Experiments · E172b
Yes. At stricter settings the cell came out at +99.8 meV/atom, only 2.6 from before, so the miss is the model's.
In the log: result — prediction confirmed: E172's number stands; the edge failure is the
book/assets/diagrams/exp/E172b.svg).E172b result — prediction confirmed: E172's number stands; the edge failure is the
model's. The same 54-atom Hf₁₀Mo₁₅Nb₂Ta₅Ti₂V₉W₁Zr₁₀ cell at the settings standard (60/720
Ry, k 4×4×4, MV 0.02, references refs_v3_std at the same settings): E/atom −5757.35808 eV,
E_form = +99.8 meV/atom vs +97.2 at 50/400 — Δ 2.6. JOB DONE in 24 iterations. So
(i) the augmentation-cutoff caveat on the V-containing 50/400 cells is worth ≤ 3 meV/atom
even at 9 V atoms — the tag on E172/E177/E179 is now a number; (ii) v4's +12.2 against
+99.8 is a genuine extrapolation failure at the Hf/Zr-rich edge, ×8, with no settings
excuse left. This is the flywheel's first real target: the cell enters v4's training in the
first refit (still gated on ≥ 10 QE cells; 6 compositions exist).
E172b result — prediction confirmed: E172's number stands; the edge failure is the
model's. The same 54-atom Hf₁₀Mo₁₅Nb₂Ta₅Ti₂V₉W₁Zr₁₀ cell at the settings standard (60/720
Ry, k 4×4×4, MV 0.02, references refs_v3_std at the same settings): E/atom −5757.35808 eV,
E_form = +99.8 meV/atom vs +97.2 at 50/400 — Δ 2.6. JOB DONE in 24 iterations. So
(i) the augmentation-cutoff caveat on the V-containing 50/400 cells is worth ≤ 3 meV/atom
even at 9 V atoms — the tag on E172/E177/E179 is now a number; (ii) v4's +12.2 against
+99.8 is a genuine extrapolation failure at the Hf/Zr-rich edge, ×8, with no settings
excuse left. This is the flywheel's first real target: the cell enters v4's training in the
first refit (still gated on ≥ 10 QE cells; 6 compositions exist).
Built with PRISMWebsite and visualizations made using Claude