Main

Observing the chemical composition of supernovae (SNe) and supernova remnants (SNRs) provides direct evidence of how elements are synthesized and distributed in the Universe. The elemental abundance patterns of these objects or stars reflect the physical processes occurring in stellar interiors and explosions1,2,3, offering crucial insights into the origin and evolution of elements. While even-Z elements (that is, α-elements) have been extensively studied in SNe and SNRs, observations of odd-Z elements, such as phosphorus (P), chlorine (Cl) and potassium (K), remain scarce, and their theoretical treatment is incomplete4,5. Notably, galactic chemical evolution models struggle to explain the observed abundances of these elements6,7,8,9, and discrepancies with theoretical predictions have also been reported in observations of some metal-poor stars10. These inconsistencies highlight a fundamental gap in our understanding of nucleosynthesis processes. Currently, stellar processes such as rotation, binary interactions and shell mergers (the merging of adjacent nuclear-burning shells into a single convective shell) have been proposed as plausible mechanisms to resolve this issue9,11,12. Here, the nucleosynthesis pathways for odd-Z element production, which include γ-reactions and neutron/proton capture, are complex11, and it is necessary to directly discuss the synthesis yields of these elements on the basis of observations. However, because emission lines from odd-Z elements are faint, it has been difficult to observe them in SNe/SNRs with existing detectors (particularly in X-rays), and their origin therefore remains unresolved.

Recent near-infrared spectroscopic studies of Cassiopeia A, the youngest known core-collapse SNR in our Galaxy, have reported the detection of emission from P (refs. 4,13). In particular, its [P/Fe] ratio is up to 100 times higher than the average in the Milky Way4, prompting a discussion on the nucleosynthesis processes during the SN explosion. However, the observed [P/Fe] ratios were scattered by about two orders of magnitude in the remnant because the two elements are synthesized at different locations and unevenly mixed in a SN, preventing a clear understanding of their origin. In addition, since most of the SN ejecta have already been heated by the reverse shock (≲20% of ejecta are unshocked)14, X-ray emission from the shocked ejecta most accurately reflects the true elemental composition in the remnant. To fully understand the origin of odd-Z elements, it is thus crucial to systematically measure the abundances of several odd-Z elements, including Cl and K, together with neighbouring even-Z elements (for example, Si, S, Ar and Ca) produced at the same site using X-ray spectroscopy. Therefore, high-resolution non-dispersive X-ray spectroscopy, which has the potential to observe all these elements, is a unique tool to probe the nucleosynthesis of odd-Z elements produced in SNe.

The X-Ray Imaging and Spectroscopy Mission (XRISM)15, successfully launched on 7 September 2023, is equipped with the Resolve microcalorimeter, capable of non-dispersive high-resolution spectroscopy (ΔE < 7 eV, where ΔE denotes the energy resolution) in the X-ray band (1.7–10 keV), where primary emission lines from ionized elements P, Cl and K are emitted. We conducted observations of Cas A with XRISM twice in mid-December 2023 (Fig. 1, Table 1 and ‘Observation and data reduction’ section in the Methods). These two pointings cover the majority of the remnant’s SN ejecta, with the exception of the north-east-extending jet-like structure, enabling a comprehensive investigation of odd-Z elements in this remnant. Notably, large clumps of ejecta are concentrated in the south-east, north and west regions, reflecting the explosion’s asymmetry16. As shown in the Chandra elemental maps (Fig. 1), the south-east and north blobs contain O-rich ejecta (most probably the product of hydrostatic nuclear fusion in the progenitor star’s deep interior) along with Fe- and Si-rich ejecta (the products of the explosive nucleosynthesis). In contrast, the west region is dominated by Fe and Si. We employ spatially resolved spectroscopy with XRISM/Resolve to examine and compare the elemental abundances across these distinct regions.

