Nebular Emission

Nebular emission arises from gas photoionized by hot stars (in H 2 regions) or by the AGN accretion disc (in the narrow-line region). tengri registers four selectable nebular backends: (i) three core stellar backends of increasing complexity (BakedIn, CloudyGrid, and the Cue neural emulator; §Cue Neural Emulator), and (ii) the CB_19 extended grid (CB_19 Grid Backend). Any of these may be combined with the diffuse ionized gas mixing component, which is controlled by a mixing fraction rather than selected as a backend in its own right, while shock emission (Shock Emission Backend) composes with any of them as a separate additive component. The MAPPINGS V photoionization grid (MAPPINGS V Photoionization Backend (Flury+2024)) is implemented but is not currently registered for selection through the model-construction API. AGN-ionized nebular emission is described separately in AGN-Ionized Nebular Emission. All backends are selected at model construction time and share the same interface.

Ionizing Photon Rate

All nebular models require the hydrogen-ionizing photon production rate \(Q_{\rm H}\), computed from the composite stellar population (CSP) spectrum: $$Q_{\rm H} = \int_0^{912,\text{\AA}} \frac{L_\nu}{h\nu}, d\nu,

(1)\[\]

where \(L_\nu\) is the CSP luminosity density. This integral is evaluated per SSP age bin and metallicity, then cached as a precomputed table \(Q_{\rm H}(Z, t_{\rm age})\) to avoid recomputation during inference.

Emission-Line Physics

Line luminosities assume Case B recombination at \(T_e = 10^4\) K, with \(\alpha_B = 2.58\times10^{-19}\,\mathrm{m^3\,s^{-1}}\) (Ferland 1980). The Case B cascade sets every Balmer and Lyman line relative to the ionizing photon rate.

The escape parameter neb_fesc multiplies the line and continuum amplitudes by one common factor, \(k = (1 - f_{\rm esc} - f_{\rm dust})\,/\,[1 + 0.597\,(f_{\rm esc} + f_{\rm dust})]\), where \(f_{\rm dust}\) is neb_fdust_frac \(\times\,(1 - f_{\rm esc})\). Because the factor is shared, neb_fesc leaves line ratios unchanged. A Balmer decrement that is not Case B, as expected for density-bounded gas, therefore cannot be produced by neb_fesc; the data’s decrement is absorbed by dust reddening. If you suspect density-bounded gas, treat dust_tau_bc as covariant with the nebular escape parameters when you interpret the fit.

The stellar metallicity met and the gas-phase metallicity neb={'logZ_gas': ...} are independent. met selects the stellar templates; logZ_gas alone sets the metallicity of the emitting gas, so set both when they differ.

BakedIn Backend

When SSP templates already include nebular emission (e.g., FSPS “wNE” files with fixed \(\log U\) and \(\log Z_{\rm gas}\)), the BakedIn backend is a no-op: it returns zero additional nebular contribution since the emission is already part of the SSP flux arrays. This is the default when no CLOUDY grid is specified.

CloudyGrid Backend

The CloudyGrid backend loads precomputed CLOUDY photoionization grids (Byler et al. 2017) and computes nebular emission as a function of two free parameters: ionization parameter \(\log U\) and gas metallicity \(\log Z_{\rm gas}\).

The grid has dimensions \((n_{\rm met}, n_{\rm age}, n_{\log U})\) for both emission lines and nebular continuum, stored in \(\log_{10}\) space for interpolation accuracy. Luminosities are in units of \(L_\odot\, Q_{\rm H}^{-1}\) (lines) and \(L_\odot\,{\rm Hz}^{-1}\, Q_{\rm H}^{-1}\) (continuum).

Long-wavelength continuum extension

The CLOUDY grids (Byler et al. 2017) and the Cue emulator’s continuum tables both end at 1e8 Å (1 cm). Beyond this wavelength, the continuum is analytically extended as optically thin thermal free-free: \(L_\nu \propto \nu^{-0.1}\), anchored at the last tabulated node (index constant NEBULAR_FREEFREE_TAIL_ALPHA_NU = -0.1 in tengri.components.nebular._constants). This is an analytic extension, not an emulator or table prediction. Optically thin bremsstrahlung has this index; pcigale’s tabulated nebular continuum, which runs to 1 m, measures \(\nu^{-0.096}\) over 1 cm to 1 m, and bagpipes tabulates to 3.6 cm.

