Agent-Guided Discovery of Exchange-Correlation Functionals with Physics-based Validation
Organizations: Mechanical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA · Biomedical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA · Chemical Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA · Materials Science and Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA ·
Abstract
Accurate, compact exchange--correlation functionals are central to predictive density functional theory. Symbolic search and language-model agents can propose new functional forms, but it remains unclear whether their gains survive self-consistency and external evaluation, whether the new terms carry physical content, whether agents outperform evolutionary search, and how language-model backbones compare. We connect frontier LLM agents to a GPU evaluator through the Model Context Protocol, with training-only fitting, persistent memory, and self-consistent and external evaluation. Opus-5.5 agents passed the screening gate with about 40-fold fewer tokens than Codex (GPT-6-Astra) agents and found a candidate with lower validation and held-out errors; only this candidate advanced to self-consistent evaluation. The resulting 13-parameter functional, O159, multiplies same-spin correlation by a smooth density window. Refitted controls favor a window that scales the entire same-spin factor, reducing it by 15--30% at its peak, and that acts on reaction energies mainly in near-iso-orbital regions; against a refitted no-window control, it lowers heavy-atom total-energy errors on GSCDB137 by 29%. With about 400 fits each, SyFES evolution did not improve on its GAS22 seed, whereas the agents found forms 0.10--0.13 \kcal\ lower in validation error by adding an explicit density scale. Self-consistent O159 lowers the held-out MGCDB84 test error of B97M-V by 11.7%, and its mean normalized error ratio on 36 GSCDB137 datasets (1.139) is comparable to B97M-V (1.144) and below GAS22 (1.190). Frontier-LLM agents thus produced O159, a compact functional whose improvement over B97M-V survives self-consistent, physics-based validation.
Figures & tables
| Search | Fits | Seed (val.) | Best new form | Training | Validation | GPU-h | Tokens (M) |
|---|---|---|---|---|---|---|---|
| SyFES evolution | 400 | GAS22 (3.462) | SyFES-174 | 3.444 | 3.506 | 61.1 | 0 |
| Opus agents | 397 | refits a | O159 | 3.252 | 3.332 | 11.1 | 251 |
| O396 b | 3.140 | 3.138 | |||||
| Codex agents | 400 c | refits a | C356 | 3.207 | 3.363 | 8.6 | 521 |
| Factor | Favors |
|---|---|
| Explicit density variable (or ) available; SyFES instructions act only on the scale-invariant and , from which and follow | agents |
| Prompt listed -dependent correlation and -based self-correlation corrections among example directions; reference sheet listed prior forms, including the multiplicative same- and opposite-spin corrections of SAFS26-b; second Opus campaign given window-related island themes | agents |
| Prompts quoted published validation and test errors of reference functionals and set beating GAS22 and SAFS26-a as a goal; no candidate test value was available during the first Opus campaign | agents |
| Category-resolved validation errors instead of a scalar fitness; warm starts from parent coefficients; screening penalty in the score | agents |
| Minimum taken on validation over repeated fits of the same form | agents |
