Skip to content

Observational astrophysics

Tracing black holes through
light & time.

I study black hole growth in the early Universe, with a focus on little red dots — their physical nature, their place among active galaxies, and their connection to their hosts.

I’m Zijian Zhang, a PhD student at the Kavli Institute for Astronomy and Astrophysics, Peking University, advised by Prof. Linhua Jiang. From September 2026 to September 2027, I am visiting Prof. Roberto Maiolino’s group at the University of Cambridge. I received my Bachelor’s degree from Nanjing University in September 2023. I’m also interested in machine learning and data-driven astrophysics.

中文简介

我叫张子键，是北京大学科维理天文与天体物理研究所的博士研究生，导师是江林华教授。2026年9月至2027年9月，我在剑桥大学Roberto Maiolino教授课题组访问。主要研究兴趣为活动星系核及其宿主星系的多波段特性、高红移活动星系核／星系的演化，以及机器学习／数据驱动的天体物理学研究。在来北大之前，我于2023年9月在南京大学获得学士学位。

Explore my research Curriculum vitae
Research notebook 01Central engines 02Populations 03Pixels to spectra 04Black holes & hosts Earlier work

Early black holes · from the inside out

One question.
Time, populations, spectra & space.

What can little red dots tell us about the growth of early black holes? I follow them through time and across populations, work on the JWST imaging and slitless spectroscopy behind these measurements, and am now extending the picture to their host galaxies.

I · The nature of little red dots
Field note / 01Synthetic
Variability on two timescales Two synthetic views. Large sample: 314 little red dots observed at a few epochs over a few years, plotted as the change between epochs against brightness; most lie inside the grey measurement-noise band and eight variable candidates lie above it, with an example light curve below. Lensed LRD: a quasi-periodic pulsation of one source's envelope, seen as four lensed images whose light left the source up to about 130 years apart, with a scrolling light curve marking each image's moment and a few-year JWST band. 314 LRDs · multi-epoch JWST noise (3σ) 8 candidates 0.5 0 Δmag brighter fainter → Typical LRD 0 0.4 mag 4 epochs over ~2 years →

Many LRDs, a few epochs over a few years: most stay within the noise, and a minority vary.

Synthetic data. Large sample: only the sample size and the number of candidates follow the paper. Lensed LRD: only the ~130-year span between the images does. Not measured variability.

01Time-domain astrophysics

Inside
little red dots.

What powers these compact red sources?

I use variability to probe the origin of their light and the gas around their central engines. Different timescales give complementary clues to their physical state.

Key results

  1. 2025ApJ

    1. First author

      Analysis of Multi-epoch JWST Images of ~300 Little Red Dots: Tentative Detection of Variability in a Minority of Sources

    Multi-epoch JWST imaging of 314 little red dots: on average they show no strong variability, and eight are strongly variable candidates.

    Paper Light curves

  2. 2025arXiv · under review at Science

    1. First author

      Little red dot variability over a century reveals black hole envelope via a giant Einstein cross

    A quadruply lensed LRD whose images are separated by up to ~130 years in light-travel time. Colour and brightness differ by up to 0.7 mag between images, which we interpret as pulsations of an extended, star-like envelope.

    arXiv

  3. 2026arXiv · under review at PRL

    1. First author

      Little red dots as a cosmological probe: constraining H₀ with quasi-periodic pulsations

    If such pulsations follow a period–luminosity–temperature relation that can be validated and calibrated, LRDs could measure distances. A proof-of-concept H₀ estimate, with forecasts for decade-long monitoring.

    arXiv

NowMonitoring nearby LRD analogs in photometry and spectroscopy.

Field note / 02Synthetic
Colour alone versus spectra Synthetic sources first sorted by colour alone, where a colour cut selects red candidates. With spectra, the candidates separate into broad-line little red dots, narrow-line little red dots and strong-line mimics whose lines imitate a red colour; little blue dots also show broad lines. Two ringed sources are X-ray and radio bright, and hollow sources have no spectrum yet. Below, an example spectrum is compared with its broadband photometry. colour cut red candidates Little blue dots Broad-line LRDs SF galaxies Line mimics Narrow-line LRDs no spectrum yet X-ray/radio broad Hα narrow Hα ← bluer redder → photometric colour Broad-line LRD spectrum bands flux [O III] Hα λ →

Colour only: a colour cut selects red candidates, but colour alone cannot tell them apart.

Synthetic sources and toy spectra with illustrative proportions; not a survey sample, a selection criterion or an evolutionary sequence.

02Spectroscopy & demographics

Little red dots
in context.

Where do LRDs fit among growing black holes?

I compare little red and little blue dots to test how viewing geometry and intrinsic physical state shape their appearance. Spectra let us look beyond colour.

Key results

  1. 2026ApJ

    1. First author

      JWST Insights into Narrow-line Little Red Dots

    Five LRDs with only narrow Hα (≈250 km/s): either dusty, compact star-forming galaxies or black holes of 10⁵–10⁶ M☉. About 20% of previously selected candidates owed their V-shaped SED to strong lines rather than a red continuum.

