ROTFIT: Stellar Parameter Estimation Tool

• ROTFIT is a spectrum-matching code that derives effective temperature, surface gravity, metallicity, radial velocity, and projected rotational velocity from observed stellar spectra.
• It employs chi-square minimization and weighted template matching, using both empirical and synthetic libraries tailored to various spectral resolutions.
• The framework supports applications from activity and lithium diagnostics to binary detection and cluster studies, ensuring robust calibration against reference stars.

ROTFIT is a spectrum-matching code for deriving stellar atmospheric parameters and kinematic diagnostics from observed spectra. Across its documented implementations, it is used to determine effective temperature $T_{\rm eff}$, surface gravity $\log g$, metallicity $[\mathrm{Fe}/\mathrm{H}]$, radial velocity (RV or $V_r$), and projected rotational velocity $v\sin i$ by comparing continuum-normalized target spectra with libraries of real-star or synthetic templates that have been degraded to the instrumental resolution, rotationally broadened, shifted in velocity, and ranked by a goodness-of-fit statistic. The method has been applied to high-resolution echelle spectroscopy, X-Shooter spectra of young stellar objects, low-resolution LAMOST-Kepler spectra, and medium-resolution LAMOST MRS spectra of field stars and open clusters, often in conjunction with spectral subtraction for activity and lithium diagnostics (Frasca et al., 2016, Molenda-Zakowicz et al., 2010, Frasca et al., 2017, Frasca et al., 2022).

1. Origins, scope, and scientific role

ROTFIT is described as a spectrum-matching code originally developed by Frasca et al. (2003, 2006) and subsequently adapted to multiple data sets and resolutions (Frasca et al., 2016). In the Kepler asteroseismic context, it was used on high-resolution spectroscopic observations to derive $T_{\rm eff}$, $\log g$, $[\mathrm{Fe}/\mathrm{H}]$, and $v\sin i$ for F–K targets, alongside RVs obtained by cross-correlation (Molenda-Zakowicz et al., 2010). In surveys of young stars and young stellar objects, it was configured with synthetic BT-Settl libraries to infer $T_{\rm eff}$, $\log g$0, veiling, RV, and $\log g$1 from X-Shooter spectra (Frasca et al., 2017, Manara et al., 2017). In LAMOST-based work, dedicated implementations were developed for low-resolution LAMOST-Kepler spectra and later for medium-resolution dual-arm LAMOST MRS spectra (Frasca et al., 2016, Frasca et al., 2022).

The scientific role of ROTFIT is broader than parameter estimation alone. In the LAMOST-Kepler survey it supported the construction of a catalog containing atmospheric parameters, RV, and an estimate of $\log g$2, and for cool stars it also enabled measurement of H$\log g$3 and Ca II IRT chromospheric fluxes (Frasca et al., 2016). In the medium-resolution LAMOST analysis of Kepler targets, it was used to determine atmospheric parameters, activity level, lithium atmospheric content, and binarity, including the identification of double-lined binaries, stars with variable RVs, lithium-rich giants, and emission-line objects (Frasca et al., 2022). In the Pleiades and NGC 1647 studies, the same framework underpinned cluster RV distributions, metallicity estimates, lithium-based ages, and the detection of SB1 and SB2 systems (Frasca et al., 12 Apr 2025, Frasca et al., 13 Apr 2026).

A recurrent feature of ROTFIT is that it operates empirically when real-star template libraries are used and physically through model atmospheres when synthetic grids are used. This duality is explicit in the literature: some implementations rely on ELODIE, Indo–US, or FRESCO reference spectra, while others adopt BT-Settl synthetic spectra (Molenda-Zakowicz et al., 2010, Frasca et al., 2016, Manara et al., 2017). This suggests that the code is best understood as a flexible fitting framework rather than a single fixed pipeline.

