HST/COS Far-ultraviolet Spectroscopic Analysis of U Geminorum Following a Wide Outburst*

Patrick Godon; Michael M. Shara; Edward M. Sion; David Zurek

doi:10.3847/1538-4357/aa9616

1. Introduction: U Geminorium—The Prototypical Dwarf Nova

Cataclysmic Variables (CVs) are semi-detached interacting binaries in which a white dwarf (WD) accretes matter from a Roche lobe-filling late-type companion. When the WD has a weak or negligible magnetic field, the matter is accreted by means of a disk in which it dissipates its gravitational potential energy. The most common class of CVs are the dwarf novae (DNe), found mostly in a state of low accretion, when the disk is dim and the WD dominates in the ultraviolet (UV) band. Every few weeks to months, DNe undergo semi-regular outbursts (lasting days to weeks), in which the system brightens by 3–5 visual magnitudes, as the mass transfer rate increases due to a thermal instability in the disk. During outbursts, the accretion disk totally dominates the systems in the UV and optical bands.

In 1855 December, the DN U Gem became the first CV to be discovered (Hind 1856), and as such it also became the prototypical "UG" subtype of DNe (i.e., DNe with period P > 3 hr). For that reason, it has been extensively studied, observed numerous times in the X-ray (e.g., Mason et al. 1988; Szkody et al. 1996), extreme UV (e.g., Long et al. 1996), UV (e.g., Long et al. 1993), optical (e.g., Zhang & Robinson 1987; Unda-Sanzana et al. 2006; Ecchevarría et al. 2007), and infrared (e.g., Panek & Eaton 1982), and it has even been modeled using three-dimensional hydrodynamics (e.g., Kunze et al. 2001) as well as within the context of the restricted three-body problem (e.g., Smak 2001). Consequently, we restrict ourselves here to summarizing only the characteristics of the system that are directly relevant to the present work, with a limited number of references (as the complete list is extensive).

U Gem has a period of 0.1769061911 days (∼4 hr 15 minutes; Marsh et al. 1990; Ecchevarría et al. 2007; Dai & Qian 2009) and a mass ratio q = 0.35 ± 0.05 (Long & Gilliland 1999; Ecchevarría et al. 2007) with a WD mass M_wd ∼ 1.1–1.2 M_⊙ (Sion et al. 1998; Long & Gilliland 1999). It has a distance of 100.4 ± 3.7 pc, which has been determined astrometrically by Harrison et al. (2004) based on a reanalysis of the Hubble Space Telescope (HST) Fine Guidance sensor parallaxes. The system is a partially eclipsing binary (Kraft 1962; Krzeminski 1965) in which the disk undergoes partial eclipses (around phase 0.0 ± 0.1) by the donor, the hot spot undergoes full eclipses (just after phase 0.0) by the donor, while the WD is never eclipsed (Smak 1971; Warner & Nather 1971; Marsh et al. 1990; Ecchevarría et al. 2007). For this to happen with a mass ratio q ∼ 0.35, the inclination has to be in the range 62°–74° (Long & Gilliland 1999; Long et al. 2006); the value derived in the literature is between i = 67° and i = 72° (Long & Gilliland 1999; Long et al. 2006; Smak 2001; Unda-Sanzana et al. 2006), confirming the early estimate by Zhang & Robinson (1987) of i = 69 fdg 7 ± 0 fdg 7. The system parameters are listed in Table 1 together with their references.

Table 1. System Parameters

Parameter	U Gem	References
${M}_{1}\, \langle {M}_{\odot } \rangle$	1.1–1.2	Sion et al. (1998), Long & Gilliland (1999)
${M}_{2}\, \langle {M}_{\odot } \rangle$	0.41–0.42	Long & Gilliland (1999), Smak (2001), Ecchevarría et al. (2007)
q = M₂/M₁	0.35–0.362	Long & Gilliland (1999), Smak (2001), Ecchevarría et al. (2007)
$i\, \langle \deg \rangle$	67–72	Zhang & Robinson (1987), Long & Gilliland (1999), Unda-Sanzana et al. (2006)
$P\, \langle {\rm{h}}{\rm{r}} \rangle$	4.24574858	Marsh et al. (1990), Ecchevarría et al. (2007), Dai & Qian (2009)
m_v	8.2–13.9	Szkody & Mattei (1984), Ritter & Kolb (2003)
E(B − V)	0.04 ± 0.01	Verbunt (1987), la Dous (1991), Bruch & Engel (1994)
$d\, \langle {\rm{p}}{\rm{c}} \rangle$	100.4 ± 3.7	Harrison et al. (2004)