Fig. 1: Three-colour image of the core-collapse SNR Cas A.
figure 1

Red, green and blue represent Chandra X-ray images of O- (0.6–0.85 keV), Si- (1.76–1.94 keV) and Fe-enhanced (6.54–6.92 keV) regions, respectively. The white grids indicate the fields of view of our two-pointing observations with XRISM/Resolve, which has a 6 × 6 pixel array: a pixel at a corner is only used for calibration. Pixels we selected for our spectral analysis are highlighted in yellow. In the south-east (SE) and north (N) regions, we examined the spectrum of each pixel and selected those exhibiting large equivalent widths of K emission; the spatial distribution of these pixels closely matches that of the O and Si emission lines. In contrast, the west (W) region exhibits weak O line intensity, and the K emission line in its spectrum is faint (Fig. 2).

Full size image
Table 1 Observation log of Cas A with XRISM
Full size table

Figure 2 shows the X-ray spectra of the north (N), south-eastern (SE) and western (W) regions of Cas A. In the SE and N regions, Resolve clearly detected the emission from highly ionized He-like ions of Cl and K (at ≳5σ and >6σ confidence levels, respectively). In contrast, the W region shows no clear line structures and only marginal Cl- and K-line detection significances (<2σ). Marginal evidence of P is also detected in the SE and W regions (≳4σ confidence level; Table 2 and Extended Data Table 1). These results reveal a spatially inhomogeneous distribution of odd-Z elements in Cas A, with their emission confined to the O-rich regions in the north and south-east but absent in the western region. These detections are unaffected by background emission, spatial–spectral mixing from the telescope’s point spread function, fitting range choices or other factors (Methods and Extended Data Table 1). The abundance ratios relative to hydrogen are approximately 3–5 solar values, confirming that these elements originate from the SN ejecta rather than the interstellar medium. Dielectronic recombination and inner-shell excitation lines from other elements lie close to the P and Cl line features, with the former being almost completely overlapped and the latter located immediately adjacent, making accurate abundance estimates dependent on careful plasma modelling. In the K band, there are no prominent contaminating lines, allowing for a clean separation of the emission feature. The N region shows complex velocity structures in intermediate-mass elements (see fig. 4 in ref. 17), making abundance measurements there less reliable (see the Methods for detail). We therefore adopt the K/Ar and Cl/S ratios from the SE region as representative values, with the K/Ar ratio being the most reliable owing to its clear detection and minimal contamination, and the Cl/S ratio also being considered reasonably robust. The Cl/S ratio is estimated to be (Cl/S)/(Cl/S)⊙ = 1.0 ± 0.1 and the K/Ar ratio (K/Ar)/(K/Ar)⊙ = 1.3 ± 0.2, on the basis of proto-solar values from ref. 18. Here, we use the abundance ratios relative to neighbouring even-Z elements to examine the characteristics of the synthesis processes in the subsequent discussion. Across a variety of fitting conditions, the derived abundance ratios remain consistent within a range of ~10–40%, which probably reflects the level of systematic uncertainty in our measurements and does not affect our conclusions (see ‘Spectral analysis’ section in the Methods and Extended Data Table 1).

Fig. 2: Resolve X-ray spectra of Cas A.
figure 2

The best-fit models are overplotted as a solid line, with colours representing the spectral regions in Fig. 1 (orange, N region; red, SE region; green, W region). In the top panel, the labels indicate the positions of the He α emission lines for each element. The vertical broken lines correspond to the centroid energies of the resonance lines of P (~2.152 keV), Cl (~2.790 keV) and K (~3.511 keV) in the rest frame. The bottom three panels show the zoom-in spectra around He α P (left), Cl (middle) and K (right) line with the best fit models with or without the K emission line. In the spectra of the P band, models without P emission show residuals near 2.13 keV, particularly in the SE and W regions, indicating the signature of P emission in the spectra, while no significant residual is seen in the N region. In the Cl and K bands, models lacking the corresponding lines show residuals in the N and SE regions but not in the W region. The intermediate-mass elements, particularly in the N region, exhibit complex velocity structures (see fig. 4 in ref. 17), which introduces modelling uncertainties. For clarity, the spectrum of the SE region is shifted slightly along the energy axis in the zoomed-in panels. The horizontal and vertical error bars indicate the selected energy range for each bin and the 1σ statistical uncertainty, respectively.

