WorksPhysics and Astronomy
Empirical Brightening Model for Distant Long-Period Comets
Abstract
Recent studies have shown the traditional comet brightening model, , with a constant activity index (i.e., single slope in ), fails to accurately describe heliocentric lightcurves at large distances, where water-ice sublimation is inefficient. We present a new empirical model, , for distant comet brightening in which the effective brightening slope varies with heliocentric distance, . Using a multi-year pre-perihelion dataset from uniform instrumentation for 23 comets, we found that provides the best match to long inbound arcs spanning a wide range in . The best-fit coefficients and cluster tightly for our best-observed comets, suggesting a common brightening behavior; we therefore also tested one- and two-parameter variants in which and/or are fixed to anchor values derived from these comets (, ), either individually, jointly, or via a linear relation. The three-parameter model is strongly favored by the Bayesian information criterion (BIC) for the full lightcurve, justifying its added complexity for characterization. We also tested predictive performance against . For brightness forecasting from limited initial data ( au, the heliocentric distance range spanned by observations), itself predicts poorly, whereas a one-parameter reduction, with both coefficients fixed, predicts more reliably, including outperforming with a fixed . Two-parameter reductions with only one coefficient fixed typically reach predictions within 1 mag of the observed brightness after only au of inbound coverage.
Typeset using LATEX twocolumn style in AASTeX631
(The LCO Outbursting Objects Key (LOOK) Project)
Corresponding author: Carrie E. Holt
LSST-DA Catalyst Postdoctoral Fellow
Introduction
Long-period comets (LPCs) are some of the most pristine objects in our solar system. Spending most of their lifetime in the Oort cloud until perturbed inward by passing stars or galactic tides, they are thought to have undergone minimal processing, retaining much of their original volatile inventory. Therefore, these objects are valuable probes for connecting observed cometary properties to conditions in the pre-solar nebula and its subsequent evolution.
Thanks to powerful sky surveys, comets today are routinely discovered at heliocentric distances () greater than 5 au, and discoveries of comets at au are increasing (see Figure 1 in Lister et al. [33]). These objects have a visible coma, so we know they are active in some manner. However, the drivers of said activity are not well understood. is the most abundant volatile in most comets and the main driver of cometary activity at au, but water-ice sublimation is inefficient at au ([39]). Distant activity is likely driven by more volatile ices like CO and , which are next in abundance and can lift grains at heliocentric distances of 6–8 au for and tens of au for CO ([39, 25]). Indeed, CO and have been detected in comets and Centaurs beyond 5 au (e.g., [65, 17, 56, 47, 50]). The one-time transition from amorphous to crystalline water-ice has also been suggested to drive activity from au up to au (e.g., [55, 22, 45]), and has been argued as an explanation for Centaur activity (cf. [16] (Jewitt 2009)).
While spectroscopy and narrowband imaging provide more direct constraints on activity drivers, they generally require higher signal-to-noise per exposure and more specialized setups (e.g., can only be directly measured using space-based observatories), making them less practical for sustained, high-cadence monitoring of large samples of faint distant comets. Broadband optical photometry is the most accessible means of investigating distant cometary activity for large samples of targets, allowing us to monitor brightness changes as comets approach the Sun and relate those changes to potential activity drivers. Broadband photometric monitoring observations can be carried out on small (1–2m), ground-based telescopes, which can allocate substantially more observing time across many nights than is feasible with larger-aperture telescopes. Detailed physical models can be used to connect the observed broadband brightness of a comet to the rate of outgassing of the driving species (e.g. [39]), but these require knowledge of (or more often assumptions on) a wide range of parameters, including nucleus size, dust-to-gas ratios, dust size distribution and reflectance properties. More commonly, the evolution in brightness of comets is described using empirical relationships, that allow comparison of relative brightness corrected for observing geometry, but are agnostic to the detailed physics of the activity process. The slope of the brightening rate has been used to characterize LPCs since the early development of the Oort-cloud concept ([43, 63, 1]).
The Traditional Model
Characterizing the brightness evolution of comets captured by broadband photometry requires a functional form relating apparent brightness to heliocentric and geocentric distance. A commonly used reference normalization to ensure the most accurate comparisons for photometric magnitudes is , representing the theoretical brightness for a Sun-based observer with the comet at au (thus with a comet-observer distance of 1 au and a phase angle of ). All values in this work are computed in -band. Hereafter we refer to this as . Traditionally for comets, a parameter , also known as the activity index, has been used to characterize the brightening rate of LPCs ([12]), so that the apparent magnitude can be described as:
We refer to Equation (1) as the traditional model () with two free parameters ( and ), where is the geocentric distance in au; the coefficient 5 reflects the dependence of flux on distance (). is the heliocentric distance in au, is the phase angle (the Sun-Target-Observer angle), and is the phase function, which is the ratio of the brightness at phase angle to that at . To calculate , we used the Schleicher-Marcus phase function for cometary dust1, which is a composite of the phase function from comet Halley ([53]) for smaller phase angles and the phase function derived for near-Sun comets in Marcus [35] at mid- to large-phase angles ([52]).
Today, Equation (1) is the most commonly used model for brightening rate characterization and predicting future brightness. Commonly used resources for comet magnitude estimates, including JPL Horizons2 and the Minor Planet Center, adopt this model, fitting the data using least squares. In JPL brightness calculations, the absolute magnitude () and magnitude slope parameter () correspond to Equation (1) as and . These values are for the total magnitude, including the coma. For some comets, JPL brightness calculations may also include and parameters, which correspond to the magnitude and the slope of the central condensation (or pseudo-nucleus) alone.
A Fixed-Aperture Scaling Correction
A fixed angular photometric aperture corresponds to a smaller physical radius at the comet as it approaches the Earth (smaller ), so a growing fraction of the coma falls outside the aperture. This partially offsets the inverse-square brightening a point source would show, so a scaling is more appropriate than the point-source scaling in Equation (1) ([44, 23]). Because our photometry uses a fixed 5″ aperture, we adopted in place of the term for all model fits in this work:
We discuss the limitations of this choice in § 6.5.
Limitations of the Traditional Model
Within 3 au, the model has also been used to compare brightening behavior across dynamical classes of LPCs, particularly between dynamically new comets (DNCs) and returning comets. DNCs are a subset of LPCs distinguished by their original orbital energy, proportional to , the inverse of the original semi-major axis calculated with respect to the solar system barycenter, found by integrating the orbit backward until planetary perturbations become negligible ( au). The distribution of shows an excess of comets with au, a “spike” that motivated the postulation of the Oort cloud ([42, 43]). However, there is no definitive threshold value of separating new from returning comets, and recent work ([11, 10]) has shown that not all comets in the Oort spike are truly dynamically new (and therefore compositionally pristine). In practice, a common threshold of has been proposed to identify DNCs ([30]), though the exact value varies between studies.
Historically, DNCs have been observed to have shallower brightening slopes than returning comets ([63]), a result recently confirmed by Lacerda et al. [31] using a large collection of Minor Planet Center photometry. To account for heterogeneity between observers and instruments, Lacerda et al. [31] grouped
observations into “strands,” consecutive measurements from a single observatory and filter, and applied quality cuts before fitting each strand independently, isolating relative brightening trends from absolute calibration differences between sources. Lacerda et al. [31] found median values of of 6.7, 7.8, and 13.4 () for new, intermediate, and old LPCs (i.e., values of 1, 20, and ), respectively, on their inbound legs where au. With modern telescopes and sky surveys now routinely discovering comets beyond au, where water-ice sublimation becomes inefficient and more volatile species are thought to drive activity, it has become possible to test whether the traditional model remains an adequate description at these distances. Previous broadband photometric studies provided early insight into this regime, but either lacked comprehensive heliocentric distance coverage ([49, 40]) or focused on individual objects ([64]).
Recent multi-object, multi-epoch studies have begun to characterize the brightening behavior of LPCs beyond au. Holt et al. [19] used uniformly processed broadband photometry of 21 LPCs, finding that systematically decreases with decreasing for individual comets and across the population as a whole (Figure 1; discussed further in § 3.1), with no clear difference between DNCs and returning comets, though the sample included only three returning comets. Lacerda et al. [31] further showed that the brightening slope of new and intermediate comets varies significantly at au, while old comets show no such trend. It is worth noting that the two studies used different thresholds to define DNCs: Holt et al. [19] adopted the threshold of from Królikowska & Dybczyński [30] (which we also adopt here), while Lacerda et al. [31] followed the Oort & Schmidt grouping with for new comets ([43]). Across both studies, a constant proves insufficient to describe the full inbound lightcurve, and post-perihelion behavior at large distances ( au) remains poorly characterized.
