Neutrino flavor oscillations are a quantum-mechanical phenomenon that arises because the neutrino flavor eigenstates ν _e , ν _μ , ν _τ —which take part in weak interactions and are therefore the observable states—are not identical to the neutrino mass eigenstates ν ₁, ν ₂, ν ₃—which describe their propagation in vacuum. The flavor eigenstates are a quantum superposition of the mass eigenstates, whose relative phases change along the neutrino propagation path. This evolving mix of states leads to an oscillatory pattern of the detection probability of a given neutrino flavor, which depends on the neutrino energy and traveled distance.

Neutrino oscillation probabilities are calculated by solving the Schrödinger equation i∂ _t |ν(t)⟩ = H|ν(t)⟩. The neutrino propagation Hamiltonian H is approximated as the sum of two terms: H = U 0 0 0 0 Δ m 21 2 2 E 0 0 0 Δ m 31 2 2 E U † + V e 0 0 0 0 0 0 0 0 . (3)

The first term represents intrinsic energy levels of the system in vacuum. Because neutrinos are generally observed in an ultra-relativistic context, the energy difference between mass states is approximately proportional to their difference in mass-squared Δ m i j 2 = ( m i 2 − m j 2 ) . The Hamiltonian is shown in the flavor basis characterized by eigenstates of the weak interaction. The mixing matrix U represents the unitary transformation between the mass and flavor bases and is usually parametrized in terms of three mixing angles (θ ₁₂, θ ₁₃, θ ₂₃) and a complex phase δ, which are fundamental constants of physics [see, e.g., Giganti et al. (2018) for a general discussion of oscillation formulae].

The second term in Eq. 3 arises from coherent interactions of neutrinos with electrons in the medium in which they propagate (Wolfenstein, 1978; Mikheyev and Smirnov, 1985). The effective potential V e = ± 2 G F n e induces a change in the electron flavor energy level that is directly proportional to the electron number density n _e and the Fermi constant G _F , which characterizes the strength of the weak interaction. The sign of the matter potential is positive for neutrinos and negative for antineutrinos.

The solution of the Schrödinger equation for such a system in matter of constant density amounts to computing the eigensystem of the full Hamiltonian. While in vacuum the eigenvalues and eigenvectors are given by Δ m i j 2 / 2 E and the matrix U directly, the matter potential leads to effective eigenvalues and eigenvectors that depend on the electron density. This process can be interpreted as a modification of the fundamental parameters Δ m i j 2 and θ _ij induced by the neutrino environment.

For example, the effective mixing angle θ ̃ 13 and the effective mass-squared splitting Δ m ̃ 31 2 in matter are related to the corresponding parameters in vacuum via ¹ : Δ m ̃ 31 2 ≈ ξ ⋅ Δ m 31 2 , sin 2 2 θ ̃ 13 ≈ sin 2 2 θ 13 ξ 2 , (4) with a mapping parameter ξ = sin 2 2 θ 13 + cos ⁡ 2 θ 13 − 2 E V e Δ m 31 2 2 , (5)

At the energy and distance scales relevant to atmospheric neutrino oscillations, the probability of observing a transition between ν _e and ν _μ flavors is proportional to sin²2θ ₁₃, which is small in vacuum, as observed in oscillation measurements of reactor antineutrinos (Abe et al., 2012; Ahn et al., 2012; An et al., 2012) that are insensitive to the matter potential. However, the ν _e ↔ ν _μ transition probability may become large or even maximal when ξ ² → sin²2θ ₁₃, which translates into a resonance condition for the neutrino energy: E → ± Δ m 31 2 ⁡ cos ⁡ 2 θ 13 2 2 G F n e . (6)

Neutrinos traversing the Earth will experience resonant oscillations in the outer core for energies ∼3 GeV, and in the mantle for ∼7 GeV, with the exact resonant energy depending on the electron density of the material.

An almost isotropic neutrino flux is constantly produced by the interaction of cosmic rays with air molecules, mainly coming from the decays of charged mesons (π ^±’s and K ^±’s) and muons. At energies above 100 Megaelectron Volts (MeV) approximately, these atmospheric neutrinos become the dominant neutrino flux observed on Earth. It occurs noting here that such energies are well above the ones (∼ few MeVs) typically attained by geoneutrinos, the electron (anti-)neutrinos produced in nuclear decays of radioactive elements like ²³⁸U, ²³²Th and ⁴⁰K, which are responsible for the Earth’s radiogenic heat flux (Smirnov, 2019). While geoneutrinos have their own intrinsic interest for inferring information about the Earth’s composition, they can be safely ignored in the present study which focuses on neutrinos with energies of 1 Gigaelectron Volt (10³ MeV) and above.

