Mapping the Invisible: What Multi-Wavelength Astronomy Reveals About the Cosmos

Space2 min

Overlapping translucent sheets, each with a partial pattern, form a clear nebula where they align; a weathered hand holds an old creased one.

Human eyes see from roughly 380 to 750 nanometres. That is a factor of two in wavelength, one octave, and astronomers now work across something like forty of them, from radio waves a metre long to gamma rays far smaller than an atom. If the electromagnetic spectrum were a piano with forty octaves, we were born hearing one key’s worth of it and spent most of history assuming that was the song.

That is the usual pitch for multi-wavelength astronomy, and it is half right. The “sliver” framing is true on a log scale and misleading on an energy scale: the Sun pours a large share of its output into exactly the octave we see, which is not a coincidence given whose eyes evolved under it. What makes the other thirty-nine octaves matter is not that there is more light out there. It is that every band answers a different question, and the interesting answers only show up when you hold two of them up against each other.

I think the pretty composite images, the ones with X-rays painted purple over infrared painted orange, are the least interesting product of this whole enterprise. They are a UI. The real product is the join.

§Most of the starlight ever made was eaten by dust

Start with the strongest version of the “drowning in light we can’t see” claim, because there is one, and it is better than the slogan.

Galaxies are dusty. Dust grains absorb ultraviolet and visible starlight, warm up a little, and re-radiate that energy in the far infrared. Add up that glow across the whole sky and the history of the universe and you get the cosmic infrared background. In 2000, Gispert, Lagache and Puget estimated that the dust-reemitted part carries about 1 to 2.6 times as much power as the optical and near-infrared background combined.[1] Do the division and somewhere between half and nearly three-quarters of all the energy stars have ever put out reaches us as far-infrared glow from dust, not as starlight. Locally the ratio is about 30%, which tells you the early universe was a much smokier place than the one outside your window.

So the honest version of the hook is this: an optical telescope does not see a sliver of the cosmos, it sees a censored copy. The most violent star formation happens inside the densest dust, which means the regions with the most going on are precisely the ones that hide best from visible light.

The second reason to go infrared is the expansion of space, and it is where the arithmetic gets fun. In January 2026 NASA confirmed a galaxy called MoM-z14 at a redshift of 14.44, seen about 280 million years after the Big Bang.[2] Redshift zz stretches every wavelength by a factor of 1+z1+z, so here by about 15.4. Hydrogen’s Lyman-alpha line, emitted at 121.6 nanometres in the far ultraviolet, arrives at roughly 1.88 microns — more than twice as far into the infrared as the reddest light your eye can register. The galaxy was blazing in ultraviolet. By the time that light reaches us it is invisible heat. JWST is not an infrared telescope because infrared is exotic; it is one because the early universe has been redshifted out of the visible entirely.

§A green monster, settled by a second opinion

My favourite recent example of the join is small and almost comic. Cassiopeia A is the remnant of a star whose explosion became visible from Earth about 340 years ago. When JWST’s mid-infrared instrument imaged it in 2023, the team found a large, strange loop crossing the central cavity and nicknamed it the Green Monster, after the wall at Fenway Park.[3]

The infrared data alone could not say what it was. Its position suggested stellar debris from the explosion. Its infrared properties suggested something else: circumstellar material, gas the star had shed before it died. Two readings of one instrument, pointing in different directions.

Chandra had been watching Cas A in X-rays for decades, so the archive already held the answer. Shock waves in the remnant heat gas to tens of millions of degrees, and at those temperatures the X-ray spectrum is a chemical fingerprint. The debris from deep inside the star is rich in silicon and iron. The Green Monster was not: its X-ray spectrum matched the filaments along the outer blast wave, where the shock is ploughing into gas around the star. Its spectrum was also blueshifted by roughly 2,300 km/s, which places it on the near side of the remnant, moving towards us.[4] Jacco Vink, who led the X-ray analysis, put it plainly: “Chandra helped us clinch the case.”