2. Core fitting formalism

The central operation of ROTFIT is $\log g$4 minimization between an observed spectrum and a set of prepared templates. In the LAMOST MRS Kepler implementation, the goodness-of-fit statistic is defined as

$\log g$5

The target spectrum is continuum-normalized, shifted to the rest frame using an RV estimate from cross-correlation, and then compared with templates spanning parameter space; the best solutions are combined through weighted means, with weight $\log g$6 for the ten best template–rotation combinations in the LAMOST MRS Kepler analysis (Frasca et al., 2022).

The RV step is commonly handled through cross-correlation. In one formulation,

$\log g$7

where a Gaussian is fit to the CCF peak and the centroid is taken as the RV (Frasca et al., 2022). Closely related CCF formulations appear in the high-resolution Kepler asteroseismic work, the low-resolution LAMOST-Kepler pipeline, the Lupus X-Shooter analysis, and the RasTyc survey (Molenda-Zakowicz et al., 2010, Frasca et al., 2016, Frasca et al., 2017, Frasca et al., 2018).

Rotational broadening is incorporated by convolving each template with a Gray rotational profile. For the LAMOST MRS Kepler work, the profile is written schematically as

$\log g$8

with $\log g$9 (Frasca et al., 2022). In X-Shooter applications, the Gray kernel is used with linear limb darkening and $[\mathrm{Fe}/\mathrm{H}]$0 scanned over an explicit trial grid (Frasca et al., 2017, Frasca et al., 2020). In high-resolution echelle work, trial broadenings are applied order by order, and the final $[\mathrm{Fe}/\mathrm{H}]$1 is the weighted mean over orders or the value minimizing the summed $[\mathrm{Fe}/\mathrm{H}]$2 (Molenda-Zakowicz et al., 2010, Frasca et al., 2018).

Weighted combination schemes are a further defining element. In the Molenda-Żakowicz et al. Kepler asteroseismic implementation, per-order parameters are averaged using weights proportional to $[\mathrm{Fe}/\mathrm{H}]$3, where $[\mathrm{Fe}/\mathrm{H}]$4 is an information-content factor proportional to total line absorption in the order; the order-wise results are then combined into a global stellar value (Molenda-Zakowicz et al., 2010). In the RasTyc analysis, order weights are similarly tied to total line absorption and best-fit $[\mathrm{Fe}/\mathrm{H}]$5 (Frasca et al., 2018). In several LAMOST MRS studies, blue- and red-arm solutions are combined as weighted means with $[\mathrm{Fe}/\mathrm{H}]$6 (Frasca et al., 12 Apr 2025, Frasca et al., 13 Apr 2026).

3. Template libraries and observational regimes

ROTFIT has been deployed with both empirical and synthetic template libraries, and the choice of library is strongly tied to spectral type, wavelength coverage, and instrumental configuration.

For high-resolution spectroscopy of Kepler asteroseismic targets, two libraries were used in parallel: the ELODIE archive, degraded from $[\mathrm{Fe}/\mathrm{H}]$7 to $[\mathrm{Fe}/\mathrm{H}]$8, and a FRESCO library of 122 slowly rotating stars calibrated on the same instrumental setup (Molenda-Zakowicz et al., 2010). In the RasTyc survey, 270 high-S/N, slowly rotating ELODIE spectra with parameters from the PASTEL catalog were employed and convolved to match SARG, FRESCO, or AURELIE resolution (Frasca et al., 2018). The low-resolution LAMOST-Kepler pipeline instead used a wide and homogeneous collection of real-star spectra from the Indo–US library, degraded to $[\mathrm{Fe}/\mathrm{H}]$9 and re-sampled onto the LAMOST wavelength scale (Frasca et al., 2016).