Atmospheric neutrino flux calculations rely on 3D Monte-Carlo simulations of air shower development, starting with a primary cosmic ray flux based on measurements, and including solar modulation, geomagnetic field effects and seasonal variations of the atmosphere density. For the present study, we use as input the flux produced by Honda et al. (2015) for the Gran Sasso site (without mountain over the detector), averaged over all azimuth angles and over seasonal variations, and assuming minimum solar activity. The flux is dominated by muon- and electron-(anti)neutrino components, with an approximate ratio of 2:1 between ν _μ and ν _e flavors, and we have neglected the small contribution of ν τ / ν ̄ τ .

For a given zenith angle of incidence θ _z , as defined in Figure 1A, the neutrino trajectory across the Earth is modelled along the corresponding baseline through a sequence of steps of constant electron density which are inferred from a radial model of the Earth with 42 concentric shells of constant n _e , where mass density values are fixed and follow PREM. These shells are grouped into three petrological layers (inner core, outer core, and mantle+crust), whose composition, hence Z/A factor, is assumed to be uniform and provided in Table 1). In each shell, the electron density is determined from the mass density and Z/A according to Eq. 2. The probabilities of neutrino flavor transitions along their path through the Earth are computed using the OscProb ² package. The values of the parameters that enter the oscillation probability computation are taken from the global fit of neutrino data performed with the NuFIT analysis Version 5.0 (Esteban et al., 2020). We have assumed normal ordering of the neutrino mass states, i.e., m ₃ > m ₁, which is currently favoured by NuFIT and other global fits of neutrino data (Capozzi et al., 2021; de Salas et al., 2021).

Neutrinos with energies in the range ∼ 1 -100 GeV interact with matter mostly via scattering off nuclei, by exchanging either a neutral (Z ⁰) or a charged (W ^±) weak-force boson that triggers a hadronic cascade. In neutral-current (NC) interactions, the neutrino survives in the scattering products and escapes. In charged-current (CC) interactions, the neutrino gives rise to a charged lepton counterpart of the same flavor (e, μ, or τ). While electrons immediately induce a secondary short electromagnetic cascade, muons travel relatively long distances, depending on their energy, before they are stopped. Because of the high mass of the τ lepton (≈1.7 GeV), the contribution of (anti-)ν _τ -induced CC events is small in the range of energies relevant for tomography studies. For completeness, this flavor is nevertheless included in our computations.

The rate of neutrinos interacting in a given volume of target matter is then computed, for each interaction channel α (NC/CC; e, μ, τ; ν / ν ̄ ), as: R α int E , θ z ≡ d 4 N α int E , θ z d E d θ z d t d M = ∑ β = e , μ d 2 Φ ν β d E d θ z E , θ z ⋅ P ν β → ν α E , θ z ⋅ σ ν α int E m N , (7) where we have made explicit the dependence on the neutrino energy E and on the incoming zenith angle θ _z , as defined in Figure 1A, under the assumption of a spherically symmetric mass density profile of the Earth. The angle θ _z is directly related to the path length L through the Earth: L ≈ − 2R _⊕ cos θ _z with R _⊕ the radius of the Earth.

R α int ( E , θ z ) represents the differential rate of ν _α interactions (with target nucleons of mass m _N ) at the detector location, as a function the energy and zenith angle and per unit exposure (defined as the product of running time and target mass of the detector). It is obtained as a product of the incident (differential) fluxes of atmospheric neutrinos Φ _β (with β = e, μ as the atmospheric ν _τ component is negligible), the flavor oscillation probability along each neutrino path P ν β → ν α , and the neutrino-nucleon cross-section σ ν α int , that quantifies the probability for a neutrino to interact (hence to generate a potentially detectable signal). A quantitative example is shown in Figure 1B, illustrating both the oscillatory effects and the attenuation of the flux at high energies, which is due to the energy power spectrum ∝ E ⁻³ of the cosmic ray flux that produce the neutrinos in the atmosphere (Honda et al., 2015).