As an engineer this is the part I love, because it is exactly how you debug a distributed system. One log stream gives you an anomaly and two plausible stories. You do not stare harder at the same log. You go find a second source with independent failure modes and correlate on a shared key — here, position on the sky. The infrared told you there was something. The X-rays told you what it was made of and which way it was moving. Neither alone settles it.

§The Bullet Cluster is a join that found a missing column

The grander version of the same move is the Bullet Cluster, about 3.8 billion light years away: two clusters of galaxies that passed through each other.[5]

Chandra sees the hot gas between the galaxies, which holds most of the ordinary matter in a cluster. When the clusters collided, the gas clouds hit each other, felt drag, and slowed down, leaving two glowing X-ray blobs stalled near the middle. Separately, astronomers mapped where the mass is by gravitational lensing, measuring how the cluster bends light from galaxies far behind it. That map is not an image in any band. It is inferred from distortions in an optical image.

Lay the two maps over each other and they disagree. Most of the mass is not where most of the ordinary matter is. It sailed straight through the collision with the galaxies, while the gas got stuck. If you believed the hot gas dominated the mass, as some alternatives to dark matter would like, this picture should not exist.

In database terms, you joined the X-ray table with the lensing table on sky position and found rows that only exist on one side. Neither table contains dark matter. The mismatch does. That, to me, is the deepest thing multi-wavelength astronomy has to say: sometimes the object you are looking for does not shine in any band, and the only way to see it is as a disagreement between bands that do.

§The “Earth-sized telescope” is mostly missing

Then there is the Event Horizon Telescope, which gets described everywhere as an Earth-sized telescope. It deserves the awe and not quite the description.

The physics of resolution is unforgiving. The finest angle a telescope can separate scales with wavelength over aperture:

θ≈λD=1.3×10−3 m1.27×107 m≈1.0×10−10 rad≈21 μas\theta \approx \frac{\lambda}{D} = \frac{1.3\times10^{-3}\ \text{m}}{1.27\times10^{7}\ \text{m}} \approx 1.0\times10^{-10}\ \text{rad} \approx 21\ \mu\text{as}

That is 1.3-millimetre radio waves across the diameter of the Earth, and it lands close to the roughly 20 microarcseconds the collaboration quotes. To feel that number: the human eye resolves about one arcminute, which makes this roughly three million times sharper. The ring around M87’s black hole came out at 42 ± 3 microarcseconds, around a mass of about 6.5 billion suns.[6] At the Moon’s distance, an angle of 42 microarcseconds spans about eight centimetres. The EHT photographed something that, on the sky, is the size of a doughnut sitting on the lunar surface.

But there is no Earth-sized dish. There are eight observatories, spread from Hawaii to Spain to the South Pole, each recording radio waves against an atomic clock, with the recordings shipped and correlated later.[7] Each pair of stations samples one spatial frequency of the image at a time, and the Earth’s rotation sweeps those samples across the sky over a night. A real dish of that size would collect every spatial frequency at once. The EHT collects a sparse scatter of them and leaves enormous holes, and the image is a reconstruction that has to fill those holes with assumptions about what images are likely to look like.

This is not a scandal; it is how all interferometry works, and the collaboration spent years stress-testing its reconstructions. It does mean the right mental model is closer to compressed sensing than to a camera. When the picture you are looking at is barely twice the instrument’s resolution, you should hold it a little more loosely than the headlines did. The thing to trust most is the ring’s size, which the data constrains directly, rather than the exact shape of every bright knot.

And Sagittarius A*, our own galaxy’s black hole, is harder still. It is about four million solar masses and 27,000 light years away. Because it is so much smaller than M87*, the gas around it orbits fast enough that the source changes over minutes, during an observation built on the assumption that the source holds still.[7] Imaging it at all is a small miracle of statistics.

What I like is that even the EHT is not really a single-band story. During the 2017 campaign, the M87 observations were joined by telescopes across the spectrum, from radio arrays to gamma-ray observatories, to build the most complete simultaneous spectrum of the jet ever taken. The paper’s conclusion was that no simple one-zone model fits it; you need a structured jet.[8] The ring is the photograph everyone has seen. The join is where the physics came out.

