Solid State Fusion & Computational Physics

I — The reply that wasn’t an experiment

When the cold-fusion claim broke in the spring of 1989, the most damaging responses did not come from anyone trying and failing to boil water in a beaker. They came from people with pencils. Within weeks, Steven Koonin and Michael Nauenberg computed the rate at which two deuterons would fuse inside an ordinary hydrogen-isotope molecule and found it preposterously small. To reproduce the heat being reported, they noted, you would need the two nuclei screened by a particle roughly ten times the mass of an electron, and palladium contains no such thing.² A few months later Anthony Leggett and Gordon Baym went further. They derived an upper bound on how fast deuterons could tunnel together inside a metal, a bound that holds for any fully interacting many-body system in equilibrium, almost regardless of the model you choose. The bound sat orders of magnitude below the rate the claimed heat demanded.³

This is worth pausing on, because it is the part of the story that gets flattened. The field was not dismissed by a sneer. It was dismissed by a calculation, and the calculation was good. Tunneling through the Coulomb barrier is suppressed by the Gamow factor, which falls off exponentially with the area under the barrier, and for two deuterons at room temperature that suppression is total for all practical purposes.¹ Nothing in the decades since has overturned that arithmetic.

But look at what both calculations assumed. Koonin and Nauenberg put two nuclei in a static molecule. Leggett and Baym took the lattice in equilibrium. Those are the honest, conservative choices a theorist makes when the burden of proof sits with the claimant, and in 1989 it did. They are also, precisely, the choices that leave one term of the problem blank: a metal that is loaded past its comfortable limit and driven hard, with current running through it, vacancies multiplying, and the deuterons anything but at rest. Whether the bound survives that case has not, in the verifiable literature, been worked out either way.⁹ That blank is a computational-physics problem. This article is about why it is yours to fill.

II — Why this is your problem

Start with something nobody disputes: even the dull, non-nuclear behavior of hydrogen in palladium strains our best methods. The Pd–H system splits into a dilute and a concentrated phase across a miscibility gap, and reproducing that gap has frustrated interatomic-potential builders for years.⁴ Standard dispersion-corrected DFT does not get off lightly either; one careful study found that a widely used flavor produces sizeable errors in the tensile strength of palladium hydride, the kind of error that quietly corrupts anything you compute downstream.⁴ The protons themselves refuse to behave classically. Their zero-point motion is large enough that the hydride’s vibrational physics shows an inverse isotope effect, calculable only when you treat the nuclei as quantum, anharmonic objects from first principles.⁵

Now push the loading toward and past one hydrogen per metal atom, throw in the superabundant vacancies that high chemical potential stabilizes, and add a driving current. You have walked from the textbook into the part of the phase diagram where the textbook stops being reliable. When a Google-convened team spent two years and real money trying to reproduce cold-fusion signatures, they reported no excess heat — but their most durable contribution was a list of how badly the extreme-loading regime is mapped, and an invitation to map it.⁷

Here is the part that should make a simulator lean forward. The tools for this did not exist in 1989, and they exist now. Machine-learned interatomic potentials trained on first-principles data reproduce DFT accuracy at length and time scales DFT cannot touch, and on-the-fly schemes now build those potentials during the simulation itself.⁸ Path-integral methods give you the quantum spread of a proton’s position rather than a single fixed coordinate. Floquet techniques describe what periodic driving does to the available states of a system — which is the formal way of saying that “driven” and “at rest” are genuinely different physics, not a rhetorical dodge.¹⁰ The blank term that Leggett and Baym left for a later generation is a problem that later generation can actually attack.

III — What the numbers already say

An advocate who skips this section is selling something. So, plainly, here is the wall any proposed mechanism has to clear.

The barrier is real and the bound is real. The equilibrium Leggett–Baym bound is not a hand-wave; it is a rigorous, near model-independent ceiling, and it sits far below the rate that reported excess heat would require.³ Any claim of a lattice effect has to explain how it lives above that ceiling, which means explaining how the lattice leaves equilibrium in the relevant way.

Screening helps, and not nearly enough. Electrons in a metal do screen the deuterons, lowering the barrier. The effect is measurable and, interestingly, larger than first-principles theory predicts: a free deuterium molecule corresponds to a screening energy near 25 eV, metallic theory lands between 50 and 150 eV, and beam experiments on metals report figures around 150 to 300 eV, with palladium toward the top of that range.⁶ That gap between measured and predicted screening is itself a respectable open question. It is also nowhere near large enough to bridge the many orders of magnitude to a calorimetric heat rate.