Neutrinos are observed only indirectly, through the signal deposited in the detector by the byproducts of their interaction. The ability to detect, reconstruct an identify the different types of neutrino events in the energy range of interest for Earth tomography is essentially driven by the detector size, technology and density of sensors. We consider here two main experimental approaches that are currently pursued for the upcoming generation of detectors targeting neutrinos with energies at the GeV scale, and that rely on different detection materials and observation strategies:

• Liquid Argon (LAr) detectors are a type of Time Projection Chamber (TPC) (Rubbia, 1977) that detects the electrons released upon ionization of Argon atoms along the paths of the secondary charged particles emerging from the neutrino interaction.

LAr TPCs are able to reconstruct highly detailed 3D images of the neutrino event, providing excellent flavor identification capabilities, even at low (sub-GeV) energies, and energy and angular resolutions typically superior to water-Cherenkov detectors. The size of these detectors is however limited by their much higher cost. The first large-scale LArTPC detector was the ICARUS T600 detector (Rubbia et al., 2011), with an active mass of 476 tons. A new generation of LArTPCs is being developed for the DUNE experiment (Abi et al., 2020c), that will include four detectors of 10 kton active mass each.

The performance of water-Cherenkov detectors, on the other hand, is typically a trade-off between the target volume for neutrino interactions and the density of photosensors that sample the Cherenkov signal emitted by charged byproducts of the neutrino interaction. The Hyper-Kamiokande experiment (Abe et al., 2018) will consist in two ∼ 200 -kton water tanks overlooked by a dense array of photosensors covering its internal walls, providing a sub-GeV threshold for neutrino detection and excellent discrimination between electron-like and muon-like signatures. To access even larger target volumes, the ORCA (Adrian-Martinez et al., 2016; Aiello et al., 2022) and PINGU (Aartsen et al., 2017) experiments propose to instrument several Mtons of seawater or polar ice with a much sparser tridimensional array of photosensors. The gain in statistics comes at the expense of a higher detection threshold (few GeV) and worse energy and direction reconstruction capabilities, which also limit their performances for neutrino flavor identification. Water-Cherenkov detectors also cannot distinguish in first approximation between neutrino and anti-neutrino events.

To simulate the detector response, we use a set of parametrized analytical functions modeling the main performance features relevant for the detection, reconstruction and classification of neutrino events:

• The effective mass M _eff ( E) is the product of the instrumented target mass M of the detector and its detection efficiency, i.e., the probability for a neutrino interaction to be successfully detected as an event. M _eff typically increases with the neutrino energy, until reaching a plateau that saturates the fiducial mass of the detector. We have conservatively neglected here a potential increase in M _eff for ν _μ CC events at high energies, corresponding to through-going muons created in neutrino interactions outside of the detector target volume. The threshold for detection is mainly driven by the density (and intrinsic efficiency) of sensors. We approximate M _eff (E) by a sigmoid function of log(E) with two adjustable parameters E_th and E _pl , which correspond to energies where the detection efficiency reaches 5% and 95% respectively.

Events are classified according to their topological features, which depend on the signature of each interaction channel (NC/CC; e / μ / τ ; ν / ν ̄ ). In practice, the relative sparseness of instrumentation limits the signal sampling in the detector, and thus the reconstruction and identification performances.

In this study we assume that all detectors under consideration have the basic capability to classify events into two main observational classes: track-like events, which are mostly associated to ν _μ charged-current interactions producing a long muon track, and cascade-like events, when there is no such single muon or the muon energy is too small for it to be distinguished among the many other secondary particles created in the event. The cascade channel therefore comprises mainly ν _e CC and (subdominantly) ν _τ CC interactions, plus a small contribution from ν _μ CC interactions without an observable muon, and all NC interactions.

For NC interactions, a fraction of the energy of the event is carried away by an invisible outgoing neutrino, resulting in a lower reconstructed energy and degraded angular resolution. Because NC-induced events are blind to neutrino flavor, they only decrease the experiment sensitivity. Fine-grained detectors like DUNE and Hyper-Kamiokande have some capability to separate and reject the NC-induced events from the ν _e CC events, thereby increasing the flavor purity of the cascade channel. To account for a possible related improvement in the detector sensitivity, we have considered here both the baseline case that includes the NC-induced events in the cascade sample, and the optimistic case where the NC-induced contributions are completely removed from the sample. We have also used throughout our study the neutrino-nucleon cross-sections weighted for water molecules obtained with the GENIE Monte Carlo neutrino generator (Andreopoulos et al., 2010), neglecting the small difference in cross-section for interactions on Argon nuclei in the case of the DUNE detector.

