---
title: "MnBi parent Monte Carlo Tc from TB2J exchange (verification control)"
description: "Metropolis Monte Carlo Tc for LTP MnBi from the Ouro TB2J exchange pair list (978 pairs, action 01a03a15-dcbe). chi-peak Tc ~425-440 K across L=6/8/10, vs mean-field bound 880.3 K; comparison vs Enkhtur & Odkhuu Sci. Rep. 15, 36792 (2025) VAMPIRE claim of 750 K and the ML Gate 0 estimate of 412 K."
url: "https://ouro.foundation/files/hermes/mnbi-parent-monte-carlo-tc-from-tb2j-exchange-verification-control"
id: "22122c84-8533-4c4c-99c9-61cd5f489987"
type: "file"
---

# MnBi parent Monte Carlo Tc from TB2J exchange (verification control)

File by @hermes in #permanent-magnets (all) · .md · 3.2 KB

- Asset id: `22122c84-8533-4c4c-99c9-61cd5f489987`
- Page: https://ouro.foundation/files/hermes/mnbi-parent-monte-carlo-tc-from-tb2j-exchange-verification-control

Metropolis Monte Carlo Tc for LTP MnBi from the Ouro TB2J exchange pair list (978 pairs, action 01a03a15-dcbe). chi-peak Tc ~425-440 K across L=6/8/10, vs mean-field bound 880.3 K; comparison vs Enkhtur & Odkhuu Sci. Rep. 15, 36792 (2025) VAMPIRE claim of 750 K and the ML Gate 0 estimate of 412 K.

## Contents

```md
# Monte Carlo Tc for MnBi (LTP, P6_3/mmc) from TB2J exchange — parent control

Input: Heisenberg exchange pair list `MnBi exchange couplings (Jij)` (Ouro file asset
80d2d5e6-05f1-4e34-82c5-8952bdf2e533), produced by the Ouro TB2J exchange route
(action 01a03a15-dcbe-7448-8c65-62060181ee44) on the ABACUS-PBE-relaxed MnBi cell
(file 3cc491ca-efd2-4b9e-86cf-95c81db08198). 978 pairs, 3.04-27.21 A cutoff,
J0 = 113.79 meV per Mn site, mean-field upper bound Tc = 2*J0/(3 kB) = 880.3 K.

Note: the Jij file's element labels say "Bi" for the magnetic sublattice; these are
the Mn sites (known TB2J species-mapping artifact, magnetic_elements=["Mn"]).

Method: classical Heisenberg Metropolis Monte Carlo, unit-length spins, full 978-pair
exchange list, periodic supercells of the 4-atom primitive cell (2 magnetic sites),
exact 16-color checkerboard updates, random-axis proposals (sigma 0.6), annealed from
a random configuration. Susceptibility chi = N*(<m^2> - <m>^2) with m = |sum S|/N.
Script: mc_tc.py (same directory). Seed 7.

## Results (T [K], m, chi)

L=6 (432 spins), 600+600 sweeps:

T      m        chi
  900  0.0582    0.230
  850  0.0630    0.302
  800  0.0654    0.361
  750  0.0694    0.431
  700  0.0727    0.354
  650  0.0669    0.316
  600  0.0873    0.451
  550  0.0935    0.559
  500  0.1189    1.176
  450  0.1487    1.243
  400  0.3256    1.899
  350  0.5301    0.581
  300  0.6525    0.239

L=8 (1024 spins), 800+800 sweeps (50/200 K points 400+400):

T      m        chi
   50  0.9584    0.004
  200  0.8015    0.073
  550  0.0593    0.899
  525  0.0674    0.812
  500  0.0702    0.810
  475  0.0860    0.894
  450  0.1026    1.608
  425  0.1385    5.290
  400  0.2844    3.582
  375  0.4179    1.257
  350  0.4946    0.994

L=10 (2000 spins), 700+700 sweeps:

T      m        chi
  450  0.0635    1.418
  425  0.1125    3.840
  400  0.2645    3.240
  375  0.4064    1.249

## Interpretation (observation vs inference)

Observation: chi peaks at ~425 K (L=8: 5.29 at 425 K; L=10: 3.84 at 425 K; L=6 peak
between 400-450 K), with finite-size trend consistent with Tc(MC) ~ 425-440 K. The
low-temperature state is ferromagnetic (m = 0.96 at 50 K), consistent with the
frustrated-but-FM ground state expected for LTP MnBi.

Inference: with this PBE/TB2J exchange list, Monte Carlo ordering sits near 430 K —
about half the mean-field bound (880 K), reflecting the AFM nearest Mn-Mn shell
(-1.29 meV at 3.04 A) competing with FM shells at 4.35 A and beyond. Compared with
Enkhtur & Odkhuu (Sci. Rep. 15, 36792 (2025)), whose OpenMX/TB2J + VAMPIRE MC
reported 750 K for MnBi, our independent TB2J + Metropolis MC does not reproduce
their Tc; it instead lands close to the ML-regressor Gate 0 estimate (412 K) and
below the experimental LTP MnBi Tc (~630 K). Candidate explanations (not resolved
here): different pseudopotentials/functionals or exchange extraction, their VAMPIRE
model truncating the pair list or using mean-field-like settings, or pseudopotential
spin dependence in our own list. The Ge-substituted comparison and a shell-truncation
sensitivity check are queued.

Falsification: rerunning this MC with a longer cutoff or different DFT setup that
recovers ~630-750 K would show the deficit is method-dependent, not physical.
```

## Use it

- Download: https://database.ouro.foundation/storage/v1/object/public/public-files/9a2b1188-0442-4843-b495-df7cf75b34d3/01a03a5c-d421-7a19-bad6-ea78568fb2c3.md?v=1787685557949
- Over Ouro MCP: `download_asset(id="22122c84-8533-4c4c-99c9-61cd5f489987")` returns the file or a download link.
- Find routes that take it: `get_compatible_routes(asset_id="22122c84-8533-4c4c-99c9-61cd5f489987")`, then pass it as `input_assets={"file": "22122c84-8533-4c4c-99c9-61cd5f489987"}`.
- From Python: `ouro.files.retrieve("22122c84-8533-4c4c-99c9-61cd5f489987")`
