We present a framework for quantifying gerrymandering using persistent homology, extending the level-set filtration of Feng & Porter to compare precinct and district-level voting data. Because precincts are the smallest electoral subunits and are too small to be gerrymandered individually, discrepancies between precinct and district-level persistence barcodes provide evidence of partisan manipulation. First, we analyze North Carolina House of Representatives elections (2012-2024). Comparing persistence barcodes via the bottleneck distance shows that precinct-level voting patterns remain stable across election cycles while district-level patterns shift significantly in redistricting years, providing topological evidence of gerrymandering that compactness-based scores (Polsby-Popper, Reock) fail to detect. Second, we extend the analysis to the 2025 congressional maps for Texas and California. The Texas 2025 plan, which was struck down by a federal court as a racial gerrymander, shows a bottleneck distance of 12.00 (vs. 8.00 for the 2024 plan), with approximately 748,000 Democratic votes-48% of the total Democratic margin-affected. We additionally introduce persistent homology applied to racial vote-margin rasters (Black and Hispanic Voting Age Population vs. White Voting Age Population), providing the first topological framework for quantifying racial gerrymandering. For California, district-level persistence exceeds precinct-level persistence, consistent with the packing of Republican voters under the proposed 2025 map. We further apply the framework to the Louisiana 2026 congressional map, which emerged from Louisiana v. Callais (2025). The Louisiana analysis reveals a bottleneck distance of 3.0 between racial and partisan barcodes, indicating a strong correlation between Democratic vote share and minority VAP-providing topological context for the Supreme Court's ruling.

翻译：暂无翻译