This preprint studies how superfluid helium-4 can test spontaneous wave-function collapse models (CSL and Diósi–Penrose) through bulk heating. White-noise collapse heats every material equally per unit mass and is excluded at its natural parameters by X-ray searches. Bulk heating therefore constrains collapse only if the noise is colored, and the ratio of the heating per unit mass of helium to that of a solid then classifies the models. Main results:• Rest-frame colored CSL: with a Lorentzian spectrum, He-4 is heated more than a solid by up to (c_s/c_He)² = 85–2170, robustly against phonon damping. For any spectrum that does not rise with frequency, helium is heated at least as much as any solid. A helium calorimeter at 10 pW/kg would improve the best existing bound by a factor of 5.4 at r_C = 10⁻⁷ m and about 72 for r_C ≲ 10⁻⁸ m. This requires a noise rest frame that co-moves with the Earth.• Collapse wind: if the noise has a CMB rest frame, heating becomes independent of the material and shows a 7.9% annual modulation peaking on 12–13 June; the lunar bound gives λΩ ≲ 30 s⁻² at r_C = 10⁻⁷ m.• Diósi–Penrose: with a noise cutoff above ≈10¹⁵ s⁻¹ and R₀ set to the zero-point width, DP predicts 12.5 pW/kg for Cu and 66–69 pW/kg for W. A factor-of-two convention in published solid He-4 Debye–Waller data is corrected. Penrose's prescription does not fix a heating rate for the liquid.• Readout: quantum-evaporation efficiency is set by the cell walls (η ≈ 0.07 with the measured roton reflectivity, 0.72 with low-loss diffuse walls). A fill-level calorimetric method with an interleaved fill order is proposed, with a background budget for surface and underground operation. Verification: the Supplementary Material proves the 15 theorems behind the paper's formulas (44 symbolic and numerical checks) and tests the inputs that cannot be proven against data and independent models. Every number printed in the paper is recomputed independently (248 of 248), and the quantum-evaporation Monte Carlo is reproduced by a second, independent implementation. All results are conditional on the stated collapse models and noise properties; whether spontaneous collapse exists can only be decided by experiment.
This preprint studies how superfluid helium-4 can test spontaneous wave-function collapse models (CSL and Diósi–Penrose) through bulk heating. White-noise collapse heats every material equally per unit mass and is excluded at its natural parameters by X-ray searches. Bulk heating therefore constrains collapse only if the noise is colored, and the ratio of the heating per unit mass of helium to that of a solid then classifies the models. Main results:• Rest-frame colored CSL: with a Lorentzian spectrum, He-4 is heated more than a solid by up to (c_s/c_He)² = 85–2170, robustly against phonon damping. For any spectrum that does not rise with frequency, helium is heated at least as much as any solid. A helium calorimeter at 10 pW/kg would improve the best existing bound by a factor of 5.4 at r_C = 10⁻⁷ m and about 72 for r_C ≲ 10⁻⁸ m. This requires a noise rest frame that co-moves with the Earth.• Collapse wind: if the noise has a CMB rest frame, heating becomes independent of the material and shows a 7.9% annual modulation peaking on 12–13 June; the lunar bound gives λΩ ≲ 30 s⁻² at r_C = 10⁻⁷ m.• Diósi–Penrose: with a noise cutoff above ≈10¹⁵ s⁻¹ and R₀ set to the zero-point width, DP predicts 12.5 pW/kg for Cu and 66–69 pW/kg for W. A factor-of-two convention in published solid He-4 Debye–Waller data is corrected. Penrose's prescription does not fix a heating rate for the liquid.• Readout: quantum-evaporation efficiency is set by the cell walls (η ≈ 0.07 with the measured roton reflectivity, 0.72 with low-loss diffuse walls). A fill-level calorimetric method with an interleaved fill order is proposed, with a background budget for surface and underground operation. Verification: the Supplementary Material proves the 15 theorems behind the paper's formulas (44 symbolic and numerical checks) and tests the inputs that cannot be proven against data and independent models. Every number printed in the paper is recomputed independently (248 of 248), and the quantum-evaporation Monte Carlo is reproduced by a second, independent implementation. All results are conditional on the stated collapse models and noise properties; whether spontaneous collapse exists can only be decided by experiment.