Full size image
Table 2 Best-fit parameters of spectral fits
Full size table

The observed ratios close to the solar value suggest abundant production of Cl and K in the remnant, in contrast to the scarcity in the theoretical models8. Figure 3 shows the elemental mass ratios from Si to Ca (relative to Ar) obtained from our observation of the SE region. It is noteworthy that the observed K and Cl abundances in Cas A, whose initial progenitor mass is estimated to be close to 15 M⊙ (ref. 19), are significantly higher than those in SN models within the initial mass range of 13–25M⊙ from ref. 20 (hereafter NKT13), whereas the observed abundances of the even-Z elements roughly follow the trend of the models (see Extended Data Fig. 1 for a more detailed comparison). This implies that some mechanisms not included in the models enhanced the abundance of the odd-Z elements in Cas A. The fact that these Cl and K emission are spatially associated with the O-rich ejecta suggests that their production was amplified during the stellar evolution stage rather than being solely the result of explosive nucleosynthesis. Regarding the P abundance, both observational uncertainties and large variations among theoretical models make quantitative discussion difficult at this stage (Extended Data Table 1), even though the measured value falls within the range of theoretical predictions in Fig. 3.

Fig. 3: Elemental mass ratios in Cas A compared with those in SN nucleosynthesis models.
figure 3

Red circles with error bars represent the mass ratios relative to Ar derived from different regions shown in Fig. 1. The error bar shows a 1σ statistical uncertainty. The grey region indicates the range predicted by core-collapse SN nucleosynthesis models in ref. 20, assuming single-star, non-rotating progenitors with solar metallicity and initial masses in the range of 13–25M⊙. To facilitate comparison between the observed values and the model values, only the model values are connected and that range is shaded. See Extended Data Fig. 2 for a more comprehensive comparison across a wider range of model parameters.

Full size image

Some stellar activities have been proposed to explain the enhanced production of odd-Z elements in massive stars, most notably stellar rotation9, binary interactions12 and the shell merger phenomenon11. To investigate these effects, we compare the observed Cl/S and K/Ar ratios with predictions from several core-collapse SN nucleosynthesis models in Fig. 4: non-rotating single-star models from NKT13, rotating models from ref. 21, shell merger candidates from ref. 22 and binary-star models from ref. 12 (see Extended Data Fig. 2 for a more detailed comparison). The models with the stellar effects tend to have higher Cl/S and K/Ar ratios than those of the NKT13 models, implying that these processes are required to explain our observations. In particular, some of the shell merger and binary models predict ratios exceeding those measured in Cas A, indicating an efficient odd-Z production mechanism in these scenarios. However, we note that these models involve large uncertainties. In the case of shell merger models, the predicted yields are highly sensitive to the multi-dimensional mixing in the burning layer11. Furthermore, Farmer et al.12 point out that, in their binary models, the convective structure near the nucleosynthesis region may introduce substantial uncertainties in the production of these elements. Since our work presents the observational constraint on Cl and K abundances in a core-collapse SN, further theoretical development will be crucial to elucidating their production mechanisms. Detailed comparison and extended discussion of our model analysis are provided in ‘Model dependence on odd-Z yields’ section in the Methods. These interpretations are constrained by current model uncertainties, highlighting the need for future efforts to fully integrate observational and theoretical constraints across diverse progenitor scenarios, as well as for population studies combined with theoretical modelling to determine the prevalence of candidate mechanisms among massive star progenitors.

