In this paper
AbstractIntroductionAxion–Photon OscillationSingle Domain: Constant Magnitude and DirectionMulti-Domain: Fixed Domain LengthMulti-Domain: Variable Domain LengthUltra High Energy Photons from GRB 221009APropagation of UHE Photons in the Presence of AxionsResults and DiscussionAcknowledgmentsReferencesWorksPhysics and Astronomy
Axion-Photon Conversion in Variable-Domain Magnetic Fields and the Highest-Energy Photons from GRB 221009A
Abstract
We investigate photon-ALP conversion as a simultaneous explanation of the ultra-high-energy photons observed from GRB 221009A by LHAASO and of the TeV photon reported by Carpet-3. We model the magnetic field of the host galaxy as a set of domains with randomly oriented field directions and variable domain lengths, and propagate the photon-ALP system through the host galaxy, the intergalactic medium and the Milky Way. Fitting the combined LHAASO and Carpet-3 spectra, we find that a broad distribution of domain lengths raises the photon survival probability at the highest energies and accommodates the Carpet-3 point with photon-ALP conversion alone. The best fit, ~eV and , lies below the CAST bound, and parts of the and regions are also compatible with it.
Introduction
Axions and axion-like particles (ALPs) are among the most well-motivated extensions of the Standard Model, arising generically as pseudo-Nambu-Goldstone bosons (pNGBs) of a spontaneously broken global chiral symmetry [1, 2, 3]. While the QCD axion initially arose as a by-product of the Peccei-Quinn solution to the strong CP problem [4, 5, 6, 7, 8], a much broader class of ALPs emerge naturally in string-theoretic constructions and other ultraviolet completions [9, 10], where they need not carry any relation to the QCD scale. Consequently, ALPs1 span an enormous range of viable masses, from eV to several GeV, and couple only extremely weakly to Standard Model fields, thereby also making them good candidates for dark matter [11, 12, 13, 14]. A generic and phenomenologically central feature of these particles is their coupling to two photons, parametrized by the coupling constant [15, 16]. This coupling gives rise to two phenomenologically important processes: (a) in the presence of an external magnetic field, the photon can convert into an axion and vice versa, leading to axion–photon oscillation [17, 18], and (b) an axion with sufficient mass can decay into two photons [16, 19, 20]. These two processes serve as the basis of the majority of the astrophysical searches for axions and ALPs.
Ultra-high-energy (UHE) photons with energies in the range of a few GeV to several TeV are produced in some of the most extreme astrophysical environments in the universe, such as active galactic nuclei (AGN), gamma-ray bursts (GRBs), and quasars [21, 22, 23]. In these environments, which typically host strong magnetic fields [24, 25], charged particles are accelerated to ultra-relativistic energies and subsequently interact with the surrounding medium and radiation fields, producing UHE photons via processes such as synchrotron emission, inverse Compton scattering, and hadronic interactions [26, 27]. UHE photons having energies above 1 TeV, while traveling through the intergalactic medium, interact with the extragalactic background light (EBL) and get heavily attenuated. The dominant process for absorption of a photon of energy few TeV is electron-positron pair production [28, 29]. Hence the flux of UHE photons from distant sources is expected to be negligible because of this attenuation.
On 9 October 2022, an extremely powerful gamma-ray burst, GRB 221009A, located at RA , Dec [30, 31] with redshift [31], was detected by several observatories. The Large High Altitude Air Shower Observatory (LHAASO) registered more than 5,000 very-high-energy photons above 0.2 TeV, with individual events extending up to 18 TeV [33, 34]. Subsequently, a 251 TeV photon-like air-shower event in spatial and temporal coincidence with the burst was reported by the Carpet-2 array [35], later revised to 300 TeV following reanalysis with the upgraded Carpet-3 detector [36], making it the highest-energy photon-like event ever associated with a GRB. Given the distance of the source from Earth, such a UHE photon should have been attenuated before reaching the Earth. Hence the observation of this UHE photon is a mystery.
UHE photons from extragalactic sources can undergo photon–ALP mixing in the presence of the magnetic field near the source galaxy, converting a fraction of the flux into a photon–ALP beam [37, 38]. While propagating through the intergalactic medium, the photon component continues to suffer EBL attenuation as described above, whereas the axion component, being weakly interacting with the background radiation fields, propagates unimpeded. The surviving photons and the unattenuated axion flux thus reach our Milky Way together, where passage through the Galactic magnetic field allows a fraction of the axions to reconvert into photons. This photon→ALP→photon conversion can therefore replenish the UHE photon flux at Earth beyond what EBL attenuation alone would permit, offering a possible resolution to the puzzle described above.
There have been attempts to explain these anomalous observations through photon–ALP conversion. Refs. [39, 40, 41, 42, 43, 44] argue that a pure photon–ALP oscillation scenario, assuming a fixed domain length for the source-galaxy magnetic field, cannot simultaneously account for the TeV Carpet-3 event together with the LHAASO spectrum extending up to TeV, and consequently invoke Lorentz invariance violation (LIV) in conjunction with photon–ALP mixing to accommodate the 300 TeV photon. In this work however, we show that if one incorporates a variable magnetic field domain length in the source-galaxy, a purely ALP-based scenario, without invoking LIV, can accommodate both the LHAASO and Carpet-3 measurements while still remaining outside the CAST excluded region.