A model with Cue nebular emission and no radio block declares wavelength nodes to 1 m: native_wave_nebular("cue") appends 40 logarithmically spaced nodes (20 per decade) from 1e8 to 1e10 Å (NEBULAR_CONTINUUM_WAVE_MAX = 1e10 Å in tengri.utils.wavelength). With a radio block, the tail continues to the radio grid end (3e11 Å). The CloudyGrid backend declares no native wavelength grid; the free-free tail is carried by components that supply nodes (a dust emission template or the radio wing). This analytic extension via interp_continuum_with_freefree_tail is in tengri.components.nebular._shared.

For a given CSP with mass weights \(w_j\) at SSP ages \(t_j\): $\(\begin{aligned} L_{\rm line} = {} & (1 - f_{\rm esc}) \sum_{j} w_j \, Q_{\rm H}(Z, t_j) \nonumber \\ & \times \ell_{\rm grid}(\log Z_{\rm gas}, \log t_j, \log U), \end{aligned}\)\( where \)\ell_{\rm grid}\( is the grid luminosity per \)Q_{\rm H}\( interpolated at \)(\log Z_{\rm gas}, \log t_j, \log U)\(, and \)f_{\rm esc}\( is the ionizing photon escape fraction. The default interpolant is the node-exact, shape-preserving monotone cubic (PCHIP, Fritsch and Carlson 1980) applied to \)\log_{10}\ell_{\rm grid}\( over the three log axes, the same interpolant the library-template lookups use (owner ruling 2026-10-11). It returns the tabulated value at every node and never overshoots between nodes. The lookup clamps to the tabulated hull on every axis; the declared prior \)\log U \in [-5, 0]\( reaches two node spacings past the grid's \)[-4, -1]\(, a support mismatch that the build-time grid-support check narrows or warns on, and the metallicity prior lies inside the hull. `grid_interp="linear"` selects the earlier trilinear lookup and `grid_interp="triweight"` the smoothing kernel; both remain available. Only young SSP bins (\)\log_{10}(t_{\rm age}/{\rm yr}) < 8.0\() contribute, reducing the sum from \)\mathord{\sim}90\( to \)\mathord{\sim}10$ terms.

Ly-\(\alpha\) (\(\lambda = 1215.67\) Å) receives a separate escape fraction \(f_{\rm esc,Ly\alpha}\) to account for resonant scattering: $\(L_{\rm Ly\alpha} \to L_{\rm Ly\alpha} \cdot \frac{1 - f_{\rm esc,Ly\alpha}}{1 - f_{\rm esc}}.\)$

Cue Neural Emulator

The Cue backend (Li et al. 2025) is a neural network emulator trained on \(2\times10^6\) CLOUDY v22.00 photoionization simulations. The original Cue code is implemented in TensorFlow; tengri provides a complete JAX reimplementation that loads the same pre-trained network weights but evaluates them using jax.numpy operations, enabling JIT compilation and automatic differentiation through the entire emulator. The reimplementation is validated to reproduce the original TensorFlow predictions to machine precision (relative error \(< 10^{-6}\)).

Unlike the CloudyGrid backend, which requires a pre-computed grid tied to specific stellar population models, Cue accepts an arbitrary ionizing spectrum parameterized as a 4-segment piecewise power law below the Lyman limit. This decoupling of the ionizing source from the gas physics is what enables Cue to model nebular emission from stellar populations, AGN, post-AGB stars, Pop III stars, or any composite source.

Ionizing spectrum parameterization.

The ionizing spectrum below 912 Å is decomposed at four ionization edges into piecewise power-law segments: $$F_\nu = A_k,\lambda^{\alpha_k}, \quad k = 1,\ldots,4,

(2)\[\]

where the segment boundaries are set by key ionization potentials:

Segment

Wavelength range

Ion edge

1

\(1\)–\(228\) Å

He ii (54.4 eV)

2

\(228\)–\(353\) Å

O ii (35.1 eV)

3

\(353\)–\(504\) Å

He i (24.6 eV)

4

\(504\)–\(912\) Å

H i (13.6 eV)

The 4 slopes \(\alpha_1\)–\(\alpha_4\) and 3 log-integrated-flux ratios \(\log(F_2/F_1)\), \(\log(F_3/F_2)\), \(\log(F_4/F_3)\) together give 7 parameters that fully characterize the ionizing spectrum shape. The overall normalization is set by the ionizing photon production rate \(Q_H = \int_0^{912\,\text{\AA}} (L_\nu / h\nu)\,d\nu\).

