From out of the Blue: Swift Links 2002es-like, 2003fg-like, and Early Time Bump Type Ia Supernovae

We collect a sample of 42 Type Ia supernovae (SNe Ia) with Swift UV photometry and well-measured early time light-curve rises and find that 2002es-like and 2003fg-like SNe Ia have different prepeak UV color evolutions compared to normal SNe Ia and other spectroscopic subtypes. Specifically, 2002es-like and 2003fg-like SNe Ia are cleanly separated from other SNe Ia subtypes by UVM2 − UVW1 ≳ 1.0 mag at t = − 10 days relative to B-band maximum. Furthermore, the SNe Ia that exhibit nonmonotonic bumps in their rising light curves, to date, consist solely of 2002es-like and 2003fg-like SNe Ia. We also find that SNe Ia with two-component power-law rises are more luminous than SNe Ia with single-component power-law rises at prepeak epochs. Given the similar UV colors, along with other observational similarities, we discuss a possible progenitor scenario that places 2002es-like and 2003fg-like SNe Ia along a continuum and may explain the unique UV colors, early time bumps, and other observational similarities between these objects. Ultimately, further observations of both subtypes, especially in the near-infrared, are critical for constraining models of these peculiar thermonuclear explosions.

Several progenitor scenarios may explain the origin of SNe Ia, including the single-degenerate (SD), doubledegenerate (DD), and core-degenerate (CD) scenarios.The SD scenario consists of a CO WD with a nondegenerate companion such as a main sequence or red giant star (e.g., Whelan & Iben 1973), the DD scenario consists of two CO WDs or a CO WD and a He WD (e.g., Nomoto 1980), and the CD scenario consists of a CO WD and a degenerate CO core of an AGB star (e.g., Hoeflich & Khokhlov 1996).
A variety of explosion mechanisms for each progenitor scenario may produce SNe Ia below, at, or above the Chandrasekhar mass.In either the SD or DD progenitor scenario, material from a companion can accrete onto the CO WD.It can trigger an explosion through central carbon ignition as the CO WD approaches the Chandrasekhar mass (Hoyle & Fowler 1960;Whelan & Iben 1973;Nomoto 1982;Piersanti et al. 2003) or through a He detonation on the surface of a CO WD below the Chandrasekhar mass (Nomoto 1980;Livne 1990;Woosley & Weaver 1994;Hoeflich & Khokhlov 1996;Hoeflich et al. 2017;Polin et al. 2019).In addition to the aforementioned explosion mechanisms, the DD scenario has additional explosion mechanisms which include mergers below, at, or above M ch (Iben & Tutukov 1984;Webbink 1984;van Kerkwijk et al. 2010;Scalzo et al. 2010;Pakmor et al. 2010Pakmor et al. , 2013;;Kromer et al. 2013Kromer et al. , 2016) ) and third-or fourth-body induced collisions (Thompson 2011;Shappee & Thompson 2013;Pejcha et al. 2013).Finally, the CD scenario may explode via the merger of a degenerate CO core of an asymptotic giant branch (AGB) star and a CO WD (Hoeflich & Khokhlov 1996;Noebauer et al. 2016;Maeda et al. 2023).
Despite differences in the explosion mechanism and potentially the progenitor system(s) for SNe Ia, Phillips (1993) find an empirical relationship between the decline rate of SNe Ia light curves and their absolute magnitudes, which holds for a majority of SNe Ia.Conversely, UV observations of SNe Ia show greater spectral diversity than in the optical (Ellis et al. 2008;Foley et al. 2008;Walker et al. 2012), and photometrically, SNe Ia can be grouped into NUV-red and NUV-blue classes based on the UV−optical color curves (Milne et al. 2013).Furthermore, the UV colors have an intrinsic scatter that is incompletely explained by extinction and redshift (Brown et al. 2017) and are redder than the Kasen & Plewa (2007) predictions for asymmetric explosions (Brown et al. 2018).Pan et al. (2020) claim a correlation between UV flux and host-galaxy metallicity; however, Brown & Crumpler (2020) were unable to confirm this correlation.
In addition to UV differences between spectroscopically normal SNe Ia, some SNe Ia show significant spectroscopic differences from normal SNe Ia yet remain on the Phillips (1993) relation (e.g., SN 1991T Phillips et al. 1992Filippenko et al. 1992a, and SN 1991bg Filippenko et al. 1992b), and other spectroscopically different SNe Ia deviate from the Phillips (1993) relationship (e.g., SN 2002es Ganeshalingam et al. 2012, SN 2003fg Howell et al. 2006, and SN 2006bt Foley et al. 2010).The growing number of these extreme SNe Ia offers a unique chance to probe the generally homogeneous nature of SNe Ia.
Two particularly interesting SNe Ia subtypes are 2002es-like and 2003fg-like SNe Ia. 2002es-like SNe Ia are subluminous and have spectra that are similar to 1991bg-like SNe Ia with Si ii λ5972, O i, and Ti ii features near t max B .They lack a secondary i-band rebrightening, similar to other subluminous SNe Ia.However, unlike other fast-declining, subluminous SNe Ia, 2002eslike SNe Ia decline at a rate similar to normal SNe Ia (Taubenberger 2017).One model for these objects is in the DD scenario with CO WDs whose masses sum greater than M ch and violently merge (Pakmor et al. 2010;Kromer et al. 2016).Unfortunately, this class of objects is small with ∼10 objects, and a homogeneous data set does not yet exist.Conversely, 2003fg-like SNe Ia have overluminous and broader optical light curves than normal SNe Ia yet also lack the secondary i-band peak (Taubenberger 2017;Ashall et al. 2021).In the UV, 2003fg-like SNe Ia have different colors than normal and 1991T-like SNe Ia (Brown et al. 2014b), and spectroscopically, they have weaker Ca ii and stronger O i and C ii features.For the first members discovered in this class, the Arnett (1982) relationship yields 56 Ni masses for 2003fg-like SNe Ia exceeding M ch (e.g., Howell et al. 2006;Tanaka et al. 2010;Scalzo et al. 2010;Taubenberger et al. 2011), leading to the colloquial "Super-Chandresekhar" designation (e.g., Howell et al. 2006;Chen & Li 2009;Scalzo et al. 2010;Silverman et al. 2011;Scalzo et al. 2012;Das & Mukhopadhyay 2013;Taubenberger et al. 2013b;Hsiao et al. 2020).However, lower luminosity 2003fg-like SNe Ia with derived 56 Ni masses below M ch have also been discovered, suggesting that not all members of this class are "super-Chandrasekhar" explosions (e.g., Hicken et al. 2007;Chakradhari et al. 2014;Chen et al. 2019;Lu et al. 2021).Potential models for 2003fg-like SNe Ia included a rapidly spinning CO WD (Langer et al. 2000;Yoon & Langer 2005;Hachisu et al. 2012), the merger of two WDS (Bulla et al. 2016), and the CD scenario (e.g., Hoeflich & Khokhlov 1996, Ashall et al. 2021, Maeda et al. 2023).
Despite their spectroscopic and photometric differences, both 2002es-like and 2003fg-like have had observations of rising light curve bumps1 (Cao et al. 2015;Miller et al. 2020;Jiang et al. 2021;Dimitriadis et al. 2023;Srivastav et al. 2023a,b); conversely, other spectroscopic subtypes and normal SNe Ia show deviations from a single-component power-law rise, and it remains unclear if these events are connected to the rising light curve behavior of 2002es-like and 2003fg-like SNe Ia.While theoretical models of normal SNe Ia are similar at peak B-band magnitude, at early times (0-5 days after explosion), various models make different predictions about the light curve shape (e.g., Kasen 2010; Piro & Nakar 2013, 2014;Piro & Morozova 2016;Polin et al. 2019;Magee & Maguire 2020;Maeda et al. 2023).Thus, probing the earliest stages of SNe Ia explosions can provide otherwise unavailable information about the underlying physics.
In this paper, we analyze the UV absolute luminosities and colors of SNe Ia with different rising light curve behavior.§2 details the observations, data reduction process, and sample selection.§3 presents the absolute magnitudes of our sample, and §4 presents the UV and optical color curves.Discussion is in §5, and conclusions are presented in §6.Throughout this work, we adopt H 0 = 73 km sec −1 Mpc −1 , Ω m = 0.30, and Ω vac = 0.70.