The rest of this paper is organized as follows. In Sec. II, we review photon–ALP oscillations in uniform and fixed-length multi-domain magnetic fields, and extend the formalism to domains of variable length. The conversion probability then follows from a Gaussian average of the fixed-domain result over the domain length, and the standard deviation of this distribution enters as an additional term that raises the conversion probability. In Sec. III, we summarize the GRB 221009A observations and the standard EBL attenuation of the photon flux. In Sec. IV, we include photon–ALP conversion in the source, the intergalactic medium and the Milky Way, and describe the fit to the LHAASO and Carpet-3 spectra. In Sec. V, we present the best-fit and for fixed and variable domain lengths, compare them with the CAST bound, and conclude.
Axion–Photon Oscillation
The Lagrangian of the axion field relevant for our discussion is given by [16, 18]
where is the axion field, its mass, the electromagnetic field strength tensor, and its dual.
Starting from the Lagrangian in Eq. (1), we can derive the equations of motion for the axion and the photon fields as
and
Working in the radiation gauge and following Ref. [18] we obtain the Schrödinger-like linearized equation of motion as
where we have defined
and
is the axion–photon mixing matrix in vacuum, where for simplicity we have assumed that the external magnetic field is transverse to the direction of propagation with being its magnitude and is the axion energy which is equal to the photon energy . The presence of the off-diagonal terms in the mixing matrix results in the axion and photon modes getting mixed as they propagate through a magnetized medium resulting in the conversion of axions into photons and vice-versa.
In realistic astrophysical environments, photon propagation is modified by the presence of background magnetic fields and plasma. Consequently, the axion–photon mixing matrix is of the form
with and .
The diagonal entries and receive contributions from vacuum polarization and the plasma [45]
The plasma contribution is
where
is the plasma frequency and is the ambient electron number density.
For the astrophysical systems considered in this work, the QED correction is much smaller than the plasma contribution and is therefore neglected.
With the magnetic field and electron density profiles specified, the mixing matrix in Eq. (7) can be constructed explicitly, and Eq. (4) solved to obtain the axion-to-photon conversion probability.
Single Domain: Constant Magnitude and Direction
We first recall the standard single-domain result [18], in which the magnetic field has constant magnitude and fixed direction over the full propagation path, so that the mixing term is independent of . In this simplest case, the axion mixes only with the parallel photon polarization , and the mixing matrix reduces to the system
Diagonalizing the mixing matrix in Eq. (11) with mixing angle , where
one obtains the photon-to-axion conversion probability for , ,
where the oscillation wavenumber is
Using , Eq. (13) can equivalently be written as
Conversion is most efficient in the strong-mixing regime , where and , giving the probability
Multi-Domain: Fixed Domain Length
Astrophysical magnetic fields are rarely uniform over the full path traversed by a photon or axion. In many environments of interest such as galaxy clusters, the intracluster and intergalactic medium, stellar and neutron star magnetospheres, the field is better described as coherent only over finite length scales, known as domains or cells, within which the magnitude and direction of the magnetic field can be treated as approximately constant. Beyond a single domain, the field is expected to change, either in direction, in magnitude, or both, reflecting the turbulent or structured nature of the underlying plasma. This domain-like structure has a direct impact on axion–photon conversion: since the mixing angle and oscillation length both depend on the local field configuration, a photon or axion traversing multiple domains experiences a series of distinct mixing conditions rather than a single coherent oscillation.
We recall the simple multi-domain treatment [46, 47] in which photons traverse a series of domains of equal length , in each of which the magnetic field is constant and homogeneous. The transverse component of the field is assumed to have the same magnitude in every domain, but a random orientation from one domain to the next, with domain separated by intervals over which conversion is negligible.
Consider an initial state
where and denote the photon polarization states perpendicular and parallel to the magnetic field in the first domain, so that the initial photon and axion fluxes are and . In the -th domain the field is rotated by an angle relative to the first domain, and the local polarization basis is related to the fixed basis by
with , .
Within each domain, the axion mixes only with the locally parallel polarization according to the mixing matrix of Eq. (11), with transition amplitudes and that are identical in every domain, since the field magnitude is fixed. The single-domain conversion probability is therefore , given by Eq. (13), with photon survival probability . Propagating the fluxes domain-by-domain and averaging over the random orientations – so that and , gives the recursion relation
Diagonalizing this transfer matrix and taking the limit of a large number of domains, so that , the fluxes after propagating a distance are
from where we can read off the axion to photon conversion probability as [46, 47]
Multi-Domain: Variable Domain Length
In Sec. II B we have reviewed the axion–photon oscillation probability for multiple magnetic field domains of equal length. But in astrophysical scenarios, it is rarely the case that the domains will have equal length. In this section we present a simplified treatment of axion–photon conversion in a region with multiple magnetic field domains where not only the magnetic field orientation changes but also the domain lengths vary from one domain to another.
We assume that the total path length traversed by the axion/photon spans multiple magnetic domains. Within each domain, the transverse magnetic field has the same magnitude across all domains, while its orientation is randomized independently in each domain, as before. Unlike the previous section, however, we now let the domain lengths vary: the length of the -th domain is drawn independently from a truncated Gaussian distribution with mean and standard deviation .