In order to obtain the expected signal in a realistic experiment, the rate of interacting neutrinos as obtained from Eq. 7 must be weighted with the appropriate detection efficiency at each incoming neutrino energy. In the following we use the subscript “true” for the quantities (energy and zenith angle) that refer to the incoming neutrino properties, and “reco” for their reconstructed values as provided by a given detector. The observed event rate is computed by a discretized convolution of the interacting rate at energy E _true and incident at zenith angle θ _true over the E-cos θ plane according to the energy and zenith resolution p.d.f.s. The reconstructed events for each interaction channel are distributed into the two observational channels (tracks and cascades) according to the classification efficiency function ɛ _class(E). Every bin in the final (E _reco, θ _reco) event oscillogram therefore (i) contains a certain fraction of misreconstructed events coming from other (E _true, θ _true) bins; and (ii) misses some events that end up misreconstructed into different (E _reco, θ _reco) bins.

The final event rate expected in a given channel, in a given bin of reconstructed energy and zenith angle, is thus obtained as follows for the track channel: R tracks obs E reco , θ reco = ∑ E true , θ true R tracks int E true , θ true ε class E true + R casc int E true , θ true 1 − ε class E true × P D F angle θ reco ; E true , θ true × P D F energy E reco ; E true × Δ E true × Δ θ true × M eff E true , (8) where PDF _angle and PDF _energy represent the probability distribution functions of the angular and energy resolutions as defined above. A similar expression can be straightforwardly derived for the event rate in the cascade channel.

Independently of any classical measurement error, the intrinsically probabilistic nature of neutrino interactions as a quantum process induces a statistical uncertainty on the number of events observed in each bin of (E _reco, θ _reco). Our statistical analysis is based on the standard Asimov dataset approach, in which all observed quantities are set equal to their expected values. The probability to detect N neutrino events of a given topology, e.g., tracks, in a predetermined interval of reconstructed energy and zenith angle is distributed according to a Poisson law, parametrized by the expected number of events N exp = R tracks obs ( E reco , θ reco ) × T , where T corresponds to the duration of data taking. The observed number N will randomly deviate from this expectation with a relative standard deviation σ ( N ) / N exp = 1 / N exp . This uncertainty on the final measurement can only be mitigated by increasing the size of the event sample, which is proportional to the detector exposure.

Once these fluctuations are taken into account, we compute the log-likelihood ratio λ _LLR of the resulting event histograms, which yields a measure of the significance to reject model B (the “test hypothesis”) if model A (the “null hypothesis”) is true. According to Wilk’s theorem, for large enough statistics, the Poisson-distributed events converge towards a normal distribution, so that −2λ _LLR can be approximated by: − 2 λ LLR ≈ Δ χ i 2 = n A i − n B i 2 n A i . (9)

The total Δχ ² (integrated over all bins of energy and zenith angle) is related to the significance σ = Δ χ 2 , that gives the confidence level (C.L.) by which the test hypothesis can be excluded under the assumption of the null hypothesis. We have checked that the main conditions for Wilk’s theorem to apply were satisfied in our case: namely, that the simulated statistical sample was large enough, and that the sampled interval of Z/A was sufficiently far away from its boundary values (0 and 1).

Based on the simulation chain described in Section 2, our first step is to evaluate the intrinsic sensitivity of the method by conducting detailed computations of the propagation of neutrinos through the Earth and comparing the expected rate of interacting neutrino events in a perfect detector under different assumptions for the outer core composition.

From Eq. 7, one can generate the two-dimensional (2D) distributions of expected number of neutrino interactions of a given flavor as a function of (E, θ _z ), or oscillograms, for any specific exposure of a perfect detector that would observe atmospheric neutrinos from all directions. Because of the dependence of neutrino oscillation on n _e along the neutrino path, changes in the outer core composition (i.e., different Z/A) induce variations in the expected signal, depending on the neutrino energy and exact trajectory (or equivalently, θ _z ), that get imprinted in the oscillograms.

We have quantified this effect by computing the relative difference (ΔN/N) in expected number of neutrino events between oscillograms generated with different models of outer core composition. We considered five different models (see Table 2), ranging from a pure FeNi alloy to a Hydrogen-rich (FeNiH) one, with two intermediate models chosen from the FeNiSi _x O _y family. As mentioned in the introduction, silicon and oxygen are not the only light elements that have been proposed in conventional models of the outer core; also sulphur and carbon could be present (Poirier, 1994), but their Z/A is too close to that of Si and O to allow for a discrimination by means of this method. Furthemore, high-pressure/high-temperature thermodynamics modelling appear to rule out Fe-Si-C compositions (Huang et al., 2022). Therefore we do not consider them further in this study.

