WorksPhysics and Astronomy
The imprint of neutron escape on non-thermal protons in AGN coronae
Abstract
The coronae of active galactic nuclei (AGN) are promising sites of proton acceleration to hundreds of TeV, potentially accounting for the neutrino excess from NGC 1068 reported by IceCube. We show that photohadronic proton-to-neutron conversion provides an important proton escape channel, as neutrons can leave the corona well before decaying. In magnetized turbulent coronae, neutron escape can reduce the characteristic neutrino energy by up to a factor of four and the integrated neutrino luminosity by approximately 30% at a fixed non-thermal proton energy density. More strikingly, proton-to-neutron conversion offers an escape route roughly two orders of magnitude faster than proton diffusion, requiring a correspondingly larger proton injection rate to sustain the non-thermal population. The effect is considerably weaker when proton escape is already ballistic, as expected in magnetic-reconnection scenarios. Therefore, neutron escape is essential for quantitatively connecting neutrino emission from AGN coronae to the efficiency of proton injection into the acceleration process.
Introduction
A diffuse extragalactic neutrino flux, with energies ranging from a few TeV to several PeV, was discovered by the IceCube Collaboration in 2013, opening a new window on the non-thermal Universe [1]. Unlike photons and charged cosmic rays (CRs), neutrinos propagate over cosmological distances without absorption or deflection, carrying direct information about their production sites. Neutrinos escape freely even from those cosmic accelerators where gamma radiation is strongly attenuated, often termed hidden or gamma-ray opaque sources.
These hidden sources play a major role because of the mismatch between the large neutrino flux observed at a few TeV, in comparison with the Fermi-LAT measurements [2] of the extragalactic diffuse gamma-ray flux at hundreds of GeV. The natural explanation is that the neutrino flux primarily arises in hidden sources [3, 4, 5, 6, 7, 8, 9] (see also [10] and references therein). If neutrinos are produced through the interaction of non-thermal protons with a dense radiation field, that same field offers a natural target to attenuate gamma rays [3], realizing the hidden source scenario suggested by multimessenger observations.
An independent hint towards the hidden-source paradigm has come from the IceCube discovery of a neutrino excess directionally coincident with an active galactic nucleus (AGN), the Seyfert galaxy NGC 1068 [11]. No comparable gamma-ray flux has been observed, with the strongest upper bound by the MAGIC telescope roughly an order of magnitude lower [12], so neutrinos are expected to be produced in a compact, optically thick region where gamma rays can be absorbed [13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27]. A natural candidate is the corona surrounding the central black hole of the AGN, exhibiting a prominent X-ray emission up to hundreds of keV which naturally absorbs gamma rays between GeV and TeV energies [28].
Even after specifying the candidate region for non-thermal emission, the dynamics of non-thermal particles remains largely uncertain. Neutrinos are emitted by the decay of high-energy pions produced in non-thermal proton interactions (see Ref. [29] for a critical discussion of alternative scenarios), crucially depending on what mechanism accelerates non-thermal protons, and their dominant interactions. The spectral properties of the neutrino excess observed by IceCube, showing a neutrino spectrum steeply falling after a few TeV, may offer some guidance in constraining candidate scenarios. Currently, several possibilities consistently explain the broad properties of the signal, including diffusive acceleration in a coronal shock [13], stochastic acceleration in magnetized turbulence [14, 20, 26, 30], and a magnetized reconnection layer in the black hole magnetosphere [18, 19, 31].
Associated with the uncertain acceleration mechanism, the coronal size also affects the nature of hadronic interactions; a relatively wide corona spanning hundreds of gravitational radii is permeated by the strong optical-ultraviolet (OUV) field produced from the surrounding disk, so Bethe–Heitler (BH) pair production strongly cools the protons [30]. A more compact corona spanning up to a few tens of gravitational radii is likely less affected by the diluted OUV radiation produced in much outer regions [18, 21, 32], so the dominant proton cooling may come from interactions with the X-ray coronal field.
In most of the scenarios discussed above, pions are predominantly produced in interactions. The latter thus play two roles: first, they furnish a means of neutrino production through pion decay; second, they drain the non-thermal proton population of energy. The latter feature is quite consequential, since it competes with the escape and acceleration of protons, shaping their spectrum. Usually, interactions are treated as a continuous energy loss, e.g. in Refs. [14, 20, 21, 23, 24, 26, 30, 33]. However, inelastic photohadronic interactions turn protons into neutrons [34, 35], which are not magnetically confined in the corona, and in fact escape much faster than they can decay back to protons. Thus, proton-to-neutron conversion via , beyond cooling protons, offers an effective escape channel to the otherwise confined protons (see for instance early works by Refs. [36, 37, 38] where the impact of neutrons on the AGN dynamics is discussed).
Here we consider the impact of this effect on stochastic acceleration in magnetized coronae. To do this, we consider the coupled evolution of protons and neutrons, rather than considering the cooled protons alone. Our methods are borrowed from the trickle-down (TD) approach applied in the ultra-high-energy regime [39], where protons and neutrons are catastrophically removed from their population every time they interact through scattering, and re-injected at lower energies according to their interaction probabilities. We consider the impact of proton-to-neutron conversion both in models of stochastic proton acceleration in magnetized turbulence, and magnetized reconnection. The key effects are a distortion of the neutrino spectrum, due to the increased losses suffered by protons in inelastic interactions, and an enhanced proton escape channel. For magnetized turbulence, the latter effect is quite dramatic, since protons, which would otherwise be magnetically confined in the turbulent region for days, can convert to neutrons and escape the region at the speed of light roughly in a few minutes. Consequently, a much larger rate of injected protons is required to saturate the IceCube neutrino signal, although the overall number density of non-thermal protons is still much smaller than the available baryon density. For a magnetized reconnection layer, where protons anyway escape within roughly a light-crossing time, the impact of proton-to-neutron conversion is more limited.
The article is organised as follows. In § 2, we introduce the background photon fields, the coronal geometries, and the relevant cosmic-ray timescales. In § 3, we discuss the transport equations for both the magnetized turbulent-corona and magnetic-reconnection models for the continuous-energy-loss (CEL) approximation and the trickle-down approach, respectively. § 4 discusses how we implement the neutrino production rates after determining the equilibrium densities of cosmic rays. In § 5 and § 6, we present the results for the two models, respectively. In § 7, we discuss the implications of our findings and draw our conclusions. Appendices § 9 and § 10 provide details on the photopion and hadronuclear simulations and the calculation of the interaction quantities employed in this work, respectively. Appendix § 11 describes the numerical implementation and the boundary conditions used to solve the transport equations. Finally, in Appendix § 12, we assess the impact of secondary cooling including synchrotron losses of charged pions and muons on the neutrino yields.
Cosmic-ray dynamics in coronal environments
The dynamics of protons depends rather strongly on the environment—how strongly magnetized the corona is, and the radiation field that protons experience inside of it. We discuss these model choices in this section, starting from the radiation field and then particle acceleration, escape and energy losses mechanisms.
Radiation field
The radiation in the corona is crucial to assess how quickly cosmic rays are cooled. We model the corona as a spherical region with a radius as a fixed multiple of the gravitational radius of the black hole ()
We focus on NGC 1068 as a benchmark with a black hole mass [40]. For the background radiation, we consider different scenarios:
scenario I, where coronal X-rays coexist with the OUV photons from the accretion disk, leading to the characteristic big blue bump in the AGN spectra. We use the luminosity per unit energy parameterized in Ref. [41], including both the X-ray and the OUV component, normalizing it to a luminosity integrated between 2 keV and 10 keV, , of [40] as in Refs. [23, 26].
scenario II, where only corona X-rays are included, as in Ref. [18]. The X-ray spectral luminosity is modeled as
with and . The normalization is fixed by requiring . Indeed, if the OUV flux is produced significantly outside of the corona, at radii of the order of (see, e.g., Eq. (A9) in Ref. [18], obtained from the Shakura–Sunyaev theory of geometrically thin accretion disks [42, 43]), the OUV field inside the corona is geometrically suppressed by .
Thus, the two configurations bracket the uncertainties on the overall geometry of the system. The differential number density of target photons reads
The resulting target photon densities for scenarios I and II are shown in Figure 1 for the same coronal radius, . This comparison isolates the effect of the additional OUV component on the photon field encountered by the accelerated particles.
Figure 1. Differential number density of target photons for scenario I, including both the coronal X-ray and accretion-disk OUV components, and scenario II, including only the coronal X-ray component. Both distributions are shown for , , and .
Particle acceleration and escape
We consider two acceleration prescriptions: stochastic acceleration in a magnetized turbulent corona and particle acceleration in a magnetic-reconnection layer. These prescriptions are independent of the two radiation-field configurations defined above. The turbulent-corona model is studied with both photon-field scenarios I and II, whereas magnetic reconnection is considered with the X-ray-only photon field (scenario II). The two acceleration prescriptions also differ in the size and magnetization of the interaction region.