The lattice can touch a nuclear rate. We have watched it happen. In 2025 a clean, mainstream, peer-reviewed experiment loaded deuterium electrochemically into a palladium target and then bombarded it with a deuteron beam at tens of keV. The loading raised the deuterium–deuterium fusion rate by 15 ± 2 percent.^6b What it establishes is narrow and undisputed: an electronvolt-scale chemical environment can measurably change the screening, and so shift a megaelectronvolt-scale nuclear rate by a small factor. It is hot, beam-driven fusion that produced exactly the neutrons it should have, and it is not excess heat or the lattice anomaly.

The missing ash is the hardest constraint of all. When two deuterons fuse, they go roughly half the time to tritium plus a proton and half to helium-3 plus a neutron. The channel that produces helium-4 and a 23.8 MeV gamma is suppressed to about one event in ten million.^1b So a real D–D fusion process at any serious rate should flood its surroundings with neutrons, breed tritium, and glow at a specific gamma energy. The reported heat claims, by and large, do not come with commensurate neutrons, tritium, or that gamma line. This is the central nuclear objection, and it remains unanswered.

IV — The one case nobody has closed

If the equilibrium ceiling is firm, the only intellectually live question is whether driving the system breaks the assumption the ceiling rests on. There are serious attempts to argue that it does, and the discipline owes them an honest hearing and an honest label.

One line of work integrates the tunneling rate not over a fixed deuteron separation but over the separation’s quantum fluctuation. Allow the two nuclei to sample positions about a tenth of a Bohr radius apart, roughly 5 picometers, and the static rate climbs by something like eight orders of magnitude.¹² That is a real estimate from a peer-reviewed paper, and it is also a single model estimate that, by the authors’ own framing, does not close the gap to the calorimetric rates. A second, more ambitious line proposes a generalized nuclear Dicke model in which lattice vibrations carry the 23.8 MeV away collectively, and claims rate enhancements exceeding forty orders of magnitude in palladium.¹³ That is a proposal, presented as a proposal. It is not an observation, and forty orders of magnitude is the kind of number that should raise every hair on a careful reader’s neck until the ash question is answered alongside it.

Notice the asymmetry, because it matters for calibration. The constraints in the previous section are independent mainstream physics. The two enhancement estimates share an author, and the solid-state screening synthesis they lean on comes from the same group.^13b The underlying screening measurements are independent work; the interpretation and the large enhancement numbers are largely one cluster. The move that would shift opinion is the driven-case calculation done with the same rigor Leggett and Baym brought to the equilibrium case.

It also means being honest about the prior, and the prior is low. Driving a system out of equilibrium changes reaction rates by modest factors, not by tens of orders of magnitude. The realistic expectation, stated flatly, is that a rigorous driven calculation lands far below the heat rate, much as the equilibrium one did. The reason to do it anyway is that the equilibrium result does not formally cover the driven case, and an open formal gap is exactly the thing a rigorous field is supposed to close rather than wave at.

The encouraging news is that the home field has already started picking the question back up. A 2025 quantum-chemistry study computed Gamow factors and D–D rate estimates across a range of systems using density functional theory. Its verdict on palladium clusters was negative, consistent with every bound above. It also reported, for some carbon nanostructures, geometries that might assist barrier penetration.¹¹ Take the specific nanocarbon claim with appropriate salt. Take the methodology as the proof of concept it is: the calculation is doable, people are doing it, and it adjudicates rather than cheerleads.

V — The calculation itself

Suppose you wanted to close the blank term. The individual pieces exist, because they are being built right now for hydride superconductors and hydrogen-storage kinetics. Assembling them into a verdict on the high-loading, driven D/Pd corner is harder than it sounds, for a reason worth stating before the recipe: the answer lives in the rarest configurations the simulation produces, and those are exactly where every tool is weakest.

First, train a machine-learned interatomic potential for palladium and deuterium on high-level electronic-structure data, and insist that it reproduce what we already know: the lattice constants, the elastic data, and that stubborn miscibility gap. A potential that fails the known physics has no business extrapolating into the unknown. The trouble is that the very quantity you most want is the one the potential was never constrained to get right — no training set samples two deuterons at sub-10-picometer separations. Second, run path-integral molecular dynamics so that the deuterons appear as the smeared quantum distributions they actually are, including the vacancies and defects that high loading stabilizes. The output you care about is the probability distribution of deuteron–deuteron separations, and specifically its extreme small-separation tail, which is a rare-event problem where statistics are worst and the potential is least trustworthy at the same time. Third, fold the Gamow factor over that distribution while the system is driven, not while it idles. This is the step that either extends the Leggett–Baym bound into the non-equilibrium regime or shows why driving cannot evade it.