| Seed is the validation minimum of the original 29,628-form search and keeps its published parameters, whereas children receive only a training-set CMA-ES fit | SyFES seed |
| Functional | Mode | MGCDB84 val. b | MGCDB84 test b | GSCDB NER c |
|---|---|---|---|---|
| B97M-V | SCF | 3.533 | 4.309 | 1.144 |
| GAS22 | non-SCF | 3.462 | 3.599 | 1.212 |
| GAS22 | SCF | 3.470 | 3.589 | 1.190 |
| O159 | non-SCF | 3.332 | 3.818 | 1.141 |
| O159 | SCF | 3.330 | 3.805 | 1.139 |
| O396 a | non-SCF | 3.138 | 3.841 | 1.147 |
| Datasets | O178 | O159 | B97M-V | 95% CI | |||
|---|---|---|---|---|---|---|---|
| New (primary) | 20 | 1.312 | 1.270 | 1.290 | 0.55 | ||
| New, without AE11 | 19 | 1.193 | 1.204 | 1.191 | 0.80 | ||
| With MGCDB84 antecedents | 16 | 1.036 | 0.979 | 0.961 | 0.018 | ||
| All | 36 | 1.189 | 1.141 | 1.144 | 0.041 | ||
| All, without AE11 | 35 | 1.122 | 1.101 | 1.086 | 0.067 |
| Parameter | Value | Parameter | Value |
|---|---|---|---|
| 0.24549733 | 1.14502582 | ||
| 1.00304439 | 3.75266635 | ||
| -4.69984961 | -3.35815548 | ||
| -4.88244588 | -10.59006317 | ||
| -1.96642642 | 10.37863864 | ||
| 7.80561285 | -0.25463509 |
| Functional | Mode | train | val. | test | NER 36 | NER 20 |
|---|---|---|---|---|---|---|
| O159 | non-SCF | |||||
| O159 | SCF | |||||
| GAS22 | non-SCF | |||||
| GAS22 | SCF | |||||
| O396 | non-SCF | |||||
| C356 | non-SCF | — | — |
| Record (parent) | Change relative to parent | Params | Train | Val. | |
| O145 (O143) | reference architecture, window | 13 | 3.226 | 3.333 | — |
| O159 (O145) | 13 | 3.252 | 3.332 | ||
| O178 (O145) | window removed | 11 | 3.386 | 3.481 | |
| O185 (O178) | analogous window ( , ) on opposite-spin channel only | 13 | 3.172 | 3.682 | |
| O186 (O178) | one shared window on both channels | 13 | 3.181 | 3.569 | |
| O114 (O92) | total-density instead of | 16 | 3.199 | 3.453 |
| Change | Record (parent) | Inference | |
|---|---|---|---|
| Robust ( kcal mol -1 ) | |||
| Window removed | O178 (O145) | 0.148 of O145’s 0.200 gain over B97M-V | |
| Analogous window ( , ) on opposite-spin channel only | O185 (O178) | specific to same-spin correlation | |
| Window on the uniform-gas same-spin term only ( B97M-V base) | O136 (O125) | must scale the whole same-spin factor | |
| Gate (off in one-orbital regions) | O166 (O152) | needed where is large | |
| Gradient-corrected radius | O189 (O152) | shifting the window to lower density at large gradient is harmful | |
| Split, | Category contributions to | Subsets, O178 O159 |
|---|---|---|
| Validation, ( ) | BH , RG10 , TCE , ID , NCEC , TCD , IE , NCD , NCED | improved: Shields38 0.49 0.35, HTBH38 1.71 1.49, HAT707nonMR 2.67 2.59, WATER27 0.49 0.42, ISOMERIZATION20 2.32 2.18, CR20 0.81 0.57; worsened: SW49Bind6 0.55 0.70, SW49Bind345 0.25 0.32, EA13 1.78 1.92, BzDC215, S66x8 |
| Training, ( ) | AE18 , ID , BH , NCED , NCD , TCE , NCEC , IE | improved: AE18 7.60 5.56, Styrene45 2.03 1.89, BHPERI26 1.24 1.02, DBH24 1.58 1.46; worsened: CRBH20 1.02 1.37, G21EA 2.48 2.66 |
| Test, ( ) | NCEC , IE , BH , TCE , TCD , NCED , ID | improved: H2O20Bind4 1.26 0.69, WCPT27 1.86 1.70, PX13 1.92 1.82, YMPJ519, H2O20Bind10; worsened: C20C24 5.39 5.72 (8 reactions), Ionic43 0.61 0.67, BSR36 |
| Atom | O178 | O159 | Atom | O178 | O159 | ||
|---|---|---|---|---|---|---|---|
| H | 2.25 | 2.98 | Ne | 3.54 | |||
| He | 0.94 | 1.65 | Na | 3.69 | |||
| Li | 0.62 | 2.51 | Mg | 4.54 | |||
| Be | 4.22 | Al | 5.62 | ||||
| B | 4.81 | 0.66 | Si | 6.81 | |||