For X-Shooter studies of pre-main sequence stars and young stellar objects, ROTFIT was configured with BT-Settl synthetic spectra. In the Lupus work, the grid spans solar metallicity, $V_r$0 to $V_r$1 in steps of $V_r$2, and $V_r$3 to $V_r$4 dex in steps of $V_r$5 dex, with selected UVB and VIS windows analyzed after masking accretion and chromospheric lines (Frasca et al., 2017). In the Class III X-Shooter template study, BT-Settl models were again adopted, degraded to match each arm, and analyzed segment by segment with veiling fixed to zero (Manara et al., 2017). For ISO-ChaI 52, the VIS arm alone was used, with BT-Settl templates at solar metallicity spanning $V_r$6–$V_r$7 and $V_r$8–$V_r$9 dex (Frasca et al., 2020).

The medium-resolution LAMOST MRS implementations are characterized by dual-arm fitting. In the Kepler-field analysis, the code was modified to accept blue-arm spectra covering 495–535 nm and red-arm spectra covering 630–680 nm at $v\sin i$0 (Frasca et al., 2022). The template library consists of 388 high-S/N, slowly rotating real-star spectra from ELODIE covering $v\sin i$1, $v\sin i$2, and $v\sin i$3 space, convolved to the MRS resolution and resampled onto the MRS wavelength grid (Frasca et al., 2022). The Pleiades work used the same 388 ELODIE spectra and a similar dual-arm strategy (Frasca et al., 12 Apr 2025). In NGC 1647, cool stars with $v\sin i$4 were fitted with the same 388 inactive ELODIE templates, whereas hot stars with $v\sin i$5 were fitted to a grid of synthetic BT-Settl models with $v\sin i$6–$v\sin i$7, $v\sin i$8–$v\sin i$9 dex, and $T_{\rm eff}$0 fixed at 0 (Frasca et al., 13 Apr 2026).

A consistent pre-processing step across these regimes is resolution matching by Gaussian convolution. One LAMOST MRS formulation gives

$T_{\rm eff}$1

for degrading ELODIE templates to $T_{\rm eff}$2 before resampling (Frasca et al., 2022). In the Pleiades analysis the same idea is written in velocity form, $T_{\rm eff}$3 (Frasca et al., 12 Apr 2025). In NGC 1647, the true LAMOST MRS resolving power was measured from Th-Ar lamps as $T_{\rm eff}$4 and $T_{\rm eff}$5, and all templates were convolved to these values before fitting (Frasca et al., 13 Apr 2026).

4. Derived quantities beyond atmospheric parameters

Although the core products of ROTFIT are atmospheric parameters, RV, and $T_{\rm eff}$6, several studies use the same fitted template as a photospheric reference for subtraction analyses.

In the low-resolution LAMOST-Kepler survey, for stars with $T_{\rm eff}$7 the nearest non-active template, shifted to the stellar RV and broadened, is subtracted from the target to derive residual H$T_{\rm eff}$8 emission and Ca II IRT fluxes (Frasca et al., 2016). The residual equivalent width is measured over defined integration windows and converted to surface flux through

$T_{\rm eff}$9

with the activity index

$\log g$0

This application led to the identification of 442 chromospherically active stars and one likely accreting object (Frasca et al., 2016).

In the LAMOST MRS Kepler-field work, H$\log g$1 and Li I 6708 equivalent widths were calculated for cool stars with $\log g$2 (Frasca et al., 2022). The analysis yielded 327 active stars on the basis of H$\log g$3 flux and detections of the Li I 6708 line for 1657 stars, both giants and stars on the main sequence (Frasca et al., 2022). Among the giants, 195 Li-rich stars were found, 161 of them reported there for the first time; among main-sequence stars, a discrete age classification was performed based on atmospheric lithium abundance and upper envelopes of a few open clusters (Frasca et al., 2022).

The same template-subtraction logic is used in cluster studies. In the Pleiades analysis, late-type stars with $\log g$4 were characterized through H$\log g$5 and Li I-6708 net equivalent widths by subtraction of inactive photospheric templates, enabling studies of activity, lithium depletion, and flares (Frasca et al., 12 Apr 2025). In NGC 1647, H$\log g$6 and Li I-6708 net equivalent widths were measured for solar-type and cooler stars, supporting inference of a lithium-based cluster age of $\log g$7 (Frasca et al., 13 Apr 2026).