Magnetized turbulent corona. In the turbulent scenario, CR protons are stochastically energized by magnetic fluctuations close to the central supermassive black hole (SMBH). The fundamental parameters describing the coronal plasma [20] are the Thomson optical depth and the plasma magnetization . Assuming charge neutrality, the thermal proton and electron number densities satisfy
The plasma magnetization and the corresponding magnetic-field strength are
To characterize the turbulent component of the magnetic field, we introduce , where quantifies the relative strength of magnetic fluctuations. The turbulence coherence length is . In this framework, proton acceleration is primarily driven by non-resonant interactions with large-scale turbulent magnetic structures. The characteristic acceleration time is energy-independent and reads [20, 44]
At the same time, turbulence controls particle confinement, with a time given by [20, 45, 46]
This prescription ensures that the diffusive confinement always satisfies .
For our reference configuration, we adopt , , , , and , motivated by the parameters inferred from NGC 1068 neutrino emission [20, 26]. These assumptions yield , , and .
Magnetic reconnection. In this scenario, non-thermal protons are accelerated through the rapid conversion of magnetic energy into particle kinetic energy within a compact reconnection layer. Following Ref. [18], we determine the magnetic field by assuming that a fraction of the incoming Poynting flux is converted into the total coronal X-ray luminosity,
Here is the characteristic inflow speed into the reconnection region, giving
The bolometric X-ray luminosity is obtained by integrating the intrinsic X-ray spectrum over the adopted photon-energy range. In our reference configuration we use , , and , leading to [18].
The characteristic acceleration rate in the upstream reconnection region is [18]
So, the corresponding acceleration time reads
which is much faster than the other relevant dynamical processes. Differently from the magnetized-turbulent scenario, here, we assume a pair-dominated plasma, with a proton density much lower than the electron density. Therefore, once injected into the radiative region, protons are no longer actively accelerated and escape ballistically,
Energy-loss mechanisms
High-energy CRs propagating through the coronal environment undergo several energy-loss processes, including synchrotron radiation, BH pair production, and photopion interactions with the ambient radiation field. In the magnetized turbulent-corona scenario, CRs additionally undergo hadronuclear interactions with the thermal proton population. The treatment of these processes is particularly important for determining the steady-state particle distributions. In this work, we compare two different prescriptions: the conventional CEL approximation and the TD approach. While synchrotron and Bethe–Heitler losses are always treated as continuous processes, photohadronic and hadronuclear interactions are described either through their average energy losses or as discrete collisions involving nucleon regeneration and charge exchange.
Continuous-energy-loss approximation
In the conventional CEL approximation, energy losses are described by a continuous energy drift, determined by the average energy transferred from the parent particle to secondary particles. The corresponding operator to the steady-state proton transport equation is
where is the differential proton number density and denotes the total energy-loss rate, defined as
where , , , are the synchrotron, BH pair production, photopion and hadronuclear collisions cooling timescales, respectively; , instead, represents the total cooling proton timescale. As said above, the hadronuclear collision contribution is included only for the magnetized turbulent-corona scenario. The proton synchrotron cooling timescale is
where is the magnetic energy density. Although this expression is employed in both acceleration scenarios, the different magnetic-field strengths lead to different synchrotron cooling timescales. The BH pair production timescale follows from the prescription of Ref. [47], using the background photon distribution adopted for each configuration.
For photopion interactions, we use the Chromo open code with the SOPHIA 2.0 event generator [34, 48] to extract cross sections, branching probabilities, and microscopic leading-nucleon energy fractions useful to calculate the overall timescale. In particular, we start by defining the total photopion interaction rate, namely the frequency with which a given nucleon interacts with an isotropic photon background field [49]
where is the Lorentz factor of the incident nucleon, denotes the photon energy in its rest frame, is the pion-production threshold, and accounts for the background photon distribution,
The corresponding photopion energy-loss timescale is
where is the effective inelasticity, averaged over the background photon field, whose typical values are of the order of . Details about the extraction of microscopic processes and the construction of the corresponding effective quantities are given in App. § 9. Here, to get a qualitative handle on the interaction properties, we note that the cross section peaks close to its threshold, so a nucleon with energy interacts primarily with target photons with energy . Therefore, if the target photon spectrum decreases faster than , the amount of photohadronic target increases with nucleon energy , leading to a larger (see e.g. Ref. [50]). For hadronuclear collisions, we analogously use Chromo to calculate cross sections, branching fractions, and inelasticities employing the Sibyll 2.3e interaction model [34, 51, 52, 53]. In particular, the associated energy-loss timescale satisfies
where denotes the effective hadronuclear inelasticity for the nucleon . By contrast, is the interaction rate for hadronuclear collisions between the nucleon and the proton gas. Further details are provided in Appendix § 10. Here, we note that the hadronuclear cross sections are roughly energy-independent for energies larger than a few GeV, with a typical order of magnitude mb.
In the CEL approximation, these interaction rates contribute only through their average energy losses. Consequently, individual collisions are not followed explicitly, and neither the discrete redistribution of nucleon energies nor proton–neutron conversion is taken into account. In particular, only the proton distribution is evolved in this treatment.
Discrete interactions: the trickle-down approach
The TD approach [39] goes beyond the CEL approximation by treating photohadronic and hadronuclear interactions as discrete processes. Each inelastic collision removes a parent nucleon from its original energy bin and reinjects a leading proton or neutron at a lower energy, depending on the interaction channel and its inelasticity. This treatment explicitly accounts for nucleon regeneration, energy redistribution, and charge-exchange processes. Synchrotron and BH losses are always treated as continuous mechanisms, with the energy-loss rate modified as
On the other hand, inelastic photohadronic and hadronuclear collisions are described through explicit nucleon-removal and secondary-injection terms. For a nucleon , the corresponding collision operator is
where denotes either the non-thermal proton or the neutron differential number densities, and the total interaction rates sum both the hadronic and the photohadronic collisions:
The term describes the production of a daughter nucleon from a parent nucleon ,
In order to construct these secondary injection terms, we use the mean-inelasticity approximation, where the daughter energy is related to the parent energy by
where is the effective channel inelasticity, while is the mean fraction of the parent energy retained by the leading daughter nucleon. The secondary injection term is then
Here is obtained by inverting the energy mapping. The denominator is the Jacobian of the parent-to-daughter energy transformation, ensuring the correct redistribution of particles in differential energy space. represents the partial interaction rate, namely the total interaction rate weighted by the probability of producing the leading nucleon . The specific implementations of and for photohadronic and hadronuclear interactions are reported in Apps. § 9 and B, respectively.
In addition to inelastic interactions, neutrons may escape from the corona or decay into protons. Since neutrons are not magnetically confined, their escape timescale is taken to be the ballistic crossing time, . Their Lorentz-dilated beta-decay timescale is
The corresponding proton injection from neutron decay is approximated as
neglecting the small energy difference between the parent neutron and the daughter proton.
Cosmic-ray evolution equations
Having specified the acceleration, escape, and collision prescriptions, in this section, we write the complete steady-state CR transport equations. We first present the canonical CEL treatment and then discuss the TD equations. Numerical discretization and boundary conditions used to solve the equations are reported in App. § 11.
Continuous-energy-loss transport
The canonical steady-state proton distribution in corona environments can be obtained from a Fokker–Planck (FP) equation [20]
where each term represents, from left to right, the acceleration operator, the collisional energy loss term, the escape and injection rates.
For the magnetized-turbulent scenario, stochastic particle acceleration redistributes the injected particles over a broad energy range, so the high-energy steady-state spectrum is insensitive to the detailed shape of the low-energy seed population. We inject a narrow proton spectrum as,
with and GeV. The prefactor can be interpreted as the phase-space weighting of an isotropic relativistic particle distribution in the momentum space, while the Gaussian factor in logarithmic energy confines the injected particles to a narrow energy range. The injection spectrum peaks slightly above , at approximately 23 GeV. We have also verified that varying the low-energy numerical setup and the seed-injection parameters, results are unchanged. Modeling the acceleration in this way is only an approximated representation of the energization process, as discussed in Refs. [54, 55], but is expected to describe well the qualitative proton spectrum [20].
For the magnetic reconnection case, instead, the acceleration rate is so high that we do not model it as an explicit term in the transport equation for the radiative region. Following Ref. [18], we inject a broken-power-law proton spectrum
where we adopt , assuming a strong guide field [18]. The characteristic break energy is related to the proton magnetization, , where . We adopt , corresponding to TeV, following Ref. [18].