Fig. 4: Observed Cl/S and K/Ar ratios in Cas A compared with representative nucleosynthesis models.
figure 4

The abundance ratio is defined as the elemental abundance ratio normalized to the solar value: (K/Ar)/(K/Ar)⊙ and (Cl/S)/(Cl/S)⊙. Red regions indicate the 1σ statistical confidence intervals of observed abundance ratios in the SE region. Model predictions are shown for non-rotating single-star progenitors with initial masses of 13–25M⊙ from ref. 20 (grey crosses), rotating models from ref. 21 with initial masses of 13–25M⊙ and rotational velocities of 300 km s−1 (green squares), shell merger candidates from a subset of the models in ref. 22 with progenitor masses of ~19–27M⊙ (red circles) and binary star models from ref. 12 with initial masses of 13–25M⊙ and companion stars with 80% of the primary mass (blue diamonds). All models shown assume solar metallicity. See Extended Data Fig. 2 for a more comprehensive comparison across a wider range of model parameters.

Full size image

The observational characteristics of the Cas A remnant are generally more consistent with a progenitor that would support these stellar processes than the standard stellar nucleosynthesis. As the remnant of a type IIb SN, Cas A probably lost most of its hydrogen envelope through binary interaction23. Its extremely low neon (Ne) abundance (Ne/O ≈ 0.1 solar)24,25,26, in contrast to the typical Ne/O ≈ 1 solar observed in other SNRs27,28, further suggests unusual stellar processes that deplete Ne. Stellar phenomena such as rotation and shell mergers have been proposed to account for this Ne scarcity in the O-rich layers of the progenitor star26,29,30, implying a possible connection between Ne depletion and the enrichment of odd-Z elements during the stellar evolution of Cas A. In addition, our observation reveals a positive correlation between O emission intensity and odd-Z element abundances, further supporting a stellar nucleosynthetic origin. Although the odd-Z enhancement mechanism remains debated11,12,21, this observational link between stellar activity and odd-Z element synthesis provides a unique benchmark. Interestingly, both the O-rich ejecta and odd-Z elements display a biased distribution from south-east to north in the remnant, which may suggest the impact of asymmetric convection in the stellar interior on SN dynamics31,32. Future observations of the odd-Z elements in Cas A and other remnants, using XRISM or upcoming X-ray observatories, will provide deeper insights into pre-SN activities and may hold the key to unveiling the chemical history of our Galaxy.

Methods

Observation and data reduction

We carried out a two-pointing observation of Cassiopeia A (Cas A) during the XRISM commissioning phase as summarized in Table 1 refs. 17,33,34. The Resolve fields of view are shown in Fig. 1. Since the Resolve aperture door (‘gate valve’) has not opened yet, a 250-μm-thickness beryllium filter attenuates X-ray photons in the soft X-ray band, reducing the effective area. The available energy bandpass in our analysis is above approximately 1.6 keV. The data reduction was done by using calibration data archived in the High Energy Astrophysics Science Archive Research Center (HEASARC) calibration database (CALDB). Cleaned event data were obtained using the latest release version of the HEASARC software (version 6.34) with the standard screening during post-pipeline processing. We only used the highest energy resolution (‘Hp’) primary events for the spectral analysis. The redistribution matrix file was generated with the extra large size option using ‘rslmkrmf’. The ancillary response file was created using ‘xaarfgen’, assuming the surface brightness of Cas A as derived from a Chandra X-ray image in the 2.0–8.0-keV band.

