Report 169 · Lab Science
"Astronomers detect" is how these stories are always written, and the telescope gets the photo. But the 2021 discovery of the first individual polycyclic aromatic hydrocarbons in interstellar space rested on a statistical method and on a laboratory spectrum measured to about a kilohertz. For one of the two molecules, not a single line rose above the noise.
Polycyclic aromatic hydrocarbons (PAHs) are flat molecules made of fused carbon rings, the same family as the char on grilled food. Astronomers have long suspected they are everywhere in space, because broad infrared emission bands seen across the sky look like PAH vibrations. The problem is that those bands are blends; the paper below notes that the frequency differences between individual PAHs are smaller than the width of the observed bands, so no single molecule could be named from them.
In March 2021, a team led by Brett McGuire (MIT, NRAO and the Harvard-Smithsonian Center for Astrophysics) reported in Science the detection of two specific PAHs, 1-cyanonaphthalene and 2-cyanonaphthalene, in a cold dark cloud called TMC-1, using the 100-metre Green Bank Telescope. Coverage called it "the first definitive proof" of PAHs in molecular clouds. The claim is well supported. How it was supported is the interesting part, and it is mostly a laboratory story.
Radio telescopes find molecules by their rotational spectra: a polar molecule tumbling in space emits at a set of sharp frequencies fixed by its shape and mass. The paper notes that more than 80 percent of known interstellar molecules were discovered this way. The catch is the word "polar." Naphthalene itself, two fused benzene rings, has no permanent dipole moment, so it has no pure rotational spectrum and is invisible to this method. Swap one hydrogen for a nitrile (–CN) group and you get a strongly polar molecule whose rotational lines fall right in the band the survey covered.
That is why the detected molecules are cyano-PAHs and not the plain hydrocarbons. It is a constraint of the physics, not a quirk of the chemistry of the cloud.
To know which frequencies to look at, you need the molecule's spectrum measured on Earth. For the cyanonaphthalenes, the paper relies on laboratory rotational spectra published by McNaughton and co-workers in 2018. NRAO's own release for a later detection in the same cloud describes the rotational spectrum as "like a fingerprint" and credits the lab work at the Center for Astrophysics directly. That is the honest framing: the telescope collects the signal, the laboratory supplies the key.
Here is the part the headlines skipped. In the survey's first data release, the authors write, searches for both molecules "showed no individual rotational lines above the noise level of the observations." Even with the deeper second data release, their best-fit model predicted about a dozen 1-cyanonaphthalene lines above the noise in parts of the data, and none for 2-cyanonaphthalene. For 1-cyanonaphthalene they report at least five lines above four sigma. For 2-cyanonaphthalene, nothing you could point to.
So they used spectral stacking with matched filtering. In plain terms: cut a small window of spectrum around every frequency where the molecule should emit, weight each window by how strong the line is predicted to be and how noisy that stretch of data is, and average them. Random noise averages toward zero. A real signal that sits at exactly the predicted frequency in every window adds up. The averaged model is then used as a filter to measure how strongly the averaged data match it.
The stacks drew on about 1,500 catalogued transitions for 1-cyanonaphthalene and about 950 for 2-cyanonaphthalene within the survey's range. The result was a lower-limit significance of 13.5 sigma for 1-cyanonaphthalene and 17.1 sigma for 2-cyanonaphthalene. The molecule with no individually visible lines produced the stronger detection. The appendix adds a useful nuance: most of the significance comes from the strongest few lines, with 5 sigma reached after averaging only the top 5 to 10. Stacking pulls a real but faint signal out of noise; it does not conjure one from nothing.
Stacking works only if you know exactly where each line is. If your predicted frequency is off, you average noise instead of signal. The authors tested this directly in their appendix, and the numbers are the reason I wanted to write this report:
The frequencies had to be right to about a kilohertz. The authors estimate that individual line frequencies need to be known to roughly 1 kHz for the method to produce a detectable signal, which is close to the telescope data's 1.4 kHz resolution. These lines sit in the tens of gigahertz, so that is a precision of a few parts in a hundred million.
They checked what happens when you are wrong. Adding random error to the lab-derived rotational constants made the response collapse; the lab constants, accurate to about 5 parts in 100 million, sat right where the full signal was recovered. In a second test they injected an entirely fictitious molecule into the real data and found it was not reliably recovered until line frequencies were accurate to about 250 Hz.
They checked that noise alone does not do it. Stacking 1,000 randomly generated line catalogs against the real data gave a maximum response of 3.2 sigma, and 1,000 synthetic noise spectra gave at most 3.1 sigma, both far below the detections.
They guarded against impostors. Any window containing a strong feature above five sigma was thrown out so another molecule's line could not leak in, and jack-knife tests (splitting the line list in two and stacking each half separately) gave a detection in each half. As the appendix puts it, a signal emerges only when emission is present "in every window at precisely the predicted frequency."
That is why the authors can conclude that "any molecule with different constants could not explain the signal." The specificity comes from the laboratory measurement. Without it, there is no detection.
My own published research is in microwave spectroscopy, the same family of technique that produced these laboratory spectra. I had no part in these studies, and nothing here comes from my work; the evidence is the papers below. I chose this story because it is the clearest case I know of a lab measurement doing the decisive work in a discovery credited to a telescope, and because the precision involved is easy to state and hard to appreciate. It also connects to Report 163: in both cases, what you can claim depends on knowing exactly what your measurement assumes.
I read the arXiv version. My numbers and quotes come from the arXiv preprint (2103.09984, version 1, posted 18 March 2021), which lists the Science citation. The published Science article is paywalled and I did not read it, so I cannot rule out small differences in wording; I found no later arXiv version.
I did not read the 2018 laboratory paper. I confirmed its bibliographic record through Crossref and rely on the Science paper's description of its accuracy.
Stacking has critics in principle, not here in particular. The authors themselves say care is needed with interlopers and noise, and they present robustness tests. I found no published challenge to this specific detection, and later work from the same survey has reported larger PAHs in the same cloud, but I did not review that literature in depth.
The first individual PAHs in interstellar space were found by averaging hundreds of rotational lines that were, one by one, invisible. That only works because a laboratory had measured where those lines are to about a kilohertz, and the authors showed the detection falls apart when the frequencies are even slightly wrong. When you read "astronomers detect a new molecule," there is almost always a spectroscopy lab behind the photo of the telescope.
Scope note: this report explains a published detection method from the primary paper. No measurement was performed. The author's own research is in microwave spectroscopy and was not involved in any study cited here.
U.S. Army combat veteran (Counter-IED / Electronic Warfare), peer-reviewed researcher in microwave spectroscopy, and founder & CEO of Shroombiosis. Consults on laboratory operations, AI, and supplement formulation.