To calculate the conversion probability of an axion into a photon after traveling a fixed distance , we start from the recursion relation in Eq. (23). This expresses the photon and axion intensities after traversing domains, and , in terms of their values after domains
where is the length of the th domain and is the axion–photon transition probability accumulated while crossing it. Unlike the equal-length case, now varies from domain to domain, since it depends explicitly on the (random) length of each domain.
Applying the same recursion once more to express in terms of , we obtain
with the th-domain matrix acting to the left of the th-domain matrix, reflecting the fact that domains are traversed sequentially in order.
Iterating this procedure back to the initial intensities and , and keeping terms to first order in the domain probabilities, we obtain
where is the single-domain transition probability associated with the th domain, of length . Since for all , cross terms of and higher, generated by the product of the transfer matrices, are systematically neglected, leaving the leading-order result above in which only the sum appears.
Let us now suppose that, over the course of the journey from source to observer, a domain of length recurs times, a domain of length recurs times, and so on, so that the total path length can be written as , following the standard domain treatment of a randomly oriented magnetic field [48, 49]. Denoting the total number of domains traversed by , the mean domain length along this path is ; strictly, this is the sample mean of the realized domain lengths, which coincides with the population mean of the underlying truncated Gaussian distribution only in the limit , by the strong law of large numbers [50], a limit we take by the end of this paragraph. With these definitions, the sum can be recast as
where is the number of distinct domain lengths appearing along the path, and indexes them. Each factor is thus a weight, given by the fraction of domains with length ; in the limit of a large number of domains (), this weight approaches the probability density of drawing a domain of length from the underlying truncated Gaussian domain-length distribution with mean and standard deviation . Since domain lengths are taken to be positive definite, is understood throughout as the truncated Gaussian probability density,
with
where is the error function. The discrete sum over realized domains is thus, in this limit, a weighted average of the conversion probability over the domain-length distribution, and may be replaced by an integral,
where denotes the single domain conversion probability averaged over the ensemble of domain lengths. This implies that for variable domain length we can simply change . A similar kind of averaging for neutrinos has also been performed in Ref. [51]. Ref. [52] studies a related setup, in which the photon–ALP conversion probability is averaged over an ensemble of random magnetic field realizations. Unlike us, however, in their setup each realization is drawn from a power law distribution.
The final axion to photon conversion probability becomes
where . Using the expression of from Eq. (13) and evaluating the integral in Eq. (29) one can get
and
with and denoting the real and imaginary parts, respectively, of
For
Here we neglected the higher order terms and assumed and . We can see that for is greater than the usual probability by a term compared to Eq. (16).
Now when , which means uniform domain length, we recover the standard oscillation probability given by Eq. (13)
As , , where denotes the Dirac delta function.
Figure 1 shows the axion–photon conversion probability as a function of mass for three distinct magnetic-domain configurations. The simple case (blue solid line), in which the domains are fixed in both size and orientation, exhibits a steady probability at lower masses before transitioning into rapid, deep oscillations for masses above . Randomizing the domain orientation alone (orange solid line) yields a slightly lower initial probability and damps the oscillatory behavior at higher masses. Randomizing both the domain orientation and size (green solid line) suppresses the oscillatory structure entirely, producing smooth, non-oscillatory behavior across the full mass range shown.

Figure 1. Transition probability as a function of mass (in eV), simulated for an energy of 1 TeV, a coupling constant of , and a magnetic field of .
A complementary comparison of the conversion probability as a function of energy is presented in Figure 2 for several values of . The solid green curve, corresponding to , recovers the fixed-domain case. For nonzero , the conversion probability is clearly relative to this baseline, demonstrating that domain-length fluctuations systematically boost the axion–photon conversion probability.

Figure 2. Energy dependence of the photon–axion conversion probability, , for different values of the standard deviation of the magnetic-domain length distribution, normalized to the mean coherence length . The mass , and the coupling , these values are taken from [39].
Ultra High Energy Photons from GRB 221009A
GRB 221009A is an extremely bright gamma-ray burst [31] associated with a Type Ic supernova [53] located in the constellation Sagitta at redshift [54, 55]. The LHAASO observatory reported a huge flux of very high energy and ultra-high energy (UHE) photons from this GRB, with the highest recorded photon energy being 18 TeV [34, 56]. Additionally, the Carpet-3 observatory reported the observation of a UHE photon with energy TeV [35, 36]. In Figure 3 we show the reconstructed photon flux as reported by these experiments.

Figure 3. Observed spectral energy distribution, of GRB 221009A as observed by LHAASO and Carpet-3.
However, the observation of these UHE photons from GRB 221009A presents a direct challenge to our understanding [43]. UHE photons emitted from distant astrophysical sources such as GRB 221009A travel enormous distances through the intergalactic medium (IGM) and reach the Earth. The IGM is permeated by a background of low-energy photons constituting the Cosmic Microwave Background (CMB) [57], the Extragalactic Background Light (EBL) [58, 59] and the Cosmic Radio Background [60, 61]. While traveling through the IGM, UHE photons can interact with these low-energy background photons to produce electron-positron pairs via pair production () [28, 29], which can lead to their extinction and a significant loss of flux over distances spanning several megaparsecs [62].
