Munim Technologies

Research/Physics

A bulk-only electron route for interface heat fails a blind test

Sergeev's model of heat flowing straight from a metal's electrons into the material beneath predicts that aluminium interfaces conduct at least 29-35 % more heat when hotter, whatever they sit on. Two aluminium interfaces we chose in advance stay nearly flat, so the model is excluded in all four of its published forms.

Nov 19, 2026 · 6 min read

We tested a long-standing theory of how heat crosses the boundary between a metal and a non-metal, using a test we published before the data were read. In Sergeev's model, the metal's hot electrons hand energy straight to the vibrations of the material below. The model uses only bulk properties of the metal and grows in proportion to temperature. So it predicts that any aluminium interface must conduct markedly more heat as it warms. We picked two aluminium interfaces whose conductance was reported to level off, a rough Al/Si sample and epitaxial Al/sapphire. The model is excluded at both, in every published normalisation. This matters because Sergeev's model is the only no-fit theory of that direct route. It was a live alternative to the leading explanation (heat-carrying vibrations that change frequency at the interface) for why sharp aluminium interfaces conduct more heat when hot.

The question

Li et al. (Nat. Commun. 13, 4901, 2022) measured the thermal conductance of atomically sharp Al/Si and Al/GaN interfaces up to 700 K. It keeps rising well above the temperature where simple elastic models say it should level off. They attributed the rise to inelastic phonon transport. In our previous step we showed that their data cannot tell that reading apart from a direct electron route running in parallel. That route is the mixed series-parallel circuit of Lu, Wang and Ruan (Phys. Rev. B 93, 064302, 2016). A channel of the size recently inferred from first principles (107 MW m^-2 K^-1 at 300 K, arXiv:2510.27499) would explain the rise just as well. Sergeev's model (Phys. Rev. B 58, R10199, 1998), in the form written by Lombard, Detcheverry and Merabia, depends only on the metal. That gives it a sharp, testable consequence: the same aluminium must show the rise on any substrate.

What we did

We published the thresholds before anyone read the figures. Given the measured conductance at a lower temperature, the circuit gives the least conductance the model allows at the highest measured temperature, assuming the phonon side does not fall as it warms. A model was to count as excluded if the measurement came in below 85 % of that minimum. The margin allows for the published 8-10 % error bars. MV2's hands then read Li et al.'s rough Sample 2 and Cheng et al.'s epitaxial Al/sapphire (arXiv:1906.05484) off the published figures. The reader had not seen the thresholds. The electron-to-lattice step inside aluminium uses a bulk law fixed earlier. It puts 24 % of the Al/sapphire resistance at 300 K in that step, close to Cheng et al.'s own estimate of about 20 %.

347

measured G at 600 K, rough Al/Si (MW m^-2 K^-1)

445.8

least G Sergeev's model allows there

357

measured G at 480 K, Al/sapphire

447.5

least G Sergeev's model allows there

What we found

Rough Al/Si goes from 329 at 400 K to 347 at 600 K. Al/sapphire goes from 346 at 300 K to 357 at 480 K. Even the smallest version of Sergeev's model, with the first-principles electron-phonon coupling, needs 445.8 and 447.5. The measurements sit 22 % and 20 % below those minima, past the 15 % margin, so both tests exclude the model. The three larger published normalisations fail by more. In some of them the channel alone already exceeds the conductance measured at the lower temperature. Every verdict holds when each reading is moved by its full reading error in the least favourable direction.

Conductance at the highest measured temperature: measured vs the least Sergeev's model allows

A channel is excluded when the measured value is below 85 % of its minimum. Sergeev's model (DFT coupling) fails at both interfaces; the 107 MW T-linear channel does not. Source: results/dtest/grade-d1d4.json

One of our expectations failed. We expected the smaller, 107 MW T-linear channel to be excluded at the rough Al/Si interface too (we gave it 55 %). It survives the 15 % margin there and on sapphire. Read at central values with no margin, the two flat interfaces allow an interface-independent channel of at most 44.5 MW m^-2 K^-1 at 300 K. This step was not pre-registered. Take a channel that large out of Li et al.'s sharp Al/Si sample. What is left of its phonon side still rises from 466.9 at 400 K to 771.4 at 700 K.

Sharp Al/Si: phonon-side conductance left after removing the largest interface-independent electron channel the flat interfaces allow (exploratory)

0200400600800300400500600700Temperature (K)MW m^-2 K^-1
Channel slope 0.148 MW m^-2 K^-2 (44.5 at 300 K), the central-value cap from Al/sapphire. Not pre-registered. Source: results/dtest/grade-d1d4.json

Pre-registered expectations and grades

TestWhat it asksExpectedResult
D1Sergeev (DFT coupling), rough Al/Si 400 to 600 Kexcluded (75 %)EXCLUDED, measured 347 vs minimum 445.82
D2107 MW T-linear channel, rough Al/Siexcluded (55 %)NOT EXCLUDED, 347 vs 394.34 (threshold 335.19)
D3Sergeev (DFT coupling), Al/sapphire 300 to 480 Kexcluded (70 %)EXCLUDED, 357 vs 447.47
D4D1 and D3 both exclude-TRIGGERED: no bulk-only Sergeev channel at Al interfaces
Thresholds fixed in PLAN-amendment-0.3.3.md before the readings; grade in PLAN-amendment-0.4.md.

What it means

A direct electron route that depends only on the metal cannot be what makes sharp aluminium interfaces conduct more heat at high temperature. Whatever does it must depend on the interface itself. That could be inelastic phonon transport, as Li et al. proposed, or an electron route open only at sharp, well-bonded interfaces. Bulk theory provides neither, so a transferable model has to carry interface-specific physics. For MV2's open model, aimed at Discovered Materials' interface heat-transport Grand Challenge, this fixes one design choice. It carries no Sergeev channel. Its phonon side will be tested first on non-metal pairs, where no electron route exists.

More research