For a single power law \(f_\nu \propto \nu^{\alpha_{\rm pl}}\) (as used for AGN; AGN-Ionized Nebular Emission), all four segments have the same slope \(\alpha_k = -\alpha_{\rm pl}\) in wavelength space, and the flux ratios follow analytically from the segment boundaries. For composite stellar populations, the 7 parameters are fitted from the SSP ionizing spectrum using a least-squares procedure that preserves both the spectral shape and the total \(Q_H\).

Gas parameters.

Five additional parameters describe the photoionized gas:

  • \(\log U \in [-4, -1]\): ionization parameter at the illuminated face of the cloud.

  • \(\log n_{\rm H} \in [1, 4]\): hydrogen number density (\(\mathrm{cm}^{-3}\)).

  • \(\log({\rm O/H})/({\rm O/H})_\odot \in [-2.2, +0.5]\): oxygen abundance. All metals except C and N scale with O.

  • \(({\rm C/O})/({\rm C/O})_\odot \in [0.1, 5.4]\): carbon-to-oxygen ratio, treated as a free parameter (not tied to any abundance relation).

  • \(({\rm N/O})/({\rm N/O})_\odot \in [0.1, 5.4]\): nitrogen-to-oxygen ratio, also free.

The free C/O and N/O are a key advantage over traditional CLOUDY grids (e.g. Gutkin et al. (2016; Feltre et al. 2016)), which tie nitrogen to oxygen via a fixed secondary-production relation. This flexibility is essential for fitting BPT line ratios (Baldwin et al. 1981), where the \N [ii]/H\(\alpha\) ratio depends sensitively on N/O.

Neural network architecture.

Cue uses the Speculator architecture (Alsing et al. 2020), with 14 sub-networks grouped by element species and ionization potential, each predicting PCA coefficients that are inverse-transformed to \(\log_{10}\) luminosities per ionizing photon. The 128 predicted emission lines span H, He, C, N, O, Ne, S, and Ar species. The tengri reimplementation evaluates all sub-networks in parallel as a single JIT-compiled JAX forward pass, differentiable with respect to all 12 input parameters. See Li et al. (2025) for full architectural details.

Accuracy.

Hydrogen recombination lines are predicted to \(<1\%\) accuracy. Of lines contributing \(>1\%\) of the total line flux, 93% have \(<5\%\) error relative to CLOUDY. Three high-ionization lines (\Ne [iv] \(\lambda 4720\), \Ar [iv] \(\lambda 7331\), \S [iv] \(10.5\,\mu\)m) have \(\sim\)20% errors due to their sensitivity to the hardest ionizing photons. The \(\sim\)5% typical accuracy is comparable to differences between CLOUDY versions (Li et al. 2025), and is well below typical spectroscopic calibration uncertainties.

Which stellar ages are counted.

In the high-level mode \(Q_H\) and the ionizing-spectrum shape are combined over every age bin of the SSP, weighted by the population’s mass: segment luminosities add linearly across ages and each slope is the luminosity-weighted mean of the bins’. Old bins (post-AGB and extreme horizontal-branch stars) therefore contribute, and a quiescent galaxy has nebular emission at about the case-B H\(\alpha\) per ionizing photon. neb={'type': 'cue', 'lyc_max_age_yr': 1e8} limits the sum to ages below 100 Myr. Cue’s forbidden-line ratios for old (post-AGB) ionizing shapes are not validated against CLOUDY.

The stellar SED must give up the LyC of every age Cue re-emits, so with Cue counting every age the dust key lyc_reprocessed_by defaults to 'all' (it defaults to 'young' for the other backends, and for Cue with a finite lyc_max_age_yr, where the stellar SED keeps the LyC of the ages Cue does not count). 'young' removes only the LyC of ages below t_birth_yr from the stellar SED, so pairing it with Cue counting older ages would create the older ages’ energy twice (94-96% of the old LyC energy of a quiescent galaxy, about 0.2% of its bolometric luminosity); that pairing raises an error. Set lyc_max_age_yr <= t_birth_yr to use 'young' with Cue. 'all' with a finite lyc_max_age_yr is refused: it would remove the LyC of the uncounted ages from the SED and re-emit none of it. The former 100 Myr cut with 'young' leaves a small gap (ages between t_birth_yr and 100 Myr keep their LyC in the SED while Cue re-emits it) and warns once.

