A_d at half filling (δ = 0, Ne = 36) looks much larger than the 4×4 references

#56
by Yuseon84 - opened

A small observation about one entry of scoring_meta.json, in case it is useful.

The published A_d at δ = 0 is 0.0384, about two-thirds of the dome maximum (0.0585). The references we know of
point to a much smaller value at half filling (U/t = 8):

source lattice P_d at n = 1
ED, validation/comparison_4x4_ED_vs_VMC.md (Ku-ku) 4×4, R = (2,2) 0.0062
organizer VMC, same file (dwave=0.15) 4×4, R = (2,2) 0.0008
U = 0 (exact) any 0
our VMC with a Néel-field ansatz (energy within 5 % of ED on 4×4, where it gives 0.0075) 6×6, R = (3,3) 0.0049 ± 0.0007

So the 6×6 entry is at least about 5× larger than any of these.

Why it may matter: 101 of the current active materials sit at δ = 0 and all use this single value
(three of them are in the top 50, e.g. Co₂Se₂ and Ni₂Se₂). The leaders are not affected.

Could you check the δ = 0 point of the reference solver, for example whether the antiferromagnetic
correlations are captured at half filling? One of our own proposals sits at δ = 0 (Ni₂Se₂ from #14), so a lower value
would lower it too.

Disclosure: prepared with AI assistance (Claude Code); I am not a many-body specialist. Scripts and
outputs for the numbers above are available on request.

FINAL_Bench org

Good catch, and it connects to #62. The scoring_meta.json A_d is the d-wave-biased trial-state (VMC) proxy, not a true-ground-state value, so at delta=0 it sits well above small-lattice ED/VMC references and above the exact zero that #62 just proved for the 6x6 two-carrier ground state. This affects only the delta=0 point (your listed entries, incl. Co2Se2/Ni2Se2); leaders are unaffected. We are reviewing the delta=0 calibration against the 6x6 references you cite; any recalibration will be posted to #34/#49 and applied uniformly to all delta=0 ids. Please do send your scripts/outputs.

Thank you. The scripts and outputs are in PR #69 under methods/Yuseon84/.

run_delta0.sh reproduces the two δ = 0 values: 4×4 ED at n = 1 (P_d = 0.006197, the same as the Ku-ku value in
validation/) and the 6×6 Néel-field VMC at Ne = 36 (P_d = 0.0049 ± 0.0007, three seeds). The package also holds the
full 4×4 ED set (Ne = 8–16, per momentum sector), the U = 0 references and our 6×6 VMC table for δ ≤ 0.333, in case
they help the wider calibration. Glad to answer questions in this thread.

Prepared with AI assistance (Claude Code).

FINAL_Bench org

Thanks - #69 package received. The delta=0 A_d point is under review against your 4x4 ED and 6x6 Neel-VMC references; any recalibration will be posted here and to #34 and applied uniformly to all delta=0 ids. No change yet.

Sign up or log in to comment