Coupled proton–neutron trickle-down transport
In the TD approach, the steady-state proton and neutron equations are coupled. For the protons, the equation becomes
The steady-state neutron equation is given, instead, by
Normalization of the particle distributions. We employ two different physically motivated normalizations for the acceleration prescriptions. For the magnetized turbulent corona, we follow the energy-budget arguments of Ref. [26], where the ratio between the non-thermal proton energy density () and the turbulent magnetic energy density () is related to the fraction of magnetic dissipation power transferred to non-thermal protons as
where is the power transferred to non-thermal protons, is the turbulent magnetic dissipation power, and is the magnetic reconnection rate. Adopting [26], and the maximal proton acceleration efficiency , we obtain
This normalization follows from the assumption of fast magnetic reconnection in a strongly magnetized turbulent plasma, adopting the approximate Alfvén-speed prescription of Ref. [26]. We consider this condition as a benchmark and impose the same energy-density normalization for both the CEL and TD solutions to isolate the impact of the different energy-loss treatments.
For magnetic reconnection, we normalize the injection spectrum by assuming that a fraction of the magnetic energy budget is supplied to non-thermal protons over the coronal crossing time [18],
The same injected spectrum and normalization are adopted for the CEL and TD calculations, with no subsequent rescaling of the steady-state distributions. Therefore, the comparison between continuous and TD transport is performed at fixed steady-state proton energy density in the turbulent corona, but at fixed injected proton power in magnetic reconnection. These normalizations allow us to study the effect of the particle transport in the two approaches.
Neutrino production rates
Equipped with the steady-state proton and neutron distributions, we compute the differential neutrino production rates from photohadronic and hadronuclear interactions. While photohadronic neutrino production is considered in both acceleration scenarios, hadronuclear contributions are included only in the magnetized turbulent-corona model. The neutrino production rate from photohadronic interactions is estimated following the analytical parametrizations of Ref. [56]. For an isotropic target photon field, the differential production rate of neutrinos of flavor is
where is the differential steady-state proton distribution and is the differential number density of target photons. The dimensionless kinematic variables are
and the lower integration boundary is determined by the threshold for pion production,
The functions describe the energy distributions of secondary neutrinos, including contributions from the decay of secondary mesons and muons.
In the TD treatment, both proton and neutron populations contribute to neutrino production. For neutron-induced photohadronic interactions, we adopt the same neutrino production kernel as for protons, assuming approximate isospin symmetry. Accordingly, the total photohadronic neutrino production rate is approximated as
where the neutron contribution is evaluated by replacing the proton distribution with the corresponding neutron distribution in Eq. (35).
For the magnetized turbulent-corona scenario, we additionally consider neutrino production from inelastic collisions of relativistic nucleons with the thermal proton population. The corresponding differential production rate is given by [57]
where is the inclusive differential cross section for neutrino production.
We evaluate the hadronuclear neutrino spectra using the AAfrag parametrizations discussed in Ref. [57], which provides the energy distributions of secondary neutrinos produced in inelastic proton–proton collisions.
For neutron–proton interactions, we approximate the inclusive neutrino production cross section by its proton–proton counterpart,
This approximation is motivated by the approximate isospin symmetry of hadronic interactions and the similarity of inclusive pion production at high energies. Furthermore, since neutron-induced interactions provide a subdominant contribution to the total all-flavour neutrino emission compared to proton-induced interactions, we consider these approximations acceptable for the goal of our analysis.
This allows us to consistently include neutrino production from the neutron population generated in the TD treatment. Therefore, the total neutrino production rate is
Throughout this work, we consider the all-flavor neutrino production rate, obtained by summing over neutrinos and antineutrinos of all flavors,
This approach, however, assumes that secondary particles, such as pions and muons produced in photohadronic and hadronuclear interactions, decay without undergoing significant energy losses. Given the strong magnetic fields expected in AGN coronae, synchrotron cooling may affect this assumption. For charged pions, the synchrotron cooling timescale becomes comparable to the decay timescale at energies of approximately 2.9 PeV and 0.76 PeV for the magnetized turbulent corona and magnetic reconnection scenarios, respectively. Therefore, pion cooling can be safely neglected over the energy range relevant to the main neutrino emission. For muons, instead, the corresponding critical energies are approximately 156 TeV and 41 TeV. Since neutrinos from muon decay typically carry about one-third of the parent muon energy, synchrotron cooling may become relevant for neutrino energies approaching approximately 50 TeV and 14 TeV in the two cases, respectively. As the predicted neutrino spectra peak below these characteristic energies, the prompt-decay approximation is expected to provide a reasonable description of the main emission features. Nonetheless, in App. § 12, we assess the impact of secondary cooling including synchrotron losses of charged pions and muons. The results show that neutrino fluxes are modified only after the peak, without changing our conclusions about the differences between the TD and CEL treatments.
Magnetized turbulent corona: numerical results
We now compare the CEL and TD treatments for a magnetized turbulent corona, considering both prescriptions for the target photon field. We start by discussing the timescales, and then we show the difference in the CR number densities and in the neutrino production rates.
Physical timescales and neutron escape
Figure 2 and Figure 3 show the characteristic timescales for scenarios I and II, respectively. In both cases, stochastic acceleration takes place on an approximately energy-independent timescale, , whereas diffusive proton escape is considerably slower. Neutrons, instead, escape ballistically on the light-crossing time for our benchmark corona. In the relevant energy range, their escape is nearly instantaneous in comparison with their decay or photohadronic interactions. So proton-to-neutron conversion opens a much faster escape route for particles that would otherwise remain magnetically confined; this route is artificially excluded in the CEL treatment.
Figure 2. Characteristic timescales as a function of particle energy for protons (left panel) and neutrons (right panel) in Scenario I of the magnetized turbulent corona. The left panel shows the proton photohadronic and hadronuclear interaction timescales, together with the total cooling, acceleration, and escape timescales. The right panel shows the neutron photohadronic and hadronuclear interaction timescales, together with the escape and decay timescales. The adopted parameters are , , and .
Figure 3. Same as Figure 2, but for scenario II, in which the background photon field contains only the coronal X-ray component.
In addition to making protons escape faster, the TD approach also quantitatively alters the proton interaction rate, since the photohadronic and hadronic interactions leading to proton-to-neutron conversion are neglected in the CEL approach. The quantitative importance of this effect depends on what limits proton acceleration. In scenario I (Figure 2), the additional optical–UV radiation from the accretion disk enhances BH cooling, which contributes substantially to the total proton energy losses. Consequently, the continuous cooling time already becomes comparable to the acceleration time around the energy at which the proton distribution turns over. In scenario II (Figure 3), the optical–UV component is absent and BH losses are less important. Photopion interactions then play the key role in limiting proton acceleration: around , their interaction time approaches the acceleration time while remaining shorter than the corresponding continuous energy-loss time. Thus, in scenario II, we anticipate a larger correction from the TD approach to the maximum energy reached by non-thermal protons.
This distinction matters because an inelastic collision does more than remove the average energy fraction accounted for by continuous cooling. It removes the parent nucleon from its original energy bin and reinjects a leading nucleon at lower energy; when the daughter is a neutron, it can subsequently escape. Hence, we expect vast changes in the proton residence time in both scenarios, but a more pronounced shift of the high-energy cutoff in scenario II, where photohadronic interactions are the principal process competing with acceleration.
Steady-state nucleon distributions and injection requirements
Figure 4 compares the stationary nucleon distributions for the two scenarios. The difference between the TD and CEL treatment is visually quite striking. The latter shows a pronounced pile-up effect close to the maximum energy, with the number density of low-energy protons close to injection suppressed by almost two orders of magnitude compared to the TD approach. In other words, in the physical TD approach, we have to inject almost two orders of magnitude more protons compared to the artificial CEL case, although both solutions are normalized to the same energy density. The large effect stems from the long confinement time of charged particles. In the CEL treatment, protons can undergo repeated stochastic acceleration and accumulate near the energy where acceleration is balanced by cooling; ultimately they are removed by the comparatively slow diffusive escape. In the TD treatment, instead, protons are removed by their conversion to neutrons which immediately escape or by their conversion to protons at lower energies; around , where , this creates an effective escape channel over a timescale of minutes to hours.
Figure 4. Steady-state proton distributions with the trickle-down treatment (dark-red solid line), the CEL approximation (green dashed line) and the neutron contribution to the trickle-down solution (sky blue line) as functions of the nucleon energy. Left: scenario I. Right: scenario II.
Therefore, the ratio between the non-thermal proton number densities normalizations in the TD and the CEL cases is approximately
This effect can be qualitatively understood with a very simple argument: for a fixed proton energy density and a fixed maximal proton energy , the total number density of protons must come from the balance between the injection rate and the residence time of the protons . Since the latter is substantially overestimated by the CEL approximation, the injection rate that fulfils the required energy density is vastly underestimated in that approach.
In addition, we also find as expected a strong change in the maximum proton energy in scenario II. The additional interaction channel, artificially neglected in the CEL approximation, limits acceleration to lower energies, with protons reaching a cutoff already below , more than a factor 3 below the cutoff retrieved in the CEL approximation. In scenario I, where the dominant energy losses are anyway from BH interactions, this latter effect is less important.