Figure 1. Activity index for long-period comets, where each color and marker corresponds to a different comet and has a changing value for piece-wise linear fits over different heliocentric ranges across a comet’s lightcurve, adapted from Fig. 6 of Holt et al. [19]. The horizontal error bar for each point corresponds to the heliocentric distance range over which the was fit and the vertical error bar corresponds to the uncertainty in the fit, and errorbars not visible are smaller than the plotted points. The black line, discussed further in § 3.1, is the effective slope of the three-parameter model (; Equation (3)) using the fixed reference values and as and . The light green dotted line shows the of the three-parameter model fit to comet C/2021 S3, whose multiple values are the light green points. Two lightcurve segments that fall outside of this trend, near within au (circled in red), belong to C/2019 L3 and C/2022 A2, which experienced a “surge” in brightness near perihelion.
This Work
The inadequacy of a constant at large heliocentric distances presents practical problems when using the model to predict the distant behavior of comets, such as attempting to assess the likely yield of new surveys (e.g., LSST, Donaldson et al., in prep; NEOSurveyor, Levine et al. [32]), or to extrapolate the observed brightness of a comet found at large distances to predict its perihelion activity level. The latter is important for the European Space Agency mission Comet Interceptor (Jones et al. [26]), which is scheduled to launch between August 2028 and July 2029, and will target a yet-to-be-discovered LPC. Based on initial observations of a comet discovered far from the Sun (ideally at au, to give sufficient warning time to have the spacecraft position itself for a flyby) the mission team must make predictions on the expected evolution of the comet’s activity level to select a good target: a comet with too high levels of dust production may be unsafe for the spacecraft, while a weakly outgassing comet may not provide enough material for in situ instruments to sample properly ([9, 57]).
A -dependent model is needed to improve accuracy in both characterization and predictions. In this paper, we extended the analysis of Holt et al. [19] using recent observations of comets discovered at au to develop the model, an empirical brightening model in which the activity index varies with heliocentric distance. We test how well the model fits the full data sets relative to the traditional power law model (), and whether the tight clustering of its best-fit parameters justifies simpler, constrained variants with fewer parameters. We then compare the predictive performance of all of these models from limited early data.
Data
The dataset used in this study is an extension of the one presented in Holt et al. [19]. Specifically, we added eight additional comets to the 15 LPCs from the original study with observational coverage spanning a range of heliocentric distance of at least 1 au ( au), while still pre-perihelion and beyond au, for a total of 23 comets (Table 1). We also extended the pre-perihelion data arcs for seven of the original 15 comets. Observational parameters and data reduction procedures remain unchanged from Holt et al. [19], where they are described in full.
| Comet Name | (au) | (au) | (au) | T | (mag) | (au) | ||
| C/2023 A3 (Tsuchinshan-ATLAS) | 7.25 | 2.83 | 4.69 | 0.39 | 2024 Sep 27 | 162 | 0.51 | 7 |
| C/2020 R7 (ATLAS) | 6.93 | 2.34 | 3.97 | 2.95 | 2022 Sep 15 | 185 | 0.51 | 10 |
| C/2021 T4 (Lemmon) | 6.73 | 2.30 | 3.81 | 1.48 | 2023 Jul 31 | 58 | 0.50 | 34 |
| C/2022 L2 (ATLAS) | 6.04 | 2.24 | 3.35 | 2.69 | 2024 Mar 11 | 117 | 0.51 | -1 |
| C/2022 E2 (ATLAS) | 8.08 | 2.20 | 4.42 | 3.67 | 2024 Sep 14 | 179 | 0.48 | 72 |
| C/2021 S3 (PANSTARRS) | 8.58 | 2.16 | 4.62 | 1.32 | 2024 Feb 14 | 140 | 0.56 | -1 |
| C/2023 C2 (ATLAS) | 5.97 | 2.14 | 3.18 | 2.37 | 2024 Nov 16 | 38 | 0.51 | 54 |
| C/2022 S4 (Lemmon) | 6.44 | 1.99 | 3.21 | 2.76 | 2024 Jul 18 | 126 | 0.52 | 73 |
| C/2021 G2 (ATLAS) | 9.42 | 1.89 | 4.43 | 4.98 | 2024 Sep 9 | 208 | 0.49 | 23 |
| C/2024 E1 (Wierzchos) | 7.71 | 1.82 | 3.47 | 0.57 | 2026 Jan 20 | 64 | 0.50 | 28 |
| C/2019 L3 (ATLAS) | 5.70 | 1.60 | 2.15 | 3.56 | 2022 Jan 10 | 77 | 0.48 | 26 |
| C/2022 QE78 (ATLAS) | 8.73 | 1.56 | 3.14 | 5.48 | 2025 Sep 12 | 57 | 0.49 | 15 |
| C/2022 A3 (ATLAS) | 5.88 | 1.54 | 2.07 | 3.71 | 2023 Sep 29 | 98 | 0.51 | 1148 |
| C/2023 R1 (PANSTARRS) | 8.11 | 1.47 | 2.61 | 3.57 | 2026 Apr 13 | 88 | 0.50 | 6 |
| C/2024 A1 (ATLAS) | 5.75 | 1.47 | 1.84 | 3.87 | 2025 Jun 12 | 87 | 0.52 | 332 |
| C/2022 E3 (ZTF) | 3.75 | 1.47 | 1.19 | 1.12 | 2023 Jan 12 | 24 | 0.54 | 713 |
| C/2023 T3 (Fuls) | 5.18 | 1.46 | 1.63 | 3.55 | 2025 Jan 25 | 53 | 0.50 | 1648 |
| C/2021 E3 (ZTF) | 4.46 | 1.45 | 1.37 | 1.78 | 2022 Jun 10 | 37 | 0.48 | -11 |
| C/2022 R6 (PANSTARRS) | 9.31 | 1.40 | 2.65 | 6.57 | 2025 Aug 27 | 177 | 0.50 | 21 |
| C/2021 O3 (PANSTARRS) | 4.18 | 1.39 | 1.17 | 0.29 | 2022 Apr 21 | 24 | 0.48 | -42 |
| C/2022 U3 (Bok) | 6.70 | 1.36 | 1.78 | 4.83 | 2024 Jul 28 | 53 | 0.56 | 25 |
| C/2022 T1 (Lemmon) | 4.68 | 1.36 | 1.24 | 3.44 | 2024 Feb 17 | 43 | 0.52 | 11 |
| C/2021 C5 (PANSTARRS) | 5.23 | 1.29 | 1.18 | 3.24 | 2023 Feb 11 | 46 | 0.47 | 16 |
Table 1. Comet Sample Sorted by
As in our previous work, we leverage Las Cumbres Observatory’s (LCO) global network [7] of 1 m robotic telescopes as part of the LCO Outbursting Objects Key (LOOK) Project [33]. This network provides uniform instrumentation across both hemispheres and enables frequent monitoring (here typically every three days) over extended periods (usually years). Observations were taken using the Sloan Digital Sky Survey (SDSS) and filters, photometrically calibrated to the Pan-STARRS1 system using calviacat [27] and the ATLAS-RefCat2 catalog [58], with color corrections derived from LOOK Project photometry; photometry calibrated to the PS1 system using AB magnitudes is hereafter denoted and . Since 2020, we have observed all comets discovered inbound at au with a heliocentric distance range of au before perihelion. Our strategy allows for detailed characterization of brightness evolution as comets approach the Sun from beyond au. Because the LCO network enables observations from both the northern and southern hemispheres, the only major observational gaps occur during solar conjunction.
For our data reduction, we applied the same image quality cuts described in Holt et al. [19]: excluding images with point-source full widths at half maximum (FWHM) greater than or photometric uncertainties exceeding 0.15 mag. Images were also manually inspected and removed from consideration if affected by background source contamination. Photometric calibration, including standard color corrections, followed the procedures established in that study as well. Finally, we adopted the same justification for using a photometric aperture, which provides the most consistent results for photometric monitoring across a broad range of heliocentric distances.
In this study, we focused on the pre-perihelion evolution and predictive modeling of cometary brightness. To isolate activity unrelated to water-ice sublimation, our analysis used only observations from au. This cutoff was motivated by the findings of Holt et al. [19], who showed that nearly all comets in the sample with perihelion distance au exhibited a sharp increase in brightening at au, likely due to the onset of water-driven activity. All comets in our sample were observed pre-perihelion through at least our 2025 May 1 cutoff; some objects had additional pre-perihelion data available beyond this date that are not included here. We reserve analysis of the au region, including post-perihelion data, for future work.