To understand why this is happening, let us consider that UHE photons are produced at some distant extragalactic source at redshift . The observed intensity of
photons at energy [59, 62] at Earth is given by
where is the initial intensity, is the initial energy which is redshifted down to . The factor is the photon survival probability with being the optical depth (opacity) of the IGM to photons of observed energy emitted by a source at redshift . It accounts for the cumulative probability of pair production along the line of sight, and is given by
where is the energy of the background photon at redshift , is the cosine of the angle between the momenta of the two interacting photons and is the differential number density of the background photons, which can be obtained from [63]. The line element in a flat CDM cosmology is
where is the Hubble constant, is the matter density parameter and is the dark-energy density parameter, with values taken from Ref. [64]. The interaction is governed by the Breit-Wheeler cross-section [29, 65],
with the Thomson cross-section and being the center of mass energy.
Now at a fixed redshift where the photon has energy , pair production is kinematically allowed only for . This fixes the threshold energy of the background photons that can pair produce electrons to
From Eq. (40) one can see that a photon of energy is absorbed most efficiently by background photons with . Thus depending on the incoming UHE photon energy, different backgrounds dominate in different energy windows. The EBL [58, 59] with energies of the order of few to tens of eV is responsible for the attenuation of -100 TeV photons, whereas for TeV the relevant target photons are the CMB photons with energies of the order of eV. At even higher energies ( eV), the cosmic radio background [60] becomes the dominant target.
Figure 4 shows the survival probability as a function of the observed photon energy for a source at . One can clearly see that the Universe is transparent for TeV, while falls steeply above TeV. At 18 TeV (red dashed line), , corresponding to a suppression of almost nine orders of magnitude. The curve drops below the plotted range by TeV and never reaches the 300 TeV line (orange dashed), so a photon at that energy cannot be explained by standard propagation.

Figure 4. Photon survival probability during propagation through the intergalactic medium accounting for photon attenuation. At TeV, the survival probability is of order , while at TeV it becomes . Thus, the survival probability becomes extremely small at high energies and is insufficient to account for a significant flux of ultra-high-energy photons reaching Earth. The vertical dashed lines indicate TeV and TeV.
Propagation of UHE Photons in the Presence of Axions
To explain the observation of UHE photons from GRB 221009A one needs physics beyond the Standard Model, and axions present themselves as a natural candidate [39, 41, 43]. Host galaxies of GRBs, just like our Milky Way, are expected to host large scale magnetic fields, in which photons can convert into axions. Being weakly interacting, the axions propagate through the IGM unimpeded, since they cannot pair-produce on the EBL/CMB, and can reconvert into photons in the Galactic magnetic field.
Including photon–ALP conversion in the source (S), intergalactic (I) and Milky Way (M) regions, the total photon survival probability is modified to
where is the probability that a state entering region exits as state . Each term is a product along one of the possible paths S I M. For example, describes a photon converting to an ALP in the source, crossing the IGM as an ALP, and reconverting to a photon in the Milky Way. Since these environments host strong large-scale magnetic fields, the survival probability, and hence the observed UHE photon flux, can be modified significantly. In this section we show that this effect can help explain the UHE photons from GRB 221009A observed by LHAASO and Carpet-3.
To do so, we model the intrinsic, unattenuated photon spectrum of GRB 221009A as a simple unbroken power law [39],
where is the normalization, TeV, and is the spectral index. The observed spectrum, modified by photon-ALP conversion and reconversions as well as absorption on background photons, is
where is the total survival probability of Eq. (41). The corresponding observed spectral energy distribution is
This is the flux of UHE photons expected to reach Earth, including the effects of photon–ALP conversion.
To model photon–ALP oscillations in the Milky Way, we use the gammaALPs [66] Python package, which builds the Galactic magnetic field from the Jansson & Farrar model [67]. The photon survival probability and the conversion probability for different values of the ALP mass and coupling are shown in Figure 5. Photon–ALP mixing becomes efficient at energies above – TeV, reducing and enhancing correspondingly. The size of the effect depends strongly on and .

Figure 5. Energy dependence of the photon survival probability, (solid lines), and the photon–ALP conversion probability, (dashed lines), for photon propagation through the Galactic magnetic field. Colors denote representative ALP parameter sets as indicated in the legend.
We also use gammaALPs to compute the photon survival probability and the conversion probability in the IGM. The total propagation distance of 745 Mpc is modeled as a sequence of magnetic domains of fixed length 10 Mpc, each with field strength 1 nG [39, 68] and a random orientation. gammaALPs includes the attenuation of UHE photons by EBL scattering through the factor shown in Figure 4. We have checked that photon–ALP conversion in the IGM has a negligible effect on the final photon flux, but we retain it in our treatment.
Finally, we model the magnetic field of the host galaxy (S) as a collection of domains within a radius of 30 kpc, each with a fixed field strength and a random orientation. The domain lengths are allowed to vary, with mean kpc and relative dispersion . The photon–ALP conversion probability in the source is obtained from the averaged oscillation probability of Eq. (31). For all domains have the same length, which reproduces the fixed-domain-length scenario studied in Refs. [39, 44]. The variable-domain-length formalism is therefore a generalization of that setup.
Figure 6 shows the total photon survival probability from GRB 221009A as a function of the photon energy. We see that photon–ALP conversion lifts the survival probability at high energies, and that a broader domain-length distribution in the source enhances this rise. As mentioned before, earlier studies have also considered axions as an explanation of the UHE photons from this burst, but with fixed domain lengths in the source they could not accommodate the Carpet-3 data [39], and had to invoke LIV as an additional ingredient. Allowing the domain lengths to vary raises the survival probability at the highest energies, so that the data, including Carpet-3, can be explained by photon–ALP conversion alone, without invoking LIV or any other new effect.

