NBursts: Full Spectrum Fitting Method

• The paper demonstrates that NBursts simultaneously retrieves stellar population parameters and internal kinematics by fitting full spectra.
• The methodology parameterizes star formation history as discrete bursts and integrates both spectroscopic and photometric data for improved accuracy.
• The technique effectively breaks age–metallicity and dust degeneracies, offering practical insights despite challenges in model consistency.

Searching arXiv for the specified NBursts paper and closely related full-spectrum fitting work. NBursts is a parametric full spectrum fitting technique designed to extract the internal kinematics and star formation histories of unresolved stellar systems by fitting stellar population models directly to spectroscopic data, and, in its spectro-photometric extension NBursts+phot, by fitting spectra and broadband spectral energy distributions simultaneously (Chilingarian et al., 2011). The method is organized around a pixel-space $\chi^2$ minimization in which the observed spectrum is represented as a linear combination of one or more simple stellar population components, convolved with a line-of-sight velocity distribution and supplemented by low-order polynomials to absorb residual continuum differences. Its defining feature is the simultaneous recovery of stellar population parameters and internal kinematics from the full information content of the data rather than selected spectral indices (Chilingarian et al., 2011).

1. Conceptual basis and scope

NBursts was introduced as a parametric full spectrum fitting method aimed at unresolved stellar systems such as galaxies and star clusters (Chilingarian et al., 2011). In its original form, the method starts from optical spectra alone and jointly recovers line-of-sight velocity distribution parameters—$v$, $\sigma$, and optionally higher Gauss–Hermite moments $h_3$ and $h_4$—together with ages, metallicities, and weights of one or more simple stellar population components that parametrize the star formation history (Chilingarian et al., 2011). The star formation history is therefore not reconstructed non-parametrically, but instead described as a small number of discrete bursts.

The method’s operating philosophy is to “fit everything at once” using the full spectrum in pixel space (Chilingarian et al., 2011). This places NBursts in the same broad methodological family as pPXF, STARLIGHT, ULySS, and STECKMAP, all of which fit full spectra with stellar population templates and line broadening functions, but differ in implementation details and in the treatment of star formation history and regularization (Chilingarian et al., 2011). A plausible implication is that NBursts is best understood not as a single inversion formula but as a design pattern: a parametric stellar population model, a LOSVD parameterization, and a global spectral fit carried out simultaneously.

NBursts+phot extends this framework by combining high-resolution spectra and broadband SEDs in a single minimization loop (Chilingarian et al., 2011). Its motivation is explicitly the breaking of classical degeneracies—age–metallicity, multiple-burst, and dust attenuation versus age—through the complementary information carried by spectral line structure and broadband continuum shape (Chilingarian et al., 2011).

2. Mathematical formulation

The core NBursts+phot objective function is the sum of a spectral term and a photometric term, with a tunable scalar $\alpha$ controlling their relative weight (Chilingarian et al., 2011):

$\chi^2 = \sum_{N_{\lambda} } \frac{ \left(F_{i} - P_{1p} \left[ T_{i}(\mathrm{SFH}) \otimes \mathcal{L}(v,\sigma,h_3,h_4) + P_{2q} \right] \right)^2 }{\Delta F_{i}^2} + \alpha \sum_{N_{ph} \frac{ \left( f_{j} - w_{j}\, t_{j}(\mathrm{SFH})\, A_{j} \right)^2 }{\Delta f_{j}^2}.$

Here the spectral model is generated from a parametric star formation history written as a sum of $N_{\mathrm{bursts}}$ SSPs (Chilingarian et al., 2011):

$$T_{i}(\mathrm{SFH}) = \sum_{n=1}^{N_{\mathrm{bursts}}} k_n\,T_i(t_n, Z_n).$$

The fitted spectrum is then the SSP mixture convolved with the LOSVD $\mathcal{L}(v,\sigma,h_3,h_4)$ and modified by multiplicative and additive Legendre polynomials $v$0 and $v$1 to absorb residual continuum-shape mismatches (Chilingarian et al., 2011). In conceptual form the model is

$v$2

The photometric term uses the same star formation history to predict fluxes in broadband filters (Chilingarian et al., 2011):

$v$3

with attenuation applied as

$v$4

The parameter $v$5 is therefore explicit in NBursts+phot but absent from pure spectroscopy-only NBursts (Chilingarian et al., 2011). The scalar $v$6 is also central: $v$7 reduces the method to pure NBursts, whereas larger values increase the leverage of the photometric data (Chilingarian et al., 2011).