For our statistical tests, we first focused on the seven comets with the largest ratio of maximum to minimum heliocentric distance (), since a larger provides greater leverage for discriminating between models in logarithmic space. Because our analysis is restricted to observations at au, and a comet can never be observed pre-perihelion below its own perihelion distance , is bounded below by 2.5 au, or by itself for comets with au. The maximum possible for a given comet is therefore divided by whichever of these two limits is larger. A comet’s actual falls short of this ceiling whenever observations stopped before reaching this limit, for example due to a gap during solar conjunction. These seven high- comets have data starting at –9 au with . We then expanded our analysis to include the full sample of 23 comets. For testing the predictive power of the models, we used those same seven comets that have . Table 1 details the comets used for our analyses.
An Empirical Brightening Model
The Hab Model
Motivated by the findings of Holt et al. [19] and Lacerda et al. [31] that a constant fails to describe distant lightcurves, we replaced the brightening slope () in Equation (2) with a linear function of heliocentric distance (), where and are independent model parameters. The improvement is visible directly in the residuals: for comet C/2021 S3, one of our best-sampled objects, the Hab fit tracks the curvature of the observed lightcurve, while the single-slope fit does not (Figure 2).
Figure 2. Pre-perihelion heliocentric lightcurve of C/2021 S3 in -band. The upper panel shows the variation of pre-perihelion heliocentric magnitude with heliocentric distance using a 5″ aperture. The black points represent the and magnitudes, where the magnitudes are converted to the -band by shifting them by the median color of 0.56; observations in other bands, if available, are not included here. The lines represent the best fits for both the model (dashed blue) and model (solid purple), determined by the least-squares method. The lower panel shows the residuals for both model fits. The -axis is in log scale.
The term in this model should be interpreted as a heliocentric-distance-dependent activity index rather than as the instantaneous analogue of the constant- slope in the traditional model. We adopted this form because it provides a simple way to capture the observed curvature of distant lightcurves over a wide range of , while remaining comparable to the traditional constant- parameterization. In the traditional model, where is constant, the slope of the lightcurve with respect to is simply . In contrast, when the activity index varies with , differentiating the magnitude with respect to introduces additional terms from the -dependence. For comparison with the traditional model, we therefore define an effective slope for the Hab model:
The Hab model has three free parameters, compared to two for the traditional model (), which improves flexibility but can also lead to poorly constrained fits when the heliocentric distance ratio sampled, , is lim: heliocentric distance of the first observation with LOOK. : ratio of the largest to smallest heliocentric distance. : pre-perihelion heliocentric distance range with au. : perihelion distance. T: perihelion date. : number of binned observations as of 2025 May 1; observations were binned by night and by filter, so the number of visits is approximately . : median color using a 5″ aperture. : reciprocal original semi-major axis from the Nakano Notes (https://www.oaa.gr.jp/~oaacs/nk.htm), used as a proxy for dynamical age. The horizontal line separates the seven comets with (above) from the rest of the sample.
ited. The best-fit and parameters cluster in a relatively narrow region for the best-observed comets (Figure 3): the six comets with the largest are well described by a distribution centered on and , with standard deviations of 0.09 and 0.68, respectively. Hereafter we denote these fixed reference values, distinct from the free parameters and . These six comets are a subset of the seven comets used as the high-quality sample in this work (Table 1); the seventh, C/2023 C2, has the smallest (2.14) of the group and only 38 observations, and is retained in the broader sample used for the predictive tests in Section 5.2. These six comets also appear to have their own linear correlation, which we fit with . This clustering suggests that simplified variants with fewer free parameters may capture the general brightening behavior of most LPCs; in Section 3.2 we use , , and this correlation to construct one- and two-parameter variants of the model. Part of this correlation may reflect the natural covariance between and when fitting over a limited range rather than a purely physical relationship; we use it as an empirical constraint, and leave a more detailed covariance or pivot-point analysis with a larger sample to future work. These anchor values are robust to the choice of comets: recomputing them six times, each time excluding one of the six comets, gives to 0.43 and to , well within their spread, and – relations that remain within the population ellipse over the observed range of .

Figure 3. The distribution of and parameters with error bars for the model. Points are colored according to the ratio of the data used to fit the model. In the inset box, we highlight the 6 comets with the largest . The center of the distribution of the small group is marked by a red symbol, and the gray-shaded ellipse shows the standard deviation. A diagonal magenta line shows the linear regression of the small group.
Hab Model Variants
We constructed four constrained versions of the model (Table 2): a one-parameter model with both and fixed to and (Fixed-), two two-parameter models with only fixed to or only fixed to (Fixed- and Fixed-, respectively), and a two-parameter model in which is fixed as a linear function of (Correlated-).
| Name | Activity Term^a | Free Parameters | Description |
| , | Traditional model; no dependence | ||
| , , | Empirical 3-parameter model | ||
| Fixed- | Fully constrained: and are defined | ||
| Fixed- | , | Partially constrained: Only is defined | |
| Fixed- | , | Partially constrained: Only is defined | |
| Correlated- | , | Partially constrained: is correlated to | |
| \multicolumn{4}{l}{Replaces in Equation (2), but is not equivalent to (see , Equation (3)).} | |||
| \multicolumn{4}{l}{The four models below the horizontal line are fully or partially constrained variants of the unrestricted 3-parameter model, where and .} |
Table 2. Tested Models
The Fixed- model, in which both the slope () and intercept () of the heliocentric brightening term are fixed to and , has only a single free parameter, . This is analogous to the traditional model with a fixed (typically assumed to be –4), but with a distance-dependent brightening term in place of a constant slope.
The Fixed- and Fixed- models instead fix only to , or only to , respectively, leaving the other coefficient free along with . In the Correlated- model, we instead use the correlation found in the distribution of , parameters to constrain as a function of , such that the activity term becomes . These two-parameter models allow us to assess whether partial flexibility in the brightening behavior improves performance relative to the fully fixed Fixed- model, while still avoiding the potential overfitting risk of the full three-parameter model.
A summary of the parameters found for each comet for each model using a least-squares fit is shown in Table 5 (Appendix § 11). Figure 4 shows an example of a single comet lightcurve with all model fits and the complete figure set (23 images) is available in the online journal.
Figure 4. Pre-perihelion heliocentric lightcurve of C/2021 T4 in -band. The upper panel shows the variation of pre-perihelion heliocentric magnitude with heliocentric distance using a aperture; the -axis is in log scale, with a secondary axis along the top giving the corresponding ratio , where is the comet’s own most distant observed heliocentric distance. The black points represent the and magnitudes, where the magnitudes are shifted by the median color reported for each comet in Table 1. The colored lines show the best-fit for the , , Fixed-, Fixed-, Fixed-, and Correlated- models, determined by the least-squares method. The lower panel shows the residuals for each model. The complete figure set (23 images) is available in the online journal.
| Comet | Name | |||||||||||||
| C/2019 L3 | ||||||||||||||
| C/2020 R7 | ||||||||||||||
| C/2021 C5 | ||||||||||||||
| C/2021 E3 | ||||||||||||||
| C/2021 G2 | ||||||||||||||
| C/2021 O3 | ||||||||||||||
| C/2021 S3 | ||||||||||||||
| C/2021 T4 | ||||||||||||||
| C/2022 A3 | ||||||||||||||
| C/2022 E2 | ||||||||||||||
| C/2022 E3 | ||||||||||||||
| C/2022 L2 | ||||||||||||||
| C/2022 QE78 | ||||||||||||||
| C/2022 R6 |
Table 5. The values determined for each model using least-squares fit. All uncertainties are . Goodness-of-fit comparisons (RMSE and BIC) for these fits are presented in § 5.1 and Table 3.
Methods
Goodness of Fit
To identify which model best reproduces each comet’s full lightcurve, we applied the model, its variants, and the traditional model () to the complete dataset for that object with least-squares and no statistical outlier rejection, and evaluated performance with the root mean squared error (RMSE). This metric is the square root of the average squared difference between the observed and predicted magnitudes, providing an interpretable measure of model accuracy in the same units as the data. The residual sum of squares (RSS), which forms the basis of the RMSE calculation, is defined as:
where and are the observed and model-predicted magnitudes for each of the observations. The RMSE is then given by:
Lower RMSE values indicate a closer match between the model and the observed lightcurve. Photometric uncertainties were not incorporated into any of the RMSE calculations in this work, as they were generally at the 10 mmag level and did not significantly affect the results. By comparing RMSE values across models, we determined which model most accurately captured the overall brightness evolution for each comet. We note that frequent outbursts could bias this value with larger average residuals even with a generally good fit to the quiescent periods of activity. However, none of the comets in our sample had clear outbursts beyond au. We discuss outbursts, along with other possible causes of deviations of our measurements from model fits, in § 6.