Be clear about what this computes and what it does not. The program addresses one wall only: the Coulomb-barrier, can-the-deuterons-get-close-enough question. It says nothing about the missing ash. Whether a fused pair emits a neutron, breeds tritium, or routes its energy into a 23.8 MeV gamma is nuclear-structure and branching-ratio physics, and the machinery here does not contain it. So the ash objection from Section III survives this calculation completely intact.

With that scope fixed, the deliverable is a single comparison: the computed driven rate against the rate the reported heat would require. A rigorous driven bound that still sits far below the heat rate would do for the non-equilibrium case what Leggett and Baym did for equilibrium, and would retire the barrier half of a thirty-five-year argument on grounds nobody could wave away. A mechanism that credibly reaches a meaningful fraction of the rate would be a surprise worth having. The dull outcome and the startling one are both publishable. The only unpublishable outcome is the one we have now, which is the blank.

VI — The invitation

That is the honest pitch, and it is a better one than belief. The case for spending computational effort here does not rest on accepting any contested result. It rests on a well-posed problem that two of the field’s foundational 1989 papers left explicitly unfinished, sitting in a material whose extreme-loading regime independent groups have flagged as poorly mapped, and approachable now — for the first time and with eyes open about how hard the rare-event tail will be — using potentials and path-integral methods that did not exist when the question was first asked.

The institutional weather has shifted to match. In 2023 a U.S. federal energy agency committed on the order of ten million dollars to eight teams, hosted at places like MIT, Stanford, and a national laboratory, to look hard and skeptically at low-energy nuclear reactions.¹⁴ That funding does not certify that anything works. It certifies that serious institutions decided the question is no longer beneath asking. A simulator who builds the validated Pd–D potential and runs the driven calculation is not endorsing a heresy. They are doing the most ordinary thing a computational physicist can do, which is to take a number that someone asserted and find out, from first principles, whether it can possibly be true. Whichever way it falls, the field is better off knowing.