Two calling modes.

tengri’s Cue implementation supports two modes: (i) high-level, where ionizing spectrum parameters are derived automatically from the SSP composite spectrum and the 7 ionspec values are precomputed at Model initialization; and (ii) low-level, where all 12 parameters are specified directly by the caller. The low-level mode is used for AGN-ionized nebular emission (AGN-Ionized Nebular Emission), where the ionizing spectrum comes from the accretion disc rather than from stellar populations.

CB_19 Grid Backend

The CB19 backend implements the Charlot & Bruzual 2019 CLOUDY photoionization grid from the 3MdB_17 database [1]. The grid comprises 2,358,330 models computed with CLOUDY c17.01 using C&B 2019 SSP and CSF ionizing spectral energy distributions. It extends the CloudyGrid backend with two additional abundance axes: the carbon-to-oxygen ratio \(\log(\mathrm{C/O})\) and a nitrogen-to-oxygen offset \(\Delta(\mathrm{N/O})\), making it better suited for abundance-sensitive diagnostics.

The six-dimensional interpolation grid spans: \(\log(\mathrm{O/H})\), \(\log t_{\rm age}\), \(\log U\), \(\log n_{\rm H}\), \(\log(\mathrm{C/O})\), and \(\Delta(\mathrm{N/O})\). A seventh discrete axis, HbFrac, parametrizes matter-bounded nebulae: $$\mathrm{HbFrac} = \frac{L_{H\beta}(\text{matter-bounded})}{L_{H\beta}(\text{radiation-bounded})},

(3)\[\]

with \(\mathrm{HbFrac} = 1\) the default radiation-bounded case and \(f_{\rm esc,\,LyC} \approx 1 - \mathrm{HbFrac}\) for matter-bounded models.

Unit convention. CB_19 stores all line fluxes as dimensionless H\(\beta\) ratios. The conversion to \(L_\odot\,Q_{\rm H}^{-1}\) (consistent with the CloudyGrid interface) uses the Case B recombination coefficient (Osterbrock and Ferland 2006): $$\frac{L_{H\beta}}{Q_{\rm H}} = 4.78 \times 10^{-13}\ \mathrm{erg,photon}^{-1} \approx 1.249 \times 10^{-46}\ L_\odot,\mathrm{s,photon}^{-1},

(4)\[\]

in agreement with Byler et al. (2017). The final line luminosity is $$\begin{aligned} L_{\lambda} = {} & (1 - f_{\rm esc}) \sum_{j} w_j , Q_{\rm H}(Z, t_j) \nonumber \ & \times \tilde{F}{\lambda}(\boldsymbol{\theta}{\rm gas}) \cdot \frac{L_{H\beta}}{Q_{\rm H}},

\end{aligned}

(5)\[\]

where \(\tilde{F}_\lambda(\boldsymbol{\theta}_{\rm gas})\) is the 6D linearly interpolated grid ratio with \(\boldsymbol{\theta}_{\rm gas} = (\log\mathrm{O/H},\, \log t_j,\, \log U,\, \log n_{\rm H},\, \log\mathrm{C/O},\, \Delta\mathrm{N/O})\).

No nebular continuum. The 3MdB_17 database does not include continuum-level output for the CB_19 run. The backend therefore returns zero nebular continuum. For science cases that require the free–bound and two-photon continuum, the CloudyGrid or Cue backend should be stacked alongside CB19.

Metallicity axis. The CB_19 grid uses the CLOUDY c17.01 solar oxygen abundance \(12 + \log(\mathrm{O/H})_\odot = 8.93\), whereas tengri uses absolute \(\log_{10}Z\) with \(\log_{10}Z_\odot = -1.848\) (Asplund et al. 2009). The internal conversion (assuming \(\mathrm{O/H} \propto Z\)) is $$\log(\mathrm{O/H}) = \log_{10}Z_{\rm gas} + \Delta_{\rm OH},

(6)\[\]

where \(\Delta_{\rm OH} = -3.07 - (-1.848) = -1.222\).