Download table as: ASCII Typeset image

U Gem goes into outburst every ∼118 days, brightening from m_v = 14.9 to m_v = 8.2 for an average outburst time of ≈12 days (Szkody & Mattei 1984; Ritter & Kolb 2003). However, the system actually exhibits two kinds of outbursts: narrow outbursts lasting about a week (∼4–7 days) and wide outbursts lasting about two weeks (∼14 days; Sion et al. 1998). Ultraviolet spectroscopic observations obtained during a wide outburst are consistent with a steady-state disk with a mass accretion rate ranging from $\dot{M}=2\times {10}^{-9}\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ (Sion et al. 1997) to $\dot{M}=(7\mbox{--}8)\times {10}^{-9}\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ (Panek & Hom 1984; Froning et al. 2001). During the decline from outburst, the mass accretion rates decreases ( $\dot{M}=6\times {10}^{-10}\,{M}_{\odot }\,{\mathrm{yr}}^{-1}$ ; see Sion et al. 1997), and eventually, the metal-enriched hot WD dominates the spectrum. A second component appears on the WD, possibly an optically thick hot boundary layer or an "accretion belt" (Long et al. 1993), improving the model fit in the shorter wavelengths of Far Ultraviolet Spectroscopic Explorer (FUSE) spectra of U Gem (<960 Å; Froning et al. 2001).

In quiescence, shortly after the outburst, the WD has a temperature still above 40,000 K (Long et al. 2006) and cools to ∼30,000 K within several months (Kiplinger et al. 1991; Long et al. 1994, 1995). The length of the quiescence dictates the post-outburst temperature reached, which has occasionally been observed to be anomalously low (Sion et al. 1998) following a very long quiescence period. The exact WD temperature derived from the data depends also on the WD effective surface gravity log(g) assumed in the modeling.

Nonsolar metal abundances attributed to the WD photosphere have been found in the studies of U Gem (Sion et al. 1998; Long & Gilliland 1999; Froning et al. 2001), revealing evidence of CNO processing, namely, subsolar carbon and silicon abundances and suprasolar nitrogen abundance. Subsolar aluminum has also been recorded (Long & Gilliland 1999). However, phase-dependent absorption has been shown to be responsible for the time-variable absorption seen in the FUV, which complicates the analysis of the WD spectra (Long et al. 2006).

Although U Gem was observed previously with HST in the UV band with the HST/Goddard High-Resolution Spectrograph (GHRS; Sion et al. 1998; Long & Gilliland 1999), we obtained HST/COS FUV observations of U Gem extending down the Lyman limit (915–2148 Å). The only previous observations covering the Lyman region were made with the Hopkins Ultraviolet Telescope (HUT; Long et al. 1993) and FUSE (Froning et al. 2001; Long et al. 2006).

We present here the FUV spectral analysis of four HST/COS spectra obtained following a two-week outburst. The data obtained at the first epoch were purposely collected to span a complete binary orbit (∼4 hr), while the data obtained at the remaining three epochs were each collected for only a quarter of an orbit. In this manner, we were able to analyze the data to detect any variation as a function of the orbital phase as well as a function of the time passed since outburst. In the next section, we describe the observations and data; in Section 3, we present and discuss the results from our spectral analysis; we then conclude in Section 4 with a summary.

2. The HST/COS Spectra

2.1. Observations

U Gem went into outburst on 2013 November 29, for about 15 days, increasing its brightness from ∼14.2 to ∼9.3. The outburst, shown in Figure 1, was fairly normal for a "wide" outburst; however, it followed a quiescence period of only 69 days (the previous outburst was a short seven-day outburst that started on September 14). The HST/COS observations of U Gem were taken on 2012 December 20 and 26, and on 2013 January 7 and 30, following the wide November outburst. The time at which the observations were taken is marked on Figure 1 with vertical arrows pointing down. The observation log is given in Table 2.

Table 2. U Gem Cos G140L (1280 Å) Spectra Observation Log

Epoch	DATAID	Obs. date	Obs. Time (UT)	Exp.time	Days^a
Number		yyyy mm dd	hh:mm:ss	s
1	LC1U01010	2012 Dec 20	11:54:12	5761.6	15
2	LC1U02010	2012 Dec 26	04:59:28	960.0	21
3	LC1U06010	2013 Jan 07	23:26:05	960.1	33
4	LC1U04010	2013 Jan 30	22:03:00	960.1	56

Note.

^aIn the sixth column, we indicate the days counted since the peak of the outburst that started on 2012 November 29 and lasted 15 days.

Download table as: ASCII Typeset image

COS was set up in the FUV configuration with the G140L grating centered at 1280 Å in time tag mode. The data were collected through the primary science aperture of 2.5 arcsec diameter. This configuration produces two spectral segments covering ∼915–1193 Å and ∼1266–1248 Å, with a gap between 1193 and 1266 Å. Although the resulting spectrum covers the Lyman series down to its cutoff, it does not cover the Lα region.