DATA
We perform an exhaustive literature search and identify 42 SNe Ia with pre-peak ( t < −5 days) Swift UV photometry and early-time (t < −14 days) optical observations which can be used to constrain the shape of the rising light curve.We relax the optical phase constraint if the SN Ia rises over four magnitudes after discovery (and thus must be young; iPTF13ebh Hsiao et al. 2015and 2019ein Pellegrino et al. 2020;Kawabata et al. 2020) or if there is a non-detection within two days of the first detection (this is only for some SNe Ia in Burke et al. 2022b who use DLT40 discoveries).Of the 42 SNe Ia with pre-peak Swift observations, we exclude five due to issues with host-galaxy contamination or low signalto-noise.Details for all SNe Ia in our sample, including reasons for exclusion from the subsequent analyses, are provided in Appendix A along with values for t max B and extinction.Table 1 lists the SNe Ia in the final sample along with their redshift, distance, early-time light curve category, and host galaxy name and morphology.
Swift photometry is taken from the Swift Optical/UV Supernova Archive (SOUSA; Brown et al. 2014a), which uses the Breeveld et al. (2011) Vega magnitude system zero points that update the original (Poole et al. 2008) zero points.Either a 3-or 5-arcsecond aperture is used to perform photometry with aperture size chosen to maximize S/N.The host galaxy counts from a post-SN template image are subtracted to produce the final photometry.Three exceptions for host-galaxy subtraction are SN 2021hpr and SN 2021aefx, which had galaxy flux from pre-explosion Swift observations subtracted, and SN 2022ilv which does not have a clear host galaxy.Observations obtained within 0.75d are combined with a weighted average to increase the S/N in the final photometry.
Two Swift filters, U V W 2 and U V W 1, have transmission functions extending into the optical wavelengths (i.e., a red leak).This creates a broader distribution of photons from the UV to the optical (Brown et al. 2010).The relative similarity of SNe Ia in the optical means that any peculiar behavior comes from the UV regime.Still, the effect of UV spectral variations is diluted in the U V W 2 filter compared to the neighboring U V M 2 filter.We use all filters in this work.
To calculate absolute magnitudes in each of the six Swift filters, we first correct for the distance and then Milky Way and host-galaxy extinction using values derived in the literature for each object.We adopt a Cardelli et al. (1989) extinction law to convert A V into a filter-specific extinction estimate for each Swift filter.
We also include a comparison sample of 2003fg-like SNe Ia that lack early-time optical observations yet still have Swift photometry.The comparison sample consists of SN2009dc (Yamanaka et al. 2009;Silverman et al. 2011;Taubenberger et al. 2011), SN2012dn (Chakradhari et al. 2014;Taubenberger et al. 2019)