DIG Mixing Backend

The DIG backend mixes HII-region emission at ionization parameter \(\log U_{\rm HII}\) with diffuse ionized gas at lower ionization \(\log U_{\rm DIG} = \log U_{\rm HII} + \Delta\log U\) (default \(\Delta\log U = -1\) dex): $\(L_{\rm neb} = (1 - f_{\rm DIG})\, L(\log U_{\rm HII}) + f_{\rm DIG}\, L(\log U_{\rm DIG}),\)\( where \)f_{\rm DIG} \in [0, 1]\( is the DIG fraction. In local galaxies, 30--60% of H\)\alpha$ emission originates from diffuse gas outside HII regions (Haffner et al. 2009); ignoring this component biases ionization-parameter estimates and line-ratio diagnostics.

Shock Emission Backend

The Shock backend adds emission lines from radiative shocks using the MAPPINGS V 3MdBs grid (Sutherland and Dopita 2017; Alarie and Morisset 2019). Shock emission is relevant for galaxies with AGN-driven outflows, post-starburst superwinds, and mergers, where fast shocks (\(v_s \gtrsim 150\) km s\(^{-1}\)) produce extreme optical line ratios above the maximum-starburst demarcation on BPT diagrams (Kewley et al. 2001; Kewley et al. 2019).

The 3MdBs grid spans \(v_s \in [100, 1000]\) km s\(^{-1}\), eight magnetic field strengths, five abundance patterns, and six pre-shock densities (Allen et al. 2008; Alarie and Morisset 2019). Shock velocity is interpolated continuously (differentiable); discrete parameters are snapped to the nearest grid point. The absolute scale is set by the shock fraction of total H\(\alpha\); all other line luminosities follow from the MAPPINGS V ratios.

Shock emission is attenuated by whichever of the two-component dust screens the dust_attenuation group’s shock_screen selector names, mirroring the equivalent choice for the nebular continuum and line catalog (nebular_screen). The selector accepts three values:

  • "diffuse" (default): the diffuse-ISM screen only, $\(L_{\rm shock}^{\rm att}(\lambda) = L_{\rm shock}(\lambda)\; \exp\!\bigl(-\tau_{\rm diff}\,k_{\rm diff}(\lambda)\bigr).\)$ (eq-shock-attenuation) Shocked gas from an AGN-driven outflow (shock={'norm': 'lhalpha', ...}) is not, in general, still confined to the compact star-forming birth cloud the young-star screen models, so the diffuse-only form is the default for every shock normalization, including the star-formation-coupled one (norm='frac').

  • "birth_cloud": both screens, \(\exp\!\bigl(-(\tau_{\rm bc}\,k_{\rm bc}(\lambda) + \tau_{\rm diff}\,k_{\rm diff}(\lambda))\bigr)\) – the same young-limit transmission the youngest stars and the nebular continuum receive by default. Appropriate when the shock is known to be embedded in the star-forming birth cloud (e.g. norm='frac' in a compact starburst).

  • "none" (synonym "off"): unattenuated, resolved once at build time.

Table 1 lists the registered parameters.

Parameter

Default

Range

Description

shock_frac

0.0

\([0, 1]\)

Fraction of total H\(\alpha\) from shocks

shock_velocity

300 km s\(^{-1}\)

[100, 1000] km s\(^{-1}\)

Shock velocity

shock_log_density

0.0

\([-2, 3]\)

\(\log_{10}(n/\mathrm{cm}^{-3})\)

shock_b_over_sqrt_n

1.0

grid values

\(B/\sqrt{n}\) (\(\mu\)G cm\(^{3/2}\))

shock_abundance

solar

5 choices

Pre-shock abundance pattern

shock_component

combined

3 choices

shock, precursor, or combined

: Shock emission parameters. shock_frac and shock_velocity are typically set free; others are fixed by default.

MAPPINGS V Photoionization Backend (Flury+2024)

The MappingsPhotoStellar and MappingsPhotoAGN backends implement the MAPPINGS V v5.2.1 photoionization grids from Flury et al. (2024), providing stellar and AGN-ionized nebular emission with homogeneous treatment of abundance patterns and dust depletion. Unlike traditional Cloudy grids (which use scaled-solar abundances), these grids leverage empirical stellar abundance patterns (Nicholls et al. 2017) with Jenkins dust depletion (\(F_\star = 0.43\)) and CHIANTI v10 atomic data, making them directly comparable to the shock grids (Shock Emission Backend).