The optimization is a non-linear least-squares minimization in which ages, metallicities, LOSVD parameters, extinction, and polynomial coefficients are non-linear variables, while the SSP weights $v$8 and photometric normalization coefficients $v$9 may be handled partly linearly (Chilingarian et al., 2011). The short paper does not describe explicit regularization; instead, the use of a small number of bursts acts as a strong implicit regularization (Chilingarian et al., 2011).

3. Stellar population parameterization and recovered quantities

The defining stellar-population assumption in NBursts is a low-dimensional star formation history represented as a sum of a small number of SSPs (Chilingarian et al., 2011). In practice, $\sigma$0 is small; the example application in the conference paper uses two components, corresponding to an older population and a recent burst (Chilingarian et al., 2011). This parametric structure makes the method well suited to simple or moderately complex histories, such as one old plus one young component, but not to detailed non-parametric age distributions (Chilingarian et al., 2011).

The recovered parameters can be grouped into four classes (Chilingarian et al., 2011):

Class	Parameters	Role
Stellar populations	$\sigma$1, $\sigma$2, $\sigma$3	Ages, metallicities, and component weights
Kinematics	$\sigma$4, $\sigma$5, $\sigma$6, $\sigma$7	LOSVD characterization
Dust attenuation	$\sigma$8	Photometric attenuation parameter in NBursts+phot
Continuum and normalization	$\sigma$9, $h_3$0, $h_3$1	Spectral mismatch absorption and spectro-photometric scaling

In spectroscopy-only mode, the absence of an explicit dust term means that continuum mismatches are handled by multiplicative and additive polynomials rather than by a physically parameterized attenuation law (Chilingarian et al., 2011). This is methodologically important because it separates pure NBursts from NBursts+phot: the former uses the spectrum to recover kinematics and a coarse parametric SFH, while the latter additionally attempts to recover internal extinction consistently with the stellar population solution (Chilingarian et al., 2011).

A plausible implication is that the meaning of the fitted weights depends on template normalization. The paper states that the $h_3$2 represent stellar mass or light fractions depending on how the SSP library is normalized (Chilingarian et al., 2011). That distinction matters for interpretation of burst strengths, especially when comparing results across different model libraries.

4. Data requirements and model ingredients

NBursts was developed for high-resolution optical spectra, with the conference-paper example based on SDSS DR7 spectroscopy at resolution $h_3$3 and wavelength coverage $h_3$4 (Chilingarian et al., 2011). Reasonably high signal-to-noise per pixel is assumed so that full spectrum fitting is meaningful (Chilingarian et al., 2011). Data preparation typically includes wavelength calibration, rebinning to a log-$h_3$5 scale, matching of the instrumental line-spread function, and masking of bad pixels, sky residuals, and strong emission lines (Chilingarian et al., 2011).

NBursts+phot supplements these spectra with broadband photometry. The example in the paper uses one GALEX NUV band and SDSS $h_3$6 photometry (Chilingarian et al., 2011). The photometric measurements must be $h_3$7-corrected to the rest frame and measured in an aperture corresponding closely to the spectroscopic aperture, such as the $h_3$8 SDSS fiber, so that the spectrum and photometry refer to the same physical region of the galaxy (Chilingarian et al., 2011). The method is described as flexible and can in principle incorporate NIR photometry or other bands provided appropriate SSP SED models exist (Chilingarian et al., 2011).

A central technical caveat is that the spectral and photometric predictions come from different model families (Chilingarian et al., 2011). Spectral fitting uses PEGASE.HR with the ELODIE 3.1 empirical stellar library, whereas photometric modeling uses PEGASE.2 with the BaSeL synthetic library (Chilingarian et al., 2011). The paper explicitly notes that this mismatch means that the spectroscopic and photometric parts are not fully self-consistent and can introduce systematic biases (Chilingarian et al., 2011). This is one of the most important limitations of the early implementation.