OPTICAL AND UV ABSOLUTE MAGNITUDES
The 2002es-like and 2003fg-like SNe Ia in our sample have unique light curves, as shown in Figure 1.First, the two 2002es-like SNe Ia (iPTF14atg and SN 2019yvq) are underluminous at optical wavelengths yet distinctly overluminous in the UV, especially in the U V M 2 band.Second, the 2003fg-like SNe 2020hvf, 2021zny, and 2022ilv are overluminous in both optical and UV wavelengths.This behavior conflicts with the paradigm that more luminous SNe Ia should be powered by more 56 Ni which in turn increases the opacity, reducing the ratio of UV to optical emission (Lentz et al. 2000;Walker et al. 2012;DerKacy et al. 2020).The observed UV brightness may be from the shock heating of an envelope of H/He-devoid material around the SN (Piro & Morozova 2016;Maeda et al. 2023), which is consistent with the findings of Ashall et al. (2021) who found the most likely progenitor system was one within an envelope.
Finally, SNe 2020hvf and 2021zny peak2 in the UV much earlier than the other SNe Ia.While iPTF14atg    and SN 2019yvq both have UV peaks that are approximately concurrent with their optical maxima, the U V M 2 peaks of SN 2020hvf, and SN 2021zny are at least ∼10 days earlier than the optical peak.We fit the U V M 2-and B-band light curves with the templateindependent Gaussian process method in SNooPy (Burns et al. 2011(Burns et al. , 2014) ) to find the respective peak times, t U V M 2 max and t B max .For iPTF14atg and SN 2019yvq, t U V M 2 max is ∼3.6 days and ∼0.9 days prior to t max B , respectively.For SN 2020hvf and SN 2021zny, t U V M 2 max is 8.9 days and 10.8 days prior to t max B , respectively.

OPTICAL AND UV COLORS
Figure 2 shows each unique color permutation of the Swift filters.We observe two disparate UV color evolutions in Figure 2, so we split our sample into two groups based on the UV color evolution.To avoid confusion with the existing terms of UV-blue and UV-red from Milne et al. (2013), we adopt different terms to refer to these groups.We define Group 1 as SNe Ia with U V M 2 − U V W 1 > 1.5 mag at t = −10 days, and the inverse is true for Group 2. Interestingly, these groupings also correspond to differentiation by spectral classification, with Group 2 consisting solely of 2002eslike and 2003fg-like SNe Ia, whereas Group 1 consists of other spectral subtypes (e.g., normal SNe Ia, 1999aa-like SNe Ia, fast-declining/transitional SNe Ia).The majority of SNe Ia are in Group 1. Group 2 consists of the 2003fg-like SNe Ia reference sample and iPTF14atg, SN 2019yvq, SN 2020hvf, SN 2021zny, and SN 2022ilv.

UV Colors
In the top row of Figure 2, Group 2 SNe Ia are bluer than Group 1 SNe Ia in the U V W 2 − U V W 1 and U V M 2 − U V W 1 color curves, whereas in the U V W 2 − U V M 2 Group 2 SNe Ia are redder.This is due to the dilution of the UV excess due to the Swift U V W 2 filter transmission, i.e., the same amount of UV flux in the U V W 2 filter will have a small effect on the UV+optical flux, whereas the excess flux is a larger percentage in the U V M 2 filter.In the After t max B , these same SNe Ia remain redder than other SNe Ia.
While Group 2 SNe Ia are slightly separated from the rest of the sample in the U V W 2 − U V W 1 color curve, they are closer to the rest of the sample in The increased similarity between Group 1 and Group 2 SNe Ia in the U V W 2 − U V W 1 color curve may originate from the red leak.In extreme cases, the optical component from the red leak provides over half of the total flux (e.g., Brown et al. 2010).Thus, contamination from optical light dilutes the observed difference from additional UV flux.The difference is still observed with filters with the red leak, which demonstrates that abnormal behavior does not arise from the red leak.

UV−Optical Colors
The U V W 2 − U/B/V color curves are all characterized by the same rapid redward ascent of Group 2 SNe Ia.Prior to t max B , Group 2 SNe Ia are all ≲2.6 mag, whereas Group 1 SNe Ia have 2.5 mag ≤ (U V W 2 − U ) ≤ 3.5 mag.After t max B , Group 2 SNe Ia are on the blue edge of the Group 1 distribution.Like the U V W 2 − U/B/V color curves, Group 2 SNe Ia become similar to the rest of our sample in the U V W 1 − U/B/V shortly after explosion.Overall, the U V W 1 colors evolve in the same manner as the U V W 2 filter.
Because the U V M 2 filter does not have a red leak, this filter is the best UV probe, so the U V M 2−U/B/V color curves are the most important to consider.Similar to the other UV−optical colors, Group 2 SNe Ia quickly rise redward in the U V M 2 − U/B/V color curves, initially dominated by the UV.The U V M 2−U color curve shows that even at t max B , Group 2 SNe Ia have different colors, and this difference persists until t ≈ +15 days.The U V M 2 − B and U V M 2 − V color curves are generally similar to the U V M 2 − U color curve.

Optical Colors
There are two phenomena in the optical color curves shown in the bottom row of Figure 2. First, Group 2 SNe Ia are not as uniform as in the UV, and second, Group 2 SNe Ia are not wholly different than Group 1 SNe Ia near and after t max B .There are several interesting features in the optical colors.First, in the U − B and U − V colors, SN 2019yvq is redder than other Group 2 SNe Ia.Second, iPTF14atg evolves similarly to SN 2019yvq when −15 ≤ t ≤ −7 days, but then changes to evolve similarly to the other Group 1 members when t ≥ −7.Either SN 2019yvq is evolving on an earlier time scale than the other bump SNe Ia and other early-time SNe Ia, similar to the V − r color evolution in subluminous SNe Ia (see Figure 8 in Hoogendam et al. 2022), or the mechanism driving the first inflection point in the U − B and U − V colors for iPTF14atg may be absent from SN 2019yvq.Finally, in the B − V color curve, it is difficult to differentiate Group 2 SNe Ia and the 2003fg-like SNe Ia from Group 1 SNe Ia in our sample.

