Metallicity Parameterization¶
tengri models stellar metallicity through three interpolation modes of increasing flexibility, and optionally couples metallicity to the SFH via a self-consistent chemical evolution model. The extension to \(\alpha\)-enhanced abundance patterns is described separately in Appendix Alpha-Element Enhancement.
Metallicity Interpolation Modes¶
Three metallicity interpolation modes are provided:
Fixed (linear).¶
Linear interpolation in \(\log_{10}(Z/Z_\odot)\) between the two nearest grid points: $\(F_{\nu}(Z) = (1 - f) \, F_{\nu}(Z_i) + f \, F_{\nu}(Z_{i+1}),\)\( where \)f = (\log Z - \log Z_i) / (\log Z_{i+1} - \log Z_i)\( and \)i$ is the index of the lower bracketing metallicity.
Linear ramp.¶
A time-dependent metallicity that increases linearly from an initial value \(\log Z_0\) at the oldest age to a final value \(\log Z_f\) at the present. The SSP flux at each age \(t_j\) is interpolated to the metallicity \(\log Z(t_j)\) using the fixed linear method above.
Triweight kernel (DSPS-compatible).¶
The metallicity is distributed across multiple grid points using the triweight kernel CDF from DSPS(Hearin et al. 2023, Eq. 10): $\(\begin{aligned} w_k = {} & \Phi_{\rm tw}\!\left(\frac{\log Z_{\rm edge,k+1} - \log Z}{\sigma_Z}\right) \nonumber \\ & - \Phi_{\rm tw}\!\left(\frac{\log Z_{\rm edge,k} - \log Z}{\sigma_Z}\right), \end{aligned}\)\( where \)\Phi_{\rm tw}\( is the cumulative triweight kernel with compact support \)|x| < 3\(, and \)\sigma_Z\( (default 0.1 dex) controls the scatter. The kernel density is the \)C^2\(-smooth polynomial \)\(K_{\rm tw}(x) = \frac{35}{96}\left(1 - \frac{x^2}{9}\right)^{\!3}, \quad |x| < 3,\)\( with \)x = (\log Z - \log Z_{\rm grid})/\sigma_Z\(. Compact support limits each query to a finite number of contributing grid points. The \)C^2\( smoothness guarantees that second derivatives \)\partial^2 f/\partial Z^2$ are continuous across grid boundaries, linear interpolation, by contrast, has discontinuous first derivatives that create kinks in the log-likelihood and slow gradient-based samplers. tengri uses the same triweight kernel for mass-remaining fraction interpolation (Mass-Remaining Fraction) and free-redshift photometry tables (Photometric Precomputation).
Grid interpolation for tabulated models.¶
Nebular, AGN torus, and dust emission grids use multilinear (bilinear or trilinear) interpolation in log-parameter space. All interpolation operations are implemented as JAX array operations compatible with jax.jit and jax.grad, so gradients propagate through the grid lookups. Multilinear interpolation has discontinuous second derivatives at grid boundaries; for models where this matters (e.g., the SKIRTOR torus grid), the grid spacing is chosen fine enough that the discontinuities do not affect sampler convergence.
Chemical Evolution Models¶
Beyond the fixed and linearly ramped metallicity modes (Metallicity Interpolation Modes), tengri provides a self-consistent chemical evolution model (Bellstedt et al. 2021; Robotham et al. 2020). In the closed-box limit, the gas-phase metallicity evolves as \(Z(t) = -y\,\ln\mu(t)\), where \(y\) is the nucleosynthetic yield and \(\mu(t) = M_{\rm gas}(t)/M_{\rm tot}\) is the gas fraction derived from the cumulative SFH. A leaky-box extension replaces \(y\) with an effective yield \(y_{\rm eff} = y/(1+\eta)\), where \(\eta\) is the mass-loading factor. The single free parameter is the present-day gas metallicity \(\log(Z_{\rm gas}/Z_\odot)\); the yield is solved to match this anchor. The resulting \(Z(t)\) is interpolated onto the SSP metallicity grid at each age bin, adding negligible computational cost.
Alpha-Element Enhancement¶
Real galaxies do not have solar-scaled abundance patterns. Alpha elements (O, Mg, Si, Ca, Ti) are produced primarily by core-collapse supernovae on short timescales (\({\sim}3\)–\(30\) Myr), while iron-peak elements are produced predominantly by Type Ia supernovae whose delay-time distribution extends to Gyr timescales. Stars formed before significant Ia enrichment therefore exhibit elevated \([\alpha/\mathrm{Fe}]\) ratios of \(+0.3\) to \(+0.5\) dex (Thomas et al. 2003; Conroy et al. 2014). This \(\alpha\)-enhancement is ubiquitous in massive ellipticals, the Milky Way thick disc, and high-redshift quiescent galaxies at \(z > 2\) (Beverage et al. 2024). Ignoring \(\alpha\)-enhancement biases stellar population parameters: at fixed total metallicity, an \(\alpha\)-enhanced population has weaker Fe lines and stronger Mg/Ca/Ti features, which can be misinterpreted as an age or metallicity shift when only solar-scaled templates are available.