Dust attenuation in NBursts+phot is parameterized by a single free parameter $h_3$9 and applied through band-dependent attenuation factors $h_4$0 (Chilingarian et al., 2011). The exact functional form of the attenuation law is not specified in the short paper, but the role of the parameter is clear: dust is constrained primarily by the broadband SED shape, while the spectrum anchors ages, metallicities, and kinematics (Chilingarian et al., 2011).

5. Degeneracy structure, precision, and practical limits

The principal scientific motivation for NBursts+phot is the breaking of degeneracies that remain strong in either spectroscopy-only or photometry-only analyses (Chilingarian et al., 2011). Spectra alone are affected by age–metallicity and multi-population degeneracies, while photometry alone is strongly affected by age–dust and age–metallicity degeneracies (Chilingarian et al., 2011). By fitting both simultaneously, NBursts+phot uses line features to constrain ages and metallicities and broadband SED shape—especially in the UV—to constrain dust and small young stellar components (Chilingarian et al., 2011).

Broader work on full-spectrum fitting provides quantitative context for the intrinsic precision of NBursts-style analyses. For mock SSPs fit over the full optical range $h_4$1, ages can be recovered to an overall precision of $h_4$2 for $h_4$3 and $h_4$4 when $h_4$5 per wavelength pixel (Asa'd et al., 2020). In the ranges $h_4$6 and $h_4$7, where AGB and RGB stars dominate significant parts of the light, the age uncertainty rises to about $h_4$8 dex (Asa'd et al., 2020). The same work shows that metallicity is generally less precisely recovered than age, and that systematic shifts between Padova- and MIST-based fits can reach $h_4$9–$\alpha$0 dex in favorable ranges and up to $\alpha$1–$\alpha$2 dex in problematic intervals (Asa'd et al., 2020).

Work on detecting age spreads with NBursts-style two-SSP fitting shows that the dominant component’s age is generally recovered to within $\alpha$3 dex up to $\alpha$4, but that detailed age spreads are difficult to recover reliably from integrated-light spectra (Asa'd et al., 2021). The derived age spreads are often larger than the true ones, especially for $\alpha$5 and high mass fractions of the younger component, and the success rate for recovering age gaps within $\alpha$6 dex is typically around $\alpha$7 for both $\alpha$8 and $\alpha$9 dex tests (Asa'd et al., 2021). This suggests that NBursts decompositions with multiple closely spaced bursts should be interpreted conservatively.

The treatment of kinematics introduces an additional technical limitation. Standard discrete-kernel LOSVD convolution becomes inaccurate when the intrinsic velocity dispersion is smaller than the velocity sampling, leading to undersampling artifacts (Cappellari, 2016). For full-spectrum fitting codes with Gauss–Hermite LOSVDs, an analytic Fourier transform approach avoids discretizing the narrow kernel in pixel space and provides accurate velocities regardless of $\chi^2 = \sum_{N_{\lambda} } \frac{ \left(F_{i} - P_{1p} \left[ T_{i}(\mathrm{SFH}) \otimes \mathcal{L}(v,\sigma,h_3,h_4) + P_{2q} \right] \right)^2 }{\Delta F_{i}^2} + \alpha \sum_{N_{ph} \frac{ \left( f_{j} - w_{j}\, t_{j}(\mathrm{SFH})\, A_{j} \right)^2 }{\Delta f_{j}^2}.$0 (Cappellari, 2016). A plausible implication is that NBursts implementations based on discretized LOSVD kernels can inherit low-$\chi^2 = \sum_{N_{\lambda} } \frac{ \left(F_{i} - P_{1p} \left[ T_{i}(\mathrm{SFH}) \otimes \mathcal{L}(v,\sigma,h_3,h_4) + P_{2q} \right] \right)^2 }{\Delta F_{i}^2} + \alpha \sum_{N_{ph} \frac{ \left( f_{j} - w_{j}\, t_{j}(\mathrm{SFH})\, A_{j} \right)^2 }{\Delta f_{j}^2}.$1 biases unless a Fourier-space analytic treatment is adopted.