Neutrino emission
These differences propagate into the all-flavor neutrino spectra shown in Figure 5. Since the comparison is performed at fixed stationary proton energy density, the figure isolates changes in the energy distribution and interaction history of the nucleons: the much larger difference in the injection efficiency is basically invisible in the neutrino spectrum.
Figure 5. Neutrino production rates summed over all flavors, comparing the TD (solid dark-red lines) and CEL (dashed green lines) treatments. The left and right panels correspond to scenarios I and II, respectively.
For scenario I (left panel), the two neutrino spectra remain relatively similar. The TD treatment modestly enhances the emission below the spectral maximum, while reducing the high-energy contribution and shifting the maximum to somewhat lower energy. The changes are much more pronounced in scenario II (right panel), as expected. The TD spectrum is enhanced at lower neutrino energies but declines much earlier at high energies; its maximum moves from several tens of TeV to less than 10 TeV, a shift by approximately a factor of four. For the benchmark parameters, the integrated neutrino luminosity is approximately 35% lower than that obtained with the CEL approach, despite the identical normalization of the stationary proton energy density. As discussed in App. § 12, accounting for synchrotron cooling of secondary pions and muons reduces this difference to approximately mately 30%, which we take as a representative estimate of the luminosity reduction when secondary cooling is included.
Thus, neutron escape has two distinct consequences in a turbulent corona. At fixed proton energy density, it modifies the neutrino spectrum, especially when photopion interactions limit acceleration. It also requires in both scenarios a much larger injection rate of protons to sustain a comparable neutrino emission, due to the inefficient confinement of the protons converted to neutrons. This sensitivity to confinement provides a useful contrast with the magnetic-reconnection scenario considered in the next section, where primary protons can already escape ballistically.
Magnetic reconnection: numerical results
In this section, we discuss the results for the magnetic reconnection model. Figure 6 shows the characteristic proton and neutron timescales.
Figure 6. Characteristic timescales for the magnetic-reconnection model with the background photon field of scenario II. Left: proton cooling, photopion interaction and escape timescales. Right: neutron photopion interaction, escape and decay timescales.
Neutron escape is considerably faster than neutron decay, as in the magnetized turbulent-corona model. This effect is enhanced by the compactness of the region, which further reduces the crossing timescale. Moreover, the absence of the OUV photon component increases the relative importance of photopion interactions compared with Bethe–Heitler cooling, as already discussed for scenario II of the turbulent-corona model. However, the impact of neutron escape on the steady-state nucleon distribution is substantially reduced in the magnetic-reconnection model. Unlike in the turbulent corona, where repeated stochastic acceleration and energy-dependent confinement can lead to a pronounced high-energy pile-up, here the accelerated protons are injected with a prescribed broken power-law spectrum and escape on a timescale comparable to the light-crossing time of the interaction region. Therefore, although photohadronic collisions provide an additional channel for removing protons from their original energy bins, ballistic proton escape already limits the residence time of the accelerated particles. The resulting nucleon distributions are shown in the left panel of Figure 7. In contrast to the magnetized turbulent-corona scenario, the CEL and TD treatments predict similar spectral shapes, with the peak of the total nucleon number density per logarithmic energy interval, , shifted towards lower energies by approximately 20% in the TD case. Nevertheless, the two treatments predict different nucleon compositions: while the CEL approximation considers only protons, neutrons account for approximately 25% of the total nucleon energy density in the TD solution. Despite this substantial neutron contribution, the total nucleon energy densities differ by only about 1.6% between the two treatments. These differences become even less pronounced in the neutrino sector, where both approaches predict nearly identical all-flavor neutrino
Figure 7. Results for the magnetic-reconnection model. Left: Steady-state proton distributions with the TD treatment (dark-red solid line), the CEL approximation (green dashed line) and the neutron contribution to the trickle-down solution (sky blue line) as functions of the nucleon energy. Right: all-flavor neutrino production rates calculated using the normalized nucleon distributions for the TD (dark-red solid line) and CEL (green dashed line) treatments.
production spectra, as illustrated in the right panel of Figure 7.
Discussion and conclusions
Neutrons are a crucial degree of freedom in non-thermal environments. While they do not directly participate in particle energization in time-dependent magnetic fields, they take part in the dynamics through hadronic and photohadronic interactions that convert protons into neutrons. This dynamics is well known in several cosmic accelerators: we have highlighted its role in the coronae of AGN, a prominent candidate source for the emission of TeV neutrinos. Here, the simultaneous modeling of proton energization through interactions with magnetic turbulence and proton losses through hadronic and photohadronic interactions, has often led to modeling the latter as a CEL process. We go beyond this approximation and show that surprisingly large effects emerge.
On the one hand, modeling only the CEL part of the photohadronic interactions overestimates the characteristic energy-bin-removal timescale, because it only accounts for the average energy-loss frequency. Since the maximum energy that protons reach is determined by the competition between acceleration and interaction, the CEL treatment generally overestimates the maximum proton energy. On the other hand, the entity of this effect depends on whether photohadronic interactions are the dominant energy losses to begin with. For example, in our scenario I relating to the magnetized turbulent corona, the stronger BH losses are anyway dominant, and they are well-modeled by a CEL treatment, so the reduction in maximal energy is more modest.
A stronger effect comes instead from the proton-to-neutron conversion offering an escape route for protons that would otherwise be diffusively confined in the corona. By converting to neutrons, protons can escape from the corona within minutes to hours, rather than several days as expected from pure magnetic diffusion. As a result, a much larger injection rate of protons is required to sustain the stationary population producing high-energy neutrinos at the level suggested by the NGC 1068 signal. The enhancement is about two orders of magnitude. Of course, if protons already possess a faster escape route than magnetized diffusion, e.g. hydrodynamical turbulent diffusion [21], the corresponding enhancement would be more modest in proportion to this shorter escape timescale. This also explains why the effects of proton-to-neutron conversion are much smaller in our magnetic-reconnection scenario, where protons already escape rapidly from the corona.
The effect on nucleons is directly reflected in the neutrino emission. Compared with the CEL approach, the full treatment of photohadronic interactions shifts the neutrino spectrum towards lower energies, changes its spectral shape, and can reduce the neutrino luminosity by up to about 30% in our benchmark models. Such differences are relevant for interpreting the neutrino emission of individual Seyfert galaxies [25, 26], and for predicting the diffuse neutrino flux from their population [22, 23, 30, 31]. In particular, the characteristic neutrino energy and spectral shape are not determined by the acceleration mechanism alone, but by its competition with proton confinement, charge-exchange interactions, and the different cooling channels. Neutrino searches based on physical spectral templates, rather than simple power laws, could help disentangle these effects.
The fate of the escaping neutrons is also interesting in another respect, in that they may lead to additional neutrinos, either from their direct beta decay, or from their hadronic or photohadronic interactions within the regions outside the corona. We do not investigate this question, which depends on the outer regions beyond the one-zone coronal scenario.
Our results are motivated by the NGC 1068 case. The quantitative impact of neutron escape will depend on the properties of individual sources, notably their photon fields, magnetic fields, and particle confinement timescales. Nevertheless, our work establishes a general point: in environments where particles are magnetically confined and accelerated, photohadronic interactions cannot always be reduced to a continuous energy loss. The conversion of energetic charged particles into freely escaping neutral ones can fundamentally change the energy budget of the accelerator, and thus the physical requirements for explaining its neutrino emission.
Note added in preparation. While this work was being finalized, a recent work appeared [58] also investigating particle acceleration and nuclear transport in Seyfert coronae, accounting for proton-to-neutron conversion as a catastrophic loss channel during acceleration. In that approach, secondary nucleons are evolved in a separate nuclear cascade, disentangled from acceleration. Instead, our treatment couples proton and neutron transport directly to stochastic acceleration, including the reinjection and subsequent acceleration of protons produced in hadronic and photohadronic interactions. While Ref. [58] focuses on the impact of nuclear composition on multimessenger emission, our work investigates the impact of discrete nucleon interactions and neutron escape on the stationary proton population and the resulting neutrino emission.
Acknowledgments
AA is thankful to Fabio Convenga for sharing useful suggestions on the graphical style and layout of the figures presented in this work. AA gratefully acknowledges the Pierre Auger Collaboration for helpful discussions that contributed to this work. DFGF is grateful to Luca Comisso, Maria Petropoulou, Enrico Peretti, and Federico Testagrossa for helpful discussions on this work. AA and DFGF acknowledge the support by the research project TAsP (Theoretical Astroparticle Physics) funded by the Istituto Nazionale di Fisica Nucleare (INFN).
Details of the photopion treatment
In this appendix, we specify how the microscopic photopion properties are extracted from event simulations and converted into the coefficients entering the CEL and TD prescriptions. In particular, we focus on the microscopic channel identification, the photon-field averages, and the specific assumptions underlying the implementation.
