A particle that light can become
The axion was proposed in 1977 and 1978, by Roberto Peccei and Helen Quinn and then by Steven Weinberg and Frank Wilczek, to explain a puzzle in the theory of the strong nuclear force. The theory allows the strong force to treat a process differently from its mirror-image antimatter version, yet experiments find no such difference. A new field that removes the difference would come with a new, very light particle. Physicists now also study a wider family of axion-like particles that share its main property without solving that puzzle. PDG review of axions, 2025
For this story only one property matters. An axion can couple to two photons at once. If a magnetic field supplies one of them, a travelling photon can become an axion with the same energy, and the axion can turn back into a photon further on. The strength of that coupling is a single number, written gaγ, and every search described below is a way of putting an upper limit on it.
A photon only takes part when its electric field points along the magnetic field across its path. Light polarized the other way passes untouched. That detail will come back twice: in how a tangled field averages out, and in why the surviving light ends up slightly polarized.
Why a galaxy cluster is the right place to look
Each short stretch converts almost nothing. Over a short stretch the chance of conversion grows as the square of field strength times distance, so a search needs a long magnetized path. Galaxy clusters supply one. The gas between their galaxies carries fields of a few millionths of a gauss, roughly a hundred thousand times weaker than the Earth's field at the ground, but the light crosses hundreds of thousands of light-years of it.
The same gas also slows the conversion. It is a hot plasma full of free electrons, and the electrons make photons behave as if they had a tiny mass. Photon and axion then travel slightly out of step, and the mismatch is worse for low-energy photons. An axion with its own mass adds a second mismatch. Both effects push the conversion towards higher energies.
A tangled field makes a ripple pattern
A cluster field is not one straight field. It breaks into patches, each with its own direction and strength, roughly ten to thirty thousand light-years across. In each patch only one polarization of the light can convert, and the pieces of converted light from different patches add with phases that depend on energy. The result for one line of sight is an irregular pattern of dips across the spectrum, different for every possible tangle of the field.
Averaged over many tangles, the pattern settles into a simple rule. After enough patches the light shares itself evenly among three states, two photon polarizations and one axion, so at most one third ends up as axions. Because each patch removes light of one polarization only, the light that survives also becomes partly polarized. Mirizzi, Raffelt and Serpico, 2007
What Chandra measured
The bright galaxy at the centre of the Perseus cluster, NGC 1275, has an active nucleus that shines in X-rays. Christopher Reynolds and colleagues collected about 490 kiloseconds, close to six days, of Chandra grating spectra of it. Between 1 and 9 keV the spectrum is a smooth power law dimmed slightly by gas in our own Galaxy. The data scatter about that smooth curve by less than three percent.
They then built two models of the Perseus magnetic field, guided by measured electron densities and radio measurements of the field. For each model they generated 500 random field tangles and computed the ripple patterns an axion would leave, across a grid of masses and couplings. Couplings that would have made ripples bigger than the data allow are ruled out. For axions lighter than 10−12 eV they found gaγ<7.9×10−13GeV−1 at 99.7 percent confidence for their second field model. Reynolds et al., 2020
That number rests on the field model. When James Matthews and colleagues redid the analysis with other descriptions of the Perseus field, the limit at the lowest masses came out weaker by 0.1 to 0.3 in the base-ten logarithm, that is by a factor of about 1.3 to 2. Matthews et al., 2022
Stars give a separate check
Axions with this coupling would also be made inside stars, where photons in the hot core can convert in the electric fields of charged particles. The escaping axions carry energy away. Stars that burn helium in their cores, called horizontal-branch stars, would then use up their fuel faster, and an old star cluster would hold fewer of them compared with red giants. Counts in 39 globular clusters give gaγ<6.6×10−11GeV−1 at 95 percent confidence, with no magnetic field model involved. Ayala et al., 2014
The two methods answer different questions. The stellar limit is about 80 times looser but holds for a wide range of axion masses, up to roughly a few keV. The X-ray limit is far tighter, holds only for extremely light axions and inherits the uncertainty of the cluster field. So far neither the spectra nor the stars show an axion.
Try it in the Lab: Run the default tangled field and watch photons leave and rejoin their lane.