Notes

1. On Coulomb-barrier tunneling and the Gamow factor as the quantity governing all low-energy fusion rates, see G. Gamow, “Zur Quantentheorie des Atomkernes,” Zeitschrift für Physik 51 (1928): 204–212; the framework is uncontested textbook nuclear physics. ↩
2. D+D branching: the reaction proceeds roughly 50% to tritium + proton and 50% to helium-3 + neutron, with the radiative helium-4 + 23.8 MeV gamma channel suppressed to a branching ratio of about 10⁻⁶ to 10⁻⁷. Evaluated nuclear data: D. A. Brown et al., “ENDF/B-VIII.0: The 8th Major Release of the Nuclear Reaction Data Library,” Nuclear Data Sheets 148 (2018): 1–142. ↩
3. S. E. Koonin and M. Nauenberg, “Calculated Fusion Rates in Isotopic Hydrogen Molecules,” Nature 339 (1989): 690–691. Reproducing the claimed rates requires a screening particle of roughly ten electron masses. ↩
4. A. J. Leggett and G. Baym, “Exact Upper Bound on Barrier Penetration Probabilities in Many-Body Systems: Application to ‘Cold Fusion,’” Physical Review Letters 63 (1989): 191–194; and “Can Solid-State Effects Enhance the Cold-Fusion Rate?” Nature 340 (1989): 45–46. The bound is derived for a fully interacting many-body system in equilibrium and lies orders of magnitude below the claimed-heat rate. ↩
5. On the difficulty of capturing the Pd–H miscibility gap with classical potentials, and on first-principles errors in palladium hydride, see N. V. Ilawe, J. A. Zimmerman, and B. M. Wong, “Breaking Badly: DFT-D2 Gives Sizeable Errors for Tensile Strengths in Palladium-Hydride Solids,” Journal of Chemical Theory and Computation 11 (2015): 5426–5435. ↩
6. On the quantum, anharmonic character of hydrogen nuclei in palladium hydride and the computed inverse isotope effect, see I. Errea, M. Calandra, and F. Mauri, “First-Principles Theory of Anharmonicity and the Inverse Isotope Effect in Superconducting Palladium-Hydride Compounds,” Physical Review Letters 111 (2013): 177002. ↩
7. Screening energies: free D₂ ≈ 25 eV; metallic theory ≈ 50–150 eV; beam-target experiments on metals ≈ 150–300 eV. F. Metzler et al., “Known Mechanisms That Increase Nuclear Fusion Rates in the Solid State,” New Journal of Physics 26 (2024): 101202; primaries F. Raiola et al., European Physical Journal A 19 (2004): 283, and A. Huke et al., Physical Review C 78 (2008): 015803. These values do not span the gap to calorimetric rates. ↩
8. A. C. Chen et al., “Electrochemical Loading Enhances Deuterium Fusion Rates in a Metal Target,” Nature 644 (2025): 640–645, doi:10.1038/s41586-025-09042-7. A 15(±2)% increase in a ∼30 keV beam-target D–D rate; hot, beam-driven fusion that produced the expected neutrons, not lattice excess heat. ↩
9. C. P. Berlinguette et al., “Revisiting the Cold Case of Cold Fusion,” Nature 570 (2019): 45–51. No excess heat reproduced; the work identifies the extreme-loading materials-science regime as under-characterized. ↩
10. On machine-learned interatomic potentials reproducing first-principles accuracy at large scale, see R. Jinnouchi, F. Karsai, and G. Kresse, “On-the-Fly Machine Learning Force Field Generation: Application to Melting Points,” Physical Review B 100 (2019): 014105. ↩
11. Whether the equilibrium Leggett–Baym bound extends to, or is evaded by, a driven non-equilibrium cathode is an open question in the verifiable literature. A May 2026 reviewer lit-check found no rigorous, model-independent driven-case bound and no peer-reviewed on-its-own-terms rebuttal. ↩
12. On periodic driving changing the accessible states of a quantum system, see T. Oka and S. Kitamura, “Floquet Engineering of Quantum Materials,” Annual Review of Condensed Matter Physics 10 (2019): 387–408. ↩
13. S.-K. Pang, “Feasibility of D–D Nuclear Fusion Achieved by Chemical Methods: Quantum Chemical Analysis,” ACS Omega 10, no. 20 (2025): 20705. DFT Gamow-factor estimates; palladium clusters do not facilitate D–D fusion, while some carbon nanostructures are reported to assist. A single computational estimate; the nanocarbon finding is a hypothesis, not a result. ↩
14. F. Metzler et al., “Known Mechanisms That Increase Nuclear Fusion Rates in the Solid State,” New Journal of Physics 26 (2024): 101202, §S2.4: fluctuations of order 0.1 Bohr radius (∼5 pm) raise the static D–D rate by ∼8 orders of magnitude. Presented as one model estimate that does not close the gap to calorimetric rates. ↩
15. P. L. Hagelstein et al., “Models for Nuclear Fusion in the Solid State,” arXiv:2501.08338 (2025): a generalized nuclear Dicke model proposes D–D rate enhancements exceeding 40 orders of magnitude in palladium via lattice-mediated energy transfer. A proposed theoretical mechanism, not an observation, and not independently confirmed. ↩
16. On shared provenance: the fluctuation estimate (note 12) and the nuclear-Dicke model (note 13) share an author (F. Metzler), and both lean on the same group’s solid-state screening synthesis (note 6). The screening measurements behind note 6 (Raiola et al. 2004; Huke et al. 2008) are independent of that group. ↩
17. U.S. Advanced Research Projects Agency–Energy (ARPA-E), program announcement, February 2023: approximately $10 million across eight teams, including groups at MIT, Stanford, and Lawrence Berkeley National Laboratory, to investigate low-energy nuclear reactions. ↩

I — The Reply That Wasn’t an Experiment

A theory question got a theory answer, and one term was left blank.

When the cold-fusion claim broke in the spring of 1989, the most damaging responses did not come from anyone trying and failing to boil water in a beaker. They came from people with pencils. Within weeks, Steven Koonin and Michael Nauenberg computed the rate at which two deuterons would fuse inside an ordinary hydrogen-isotope molecule and found it preposterously small. To reproduce the heat being reported, they noted, you would need the two nuclei screened by a particle roughly ten times the mass of an electron, and palladium contains no such thing.² A few months later Anthony Leggett and Gordon Baym went further. They derived an upper bound on how fast deuterons could tunnel together inside a metal, a bound that holds for any fully interacting many-body system in equilibrium, almost regardless of the model you choose. The bound sat orders of magnitude below the rate the claimed heat demanded.³

This is worth pausing on, because it is the part of the story that gets flattened. The field was not dismissed by a sneer. It was dismissed by a calculation, and the calculation was good. Tunneling through the Coulomb barrier is suppressed by the Gamow factor, which falls off exponentially with the area under the barrier, and for two deuterons at room temperature that suppression is total for all practical purposes.¹ Nothing in the decades since has overturned that arithmetic.