In young-star spectroscopy, the photospheric template also anchors accretion and chromospheric diagnostics. In the Lupus X-Shooter study, the best-fit BT-Settl spectrum, rotationally broadened and veiled, was subtracted from the observed spectrum to recover net emission in H$\log g$8, H$\log g$9, Ca II K and IRT, and Na I D (Frasca et al., 2017). Residual line equivalent widths were converted to surface fluxes using model continuum fluxes,

$[\mathrm{Fe}/\mathrm{H}]$0

and these diagnostics were combined with $[\mathrm{Fe}/\mathrm{H}]$1, RV, and Li I 6708 equivalent width for membership assessment (Frasca et al., 2017).

5. Calibration, uncertainty estimation, and validation

Uncertainty estimation in ROTFIT depends on implementation, but several patterns recur: internal errors from the scatter among orders or segments, formal errors from the curvature of the $[\mathrm{Fe}/\mathrm{H}]$2 minimum, and external checks against independent surveys or blue–red arm comparisons.

For the Kepler asteroseismic analysis at $[\mathrm{Fe}/\mathrm{H}]$3, validation against ELODIE-based and FRESCO-based solutions showed agreement to within $[\mathrm{Fe}/\mathrm{H}]$4, $[\mathrm{Fe}/\mathrm{H}]$5 dex, and $[\mathrm{Fe}/\mathrm{H}]$6 dex, while independent synthetic-spectrum fitting for a subset agreed with ROTFIT to better than the adopted grid spacing of $[\mathrm{Fe}/\mathrm{H}]$7, $[\mathrm{Fe}/\mathrm{H}]$8 dex, and $[\mathrm{Fe}/\mathrm{H}]$9 (Molenda-Zakowicz et al., 2010). In the X-Shooter Class III analysis, typical internal uncertainties are quoted as $v\sin i$0–$v\sin i$1, $v\sin i$2–$v\sin i$3 dex, $v\sin i$4–$v\sin i$5, and $v\sin i$6–$v\sin i$7 (Manara et al., 2017). In the Lupus study, typical internal errors are $v\sin i$8, $v\sin i$9 dex, $T_{\rm eff}$0, and $T_{\rm eff}$1 (Frasca et al., 2017).

The low-resolution LAMOST-Kepler pipeline reported internal repeat-based precisions of $T_{\rm eff}$2, $T_{\rm eff}$3 dex, and $T_{\rm eff}$4 dex, but external accuracies against high-resolution literature values were about $T_{\rm eff}$5 in RV, $T_{\rm eff}$6 in $T_{\rm eff}$7, $T_{\rm eff}$8 dex in $T_{\rm eff}$9, and $\log g$00 dex in $\log g$01 (Frasca et al., 2016). This distinction between internal consistency and external accuracy is central to interpreting ROTFIT outputs in low-resolution surveys.

The medium-resolution LAMOST MRS analyses introduced explicit blue–red arm cross-validation. In the Kepler-field study, blue–red differences of $\log g$02 in $\log g$03, $\log g$04 in $\log g$05, $\log g$06 in $\log g$07, $\log g$08 in $\log g$09, and $\log g$10 in RV imply per-arm uncertainties of about $\log g$11, $\log g$12 dex, $\log g$13 dex, $\log g$14, and $\log g$15, respectively (Frasca et al., 2022). Comparison with APOGEE DR16 for about 2500 stars gave rms dispersions of about $\log g$16 in $\log g$17, $\log g$18 dex in $\log g$19, and $\log g$20 in RV including genuine RV variables, or $\log g$21 if variables were removed, corresponding to an accuracy of about $\log g$22 on a clean sample (Frasca et al., 2022).