Continuous Spontaneous Localization (CSL) and related collapse models predict a universal heating of matter. If the collapse noise is colored, i.e. has a spectral cutoff Ω far below X-ray frequencies, the X-ray bounds that exclude white-noise CSL no longer apply, and bulk heating of a medium samples the noise at the phonon frequency c_s/r_C. Superfluid helium-4, with a sound speed of about 238 m/s, samples it roughly 17 times lower in frequency than typical solids. For Lorentzian (cCSL) noise, helium is therefore more sensitive than a solid by (c_solid/c_He)^2 ≈ 280 when Ω ≪ c_He/r_C. This advantage is robust against phonon damping; the much larger ratio obtained for an ideal sharp cutoff is not. We compare projected helium constraints with all cutoff-robust bounds over the full (λ, r_C, Ω) parameter space: LISA Pathfinder and its re-analysis, the layered ultracold cantilever, cold atoms, and the heat budgets of the Earth and Moon. The mechanical bounds are scaled with their exact form factors, and the space is limited from below by a classicality requirement. At r_C = 10^-7 m, a helium calorimeter with residual unexplained heating of 10 pW/kg improves on the strongest peer-reviewed bound only by a factor of about 5 (2.4 including preprint bounds). The advantage is concentrated at small localization lengths: for r_C ≲ 3×10^-8 m the gain is about 55–75 at 10 pW/kg and about 7 even at the demonstrated level of ~100 pW/kg. For r_C ≲ 3 nm, helium would close the remaining open parameter space in a band of Ω. We propose a discriminating two-medium signature. CSL heating per unit mass is material independent for white noise, but colored noise raises the helium-to-solid ratio up to (c_solid/c_He)^2, i.e. 85–2200 for Pb to sapphire. We also show that single-mode quantum acoustics is not competitive with calorimetry, and we give a quantitative systematic budget for an underground, differential calorimeter. Version 2 adds the layered-cantilever bound, the LISA Pathfinder re-analysis, the lunar bound, a classicality lower bound, a damping analysis, a systematic budget and the two-medium signature, and corrects the novelty claims of version 1. All numbers and figures are reproducible with the accompanying Python scripts (physics.py, analysis.py), which run a built-in validation suite first.
Dynamical collapse models such as Continuous Spontaneous Localization (CSL) address the quantum measurement problem and predict a tiny spontaneous heating of matter. The originally proposed parameter range of white-noise CSL has been excluded by X-ray searches, most recently XENONnT (2026). These X-ray bounds, however, lose their validity if the collapse noise is "colored", i.e. cut off below X-ray frequencies. In that regime the strongest remaining constraints come from bulk heating of solids and from low-frequency mechanical experiments. This feasibility study applies the colored-noise heating formalism of Adler and Vinante to superfluid helium-4. Because the sound speed of helium (~238 m/s) is about 17 times lower than that of solids, helium samples the noise spectrum at correspondingly lower frequencies. This gives a sensitivity advantage over solids of ~280 for Lorentzian (cCSL) noise and up to ~10^6 for a sharp cutoff. For r_C = 10^-7 m, a helium calorimeter limiting unexplained heating to 10 pW/kg (1 pW/kg) would improve on the strongest existing bound by a factor of ~2 (~25) in the window Ω ≈ 3×10^9 – 5×10^10 s^-1. This window overlaps with cosmologically motivated cutoff frequencies for cCSL. The note includes closed-form results, validated numerical calculations, and a proposed experimental concept: underground DC calorimetry with tens of kilograms of superfluid helium, using a fill-level differential measurement to separate wall-induced backgrounds. Limitations and open questions are discussed.