But look at what both calculations assumed. Koonin and Nauenberg put two nuclei in a static molecule. Leggett and Baym took the lattice in equilibrium. Those are the honest, conservative choices a theorist makes when the burden of proof sits with the claimant, and in 1989 it did. They are also, precisely, the choices that leave one term of the problem blank: a metal that is loaded past its comfortable limit and driven hard, with current running through it, vacancies multiplying, and the deuterons anything but at rest. Whether the bound survives that case has not, in the verifiable literature, been worked out either way.⁹ That blank is a computational-physics problem. This article is about why it is yours to fill.

II — Why This Is Your Problem

Hydrogen in palladium is already hard to compute. The interesting regime is the hard part of the hard problem.

Start with something nobody disputes: even the dull, non-nuclear behavior of hydrogen in palladium strains our best methods. The Pd–H system splits into a dilute and a concentrated phase across a miscibility gap, and reproducing that gap has frustrated interatomic-potential builders for years.⁴ Standard dispersion-corrected DFT does not get off lightly either; one careful study found that a widely used flavor produces sizeable errors in the tensile strength of palladium hydride, the kind of error that quietly corrupts anything you compute downstream.⁴ The protons themselves refuse to behave classically. Their zero-point motion is large enough that the hydride’s vibrational physics shows an inverse isotope effect, calculable only when you treat the nuclei as quantum, anharmonic objects from first principles.⁵

Now push the loading toward and past one hydrogen per metal atom, throw in the superabundant vacancies that high chemical potential stabilizes, and add a driving current. You have walked from the textbook into the part of the phase diagram where the textbook stops being reliable. The honest description of that corner is not “well understood.” It is “under-characterized,” and the people who said so most clearly were not advocates. When a Google-convened team spent two years and real money trying to reproduce cold-fusion signatures, they reported no excess heat. The result they did hand the community was a list of how badly the extreme-loading regime is mapped, and an invitation to map it.⁷

Here is the part that should make a simulator lean forward. The tools for this did not exist in 1989, and they exist now. Machine-learned interatomic potentials trained on first-principles data reproduce DFT accuracy at length and time scales DFT cannot touch, and on-the-fly schemes now build those potentials during the simulation itself.⁸ Path-integral methods give you the quantum spread of a proton’s position rather than a single fixed coordinate. Floquet techniques describe what periodic driving does to the available states of a system, which is the formal way of saying that “driven” and “at rest” are genuinely different physics, not a rhetorical dodge.¹⁰ The blank term that Leggett and Baym left for a later generation is a problem that later generation can actually attack.

III — What the Numbers Already Say

Be honest about the constraints first. They are strong, and pretending otherwise is how the field lost its credibility the first time.

An advocate who skips this section is selling something. So, plainly, here is the wall any proposed mechanism has to clear.

The barrier is real and the bound is real. The equilibrium Leggett–Baym bound is not a hand-wave; it is a rigorous, near model-independent ceiling, and it sits far below the rate that reported excess heat would require.³ Any claim of a lattice effect has to explain how it lives above that ceiling, which means explaining how the lattice leaves equilibrium in the relevant way.

Screening helps, and not nearly enough. Electrons in a metal do screen the deuterons, lowering the barrier. The effect is measurable and, interestingly, larger than first-principles theory predicts: a free deuterium molecule corresponds to a screening energy near 25 eV, metallic theory lands between 50 and 150 eV, and beam experiments on metals report figures around 150 to 300 eV, with palladium toward the top of that range.⁶ That gap between measured and predicted screening is itself a respectable open question. It is also nowhere near large enough to bridge the many orders of magnitude to a calorimetric heat rate, and the appendix records that bound explicitly. Screening is real but bounded; it is not evidence for the heat claim.

The lattice can touch a nuclear rate. We have watched it happen. In 2025 a clean, mainstream, peer-reviewed experiment loaded deuterium electrochemically into a palladium target and then bombarded it with a deuteron beam at tens of keV. The loading raised the deuterium–deuterium fusion rate by 15 ± 2 percent.^6b Read that carefully, because it is easy to over-read. What it establishes is narrow and undisputed: an electronvolt-scale chemical environment can measurably change the screening, and so shift a megaelectronvolt-scale nuclear rate by a small factor. That is all it establishes. It is hot, beam-driven fusion that produced exactly the neutrons it should have, the enhancement is a percent-level effect rather than the many orders of magnitude the heat claim needs, and it is not excess heat or the lattice anomaly.

