Research

Decaying Dark Matter with an Integral-Equation Approach

2026 · with A. Bencke†, M. Kamionkowski, J. L. Bernal

We study models where a dark matter particle decays into two lighter daughter particles with arbitrary masses, naturally spanning both the massless limit (dark radiation) and the massive limit (warm decay products). Building on our integral-equation framework for non-cold relics, we developed CLASSIER-DDM, an extension of CLASSIER that evolves the decay products' perturbations without truncating a Boltzmann hierarchy or resorting to fluid approximations. The method handles the momentum-dependent decay time of each product, achieves sub-0.1% accuracy in both the matter power spectrum and the CMB lensing potential power spectrum, and runs in about a minute per evaluation — making DDM parameter estimation numerically tractable across a wide range of decay rates and kick velocities.

Using this framework, we also derived observational constraints from current data. Motivated by DESI's ~5% lower late-time matter density relative to Planck, we combined DESI DR2 BAO with Planck CMB data including lensing and found that decaying dark matter is not favored over ΛCDM: CMB lensing tightly constrains the recoil velocity imparted to decay products (to within 10⁻²–10⁻³ of the speed of light at 1σ), though part of the allowed parameter space could still help address small-scale structure anomalies like dwarf-galaxy discrepancies.

arXiv:2607.24910 → arXiv:2606.14849 → link to CLASSIER-DDM

Non-Cold Relics Without a Boltzmann Hierarchy

2025 – 2026 · with M. Kamionkowski et al.

Traditional Boltzmann codes track non-cold relics (like massive neutrinos) by truncating an infinite hierarchy of multipole equations, which becomes costly and can introduce numerical artifacts on small scales. We developed CLASSIER (CLASS Integral Equation Revision), which replaces that hierarchy with a set of integral equations solved iteratively, matching fully-converged accuracy while avoiding truncation artifacts. A follow-up analytic approximation for the small-scale, quasi-stationary regime pushes this further, giving a 3–6× speedup over standard CLASS while preserving sub-0.1% accuracy in the matter power spectrum — removing massive neutrinos as a computational bottleneck for high-precision, small-scale cosmological analyses. Together, these results lay the groundwork for fast, accurate treatment of non-cold relics in future cosmological analyses.

arXiv:2506.01956 → arXiv:2510.20821 → link to CLASSIER

Data-Driven Solutions to the Hubble Tension

2022 – 2026 · with Y. Ali-Haïmoud, N. Schöneberg, V. Poulin, M. Braglia, T. Zhou†

A three-part series using the Fisher-bias formalism to ask, quantitatively, what data-driven modifications to ΛCDM could resolve the Hubble tension without degrading the fit elsewhere. Paper 1 showed a perturbative time-varying electron mass can fully resolve the tension using Planck CMB data alone, though adding BAO and supernova data breaks the solution. Paper 2 instead explored scale-dependent modifications to the primordial power spectrum, again finding solutions that work for Planck alone but conflict with BAO and supernova data. Paper 3 revisited the recombination approach with newer ACT DR6 and DESI DR2 data, confirming the same oscillatory recombination modification is robust to Planck alone, but still fails once DESI DR2 BAO is included. Across all three, the same tension recurs: any modification that raises H₀ tends to lower the total matter density Ω_m, in conflict with late-time observations.

arXiv:2212.04494 → arXiv:2504.07966 → arXiv:2606.06495 →

Compressed Gaussian Likelihood for the Planck Low-ℓ Data

2026

Fisher-matrix analyses require an analytic Gaussian χ², but CMB low-ℓ likelihoods — including Planck's SRoll2 EE data, currently the tightest constraint on the reionization optical depth τ — are non-Gaussian. We show that an offset log-normal likelihood is exactly Gaussian in the log-transformed power spectrum amplitude, letting it serve as a proxy for the true likelihood in Fisher-matrix analyses without any explicit change of variables. Building on this, we compress the SRoll2 likelihood into a small number of piecewise offset log-normal fits, validate it against the full SRoll2 likelihood via MCMC (with Planck and ACT DR6 data) across standard and extended ΛCDM models, and release it as planck-gaussian-lowl, a lightweight public Python package.