The Pleiades study used the rms of blue–red differences divided by $\log g$23 as an external empirical error estimate and found $\log g$24, $\log g$25 for $\log g$26, $\log g$27 dex, $\log g$28 dex, and $\log g$29–$\log g$30 for $\log g$31 (Frasca et al., 12 Apr 2025). In NGC 1647, blue–red arm comparisons yielded rms $\log g$32 overall, or about $\log g$33 for $\log g$34, corresponding to a single-arm precision of about $\log g$35; for cool stars, the rms scatters were about $\log g$36 in $\log g$37, $\log g$38 dex in $\log g$39, $\log g$40 dex in $\log g$41, and $\log g$42 in $\log g$43, reduced to about $\log g$44 for $\log g$45 (Frasca et al., 13 Apr 2026).

RV zero-point calibration is especially important in LAMOST MRS work. In the Kepler-field analysis, plate-by-plate zero-point offsets of up to about $\log g$46 before May 2018 were corrected by comparison to Gaia RVs of non-variable stars on the same plate (Frasca et al., 2022). The Pleiades work states that early MRS runs before May 2018 were calibrated with Sc lamps and later ones with Th–Ar, causing $\log g$47 RV offsets; corrections were applied plate by plate and arm by arm using contemporaneous RV standards or Gaia DR3 stars with $\log g$48 (Frasca et al., 12 Apr 2025). NGC 1647 likewise required correction for nightly wavelength-calibration zero-point shifts using Th-Ar lamp runs or DR12 recipes (Frasca et al., 13 Apr 2026).

6. Resolution limits, biases, and methodological constraints

A major practical limitation of ROTFIT is the resolving-power floor on $\log g$49. In the medium-resolution LAMOST MRS Kepler analysis, Monte Carlo tests with BT-Settl spectra at S/N = 50–100 showed that $\log g$50 cannot be reliably resolved, and values below that threshold are reported as “$\log g$51” (Frasca et al., 2022). The Pleiades and NGC 1647 studies adopt the same lower limit for LAMOST MRS (Frasca et al., 12 Apr 2025, Frasca et al., 13 Apr 2026). In X-Shooter work, the minimum detectable $\log g$52 is about $\log g$53 for the VIS slit at $\log g$54 and about $\log g$55 at $\log g$56, as verified by Monte Carlo tests (Frasca et al., 2017). At low LAMOST resolution, projected rotation is only robustly measurable for $\log g$57 (Frasca et al., 2016).

Template-grid structure also imposes systematic effects. The LAMOST MRS Kepler study notes that template-grid non-uniformity leads to clustering of returned parameters at nodes, especially at low $\log g$58 and $\log g$59, and that no interpolation or regularization between templates produces discrete jumps in output (Frasca et al., 2022). The Pleiades paper describes analogous “boxy” behavior in parameter space because ROTFIT picks the single best non-interpolated template, producing clusters of solutions at discrete grid points (Frasca et al., 12 Apr 2025). In the NGC 1647 analysis, the hot-star template grid is restricted to solar metallicity and coarse $\log g$60 steps, and early-type results carry larger uncertainties (Frasca et al., 13 Apr 2026).

Metallicity calibration is another recurrent issue. In the low-resolution LAMOST-Kepler survey, comparison with APOKASC and SAGA revealed systematic biases, leading to the relation

$\log g$61

while the raw ROTFIT values were retained in the catalog with advice that users apply the correction as needed (Frasca et al., 2016). In the medium-resolution LAMOST MRS Kepler analysis, comparison with APOGEE showed overestimation at $\log g$62 and underestimation at $\log g$63, leading to

$\log g$64

The same study states that better template coverage at $\log g$65 and $\log g$66 dex would help (Frasca et al., 2022).