In the Diósi–Penrose (DP) model a superposition of two mass distributions decays at the rate E_G/ħ. Diósi's master equation uses the mass-density operator smeared over a length R0; Penrose's formulation uses the density of the stationary state, and the "parameter-free" choice R0 = u_rms (the zero-point amplitude) links the two. This preprint examines what each choice implies for spontaneous heating and for collapse rates, and states separately which results are standard and which are new. Main results:• Heating per unit mass is 2πGħ∫g², i.e. Għ/(4√π σxσyσz) for Gaussian smearing of per-axis widths σi. With R0 equal to the per-axis zero-point amplitude, Diósi's functional reproduces Penrose's E_G for displacements smaller than the lattice spacing.• For a crystal ball, E_G splits exactly into a continuum term and a per-atom granular term. The continuum term is already contained in the per-atom term (the gravitational analogue of the Lorentz local field), the granular term is lattice periodic, and it carries a Madelung reduction of 0.90 (Cu) and 0.29 (⁴He).• Copper is predicted to heat at 12.1–12.8 pW/kg. This is in mild tension with the 10 pW/kg figure adopted after background subtraction from the Bayreuth heat leak.• The published Debye–Waller values of hcp ⁴He follow a convention in which the reported ⟨u²⟩ is half the one-dimensional mean-square displacement. One direct cross-check and four consistency checks support this reading. With it, the solid-helium prediction becomes 6–8 fW/kg instead of 21 fW/kg.• For any liquid (⁴He or ³He) the one-body density is uniform, so Penrose's granular E_G vanishes. The solid/liquid ⁴He collapse-rate ratio is ≈6 for d ≪ u, 2.8 at 0.2 nm and 1.15 at 1 nm, which quantifies a conjecture of Diósi (2013).• Heating under Penrose's prescription depends on the dynamical completion. For every completion that splits ⟨ρ⟩ into non-negative pieces, the liquid heats at most 2πGħn = 1.15 fW/kg; maximal localization then gives a liquid/solid ratio of 0.155. Density-matrix and configuration-based completions give anything from ~10⁻²⁶ to order one.• Experiment: – Ionizing radiation deposits nearly the same power per kilogram in every material and therefore mimics the fixed-R0 signal. – Underground at LNGS, unshielded environmental gammas still deposit 0.6–8 pW/kg, so ≥20 cm of lead is needed. – The nearest-term test of the zero-point prescription is to compare the heat leaks of heavy, stiff metals with copper; tungsten is predicted at ≈77 pW/kg, six times copper. – Underground helium calorimetry at 1–0.1 pW/kg would test colored fixed-R0 models up to R0 ≈ 10–21 pm, provided the cell-wall heat release can be subtracted to 5–0.5%.
Spontaneous-collapse models modify the Schrödinger equation with a weak stochastic noise that suppresses macroscopic superpositions. A physically generated noise cannot be white: it must have a spectral cutoff Ω_c. A cutoff in turn breaks Lorentz invariance and singles out a rest frame, most naturally that of the cosmic microwave background (CMB). This preprint works out what happens when a laboratory moves through such a colored collapse noise at v ≈ 370 km/s, an effect we call the "collapse wind". Because the CSL kernel weights wave numbers q ~ 1/r_C, the Doppler shift q·v ~ v/r_C ≈ 3.7 × 10^12 s^-1 dominates every phonon and mechanical frequency, provided the noise pattern is frozen (Ω_c r_C ≪ v). Main results: • Bulk heating: an exact suppression factor with the closed-form asymptote F ≈ (2√π/3) Ω_c r_C / v. Heating per unit mass becomes independent of the material, whereas rest-frame colored CSL predicts ratios up to 10^2–10^4 between materials.• Bounds: existing bulk-heating data imply λ Ω_c ≲ 10^2 s^-2 at r_C = 10^-7 m.• Annual modulation: 7.9% amplitude, peaking on 12–13 June (Planck dipole + ERFA ephemerides).• Mechanical tests: an exact real-space method for layered and cuboidal test masses, applied to the LISA Pathfinder cube and to the multilayer cantilever that sets the strongest mechanical bound. The collapse force noise is suppressed to a sidereal average of ~5 × 10^-3 at Ω_c = 10^10 s^-1 and concentrated in two ~15-minute windows per sidereal day. Bulk heating becomes the binding constraint, and part of Adler's enhanced-collapse region is reopened. A sidereal-phase-gated reanalysis of existing force-noise data could reach it.• A norm-preserving, Born-rule-compliant directional collapse equation: superpositions separated perpendicular to the wind collapse ~15 times faster than those separated along it (10 μm sphere). All numbers and figures are reproduced by the accompanying Python code (included in the source archive).