The data were processed with CALCOS version 3.1.8 through the pipeline that produces for each HST orbit four sub-exposures generated by shifting the position of the spectrum on the detector by 20 Å each time. This strategy is to reduce detector effects. It is left to the observer to compare the four sub-exposures and identify all of the detector artifacts that appear on all sub-exposures but with a shift of 20 Å from sub-exposure to sub-exposure. The first epoch data cover three HST orbits and consist of 12 sub-exposures; the remaining data obtained at epochs #2, #3, and #4 all consist of one HST orbit each, and made of four sub-exposures each. The second sub-exposure of epoch #2 was lost due to the COS light path being blocked during that exposure. We therefore collected a total of 23 sub-exposures, each consisting of two spectral segments with a gap in the Lα region. In the first sub-exposure of each exposure (HST orbit), the gap spans the region λ ∼ [1195–1318] Å, in the second sub-exposure λ ∼ [1175–1298] Å, in the third sub-exposure λ ∼ [1155–1278] Å, and in the fourth and last sub-exposure λ ∼ [1135–1258] Å. The extracted spectra have a signal that deteriorates toward the edges of the spectrum (see Figure 2).

**Figure 2.** Sub-exposure (from a single detector position) presented to show the poor signal in the very short and very long wavelengths (sub-exposure #1 of (*HST*) orbit #2 obtained at epoch #1). In this spectrum, the gap occurs between 1195 and 1318 Å, though some detector edge effects affect the spectrum and make the actual gap wider. We discard the region >2000 Å; however, it is likely that the spectrum is affected at even shorter wavelengths (∼1900 Å).
Download figure:
Standard image High-resolution image

For the first orbit of epoch #1 and for the single orbit of epochs #2, #3, and #4, the effective good exposure time of each sub-exposure is a little more than ∼200 s. For the second and third orbits of epoch #1, the exposure time for each sub-exposure is ∼600 s. In Figure 3, we display the 23 spectra extracted from the sub-exposures in the Si iv doublet (∼1400 Å) region to show the quality of the spectra.

**Figure 3.** Si iv doublet (1392.8, 1402.8) shown for the individual sub-exposure spectra. The *HST*/COS data were obtained at four epochs, covering six *HST* orbits, as shown on the right. Each *HST* orbit generated four sub-exposures (one for each of the four COS detector positions), totaling 23 spectra (sub-exposure 2 of the second epoch contains no data). The sub-exposures of the second and third orbits obtained at the first epoch have an exposure time of ∼600 s each; the remaining sub-exposures have an exposure of only ∼200 s each. The difference in the signal-to-noise ratio (S/N) can be seen in the figure.
Download figure:
Standard image High-resolution image

Due to the gap between the two segments of the detector and the 20 Å shift in the detector position for each of the four sub-exposures, and due to detector artifacts (especially near its edges), the spectra are not reliable beyond the C iii (1175) absorption lines nor below ∼1300 Å. In Figure 4, we illustrate how the detector edges (and gain sag) affect some of the spectral regions, and we explain how we identify these artifacts that need to be discarded. For this reason, the spectral line analysis is carried out on the individual sub-exposures, which have a much lower S/N than the combined spectrum. Because of the sub-exposure spectra having a different wavelength coverage, the C iii (1175) and N iii (1183) absorption lines are only present in six sub-exposures.

**Figure 4.** Each COS exposure is made up of four sub-exposures that are generated by shifting the position of the spectrum on the detector by 20 Å. This strategy allows the detector artifacts to be identified and compensated for by inspecting the four sub-exposures simultaneously. We display here the first *HST*/COS observations sub-exposures vertically shifted for clarity. Detector artifacts repeat themselves at 20 Å intervals and are easily found by looking for similar patterns shifted 20 Å from sub-exposure to sub-exposure. In the present case, for illustration purposes, we isolate two such artifacts within the circle and within the ellipse. Unfortunately, the detector artifacts are stronger toward the edges of the detectors, which render the effective detector gap larger than it actually is.
Download figure:
Standard image High-resolution image

2.2. The FUV Absorption Lines

In this subsection, we present the absorption lines detected in the COS spectrum of U Gem. Although many lines can be easily identified, some other lines were identified during our spectral analysis, and the procedure we follow for line identifications is described in the analysis section. We did not detect any emission features in the COS spectrum of U Gem.

