01 · RESEARCH

From quasar light to small-scale structure

Light from a distant quasar crosses the cosmic web before it reaches us. Diffuse hydrogen follows its filaments and sheets, absorbing the light at different wavelengths. Together, these absorption features form what we call the Lyα forest. The field above is one simulated slice of this web, with each row a sightline through it.

The forest lets us probe very small scales. A running primordial spectrum changes how much structure is seeded on different scales, while warm dark matter can suppress the smallest structures.

One sightline

Follow one row across the field and it becomes a spectrum. Most of the absorption comes from the diffuse hydrogen that makes up the forest. Sometimes the sightline crosses much denser neutral hydrogen in and around galaxies, producing broad damping wings that can cover a large part of the spectrum.

These dense absorbers are often difficult to see directly in emission. Against a bright quasar, however, their absorption lets us probe neutral gas in and around galaxies that would otherwise be hard to see. Measuring them means inferring the quasar light hidden beneath the damping wings, together with the absorber's redshift and column density. Across many quasars, those inferences constrain the absorber population.

Many sightlines

Keep enough sightlines and the ξ curve above begins to develop structure. The individual absorption lines fade from view, but their correlations remain. Those correlations contain information about the small-scale clustering of matter, as well as the thermal state of the gas.

Simulations connect forest statistics to small-scale matter power.

The dense absorbers return

Dense absorbers are rare, but their broad damping wings still affect forest statistics after many sightlines are combined. Changes in the absorber population can shift the same statistics used for cosmology, sometimes in ways that resemble changes in small-scale matter power.

Their population therefore enters the cosmological likelihood.

light → forest → sightlines → statistics → matter power

02 · PROJECTS

Selected work

a sightline read from the field · simulated · PRIYA
Small-scale cosmologyJCAP 2026

Small-scale Lyα cosmology with high-resolution spectra

High-resolution quasar spectra resolve the Lyα forest out to wavenumbers of order k ~ 10 h Mpc^-1, smaller scales than those usually used in cosmological analyses. Using KODIAQ-SQUAD and XQ-100, we constrain the amplitude and slope of the small-scale matter power spectrum.

The amplitude is consistent with DESI DR1 and Planck; the inferred slope is somewhat higher.

How it works
Spectra
KODIAQ-SQUAD and XQ-100
Model
PRIYA multi-fidelity emulators, with dense absorbers included in the likelihood
Inference
Small-scale matter-power amplitude and slope
synthetic · illustrative
Absorbers · inference2020 → DESI

Finding damped absorbers automatically

Some sightlines contain more than the diffuse forest. A dense, self-shielded neutral-hydrogen system can remove enough quasar light to strongly affect downstream analyses. That is why we need a DLA finder.

Rather than making only a yes-or-no detection, it keeps a posterior over the absorber's redshift and column density. It is now one of the complementary finders used for the DESI DR2 DLA catalogue.

How it works
Missing light
Infer the unabsorbed quasar continuum with a Gaussian process
Absorbers
Compare models with zero, one, or more systems
Output
Posterior over redshift and column density

From SDSS DR12 to DR16Q and now DESI DR2.

one sightline, resolution rising · simulated · PRIYA
Emulation · uncertainty2022 – 2023

Multi-fidelity emulation

High-resolution cosmological simulations are expensive, so there are never enough of them to densely cover the parameter space needed for inference. Multi-fidelity emulation combines many lower-resolution simulations with a smaller number of high-resolution runs and learns the correction between them.

This lets the emulator use information from both resolutions without treating the lower-resolution simulations as exact.

How it works

Learn the low-to-high-fidelity correction as part of the Gaussian-process model.

First used for matter power, then extended to the Lyα forest.

simulated · PRIYA
Dense absorbers · emulationin preparation

What dense absorbers do to the Lyα forest

Even a small number of dense absorbers can change forest statistics through their broad damping wings. Changing the absorber population changes those statistics as well, sometimes in ways that resemble changes in small-scale matter power.

HCDEmu models this dependence by emulating how the forest responds to different absorber populations. That response then enters the likelihood.

How it works

Split PRIYA synthetic spectra by absorber class and emulate each class's response.

The change in ξ when you kept a dense sightline above is the same effect in a simpler setting.

Population inferencePRD 2024

One population of black holes, or two?

GWTC-3 has a very different population-inference problem. The objects are merging black holes rather than absorbers, but the inference is again about a population rather than a single object.

Using hierarchical population inference, we constrained how much two proposed black-hole populations could mix.

How it works

Hierarchical population inference on GWTC-3.

Related work includes microlensing searches for primordial black holes.

More work
03 · PLAY

Play

PRIOR → OBSERVE → POSTERIOR · illustration on a toy model

Cosmic Battleship

There is a universe you cannot see, and ten chances to look at it. Each sightline gives you a noisy piece of the map. After every observation, some possibilities become more likely and others fade away, and you have to decide where to look next.

The opponent is doing the same thing.

A second mode uses absorption lines instead of a map. Sometimes the data identify the answer. Sometimes they do not.

A concept game designed for a science exhibit. The model underneath it is simple enough to check by hand.

status · prototype
stack · web, no backend
Not yet public

Also playable: the PRIYA parameter sweep ↗, a small live widget: move a cosmological or astrophysical parameter and watch the forest power spectrum respond.

04 · PAPERS

Papers

First-author work first. Full list below.

  1. 2026
    JCAP 07 (2026) 094 · arXiv:2509.18271
  2. 2024
    M.-F. Ho, S. E. Perkins, S. Bird, W. Dawson, N. Golovich, J. R. Lu, P. McGill
    Phys. Rev. D 110, 063031 · arXiv:2408.09024
  3. 2023
    MNRAS 526, 2903 · arXiv:2306.03144
  4. 2022
    MNRAS 509, 2551 · arXiv:2105.01081
  5. 2021
    MNRAS 507, 704 · arXiv:2103.10964
  6. 2020
    MNRAS 496, 5436 · arXiv:2003.11036
Google Scholar ↗ ORCID ↗
05 · ABOUT

Ming-Feng Ho 何銘峰

NOW

I am currently a postdoctoral fellow at the University of Michigan. Most of my work is on the Lyα forest, dense absorbers, and the inference problems that come with them. I have also worked on emulation and black-hole population inference.

I did my PhD at UC Riverside with Simeon Bird, after studying physics and astrophysics at National Taiwan University. As an undergraduate, I also spent much of my time in the philosophy department. I pursued a philosophy minor for several years, although I never formally completed it, and for much of college I probably took philosophy more seriously than physics.

Outside physics, I play cello, brew coffee, cook, and make wagashi and Suzhou-style mooncakes. I also like hiking, road trips, and exploring America's natural landscapes.

Before graduate school, I also worked on text mining and digital editions of classical Chinese texts. I no longer work in philosophy, but it probably shaped how I think more than astronomy did. That training still shows up in how I do science. I tend to spend a lot of time on the assumptions behind a problem, on whether the question is framed correctly, and on things that are easy to leave unexamined.