| C | 1.54 | 1.72 | P | 5.48 | 4.48 | 8.22 |
| Split | Both | Electrons in both | |||||
|---|---|---|---|---|---|---|---|
| Training | 3.252 | 5.111 | 3.386 | — | — | — | — |
| Validation | 3.332 | 5.595 | 3.481 | 87% | 82% | 82% | 4.9% |
| Test | 3.818 | 7.188 | 3.865 | 88% | 91% | 91% | 5.4% |
| Functional | (H) |
|---|---|
| B97M-V (published) | |
| GAS22 (published) | |
| O178 (refitted, no window) | |
| O145 | |
| O159 | |
| O159, window removed at fixed coefficients |
| Dataset | Type | B97M-V | O159 | NER B97M-V | NER O159 |
|---|---|---|---|---|---|
| 3d4dIPSS | TM | 5.073 | 5.613 | 0.977 | 1.081 |
| AE11 | TC | 368.86 | 295.66 | 3.154 | 2.528 |
| ALKBDE10 | TC | 3.581 | 3.561 | 1.100 | 1.094 |
| DAPD | TM | 6.264 | 6.625 | 2.256 | 2.386 |
| Dip146 | EF | 0.0365 | 0.0391 | 1.051 | 1.125 |
| DIPCS9 | TC | 5.106 | 4.189 | 1.510 | 1.238 |
| Dataset | Type | O178 | O159 | NER O178 | NER O159 | NER B97M-V | Prediction | |
| Datasets without MGCDB84 antecedents | ||||||||
| 3d4dIPSS | TM | 5.717 | 5.616 | 1.101 | 1.081 | 0.977 | — | |
| AE11 | TC | 415.8 | 295.8 | 3.555 | 2.529 | 3.154 | better (l) ✓ | |
| ALKBDE10 | TC | 3.703 | 3.568 | 1.138 | 1.096 | 1.100 | — | |
| DAPD | TM | 7.054 | 6.632 | 2.540 | 2.388 | 2.256 | — | |
| Dip146 | EF | 0.0375 | 0.0392 | 1.079 | 1.128 | 1.051 | — | |
| Functional | Mode | Train | Validation | Test | VT ¬RG10 |
|---|---|---|---|---|---|
| B97M-V | SCF | 3.319 | 3.533 | 4.309 | 3.867 |
| GAS22 | non-SCF | 3.270 | 3.462 | 3.599 | 3.657 |
| GAS22 | SCF | 3.264 | 3.470 | 3.589 | 3.664 |
| O159 | non-SCF | 3.252 | 3.332 | 3.818 | 3.613 |
| O159 | SCF | 3.253 | 3.330 | 3.805 | 3.607 |
| Candidate | Parameters | Training | Validation | Test |
|---|---|---|---|---|
| O159 | 13 | 3.252 | 3.332 | 3.818 |
| O396 | 10 | 3.140 | 3.138 | 3.841 |
| O388 | 9 | 3.191 | 3.236 | 3.906 |
| O376 | 12 | 3.230 | 3.294 | 3.961 |
| C356 | 14 | 3.207 | 3.363 | 3.897 |
| GAS22 refit (Codex) | 12 | 3.266 | 3.489 | 3.628 |
| Functional | Training | Validation | Test | |
|---|---|---|---|---|
| A. SAFS26 paper: reported non-SCF results | ||||
| B97M-V | 3.32 | 4.02 | 3.64 | 4.07 |
| GAS22 | 3.27 | 3.74 | 3.19 | 3.77 |
| SAFS26-x | 2.91 | 3.95 | 3.49 | 3.81 |
| SAFS26-a | 2.76 | 3.65 | 3.41 | 3.70 |
| SAFS26-b | 2.75 | 3.78 | 3.44 | 3.83 |
| Functional/mode | All 36 | New 20 | Without AE11 (35) |
|---|---|---|---|
| B97M-V published | 1.135 | 1.285 | 1.077 |
| B97M-V local SCF | 1.144 | 1.290 | 1.086 |
| O159 non-SCF | 1.141 | 1.270 | 1.101 |
| O159 SCF | 1.139 | 1.269 | 1.099 |
| O396 non-SCF | 1.147 | 1.316 | 1.110 |
| GAS22 non-SCF | 1.212 | 1.344 | 1.227 |
| Form | Params | Train | Validation | Violations | Failed test |
|---|---|---|---|---|---|
| GAS22 (seed, published) | 14 | 3.2702 | 3.4623 | 1 | uniform-gas exchange |
| SyFES-174 | 14 | 3.4441 | 3.5057 | 0 | — |
| SyFES-55 | 14 | 3.2690 | 3.5542 | 1 | uniform-gas exchange |
| SyFES-699 | 14 | 3.2072 | 3.5554 | 1 | uniform-gas exchange |
| SyFES-1244 | 14 | 3.1896 | 3.5655 | 1 | uniform-gas exchange |
| SyFES-30 | 14 | 3.1934 | 3.5683 | 1 | uniform-gas exchange |
| Scope | Records | Input | Output | Cache read | Cache write | Total |
|---|---|---|---|---|---|---|
| Opus | 397 | 0.002 | 1.500 | 244.909 | 5.010 | 251.421 |
| Opus, omit ID 201 | 396 | 0.002 | 1.285 | 200.864 | 3.743 | 205.893 |
| Codex | 401 | 7.852 | 1.028 | 512.494 | 0 | 521.374 |
| Superseded Codex | 200 | 0.829 | 0.128 | 31.136 | 0 | 32.094 |