The Interesting Bump Cases
We separate the rising light curves of SNe Ia into three categories as shown in Figure 3: "single", "double", and "bump".Single SNe Ia have rising light curves well fit by a single power law, whereas double SNe Ia have rising light curves better fit by a broken or two-component power law.Bump SNe Ia have non-monotonic light curve bumps in the UV and/or the optical (i.e., the flux decreases by at least 1σ between any two epochs during the light curve rise).Figure 3 and the left-hand panel of Figure 4 elucidate these different rising light curve behaviors using idealized and observed light curves, respectively.The SNe Ia with bluer UV colors, iPTF14atg, SN 2019yvq, SN 2020hvf, SN 2021zny, and SN 2022ilv, are the only bump SNe Ia, and no spectroscopically normal SNe Ia display bumps by our definition in their rising light curve, despite composing the majority of our sample and the majority of SN Ia volumetrically (Desai et al. 2023).Thus, the color scheme in Figures 1 and 2 also correlates with rising light curve morphology.The blue and orange points correspond to SNe Ia with a rising light curve bump and the grey points correspond to SNe Ia without a rising light curve bump.However, we note that simultaneous high-cadence, highsignal-to-noise, early-time optical and UV data do not exist for most of the supernovae in our sample since the current observations come from heterogeneous survey and follow-up efforts.For example, there are no simultaneous UV observations of 2003fg-like SNe Ia since all the bumps are observed in the optical.Conversely, both 2002es-like SNe Ia in our sample have observed UV bumps but iPTF14atg does not show a clear bump in its optical light curves due to a gap in ground-based optical coverage at these epochs and low signal-to-noise in in the Swift optical observations.Ultimately, simultaneous high-cadence, high-signal-to-noise optical and UV data are needed for further study of SNe Ia rising light curves.
When considering the colors of the bump SNe Ia, the observed dichotomy of UV colors is inconsistent with the excess luminosity originating from interaction with a companion which is viewing-angle dependent (Kasen 2010; Brown et al. 2012).This implies companion interaction and other viewing-angle dependent models are insufficient to explain both two-component rising light curves and rising light curve bumps.Thus, monotonically and non-monotonically rising light curves may have different physical origins.Despite our small sample, we can disfavor companion interaction as the cause of the early-time light curve bumps.Using the same viewingangle argument as Burke et al. (2021) 2003fg-like SNe Ia is 1 in 10 5 .Underlying this claim is the assumption that the viewing angle is independent of the spectral subtype, which seems likely to be true but observations do not guarantee this is reality, and that the companion interaction results in the observables predicted by Kasen (2010).Thus, at the population level, the observed rate of SNe Ia with rising light curve bumps is too high to be fully explained by companion interaction.
Intriguingly, SN 2020hvf, SN 2021zny, and SN 2022ilv all display rising light curve bumps (Jiang et al. 2021;Dimitriadis et al. 2023;Srivastav et al. 2023a), and Figure 2 shows that the color evolutions of SN 2020hvf, SN 2021zny, and SN 2022ilv are consistent with the regular 2003fg-like SNe Ia in our comparison sample.This would naively suggest that other 2003fg-like SNe Ia have unobserved early-time light curve bumps.Such a conclusion remains speculative due to the small sample size.Initial analysis of SN 2022pul suggests there may not be a rising light curve bump Kwok et al. 2023;Siebert et al. 2023, but an in-depth study of the rising light curve is yet to be published.Similarly, iPTF14atg, SN 2019yvq are the only 2002es-like SNe Ia with UV data observed early enough to detect a bump.Three other 2002es-like SNe Ia, iPTF14dpk (Cao et al. 2016 we can obtain preliminary statistical conclusions despite the small-number statistics.
Assuming the SNe Ia in our sample are representative, we can compute the fraction of 2002es-like and 2003fglike SNe Ia that display a rising light curve bump at a 90% confidence level.For 2002es-like SNe Ia, observing 2 rising light curve bumps from our 2 object sample implies that at least 32% of early-time 2002es-like SN Ia light curves display a rising light curve bump.Similarly, for our sample of 2003fg-like SNe Ia, at least 47% display a rising light curve bump.Combining our sample of 2002es-like and 2003fg-like SNe Ia and assuming 2002es-like and 2003fg-like SNe Ia arise from the same distribution, then at least 63% of the SNe Ia display a rising light curve bump.Finally, if we include 2002eslike and 2003fg-like SNe Ia that lack UV data in our statistical analysis, then 8 out of 9 SNe display a rising light curve bump, and thus at least 63% of SNe Ia in these subtypes display a rising light curve bump.

