Explicit methods

Explicit Runge-Kutta tableaus scored end-to-end in Q15 at a fixed cycle budget.

The rk run ended on 2026-11-10, and no more compute goes to it. Epoch 2 is frozen and these pages are its final snapshot: the site is no longer rebuilt. The overview site sums up the final findings, and the harness is public for anyone to download and run.

Explicit Runge-Kutta methods build each stage from stages already computed, so a step is a fixed run of multiply-adds. For each order, stage count and cost, the grid keeps the verified tableau with the lowest held-out error, and that error, measured in Q15 at a fixed cycle budget, is the number each elite reports.

At a glance

family
Explicit fixed-step Runge-Kutta: a strictly lower triangular A and one b vector
search
MAP-Elites cells filled by exhaustive enumeration and CMA-ES islands
question
What error does a method reach inside a fixed cycle budget?
problems
The frozen search set and held-out set
arithmetic
Q15 through the pinned solver
checked by
The pinned verifier, order conditions included, then scoring on the held-out set
cost basis
The pinned cost model
sources
rk-work/archive, validation/results.json, benchmark/results.json
records
358,752
verified tableaus archived
elite cells
30/136
grid coverage, all orders; 31 cells hold an elite, 30 found by the search and 1 seeded classical, 1 of them outside the searched stage range
heldout_verified
10
top tier at insertion

Built from cycle 17,364 of the run, 358,752 archive records. Newest elite recorded 2026-10-23 23:10 CT, after the last saturation check at 2026-09-27 17:23 CT.

Cost against held-out error

Blue dots are archive elites and orange diamonds classical baselines: analytic cycles per step under m0plus_fast against held-out error at the fixed cycle budget, both log, so down-left is better. A diamond appears only when its identical tableau is archived; each dot links to its cell page, and the table under the chart carries every plotted value.
archive elite (links to its cell)classical baseline
1e01e11e21e31e-21e-11e0cycles per step (m0plus_fast, n_states=1)held-out errororder 1, 1 stage, bucket 0: 0.320268 held-out at 5 cycles (no_incumbent); ff5d6ff6768corder 2, 2 stages, bucket 0: 0.0705737 held-out at 14 cycles (heldout_verified); 892bd5b10e39order 2, 2 stages, bucket 1: 0.0530643 held-out at 16 cycles (no_improvement); 2d2c2b75d719order 2, 2 stages, bucket 2: 0.0560062 held-out at 40 cycles (heldout_verified); 9e7d5a6e28ddorder 2, 3 stages, bucket 0: 0.0779237 held-out at 13 cycles (no_incumbent); 6058c34391b1order 2, 3 stages, bucket 1: 0.0238649 held-out at 29 cycles (no_improvement); 312c38c22850order 2, 3 stages, bucket 2: 0.0435731 held-out at 60 cycles (no_improvement); ddfa9e5bdf2corder 2, 3 stages, bucket 3: 0.0456747 held-out at 64 cycles (no_improvement); 06bcc450bce4order 2, 4 stages, bucket 1: 0.0216412 held-out at 31 cycles (no_incumbent); 8b1c20657fe8order 2, 4 stages, bucket 2: 0.0158228 held-out at 45 cycles (no_improvement); 3c3c46da6ed5order 2, 4 stages, bucket 3: 0.0362323 held-out at 91 cycles (heldout_verified); 9fef8aebd015order 2, 5 stages, bucket 2: 0.0122299 held-out at 60 cycles (heldout_verified); 8677ac52b717order 2, 5 stages, bucket 3: 0.0135838 held-out at 64 cycles (no_improvement); ec6a1c6e2990order 2, 5 stages, bucket 4: 0.0260573 held-out at 138 cycles (heldout_verified); 6ede8aa89377order 2, 6 stages, bucket 2: 0.0129137 held-out at 58 cycles (no_improvement); 977d34ecabe5order 2, 6 stages, bucket 3: 0.0135838 held-out at 64 cycles (no_improvement); 4a62c0ce629forder 2, 6 stages, bucket 4: 0.0354374 held-out at 163 cycles (no_improvement); 1d4f752965adorder 3, 3 stages, bucket 2: 0.0410749 held-out at 61 cycles (no_improvement); 57af328f7be2order 3, 3 stages, bucket 3: 0.0392167 held-out at 68 cycles (heldout_verified); a4bfd55b3f91order 3, 4 stages, bucket 2: 0.0421996 held-out at 61 cycles (no_improvement); 32f5efc1ebe8order 3, 4 stages, bucket 3: 0.0281748 held-out at 105 cycles (no_improvement); cf7cbff1cb8aorder 3, 5 stages, bucket 2: 0.0631408 held-out at 63 cycles (heldout_verified); 62efd50c6ffforder 3, 5 stages, bucket 3: 0.0255085 held-out at 125 cycles (heldout_verified); 27688d49ea96order 3, 5 stages, bucket 4: 0.0251585 held-out at 156 cycles (heldout_verified); 5f506b0793cborder 3, 6 stages, bucket 3: 0.0253993 held-out at 100 cycles (no_improvement); 090b2e692801order 3, 6 stages, bucket 4: 0.0228645 held-out at 128 cycles (no_improvement); ba6c87de14bcorder 4, 4 stages, bucket 2: 0.0694907 held-out at 49 cycles (no_improvement); 016199da4761order 4, 5 stages, bucket 2: 0.060571 held-out at 63 cycles (no_improvement); 715e0970926eorder 4, 5 stages, bucket 3: 0.0371707 held-out at 88 cycles (no_improvement); 626d128171fborder 4, 6 stages, bucket 3: 0.0278667 held-out at 104 cycles (heldout_verified); b619e3e62d5eorder 4, 6 stages, bucket 4: 0.0366576 held-out at 129 cycles (no_incumbent); 0d9f381b85c3euler: 0.320268 held-out at 5 cycleseuler