Predictions of the Diósi–Penrose (DP) gravity-related collapse model depend strongly on the mass-density smoothing length R₀ and on the frequency spectrum of the gravitational fluctuations. We reanalyze the public low-energy electronic-recoil data of XENONnT (1–30 keV, 433 events, 1.16 t·yr) with an unbinned likelihood and report four results. First, when R₀ exceeds the atomic size, the equivalence principle makes the nucleus and its electrons accelerate together, and low-energy X-ray emission is strongly cancelled (about 2×10⁻⁴ at 1 keV and 0.16 at 10 keV for xenon). Including this screening, the bound on the white-noise DP model is R₀ > ~0.9 nm, much weaker than the unscreened value (~4.0 nm). Second, the reported bound R₀ > 2.8 nm of an earlier collapse analysis of XENONnT data appears to differ from the Majorana analysis in signal normalization by a factor π^(3/2). With the same normalization it corresponds to 1.56 nm, in agreement with the 1.52 nm we obtain under the same conditions. Third, an 'atomic-unit' variant, with R₀ set to the atomic radius and the fluctuation cutoff set to the electronic time scale, cannot be decided with these data: its verdict ranges from allowed (prediction/limit 0.006) to excluded (11) depending on how strongly the electrons co-move with the nucleus. Fourth, a parameter-free 'phonon-unit' variant, with both R₀ and the cutoff fixed by the lattice vibration of the same nucleus (Debye–Waller rms displacement and Debye frequency), passes the X-ray data independently of the screening treatment and predicts spontaneous heating of about 9.4 pW/kg in copper at cryogenic temperature. This lies just below the currently known unexplained heat-leak limit (~10 pW/kg). X-rays cannot separate the two variants; the (atomic mass × Debye temperature)^(3/2) heating scaling predicted by the phonon-unit variant can be tested by comparing heat leaks in tungsten and aluminium (a factor of about 16).“v2: removed comparison with a withdrawn preprint; added comparison with the XENON collaboration analysis (arXiv:2506.05507) and the 8π convention conversion.”
The possibility that the collapse of the wave function in quantum mechanics is real and ultimately connected to (classical) gravity has been debated for decades with main contributions by Diósi and Penrose (DP). In particular, Diósi proposed a noise-based dynamical reduction model, which captures the same order of magnitudes for the collapse time suggested by Penrose based on heuristic arguments. This is known in the literature as the DP model (Diósi–Penrose). A peculiarity of the DP model is the prediction of spontaneous heating of matter, which can be tested without the need for massive quantum superpositions. Notably, a very similar effect is predicted by recent theoretical approaches to gravity as a classical-only information channel. Here, we reconsider the current constraints on the DP model from spontaneous heating by analyzing experimental situations not properly considered before. We argue that the parameter-free version of the DP model is close to be ruled out by standard heat leak measurements at ultralow temperatures with a conclusive exclusion likely within reach with existing technology. This result would strengthen a recent claim of exclusion inferred by spontaneous x-ray emission experiments, which relies on the somewhat stronger assumption that the DP noise field is white up to x-ray frequencies.