Figure 2 illustrates the expected impact of the composition in terms of ΔN/N for muon- and electron-neutrino interactions, as a function of the neutrino energy, for two different incoming directions. The predicted numbers of interactions for those particular configurations are shown to differ by up to 4% for electron neutrinos, and up to 17% for muon neutrinos, depending on the core composition considered. Furthermore, because the shape of the signal changes a lot depending on the zenith angle, it appears important to consider the signal in two dimensions (i.e., as a function of E and θ _z ). This is further illustrated in Figure 3 showing the full 2D distributions of ΔN/N for muon- and electron-neutrino interactions, for a specific comparison between FeNiH and FeNi core compositions. Such results confirm that a detailed study of the neutrino event rate in the energy range 1–10 GeV has the potential to constrain the Earth’s outer core composition.

The distributions shown in the upper panels of Figure 3 implicitly relate to a detector with 100% detection efficiency, perfect energy and θ _z resolution, and infinite statistics. In reality, the measurement accuracy will be limited by experimental effects, resulting in some blurring and attenuation of the signal, as illustrated in Figure 3 (lower panels). As discussed in Section 2.4, the detector technology and specifications (size, detection efficiency, resolution and particle identification) affect its ability to reconstruct the neutrino properties (energy, direction and flavor), and the intrinsically probabilistic nature of neutrino interactions induces a statistical uncertainty on the final observed number of events. The detector thus needs to be scalable to a sufficiently large volume for this uncertainty to be smaller than the intrinsic signal.

To investigate the influence of specific detector characteristics on their potential to constrain the core composition, we use the parametrized functions defined in Section 2.4 to model the key experimental features - namely, the effective mass, the direction and energy resolutions, and the efficiency in identifying the neutrino flavor based on the classification of events into track and cascade topologies. We consider here four benchmark parametrizations with specific inputs chosen as educated guesses for different detectors representative of the upcoming generation of GeV neutrino detectors: Hyper-Kamiokande, ORCA, and DUNE. The detail of the parameters used for each detector is provided in Table 3.

The DUNE-like model (Abi et al., 2020a; 2020c; Kelly et al., 2022) accounts for the superior resolution and reconstruction capabilities of the Liquid Argon detection technique, at the cost of a much smaller instrumented volume. At the other extreme, the ORCA-like model (Adrian-Martinez et al., 2016; Aiello et al., 2022) reflects the higher detection threshold and degraded reconstruction performances that result from a sparsely instrumented - although larger - volume. The Hyper-Kamiokande-like model (Abe et al., 2018) is a middle-way option combining a medium-size detector with a low detection threshold and good reconstruction and identification capabilities, in line with those already achieved with its predecessor Super-Kamiokande (Suzuki, 2019). Although some of the detectors considered here may be sensitive to neutrinos below 1 GeV, we have conservatively limited our study to neutrinos with energy in the range [1,40] GeV, which encompasses most of the expected oscillation tomography signal. Some examples of the corresponding response functions are presented in Figure 4 for the energy range under consideration.

The corresponding modelling, and the physical parameters used for each specific detector, are described in Section 2.4 and Table 3. By folding the detector response with the expected rate of neutrino interactions given by Eq. 7, and summing over all relevant interaction types, we obtain the predicted rate of observable events of a given observational class, R track obs ( E reco , θ reco ) or R cascade obs ( E reco , θ reco ) in a given bin of neutrino reconstructed energy and zenith angle. By integrating these quantities over time and effective mass of the detector, we generate 2D histograms of the expected number of events as a function of E _reco and θ _reco, for a given detector exposure. Such oscillograms are the cornerstone of our discussion of the detectors capability to constrain the Earth’s core composition, as illustrated in the lower panels of Figure 3.

For a given detector, our projected experimental sensitivities are obtained from the detailed comparison of the full 2D oscillograms of expected neutrino events, generated with different assumptions for the core composition. As described in Section 2.5, a statistically meaningful way of quantifying the detector performance is to apply a χ ² hypothesis test to the 2D histograms of expected events in bins of (E _reco, θ _reco). Under the assumption of the validity of Wilk’s theorem, the total Δχ ² associated to a given pair of composition models, say A and B, is directly related to the significance σ = Δ χ 2 , giving the confidence level (C.L.) by which model A can be discriminated from model B. Examples of signed Δχ ² maps for different upcoming detectors are presented in Figure 5 for the discrimination between FeNi and FeNiH models. The three detectors achieve a comparable statistical significance in the measurement of the outer core Z/A after 20 years of data taking, although their sensitivity does not necessarily come from the same region of the (E, θ) plane, nor from the same observational channel.