While a lower RMSE indicates a better fit, it does not account for the fact that a model with more free parameters will naturally tend to fit the data more closely. To evaluate whether the added complexity of the model was warranted relative to the traditional model () and its own constrained variants, we compared their fits using the Bayesian Information Criterion (BIC; [54]), which penalizes models for additional free parameters while rewarding goodness of fit. BIC is defined as:
where is the number of free model parameters and RSS is the residual sum of squares defined above. This formulation penalizes additional parameters more strongly than other criteria and is particularly well-suited for identifying the most appropriate model when the goal is to characterize underlying physical behavior ([8]). For nested models such as these (each is a constrained form of ), is equivalent to the Likelihood Ratio Test (LRT) statistic with an additional penalty of . We adopted BIC because, unlike an LRT significance test, it can indicate support for the simpler model as well as the more complex one. For each object, we considered a model to be significantly favored if its BIC value was more than 10 lower than that of a simpler model, a threshold commonly interpreted as strong evidence in model selection ([46]); with the term, this requires a considerably larger improvement in fit than a conventional LRT significance test.
Predictive Tests
To evaluate the predictive performance of the model and its constrained variants relative to the model, we conducted a progressive fitting analysis by incrementally increasing the data used for each least-squares fit (i.e., decreasing ) in steps of (Figure 5). We use rather than a fixed arc length as the threshold variable because our models operate in (§ 3); a fixed ratio corresponds to a fixed span in log-space and therefore comparable leverage on the fit at any heliocentric distance, whereas a fixed would represent very different amounts of log-space coverage depending on where along the arc it falls. For example, for C/2021 S3 ( au), the threshold corresponds to au, i.e., a fit arc of au; for C/2023 C2 ( au), the same threshold corresponds to au, or au. At each step, the model was fit using only the data within the specified range, and we computed the root mean squared error against the observed data up to a common endpoint for every comet (, including both the fitted and predictive portions within that range), so that predictive accuracy is compared over an equivalent relative arc across all seven comets rather than each comet’s full, unevenly extended dataset. We also calculated the average magnitude difference between the predicted and observed values for the five data points nearest to, but not short of, this same endpoint (), providing a complementary measure of end-of-curve predictive performance at a single reference point rather than averaged over the capped arc. We consider a prediction successful once falls below 0.5 mag or falls below 1 mag. gets the tighter threshold because it averages over the whole curve, while comes from just five points.
Figure 5. An example of the strategy used for assessing predictive power of each of the models, for C/2023 A3 ( au). The blue circles are the real data points used for each model fitting; the gray open circles are the rest of the real data. Each model is fit to the blue points and extrapolated out to the end of the pre-perihelion data, represented by the colored lines. The left panel uses ( au) and the right panel uses ( au).
We initially interpolated the lightcurves onto evenly spaced points to control for uneven data sampling, but found no statistically significant difference in predictive performance between the real and interpolated datasets for any model except ; for , the real data produced higher residuals, indicating that interpolation would understate its predictive limitations. We therefore report results from the real data throughout.
Results
Best-Fit Model
Using the seven comets with , we found that the model provides the best fit to the full lightcurve in all seven cases. Here, each model was fit once to the complete pre-perihelion dataset for a given comet (§ 4.1), rather than fit progressively and evaluated on its ability to predict unseen data, which we address separately in § 5.2. The full ranking of all six models by RMSE for each comet is shown in Figure A1 (Appendix § 9). It is worth noting that all models, including the traditional model, had an RMSE (i.e., average residuals) of less than 0.4 mag when fit to the complete lightcurve. For the entire sample of 23 comets, was again the best full-lightcurve fit by RMSE in every case, and the full ranking breakdown is likewise shown in Figure A1. For all models, the RMSE did not exceed 0.55 mag.
Figure A1. Figure A1. RMSE (top) and RMSE rank (bottom) for each of the six models tested, for every comet in the sample. The seven comets with are separated from the remaining sample by the vertical dashed line.
Because is the clear best fit, we compared its BIC to that of the other models to assess whether the additional parameters are warranted. A would be strong evidence that the model is significantly favored. We first evaluated the subset of seven comets with , finding that is strongly preferred over in six of seven objects, over Fixed- in all seven, over Fixed- in five of seven, and over Fixed- and Correlated- in four of seven. Extending to the full sample of 23 comets, the model strongly outperforms all alternatives, with in 16–23 of 23 objects ( of the sample) across all pairwise comparisons. Correlated- is the closest competitor to across the full sample: is strongly preferred () in 16 of 23 comets, weakly preferred in 5, and disfavored relative to Correlated- in the remaining 2 (C/2021 S3, C/2021 G2). The values for each comet and model are shown in Figure A2 (Appendix A). Table 3 summarizes the output of the statistical tests for each comet and model.
| Comet Name | RSS | RMSE | BIC | RSS | RMSE | BIC | Fixed- RSS | Fixed- RMSE | Fixed- BIC | Fixed- RSS | Fixed- RMSE | Fixed- BIC | Fixed- RSS | Fixed- RMSE | Fixed- BIC | Correlated- RSS | Correlated- RMSE | Correlated- BIC |
| C/2023 A3 | 19.49 | 0.35 | 446 | 1.20 | 0.09 | 24.71 | 0.39 | 480 | 2.05 | 0.11 | 81 | 1.25 | 0.09 | 2 | 1.48 | 0.10 | 29 | |
| C/2020 R7 | 5.35 | 0.17 | 245 | 1.38 | 0.09 | 2.44 | 0.11 | 95 | 2.16 | 0.11 | 77 | 1.95 | 0.10 | 59 | 1.66 | 0.09 | 29 | |
| C/2021 T4 | 3.82 | 0.26 | 167 | 0.20 | 0.06 | 3.22 | 0.24 | 153 | 0.32 | 0.07 | 23 | 0.21 | 0.06 | -1 | 0.22 | 0.06 | 3 | |
| C/2022 L2 | 4.77 | 0.20 | 290 | 0.38 | 0.06 | 0.59 | 0.07 | 41 | 0.39 | 0.06 | -4 | 0.38 | 0.06 | -5 | 0.41 | 0.06 | 2 | |
| C/2022 E2 | 9.03 | 0.22 | 378 | 1.06 | 0.08 | 4.12 | 0.15 | 233 | 1.10 | 0.08 | 1 | 1.26 | 0.08 | 25 | 1.46 | 0.09 | 52 | |
| C/2021 S3 | 4.23 | 0.17 | 426 | 0.20 | 0.04 | 7.91 | 0.24 | 508 | 0.55 | 0.06 | 140 | 0.25 | 0.04 | 27 | 0.20 | 0.04 | -5 | |
| C/2023 C2 | 0.38 | 0.10 | 0 | 0.35 | 0.10 | 5.56 | 0.38 | 98 | 0.92 | 0.16 | 34 | 1.17 | 0.18 | 43 | 1.64 | 0.21 | 55 | |
| C/2022 S4 | 11.40 | 0.30 | 259 | 1.40 | 0.11 | 8.31 | 0.26 | 215 | 4.42 | 0.19 | 140 | 3.71 | 0.17 | 118 | 2.98 | 0.15 | 90 | |
| C/2021 G2 | 3.01 | 0.12 | 233 | 0.96 | 0.07 | 56.89 | 0.52 | 839 | 2.66 | 0.11 | 207 | 1.18 | 0.08 | 38 | 0.96 | 0.07 | -4 | |
| C/2024 E1 | 0.55 | 0.09 | -2 | 0.53 | 0.09 | 2.95 | 0.21 | 101 | 1.55 | 0.16 | 64 | 1.34 | 0.14 | 55 | 1.21 | 0.14 | 48 | |
| C/2019 L3 | 0.16 | 0.05 | -4 | 0.16 | 0.05 | 22.71 | 0.54 | 374 | 0.28 | 0.06 | 40 | 0.50 | 0.08 | 84 | 0.96 | 0.11 | 135 | |
| C/2022 QE78 | 2.19 | 0.20 | 170 | 0.10 | 0.04 | 9.36 | 0.41 | 249 | 0.43 | 0.09 | 77 | 0.73 | 0.11 | 107 | 0.88 | 0.12 | 118 | |
| C/2022 A3 | 1.42 | 0.12 | 129 | 0.36 | 0.06 | 3.76 | 0.20 | 220 | 0.65 | 0.08 | 52 | 0.50 | 0.07 | 26 | 0.39 | 0.06 | 3 | |
| C/2023 R1 | 1.24 | 0.12 | 134 | 0.26 | 0.05 | 3.38 | 0.20 | 218 | 0.38 | 0.07 | 31 | 0.47 | 0.07 | 50 | 0.53 | 0.08 | 59 | |
| C/2024 A1 | 0.87 | 0.10 | 0 | 0.83 | 0.10 | 1.14 | 0.11 | 19 | 1.12 | 0.11 | 21 | 1.12 | 0.11 | 22 | 1.13 | 0.11 | 22 | |
| C/2022 E3 | 0.09 | 0.06 | 21 | 0.03 | 0.04 | 3.62 | 0.39 | 105 | 0.04 | 0.04 | 3 | 0.04 | 0.04 | -2 | 2.60 | 0.33 | 100 | |
| C/2023 T3 | 2.31 | 0.21 | 108 | 0.28 | 0.07 | 2.35 | 0.21 | 105 | 1.58 | 0.17 | 88 | 1.46 | 0.17 | 84 | 1.25 | 0.15 | 76 | |
| C/2021 E3 | 0.11 | 0.05 | 26 | 0.05 | 0.04 | 3.59 | 0.31 | 151 | 0.05 | 0.04 | -1 | 0.05 | 0.04 | 0 | 0.24 | 0.08 | 55 | |