Energy scale reconstruction

The gain of the XRISM/Resolve calorimeter detectors is intrinsically unstable since the thermal detectors respond to their thermal environment. The XRISM observatory includes multiple on-board calibration sources to track the time-dependent gain of the Resolve instrument and reconstruct the energy scale for each observation. The Cas A observations were performed early in the performance and verification phase of the mission while the gain reconstruction strategy was still being formalized. Currently, gain fiducials are acquired sparsely, approximately twice per day on average, to sample the slow drift of the detector gain due to changes in spacecraft attitude and the thermal recovery from recycling the sub-kelvin adiabatic demagnetization refrigerator. However, the Cas A observations were performed with much more frequent sampling of the gain: once every other orbit, or approximately every 3 h, during Earth occultation. One position of the instrument’s filter wheel contains a set of 55Fe radioactive sources providing a fiducial x-ray line at 5.9 keV. During Earth occultation, the filter wheel sources were rotated into the aperture, providing 30-min exposures with 500 counts in the 5.9-keV line per pixel per fiducial measurement. The fiducial measurements were then used to reconstruct the energy scale at that time step using the nonlinear reconstruction method described in ref. 35. The energy scale was linearly interpolated between fiducial steps, which has been shown to be sufficient to reconstruct the energy scale to better than 0.3 eV at 5.9 keV (ref. 36), especially for the frequently sampled fiducials during this early observation. The summed, reconstructed fiducial measurements for the Cas A observations give a line shift error of <0.04 eV at 5.9 keV for the entire detector array. More detail can be found in appendix 1 of refs. 37.

Spectral Analysis

As shown in Fig. 1, we divided the field of view into three regions to investigate the spatial variation of odd-Z elements in the spectral analysis described below. For the SE region, we checked the spectrum of each pixel and selected pixels with a large equivalent width of K emissions (pixel IDs 17, 18, 25, 32, 33 and 34), where the distribution of the selected pixels is in good agreement with that of the O emissions. Each spectral fit was performed using SPEX software38 version 3.08.01 and Xspec software39 version 12.14.1 (AtomDB 3.1.0.v7) with the maximum-likelihood C statistic40: we analysed the Resolve spectrum in the 1.6–5.0-keV band, where any background component is negligible (but, of course, we considered the non-X-ray background emissions in our analysis, as described below). The spectral fit used in the main text is the result with Xspec. To evaluate the effect of the non-X-ray background (NXB), we use a temporal NXB spectral model provided by the XRISM calibration team. The NXB model was constructed from a stacked NXB event file based on night-Earth observations with a total exposure of 785 ks, and has a power law and 17 Gaussian lines for Al Kα1/Kα2, Au Mα1, Cr Kα1/Kα2, Mn Kα1/Kα2, Fe Kα1/Kα2, Ni Kα1/Kα2, Cu Kα1/Kα2, Au Lα1/Lα2 and Au Lβ1/Lβ2. In Fig. 2, the NXB contribution is taken into account in the modelling but is below the displayed range. The NXB model we used has no line features around the K He-like Kα emission. We have confirmed that the K/Ar ratios agree within the statistical error (only a few per cent change), whether or not the NXB model is considered.

On the basis of previous observations25, the spectra were fitted with a two-component non-equilibrium ionization (NEI) model absorbed by the interstellar medium with the solar abundances18. For the ejecta component, we use a plane-parallel shock (pshock) plasma model (in the case of SPEX, ‘Mode=3’ in the neij model was used). The pshock and single NEI models give consistent results with respect to K detection, although the former gives a slightly better fit (Extended Data Table 1). We allowed the volume emission measure VEMNEI (= nenpVNEI, where ne, np and VNEI are the electron density, proton density and the volume of the plasma, respectively), the electron temperature kTe and the ionization timescale net to vary. The abundances of Si, P, S, Cl, Ar, K and Ca were left free and tied between the two NEI components, while the others were fixed at the solar values. The absorption column density NH was fixed at 1.9 × 1022 cm−2, 2.0 × 1022 cm−2 and 1.5 × 1022 cm−2 for the SE, W and N region, respectively, based on previous observation25. Supplementary Fig. 1 shows the spectrum of the SE region along with the best-fit model consisting of two plasma components. These components, having different electron temperatures and line-of-sight velocities, both contribute to the observed line structures, complicating abundance measurements. For example, the N region exhibits a complex velocity structure, with redshifts varying across elements and ionization states17, which makes it difficult to accurately determine the abundances of Cl and K, even though their spectral features are clearly present. In contrast, the SE region shows a simpler velocity structure and narrower line widths compared with the N region (Fig. 2), allowing for more reliable abundance measurements. We confirmed that additional components such as non-thermal emission or another NEI model do not substantially affect the results for the SE region. On the other hand, in the N and W regions, if non-thermal emission is not considered, the temperature of the high-temperature plasma component exceeds 5 keV, which is higher than in other regions. Therefore, we added a power-law (non-thermal) component for the spectral fitting of the N and W regions. In these regions, previous studies have reported that non-thermal emission is strong41,42,43,44, supporting the validity of adding the power-law component in this study. All the results are summarized in Table 2, showing that the two-component NEI model effectively explains all the Resolve spectra with statistically acceptable fits.