Figure 6. Energy dependence of the photon survival probability, , for a Gaussian distribution of magnetic-domain lengths characterized by a mean coherence length and standard deviation . The legend indicates the normalized dispersion parameter . Increasing the spread in domain lengths modifies the oscillation pattern and enhances the high-energy survival probability relative to the uniform-domain case .
To quantify this, we fit the spectrum of Eq. (44) to the LHAASO [33, 34] and Carpet-3 [35, 36] data of GRB 221009A, and extract the best-fit ALP parameters together with the and viable regions in the plane. We perform the fit for two cases: , which reproduces the fixed-domain-length benchmark of Ref. [39], and , the maximal domain-length dispersion we consider. The intermediate values 0.25 and 0.5 in Figure 6 are shown for illustration only and are not fitted. In both cases is computed along SIM as described above, so the comparison isolates the effect of the domain-length distribution on the viable region.
We use the Gaussian
where is the measured in the -th energy bin and its uncertainty. For asymmetric uncertainties, we conservatively take to be the larger of the two. The source spectral parameters of Eq. (42) are nuisance parameters, which we profile out by minimizing the over them at each . The best-fit ALP parameters minimize the resulting profiled , with minimum , and the goodness of fit is quantified by , where is the number of data points minus the number of fitted parameters. The and regions are defined by and respectively [69, 70].
Results and Discussion
We scan the ALP parameter space and , evaluating the spectrum and the profiled at each point. Figure 7 and Figure 8 show the best-fit spectra and the confidence regions in the plane for (variable domain length) and (fixed domain length), respectively. The fixed-domain fit is poor, mainly because it cannot reach the Carpet-3 point, whereas the variable-domain fit describes both the LHAASO and Carpet-3 data.

Figure 7. Top panel: Best-fit spectrum for the variable-domain-length formalism, with spectral index . Bottom panel: Confidence-level contours in the plane for the variable-domain-length formalism with a single power-law spectrum. The CAST exclusion region is also shown.

Figure 8. Top panel: Best-fit spectrum for the fixed-domain-length formalism, with spectral index . The black line is the ALP-only best fit of Ref. [39]. Bottom panel: Confidence-level contours in the plane for the fixed-domain-length formalism with a single power-law spectrum. The CAST exclusion region is also shown.
For variable domain lengths, the minimum is obtained at
with a fitted spectral index , i.e. . The best-fit spectrum is shown in the top panel of Figure 7, and the , and regions in the bottom panel.
For fixed domain lengths, the best fit is at
with , i.e. (Figure 8). This spectrum follows the LHAASO points but falls far below the Carpet-3 point at 300 TeV.
Figure 8 also shows the best fit of Ref. [39], and , obtained for photon–ALP conversion only. Our analysis differs from theirs in the modeling of all three propagation regions. For the host galaxy, we use a 30 kpc region with divided into 5 domains of random orientation and varying length, whereas they use 10 kpc and without specifying the domain size. For the IGM, the field is weak enough that its effect is negligible, as also noted in their work. For the Milky Way, we construct the field with the Jansson & Farrar model in gammaALPs along the line of sight to GRB 221009A, whereas they approximate it as over kpc and take , i.e. they neglect conversion in the Milky Way. As shown in Figure 5, however, this conversion is not negligible at high energies, so it must be included when computing the flux at Earth.
We compare the viable regions with the CAST bound [71], , which is independent of astrophysical modelling and holds for ALP masses up to , far above the masses relevant
here. For fixed domain lengths, the best-fit spectrum falls below the Carpet-3 point (Figure 8, top panel), and the best-fit coupling, , is an order of magnitude above the CAST bound (Figure 8, bottom panel). The confidence regions around it are too small to be visible in the figure, but they lie entirely within the CAST-excluded region. The fixed-domain model therefore cannot accommodate the Carpet-3 data without conflicting with laboratory limits.
The situation is different for variable domain lengths. The best-fit spectrum reaches the Carpet-3 point while following the LHAASO data (Figure 7, top panel). The best-fit coupling, , lies a factor of about two below the CAST bound, and the , and regions all extend across the CAST line (Figure 7, bottom panel). Part of each region therefore lies at couplings allowed by CAST, so the variable-domain model is not excluded. The portions above the CAST line are excluded, and the viable parameter space is the intersection of the confidence regions with the CAST-allowed region. Compatibility with laboratory bounds thus cannot be judged from the best-fit point alone, but requires the full confidence region.
Our results also bear on the earlier literature. The ALP-only best fit of Ref. [39] with fixed magnetic domains, , lies a factor of about two above the CAST bound, and fails to reproduce the Carpet-3 point (Figure 8, top panel) requiring the introduction of LIV as an additional ingredient. Our fixed-domain analysis, which includes the Milky Way conversion neglected there, reaches the same conclusion: a single domain length cannot simultaneously fit the data and respect the CAST bound. Allowing a variability of the domain lengths of the source removes this tension, so that the Carpet-3 data can be explained by photon–ALP conversion alone, without LIV or any other physics beyond the axion.