arXiv:2606.05166 → link to planck-gaussian-lowl

Magnetic Fields from Small-Scale Primordial Perturbations

2024 · with Y. Ali-Haïmoud

Weak cosmic magnetic fields can be seeded before recombination via the Biermann-battery mechanism, sourced by density and temperature gradients acting on free electrons. We computed these fields self-consistently — including baryon-dark matter relative velocities — for both standard adiabatic perturbations and non-standard small-scale isocurvature perturbations. Standard adiabatic perturbations yield fields with rms ∼10⁻¹⁵ nG on ∼kpc comoving scales at cosmic dawn, a plausible seed for present-day galactic and cluster magnetic fields. Pushing to current upper limits on small-scale isocurvature perturbations could enhance this considerably, suggesting magnetic fields as a novel, if currently unobservable, probe of poorly constrained small-scale initial conditions.

arXiv:2404.03655 →

Probing Light Relics with 21-cm Cosmic Dawn Surveys

2023 – 2024 · with S. C. Hotinli

We assessed how much upcoming 21-cm surveys of cosmic dawn (12 ≲ z ≲ 30), like the Square Kilometre Array, can sharpen constraints on the effective number of relativistic species N_eff — a key probe of light particles beyond the Standard Model — on top of CMB-S4, the Simons Observatory, and DESI. Including SKA cosmic-dawn data alongside CMB-S4 tightens constraints to 2σ(N_eff) = 0.034, and once the degeneracy with the primordial helium fraction Y_p is accounted for, sensitivity to both N_eff and the dark matter density improves by more than a factor of two.

arXiv:2309.15119 →

Cosmic Birefringence via Polarized Sunyaev-Zel'dovich Tomography

2022 · with S. C. Hotinli, M. Kamionkowski

If the physics of the dark sector violates parity symmetry, CMB photons' linear polarization can rotate as they traverse the dark-sector background — "cosmic birefringence," for which recent CMB EB-spectrum measurements show ~3σ hints. We show that polarized Sunyaev-Zel'dovich (pSZ) tomography, which reconstructs the CMB quadrupole seen by free electrons at different redshifts, can independently probe the redshift dependence of this rotation and help calibrate instrumental polarization-angle systematics. As an example, pSZ tomography could probe axion-like dark energy with masses ≲ 10⁻³² eV, sourcing ∼0.1° of rotation between reionization and recombination.

arXiv:2207.05687 →

Probing Small-Scale Isocurvature Perturbations with the CMB

2021 · with Y. Ali-Haïmoud

Baryon and cold dark matter isocurvature perturbations are essentially unconstrained on sub-Mpc scales. We developed a formalism for how small-scale baryon perturbations, with arbitrary time and scale dependence, alter the mean free-electron abundance during recombination and thus imprint on CMB anisotropies. Applying this to Planck data across four isocurvature scenarios (pure baryon, pure CDM, compensated, and joint baryon-CDM), we found no evidence for such perturbations and set upper limits on their initial power spectrum on comoving scales 1 Mpc⁻¹ ≤ k ≤ 10³ Mpc⁻¹ — and showed this ingredient does not resolve the Hubble tension. A generalized Fisher forecast indicates a CMB Stage-4 experiment could probe 3–10× deeper than current Planck limits.

arXiv:2108.07798 →

Cosmological Recombination Code: HYREC-2

2020 · with Y. Ali-Haïmoud

Precision CMB analyses depend on an accurate recombination history, but the most accurate codes were too slow to run inside cosmological parameter searches, and the fast ones (e.g. RECFAST) weren't accurate enough. HYREC-2 closes that gap: an effective 4-level atom model captures the non-equilibrium behavior of highly excited hydrogen states, with a tabulated correction for Lyman-α radiative transfer, reproducing the accuracy of the original HYREC and COSMOREC codes while running in under a millisecond. It introduces no detectable bias in cosmological parameters even for an ideal, cosmic-variance-limited experiment out to ℓ = 5000, and has since been incorporated into CLASS and CAMB.

arXiv:2007.14114 → link to HyRec2