A Link Between 2002es-like SNe Ia and 2003fg-like SNe Ia
Brown et al. (2014b) show that 2003fg-like SNe Ia are different from normal as well as 1991T-like SNe Ia in the U V W 1 − V and U V M 2 − U V W 1 colors.This work expands that analysis with color evolution data at t ≤ −10 days for more SNe Ia and the inclusion of three additional 2003fg-like SNe Ia and two 2002es-like SNe Ia.
There are several observational similarities between 2002es-like and 2003fg-like SNe Ia.Both classes lack a secondary i-band maximum (Ashall et al. 2021;Burke et al. 2021), have some members with nebular [O i] emission (Taubenberger et al. 2013b,a;Kromer et al. 2016;Taubenberger et al. 2019;Dimitriadis et al. 2023;Siebert et al. 2023), and have members with C ii absorption of varying strengths in their near-peak optical spectra (Kromer et al. 2013;Cao et al. 2015;Ashall et al. 2021;Li et al. 2023;Siebert et al. 2023) Of the observational similarities between the two subtypes, the shared carbon absorption is perhaps the most intriguing.One proposed source for the C ii feature is a C-rich envelope either from the merger of two CO WDs in the DD scenario (e.g., Moll et al. 2014;Raskin & Kasen 2013;Raskin et al. 2014) or a CO WD and an AGB core in the CD scenario where the AGB star has lost its H and/or He envelope (e.g., Lu et al. 2021;Hsiao et al. 2020;Ashall et al. 2021).Ashall et al. (2021) find a weak linear relationship between the pEW of C ii λ6580 Å and ∆m 15 (B) where faster-declining 2003fg-like SNe Ia have a smaller pEW value.Therefore, if there is a link between the progenitor scenario and/or explosion mechanism of 2002es-like and 2003fg-like SNe Ia, one may expect 2002es-like SNe Ia to follow this relationship with 2002es-like SNe Ia showing weak C ii features.
Observationally, only some 2002es-like SNe Ia display C ii absorption (e.g, SN 2010lp, Kromer et al. 2013;iPTF14atg, Cao et al. 2015;andSN 2016ije, Li et al. 2023), while others completely lack C ii (e.g., PTF10ops, Maguire et al. 2011;PTF10ujn, White et al. 2015;PTF10acdh, White et al. 2015;andSN 2019yvq, Miller et al. 2020).However, 2002es-like SNe Ia without C ii may still have an envelope which could be observed in a lower ionization state of carbon.C i λ10693 Å is the strongest C i feature, and visual inspection of Figure 2 in Burke et al. (2021) suggests the presence of C i in SN 2019yvq.Unfortunately, no other 2002es-like SNe Ia have published NIR spectra near peak light.
If we assume 2002es-like and 2003fg-like SNe Ia are linked, different scenarios can be constructed to explain a potential continuum in either the DD scenario or the CD scenario.The first-order explosion parameters would be the mass of the carbon envelope and the mass of the primary degenerate object (CO WD or CO AGB core, depending on the progenitor scenario).We do not consider potential higher-order effects such as flame speed or burning efficiency in our brief qualitative comparisons.
A larger core mass will produce a more optically luminous explosion with higher ionization features.In this picture, 2002es-like SNe Ia will have smaller core masses because they have lower optical luminosities (cf. Figure 1) and show lower ionization lines (e.g., Taubenberger 2017), and conversely, 2003fg-like SNe Ia will have larger core masses due to their larger optical luminosities (cf. Figure 1) and higher-ionization spectral features (e.g., Taubenberger 2017).Less massive circumstellar material results in weaker C features (i.e., smaller pEWs) since these features trace envelope mass (Ashall et al. 2021), whereas a larger envelope should produce stronger C ii and C i absorption, more luminous peak magnitudes, and lower Si ii velocities.
In addition to a nearby envelope, both a violent merger in the DD scenario and the CD scenario may have material extending out to larger distances (Kashi & Soker 2011;Raskin & Kasen 2013;Hsiao et al. 2020).If that material is either launched in a wind or dynamically ejected at a constant velocity, then the extended material will have a r −2 density profile (e.g., Moriya   et al. 2023).That can be compared to the expected r −3 density profile of the nearby envelope (Piro & Morozova 2016).Studies have investigated the effects of shock heating the envelope and wind separately, but to our knowledge, no study has simultaneously simulated both.Piro & Morozova (2016) and Maeda et al. (2023) show that a nearby envelope with an r −3 density profile is expected to produce a distinct bump in the early-time light curves whereas Moriya et al. (2023) show that a diffuse, r −2 wind may provide persistent additional UV luminosity through maximum light.Qualitatively, the combination of these two effects may reproduce the observed 2003fg-like and 2002es-like light curves with their observed rising light curve bumps and persistent blue UV colors through maximum light.A combined envelope and wind-driven circumstellar medium may produce a correlation between the size of the ∼ 2-day rising light curve bump or the UV luminosity and the combined pEW of the C i and C ii features at early times.Future theoretical and observational work is needed to test these qualitative considerations.
Finally, determining if 2003fg-like SNe Ia originate from the DD or CD scenario is another important open question.There are several predicted observational differences between the DD and CD scenarios.First, the DD scenario is predicted to show [O i] emission during the nebular phase (Pakmor et al. 2012) which may arise from low-velocity, asymmetric oxygen distributions caused by incomplete burning during a violent merger (Mazzali et al. 2022), whereas the CD scenario does not predict [O i] emission.Second, a violent merger in the DD scenario is predicted to be aspherical (Bulla et al. 2016); currently, no CD models offer predictions about polarization but intuitively one could expect these explosions to be spherical.Third and finally, the CD model of Lu et al. (2021) predicts a high X-ray flux due to a fast-receding photosphere and low opacity.This X-ray flux is as of yet undetected in any SN Ia to date (Lu et al. 2021), and future X-ray studies of 2002fg-like and 2003fg-like SNe Ia are also needed.