The stellar grids are 5D (\(\zeta_{\rm O}\), \(\log U\), \(\log t_{\rm age}\), \(\log n_{\rm H}\), SFH mode) with Starburst99 and BPASS v2.2 populations. The AGN grids add \(\log M_{\rm BH}\) and \(\log(L/L_{\rm Edd})\) axes, normalized by ionizing luminosity rather than \(Q_H\). Both include 140 emission lines from UV to far-infrared.

For stellar grids, the line luminosity is: $\(L_{\rm line} = \sum_i w_i \cdot Q_{\rm H}(Z_\star, t_i) \cdot \frac{L_{\rm line}}{Q_{\rm H}}(\zeta_{\rm O}, \log t_i, \log U, \log n_{\rm H}),\)\( where \)w_i\( are the CSP mass weights and only young SSP bins (\)\log_{10}(t_{\rm age}/{\rm yr}) < 8.0\() contribute. For AGN grids, \)Q_H$ is replaced by the ionizing luminosity from the disc model with 5D interpolation. Grids are available from Zenodo (doi:10.5281/zenodo.14140949). Table 2 lists the registered parameters.

Parameter

Default

Range

Description

neb_logu

\(-3.0\)

\([-4, -0.5]\)

Ionization parameter

neb_logn

\(2.0\)

\([1, 4]\)

\(\log_{10}(n_{\rm H}/{\rm cm}^{-3})\)

neb_logz_gas

stellar

\([-2, 0.5]\)

Gas metallicity \(\log_{10}(Z/Z_\odot)\)

AGN-specific (MappingsPhotoAGN only)

agn_logmbh

\(7.0\)

\([4, 7]\)

\(\log_{10}(M_{\rm BH}/M_\odot)\)

agn_logedd

\(-1.0\)

\([-2, -1]\)

\(\log_{10}(L/L_{\rm Edd})\)

: MAPPINGS V photoionization parameters.

Parameter support

A template-backed backend interpolates over the axes of its grid, and the lookup clips outside them. A value past the edge is therefore bit-identical to the edge node, with a zero gradient, so the parameter silently does nothing. The build applies four rules to every installed grid:

  1. A Fixed value outside the backend’s support is refused at build, naming the parameter, value, component and extent.

  2. A FREE parameter whose prior is declared by the registry resolves to the selected backend’s support: the prior is the backend’s full extent, even where that is wider than the declaration. The parameter summary reads [resolved to <backend> grid: lo..hi]. The declaration remains the prior only for a backend with no registered support.

  3. An explicit prior that extends beyond the support is refused, naming the extent.

  4. An explicit prior inside the support is unchanged.

Cue is the exception. It is a neural emulator, so it extrapolates past its trained range rather than clipping. Out-of-footprint values warn and never raise.

Gas-phase and stellar metallicity are independent knobs. neb={'type': 'cloudy', 'logZ_gas': FREE} infers the gas-phase metallicity, and its prior spans the grid’s neb_logZ_gas support whatever met_logzsol is.

The installed extents, read from the grid files by grid_support('neb', <backend>):

Backend

Data file

neb_logU

neb_logZ_gas

cloudy

cloudy_grid_{mist,prsc,pdva,bpss}.h5

[-4, -1]

[-1.98, 0.2]

cb19

cb19_templates.h5

[-4, -1.5]

[-3.84, -1.36]

mappings

flury2024_grids.h5 (sb99/cpr)

[-4, -0.5]

not registered (see below)

mappings_agn

flury2024_grids.h5 (agn_oxaf/cpr)

[-3.5, -0.5]

not registered (see below)

cue

cue_weights.npz (trained range, Li et al. 2025, Table 1)

[-4, -1] (extrapolating)

[-2.2, 0.5] (extrapolating)

neb_logZ_gas is log10(Z/Zsun). The cb19 extent is its 12+log(O/H) axis converted with the backend’s solar oxygen abundance. The CB19 nebular grid ships without data, so a cb19 build refuses before any lookup (see the known limitations page); the extents above are its axes. The MAPPINGS backends take no gas-metallicity key, so neb_logZ_gas has no knob there.

Differences from CIGALE’s Nebular Module

  • Density. CIGALE’s ne selects one of three tabulated electron densities (10, 100 and 1000 cm\(^{-3}\)). tengri’s density axes are hydrogen densities: gas_logn (\(\log_{10}n_{\rm H}\); the Cue gas parameters above give its range) for the cue backend and neb_log_nH (grid range 1 to 4) for cb19. The cloudy grid backend carries no density axis.