The 4D SSP Grid¶
A four-dimensional SSP grid with axes \((\log Z_{\rm total},\, [\alpha/\mathrm{Fe}],\, t_{\rm age},\, \lambda)\) gives \(\alpha\)-enhanced stellar population synthesis. met_alpha_fe acts only on a grid that has an \([\alpha/\mathrm{Fe}]\) axis. The alpha-enhanced grids that tengri hosts (sMILES, BPASS v2.3, and the \(\alpha\)-MC library of Park et al. 2025, all named with _afe in tengri.list_remote_ssps()) have one, at \([\alpha/\mathrm{Fe}] = \{-0.2, 0.0, +0.2, +0.4, +0.6\}\) dex. A 3D grid has none, so met_alpha_fe is refused at build on it, whatever its metallicity mode. On such a grid the interpolation is bilinear in \((\log Z, [\alpha/\mathrm{Fe}])\), four multiplications per wavelength pixel, and differentiable with respect to both \(\log Z_{\rm total}\) and \([\alpha/\mathrm{Fe}]\).
met_alpha_fe is fixed at 0.0 by default on every grid and with every observation. It is free only through an explicit prior, for example met={'alpha_fe': Uniform(-0.2, 0.4)}, which is checked against the grid’s \([\alpha/\mathrm{Fe}]\) nodes: a prior with no overlap is refused, and one that extends past them is kept with a GridSupportWarning, since the lookup clips at the edge nodes. A bare FREE is refused, and a wildcard does not free it. Broadband photometry constrains \([\alpha/\mathrm{Fe}]\) weakly at fixed total \(Z\): on the \(\alpha\)-MC Kroupa grid, raising it from 0 to +0.4 changes the \(g\)-band flux of a dusty delayed-SFH model by 0.4%. The Mg b and Fe features near 5175–5335 Å carry the signal, so a spectrum that resolves them or Lick indices are what constrain it.
Surviving mass.¶
The sMILES, MILES \([\alpha/\mathrm{Fe}]\) and \(\alpha\)-MC grids embed a surviving-mass table on the \([\alpha/\mathrm{Fe}]\) axis, so stellar_mass_surviving follows \([\alpha/\mathrm{Fe}]\) on them. The hosted BPASS v2.3 grids (bpss_v23_{bin,sin}_imf135_300_afe) have no surviving-mass table: their ssp_mass_living_stars counts living stars only, without remnants, and tengri does not read it. Loading one is refused by default, naming the opt-in load_ssp_data(..., mass_remaining='dsps_fit'). With the opt-in, the surviving mass comes from the DSPS fit for the Chabrier IMF, which is not BPASS’s IMF (a warning says so); that fit depends on age only, so the surviving mass follows neither \([\alpha/\mathrm{Fe}]\) nor \(Z\).
Metallicity convention.¶
Every grid holds its metallicity nodes at a fixed total metallicity. ssp_lgmet is \(\log_{10} Z\) with \(Z\) the total metallicity, and met_logzsol is \(\log_{10}(Z/Z_\odot)\) with \(Z_\odot = 0.0142\) on every grid. The loader reads the file’s metallicity_axis attribute. A file on a fixed-\([\mathrm{Fe/H}]\) axis is refused, with a pointer to its _totalz variant, which holds the same spectra resampled to fixed total \(Z\). tengri does not convert between the two conventions. On a fixed total-\(Z\) grid, \([\alpha/\mathrm{Fe}]\) changes at fixed total \(Z\), so an \(\alpha\)-rich population has less iron at the same met_logzsol.