| C/2022 R6 | 3.80 | 0.15 | 313 | 0.63 | 0.06 | 6.39 | 0.19 | 399 | 1.53 | 0.09 | 151 | 1.84 | 0.10 | 184 | 1.96 | 0.11 | 195 | |
| C/2021 O3 | 0.19 | 0.09 | 54 | 0.02 | 0.03 | 2.24 | 0.31 | 110 | 0.11 | 0.07 | 41 | 0.06 | 0.05 | 27 | 0.03 | 0.04 | 10 | |
| C/2022 U3 | 0.97 | 0.14 | 45 | 0.39 | 0.09 | 1.19 | 0.15 | 52 | 0.77 | 0.12 | 33 | 0.75 | 0.12 | 31 | 0.74 | 0.12 | 30 | |
| C/2022 T1 | 0.43 | 0.10 | 94 | 0.04 | 0.03 | 0.63 | 0.12 | 106 | 0.27 | 0.08 | 74 | 0.30 | 0.08 | 79 | 0.37 | 0.09 | 87 | |
| C/2021 C5 | 0.11 | 0.05 | 0 | 0.10 | 0.05 | 1.33 | 0.17 | 111 | 0.10 | 0.05 | -3 | 0.11 | 0.05 | 1 | 0.13 | 0.05 | 9 |
Table 3. Model-fit statistics summary. RSS: The residual sum of squares for each model and each comet (Equation (4)) in units of mag. The RSS forms the basis of the RMSE equation. A smaller value indicates a better fit. RMSE: The root mean squared error based on the RSS for each model and each comet (Equation (5)) in units of mag. A smaller value indicates a better fit. BIC: BIC − BIC. Positive values indicate the model is preferred; BIC > 10 indicates strong preference. Negative values indicate the alternative model is preferred over . The BIC column is omitted since BIC = 0 by definition. The horizontal line separates the seven comets with the largest (above) from the rest of the sample (below).
Figure A2. for each of the five alternative models, for every comet in the sample, plotted on a symmetric-log (-)scale so that negative and zero values are visible alongside the more extreme positive ones. The solid gray horizontal line marks , where negative values indicate that the alternative model is preferred over ; the dashed red horizontal line marks , the threshold for strong evidence in favor of . The seven comets with are separated from the remaining sample by the vertical dashed line.
Predictive Power
Figure 6 and Figure 7 show the mean and mean at (§ 4.2), respectively, across our seven comets as a function of . With only seven comets contributing to the mean, each sampled at its own irregular real cadence with gaps during solar conjunction, both aggregate curves show modest non-monotonic structure around – rather than a perfectly smooth trend. Appendix § 10 shows this is a small-sample sampling effect rather than model instability: the individual per-comet curves (Figure A3) and the corresponding fit-arc length vs. (Figure A4) show that the fit-arc used at a given changes in discrete steps as the threshold sweep crosses real observation epochs, not continuously.
Figure 6. Mean root-mean-squared error (RMSE) up to the common endpoint, as a function of the ratio threshold used for fitting, averaged across the seven comets. The curves are not perfectly smooth, reflecting the small number of comets (N = 7) contributing to the mean, each sampled at its own irregular real cadence (see Appendix B, Figures A3 and A4). Fixed- and Correlated- reach RMSE below 1 mag for all seven comets from the earliest threshold tested, while and require substantially more data to converge. At large , and Correlated- converge to comparable RMSE; note that this predictive comparison uses only data up to a given threshold, capped at , whereas the full-lightcurve comparison in Section 5.1 fits and evaluates each model using the complete dataset for each comet (extending to each comet’s full observed ); the two are therefore not evaluating the models against the same amount of data.
Figure 7. Mean absolute magnitude difference between each model’s prediction and the observed brightness at a fixed evaluation point, , as a function of the ratio threshold used for fitting. This metric isolates predictive accuracy at a single distant reference point rather than averaging over the full extrapolated curve as in Figure 6. It shows a broadly consistent picture: Fixed- and Correlated- converge earliest, while and remain unreliable until .
Figure A3. Individual per-comet RMSE curves (gray) underlying the sample mean shown in Figure 6, one panel per model. Individual comets can show non-monotonic jumps where the fit-arc length does not grow smoothly with (Figure A4); these average out in the sample mean but are visible here directly, demonstrating that the modest structure in Figure 6 reflects small-sample sampling behavior rather than instability in any single model’s fit.
Figure A4. Heliocentric distance range of the data actually used for each fit () as a function of the ratio threshold , for each of the seven comets (colored lines) and the sample mean (black). The dashed purple line shows the smooth theoretical expectation, , evaluated at the sample-mean au; it is not a per-comet prediction, so comets with individual above/below this average (5.97–8.58 au across the sample) trace systematically above/below it. Within a given comet, the fit-arc length only increases when crosses an observed epoch, producing the step-like behavior seen here rather than a smooth curve. This discreteness, not model instability, is the origin of the non-monotonic structure in Figure 6 and Figure 7.
is the best-fit model for characterizing the full lightcurve (§ 5.1) but the worst-performing for prediction from limited early data: at (– au), fails badly by both measures (mean exceeds 10 mag throughout; mean exceeds 10 mag at and stays high, 7 mag, through 1.20), reflecting the difficulty of constraining three free parameters from a short arc. performs similarly poorly at small , succeeding for only 2–4 of 7 comets by either metric until .
Fixed- is the only model to predict brightnesses for all 7 comets to within 1 mag at the very first threshold tested (, mean mag, mean mag), since fixing both and leaves only free: with and fixed, sets only the vertical offset of the curve rather than its shape, so it can be constrained from even a single data point. This advantage is short-lived, however: by the next threshold (), Correlated- is already more accurate than Fixed- by both metrics ( vs. 0.33 mag; vs. 0.53 mag), and the gap widens steadily thereafter, reaching roughly a third of
Fixed-’s error by (RMSE = 0.07 vs. 0.20 mag). Fixed-’s value is therefore specifically in guaranteeing a usable prediction from the very first available epoch, not in being the most accurate choice once even slightly more data exist. Fixed- succeeds for only 2/7 comets at by both metrics, catching up to Correlated- by . Fixed- is the slowest of the four constrained variants under both metrics, not reaching full success until .
All models converge toward good predictive performance by both metrics once enough data are available (), consistent with ’s superior full-curve characterization once sufficiently constrained.
Characterization
After assessing the distribution of the and parameters in the model (§ 3.1, Figure 3), we examined the distribution of in the two-parameter models. We focused on the Fixed- and Correlated- models. Fixed- was the worst-performing two-parameter variant in our predictive tests (§ 5.2) and is not discussed further here. Fixed- was the second-best fit for three of the seven high-quality () comets (§ 5.1). For this model, has a mean of 0.46 (median of 0.45) with a standard deviation of 0.14 across the full sample; for comparison, the Correlated- model gives a similar but slightly broader distribution, with a mean of 0.44 (median of 0.36) and a standard deviation of 0.18, after outlier rejection excluded 4 of the 23 comets. We also found that larger values correspond to comets sampled over smaller heliocentric distance ranges closer to the Sun (Figure 8, left panel).
Figure 8. (left) Heliocentric distance of the data for each comet, plotted against the coefficient from the Fixed- model fit. Comets are colored and shaped by dynamical class inferred from : New (; blue circles), Uncertain (; gray squares), and Returning (; green triangles). Open symbols mark the seven comets with used in our high quality tests. The dashed red line and shaded band show the sample mean () and range. Comets sampled closer to the Sun tend toward systematically larger , and these are also the comets with some of the smallest coverage. (right) The -model parameters and for the same sample, colored and marked as in the left panel; open symbols again denote the seven comets. The red cross and shaded ellipse show the mean and covariance region for the entire sample. For both models, we observe no trend with dynamical class.
