✅ Verification: H3S — Paper vs Simulation [2026-05-26]

The Paper's Central Claim

In 2015, something remarkable happened in a diamond anvil cell — a tiny chamber designed to crush materials under pressures rivaling those found near Earth's core. A team led by Mikhail Eremets at the Max Planck Institute squeezed hydrogen sulfide (H₃S) to approximately 200 gigapascals — about two million times atmospheric pressure — and watched its electrical resistance drop to zero. The critical temperature? Above 200 kelvin, or roughly -70°C. That might not sound warm, but for superconductor physics, it was seismic.

As noted by physicist Inna Vishik on X (formerly Twitter), this result represented "the first major empirical success" in the modern hunt for high-temperature superconductivity in hydrogen-rich compounds — so-called hydride superconductors. For decades, theorists had predicted that hydrogen, if coaxed into the right crystal structure under extreme pressure, could superconduct at temperatures far above those achieved by cuprates or iron-based superconductors. H₃S was the proof of concept. It told the community: the theory works, the materials are real, and the ceiling might be much higher than we thought.

The significance extends beyond the number itself. H₃S validated a theoretical framework — BCS-Migdal-Eliashberg theory applied to hydrogen-dominant lattices — that has since guided the discovery of additional hydride superconductors, including LaH₁₀ with a reported Tc near 250K. It opened a door. The question now is whether that door leads anywhere we can actually live — namely, ambient pressure.

How Our Simulation Approaches This

At AI Future Lab, we run AI-augmented computational analyses that combine machine-learned interatomic potentials, neural network-accelerated phonon calculations, and Eliashberg spectral function estimation to predict superconducting properties of materials. We want to be transparent about what this means — and what it doesn't.

Our approach is not a replacement for full ab initio density functional theory (DFT) calculations or, certainly, for experimental measurement. State-of-the-art DFT studies of H₃S, such as those by Duan et al. (2014) and Errea et al. (2015), solve the quantum mechanical equations of electron-phonon coupling with painstaking precision. Experimental groups physically synthesize samples and measure resistance and magnetic susceptibility under extreme conditions. We do neither of those things directly.

What we do is build surrogate models trained on existing DFT datasets and experimental results for hydride systems, then use those models to rapidly estimate key superconducting parameters — Tc, electron-phonon coupling strength (λ), stability regime, and the dominant pairing mechanism. Think of it as a fast, informed interpolation engine. It's powerful for cross-checking claims, identifying outliers, and flagging results that deviate from established physics. But it inherits the biases and boundaries of its training data, and it cannot capture genuinely novel physics that lies outside the distribution it has learned.

For a well-studied system like H₃S, our model operates in a regime where it should perform well. This is familiar territory. That said, we report confidence intervals, not certainties.

What Our Analysis Found

Our simulation predicts a critical temperature of 203K for H₃S, with the superconducting state emerging at a pressure of approximately 150 GPa. The electron-phonon coupling constant λ is estimated at 2.19, placing H₃S firmly in the strong-coupling regime — well beyond the λ ≈ 0.5–1.0 range typical of conventional metallic superconductors.

The predicted mechanism aligns with the established theoretical picture: strong electron-phonon coupling driven by high-frequency hydrogen vibrational modes — specifically stretching and bending modes — within the Im̄3m body-centered cubic crystal structure. In this phase, sulfur atoms sit on the BCC lattice sites and provide a substantial electronic density of states at the Fermi level, while the hydrogen sublattice contributes the dominant phonon modes that mediate Cooper pairing. The framework is conventional BCS-Migdal-Eliashberg theory, and our model identifies no need to invoke unconventional pairing symmetry or exotic mechanisms.

Stability classification: metastable. This is an important nuance. H₃S in the Im̄3m phase is not the ground state at all pressures; it becomes dynamically stable only above a threshold pressure and may decompose upon decompression. Our model flags it as metastable rather than thermodynamically stable, consistent with the experimental reality that these phases exist only under sustained high pressure.

Overall confidence: high. This is a system where our training data is dense and the underlying physics is well-characterized.

✅ Strong Agreement: Reading the Gap

The headline comparison: the paper reports Tc above 200K at ~200 GPa; we predict 203K at ~150 GPa. The Tc values are in excellent agreement — within the noise floor of both experimental measurement and computational estimation. The pressure, however, shows a meaningful gap of roughly 50 GPa.