Other limitations are regime-specific. In the X-Shooter Class III analysis, BT-Settl synthetic spectra are said to show small mismatches in certain molecular bands, notably TiO and VO, at $\log g$67–4.0; this introduces systematic uncertainties of about $\log g$68 in $\log g$69 and about $\log g$70 dex in $\log g$71 for the coolest objects beyond the quoted statistical errors (Manara et al., 2017). In the Lupus implementation, veiling below $\log g$72 is not significant, and under-luminous objects with edge-on disks require special treatment when placed on the HR diagram (Frasca et al., 2017). In NGC 1647, double-lined spectroscopic binaries are not handled by ROTFIT in its single-template mode and require dedicated SB2 analysis after identification through double peaks in the CCF (Frasca et al., 13 Apr 2026).

7. Representative applications and evolution of the framework

ROTFIT has been used in diverse observational programs, and its scientific outputs illustrate the evolution of the code from stellar parameter estimation to broader survey analysis.

In Kepler-related spectroscopy, the high-resolution asteroseismic study derived RV, $\log g$73, $\log g$74, $\log g$75, and metallicity for 44 targets, discovered three double-lined spectroscopic binaries and two suspected single-lined systems, and emphasized the need for verification and refinement of atmospheric parameters in the Kepler Input Catalog (Molenda-Zakowicz et al., 2010). The later low-resolution LAMOST-Kepler application produced atmospheric parameters and RVs for 61,753 spectra of 51,385 stars, identified RV variables, ultrafast rotators, and emission-line objects, and found 442 chromospherically active stars (Frasca et al., 2016). The medium-resolution LAMOST Kepler-field analysis then extended the same line of work to 16,300 spectra, deriving RV and atmospheric parameters for 14,300 spectra of 7443 stars and reporting 327 active stars, 1657 stars with Li I 6708 detections, and a fraction of Li-rich giants of about 4%, higher than expected (Frasca et al., 2022).

In young-star studies, ROTFIT was integrated with X-Shooter analyses of pre-main sequence and accreting systems. The Lupus survey used it to determine $\log g$76, $\log g$77, veiling, RV, and $\log g$78 for 102 YSO candidates, rejecting 13 as non-members primarily through discrepant RV or very low $\log g$79 (Frasca et al., 2017). The Class III template work used ROTFIT-derived stellar parameters as part of a library of reduced, flux-calibrated, and telluric-corrected spectra for disk-less pre-main sequence stars (Manara et al., 2017). In the ISO-ChaI 52 study, ROTFIT on the X-Shooter VIS spectrum delivered $\log g$80, $\log g$81 dex, $\log g$82, $\log g$83, and veiling $\log g$84 in the red VIS, anchoring the subsequent SED and disk-warp analysis (Frasca et al., 2020).

In open-cluster spectroscopy with LAMOST MRS, the framework became closely tied to rotation, activity, lithium depletion, and binarity. The Pleiades analysis applied ROTFIT to 1581 spectra of 283 stars, finding an RV distribution that peaks at $\log g$85 with a dispersion of $\log g$86, an average metallicity of $\log g$87, 39 possible SB1 and ten SB2 systems, and a lithium-isochrone age of $\log g$88 (Frasca et al., 12 Apr 2025). The NGC 1647 work used ROTFIT for 158 cluster members, found an RV distribution peaking at $\log g$89 with a dispersion of $\log g$90, an average metallicity of $\log g$91 dex, four double-lined spectroscopic systems, and a lithium-based age of $\log g$92 (Frasca et al., 13 Apr 2026).

Taken together, these implementations show a stable methodological core—cross-correlation for RV, rotational broadening, $\log g$93-based template matching, and weighted parameter aggregation—combined with instrument-specific calibration, library selection, and post-fit spectral subtraction. A plausible implication is that ROTFIT’s enduring utility derives less from a single fixed parameterization than from its adaptability to different resolutions, spectral types, and survey objectives, provided that template coverage, instrumental resolution, and zero-point calibration are explicitly controlled.