Thus our results highlight the fact that the treatment of the magnetic-domain structure strongly affects the conversion probability, the spectral fit and the inferred ALP parameter space.
Acknowledgments
The authors sincerely acknowledge Satyaki Bhattacharya for the valuable discussions and kind help with the statistical analysis methodology adopted in this paper. RSG sincerely thanks the Indian Association for the Cultivation of Science (IACS), Kolkata, for financial and institutional support during the course of this research.
The work of TK was supported by Indian Institute of Technology Kanpur.
References
References
- [1]D. J. E. Marsh, Axion Cosmology, Phys. Rept. 643, 1 (2016), arXiv:1510.07633 [astro-ph.CO].DOI
- [2]I. G. Irastorza and J. Redondo, New experimental approaches in the search for axion-like particles, Prog. Part. Nucl. Phys. 102, 89 (2018), arXiv:1801.08127 [hep-ph].DOI
- [3]K. Choi, S. H. Im, and C. S. Shin, Recent Progress in the Physics of Axions and Axion-Like Particles, Ann. Rev. Nucl. Part. Sci. 71, 225 (2021), arXiv:2012.05029 [hep-ph].DOI
- [4]R. D. Peccei and H. R. Quinn, CP Conservation in the Presence of Instantons, Phys. Rev. Lett. 38, 1440 (1977).
- [5]R. D. Peccei and H. R. Quinn, Constraints Imposed by CP Conservation in the Presence of Instantons, Phys. Rev. D 16, 1791 (1977).
- [6]S. Weinberg, A New Light Boson?, Phys. Rev. Lett. 40, 223 (1978).DOI
- [7]F. Wilczek, Problem of Strong P and T Invariance in the Presence of Instantons, Phys. Rev. Lett. 40, 279 (1978).DOI
- [8]R. D. Peccei, The Strong CP problem and axions, Lect. Notes Phys. 741, 3 (2008), arXiv:hep-ph/0607268.DOI
- [9]A. Arvanitaki, S. Dimopoulos, S. Dubovsky, N. Kaloper, and J. March-Russell, String Axiverse, Phys. Rev. D 81, 123530 (2010), arXiv:0905.4720 [hep-th].DOI
- [10]P. Svrcek and E. Witten, Axions In String Theory, JHEP 06, 051, arXiv:hep-th/0605206.
- [11]J. Preskill, M. B. Wise, and F. Wilczek, Cosmology of the Invisible Axion, Phys. Lett. B 120, 127 (1983).DOI
- [12]L. F. Abbott and P. Sikivie, A Cosmological Bound on the Invisible Axion, Phys. Lett. B 120, 133 (1983).DOI
- [13]M. Dine and W. Fischler, The Not So Harmless Axion, Phys. Lett. B 120, 137 (1983).
- [14]M. Y. Khlopov, A. S. Sakharov, and D. D. Sokoloff, The nonlinear modulation of the density distribution in stan dard axionic CDM and its cosmological impact, Nucl. Phys. B Proc. Suppl. 72, 105 (1999).
- [15]J. E. Kim and G. Carosi, Axions and the Strong CP Problem, Rev. Mod. Phys. 82, 557 (2010), [Erratum: Rev.Mod.Phys. 91, 049902 (2019)], arXiv:0807.3125 [hep-ph].DOI
- [16]L. Di Luzio, M. Giannotti, E. Nardi, and L. Visinelli, The landscape of QCD axion models, Phys. Rept. 870, 1 (2020), arXiv:2003.01100 [hep-ph].DOI
- [17]P. Sikivie, Experimental Tests of the Invisible Axion, Phys. Rev. Lett. 51, 1415 (1983), [Erratum: Phys.Rev.Lett. 52, 695 (1984)].
- [18]G. Raffelt and L. Stodolsky, Mixing of the Photon with Low Mass Particles, Phys. Rev. D 37, 1237 (1988).pubmed.ncbi.nlm.nih.gov/9958803
- [19]D. Cadamuro and J. Redondo, Cosmological bounds on pseudo Nambu-Goldstone bosons, JCAP 02, 032, arXiv:1110.2895 [hep-ph].DOI
- [20]A. Caputo and G. Raffelt, Astrophysical Axion Bounds: The 2024 Edition, PoS COSMICWISPers, 041 (2024), arXiv:2401.13728 [hep-ph].DOI
- [21]G. G. Madejski and M. Sikora, Gamma-Ray Observations of Active Galactic Nuclei, Annu. Rev. Astron. Astrophys. 54, 725 (2016).DOI
- [22]P. Padovani et al., Active galactic nuclei: what’s in a name?, Astron. Astrophys. Rev. 25, 2 (2017), arXiv:1707.07134 [astro-ph.GA].DOI
- [23]R. Blandford, D. Meier, and A. Readhead, Relativistic Jets from Active Galactic Nuclei, Ann. Rev. Astron. Astrophys. 57, 467 (2019), arXiv:1812.06025 [astro-ph.HE].DOI
- [24]R. E. Pudritz, M. J. Hardcastle, and D. C. Gabuzda, Magnetic Fields in Astrophysical Jets: From Launch to Termination, Space Sci. Rev. 169, 27 (2012), arXiv:1205.2073 [astro-ph.HE].DOI
- [25]P. Kumar and B. Zhang, The physics of gamma-ray bursts \& relativistic jets, Phys. Rept. 561, 1 (2014), arXiv:1410.0679 [astro-ph.HE].DOI
- [26]P. Mészáros, Gamma-Ray Bursts, Rept. Prog. Phys. 69, 2259 (2006), arXiv:astro-ph/0605208.DOI
- [27]M. Boettcher, A. Reimer, K. Sweeney, and A. Prakash, Leptonic and Hadronic Modeling of Fermi-Detected Blazars, Astrophys. J. 768, 54 (2013), arXiv:1304.0605 [astro-ph.HE].