Source: rk-work/archive, the elite in each cell when this page was built. Arithmetic: Q15 with floor rounding; cycles per step from the pinned m0plus_fast model.

Every plotted mark, in one table (32 rows)
markorderstagesbuckettableaucycles per stepheld-out error
archive eliteorder 11 stagebucket 0ff5d6ff6768c50.320268
archive eliteorder 22 stagesbucket 0892bd5b10e39140.0705737
archive eliteorder 22 stagesbucket 12d2c2b75d719160.0530643
archive eliteorder 22 stagesbucket 29e7d5a6e28dd400.0560062
archive eliteorder 23 stagesbucket 06058c34391b1130.0779237
archive eliteorder 23 stagesbucket 1312c38c22850290.0238649
archive eliteorder 23 stagesbucket 2ddfa9e5bdf2c600.0435731
archive eliteorder 23 stagesbucket 306bcc450bce4640.0456747
archive eliteorder 24 stagesbucket 18b1c20657fe8310.0216412
archive eliteorder 24 stagesbucket 23c3c46da6ed5450.0158228
archive eliteorder 24 stagesbucket 39fef8aebd015910.0362323
archive eliteorder 25 stagesbucket 28677ac52b717600.0122299
archive eliteorder 25 stagesbucket 3ec6a1c6e2990640.0135838
archive eliteorder 25 stagesbucket 46ede8aa893771380.0260573
archive eliteorder 26 stagesbucket 2977d34ecabe5580.0129137
archive eliteorder 26 stagesbucket 34a62c0ce629f640.0135838
archive eliteorder 26 stagesbucket 41d4f752965ad1630.0354374
archive eliteorder 33 stagesbucket 257af328f7be2610.0410749
archive eliteorder 33 stagesbucket 3a4bfd55b3f91680.0392167
archive eliteorder 34 stagesbucket 232f5efc1ebe8610.0421996
archive eliteorder 34 stagesbucket 3cf7cbff1cb8a1050.0281748
archive eliteorder 35 stagesbucket 262efd50c6fff630.0631408
archive eliteorder 35 stagesbucket 327688d49ea961250.0255085
archive eliteorder 35 stagesbucket 45f506b0793cb1560.0251585
archive eliteorder 36 stagesbucket 3090b2e6928011000.0253993
archive eliteorder 36 stagesbucket 4ba6c87de14bc1280.0228645
archive eliteorder 44 stagesbucket 2016199da4761490.0694907
archive eliteorder 45 stagesbucket 2715e0970926e630.060571
archive eliteorder 45 stagesbucket 3626d128171fb880.0371707
archive eliteorder 46 stagesbucket 3b619e3e62d5e1040.0278667
archive eliteorder 46 stagesbucket 40d9f381b85c31290.0366576
classical baselineeuler50.320268

Elite grids

One grid per order. Rows are stage counts, columns are m0plus_fast cycle buckets, and each filled cell prints its elite's held-out error. The further a cell's color is from the page background, the lower the error within that grid. Click a cell for the record, or open the table below the grids, which carries every occupied cell as a row.