Microscopic photopion properties
We use Chromo [34] with the SOPHIA 2.0 event generator [48] to determine the inelastic photopion cross section and the properties of the final-state baryons. Photon–proton collisions are simulated for different photon energies in the proton rest frame. Following Refs. [48, 59], we assign to each final-state particle the light-cone energy fraction
where the incident photon defines the positive longitudinal direction. In the ultrarelativistic limit, approximates the fraction of the incident proton energy carried by particle in the coronal frame. For TD transport, we select the final-state proton or neutron with the largest as the leading nucleon. The microscopic probability for an incident proton to produce a leading nucleon of species and its conditional mean energy fraction are
where is the number of simulated inelastic interactions at a given and counts events assigned to channel . The average is taken only over events assigned to that channel. The corresponding microscopic channel inelasticity, , describes the mean energy not retained by that leading nucleon.
For the CEL, we define
with the average taken over events for which a leading baryon is identified. In other words, we do not restrict the leading particle to be a proton or neutron, allowing for other baryonic species in the final state. This choice is motivated by the CEL approximation, in which inelastic interactions are described through their average energy losses, while the proton identity is assumed to remain unchanged throughout the transport. This quantity is extracted directly from the Monte Carlo events, rather than reconstructed from the two propagated leading-nucleon channels. Consequently, the microscopic inputs used for CEL and TD treatments do not coincide exactly.
As shown in Figure 8, both the channel probabilities and energy fractions depend on , particularly near the photopion threshold. At higher photon energies, and approach approximately and , respectively. The channel inelasticities are typically of order 20% near threshold and increase with . The leading-baryon quantity is not, in general, identical to either channel inelasticity or to their probability-weighted average, since it is defined using a different final-state selection.
Figure 8. Microscopic photopion properties as functions of the incident photon energy in the proton rest frame. Left: probabilities and for the leading-nucleon channels. Right: corresponding channel inelasticities and , compared with the leading-baryon inelasticity employed for CEL cooling.
Effective coefficients for CEL and TD transport
The microscopic coefficients determined in the previous subsection must be averaged over the radiation field of the coronal configuration, in order to determine the corresponding macroscopic quantities. Using the photon distribution and the function defined in Eq. (18), we introduce the differential interaction weight
where . This weight has dimensions of an interaction rate per unit ; its integral over gives the total rate already defined in Eq. (17). All photon energies and tabulated cross sections are evaluated in consistent units.
For TD transport, weighting the interaction probability by the microscopic channel probability yields the partial interaction rate (directly evaluated for ; the neutron channels are constructed as described below)
and the channel-conditioned mean daughter energy fraction is
The corresponding channel inelasticity is , consistently with Eq. (25). The rate and the energy fraction above are inserted into the general reinjection formula, Eq. (26). In particular, the Jacobian contains the derivative of the photon-field-averaged function , not the derivative of the microscopic Monte Carlo table.
The effective photopion inelasticity entering CEL cooling is computed separately from the leading-baryon observable,
This coefficient enters the CEL loss time through Eq. (19). It is important to distinguish from the TD channel inelasticities : the former follows the energy of the leading baryon irrespective of species, while the latter follow only explicitly propagated leading proton or neutron channels. Both prescriptions use the same underlying interaction simulations and photon-field weight, but they employ different microscopic observables. For diagnostic purposes, we define the effective branching fractions
They are not independent inputs when the partial photopion rates are calculated directly through Eq. (A.5).
Figure 9 illustrates the effect of averaging over the target photon distribution in Scenario I. The two TD energy fractions are conditional on different final-state channels, whereas is averaged from the leading-baryon observable. These distinctions are preserved when constructing the transport coefficients.
Figure 9. Photopion coefficients after convolution with the photon field of Scenario I, as functions of the parent proton energy. Left: effective branching fractions and and their sum. Right: TD channel inelasticities and , compared with the CEL coefficient .
Unassigned channels and neutron-initiated interactions
As said above, not every simulated photopion event can be assigned to a leading proton or neutron. In this subsection, we investigate their impact over the TD equations. To this purpose, we define the microscopic residual probability and its photon-field-averaged counterpart as,
Such interactions are retained in the total interaction rate, but do not produce an explicit leading-proton or leading-neutron reinjection. The microscopic residual reaches approximately 7% at some photon energies in the interpolated input tables. After averaging over the photon field of Scenario I, the residual remains below approximately 0.5% in the energy range relevant for photopion transport. Its fraction among interactions sampled by the stationary TD proton population is
We obtain for Scenario I. Therefore, we conclude that the impact of unassigned events cannot impact the solutions of the TD transport equations.
The microscopic channel tables described above are obtained for photon–proton collisions. For neutron-initiated interactions, the total rate is evaluated independently, while the effective branching fractions and daughter energy fractions are related to the proton channels by approximate isospin symmetry,
Therefore, the neutron partial rates are constructed as . The energy derivatives required by the TD Jacobian are inherited from the corresponding proton energy fractions. This approximation does not assume that .
Details of the hadronuclear treatment
In this section, we describe the microscopic inputs and their use in the CEL and TD transport prescriptions for hadronuclear collisions obtained from Chromo [34] with Sibyll 2.3e [51, 52, 53]. In particular, we specify the cross sections, the leading-nucleon selection, and the effective coefficients entering those expressions.
Inelastic hadronuclear cross sections
We extract the inelastic cross sections from the event generator as functions of the projectile energy. Since in the Chromo–Sibyll setup, the energy range starts at GeV [53], we construct a hybrid cross section by combining the Monte Carlo predictions with the analytical parametrization of Ref. [60]. In particular, we adopt
where GeV. In other words, we allow the analytical prescription to describe the lower energies while retaining the Monte Carlo predictions at higher energies. The analytical cross section is given by [60]
where and GeV is the kinematic energy threshold. For neutron–proton interactions, we employ the same treatment simply assuming that for GeV. The adopted prescription, for pp, is shown in Figure 10 for reference.
Figure 10. Inelastic hadronuclear cross sections as functions of the projectile energy for pp interactions. The function used in this work is represented by a solid green line while the analytical parametrization of Ref. [60] is shown as a dashed blue line. The vertical dotted line indicates the matching energy of .
Microscopic leading-nucleon properties
To determine the energy redistribution in each channel, we generate inelastic events for both proton and neutron projectiles. We identify the leading final-state nucleon among outgoing protons and neutrons with positive longitudinal momentum (), selecting the one that maximizes the forward light-cone fraction [53]
where and are the energy and longitudinal momentum of the final-state nucleon, and and refer to the incident nucleon. The fraction of the projectile energy retained by the selected nucleon is
This quantity, rather than the light-cone fraction itself, is used to determine the energy of the daughter nucleon in the TD transport calculation.
The microscopic branching fraction and conditional mean daughter energy fraction are given by
The denominator includes all simulated inelastic events, whereas the conditional average is restricted to events with a leading nucleon of species . The channel inelasticity is . As for photohadronic interactions, also in this case, a small fraction of events cannot be assigned to either leading-nucleon channel. The corresponding residual probability is given by
Such events are included in the total interaction rate but not in either reinjection channel. The effective inelasticity associated with the retained energy of the explicitly propagated leading nucleons is defined as
In this definition, interactions without an identified leading proton or neutron contribute to the total interaction rate but do not produce explicit nucleon reinjection terms. Consequently, any energy retained by baryonic secondaries in these residual channels is not tracked in the nucleon transport calculation. We find this approximation reasonable given that hadronuclear processes are subdominant to shape the overall cosmic rays dynamics in the magnetized turbulent corona scenario (see Figure 2).
Figure 11 shows that the branching fractions and mean daughter energy fractions exhibit relatively weak energy dependence across the range relevant for the present calculation. In particular, the effective leading-nucleon inelasticities constructed from the channel coefficients vary by less than approximately 1% in the range considered. Therefore, we evaluate and at and keep them constant in the transport calculation. The Monte Carlo energy dependence is smoothed using a five-point moving average and interpolated in logarithmic projectile energy before extracting the reference values. The resulting coefficients are listed in Table 1.
Figure 11. Leading-nucleon properties of inelastic and collisions obtained with Chromo and Sibyll 2.3e. Left: branching fractions for the , , , and channels. Right: corresponding mean daughter energy fractions .
| Quantity | ||
| 0.6665 | 0.6673 | |
| 0.3272 | 0.3261 | |
| 0.4934 | 0.4920 | |
| 0.3595 | 0.3607 | |
| 0.5536 | 0.5540 |
Table 1. Effective hadronuclear parameters at obtained from Sibyll 2.3e. In the column the initial nucleon is ; in the column it is . The effective inelasticity is constructed from the same leading-nucleon coefficients used in the TD reinjection.
Equipped with all these quantities, analogously to the photohadronic collisions, we define the partial interaction rates entering the TD reinjection terms by
where is obtained from the cross section of Eq. B.1. Likewise, the energy mapping between parent and daughter nucleons becomes
Consequently, the TD reinjection term takes the simplified form
where .