The missing ash is the hardest constraint of all. When two deuterons fuse, they go roughly half the time to tritium plus a proton and half to helium-3 plus a neutron. The channel that produces helium-4 and a 23.8 MeV gamma is suppressed to about one event in ten million.^1b So a real D–D fusion process at any serious rate should flood its surroundings with neutrons, breed tritium, and glow at a specific gamma energy. The reported heat claims, by and large, do not come with commensurate neutrons, tritium, or that gamma line. This is the central nuclear objection, and it remains unanswered. No calculation of an enhanced tunneling rate is worth much until it also says where the ash went.

IV — The One Case Nobody Has Closed

Two careful theory groups have proposed ways around the bound. Neither is established, and saying so is the point.

If the equilibrium ceiling is firm, the only intellectually live question is whether driving the system breaks the assumption the ceiling rests on. There are serious attempts to argue that it does, and the discipline owes them an honest hearing and an honest label.

One line of work integrates the tunneling rate not over a fixed deuteron separation but over the separation’s quantum fluctuation. Allow the two nuclei to sample positions about a tenth of a Bohr radius apart, roughly 5 picometers, and the static rate climbs by something like eight orders of magnitude.¹² That is a real estimate from a peer-reviewed paper, and it is also a single model estimate that, by the authors’ own framing, does not close the gap to the calorimetric rates. A second, more ambitious line proposes a generalized nuclear Dicke model in which lattice vibrations carry the 23.8 MeV away collectively, and claims rate enhancements exceeding forty orders of magnitude in palladium while trying to address decoherence and the resonance condition.¹³ That is a proposal, presented as a proposal. It is not an observation, it has not been independently confirmed, and forty orders of magnitude is the kind of number that should raise every hair on a careful reader’s neck until the ash question is answered alongside it.

Notice the asymmetry, because it matters for calibration. The constraints in the previous section are independent mainstream physics. The two enhancement estimates are not independent of each other: the fluctuation argument and the nuclear-Dicke model share an author, and the solid-state screening synthesis they lean on comes from the same group.^13b The underlying screening measurements (the beam experiments behind note 6) are independent work; the solid-state interpretation and the large enhancement numbers are largely one cluster. That does not make them wrong. It does mean the burden of the next move sits with the proposers, and the move that would shift opinion is the driven-case calculation done with the same rigor Leggett and Baym brought to the equilibrium case.

It also means being honest about the prior, and the prior is low. Driving a system out of equilibrium, by every well-characterized example we have, changes reaction rates by modest factors, not by tens of orders of magnitude. Floquet driving reshuffles energy levels and opens nominally forbidden transitions, but a tunneling probability that is exponentially small in the barrier area does not become order-unity because you ran a current through the cathode. To bridge the gap to the heat claim, a tweak to the barrier is not enough; you would need a wholesale collective channel that hands the 23.8 MeV to the lattice coherently, and no such channel has been shown to exist. So the realistic expectation, stated flatly, is that a rigorous driven calculation lands far below the heat rate, much as the equilibrium one did. The reason to do it anyway is not optimism. It is that the equilibrium result does not formally cover the driven case, and an open formal gap, however unpromising, is exactly the thing a rigorous field is supposed to close rather than wave at.

The encouraging news is that the home field has already started picking the question back up. A 2025 quantum-chemistry study computed Gamow factors and D–D rate estimates across a range of systems using density functional theory. Its verdict on palladium clusters was negative, consistent with every bound above. It also reported, for some carbon nanostructures, geometries that might assist barrier penetration.¹¹ Take the specific nanocarbon claim with appropriate salt. Take the methodology as the proof of concept it is: the calculation is doable, people are doing it, and it adjudicates rather than cheerleads.

V — The Calculation Itself

Here is a concrete program with a number at the end of it. The number is the deliverable.

Suppose you wanted to close the blank term. The individual pieces exist, because they are being built right now for hydride superconductors and hydrogen-storage kinetics. Assembling them into a verdict on the high-loading, driven D/Pd corner is harder than it sounds, for a reason worth stating before the recipe: the answer lives in the rarest configurations the simulation produces, and those are exactly where every tool is weakest. With that caveat in front of us, the program looks like this.

First, train a machine-learned interatomic potential for palladium and deuterium on high-level electronic-structure data, and insist that it reproduce what we already know: the lattice constants, the elastic data, and that stubborn miscibility gap. A potential that fails the known physics has no business extrapolating into the unknown. The trouble is that the converse does not follow. A potential that passes on bulk lattice constants and elastic data has been validated nowhere near the sub-10-picometer separations that govern the tunneling rate, because no training set samples two deuterons that close. The very quantity you most want is the one the potential was never constrained to get right. Second, run path-integral molecular dynamics so that the deuterons appear as the smeared quantum distributions they actually are, including the vacancies and defects that high loading stabilizes. The output you care about is not a pretty trajectory. It is the probability distribution of deuteron–deuteron separations, and specifically its extreme small-separation tail, which is a rare-event problem where statistics are worst and the potential is least trustworthy at the same time. Third, fold the Gamow factor over that distribution while the system is driven, not while it idles. This is the step that either extends the Leggett–Baym bound into the non-equilibrium regime or shows, on the bound’s own terms, why driving cannot evade it.