§We keep building new eyes and starving the old ones

Here is the opinion I actually came to write. Multi-wavelength astronomy is a property of a fleet, not of any single telescope, and fleets are fragile in a way no individual mission is. The join only works if the tables overlap: same patch of sky, often the same months, sometimes the same hour. A brilliant new infrared telescope with no working X-ray partner is a database with a foreign key pointing at a dropped table.

Which is why Chandra’s budget story bothers me more than it probably should. Chandra launched in 1999 and is still the sharpest X-ray eye we have. NASA’s 2025 budget request proposed cuts deep enough that the community spent two years fighting what looked like a slow shutdown. As of the Chandra director’s report in May 2026, Congress had held the line for fiscal 2026 at $63 million, $11 million short of what full operations cost, with the observatory working normally apart from an anomaly on one detector and a planned drop to a reduced “minimum-cost mission” sometime in 2027. The 2027 request did not mention Chandra at all.[9]

I understand the logic of retiring old hardware. In software, keeping a legacy service alive forever is how you end up with a codebase nobody can move. But the Green Monster is a neat counterexample: the answer to a 2023 question from the newest telescope in space was sitting in decades of archived data from a 1999 one. Old instruments are not only legacy. They are the baseline that makes new anomalies legible. Lose the X-ray eye and every future infrared surprise gets one fewer second opinion.

So when you see the next composite image, purple and orange and impossibly detailed, look at it the way you would look at a dashboard. The colours are arbitrary. The alignment is the result. And the instruments that made it are a shared, aging piece of infrastructure that somebody has to keep paying for, long after the launch-day photos stop trending.

  1. ^R. Gispert, G. Lagache and J.-L. Puget, Implications of the cosmic infrared background for light production and the star formation history in the Universe, A&A (2000). The abstract puts dust-reemitted power at about 1 to 2.6 times the optical/near-IR background, against roughly 30% measured locally. The fractions in the text are my arithmetic on their range.
  2. ^NASA, NASA Webb Pushes Boundaries of Observable Universe Closer to Big Bang (January 2026). MoM-z14, redshift 14.44, confirmed with NIRSpec, about 280 million years after the Big Bang.
  3. ^Chandra X-ray Center, NASA Telescopes Chase Down “Green Monster” in Star’s Debris (January 2024).
  4. ^J. Vink et al., X-ray diagnostics of Cassiopeia A’s “Green Monster”: evidence for dense shocked circumstellar plasma, ApJL 964, L11 (2024). The blueshift of about −2,300 km/s is what puts the structure on the near side of the remnant.
  5. ^Chandra X-ray Center, 1E 0657-56: NASA Finds Direct Proof of Dark Matter. “Direct proof” is the press office talking; the paper behind it is more careful, but the separation of gas and mass is the result.
  6. ^The Event Horizon Telescope Collaboration, First M87 Event Horizon Telescope Results. I. The Shadow of the Supermassive Black Hole, ApJL 875, L1 (2019): a ring of 42 ± 3 μas at 1.3 mm, and a mass of (6.5 ± 0.7) × 10⁹ suns.
  7. ^MIT News, Astronomers reveal first image of the black hole at the heart of our galaxy (May 2022). Eight observatories, about five petabytes of data across both targets.
  8. ^The EHT MWL Science Working Group et al., Broadband Multi-wavelength Properties of M87 during the 2017 Event Horizon Telescope Campaign, ApJL 911, L11 (2021). The author list alone, with Fermi-LAT, H.E.S.S., MAGIC and VERITAS beside the EHT, makes the point.
  9. ^P. Slane, Director’s report to the Chandra Users’ Committee (6 May 2026). FY26 set at $63M, $11M below full-mission cost; a transition to a reduced “minimum-cost mission” sometime in 2027; the FY27 request “makes no mention at all about Chandra.” Fiscal 2027 began on 1 October and I have not seen a settled number for it.