Time-Evolving \([\alpha/\mathrm{Fe}]\)¶
In the simplest mode, a single \([\alpha/\mathrm{Fe}]\) is applied uniformly to all stellar populations. However, tengri also supports a time-evolving \([\alpha/\mathrm{Fe}](t)\) that captures the physical expectation from chemical evolution: old populations formed from \(\alpha\)-enriched gas, while young populations formed from gas with approximately solar abundance ratios. We parameterize this as a linear ramp in lookback time: $$[\alpha/\mathrm{Fe}](t_{\rm lb}) = [\alpha/\mathrm{Fe}]_{\rm young}
\bigl([\alpha/\mathrm{Fe}]_{\rm old}
[\alpha/\mathrm{Fe}]{\rm young}\bigr) \times \frac{t{\rm lb}}{t_{\rm universe}},
with two parameters, \([\alpha/\mathrm{Fe}]_{\rm old}\) and \([\alpha/\mathrm{Fe}]_{\rm young}\). Both are fixed at 0.0 by default, like the scalar; either is free only through an explicit prior (for example 'alpha_fe_old': Uniform(0.0, 0.4)), which is checked against the grid’s \([\alpha/\mathrm{Fe}]\) nodes. A bare FREE is refused and a wildcard never frees them. On a grid with an \([\alpha/\mathrm{Fe}]\) axis, each SSP age collapses the axis at its own \([\alpha/\mathrm{Fe}](t_{\rm lb})\), so the exact forward, the photometry and spectrum lookup tables, the dust energy-balance table and the spectral-index fast path all see the same per-age ramp. \(t_{\rm universe}\) is the age of the universe at the model redshift. The flag and the two parameters are spelled in the metallicity group: met={'alpha_fe_evolving': True, 'alpha_fe_old': ..., 'alpha_fe_young': ...}. It is exclusive with a scalar alpha_fe. On a grid without an \([\alpha/\mathrm{Fe}]\) axis the evolving mode is refused at build; a non-zero scalar \([\alpha/\mathrm{Fe}]\) on such a grid is refused at build too (see the next section).
References¶
Bellstedt, Sabine, Aaron S. G. Robotham, Simon P. Driver, et al. 2021. “Galaxy and mass assembly (GAMA): the inferred mass-metallicity relation from z = 0 to 3.5 via forensic SED fitting.” 503 (3): 3309–25. https://doi.org/10.1093/mnras/stab550.
Beverage, Andrew G., Mariska Kriek, Charlie Conroy, Katherine A. Suess, David J. Setton, and Rachel Bezanson. 2024. “The Heavy Metal Survey. I. The Evolution of Stellar Metallicities, Abundance Ratios, and Ages of Massive Quiescent Galaxies since z \(\sim\) 2.” 966 (1): 1. https://doi.org/10.3847/1538-4357/ad372d.
Byrne, C. M., E. R. Stanway, J. J. Eldridge, L. McSwiney, and O. T. Townsend. 2022. “The dependence of theoretical synthetic spectra on \(\alpha\)-enhancement in young, binary stellar populations.” 512 (4): 5329–38. https://doi.org/10.1093/mnras/stac807.
Conroy, Charlie, Genevieve J. Graves, and Pieter G. van Dokkum. 2014. “Early-type Galaxy Archeology: Ages, Chemical Abundances, and Star Formation Histories.” 780 (1): 33. https://doi.org/10.1088/0004-637X/780/1/33.
Hearin, Andrew P., Jonás Chaves-Montero, Alex Alarcon, Matthew R. Becker, and Andrew Benson. 2023. “DSPS: Differentiable stellar population synthesis.” 521 (2): 1741–56. https://doi.org/10.1093/mnras/stad456.
Knowles, Adam T., A. E. Sansom, C. Allende Prieto, and A. Vazdekis. 2023. “[sMILES: a library of semi-empirical MILES stellar spectra with variable [\(\alpha\)/Fe] abundances]{.nocase}.” 523 (3): 3450–70. https://doi.org/10.1093/mnras/stad1647.
Park, Minjung, Charlie Conroy, Benjamin D. Johnson, Joel Leja, Aaron Dotter, and Phillip A. Cargile. 2025. “\(\alpha\)-MC: Self-consistent \(\alpha\)-enhanced stellar population models covering a wide range of age, metallicity, and wavelength.” The Astrophysical Journal 994: 165. arXiv:2410.21375. https://doi.org/10.3847/1538-4357/ae0cba.
Robotham, A. S. G., S. Bellstedt, C. del P. Lagos, et al. 2020. “ProSpect: generating spectral energy distributions with complex star formation and metallicity histories.” 495 (1): 905–31. https://doi.org/10.1093/mnras/staa1116.
Salaris, M., A. Chieffi, and O. Straniero. 1993. “The alpha -enhanced isochrones and their impact on the FITS to the Galactic globular cluster system.” 414: 580. https://doi.org/10.1086/173105.
Thomas, Daniel, Claudia Maraston, and Ralf Bender. 2003. “Stellar population models of Lick indices with variable element abundance ratios.” 339: 897–911. https://doi.org/10.1046/j.1365-8711.2003.06248.x.
Vazdekis, A., P. Coelho, S. Cassisi, et al. 2015. “Evolutionary stellar population synthesis with MILES - II. Scaled-solar and \(\alpha\)-enhanced models.” 449 (2): 1177–214. https://doi.org/10.1093/mnras/stv151.