What Does Machine Learning Learn from Numerical Calabi--Yau Geometry? A Controlled Study on the Fermat Quintic
Abstract
We study polynomial regression of numerical metric errors on the Fermat quintic. A known hypersurface volume-ratio identity gives the exact reference relative to the induced Fubini--Study metric, where on unit-normalized homogeneous coordinates. We optimize algebraic metrics at degrees while preserving finite Fermat symmetry, then fit their remaining logarithmic volume-form error. Regression splits and bootstrap draws keep the five roots of each sampled line together, with all compared teachers evaluated on common clouds. Warm-start replication correlations exceed at . Across degrees, source-derived modes fall short of target-specific fits, including in a post-hoc quartic check, but outperform random spans of matched rank in the tested libraries. Two independent random-invariant starts have residual correlations --, while their fitted cubic modes correlate at -- and retain useful calibrated cross-start prediction. Their fitted modes agree poorly with those of the principal warm-start teacher. Adding the phase-sensitive invariant improves prediction beyond the cubic modulus-only library for all seven warm-start teachers. The gain depends on the library: at , a quartic modulus-only fit explains about 97--98% of residual variance and the conditional interval for the added gain includes zero for both replicas under both measures. The gains therefore establish neither phase independence nor irreducible phase necessity; fixed-baseline augmentation preserves this distinction. The fitted components depend on the numerical protocol; global optimality and a universal finite-degree correction remain unestablished.
Figures & tables
| character orbits | largest block | coordinates | ||
|---|---|---|---|---|
| 6 | 205 | 9 | 4 | 24 |
| 8 | 460 | 14 | 4 | 59 |
| 10 | 875 | 19 | 10 | 208 |
| 12 | 1490 | 22 | 10 | 355 |
| cloud | library | rank / columns | range |
|---|---|---|---|
| seven-teacher | cubic | 35 / 35 | |
| seven-teacher | quartic | 66–67 / 70 | |
| three-teacher | cubic | 35 / 35 | |
| three-teacher | quartic | 66–67 / 70 |
| (%) | RMSE int | RMSE Ω | |||
|---|---|---|---|---|---|
| 6 | 0.1411 | 0.002482 | 0.002021 | 0.958455 | 0.921715 |
| 8 | 0.0490 | 0.000627 | 0.000632 | 0.893214 | 0.865281 |
| 10 | 0.0436 | 0.000650 | 0.000567 | 0.729694 | 0.653304 |
| 12 | 0.0641 | 0.000903 | 0.000823 | 0.710787 | 0.624140 |
| measure | |||||
|---|---|---|---|---|---|
| 10 | intersection | 0.182 | 0.020 | 0.167 | 0.019 |
| 10 | volume weighted | 0.191 | 0.022 | 0.187 | 0.021 |
| 12 | intersection | 0.240 | 0.036 | 0.227 | 0.033 |
| 12 | volume weighted | 0.249 | 0.035 | 0.246 | 0.033 |
| correlation int | correlation Ω | |
|---|---|---|
| 8 | 0.999807 [0.999797, 0.999817] | 0.999860 [0.999852, 0.999868] |
| 10 | 0.999781 [0.999765, 0.999798] | 0.999832 [0.999824, 0.999839] |
| 12 | 0.999878 [0.999869, 0.999887] | 0.999916 [0.999912, 0.999920] |
| measure | own-fit | frozen-transfer | |
|---|---|---|---|
| intersection | 10 | 0.008710, 0.008798 | 0.008779, 0.008728 |
| intersection | 12 | 0.006854, 0.007281 | 0.007292, 0.006830 |
| volume weighted | 10 | 0.005142, 0.005192 | 0.005181, 0.005154 |
| volume weighted | 12 | 0.010844, 0.011513 | 0.011488, 0.010848 |
| measure | quartic | increment [interval] | |
|---|---|---|---|
| intersection | 6 | 0.965829 | 0.030575 [0.026138, 0.035219] |
| intersection | 8 | 0.931000 | 0.001108 [0.000680, 0.001537] |
| intersection | 10 | 0.893332 | 0.037976 [0.034151, 0.042047] |
| intersection | 12 | 0.978915 | |
| volume weighted | 6 | 0.933604 | 0.059540 [0.055291, 0.064170] |
| volume weighted | 8 | 0.903110 | 0.003861 [0.002872, 0.004807] |