We further examined whether either model’s parameters correlate with dynamical class. Comets with au are classified as New, following the threshold adopted in § 1.3, while comets with au, the New threshold used by Lacerda et al. [31] for the same purpose, are classified as Returning. Comets falling between these two literature thresholds are classified as Uncertain. As shown in Figure 8, we found no clear trend with dynamical class in the values from the Fixed- model or in the full-sample distribution (mean , ; median , ), consistent with the lack of a dynamical-class dependence found by Holt et al. [19].
Discussion
An Improved Brightening Model
Allowing the activity index to vary linearly with greatly improves the fits to the observations. This gives smoothly varying heliocentric lightcurves that fit all comets in our sample remarkably well. It is also notable that the range of fit parameters for the best sampled comets (those observed over the largest range in ) is small, and that even the Fixed- model does a good job of approximating the behavior of most of the comets. This suggests a common activity pattern across most long-period comets discovered around au, which we can describe by looking at the with for the average case (i.e., the Fixed- model; Figure 1). This has around au, corresponding to the ‘typical’ value for historically observed DNCs, where the amount of material in the coma is constant, and the change in brightness is just due to the inverse-square law illumination change. This makes sense, as most of the comets in our sample are DNCs, although we found no difference in brightening behavior between DNCs and returning comets in this sample (Section 5.3). The Fixed- model has reach zero at au (and go
negative inside this distance), showing the limits of this empirical model – it fits well for the distant evolution of comet brightness, but not in the water-dominated region, so should not be relied on to describe activity changes close to the Sun (e.g., au). At larger distances the average comet described by the Fixed- model has a rapid increase in brightness, with at au and at 15 au, suggesting activity ‘turning on’ at these distances. This appears to be corroborated by the observed range of discovery distances of comets, which in the modern survey era are regularly found out to au, but rarely beyond that distance ([33]). Around this general behavior, there are clearly subtleties for individual comets, and there are other comets that are clear outliers – for example, those showing very distant activity at 20 au or more (e.g., C/2017 K2) – which were not included in the sample used here. One interpretation of this could be that the majority of LPCs we observe are depleted in near-surface CO (perhaps due to heating before they were placed into the Oort cloud; [15]), and their activity therefore starts when CO driven activity becomes significant near 10 au, while more distantly active comets are the exceptional ones that retain significant CO. This would match JWST observations showing that C/2024 E1 was CO depleted ([56]); this comet was discovered at au and had a pre-perihelion brightness evolution that could be described by the model, where and . Testing this interpretation directly would require volatile abundance measurements for a larger fraction of the sample, since our own dataset only consists of broadband photometry. JWST is well suited to this: NIR-Spec IFU spectroscopy can constrain the CO/CO/HO abundance ratios of individual comets at the heliocentric distances relevant here, as demonstrated for C/2024 E1 ([56]). Spectroscopic follow-up of newly discovered distant comets, compared against their photometric (, ) values, would offer a direct test of whether volatile composition drives the diversity seen in Figure 3. This test would also extend to rarer distant discoveries, which may have different volatile abundances resulting in brightening behavior that differs from our sample, potentially occupying a distinct region of space.
Interpreting Across Models
While § 5.1 and § 5.2 focus on how well each model fits and predicts brightness, the recovered values themselves carry an important caveat: they are not directly comparable across models. The values recovered also depend strongly on which parameters are fixed (Table 5). Adopting the Fixed- model, for example, yields values that are fainter by mag on average relative to the traditional model, a systematic offset between models rather than a physical difference in nucleus size. This model dependence means that once and are allowed to vary, loses much of its utility as a proxy for comparing intrinsic brightness between comets, limiting its
applicability to metrics such as the Bortle limit ([6]). As with comet nuclei generally, cannot be converted to a physical size using the size-albedo formalism applied to asteroids; that said, the values from the Fixed-, Fixed-, Fixed-, and Correlated- models are broadly closer to plausible nucleus sizes under this formalism than those from , whereas the full three-parameter model yields a much wider, unphysical range of (2.3 to 46.8 mag) across the sample.
Predicting Brightness Behavior
The model, which best reproduces a long pre-perihelion lightcurve, is not always the safest choice for forecasting when only a short early arc exists. , Fixed-, Fixed-, Fixed-, and Correlated- all share the same functional form, ; they differ only in whether , , or both are fixed to the population values and , rather than left free. With both and free, is preferred by BIC for characterizing the distant evolution in the majority of our sample (§ 5.1), though Correlated- achieves comparable once enough data are available (Figure 6) and reaches that level of performance with substantially less data. 's flexibility becomes a liability when only a short arc is available to constrain three free coefficients; in our progressive fitting experiments, it is the weakest performer at forecasting further along the inbound leg by both the and fixed-point metrics. (Section 5.2). The traditional model is not markedly better, struggling even with moderate data coverage.
Table 4 translates these thresholds into a practical recommendation for observers deciding which model to trust at a given stage of an inbound arc. The pattern underlying it, evident in the metrics reported in § 5.2, is that fixing more coefficients trades full-curve fit quality for predictive stability when data are limited, and that trade becomes less necessary as the arc lengthens. With only discovery-epoch data (), Fixed- is the only model that predicts all seven comets to within 1 mag from the first available epoch, since fixing both and to and leaves as the sole free parameter. By (roughly 1 – 1.5 au of coverage), Correlated- overtakes it as the preferred two-coefficient choice. Once the arc reaches roughly – 4 au (), enough data exist to leave both and free (in addition to ), and the fully unconstrained model both meets the 1 mag threshold and gives the best characterization of the lightcurve, at least until the comet enters the water-dominated regime ( au). Even once fully converged, the RMS for the Fixed- model is roughly three times that of . Fixed- is not the most accurate model at any stage; it is simply the only one usable immediately.
Among the three two-parameter variants, Correlated- reaches full-sample convergence earliest, followed closely by Fixed-, with Fixed- the slowest (§ 5.2). Because the anchor values and the – relation were derived from six of the seven comets used in this test, we repeated the predictive analysis using anchors recomputed without each predicted comet (§ 3.1). The results were essentially unchanged ( mag for Correlated-, Fixed-, and Fixed-), since each retains a free coefficient that absorbs small shifts in the anchors; Fixed-, with no free shape parameter, was the most affected ( mag). Nonetheless, all of these anchors are derived from a small sample, and Fixed-, which does not depend on the – correlation, may be the more reliable choice if a larger dataset reveals a weaker or different relation.
Our predictive framework could in principle also be applied in the opposite direction: rather than forecasting a comet’s future brightness, one could extrapolate backward from early inbound data to estimate its brightness at larger , before discovery. This is directly relevant to precovery searches, where archival survey data are mined for earlier detections of a newly discovered comet, since a model-based estimate of the pre-discovery brightness could indicate whether such a search is worthwhile. We have not tested backward extrapolation with this dataset, though the reduced-parameter models in Table 4 may be similarly well suited to this application, given their performance in forward prediction.
Table 4. Recommended brightening models as a function of available data coverage for comets discovered at – au.
Implications for Comet Interceptor
This model can hopefully assist the Comet Interceptor mission in two ways: firstly, it can be used to look at the most general evolution of distant comet brightness, as an input for simulations for likely discovery rates with LSST (Donaldson et al., in prep), which are important to assess the likelihood of discovery of suitable targets with time, and consequently inform decisions on whether to accept a possible target or wait for a ‘better’ one ([59] (Snogdrass et al. 2026)). Secondly, and more directly, the approximations provided by the model variants can assist in making more realistic predictions for the future brightness evolution of a newly discovered comet — particularly soon after discovery, when only a short arc is available — to assess its suitability as a target.
It is worth noting that the finding that most comets brighten rapidly at au has important implications for the mission: if comets with activity at substantially larger distances are unusual, then even in the LSST era we may not expect to find many. This means that the likely mission target will be found closer to au, which means only years from perihelion for comets coming close to au (Figure 9). By contrast, a model fit to the same inbound lightcurve ( over 3–9 au, consistent with the mean for dynamically new comets from Holt et al. [19]) predicts LSST detection at au, yr earlier than . Because steepens beyond au (§ 6.1), any extrapolation of the traditional model will overestimate how early distant comets can be found. Rapid decision making and reaction will therefore likely be necessary to meet an incoming comet with relatively little warning, given typical cruise times of 2-3 years ([48]).
Figure 9. Apparent magnitude vs. heliocentric distance (, log scale) for a hypothetical comet using the real orbital geometry of a comet with au. The comet is discovered at au with (filled star), a typical current-survey limit (upper dotted line). The solid purple curve shows the model (, ); the dashed blue curve shows the model fit to the curve over 3–9 au (shaded region; ). Open symbols mark where each model reaches , the LSST single-visit depth (lower dotted line). Times are relative to , when the comet reaches au, the approximate encounter distance of Comet Interceptor.