Several factors, such as differences in region selections, fit ranges and atomic codes, can introduce errors in abundance estimation. Here, we discuss systematic errors in abundance estimation (in particular, the K/Ar ratio, which is the focus of this study) by performing spectral analysis under various conditions. The XRISM calibration team does not recommend the asymmetric region selection used here because of the difficulty in evaluating contamination from other regions. For example, it is recommended to select a symmetrical region with a collection of 2 × 2 pixels. Therefore, we also analysed the 2 × 2 pixel region at the centre of the observation in the East region (pixel IDs 0, 17, 18 and 35), where the odd-Z elements seem to be concentrated. Even in this region, we found the K/Ar ratio to be 1.4 ± 0.2 with Xspec, which agrees with the result in region A. Here, the detection of K is 7.7σ. On the other hand, we emphasize that the effect of spatial–spectral mixing due to the point spread function of the telescope does not change our conclusion. In particular, in the eastern region used in the main text, the observed value of K/Ar ≈ 1 solar does not change even when using the Resolve full-array data. Also, whether we assume a single NEI plasma or a multi-temperature plasma, the result remains the same. Therefore, slight differences in region selection or plasma parameters do not change our conclusion. In a 2 × 2 pixel region, we can estimate with the raytracing code ‘xrtraytrace’ that ~50–60% of the radiation comes from within this region. Even though there is such a large amount of contamination from the neighbouring regions, it would be difficult to change the plasma parameters to overturn the conclusion.

In Extended Data Table 1, we summarize the K/Ar (also P/Si and Cl/S) ratios under different fitting conditions. It is found that the best-fit values agree within 10% in most cases. We checked the spectral fitting using a narrower energy band of 3.0–4.2 keV, which agrees well with the results in 1.6–5.0 keV. The electron loss continuum considered in the extra-large (XL) response matrix mainly affects the continuum component below 1.8 keV, so the large (L) matrix was used for the fitting in the 3.0-4.2-keV band without its effect. As mentioned in the main text, the emission lines of Ar and K can be explained by almost a single plasma model, thus we fitted the spectrum using a single NEI (pshock or nei) model in this energy band. The NXB model was not included in this analysis, because the NXB level is three orders of magnitude lower than the emissions from the remnant. In fitting over a wide energy band, the plasma parameters of K and Ar may be affected by other components and their abundance may change. On the other hand, even with this simple approach in the narrow band, the K/Ar ratios agree very well, supporting the robustness of our K/Ar measurements. In the case of Cl/S, the best-fit values agree within 40% in most cases, which implies larger systematic errors for the Cl observations. This is influenced by the difficulty in modelling the continuum emission; in particular, using a wider fitting energy range tends to result in a larger Cl/S ratio. A similar trend is also seen in the case of the P/Si ratio. These variations probably reflect the magnitude of systematic uncertainties, but even when taking them into account, our conclusions remain robust.