| measure | residual correlation [95% interval] | fitted-mode correlation |
|---|---|---|
| intersection | 0.8032 [0.7902, 0.8163] | 0.9831 |
| volume weighted | 0.7677 [0.7567, 0.7788] | 0.9786 |
| measure | source target | own | transfer | loss |
|---|---|---|---|---|
| intersection | 0.6619 | 0.6359 | 0.0260 | |
| intersection | 0.7799 | 0.7558 | 0.0240 | |
| volume weighted | 0.5556 | 0.5274 | 0.0282 | |
| volume weighted | 0.6725 | 0.6471 | 0.0253 |
| measure | pair | residual correlation [interval] | fitted-mode correlation |
|---|---|---|---|
| intersection | |||
| intersection | |||
| volume weighted | |||
| volume weighted |
| measure | teacher | cubic | quartic | cubic increment [interval] | quartic increment [interval] |
|---|---|---|---|---|---|
| intersection | 0.71831 | 0.88581 | 0.01081 [0.00819, 0.01352] | 0.04224 [0.03820, 0.04669] | |
| intersection | 0.77986 | 0.85164 | 0.00392 [0.00229, 0.00559] | 0.01010 [0.00774, 0.01248] | |
| intersection | 0.66192 | 0.71649 | 0.01070 [0.00817, 0.01305] | 0.02118 [0.01766, 0.02469] | |
| volume weighted | 0.63357 | 0.85697 | 0.00719 [0.00500, 0.00944] | 0.04414 [0.04005, 0.04806] | |
| volume weighted | 0.67248 | 0.76932 | 0.00394 [0.00184, 0.00626] | 0.01245 [0.00882, 0.01614] | |
| volume weighted | 0.55556 | 0.62505 | 0.01128 [0.00830, 0.01423] | 0.02455 [0.02015, 0.02901] |
| intersection | volume weighted | |||
| target | transfer | target fit | transfer | target fit |
| Cubic library | ||||
| 6 | 0.284584 | 0.958455 | 0.247642 | 0.921715 |
| 8 | 0.460289 | 0.893214 | 0.525306 | 0.865281 |
| 10 | 0.407344 | 0.729694 | 0.421562 | 0.653304 |
| 12 | 0.188180 | 0.710787 | 0.229902 | 0.624140 |
| intersection | volume weighted | ||||
|---|---|---|---|---|---|
| library | target | source | reference | source | reference |
| cubic | 6 | 0.284584 | 0.258403 | 0.247642 | 0.236810 |
| cubic | 8 | 0.460289 | 0.235909 | 0.525306 | 0.223270 |
| cubic | 10 | 0.407344 | 0.198709 | 0.421562 | 0.166863 |
| cubic | 12 | 0.188180 | 0.173060 | 0.229902 | 0.158067 |
| quartic | 6 | 0.272661 | 0.132967 | 0.247400 | 0.133531 |
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
| role | identifier |
|---|---|
| principal warm-start sequence | 20260916 |
| warm-start replication at | 314159 |
| warm-start replication at | 20260921 |
| 30,000-update independent-start control | 20261011 |
| continuation control | 20261001 |
| 90,000-update independent start | 20261031 |
| comparison | sampling | split |
|---|---|---|
| 30,000-update independent-start comparison | 20261014 | 20261015 |
| long-budget paired quality check | 20261101 | — |
| line-grouped principal/independent linkage | 20261201 | 20261202 |
| line-grouped degree/replication evaluation | 20261211 | 20261212 |
| measure | role | baseline | augmented | [interval] | |
|---|---|---|---|---|---|
| intersection | 6 | 0.958455 | 0.991222 | 0.032767 [0.028133, 0.037929] | |
| intersection | 8 | 0.893214 | 0.897770 | 0.004556 [0.003471, 0.005569] | |
| intersection | 8 | 0.892466 | 0.897004 | 0.004539 [0.003536, 0.005484] | |
| intersection | 10 | 0.729694 | 0.738403 | 0.008710 [0.006529, 0.010962] | |
| intersection | 10 | 0.728297 | 0.737096 | 0.008798 [0.006633, 0.011155] | |
| intersection | 12 | 0.710787 | 0.717641 | 0.006854 [0.004726, 0.008929] |
| measure | RMSE P | RMSE I | correlation | transfer | transfer |
|---|---|---|---|---|---|
| intersection | 0.000661 | 0.004717 | -0.085216 | -0.000471 | -0.000648 |
| volume weighted | 0.000561 | 0.004206 | -0.081823 | 0.004485 | 0.002669 |
Riemannian Geometry
Riemannian geometry studies curved spaces equipped with smoothly varying inner products, enabling intrinsic notions of distance, curvature, and optimization.
See all 45 papers