Numerical implementation of the transport equations
In this appendix, we describe the numerical procedures adopted to solve the steady-state transport equations presented in the main text. Both acceleration scenarios are evaluated on a logarithmically spaced energy grid covering to , with 220 grid points. The numerical treatment differs between the magnetized turbulent-corona and magnetic-reconnection scenarios because stochastic acceleration is explicitly included only in the former.
Fokker–Planck discretization
In the magnetized turbulent-corona scenario, the proton transport equation contains both stochastic acceleration and continuous energy losses. We solve the resulting second-order differential equation using a finite-volume discretization in energy space.
The differential terms of the transport equation are expressed through the energy-space particle flux. We denote by the grid energies and define the interfaces between adjacent cells as . The corresponding cell width is , with the outer interfaces obtained by linear extrapolation from the adjacent grid points.
The stochastic-acceleration contribution to the flux at the interface between cells and is discretized as
where and .
Since continuous energy losses transport particles from higher to lower energies, we adopt a first-order upwind prescription for the corresponding flux,
where . The total energy-space flux is obtained by summing the two contributions.
The steady-state transport equation is then discretized in conservative form,
where collects the escape and, when applicable, the discrete interaction rates, while contains the primary injection and secondary reinjection terms.
The resulting linear system is solved by imposing a vanishing energy-space flux at the lower boundary and an absorbing condition at the upper boundary [24],
The first condition is implemented by excluding any flux through the lower external cell interface, while the second is imposed directly on the last row of the discretized operator.
Magnetic-reconnection discretization
In the magnetic-reconnection scenario, proton acceleration is incorporated into the prescribed injection spectrum rather than explicitly modeled in the radiative-region transport equation. Consequently, the equation contains only continuous energy losses, escape, and, in the TD approach, discrete interaction sinks and reinjection terms.
The first-order energy derivative is discretized using an upwind scheme. Denoting by the appropriate continuous energy-loss rate and by the sum of the escape and discrete interaction rates, the resulting equation reads
For the CEL calculation, contains only proton escape, whereas in the TD calculation it also includes the total photohadronic interaction rate. The discretized operator is upper bidiagonal and can therefore be solved by backward substitution, starting from the absorbing boundary . No additional lower-energy boundary condition is required.
Iterative solution and convergence for the TD equations
The TD treatment requires the numerical evaluation of the secondary source terms associated with the different nucleon production channels. For each channel , the parent-to-daughter energy mapping defined in Eq. (25) is evaluated on the energy grid. The resulting relation between parent and daughter energies is numerically inverted by interpolation to determine the parent energy corresponding to each daughter energy .
The partial interaction rate, parent particle distribution, and effective daughter energy fraction are then interpolated at . The Jacobian entering Eq. (26) is evaluated using the energy derivative of the effective daughter energy fraction. For photohadronic interactions, these derivatives are computed numerically from the energy-dependent effective fractions using second-order accurate finite differences. For hadronuclear interactions, the adopted constant energy fractions imply a vanishing derivative, reducing the Jacobian to the corresponding daughter energy fraction.
The coupled proton–neutron TD equations are solved using a fixed-point iterative procedure. Starting from an initial estimate of the particle distributions, we evaluate the photohadronic and hadronuclear reinjection terms and update the proton and neutron solutions. The updated distributions are then used to reconstruct the secondary source terms at the following iteration.
To improve numerical stability, the particle distributions are updated through underrelaxation,
where denotes the solution obtained before mixing and is the relaxation parameter.
For the magnetized turbulent-corona scenario, we initialize the proton distribution using the solution without secondary reinjection and set the initial neutron distribution to zero. We adopt , a relative convergence tolerance of , and a maximum of 2000 iterations. Convergence is monitored through the maximum relative change of both nucleon distributions, with a small numerical floor introduced to avoid artificially large relative differences in regions where the distributions are negligible.
For magnetic reconnection, both particle distributions are initially set to zero. We employ the same relaxation parameter, , with a relative tolerance of and a maximum of 1000 iterations. In this case, the convergence criterion is evaluated from the changes between the current distributions and the corresponding unmixed updates.
After the linear solves, numerical safeguards are applied to remove non-finite entries and small negative values from the particle distributions.
Impact of secondary pion and muon cooling
The neutrino production prescription adopted in the main text assumes that unstable secondaries decay without significant energy losses. In this appendix, we quantitatively assess this approximation by explicitly following the production, synchrotron cooling, and decay of charged pions and muons. We use the same parameters and the steady-state solutions as obtained in the main text for the CEL and TD treatments. We start by discussing the pions and muons timescales, then we report their transport equation and finally how we evaluate the neutrino emissions from their decay.
Synchrotron cooling and decay timescales
For a relativistic charged particle , the decay timescale scales linearly with the energy as for the decay timescale of neutrons. By contrast, the synchrotron cooling timescale decreases with energy and magnetic field strength. The critical energy at which the two timescales are equal is given by
where and are the mass and proper lifetime of the particle, respectively.
As shown in Figure 12, the critical muon energies are approximately 156 TeV for the magnetized turbulent corona and 41 TeV for magnetic reconnection. The corresponding values for charged pions are considerably higher, approximately 2.87 PeV and 0.76 PeV, respectively. Since a neutrino from muon decay typically carries a fraction of order one third of its parent’s energy, the characteristic neutrino energies associated with muon cooling are TeV and TeV.
Figure 12. Decay (solid) and synchrotron-cooling (dashed) timescales of charged muons (dark red) and pions (blue) for the magnetized turbulent corona (left) and magnetic-reconnection (right) configurations. Vertical dotted lines mark the energies at which the two timescales become equal for each particle species. The adopted magnetic fields are and , respectively.
Stationary transport of charged secondaries
In order to account for the possible energy losses of muons and pions, we solve their transport equation as
where denotes the injection rate. For pions, it comes directly from the photohadronic and hadronuclear collisions, while for muons it comes from the fraction of pions that decay rather than losing energy. They are specified in the next section. , instead, represents the equilibrium differential number density and it is given by
where
Injection of charged pions and muons
In order to calculate the injection of charged pions, we first note that neutrino emission is dominated by photohadronic interactions. Therefore, in what follows, we neglect the hadronuclear contribution. We adopt a channel-based response-function formalism analogous to that implemented in the AM framework [61]. In particular, we employ the simplified photohadronic model Sim-B of Ref. [62], including resonance, direct, and multipion-production channels. Unlike AM, which adopts the more detailed Sim-A prescription, our implementation uses Sim-B to provide a simplified treatment of secondary particle production and cooling. For either incident nucleon , each interaction channel is characterized by the average multiplicity of a charged pion () and the fraction of the incident nucleon energy carried by that pion. The charged-pion source term is
where
As in the main text, , while represents the stationary nucleon density, and is the differential target-photon number density. The so-called angle-averaged response function, reads
with . The parametrized cross sections, multiplicities, and energy fractions are taken from the Sim-B channel tables of Ref. [62].
Subsequently, the pions decay. Firstly, we determine the rate with which pions decay as
For the muons, instead, the injection rate is
where are helicity-dependent functions [63] and . For these functions we follow Ref. [63]
defined for and vanish outside. Therefore, we solve the transport equations separately for right-handed and left-handed muons.
All-flavor neutrino production
Equipped with the pion and muons number densities, we determine the all-flavor neutrino flux following the approach of [63]. In particular, the neutrino production rate from the pions decay is given by
where
The neutrino production rate from the muons decay, instead, is
The functions in this case are given by [63]
where and denote right- and left-handed muons, respectively. For the charge-conjugate decay, the sign of is reversed. The final all-flavor neutrino production rate is .
Figure 13 compares the all-flavor neutrino production rates obtained using the Kelner–Aharonian (KA) photohadronic parametrization [56], adopted in the main text, with those computed using the simplified Sim-B approach introduced in this Appendix. For this comparison, synchrotron cooling of secondary pions and muons is neglected, such that and . The two approaches show good agreement in both spectral shape and normalization, with differences in the integrated neutrino luminosities below approximately 10% for the magnetized corona scenarios. This agreement supports the reliability of the simplified framework adopted here to investigate the impact of secondary cooling.
Figure 13. Comparison of the all-flavor photohadronic neutrino production rates obtained using the Kelner–Aharonian (KA, dashed lines) and Sim-B (solid lines) parametrizations, neglecting synchrotron cooling of secondary particles. The upper panels correspond to the magnetized corona, considering scenarios I (left) and II (right), while the lower panel refers to the magnetic reconnection scenario. Dark-red and green lines represent the TD and CEL treatments, respectively. The overall agreement between the two parametrizations supports the use of the simplified Sim-B framework to investigate the impact of secondary cooling.