Be clear about what this computes and what it does not. The program addresses one wall only: the Coulomb-barrier, can-the-deuterons-get-close-enough question. It says nothing about the missing ash. Whether a fused pair emits a neutron, breeds tritium, or routes its energy into a 23.8 MeV gamma is nuclear-structure and branching-ratio physics, and the machinery here — classical-lattice potentials plus path-integral sampling — does not contain it. So the ash objection from Section III survives this calculation completely intact. A driven rate, however it comes out, is necessary-question progress, not the whole question.

With that scope fixed, the deliverable is a single comparison: the computed driven rate against the rate the reported heat would require. Two clean outcomes follow, and both are worth the effort. A rigorous driven bound that still sits far below the heat rate would do for the non-equilibrium case what Leggett and Baym did for equilibrium, and would retire the barrier half of a thirty-five-year argument on grounds nobody could wave away. This is the likely outcome, and it would be a real result. A mechanism that credibly reaches a meaningful fraction of the rate would be a surprise worth having, and even then it would still owe an answer to the ash before anyone called it nuclear heat. The dull outcome and the startling one are both publishable. The only unpublishable outcome is the one we have now, which is the blank.

VI — The Invitation

You do not have to believe the heat is real. You only have to want the question closed.

That is the honest pitch, and it is a better one than belief. The case for spending computational effort here does not rest on accepting any contested result. It rests on a well-posed problem that two of the field’s foundational 1989 papers left explicitly unfinished, sitting in a material whose extreme-loading regime independent groups have flagged as poorly mapped, and approachable now, for the first time and with eyes open about how hard the rare-event tail will be, using potentials and path-integral methods that did not exist when the question was first asked.

The institutional weather has shifted to match. In 2023 a U.S. federal energy agency committed on the order of ten million dollars to eight teams, hosted at places like MIT, Stanford, and a national laboratory, to look hard and skeptically at low-energy nuclear reactions.¹⁴ That funding does not certify that anything works. It certifies that serious institutions decided the question is no longer beneath asking. A simulator who builds the validated Pd–D potential and runs the driven calculation is not endorsing a heresy. They are doing the most ordinary thing a computational physicist can do, which is to take a number that someone asserted and find out, from first principles, whether it can possibly be true. Whichever way it falls, the field is better off knowing.