We also investigated a broader energy band from 1.6 to 12 keV, including the Fe K emissions. In this energy band, it is difficult to ignore the non-thermal emission even in the SE region, so a power-law component was added to the analysis (the NXB model is also included). Here, the photon index of the power-law component was fixed at a typical value of 2.3 (ref. 45), because it was difficult to determine. We confirmed that the mean values of the K/Ar ratios stay within the 1σ confidence range with any photon-index values in the range of 2.3–3.0 even in the cases of the W and N regions where the non-thermal components are more dominant than in the SE region. The addition of a power-law component can reduce the fitted continuum level of the thermal component, thereby resulting in higher abundance values as the line to continuum ratio increases. On the other hand, the K/Ar ratio is almost the same as those for the 1.6–5.0-keV band. The result with AtomDB has a slightly larger K/Ar ratio of 1.5, but the spectral fit around the Fe K band is worse than with SPEX. Currently, there is a discrepancy between atomic codes in the Fe band, and an update is planned. Since the uncertainty may affect the determination of the plasma parameters of K and Ar, the results in the 1.6–5-keV band are considered more reliable for the present analysis.

Supplementary Fig. 2 shows a comparison of the spectra in 3.0–4.2 keV generated by AtomDB and SPEX. In the energy band where the Ar and K emissions are present, there are some differences between the different atomic codes, but the spectral models are in good agreement. Several updates were made in these atomic databases to explain this observation of Cassiopeia A. In the case of AtomDB, the new data (version 3.1.0) include major improvements to the dielectronic recombination satellite lines for B-like to H-like ions; notably improved inner shell lines from excitation for the same, going up to n = 15; inclusion of the K-beta and K-gamma lines from inner shell excitation; and revised fluorescence yields for n = 3 and above inner shell lines. In the case of SPEX, a new calculation has been performed for Be- and Li-like Si, P, S, Cl Ar, and K, focusing on a complete set of inner-shell and dielectronic recombination transitions up to n = 7. For the dominant S He- and H-like transitions, SPEX now includes an extended calculation up to n = 52. The high Rydberg series of He-like S lines reproduce better the observed spectrum around 3.2 keV. Since the fits to the 3.2-keV feature may influence the measurement of the K abundance, a Gaussian component has been introduced into the model with AtomDB to compensate the atomic code difference. The two atomic codes currently agree on K/Ar with a difference of only a few per cent.

Model dependence on odd-Z yields

Nucleosynthetic studies have predicted that the fraction of odd-Z elements in SN ejecta depends on several stellar factors, for example, stellar rotation21, binary interaction12, internal activities11 and initial metallicity21. To investigate these effects, we plot the final yields in ejecta in theoretical models with our observational results in Extended Data Fig. 1. The top panel shows the models provided by Limongi and Chieffi in ref. 21, in which they calculated the SN yield with different rotation velocities (v = 0, 150 and 300 km s−1) and metallicities ([Fe/H] = −3, −2, −1 and 0). While the even-Z elements depend less on the parameters, the odd-Z element yields vary from the models with different parameters. In particular, K/Ar ratios appear to increase with both higher metallicities and faster rotational velocities. The middle panel shows the models of Farmer et al. from ref. 12, in which they surveyed effects of binary interactions by stellar simulations with the companions, whose initial masses were set to be 0.8 times the main stars. In the range of the plot, the K/Ar ratios are likely to be higher in the binary models, but the yields of other elements, including those with even Z, seem not to have clear dependence on whether the progenitor was a single star or a binary system. The bottom panel shows the models by Sukhbold et al.22. We analysed the mass fraction profiles at the onset of core collapse, and plot the yields of models in which the O-burning layers are merging with outer C-burning layers (since this resembles the characteristics of a shell merger29, hereafter we refer to this as a ‘shell merger candidate’) and those in which the distinct borders between the layers are maintained (see below; see Supplementary Fig. 3 and ref. 46 for more details). Consistent with ref. 11, the shell merger candidates show higher Cl and K yields than those of their models without shell merger.