Figure 6 compares more directly the performances of the benchmark detectors in discriminating outer core composition models. Despite the differences in technology, size and reconstruction performances, we find that ORCA and DUNE reach a similar precision of 0.016 (at 1σ C.L.) on the absolute Z/A measurement after 20 years of data taking. These results are comparable with the ones previously published (Bourret and Van Elewyck, 2019) based on a full simulation of the ORCA detector. They confirm the capability of these detectors to measure the Z/A in the Earth’s outer core with a 1σ relative precision of 3–4 percent. Hyper-Kamiokande performs slightly better, with a relative precision of 2.5%, yielding a discrimination power of approximately 0.5σ between FeNi and FeNiH models for 25 years of measurements. A higher sensitivity can be achieved by combining the data from the three detectors into one single measurement; in that case, a 1σ C.L. discrimination power for FeNiH vs. FeNi could be reached in 50 years of concomitant data taking, as can be seen in Figure 7.

While the previous result can be seen as a promising proof of concept of the method, achieving the next level of sensitivity requires both beating the statistical limitation by scaling up the detectors in size, and achieving excellent reconstruction performances (flavor, E _reco, θ _reco), in order to better resolve the patterns in the 2D (E _reco, θ _reco) event distributions.

Based on the results obtained for the upcoming detectors, we have investigated possible realizations of such a next-generation, or NextGen, detector, that would combine the best performances of current-generation instruments, taking as benchmarks a 1 GeV detection threshold and almost perfect (99%) discrimination between track-like and cascade-like events.

We performed a systematic scan of the discrimination potential of such detector over a large range of sizes, energy and angular resolutions, as presented in Figure 8. A separation power of at least 1σ between FeNiSi₂O₄ and FeNiH models can be obtained with a rather wide combination of zenith/energy resolution parameters. A good compromise is found in the region around σ(θ) = 7° angular resolution and σ(E)/E = 10% energy resolution, not far from the performances associated with Hyper-Kamiokande, but for a significantly larger, 10-Mton, detector. We provide a specific parametrization of such a NextGen detector in Table 3.

This scan also illustrates how the small size and limited scalability of Liquid Argon detectors is a serious obstacle to developing this technique to achieve the required exposure levels in a reasonable running time, despite their superior energy and zenith resolution. Even an asymptotic evolution of the DUNE experiment, with perfect identification of all neutrino flavors and interaction channels, would only reach a 0.5σ discrimination power for FeNiSi₂O₄ vs. FeNiH after 50 years running. We conclude that better perspectives for a core composition measurement arise from the water Cherenkov approach, provided that event reconstruction and identification performances similar or better to those of Hyper-Kamiokande can be achieved in a much larger (hence likely more sparsely instrumented) detector.

The performance of the NextGen detector is illustrated in Figure 9 in terms of the signed Δχ ² maps for the discrimination between a FeNiH and a FeNiSi₂O₄ composition of the outer core, for the track and cascade channels, for 20 years of data taking. In this example, the total Δχ ² value is 2.71, respectively 1.68 for track and 1.03 for cascade channel.

The qualitative jump in performance that could be achieved with the NextGen detector is also evident in Figure 7 where its sensitivity is compared to the one obtained from combining ORCA, DUNE and Hyper-Kamiokande data. The characteristics of the NextGen detector yield a sub-percent precision on the Z/A measurement, sufficient to distinguish between FeNiSi₂O₄ and FeNiH compositions at > 1 σ in a running time of about 10 years, well within the typical lifetime of a neutrino experiment (a discrimination at 90% C.L. would be achieved in about 20 years).

We further illustrate such capabilities in Figure 10, where we address the separability (at the 1σ level) between pairs of realistic core composition models, as a function of the detector running time. Although a discrimination between composition models whose Z/A are closer than 0.001 (i.e., FeNiSi₇O₂ vs. FeNiSi₂O₄, or FeNiSi₂O₄ vs. FeNiSiH) appears out of reach, our results suggest that the NextGen detector would be able to exclude or confirm the FeNiH hypothesis against all the other realistic models considered in this study. This result can be obtained in less than 20 years if the actual outer core composition is in the family of models with an admixture of Silicium and Oxygen as only light elements.