CONCLUSION
We present UV and optical photometry from Swift compiled in the SOUSA catalog (Brown et al. 2014a) for SNe Ia with early-time optical observations and pret max B UV photometry.The data can be categorized by either the UV color evolution or the rising light curve morphology, and we find that both categorization criteria separate 2002es-like and 2003fg-like SNe Ia from the other spectroscopic subtypes of SNe Ia.Observationally, 1. 2002es-like and 2003fg-like SNe Ia are, on average, UV brighter than other SNe Ia (Figure 1), 2. 2002es-like and 2003fg-like SNe Ia have extreme blue UV colors through 10 days after maximum (Figure 2), and 3. 2002es-like and 2003fg-like SNe Ia are the only spectroscopic subtypes that exhibit rising light curve bumps (Figure 4).
Because 2002es-like and 2003fg-like SNe Ia show similar UV colors and also are the only spectroscopic subtypes to exhibit rising light curve bumps, we examine a potential relationship between these two subtypes.We propose a potential continuum between 2002es-like and 2003fg-like SNe Ia with the following progenitor properties.These SNe Ia may originate in low-metallicity DD or CD scenarios enshrouded by a carbon-rich circumstellar medium (Kromer et al. 2016;Lu et al. 2021;Ashall et al. 2021;Siebert et al. 2023;Kwok et al. 2023).2003fg-like SNe Ia should have higher luminosities and 2002es-like SNe Ia are the lower-luminosity members of the continuum; the rising light curve bumps may arise from shock heating of an inner carbon envelope (e.g., Piro & Morozova 2016;Maeda et al. 2023) and wind-originated outer carbon material may cause the blue UV color evolution (e.g., Moriya et al. 2023).Future theoretical modeling should include two-layer circumstellar medium initial conditions.
With the advent of transient surveys (e.g., All-Sky Automated Survey for SuperNovae (ASAS-SN; Shappee et al. 2014;Kochanek et al. 2017;Hart et al. 2023), Asteroid Terrestrial-impact Last Alert System (ATLAS; Tonry et al. 2018), Panoramic Survey Telescope And Rapid Response System (Pan-STARRS;Chambers et al. 2016 and Zwicky Transient Facility (ZTF;Bellm et al. 2019), rapid classification groups (e.g., SCAT; Tucker et al. 2022c;ePESSTO+ Smartt et al. 2015) and dedicated follow-up groups (e.g., POISE; Burns et al. 2021;YSE Jones et al. 2021), many more SNe Ia will be quickly discovered and rapidly observed after the explosion.This is especially important for 2002es-like SNe Ia, which may initially be spectroscopically misclassified as 1991bg-like SNe Ia.Because of this, rapid followup programs with Swift (e.g., Brown et al. 2023) are doubly important.First, such programs increase the sample of well-observed SNe Ia in the UV, and more importantly, 2002es-like or 2003fg-like SN Ia can be differentiated from other subtypes with a single Swift epoch The majority of SNe Ia in our sample are well documented in the literature due to their early discoveries and intense follow-up campaigns compared to the majority of SNe Ia.In this section, we comment on our early-time light curve categorization for each SN Ia, which categorizes the SNe Ia sample into three groups: single, double, and bump as discussed in Section 2. We also provide a discussion on adopted A V values as well as t max B .For some SNe Ia, the only literature extinction estimate is from the Poznanski et al. (2012) Na i D pEW relationship, which we adopt with the reported uncertainties (although see Phillips et al. 2013).Lastly, spectroscopic classification data were taken from TNS or individual object papers, as available.The Milky Way extinction is E(B − V ) M W = 0.01 mag (Schlafly & Finkbeiner 2011).

A.1. SN 2009ig
SN 2011fe was an incredibly nearby normal SN Ia Nugent et al. (2011).Located in M 101 (NGC 5457) at 6.4 Mpc (Shappee & Stanek 2011), SN 2011fe is one of the most nearby SN Ia to date.As such, extensive searches for companion interaction under the Kasen (2010) models have been performed, all yielding no evidence of companion interaction (Li et al. 2011;Röpke et al. 2012;Brown et al. 2012;Shappee et al. 2013bShappee et al. , 2017;;Tucker et al. 2022a,b) or even a surviving companion (Lundqvist et al. 2015) which are predicted to be overluminous (Shappee et al. 2013a).Given the large sample of early data indicating strong agreement with the fireball model, we categorize SN 2011fe as a single SN Ia.
SN 2012cg was initially reported by Silverman et al. (2012).Subsequent studies found evidence for (Marion et al. 2016) and against (Shappee et al. 2018) companion interaction.Irrespective to the mechanism of the early-time emission, it is clear that emission beyond the predicted f ∝ t n fireball model was detected.Thus, we categorize SN 2012 in the doube category.A later study by Contreras et al. (2018) found a similar t max B on MJD 56242.6.Contreras et al. (2018) examined the host-galaxy extinction using both the Na i D line from different spectra than Childress et al. (2013) as well as high quality Carnegie Supernova Project II (Phillips et al. 2019) photometry.This analysis by Contreras et al. (2018) resulted in a final E(B − V ) = 0.03 ± 0.03 mag.We use the values from Contreras et al. (2018) in our analysis.The Milky Way extinction is E(B − V ) M W = 0.02 mag (Schlafly & Finkbeiner 2011).
There is a similarity between SN 2012fr and SN 2014J shown in Contreras et al. (2018), which suggests that the broken power law fitted to SN 2014J by Zheng et al. (2014)  to be on MJD 56295.6, and they claim host-galaxy extinction is negligible.We accept their claim as valid given the presented B − V color curve in their Figure 1, as well as the lack of visible Na i D in their spectra.However, a negligible extinction in the optical will be larger in the UV, so we assume a host galaxy extinction of A V = 0.01±0.01mag to extrapolate to the Swift UV filters.The Milky Way line of sight extinction from Schlafly & Finkbeiner (2011) Yamanaka et al. (2014) present a smooth, single component rise, so we categorize SN 2012ht as a single SN Ia, similarly to Jiang et al. (2018).
A.6.LSQ12gdj to be on MJD 56500.7 and a host-galaxy extinction from fitting the Na i D line (Poznanski et al. 2012) to be E(B − V ) = 0.15 mag.The Milky Way extinction is E(B − V ) M W = 0.13 mag (Schlafly & Finkbeiner 2011).Zheng et al. (2013) found the best fit to the earlytime light curve of SN 2013dy was a broken power law.Thus, we categorize SN 2013dy as a double SN Ia.In their work, Jiang et al. (2018)  A single power-law rise fits the early-time light curve of SN 2013gy (Holmbo et al. 2019), so we categorize it as a single SN Ia.
A.9. iPTF13dge iPTF13dge was studied by Ferretti et al. (2016)  In their analysis, Ferretti et al. (2016) did not include fits to the early-time light curve, however, we find no evidence for a two-component power-law rise hence we categorize iPTF13dge as a single SN Ia.
While Hsiao et al. (2015) did not directly fit the earlytime rise of iPTF13ebh, a single-component power-law fit is a reasonable conclusion from their comparison to normal and 1991bg-like models.Thus, like Jiang et al. (2018), we categorize iPTF13ebh as a single SN Ia.

A.11. ASASSN-14lp
SNooPY fits performed by Shappee et al. (2016) show that the B-band maximum t max B of ASASSN-14lp was on MJD 57015.3 and had a host-galaxy extinction of E(B − V ) HG = 0.33 ± 0.06 mag.Fits performed by Shappee et al. (2016) find that ASASSN-14lp is in good agreement with a single-component power-law earlytime light curve rise.The Milky Way extinction is E(B − V ) M W = 0.02 mag (Schlafly & Finkbeiner 2011).