Order 1

b0b1b2b3b4b5b6b7s=1s=2s=3s=4s=5s=6order 1, 1 stage, bucket 0: 0.320268 held-out, no_incumbent, ff5d6ff6768c0.32cycle bucket (m0plus_fast)

Order 2

b0b1b2b3b4b5b6b7s=2s=3s=4s=5s=6order 2, 2 stages, bucket 0: 0.0705737 held-out, heldout_verified, 892bd5b10e390.0706order 2, 2 stages, bucket 1: 0.0530643 held-out, no_improvement, 2d2c2b75d7190.0531order 2, 2 stages, bucket 2: 0.0560062 held-out, heldout_verified, 9e7d5a6e28dd0.056order 2, 3 stages, bucket 0: 0.0779237 held-out, no_incumbent, 6058c34391b10.0779order 2, 3 stages, bucket 1: 0.0238649 held-out, no_improvement, 312c38c228500.0239order 2, 3 stages, bucket 2: 0.0435731 held-out, no_improvement, ddfa9e5bdf2c0.0436order 2, 3 stages, bucket 3: 0.0456747 held-out, no_improvement, 06bcc450bce40.0457order 2, 4 stages, bucket 1: 0.0216412 held-out, no_incumbent, 8b1c20657fe80.0216order 2, 4 stages, bucket 2: 0.0158228 held-out, no_improvement, 3c3c46da6ed50.0158order 2, 4 stages, bucket 3: 0.0362323 held-out, heldout_verified, 9fef8aebd0150.0362order 2, 5 stages, bucket 2: 0.0122299 held-out, heldout_verified, 8677ac52b7170.0122order 2, 5 stages, bucket 3: 0.0135838 held-out, no_improvement, ec6a1c6e29900.0136order 2, 5 stages, bucket 4: 0.0260573 held-out, heldout_verified, 6ede8aa893770.0261order 2, 6 stages, bucket 2: 0.0129137 held-out, no_improvement, 977d34ecabe50.0129order 2, 6 stages, bucket 3: 0.0135838 held-out, no_improvement, 4a62c0ce629f0.0136order 2, 6 stages, bucket 4: 0.0354374 held-out, no_improvement, 1d4f752965ad0.0354cycle bucket (m0plus_fast)

Order 3

b0b1b2b3b4b5b6b7s=2s=3s=4s=5s=6order 3, 3 stages, bucket 2: 0.0410749 held-out, no_improvement, 57af328f7be20.0411order 3, 3 stages, bucket 3: 0.0392167 held-out, heldout_verified, a4bfd55b3f910.0392order 3, 4 stages, bucket 2: 0.0421996 held-out, no_improvement, 32f5efc1ebe80.0422order 3, 4 stages, bucket 3: 0.0281748 held-out, no_improvement, cf7cbff1cb8a0.0282order 3, 5 stages, bucket 2: 0.0631408 held-out, heldout_verified, 62efd50c6fff0.0631order 3, 5 stages, bucket 3: 0.0255085 held-out, heldout_verified, 27688d49ea960.0255order 3, 5 stages, bucket 4: 0.0251585 held-out, heldout_verified, 5f506b0793cb0.0252order 3, 6 stages, bucket 3: 0.0253993 held-out, no_improvement, 090b2e6928010.0254order 3, 6 stages, bucket 4: 0.0228645 held-out, no_improvement, ba6c87de14bc0.0229cycle bucket (m0plus_fast)

Order 4

b0b1b2b3b4b5b6b7s=2s=3s=4s=5s=6order 4, 4 stages, bucket 2: 0.0694907 held-out, no_improvement, 016199da47610.0695order 4, 5 stages, bucket 2: 0.060571 held-out, no_improvement, 715e0970926e0.0606order 4, 5 stages, bucket 3: 0.0371707 held-out, no_improvement, 626d128171fb0.0372order 4, 6 stages, bucket 3: 0.0278667 held-out, heldout_verified, b619e3e62d5e0.0279order 4, 6 stages, bucket 4: 0.0366576 held-out, no_incumbent, 0d9f381b85c30.0367cycle bucket (m0plus_fast)

Source: rk-work/archive, the elite in each cell when this page was built. Arithmetic: Q15 with floor rounding; cycles per step from the pinned m0plus_fast model.