Limitations of the Model
The model and its reduced variants describe the average brightening behavior of the sample, not necessarily the behavior of any single comet. Individual lightcurves can depart from the population trend for identifiable physical reasons, causing that comet’s behavior to deviate from the population-level model without necessarily invalidating it. The fitted parameters can also carry systematic biases from our specific dataset and reduction choices, which could shift the inferred distribution without appearing as outliers at all.
We selected comets that were discovered at au, observed over a pre-perihelion range of au and that have au. We additionally only considered observations from au. Comets discovered closer to the Sun, with larger perihelion distances, or observed over shorter arcs, are not represented here. Using the threshold from § 1, 16 of the 23 comets are dynamically new; the sample is weighted toward DNCs, and while we found no clear difference in brightening behavior between DNCs and returning comets, that comparison relied on only seven returning comets, only four of which would still be classified as returning under the stricter threshold used by Lacerda et al. [31].
All of our fits use pre-perihelion (inbound) data only; we do not test or validate any of the models presented here against post-perihelion lightcurves in this work. As noted in § 1, the post-perihelion brightening behavior of LPCs at large distances remains poorly characterized in the literature more broadly, and it is not yet known whether the same values, or even the same functional form, would describe a comet’s outbound evolution. We are extending the LOOK sample to post-perihelion data and plan to address this question directly in future work.
same functional form, would describe a comet’s outbound evolution. We are extending the LOOK sample to post-perihelion data and plan to address this question directly in future work.
Because our photometry uses a fixed angular aperture, the volume of coma sampled grows with geocentric distance , biasing resolved comae to appear relatively brighter at large : flux scales closer to than the point-source of Equation (1), though this can shift back toward if a tail extends beyond the aperture ([23]). We already substituted for scaling in all of our model fits (Equation (2)) to partially account for this; however, the transition between coma- and tail-dominated regimes is not captured by this single correction. Since this transition is not modeled explicitly, it is degenerate with the -dependence we fit, and would be absorbed into and rather than appearing as clear deviations in the residuals. Our reported values are therefore specific to this aperture choice.
Individual comets can also deviate from the population trend for reasons our models do not capture: outbursts, which are well documented in distant LPCs (e.g., [28, 20]) and can bias a fit or produce a temporary offset from the trend at any stage; seasonal and projection effects, where a changing illuminated area or coma/tail geometry alters apparent brightness independent of activity ([62, 51, 29]); and comets like C/2019 L3, whose rapid brightness surge beyond au departs strongly from the population trend (e.g., Fixed- RMSE mag, the worst fit in the sample). These are reminders that our models describe the typical inbound evolution of an LPC discovered near 5–10 au, not a universal law.
An outburst or a surge will likely still be visible as an outlier relative to the population trend. For brightness predictions, where the goal is the likely range of activity rather than an exact forecast, the reduced models in Table 4 remain a practical tool available at discovery.
Summary
We compared the traditional constant activity-index model () to an empirical distance-dependent brightening model () and its anchored, reduced-parameter variants, using real pre-perihelion lightcurves for 23 long-period comets from the LCO Outbursting Objects Key (LOOK) project. We summarize our results below.
Characterizing distant brightening behavior:
For inbound photometry at au, a constant activity index often cannot describe the full distant lightcurve; allowing the effective slope to vary with heliocentric distance substantially improves the fits.
The empirical three-parameter model () provides a significantly better match to our distant lightcurves than the traditional model.
The best-observed comets cluster in space (, ), motivating one- and two-parameter variants (Fixed-, Fixed-, Fixed-, and Correlated-) anchored to these population statistics; the Bayesian information criterion favors the full three-parameter model over all four variants for the majority of objects, justifying the added complexity for characterization.
Predicting future brightness:
Short early segments poorly constrain , so although it is best for full-curve characterization, it is comparatively poor for early-time magnitude forecasting.
The constrained variants are more robust predictors: Fixed- performs well even for very short initial segments, while the two-parameter variants can improve quickly with modest additional coverage (– au). Correlated- converges earliest among the two-parameter cases, with Fixed- close behind.
Applicability and outlook:
These models were fit only to pre-perihelion (inbound) data and should not be assumed to apply post-perihelion, where fading behavior remains poorly characterized. They also do not capture all pre-perihelion variability (e.g., outbursts, seasonal effects), and should not be relied on to describe activity inside the water-dominated regime ( au), where none of the models tested here were fit or validated. All fits also assume that the brightness measured in our fixed angular aperture scales as (Equation (2)), so the reported values apply to photometry fit with the same scaling.
Additional long-term lightcurves provided by LSST should tighten population constraints on and strengthen the priors behind reduced-parameter prediction for survey and mission applications.
Acknowledgments
The authors would like to sincerely thank the two anonymous reviewers, whose valuable feedback greatly strengthened this article. CEH was supported by an LSST-DA Catalyst Fellowship during this work, made possible through the support of Grant 62192 from the John Templeton Foundation to LSST-DA. CS acknowledges funding from the UK Space Agency. This work makes use of observations from the Las Cumbres Observatory global telescope network under proposal codes KEY2020B-009, LTP2023B-001, and LTP2025A-004. The raw frames and Basic Calibrated Data (BANZAI-processed) products are available at https://archive.lco.global.
The views expressed in this article are those of the authors and do not reflect the official policy or position of the U.S. Naval Academy, Department of the Navy, the Department of Defense, or the U.S. Government.
During the preparation of this work the authors used a large language model for minor copy editing. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the published article.
Facilities: LCOGT (1m Sinistro)
Software: astropy ([2, 3, 4]), numpy ([18]), pandas ([38]), astroquery ([13]), scipy ([60]), sbpy ([41]), JPL Horizons ([14]), matplotlib ([21]), seaborne ([61]), NEOexchange ([34]), BANZAI ([37]), calviacat ([27]), SExtractor ([5]).
Model Ranking and Significance Visualizations
Figure A1 shows the RMSE (top panel) and corresponding rank (bottom panel) of each model for every comet in the sample, as described in § 5. Figure A2 shows the of each model relative to for every comet, with the threshold used to define strong evidence in § 5 marked for reference.
Predictive Test Diagnostics
Figure A3 shows the individual per-comet curves underlying the sample-mean curves presented in Figure 6 (§ 5.2), and Figure A4 shows the heliocentric distance range of data actually used for each fit as a function of for each comet. Because only seven comets contribute to the sample mean, and each is sampled at its own irregular real cadence, the fit-arc length used at a given threshold grows in discrete steps as crosses real observation epochs, rather than continuously (Figure A4). This produces the modest non-monotonic structure visible in Figure 6 and Figure 7 around ; Figure A3 confirms that this behavior arises from small-sample sampling granularity, not from instability in any individual model’s predictions.
Model Fit Parameters
Table 5 lists the best-fit parameters for each comet and model described in § 3.2.