A.12. iPTF14atg
Determinations for the t max B and E(B − V ) HG for iPTF14atg come from two different sources.First, Cao et al. (2015) determine that t max B of iPTF14atg occurred on MJD 56799.2, but they do not provide an estimate for the host-galaxy extinction in their manuscript.Second, Kromer et al. (2016) determine the host-galaxy extinction for iPTF14atg is A B = 0.00 ± 0.02 mag based on the Na i D line.We adopt a slightly different value of A V = 0.01 ± 0.02 mag, which is consistent with the Kromer et al. (2015) value but is in the same band as all the other SNe Ia in our sample.The Milky Way extinction is E(B − V ) M W = 0.01 mag (Schlafly & Finkbeiner 2011).Cao et al. (2015) and Kromer et al. (2016) report early-time light curve bumps, thus we classify iPTF14atg as a bump SNe Ia.
While the early-time Swift photometry in Smitka et al. (2015) may have an early-time bump, we found no bump after redoing the photometry.Thus, we categorize iPTF14bdn as double SN Ia.

A.14. SN 2014J
We adopt t max B from Foley et al. (2014) which is MJD 56690.0.Unfortunately, determining the host-galaxy extinction is not so straightforward.
SN 2014J is one of the most heavily extincted SNe Ia to date.As such, there are a plethora of estimates on the host-galaxy extinction for this object.Amanullah et al. (2014) perform various extinction fits to their data.Their power-law fit ( ) yielded A V = 1.85 ± 0.11 mag, whereas their MW-like fit based on the Fitzpatrick (1999) parametrization yielded E(B −V ) = 1.37±0.03mag with R V = 1.4±0.1.Alternatively, Ashall et al. (2014) determine the host-galaxy extinction via selecting the A V and R V values which optimize their abundance tomography models.This yields E(B − V ) = 1.2 mag and R V = 1.38.Goobar et al. (2014) present two extinction values, one from SNooPy fits and the other based on the Phillips et al. (2013) method.The SNooPy fits yield E(B − V ) = 1.22 ± 0.05 mag with R V = 1.4 ± 0.15 and the Phillips et al. (2013) method yields A V = 2.5 ± 1.3 mag Finally, Foley et al. (2014) presents several further calculations of A V for SN 2014J.First, by using distant independent optical-infrared colors, they derive A V = 1.95 ± 0.09 mag.Second, using the Phillips et al. (2013) relation, they derive A V = 1.8±0.9mag from their high-resolution spectrum.Finally, using the color excess method, they derive A V = 1.91 mag using a Fitzpatrick (1999) parametrization and A V = 1.82 mag using a Cardelli et al. (1989) parametrization.We elect to adopt the Foley et al. (2014) value measured by the high-resolution spectrum, which is also consistent with their color excess fits as well as the power-law model from Amanullah et al. (2014).Specifically, we take the value to be 1.8 ± 0.9 mag.Siverd et al. (2015) find peak light on MJD 56690.62 using data from the Kilodegree Extremely Little Telescope North (Pepper et al. 2007;Siverd et al. 2012).Zheng et al. (2014) demonstrate a two-component power-law fit to the early-time light curve rise, and Goobar et al. (2015) also found that multiple components better fit the rising light curve.Thus, we categorize SN 2014J as a double SN Ia.Unfortunately, the extinction is so severe in the UV that SN 2014J is not comparable in the UV to other SNe Ia, thus we do not include it in plots and only include this discussion for completeness of the early-time SN Ia sample.

A.15. SN 2015F
Using MLCS2k2 (Jha et al. 2007), Im et al. (2015) find E(B − V ) hg = 0.04 ± 0.03 mag and t max B on MJD 57105.98.Cartier et al. (2017) estimate the t max B to be MJD 57106.5.Using the optical and near-infrared colors, they calculate various E(B−V ) values, the weighted average of which they calculate to be E(B − V ) HG = 0.09 ± 0.02 mag.Their individual E(B − V ) calculations use bot the optical (Phillips et al. 1999) and nearinfrared (Krisciunas et al. 2004a,b) colors.The Milky Way extinction is E(B − V ) M W = 0.17 mag (Schlafly & Finkbeiner 2011).Im et al. (2015) fit both single and double power laws to the early-time data of SN 2015F and find that the single-component power-law gives the best fit and that the double power law fit converges to the single power law result.Thus, we categorize SN 2015F as a single SN Ia, which is consistent with the categorization of Jiang et al. (2018). A.16. SN 2015bq Li et al. (2022) determined t max B was on MJD 57084.1.By performing fits using the SALT2 modelling program (Guy et al. 2007), Li et al. (2022) calculate E(B − V ) HG = 0.15 ± 0.07 mag.The Milky Way extinction is E(B − V ) M W = 0.01 mag (Schlafly & Finkbeiner 2011).
An early-time flux excess is claimed by Li et al. (2022).Their data show SN 2015bq has early-time flux which is

Figure 1 .
Figure 1.Absolute Vega magnitudes of our sample.2002es-like SNe Ia are orange, 2003fg-like SNe Ia are blue, and other SNe Ia spectral sub-types are gray.Note that errors within an individual SN light curve are correlated because the distance and extinction uncertainties are added in quadrature with photometric uncertainties.

Figure 2 .
Figure 2.Each unique Swift observed color curve of our sample.The color scheme is the same as Figure 1.Row 1: Observed UV colors of our sample.Row 2: Observed U V W 2 − U/B/V colors of our sample.Row 3: Observed U V M 2 − U/B/V colors of our sample.Row 4: Observed U V W 1 − U/B/V colors of our sample.Row 5: Observed optical colors of our sample.

Figure 3 .
Figure3.Idealized light curve categorization examples for bumps (dot-dash), double power law (dashed), and single power law (solid).The peak brightness for each category has been vertically offset for visual clarity.