Why the pressure discrepancy? Several factors likely contribute:

• Idealized crystal structure. Our simulation assumes a pristine Im̄3m phase. Real samples contain grain boundaries, defects, residual H₂S or elemental sulfur, and possible mixed phases. These imperfections can shift the pressure at which the optimal superconducting phase fully stabilizes.
• Pressure calibration. Measuring pressure inside a diamond anvil cell at 200 GPa is itself a nontrivial challenge. Ruby fluorescence scales and equation-of-state standards carry uncertainties of several GPa, and pressure gradients across the sample chamber are common. The "~200 GPa" figure in the experimental literature is an approximation.
• Anharmonic effects. Hydrogen is light. Its zero-point motion is enormous. Full quantum treatment of anharmonicity, as performed by Errea et al. using stochastic self-consistent harmonic approximation (SSCHA), shifts the stability window and Tc in ways that our surrogate model captures imperfectly. We likely underestimate the pressure required to stabilize the phase because our phonon model doesn't fully account for these quantum nuclear effects.
• Training data bias. Our model is trained partly on DFT results that themselves vary depending on the exchange-correlation functional used. PBE, PBEsol, and hybrid functionals give different equilibrium volumes and thus different pressure scales.

The λ value of 2.19 aligns well with published DFT estimates, which range from approximately 1.8 to 2.5 depending on methodology. The mechanism and crystal structure assignment are consistent with the literature consensus. We classify this as strong agreement, with the pressure offset representing a systematic rather than fundamental discrepancy.

What This Tells Us About Room-Temperature Superconductivity

H₃S is often cited as the beginning of the modern hydride superconductivity story, and for good reason. It demonstrated that phonon-mediated superconductivity can reach temperatures that were, until recently, considered the exclusive domain of unconventional (and poorly understood) mechanisms. It also demonstrated something sobering: you need extraordinary pressure to get there.

200 GPa is not a condition you encounter in daily life, or in any practical device. The ongoing challenge — the reason this field generates both excitement and controversy — is whether the same physics can operate at ambient or near-ambient pressure. For that to happen, you would need a material that satisfies several simultaneous constraints: a high hydrogen phonon frequency (requiring light atoms in stiff bonds), a large electronic density of states at the Fermi level (requiring favorable band structure), strong electron-phonon matrix elements (requiring the right orbital character and symmetry), and thermodynamic or at least kinetic stability near 1 atmosphere.

No known material satisfies all of these constraints simultaneously at ambient pressure. Claims to the contrary — most notably the LK-99 episode in 2023 and disputed nitrogen-doped lutetium hydride results — have not survived independent replication. Reproducibility in high-pressure superconductor research is notoriously difficult: samples are microscopic, pressure environments are heterogeneous, and resistance measurements can be confounded by non-superconducting artifacts. The field has learned, sometimes painfully, that extraordinary claims require not just extraordinary evidence but extraordinary reproducibility.

H₃S remains a benchmark precisely because it has been reproduced by multiple groups and because the theory-experiment agreement is robust. It is the solid ground from which more ambitious searches depart.

Our Evolving Simulation

The 50 GPa pressure gap in our H₃S analysis is a useful calibration point. It tells us where our model's approximations bite hardest — specifically, in the treatment of quantum anharmonicity and pressure-dependent phase stability. As we incorporate more training data from SSCHA calculations and from emerging machine-learned potential energy surfaces that explicitly handle nuclear quantum effects, we expect this gap to narrow.

We are also expanding our validation set beyond the "canonical" hydrides (H₃S, LaH₁₀, YH₆) to include ternary and quaternary hydrides where experimental data is sparser and our model's predictions become genuine forecasts rather than retrospective checks. That is where the real test lies.

For now, the H₃S result gives us confidence that our framework captures the essential physics of phonon-mediated hydride superconductivity in well-characterized systems. The gap today — 50 GPa in pressure, negligible in Tc — is honest, explainable, and, we believe, closable. We will keep reporting both our agreements and our misses with equal transparency. That is the only way computational tools earn trust in a field where trust is hard-won and easily lost.