Figure 14 compares the all-flavor neutrino production rates obtained with (solid lines) and without (dotted lines) synchrotron cooling of secondary pions and muons. For Scenario I of the magnetized turbulent corona, the two treatments yield nearly identical neutrino spectra, indicating that secondary cooling has a negligible impact. For Scenario II, synchrotron cooling mainly suppresses the high-energy part of the neutrino spectra, with a more pronounced effect on the CEL solution. Nevertheless, the significant difference between the TD and CEL spectra persists. In the magnetic reconnection scenario, cooling primarily affects the neutrino emission above the spectral peak, leaving the TD and CEL solutions almost unchanged relative to each other.
Figure 14. All-flavor neutrino production rates obtained using the simplified Sim-B framework, with (solid lines) and without (dotted lines) synchrotron cooling of secondary charged pions and muons. The upper-left and upper-right panels correspond to scenarios I and II of the magnetized corona, respectively, while the lower panel refers to the magnetic reconnection scenario. Dark-red and green lines represent the TD and CEL treatments, respectively.
Quantitatively, for Scenario II of the magnetized turbulent corona, the KA parametrization adopted in the main text predicts a reduction of 35% in the integrated neutrino luminosity when comparing the TD and CEL treatments. Within the simplified Sim-B framework considered here, the corresponding reduction is approximately 39% in the absence of secondary cooling, decreasing to about 30% when synchrotron losses of pions and muons are included. The difference between the two predictions without cooling reflects the different photohadronic parametrizations. Therefore, we take the 30% reduction as a representative estimate when secondary cooling is accounted for, confirming that the main conclusions remain unchanged. In the magnetic reconnection scenario, the corresponding difference remains at the percent level. Therefore, although secondary cooling modifies the neutrino spectra and their integrated luminosities, the main conclusions presented in the main text remain unchanged within the simplified framework considered here.
References
- [1]M. G. Aartsen et al. Evidence for High-Energy Extraterrestrial Neutrinos at the IceCube Detector. Science, 342:1242856, 2013.DOI
- [2]M. Ackermann et al. The spectrum of isotropic diffuse gamma-ray emission between 100 MeV and 820 GeV. Astrophys. J., 799:86, 2015.
- [3]Kohta Murase, Dafne Guetta, and Markus Ahlers. Hidden Cosmic-Ray Accelerators as an Origin of TeV-PeV Cosmic Neutrinos. Phys. Rev. Lett., 116(7):071101, 2016.DOI
- [4]Xiao-Chuan Chang, Ruo-Yu Liu, and Xiang-Yu Wang. How far are the sources of IceCube neutrinos? Constraints from the diffuse TeV gamma-ray background. Astrophys. J., 825:148, 2016.DOI
- [5]Di Xiao, Peter Mészáros, Kohta Murase, and Zi-gao Dai. Revisiting the Contributions of Supernova and Hypernova Remnants to the Diffuse High-energy Backgrounds: Constraints on Very High Redshift Injection. Astrophys. J., 826(2):133, 2016.DOI
- [6]Takahiro Sudoh, Tomonori Totani, and Norita Kawanaka. High-energy gamma-ray and neutrino production in star-forming galaxies across cosmic time: Difficulties in explaining the IceCube data. Publ. Astron. Soc. Jap., 70(3):49, 2018.DOI
- [7]Antonio Capanema, Arman Esmaili, and Kohta Murase. New constraints on the origin of medium-energy neutrinos observed by IceCube. Phys. Rev. D, 101(10):103012, 2020.DOI
- [8]Antonio Capanema, Arman Esmaili, and Pasquale Dario Serpico. Where do IceCube neutrinos come from? Hints from the diffuse gamma-ray flux. JCAP, 02:037, 2021.DOI
- [9]Ke Fang, John S. Gallagher, and Francis Halzen. The TeV Diffuse Cosmic Neutrino Spectrum and the Nature of Astrophysical Neutrino Sources. Astrophys. J., 933(2):190, 2022.DOI
- [10]Antonio Ambrosone, Marco Chianese, Damiano F. G. Fiorillo, Antonio Marinelli, Gennaro Miele, and Ofelia Pisanti. Starburst galaxies strike back: a multi-messenger analysis with Fermi-LAT and IceCube data. Mon. Not. Roy. Astron. Soc., 503(3):4032–4049, 2021.DOI
- [11]R. Abbasi et al. Evidence for neutrino emission from the nearby active galaxy NGC 1068. Science, 378(6619):538–543, 2022.DOI
- [12]V. A. Acciari et al. Constraints on gamma-ray and neutrino emission from NGC 1068 with the MAGIC telescopes. Astrophys. J., 883:135, 2019.DOI
- [13]Yoshiyuki Inoue, Dmitry Khangulyan, and Akihiro Doi. On the Origin of High-energy Neutrinos from NGC 1068: The Role of Nonthermal Coronal Activity. Astrophys. J. Lett., 891(2):L33, 2020.DOI
- [14]Ali Kheirandish, Kohta Murase, and Shigeo S. Kimura. High-energy Neutrinos from Magnetized Coronae of Active Galactic Nuclei and Prospects for Identification of Seyfert Galaxies and Quasars in Neutrino Telescopes. Astrophys. J., 922(1):45, 2021.DOI
- [15]Kohta Murase. Hidden Hearts of Neutrino Active Galaxies. Astrophys. J. Lett., 941(1):L17, 2022.DOI
- [16]Björn Eichmann, Foteini Oikonomou, Silvia Salvatore, Ralf-Jürgen Dettmar, and Julia Becker Tjus. Solving the Multimessenger Puzzle of the AGN-starburst Composite Galaxy NGC 1068. Astrophys. J., 939(1):43, 2022.DOI
- [17]Rostom Mbarek, Alexander Philippov, Alexander Chernglazov, Amir Levinson, and Richard Mushotzky. Interplay between accelerated protons, x rays and neutrinos in the corona of NGC 1068: Constraints from kinetic plasma simulations. Phys. Rev. D, 109(10):L101306, 2024.DOI
- [18]Damiano F. G. Fiorillo, Maria Petropoulou, Luca Comisso, Enrico Peretti, and Lorenzo Sironi. TeV Neutrinos and Hard X-Rays from Relativistic Reconnection in the Corona of NGC 1068. Astrophys. J., 961(1):L14, 2024.DOI
- [19]D. Karavola, M. Petropoulou, D. F. G. Fiorillo, L. Comisso, and L. Sironi. Neutrino and pair creation in reconnection-powered coronae of accreting black holes. JCAP, 04:075, 2025. [Erratum: JCAP 11, E01 (2025)].DOI
- [20]Damiano F. G. Fiorillo, Luca Comisso, Enrico Peretti, Maria Petropoulou, and Lorenzo Sironi. A Magnetized Strongly Turbulent Corona as the Source of Neutrinos from NGC 1068. Astrophys. J., 974(1):75, 2024.DOI
- [21]Martin Lemoine and Frank Rieger. Neutrinos from stochastic acceleration in black hole environments. Astron. Astrophys., 697:A124, 2025.DOI
- [22]Antonio Ambrosone. Berezinsky hidden sources: an emergent tension in the high-energy neutrino sky? JCAP, 09:075, 2024.arxiv.org/abs/2406.13336
- [23]Damiano F. G. Fiorillo, Luca Comisso, Enrico Peretti, Maria Petropoulou, and Lorenzo Sironi. The Contribution of Turbulent Active Galactic Nucleus Coronae to the Diffuse Neutrino Flux. Astrophys. J., 989(2):215, 2025.DOI
- [24]Chengchao Yuan, Damiano F. G. Fiorillo, Maria Petropoulou, and Qinrui Liu. Coupled time-dependent proton acceleration and leptonic-hadronic radiation in turbulent supermassive black hole coronae. Phys. Rev. D, 113(4):043016, 2026.DOI
- [25]Jose Alonso Carpio, Ali Kheirandish, and Kohta Murase. Multimessenger Characterization of High-energy Neutrino Emission from the Brightest Neutrino-active Galactic Nuclei. Astrophys. J., 1005(1):61, 2026.DOI
- [26]Federico Testagrossa, Damiano F. G. Fiorillo, Luca Comisso, Enrico Peretti, Maria Petropoulou, and Lorenzo Sironi. Neutrinos and Gamma Rays from Seyfert Galaxies Constrain the Properties of Coronal Turbulence. Astrophys. J., 1004(1):4, 2026.
- [27]Luana Passos-Reis, Elisabete M. de Gouveia Dal Pino, Juan C. Rodríguez-Ramírez, and Giovani H. Vicentin. A Turbulence-Driven Magnetic Reconnection Model for the High-Energy Neutrino Emission from NGC 1068. 4 2026.arxiv.org/abs/2604.10847
- [28]Sibasis Laha, Claudio Ricci, John C. Mather, Ehud Behar, Luigi C. Gallo, Frederic Marin, Rostom Mbarek, and Amelia Hankla. X-ray properties of coronal emission in radio quiet active galactic nuclei. Front. Astron. Space Sci., 11:1530392, 2024.