Figure 5 .
Figure5.Enlarged color curves of the sample colored by the shape of the rising light curve: single power law (cyan), double power law (purple), and bump (khaki).Left: The U V M 2 − U V W 1 color curve; all bump SNe Ia are below ∼1.5 mag until t ≈ 10 days after t max B .Single and double SNe Ia are above ∼1.5 mag until t ≈ 10 days.We expect future SN with U V M 2 − U V W 1 colors below this region to be 2002es-like or 2003fg-like.Right: The B − V color curve shows minimal distinction between the different groups if any at all.
then the SN Ia is either 2002es-like or 2003fg-like, with the difference between 2002es-like and 2003fg-like SNe Ia determined by peak absolute magnitude.Furthermore, the planned ULTRASAT satellite (Sagiv et al. 2014) will provide many transient discoveries and early-time UV light curves, which are crucial for SNe Ia science and may display early-time light curve bumps.Ultimately, further early-time observations across the electromagnetic spectrum are needed of 2002es-like and 2003fg-like SNe Ia.

Foley
et al. (2012) report A V = 0.01±0.01mag for the host galaxy of SN 2009ig and t max B on MJD 55080.04.The Milky Way extinction is E(B − V ) M W = 0.03 mag (Schlafly & Finkbeiner 2011).Foley et al. (2012) reported early-time observations of SN 2009ig, a normal SN Ia.After subtracting a Arnett (1982) f ∝ t 2 fireball model, SN 2009ig has positive residuals which are indicative of excess flux above what is expected with the fireball model.However, Foley et al. (2012) found the rise was well fit with a single component power law.Given this acceptable single-component fit, we categorize SN 2009ig as an SN Ia without earlytime excess flux (i.e., as a single SN Ia).This determination is similar to Jiang et al. (2018), who also categorize SN 2009ig as having no early excess in either the UV or the optical.A.2. SN 2011fe Pereira et al. (2013) report t max B on MJD 55814.51 and a host galaxy extinction of E(B − V ) = 0.03 ± 0.04 mag.
A.4. SN 2012fr   Childress et al. (2013)  presented the first study of SN 2012fr and derived t max B on MJD 56243.0 and an upper limit of E(B − V ) < 0.015 mag via the Na i D line.
Scalzo et al. (2014) find t max B on MJD 56252.5 and a host-galaxy extinction of E(B − V ) HG = 0.02 ± 0.08 from the(Lira et al. 1998) law.Using SNooPy, they find E(B − V ) HG = 0.01 ± 0.01 with R V = 1.66 ± 1.66.We opt to use the extinction from the(Lira et al. 1998) law assuming an R V = 3.1.The Milky Way extinction is E(B − V ) M W = 0.02 mag(Schlafly & Finkbeiner 2011).Scalzo et al. (2014) fit a two-component Arnett (1982) with a shock component model to the bolometric light curve of LSQ12gdj.Thus, we categorize LSQ12gdj as a double SN Ia.A.7. SN 2013dy Zheng et al. (2013) derive t max B also classify SN 2013dy as an early-excess SN Ia.A.8. SN 2013gy Holmbo et al. (2019) present the discovery and an analysis of SN 2013gy where they find E(B − V ) HG = 0.11 ± 0.06 mag and t max B on MJD 56648.5 from SNooPy (Burns et al. 2011, 2014) fits.The Schlafly & Finkbeiner (2011) Milky Way extinction toward SN 2013gy is E(B− V ) M W = 0.05 mag.
who found t max B occurred on MJD 56558.0.Ferretti et al. (2016) determined there was minimal host-galaxy toward iPTF13dge.They derived a value of E(B − V ) HG = 0.03 ± 0.04 mag.The Milky Way extinction for SN 2013gh is E(B − V ) M W = 0.08 mag (Schlafly & Finkbeiner 2011).
(Srivastav et al. 2023b 2022ywc(Srivastav et al. 2023b), and SN 2022vqz  (Xi et al. 2023), have reported rising light curve bumps but no UV data.While the current sample is small, Light curves grouped by category.For several SNe Ia, data were not made available after publication.In these cases, the figures containing the data were digitalized.Left: Apparent magnitudes with offset.The top and middle panels use B/V -band data, whereas the bottom panel uses U V M 2 data for iPTF14atg, SN 2019yvq, ATLAS data for SN 2020hvf and SN 2022ilv, and TESS data for SN 2021zny.The left panel demonstrates the differences between single, double, and bump SNe Ia.If not observed early enough, bump SNe Ia may look like double SNe Ia.Likewise, double SNe Ia not observed early enough would look like single SNe Ia.Our sample selection should prevent this issue from influencing our results because we select SNe Ia with data prior to −15 d.Right: Extinction-corrected absolute magnitudes.For the single and double panels, B-band data is used where available.Otherwise, g-or V -band data is used.For the bump panel, the data is the same as the left-hand side.The upper limits are taken from the latest available 3σ upper limit or detections with S/N < 5, regardless of the bandpass.Interestingly, at concurrent times the SNe Ia with two-component rises are more luminous than SNe Ia with single-component rises, and additionally, the bump 2003fg-like SNe Ia appear to have longer rise times than the bump 2002es-like SNe Ia.
. Lastly, 2003fg-like SNe Ia prefer metal-poor, young environments (e.g., Lu et al. 2021, Galbany et al. in prep), whereas White et al. (2015) suggest 2002es-like SNe Ia prefer older, elliptical galaxies, but an in-depth study has not yet been performed for the local environments of 2002es-like SNe Ia.
Jiang et al. (2018)018)2012fr well.We categorize SN 2012fr as a double SN Ia, which is different than the categorization ofJiang et al. (2018), who categorize SN 2012fr as having no excess.Our determination is based on information provided inContreras et al. (2018)whichJiang et al. (2018)may not have had available to the,.