    Paper

  2. 2026Nature Astronomy

    1. Co-first author

      The discovery of two little red dots in transition into quasars

    Two LRDs at z ≈ 2.9 that are also bright in X-rays, radio and the mid-infrared: hybrid properties we interpret as LRDs in transition toward typical quasars.

    Paper

NowA controlled LRD/LBD comparison.

Methods · From JWST data to measurements

03JWST imaging & slitless spectroscopy

From pixels
to spectra.

Which galaxy produced this line, and at what redshift?

Slitless spectroscopy disperses every source in the field at once. It is efficient, but traces overlap and most spectra show a single line. In COSMOS-3D, we turned NIRCam imaging and F444W grism data into a census of ~12,000 emission-line galaxies at z ≈ 0–9.

Each identification combines template cross-correlation, a Siamese network that compares the 2-D line with candidate galaxy images, and an LLM agent that weighs redshift, equivalent-width, SED and morphology evidence. Several authors then checked the results by eye.

Field note / 03Synthetic

  1. 1 Image

    Target (ringed) and a neighbour

  2. 2 Grism

    Traces overlap; one line

  3. 3 Spectrum

    One line at 4.30 μm

  4. 4 Evidence

    Shape match
    The 2-D line matches the target's shape better than the neighbour's.
    Photo-z
    Two redshift solutions, z ≈ 1.3 and z ≈ 5.5; the higher one is favoured.
    Blue bands
    Faint in F606W and F814W; detected in F150W and F444W.
    Line EW
    large

  5. 5 Decision

    LLM proposal Hα · z = 5.55
    Human review ✓ Confirmed Compact, faint in blue bands, strong line

Example 1 · A compact source that is faint in the bluest bands and has a strong line. The LLM proposes Hα at z = 5.55; review confirms it.

Synthetic sources, spectra and evidence. The workflow follows the paper; the values in this figure are illustrative, not catalogue entries.
  • ~12,000emission-line galaxies at z ≈ 0–9 in COSMOS-3D
  • 88.3%of LLM redshifts agree with the final human-validated redshifts

The most common disagreement is Paα read as Hα, a degeneracy that people resolve using morphology, imaging and equivalent width.

My contribution: NIRCam image reduction, catalogue construction and comparison, the LLM line-identification step, and part of the visual inspection.

II · Black holes and their hosts
Field note / 04Synthetic
From an instrumental velocity gradient to a dynamical mass Two synthetic steps. Slice bias: a compact source seen through the IFU image slicer; light falling off-centre within each slice shifts that slice's spectrum, so an apparent velocity gradient appears although no gas motion is put in. Dynamical mass: a rotation velocity and radius give the total mass inside that radius; because the mass scales with velocity squared, an instrumental gradient left in the data inflates it. Through the slicer Each slice's spectrum true λ λ → velocity map No gas motion the gradient comes from the slicer alone approaching receding

No gas motion is put in, yet light falling off-centre within each slice shifts that slice's wavelength, so a compact source shows an apparent velocity gradient across the slices.

Synthetic illustration of the instrumental effect and the goal. No measured cubes, velocities or test results are shown.

04Spatially resolved spectroscopy

From nuclei
to host galaxies.

Can gas motions weigh a little red dot?

Integral-field spectroscopy records a spectrum at every position. In compact sources, the JWST/NIRSpec image slicer can imprint an apparent velocity gradient, so I am developing a calibration that removes it.

The aim is a dynamical mass for LRDs: the total mass within the measured radius, not the black hole alone.

In progress

Earlier work · X-ray AGN & gamma-ray bursts

Before little red dots.

Before JWST, I used X-ray and gamma-ray observations to study variable accretion, obscuration and relativistic jets.

  1. 2023ApJ

    X-ray-weak quasars

    1. First author

      XMM-Newton Observations of Two Archival X-Ray Weak Type 1 Quasars: Obscuration Induced X-Ray Weakness and Variability

    Two quasars that were X-ray weak in archival Chandra data returned to normal X-ray levels in XMM-Newton (factors of ~4 and ~9) with little UV/optical change, pointing to variable obscuration.

    Paper

  2. 2025ApJ

    Extreme X-ray variability

    1. First author

      On the Extremely X-Ray Variable Active Galactic Nuclei in the XMM-LSS Field

    Eight AGN in the XMM-LSS field varying by factors of ~6–12: accretion-state changes in five, including a changing-look AGN confirmed with new MMT spectra, and changing obscuration in three.

    Paper

  3. 2023RAA

    GRB 220408B

    1. First author

      GRB 220408B: A Three-episode Burst from a Precessing Jet

    A three-episode gamma-ray burst detected by the GRID CubeSat mission, whose similar, correlated episodes are explained by a precessing jet.

    Paper

From questions to papers

First-author papers.

All publications

My first-author papers, grouped by theme. Collaborations are listed on the full publications page.

Little red dots

X-ray AGN & gamma-ray bursts

Along the way

Latest updates.

All updates

  1. Now out in Nature Astronomy: The discovery of two little red dots in transition into quasars (Fu, Zhang et al.).

A conversation starts here

Let's talk about the Universe.

zjz.kiaa@stu.pku.edu.cn