- [29]Abhishek Das, Kohta Murase, and B. Theodore Zhang. Multimessenger Constraints on Production Sites of High-energy Neutrinos from NGC 1068. Astrophys. J. Lett., 1009(2):L27, 2026.DOI
- [30]Kohta Murase, Shigeo S. Kimura, and Peter Mészáros. Hidden Cores of Active Galactic Nuclei as the Origin of Medium-Energy Neutrinos: Critical Tests with the MeV Gamma-Ray Connection. Phys. Rev. Lett., 125(1):011101, 2020.DOI
- [31]D. Karavola, M. Petropoulou, D. F. G. Fiorillo, A. Georgakakis, L. Comisso, and L. Sironi. Diffuse neutrino flux from relativistic reconnection in AGN coronae. Astron. Astrophys., 709:A94, 2026.DOI
- [32]Sébastien Le Bihan, Martin Lemoine, and Franck Rieger. Self-consistent modelling of neutrino production in turbulent black hole coronae. Astron. Astrophys., 712:A218, 2026.DOI
- [33]Shigeo S. Kimura, Kohta Murase, and Kenji Toma. Neutrino and Cosmic-Ray Emission and Cumulative Background from Radiatively Inefficient Accretion Flows in Low-Luminosity Active Galactic Nuclei. Astrophys. J., 806:159, 2015.arxiv.org/abs/1411.3588
- [34]Anatoli Fedynitch, Hans Dembinski, and Anton Prosekin. Chromo: A high-performance python interface to hadronic event generators for collider and cosmic-ray simulations. Comput. Phys. Commun., 321:110031, 2026.
- [35]Ellis R. Owen, Kinwah Wu, Yoshiyuki Inoue, Tatsuki Fujiwara, Qin Han, and Hayden P. H. Ng. The ‘Forgotten’ Neutrons: Implications for the Propagation of High-Energy Cosmic Rays in Magnetized Astrophysical and Cosmological Structures. Universe, 12(4):94, 2026.
- [36]Mitchell C. Begelman, Bronislaw Rudak, and Marek Sikora. Consequences of Relativistic Proton Injection in Active Galactic Nuclei. Astrophys. J., 362:38, October 1990.
- [37]Marek Sikora, Mitchell C. Begelman, and Bronislaw Rudak. Relativistic Neutrons in Active Galactic Nuclei. Astrophys. J. Lett., 341:L33, June 1989.
- [38]K. Toma and F. Takahara. Baryon Loading of Active Galactic Nucleus Jets Mediated by Neutrons. Astrophys. J., 754(2):148, August 2012.DOI
- [39]Michael Unger, Glennys R. Farrar, and Luis A. Anchordoqui. Origin of the ankle in the ultrahigh energy cosmic ray spectrum, and of the extragalactic protons below it. Phys. Rev. D, 92(12):123001, 2015.DOI
- [40]P. Padovani et al. High-energy neutrinos from the vicinity of the supermassive black hole in NGC 1068. Nature Astron., 8(9):1077–1087, 2024.arxiv.org/abs/2405.20146
- [41]A. Marconi, G. Risaliti, R. Gilli, L. K. Hunt, R. Maiolino, and M. Salvati. Local supermassive black holes, relics of active galactic nuclei and the x-ray background. Mon. Not. Roy. Astron. Soc., 351:169, 2004.
- [42]N. I. Shakura and R. A. Sunyaev. Black holes in binary systems. Observational appearance. Astron. Astrophys., 24:337–355, January 1973.
- [43]Matias Montesinos. Review: Accretion Disk Theory. arXiv e-prints, page arXiv:1203.6851, March 2012.arxiv.org/abs/1203.6851
- [44]Luca Comisso and Lorenzo Sironi. The interplay of magnetically-dominated turbulence and magnetic reconnection in producing nonthermal particles. Astrophys. J., 886:122, 2019.DOI
- [45]Philipp Kempski, Drummond B. Fielding, Eliot Quataert, Alisa K. Galishnikova, Matthew W. Kunz, Alexander A. Philippov, and Bart Ripperda. Cosmic ray transport in large-amplitude turbulence with small-scale field reversals. Mon. Not. Roy. Astron. Soc., 525(4):4985–4998, 2023.DOI
- [46]Luca Comisso, Glennys R. Farrar, and Marco S. Muzio. Ultra-High-Energy Cosmic Rays Accelerated by Magnetically Dominated Turbulence. Astrophys. J. Lett., 977(1):L18, 2024.DOI
- [47]Michal J. Chodorowski, Andrzej A. Zdziarski, and Marek Sikora. Reaction Rate and Energy-Loss Rate for Photopair Production by Relativistic Nuclei. The astrophysical Journal, 400:181, November 1992.DOI
- [48]A. Mucke, Ralph Engel, J. P. Rachen, R. J. Protheroe, and Todor Stanev. SOPHIA: Monte Carlo simulations of photohadronic processes in astrophysics. Comput. Phys. Commun., 124:290–314, 2000.
- [49]Roberto Aloisio, Denise Boncioli, Armando Di Matteo, Aurelio F. Grillo, Sergio Petrera, and Francesco Salamida. SimProp v2r4: Monte Carlo simulation code for UHECR propagation. JCAP, 11:009, 2017.DOI
- [50]Damiano F. G. Fiorillo, Arjen Van Vliet, Stefano Morisi, and Walter Winter. Unified thermal model for photohadronic neutrino production in astrophysical sources. JCAP, 07:028, 2021.
- [51]Eun-Joo Ahn, Ralph Engel, Thomas K. Gaisser, Paolo Lipari, and Todor Stanev. Cosmic ray interaction event generator SIBYLL 2.1. Phys. Rev. D, 80:094003, 2009.DOI
- [52]Anatoli Fedynitch, Felix Riehn, Ralph Engel, Thomas K. Gaisser, and Todor Stanev. Hadronic interaction model sibyll 2.3c and inclusive lepton fluxes. Phys. Rev. D, 100(10):103018, 2019.DOI
- [53]Felix Riehn, Ralph Engel, Anatoli Fedynitch, Thomas K. Gaisser, and Todor Stanev. Hadronic interaction model Sibyll 2.3d and extensive air showers. Phys. Rev. D, 102(6):063002, 2020.DOI
- [54]Virginia Bresci, Martin Lemoine, Laurent Gremillet, Luca Comisso, Lorenzo Sironi, and Camilia Demidem. Nonresonant particle acceleration in strong turbulence: Comparison to kinetic and MHD simulations. Phys. Rev. D, 106(2):023028, 2022.DOI
- [55]Martin Lemoine. First-Principles Fermi Acceleration in Magnetized Turbulence. Phys. Rev. Lett., 129(21):215101, 2022.DOI
- [56]S. R. Kelner and F. A. Aharonian. Energy spectra of gamma-rays, electrons and neutrinos produced at interactions of relativistic protons with low energy radiation. Phys. Rev. D, 78:034013, 2008. [Erratum: Phys.Rev.D 82, 099901 (2010)].DOI
- [57]S. Koldobskiy, M. Kachelrieß, A. Lskavyan, A. Neronov, S. Ostapchenko, and D. V. Semikoz. Energy spectra of secondaries in proton-proton interactions. Phys. Rev. D, 104(12):123027, 2021.DOI
- [58]Chengchao Yuan. Multimessenger Constraints on Nuclear Acceleration in Neutrino-Emitting Seyfert Coronae. 10 2026.arxiv.org/abs/2610.18671
- [59]John C. Collins. Light cone variables, rapidity and all that. 5 1997.DOI
- [60]S. R. Kelner, Felix A. Aharonian, and V. V. Bugayov. Energy spectra of gamma rays, electrons, and neutrinos produced at proton-proton interactions in the very high energy regime. Phys. Rev. D, 74:034018, 2006. [Erratum: Phys. Rev. D 79, 039901 (2009)].arxiv.org/abs/astro-ph/0606058
- [61]Marc Klinger, Annika Rudolph, Xavier Rodrigues, Chengchao Yuan, Gaëtan Fichet de Clairfontaine, Anatoli Fedynitch, Walter Winter, Martin Pohl, and Shan Gao. AM³: An Open-source Tool for Time-dependent Lepto-hadronic Modeling of Astrophysical Sources. Astrophys. J. Suppl., 275(1):4, 2024.DOI
- [62]S. Hummer, M. Ruger, F. Spanier, and W. Winter. Simplified models for photohadronic interactions in cosmic accelerators. Astrophys. J., 721:630–652, 2010.DOI
- [63]Paolo Lipari, Maurizio Lusignoli, and Davide Meloni. Flavor Composition and Energy Spectrum of Astrophysical Neutrinos. Phys. Rev. D, 75:123005, 2007.DOI