- [28]A. I. Nikishov, Absorption of High-Energy Photons in the Universe, Sov. Phys. JETP 14, 393 (1962).
- [29]R. J. Gould and G. P. Schrèder, Pair Production in Photon-Photon Collisions, Physical Review 155, 1404 (1967).DOI
- [30]S. Dicharia, J. D. Gropp, J. A. Kennea, N. P. M. Kuin, A. Y. Lien, F. E. Marshall, A. Tohuvavohu, M. A. Williams, and Neil Gehrels Swift Observatory Team, Swift J1913.1+1946 a new bright hard X-ray and optical transient, GRB Coordinates Network 32623, 1 (2022).
- [31]E. Burns et al., GRB 221009A: The BOAT, Astrophys. J. Lett. 946, L31 (2023), arXiv:2302.14037 [astro-ph.HE].DOI
- [33]Z. Cao et al. (LHAASO), A tera-electronvolt gamma-ray burst high-energy plateau powered by a structured jet, Science 380, 1390 (2023).
- [34]Z. Cao et al. (LHAASO), Very high-energy gamma-ray emission beyond 10 TeV from GRB 221009A, Sci. Adv. 9, adj2778 (2023), arXiv:2310.08845 [astro-ph.HE].DOI
- [35]D. D. Dzhappuev et al., Signatures of a 251-TeV gamma ray from GRB 221009A by Carpet-3, JETP Lett. 116, 829 (2022).
- [36]D. D. Dzhappuev et al. (Carpet-3 Group), Carpet-3 detection of a photolike air shower with estimated primary energy above 100 TeV in a spatial and temporal coincidence with GRB 221009A, Phys. Rev. D 111, 102005 (2025), arXiv:2502.02425 [astro-ph.HE].
- [37]M. Fairbairn, T. Rashba, and S. V. Troitsky, Photon-axion mixing and ultra-high-energy cosmic rays from BL Lac type objects - Shining light through the Universe, Phys. Rev. D 84, 125019 (2011), arXiv:0901.4085 [astro-ph.HE].DOI
- [38]G. Galanti and M. Roncadelli, Axion-like Particles Implications for High-Energy Astrophysics, Universe 8, 253 (2022), arXiv:2205.00940 [hep-ph].DOI
- [39]L. Qin et al., Revisiting Very High Energy Gamma-Ray Absorption in Cosmic Propagation under the Combined Effects of Axion-like Particles and Lorentz Invariance Violation, Astrophys. J. 1000, 260 (2026), arXiv:2510.23113 [astro-ph.HE].arxiv.org/abs/2510.23113
- [40]D. D. Ofengeim and T. Piran, 300 TeV photon from GRB 221009A: A hint at nonlinear Lorentz invariance violation?, Phys. Rev. D 112, 083055 (2025), arXiv:2508.07153 [astro-ph.HE].DOI
- [41]P. S. Satunin and S. V. Troitsky, Testing New-Physics Scenarios with the Combined LHAASO and Carpet-3 Fluence Spectrum of GRB 221009A: Axion-Like Particles and Lorentz-Invariance Violation, JETP Lett. 123, 73 (2026), arXiv:2510.07234 [astro-ph.HE].arxiv.org/abs/2510.07234
- [42]J. D. Finke and S. Razzaque, Possible Evidence for Lorentz Invariance Violation in Gamma-Ray Burst 221009A, Astrophys. J. Lett. 942, L21 (2023), arXiv:2210.11261 [astro-ph.HE].DOI
- [43]G. Galanti, L. Nava, M. Roncadelli, F. Tavecchio, and G. Bonnoli, Observability of the Very-High-Energy Emission from GRB 221009A, Phys. Rev. Lett. 131, 251001 (2023), arXiv:2210.05659 [astro-ph.HE].DOI
- [44]L. Wang and B.-Q. Ma, Axion-photon conversion of GRB221009A, Phys. Rev. D 108, 023002 (2023).DOI
- [45]P. Carenza, C. Evoli, M. Giannotti, A. Mirizzi, and D. Montanino, Turbulent axion-photon conversions in the Milky Way, Phys. Rev. D 104, 023003 (2021), arXiv:2104.13935 [hep-ph].DOI
- [46]Y. Grossman, S. Roy, and J. Zupan, Effects of initial axion production and photon axion oscillation on type Ia supernova dimming, Phys. Lett. B 543, 23 (2002), arXiv:hep-ph/0204216.DOI
- [47]C. Csaki, N. Kaloper, and J. Terning, Dimming supernovae without cosmic acceleration, Phys. Rev. Lett. 88, 161302 (2002), arXiv:hep-ph/0111311.DOI
- [48]P. P. Kronberg, Extragalactic magnetic fields, Rep. Prog. Phys. 57, 325 (1994).DOI
- [49]D. Grasso and H. R. Rubinstein, Magnetic fields in the early universe, Phys. Rept. 348, 163 (2001).DOI
- [50]W. Feller, An Introduction to Probability Theory and Its Applications, Vol. 1, 3rd ed. (Wiley, New York, 1968).