6. Applications, comparisons, and limitations

The conference-paper demonstration of NBursts+phot is a post-starburst E+A galaxy, SDSS J230743.41+152558.4, fit with an SDSS optical spectrum and GALEX NUV plus SDSS $\chi^2 = \sum_{N_{\lambda} } \frac{ \left(F_{i} - P_{1p} \left[ T_{i}(\mathrm{SFH}) \otimes \mathcal{L}(v,\sigma,h_3,h_4) + P_{2q} \right] \right)^2 }{\Delta F_{i}^2} + \alpha \sum_{N_{ph} \frac{ \left( f_{j} - w_{j}\, t_{j}(\mathrm{SFH})\, A_{j} \right)^2 }{\Delta f_{j}^2}.$2 photometry (Chilingarian et al., 2011). The best-fitting model is decomposed into two SSP components representing an old population and a recent burst, with the UV photometry enforcing the presence and strength of the young component while the spectrum constrains its detailed age, metallicity, and kinematics (Chilingarian et al., 2011). The paper presents this as an illustration rather than as a full validation suite.

Comparative studies with other methodologies place NBursts within a wider full-spectrum fitting ecosystem. Relative to pPXF, STARLIGHT, ULySS, and STECKMAP, NBursts shares the full-spectrum, template-based philosophy but differs in its explicitly parametric treatment of the star formation history (Chilingarian et al., 2011). Relative to photometry-only SED fitting codes, NBursts+phot is distinctive in combining high-resolution spectra and broadband photometry while explicitly recovering LOSVD parameters (Chilingarian et al., 2011). More recent work such as SEW shifts the fitting from flux space to an equivalent-width spectrum, eliminating the need for a prior attenuation law and deriving the attenuation curve as an output (Lu et al., 19 Feb 2025). This suggests an alternative route for addressing continuum-shape and dust degeneracies, though it is methodologically distinct from NBursts’s polynomial-plus-flux-space strategy.

The limitations explicitly identified for NBursts+phot are substantial (Chilingarian et al., 2011). The relative weight $\chi^2 = \sum_{N_{\lambda} } \frac{ \left(F_{i} - P_{1p} \left[ T_{i}(\mathrm{SFH}) \otimes \mathcal{L}(v,\sigma,h_3,h_4) + P_{2q} \right] \right)^2 }{\Delta F_{i}^2} + \alpha \sum_{N_{ph} \frac{ \left( f_{j} - w_{j}\, t_{j}(\mathrm{SFH})\, A_{j} \right)^2 }{\Delta f_{j}^2}.$3 between spectroscopic and photometric terms is “not evident” and must be tuned for each dataset. The spectral and photometric models are not fully self-consistent because they are built from different stellar libraries and synthesis codes. Real galaxy spectra may contain emission lines that contaminate both the spectral fit and the broadband fluxes, while the stellar population models do not include nebular emission. Multi-survey photometry may be heterogeneous in apertures, zero-points, and systematics, making homogeneous spectro-photometric fitting difficult (Chilingarian et al., 2011).

Additional limitations arise from the general behavior of full-spectrum fitting. Stochastic IMF sampling can dominate the error budget in low-mass stellar clusters outside the blue optical, producing highly non-Gaussian and sometimes catastrophic age and metallicity errors even at infinite S/N (Goudfrooij et al., 2020). Comparisons between integrated-light full-spectrum fitting and CMD analyses show generally good agreement in age but significantly worse agreement in metallicity and reddening, and also show that eMSTO-like age spreads do not necessarily translate to detectable multi-age signatures in integrated-light spectra (Asa'd et al., 2022). Taken together, these results indicate that NBursts is robust for coarse age characterization and simultaneous kinematic recovery, but that its multi-component decompositions, metallicities, and dust estimates are sensitive to data quality, wavelength coverage, model consistency, and the intrinsic information limits of stellar population synthesis.