Notes

1. On Coulomb-barrier tunneling and the Gamow factor as the quantity governing all low-energy fusion rates, see G. Gamow, “Zur Quantentheorie des Atomkernes,” Zeitschrift für Physik 51 (1928): 204–212; the framework is uncontested textbook nuclear physics. ↩
2. D+D branching: the reaction proceeds roughly 50% to tritium + proton and 50% to helium-3 + neutron, with the radiative helium-4 + 23.8 MeV gamma channel suppressed to a branching ratio of about 10⁻⁶ to 10⁻⁷. Evaluated nuclear data: D. A. Brown et al., “ENDF/B-VIII.0: The 8th Major Release of the Nuclear Reaction Data Library,” Nuclear Data Sheets 148 (2018): 1–142. ↩
3. S. E. Koonin and M. Nauenberg, “Calculated Fusion Rates in Isotopic Hydrogen Molecules,” Nature 339 (1989): 690–691, https://doi.org/10.1038/339690a0. Reproducing the claimed rates requires a screening particle of roughly ten electron masses. ↩
4. A. J. Leggett and G. Baym, “Exact Upper Bound on Barrier Penetration Probabilities in Many-Body Systems: Application to ‘Cold Fusion,’” Physical Review Letters 63 (1989): 191–194; and “Can Solid-State Effects Enhance the Cold-Fusion Rate?” Nature 340 (1989): 45–46, https://doi.org/10.1038/340045a0. The bound is derived for a fully interacting many-body system in equilibrium and lies orders of magnitude below the claimed-heat rate. ↩
5. On the difficulty of capturing the Pd–H miscibility gap with classical potentials, and on first-principles errors in palladium hydride, see N. V. Ilawe, J. A. Zimmerman, and B. M. Wong, “Breaking Badly: DFT-D2 Gives Sizeable Errors for Tensile Strengths in Palladium-Hydride Solids,” Journal of Chemical Theory and Computation 11 (2015): 5426–5435. ↩
6. On the quantum, anharmonic character of hydrogen nuclei in palladium hydride and the computed inverse isotope effect, see I. Errea, M. Calandra, and F. Mauri, “First-Principles Theory of Anharmonicity and the Inverse Isotope Effect in Superconducting Palladium-Hydride Compounds,” Physical Review Letters 111 (2013): 177002. ↩
7. Screening energies: free D₂ ≈ 25 eV; metallic theory ≈ 50–150 eV (lithium low, palladium high); beam-target experiments on metals ≈ 150–300 eV. F. Metzler et al., “Known Mechanisms That Increase Nuclear Fusion Rates in the Solid State,” New Journal of Physics 26 (2024): 101202, §S2.3; primaries F. Raiola et al., European Physical Journal A 19 (2004): 283, and A. Huke et al., Physical Review C 78 (2008): 015803. These values do not span the gap to calorimetric rates. ↩
8. A. C. Chen et al., “Electrochemical Loading Enhances Deuterium Fusion Rates in a Metal Target,” Nature 644 (2025): 640–645, https://doi.org/10.1038/s41586-025-09042-7. A 15(2)% increase in a ~30 keV beam-target D–D rate; this is hot, beam-driven fusion that produced the expected neutrons, not lattice excess heat. ↩
9. C. P. Berlinguette et al., “Revisiting the Cold Case of Cold Fusion,” Nature 570 (2019): 45–51, https://doi.org/10.1038/s41586-019-1256-6. No excess heat reproduced; the work identifies the extreme-loading materials-science regime as under-characterized. ↩
10. On machine-learned interatomic potentials reproducing first-principles accuracy at large scale, and on-the-fly generation during simulation, see R. Jinnouchi, F. Karsai, and G. Kresse, “On-the-Fly Machine Learning Force Field Generation: Application to Melting Points,” Physical Review B 100 (2019): 014105. ↩
11. Whether the equilibrium Leggett–Baym bound (note 3) extends to, or is evaded by, a driven non-equilibrium cathode is recorded as OPEN in the series ledger (QM-OPEN1). A May 2026 reviewer lit-check found no rigorous, model-independent driven-case bound and no peer-reviewed on-its-own-terms rebuttal; non-rigorous non-equilibrium proposals exist in the field literature but are unreplicated. Carried as a strong lead, not a closed file. ↩
12. On periodic driving changing the accessible states of a quantum system, cited only to legitimize the equilibrium-vs-driven distinction and not as a fusion claim, see T. Oka and S. Kitamura, “Floquet Engineering of Quantum Materials,” Annual Review of Condensed Matter Physics 10 (2019): 387–408, https://doi.org/10.1146/annurev-conmatphys-031218-013423. ↩
13. S.-K. Pang, “Feasibility of D–D Nuclear Fusion Achieved by Chemical Methods: Quantum Chemical Analysis,” ACS Omega 10, no. 20 (2025): 20705, https://doi.org/10.1021/acsomega.5c01651. DFT (PBE/def2-SVP) Gamow-factor estimates; palladium clusters do not facilitate D–D fusion, while some carbon nanostructures are reported to assist. A single computational estimate; the nanocarbon finding is a hypothesis, not a result. ↩
14. F. Metzler et al., “Known Mechanisms That Increase Nuclear Fusion Rates in the Solid State,” New Journal of Physics 26 (2024): 101202, §S2.4: fluctuations of order 0.1 Bohr radius (~5 pm) raise the static D–D rate by ~8 orders of magnitude. Presented by its authors as one model estimate that does not close the gap to calorimetric rates. ↩
15. P. L. Hagelstein et al., “Models for Nuclear Fusion in the Solid State,” arXiv:2501.08338 (2025): a generalized nuclear Dicke model proposes D–D rate enhancements exceeding 40 orders of magnitude in palladium via lattice-mediated energy transfer. A proposed theoretical mechanism, not an observation, and not independently confirmed. ↩
16. On shared provenance: the fluctuation estimate (note 12) and the nuclear-Dicke model (note 13) share an author (F. Metzler), and both lean on the same group’s solid-state screening synthesis (note 6, Metzler et al. 2024). The screening measurements behind note 6 (Raiola et al. 2004; Huke et al. 2008) are independent of that group. ↩
17. U.S. Advanced Research Projects Agency–Energy (ARPA-E), program announcement, February 2023: approximately $10 million across eight teams, including groups at MIT, Stanford, and Lawrence Berkeley National Laboratory, to investigate low-energy nuclear reactions. ↩