How to read this

The archive is a MAP-Elites structure: one grid per algebraic order 1 to 4, whose cells are keyed by (stage count, cycle bucket). A cell holds at most one record, its elite: the verified tableau with the lowest held-out error seen so far for that shape and cost; ties keep the earlier record. An empty cell means no verified tableau has landed there yet.

Stage rows span 2 to 6 by default, plus a row for any other stage count that holds an elite (order 1's single-stage cell appears as s=1). The cycle buckets are log2-spaced bins of the m0plus_fast cycles per step: bucket 0 means fewer than 16 cycles, bucket 1 is 16 to 31, bucket 2 is 32 to 63, doubling each column, and bucket 7 collects everything at 1024 cycles or more.

Every elite in one table (31 rows)

One row per occupied cell, every order together so cells can be compared across orders. The hash links to the full record. A large gap between search_error and heldout_error is overfitting to the visible search set; measured_order and the tier badge are defined in the glossary. 1 of these rows is a seeded classical baseline rather than a search result; the label column says which.

orderstagesbuckettableau_hashtierheldout_errorsearch_errorcycles fastcycles slowmeasured_orderlabel
110ff5d6ff6768cno_incumbent0.3202680.01289550.984858seeded classical baseline
220892bd5b10e39heldout_verified0.07057370.010616814142.01094exhaustive: optimal within the enumerated space
2212d2c2b75d719no_improvement0.05306430.019240416162.01094exhaustive: optimal within the enumerated space
2229e7d5a6e28ddheldout_verified0.05600620.0048126240402.01094search result
2306058c34391b1no_incumbent0.07792370.017702613132.01094search result
231312c38c22850no_improvement0.02386490.0046923129292.01094search result
232ddfa9e5bdf2cno_improvement0.04357310.0123266601842.01997search result
23306bcc450bce4no_improvement0.04567470.0154514641262.01202search result
2418b1c20657fe8no_incumbent0.02164120.010252931312.01094search result
2423c3c46da6ed5no_improvement0.01582280.0053536245452.01094search result
2439fef8aebd015heldout_verified0.03623230.00337466911841.99285search result
2528677ac52b717heldout_verified0.01222990.0026494260602.01094search result
253ec6a1c6e2990no_improvement0.01358380.0058424564642.01094search result
2546ede8aa89377heldout_verified0.02605730.004441571383551.99631search result
262977d34ecabe5no_improvement0.01291370.0080878558582.01094search result
2634a62c0ce629fno_improvement0.01358380.0058424564642.01094search result
2641d4f752965adno_improvement0.03543740.008946251632251.97865search result
33257af328f7be2no_improvement0.04107490.015794861923.04041exhaustive: optimal within the enumerated space
333a4bfd55b3f91heldout_verified0.03921670.012137168683.04041exhaustive: optimal within the enumerated space
34232f5efc1ebe8no_improvement0.04219960.0059894761613.03441search result
343cf7cbff1cb8ano_improvement0.02817480.01411721051983.04079search result
35262efd50c6fffheldout_verified0.06314080.0039985663944.07063search result
35327688d49ea96heldout_verified0.02550850.009373251253113.00774search result
3545f506b0793cbheldout_verified0.02515850.003002341563113.02822search result
363090b2e692801no_improvement0.02539930.01025621001003.0246search result
364ba6c87de14bcno_improvement0.02286450.01228331281283.09694search result
442016199da4761no_improvement0.06949070.020185849494.07063search result
452715e0970926eno_improvement0.0605710.028044663634.07063search result
453626d128171fbno_improvement0.03717070.013035388884.07063search result
463b619e3e62d5eheldout_verified0.02786670.00212391041044.07063search result
4640d9f381b85c3no_incumbent0.03665760.003600651292224.08062search result

Scored under one verifier hash: 2db0816c621c44aa5b4b8efbadfde3a37497780460be138cd8af5439ab176611 (31 rows). Each record's own hash is on its detail page.

Beyond the search problems

rk-work/validation/results.json has not been written in this work directory, so the practical problems are not reported here.

Measured against real counterparts, at matched accuracy

rk-work/benchmark/results.json has not been written in this work directory, so the matched-accuracy comparison is not stated here. It appears once rk_harness.benchmark has been run against the current code; nothing is inferred in its place. The epoch 1 comparison, run under that epoch's own pin, is on the frozen epoch 1 site.