References
- [1]A’Hearn, M. F., Millis, R. L., Schleicher, D. G., Osip, D. J., & Birch, P. V. 1995, Icarus, 118, 223, doi: 10.1006/icar.1995.1190DOI
- [2]Astropy Collaboration, Robitaille, T. P., Tollerud, E. J., et al. 2013, A&A, 558, A33, doi: 10.1051/0004-6361/201322068
- [3]Astropy Collaboration, Price-Whelan, A. M., Sipőcz, B. M., et al. 2018, AJ, 156, 123, doi: 10.3847/1538-3881/aabc4f
- [4]Astropy Collaboration, Price-Whelan, A. M., Lim, P. L., et al. 2022, ApJ, 935, 167, doi: 10.3847/1538-4357/ac7c74DOI
- [5]Bertin, E., & Arnouts, S. 1996, A&AS, 117, 393, doi: 10.1051/aas:1996164DOI
- [6]Bortle, J. E. 1991, Int. Comet Quart., 13, 89
- [7]Brown, T. M., Baliber, N., Bianco, F. B., et al. 2013, PASP, 125, 1031, doi: 10.1086/673168DOI
- [8]Burnham, K. P., & Anderson, D. R. 2004, Sociological Methods & Research, 33, 261, doi: 10.1177/0049124104268644DOI
- [9]De Keyser, J., Edberg, N. J. T., Henri, P., et al. 2025, Planet. Space Sci., 256, 106032, doi: 10.1016/j.pss.2024.106032DOI
- [10]Dybczyński, P. A., & Breiter, S. 2022, A&A, 657, A65, doi: 10.1051/0004-6361/202142227DOI
- [11]Dybczynski, P. A., & Królikowska, M. 2015, MNRAS, 448, 588, doi: 10.1093/mnras/stv013DOI
- [12]Everhart, E. 1967, AJ, 72, 716
- [13]Ginsburg, A., Sipőcz, B. M., Brasseur, C. E., et al. 2019, AJ, 157, 98, doi: 10.3847/1538-3881/aafc33DOI
- [14]Giorgini, J. D., Yeomans, D. K., Chamberlin, A. B., et al. 1996, in Bulletin of the American Astronomical Society, Vol. 28, AAS/Division for Planetary Sciences Meeting Abstracts #28, 1158
- [15]Gkotsinas, A., Nesvorný, D., Guilbert-Lepoutre, A., Raymond, S. N., & Kaib, N. 2024, PSJ, 5, 243, doi: 10.3847/PSJ/ad7f4eDOI
- [16]Guilbert-Lepoutre, A. 2012, AJ, 144, 97, doi: 10.1088/0004-6256/144/4/97
- [17]Harrington Pinto, O., Kelley, M. S. P., Villanueva, G. L., et al. 2023, PSJ, 4, 208, doi: 10.3847/PSJ/acf928DOI
- [18]Harris, C. R., Millman, K. J., van der Walt, S. J., et al. 2020, Nature, 585, 357, doi: 10.1038/s41586-020-2649-2DOI
- [19]Holt, C. E., Knight, M. M., Kelley, M. S. P., et al. 2024, PSJ, 5, 273, doi: 10.3847/PSJ/ad8e38DOI
- [20]Hui, M.-T., Farnocchia, D., & Micheli, M. 2019, AJ, 157, 162, doi: 10.3847/1538-3881/ab0e09DOI
- [21]Hunter, J. D. 2007, Computing in Science & Engineering, 9, 90, doi: 10.1109/MCSE.2007.55DOI
- [22]Jenniskens, P., & Blake, D. F. 1996, ApJ, 473, 1104, doi: 10.1086/178220DOI
- [23]Jewitt, D. 1991, in Astrophysics and Space Science Library, Vol. 167, IAU Colloquium 116: Comets in the post-Halley era, ed. R. L. Newburn, Jr., M. Neugebauer, & J. Rahe, 19, doi: 10.1007/978-94-011-3378-4_2DOI
- [25]Jewitt, D., Kim, Y., Mutchler, M., et al. 2021, AJ, 161, 188, doi: 10.3847/1538-3881/abe4cfDOI
- [26]Jones, G. H., Sondgrass, C., Tubiana, C., et al. 2024, SSRv, 220, 9, doi: 10.1007/s11214-023-01035-0DOI
- [27]Kelley, M. S. P., Bodewits, D., Ye, Q., et al. 2019, in Astronomical Society of the Pacific Conference Series, Vol. 523, Astronomical Data Analysis Software and Systems XXVII, ed. P. J. Teuben, M. W. Pound, B. A. Thomas, & E. M. Warner, 471
- [28]Kelley, M. S. P., Kokotankova, R., Holt, C. E., et al. 2022, ApJL, 933, L44, doi: 10.3847/2041-8213/ac7becDOI
- [29]Kelley, M. S. P., Protopapa, S., Bodewits, D., et al. 2025, PSJ, 6, 185, doi: 10.3847/PSJ/ade5acDOI
- [30]Królikowska, M., & Dybczyński, P. A. 2017, MNRAS, 472, 4634, doi: 10.1093/mnras/stx2157DOI
- [31]Lacerda, P., Guilbert-Lepoutre, A., Kokotankova, R., et al. 2025, A&A, 697, A210, doi: 10.1051/0004-6361/202453565DOI
- [32]Levine, W. G., Kim, Y., Spahr, T. B., et al. 2025, RNAAS, 9, 262, doi: 10.3847/2515-5172/ae0f12DOI
- [33]Lister, T., Kelley, M. S. P., Holt, C. E., et al. 2022, PSJ, 3, 173, doi: 10.3847/PSJ/ac7a31DOI
- [34]Lister, T., Gomez, E., Chatelain, J., et al. 2023, NEOExchange: Target and Observation Manager for the Solar System, Astrophysics Source Code Library, record ascl:2311.005. http://ascl.net/2311.005
- [35]Marcus, J. N. 2007a, Int. Comet Quart., 39
- [37]McCully, C., Volgenau, N. H., Harbeck, D.-R., et al. 2018, in Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series, Vol. 10707, Software and Cyberinfrastructure for Astronomy V, ed. J. C. Guzman & J. Ib sen, 107070K, doi: 10.1117/12.2313430
- [38]McKinney, W. 2010, in Proceedings of the 9th Python in Science Conference, ed. Stéfan van der Walt & Jarrod Millman, 56 – 61, doi: 10.25080/Majora-92bf1922-00aDOI
- [39]Meech, K. J., & Svoren, J. 2004, in Comets II, ed. M. C. Festou, H. U. Keller, & H. A. Weaver (Univ. of Arizona Press/Lunar Planet. Inst., Tucson, AZ/Houston, TX), 317–335
- [40]Meech, K. J., Pittichová, J., Bar-Nun, A., et al. 2009, Icarus, 201, 719, doi: 10.1016/j.icarus.2008.12.045DOI
- [41]Mommer, M., Kelley, M., de Val-Borro, M., et al. 2019, Journal of Open Source Software, 4, 1426, doi: 10.21105/joss.01426
- [42]Oort, J. H. 1950, BAN, 11, 91
- [43]Oort, J. H., & Schmidt, M. 1951, BAN, 11, 259
- [44]Opik, E. J. 1963, Irish Astronomical Journal, 6, 93
- [45]Prialnik, D., & Jewitt, D. 2022, Comets III, arXiv:2209.05907, doi: 10.48550/arXiv.2209.05907arxiv.org/abs/2209.05907
- [46]Raftery, A. E. 1995, Sociological Methodology, 25, 111. http://www.jstor.org/stable/271063
- [47]Roth, N. X., Milam, S. N., Cordiner, M. A., et al. 2025, ApJL, 986, L22, doi: 10.3847/2041-8213/add526DOI
- [48]Sánchez, J. P., Morante, D., Hermosin, P., et al. 2021, Acta Astronautica, 188, 265, doi: 10.1016/j.actaastro.2021.07.014DOI
- [49]Sárneczky, K., Szabó, G. M., Csák, B., et al. 2016, AJ, 152, 220, doi: 10.3847/0004-6256/152/6/220DOI
- [50]Schambeau, C. A., Kelley, M. S. P., Womack, M., et al. 2026, PSJ, 7, 137, doi: 10.3847/PSJ/ae685cDOI
- [51]Schleicher, D. G. 2007, Icarus, 190, 406, doi: 10.1016/j.icarus.2007.04.013DOI
- [52]Schleicher, D. G., & Bair, A. N. 2011, AJ, 141, 177, doi: 10.1088/0004-6256/141/6/177DOI
- [53]Schleicher, D. G., Millis, R. L., & Birch, P. V. 1998, Icarus, 132, 397, doi: 10.1006/icar.1997.5902DOI
- [54]Schwarz, G. 1978, Annals of Statistics, 6, 461
- [55]Smoluchowski, R. 1981, ApJL, 244, L31, doi: 10.1086/183473DOI
- [56]Snodgrass, C., Holt, C. E., Kelley, M. S. P., et al. 2025, MNRAS, 541, L8, doi: 10.1093/mnrasl/slaf046DOI
- [57]Snodgrass, C., Mazzotta Epifani, E., Tubiana, C., et al. 2026, Icarus, 447, 116887, doi: 10.1016/j.icarus.2025.116887DOI
- [58]Tonry, J. L., Stubbs, C. W., Lykke, K. R., et al. 2012, ApJ, 750, 99, doi: 10.1088/0004-637X/750/2/99DOI
- [59]Vigren, E., Eriksson, A. I., Edberg, N. J. T., & Snodgrass, C. 2023, Planet. Space Sci., 237, 105765, doi: 10.1016/j.pss.2023.105765DOI
- [60]Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature Methods, 17, 261, doi: 10.1038/s41592-019-0686-2DOI
- [61]Waskom, M. L. 2021, Journal of Open Source Software, 6, 3021, doi: 10.21105/joss.03021DOI
- [62]Weissman, P. R. 1988, Exploration of Halley’ s Comet, 873, doi: 10.1007/978-3-642-82971-0_148
- [63]Whipple, F. L. 1978, Moon and Planets, 18, 343
- [64]Womack, M., Sarid, G., & Wierzchos, K. 2017, PASP, 129, 031001, doi: 10.1088/1538-3873/129/973/031001DOI
- [65]Yang, B., Jewitt, D., Zhao, Y., et al. 2021, ApJL, 914, L17, doi: 10.3847/2041-8213/ac03b7DOI