- [51]T. Ohlsson, Equivalence between neutrino oscillations and neutrino decoherence, Phys. Lett. B 502, 159 (2001), arXiv:hep-ph/0012272.DOI
- [52]O. Khasai, L. Zadorozhna, D. Malyshev, and O. Ruchayskiy, Domain models of turbulent magnetic fields and the spectral shape of ensemble averaged axion-photon conversion (2026), arXiv:2610.29541 P. K. Blanchard et al., JWST detection of a supernova associated with GRB 221009A without an r-process signature, Nature Astron. 8, 774 (2024), arXiv:2308.14197 [astro-ph.HE].
- [54]A. de Ugarte Postigo, L. Izzo, G. Pugliese, D. Xu, B. Schneider, J. P. U. Fynbo, N. R. Tanvir, D. B. Malesani, A. Saccardi, D. A. Kann, K. Wiersema, B. P. Gompertz, C. C. Thoene, A. J. Levan, and Stargate Collaboration, GRB 221009A: Redshift from X-shooter/VLT, GRB Coordinates Network 32648, 1 (2022).
- [55]A. J. Castro-Tirado, R. Sanchez-Ramirez, Y.-D. Hu, M. D. Caballero-Garcia, M. A. Castro Tirado, E. Fernandez-Garcia, I. Perez-Garcia, G. Lombardi, S. B. Pandey, J. Yang, and B.-B. Zhang, GRB 221009A: 10.4m GTC spectroscopic redshift confirmation, GRB Coordinates Network 32686, 1 (2022).
- [56]Z. Cao et al. (LHAASO), A tera–electron volt afterglow from a narrow jet in an extremely bright gamma-ray burst, Science 380, adg9328 (2023), arXiv:2306.06372 [astro-ph.HE].
- [57]A. A. Penzias and R. W. Wilson, A Measurement of Excess Antenna Temperature at 4080 Mc/s., Astrophys. J. 142, 419 (1965).
- [58]M. G. Hauser and E. Dwek, The cosmic infrared background: measurements and implications, Ann. Rev. Astron. Astrophys. 39, 249 (2001), arXiv:astro-ph/0105539.DOI
- [59]E. Dwek and F. Krennrich, The extragalactic background light and the gamma-ray opacity of the universe, Astroparticle Physics 43, 112–133 (2013).DOI
- [60]R. J. Protheroe and P. L. Biermann, A New estimate of the extragalactic radio background and implications for ultra-high-energy gamma-ray propagation, Astropart. Phys. 6, 45 (1996), [Erratum: Astropart.Phys. 7, 181 (1997)], arXiv:astro-ph/9605119.DOI
- [61]D. J. Fixsen, A. Kogut, S. Levin, M. Limon, P. Lubin, P. Mirel, M. Seiffert, J. Singal, E. Wollack, T. Villela, and C. A. Wuensche, ARCADE 2 measurement of the absolute sky brightness at 3-90 GHz, The Astrophysical Journal 734, 5 (2011).DOI
- [62]F. W. Stecker and O. C. De Jager, Absorption of high-energy gamma-rays by low-energy intergalactic photons, Space Sci. Rev. 75, 401 (1996), arXiv:astro-ph/9501065.DOI
- [63]A. Dominguez et al., Extragalactic Background Light Inferred from AEGIS Galaxy SED-type Fractions, Mon. Not. Roy. Astron. Soc. 410, 2556 (2011).
- [64]N. Aghanim et al. (Planck), Planck 2018 results. VI. Cosmological parameters, Astron. Astrophys. 641, A6 (2020), [Erratum: Astron.Astrophys. 652, C4 (2021)], arXiv:1807.06209 [astro-ph.CO].DOI
- [65]G. Breit and J. A. Wheeler, Collision of two light quanta, Phys. Rev. 46, 1087 (1934).DOI
- [66]M. Meyer, J. Davies, and J. Kuhlmann, gammaALPs: An open-source python package for computing photon-axion-like-particle oscillations in astrophysical environments, PoS ICRC2021, 557 (2021).DOI
- [67]R. Jansson and G. R. Farrar, A New Model of the Galactic Magnetic Field, Astrophys. J. 757, 14 (2012).DOI
- [68]C. Han, Z. Lei, J. Yang, and S. Zhao, Line-of-sight magnetic-field propagation effects on axion-like particle constraints from GRB 221009A (2026), arXiv:2606.15269 [hep-ph].arxiv.org/abs/2606.15269
- [69]G. Cowan, K. Cranmer, E. Gross, and O. Vitells, Asymptotic formulae for likelihood-based tests of new physics, Eur. Phys. J. C 71, 1554 (2011), [Erratum: Eur.Phys.J.C 73, 2501 (2013)], arXiv:1007.1727 [physics.data-an].DOI
- [70]F. Takahashi et al. (Particle Data Group), Review of Particle Physics, Int. J. Mod. Phys. A 41, 2630011 (2026).
- [71]V. Anastasopoulos et al. (CAST), New CAST Limit on the Axion-Photon Interaction, Nature Phys. 13, 584 (2017), arXiv:1705.02290 [hep-ex].