).

1.4 Key Lemmas and Theorems

Theorem 8 Maximized on Balanced

Let \(D\) be a symmetric log-concave function on \(E(n,d)\). Then there exists a balanced vector \(b \in E(n,d)\) such that \(D(e) \leq D(b)\) for all \(e \in E(n,d)\).

Theorem 9 Minimized on Concentrated

Let \(D\) be a symmetric log-concave function on \(E(n,d)\). Then there exists a concentrated vector \(c \in E(n,d)\) such that \(D(c) \leq D(e)\) for all \(e \in E(n,d)\).

Lemma 10 Unimodal of Log-Concave Palindromic
#

If a sequence is log-concave and palindromic, then it is unimodal (non-decreasing up to the middle, then non-increasing).

Lemma 11 Balanced Maximizes (Pointwise)
#

For a symmetric log-concave function \(D\) and any \(e \in E(n,d)\), there exists a balanced \(b \in E(n,d)\) with \(D(e) \leq D(b)\).

Lemma 12 Concentrated Minimizes (Pointwise)
#

For a symmetric log-concave function \(D\) and any \(e \in E(n,d)\), there exists a concentrated \(c \in E(n,d)\) with \(D(c) \leq D(e)\).