  • Emission-line profile. The default tengri profile is a compact-support triweight kernel of velocity dispersion neb_eline_sigma_kms \(= 100\) km s\(^{-1}\), whose FWHM is \(2\,(3\sigma)\sqrt{1 - 2^{-1/3}} \approx 272\) km s\(^{-1}\). CIGALE broadens each line with a Gaussian of FWHM lines_width \(= 300\) km s\(^{-1}\) by default.

References

Alarie, A., and C. Morisset. 2019. “Extensive Online Shock Model Database.” 55 (October): 377–94. https://doi.org/10.22201/ia.01851101p.2019.55.02.21.

Allen, Mark G., Brent A. Groves, Michael A. Dopita, Ralph S. Sutherland, and Lisa J. Kewley. 2008. “The MAPPINGS III Library of Fast Radiative Shock Models.” 178 (1): 20–55. https://doi.org/10.1086/589652.

Alsing, Justin, Hiranya Peiris, Joel Leja, et al. 2020. “SPECULATOR: Emulating Stellar Population Synthesis for Fast and Accurate Galaxy Spectra and Photometry.” 249 (1): 5. https://doi.org/10.3847/1538-4365/ab917f.

Asplund, Martin, Nicolas Grevesse, A. Jacques Sauval, and Pat Scott. 2009. “The Chemical Composition of the Sun.” 47: 481–522. https://doi.org/10.1146/annurev.astro.46.060407.145222.

Baldwin, J. A., M. M. Phillips, and R. Terlevich. 1981. “Classification parameters for the emission-line spectra of extragalactic objects.” 93 (February): 5–19. https://doi.org/10.1086/130766.

Byler, Nell, Julianne J. Dalcanton, Charlie Conroy, and Benjamin D. Johnson. 2017. “Nebular Continuum and Line Emission in Stellar Population Synthesis Models.” 840 (1): 44. https://doi.org/10.3847/1538-4357/aa6c66.

Feltre, A., S. Charlot, and J. Gutkin. 2016. “Nuclear activity versus star formation: emission-line diagnostics at ultraviolet and optical wavelengths.” 456 (3): 3354–74. https://doi.org/10.1093/mnras/stv2794.

Flury, Sophia R. et al. 2024. “MAPPINGS V photoionization grids for stellar and AGN-ionized nebular emission.”

Gutkin, Julia, Stéphane Charlot, and Gustavo Bruzual. 2016. “Modelling the nebular emission from primeval to present-day star-forming galaxies.” 462 (2): 1757–74. https://doi.org/10.1093/mnras/stw1716.

Haffner, L. M., R.-J. Dettmar, J. E. Beckman, et al. 2009. “The warm ionized medium in spiral galaxies.” Reviews of Modern Physics 81: 969–1014. https://doi.org/10.1103/RevModPhys.81.969.

Kewley, L. J., M. A. Dopita, R. S. Sutherland, C. A. Heisler, and J. Trevena. 2001. “Theoretical Modeling of Starburst Galaxies.” 556 (1): 121–40. https://doi.org/10.1086/321545.

Kewley, Lisa J., David C. Nicholls, and Ralph S. Sutherland. 2019. “Understanding Galaxy Evolution Through Emission Lines.” 57: 511–70. https://doi.org/10.1146/annurev-astro-081817-051832.

Li, Yongda, Joel Leja, Benjamin D. Johnson, and Sandro Tacchella. 2025. “Cue: A Fast, Flexible, and Accurate Neural Emulator for Nebular Emission.” https://arxiv.org/abs/2312.12345.

Nicholls, David C., Ralph S. Sutherland, Michael A. Dopita, Lisa J. Kewley, and Brent A. Groves. 2017. “Abundance scaling in stars, nebulae and galaxies.” 466: 4403–22. https://doi.org/10.1093/mnras/stw3235.

Osterbrock, Donald E., and Gary J. Ferland. 2006. Astrophysics of gaseous nebulae and active galactic nuclei. 2nd ed. University Science Books.

Sutherland, Ralph S., and Michael A. Dopita. 2017. “Effects of Preionization in Radiative Shocks. I. Self-consistent Models.” 229 (2): 34. https://doi.org/10.3847/1538-4365/aa6541.