To compare the relative contributions of these effects to the enhancement of odd-Z elements, we constructed a more comprehensive model plot of the K/Ar and Cl/S ratios, using the models of Nomoto et al.20, Limongi and Chieffi21 Roberti et al.30, Farmer et al.12 and Sukhbold et al.22 (hereafter NKT13, LC18, R24, F23 and S16), incorporating our observational results (Extended Data Fig. 2). These models were computed using different stellar evolution codes (for example, KEPLER and FRANEC), each employing distinct nuclear reaction networks and treatments. However, as shown in Extended Data Fig. 2, the NKT13, LC18 and S16 models excepted for the shell merger candidates show similar ratios, implying that the variation due to code differences appears small compared with the range caused by physical parameters such as rotation, binarity and shell mergers. The K/Ar ratios and Cl/S ratios roughly show positive correlation, and those in shell merger candidates in the S16 and F23 models are probably able to enhance the odd-Z yields much more than rotating models. Both the K/Ar and Cl/S ratios in Cas A are close to the shell merger candidates in S16 models and both single and binary models of F23, rather than LC18 models incorporating stellar rotation and metallicity effects. R24 is an extension of LC18 to higher metallicity ([Fe/H] = 0.3, that is, twice solar). Although increased metallicity is known to enhance odd-Z element production, even the R24 models cannot reach the Cl/S and K/Ar ratios observed in Cas A. Given that the initial metallicity of Cas A is suggested to be sub-solar47, the contribution of metallicity to the nucleosynthesis of odd-Z elements in Cas A is likely to be limited. While the S16 results may suggest that progenitors with initial masses of ≳20M⊙ tend to have enhanced odd-Z yields, it is important to note that the LC18 and NKT13 models in the similar mass range do not show such enhancements. This implies that the enhanced odd-Z element yields are unlikely to be a simple function of initial mass and instead highlight the importance of stellar processes such as shell mergers. We also note that, although binary progenitors have been suggested to enhance odd-Z yields, single-star models show variable values of these ratios, and even they can exhibit notably enhanced odd-Z yields.

While the enhancement of odd-Z elements such as K and Cl is often attributed to progenitor rotation or binary interaction, a closer inspection of our model survey reveals a more nuanced picture, indicating that additional mechanisms may be at play. To explore this point, we further investigated the mass fraction profiles of the models in Fig. 3; the results are shown in Supplementary Fig. 3. The top panel shows the 14.9M⊙ model from Sukhbold et al. (S16), which retains a textbook, onion-like structure. In contrast, the bottom panel exhibits a structure where the Si-rich layer (O-burning layer) appears to be merging with the O/Ne-rich layer (C-burning layer). The mass fraction profiles of the binary and rotating progenitor models deviate from the canonical onion-skin structure, exhibiting signatures of partial mixing or merging between adjacent burning layers. It has been suggested that such layer merging is more likely to occur in rotating progenitor models30, and some of the binary models used in this study have also been reported to exhibit similar structural features48. Furthermore, in ref. 12, Farmer et al. found that their models with abundant K produced the amount during their pre-SN phase and pointed out that the convective structure outside the Fe core is the key to the production. These results suggest that the internal structure at the final stage of stellar evolution may influence the yields of odd-Z elements as much as, or even more than, the initial stellar parameters. According to Ritter et al.11, the pre-SN yields of odd-Z elements in such progenitors depend on the entrainment rate at the O/Si-burning shell. In other words, they are sensitive to the nature of the mixing processes occurring during the O/Si-burning phase shortly before or at the onset of core collapse. Furthermore, some multi-dimensional SN simulations suggest that such a convective state can result in asymmetry of the remnant49. Therefore, future deep observations of the asymmetric SNR Cas A, including spatially resolved abundance distributions, will provide further insight into the nucleosynthetic processes responsible for the excess production of odd-Z elements.