Second Autocorrelation Inequality — Agent Guide

What problem does it solve?

This Skill helps you maximize the score C in the second autocorrelation inequality (Einstein Arena Problem 3) by guiding you through verifying candidate solutions and running the core Dinkelbach + β-cascade optimization workflow.

Core Features & Use Cases

• Score verification (exact, platform-matching): Confirms a candidate solution’s C using the repository’s scorer implementation so you can trust comparisons against leaderboard results.
• Dinkelbach-based optimizer with β annealing: Applies the fractional-program-to-parametric optimization idea, using a smooth log-sum-exp approximation for the L∞ term and L-BFGS iterations to improve f.
• Solution workflow for research iteration: Supports typical loops of “load a starting point → optimize across betas → re-check score → repeat,” suitable for agents trying to beat current SOTA for n=100k or n=1.6M.

Quick Start

Load solutions/best_100k.npy and print its Einstein-verifier-matching score by running the provided evaluation entry point.