During quiescence, we expect the WD to dominate the FUV and, therefore, we expect the COS spectrum to exhibit absorption lines from a moderately hot (∼30,0000–40,000 K) WD, namely, the hydrogen Lyman lines, helium Balmer lines, low-ionization species lines, as well as possibly some higher-ionization species lines. The individual sub-exposure spectra are very noisy, especially below 1000 Å, and we consequently first proceed by combining together the sub-exposures of the first epoch to clearly display the absorption lines and identify them below 1080 Å. For clarity, the resulting epoch #1 spectrum is presented in two figures. The actual spectral analysis is carried out on the sub-exposures rather than on the combined spectrum.

In Figure 5, we display the short-wavelength segment, which has a similar spectral range to FUSE spectra. The hydrogen Lyβ absorption line is the main broad feature dominating the spectrum in the middle panel. The hydrogen Lyγ and Lyδ absorption features are also clearly apparent, but the higher orders of the series are not. The spectrum is marked by broad absorption features usually detected in the FUSE spectra of CV WDs (Godon et al. 2012) from carbon (C iii 977, 1175), nitrogen (N iii 990, 1085, 1184), silicon (Si iii 1110, 1144; Si iv 1167, 1122, 1128), and sulfur (S iv 1063, 1073). In addition, some higher-ionization species absorption lines appearing only at higher temperature (T ∼ 80,000 K), N iv (923) and the O vi doublet (1032, 1038), are also seen, an indication that a hot absorbing medium is in front of the WD. The S vi (933, 944) and P v (1118, 1128) absorption lines, previously seen in the FUSE spectra of U Gem in quiescence (Long et al. 2006), are not detected here in the spectrum. The line identification process is explained in the analysis section, where each spectrum (and its sub-exposures) is modeled with realistic photospheric WD models. We list the absorption lines in Table 3, where only the rest (laboratory) wavelength is listed for each line, since the exact position of the lines changes for each sub-exposure.

Table 3. Absorption Lines in the COS Spectra of U Gem in Quiescence

Ion	λ(Å)
N iv	921.46–924.91; 923.2
C iii	977.0
N iii	989.8, 991.56
He ii	992.4
O vi	1031.9
O vi	1037.6
S iv	1062.7
Si iv	1066.6
S iv	1073.0
He ii	1084.9
N iii	1085.5
Si iii	1108.4, 1109.9, 1113.2
Si iv	1122.5, 1128.3
Si iii	1140.6–1145.7; 1144.3
C iii	1174.6–1176.8; 1175.3
N iii*	1183.0
N iii*	1184.5–1185.6
Si iii	1294.4, 1296.7, 1298.9, 1301, 1303.2
C ii	1323.9–1324.0
C ii*	1334.5, 1335.7
Si iii	1341.5, 1342.4, 1343.4
Si iv	1392.8, 1402.8
Si iii	1417.2
Si iii	1500.2, 1501.2, 1501.9
Si ii	1526.7, 1533.4
C iv	1548.2, 1550.8
He ii	1640.3, 1640.5
Ar iii	1669.3, 1669.7
Al iii	1854.7, 1862.8
Fe	∼1450–1600

Note. All of the lines listed here have been marked in Figures 5 and 6. As described in the analysis section, many lines were identified by comparing them to low-metallicity synthetic stellar spectra while increasing the abundance of one species (metal) at a time. In this manner, we identified the sawtooth pattern between ∼1450 and ∼1600 Å as a multitude of small iron absorption lines (Fe ii). Lines annotated with an asterisk (*) are contaminated with detector artifacts.

Download table as: ASCII Typeset image

Not unsurprisingly, the COS spectrum of U Gem does not show any interstellar absorption lines as it is a nearby system.

The long-wavelength segment of the epoch #1 spectrum is displayed in Figure 6, where, for clarity, the vertical scale has been stretched. The first striking feature of the long-wavelength segment spectrum is the numerous absorption features seen at wavelengths below 1400 Å. An inspection of the sub-exposures reveals that most of the "lines" are due to detector artifacts (gain sag in regions illuminated by airglow), as they are seen in each sub-exposure shifted by 20 Å. Below 1400 Å, we detect with confidence Si iii (∼1300) and C ii (1324) lines. The C ii (1335) line is contaminated by a detector artifact. All of the other features between 1265 and 1380 Å are detector artifacts. Fortunately, the longer wavelength region is not affected, and we identify silicon lines (Si iv doublet ∼1400; Si iii ∼1342, 1417, 1500; Si ii 1527, 1533), the C iv (1550) line, the He ii (1640) line, as well as aluminum (Al iii 1855, 1863) and possibly argon (Ar iii 1670) lines. In the line analysis section, we will show that the sawtooth pattern observed from ∼1420 Å to ∼1630 Å is likely due to iron (Fe ii).

3. Spectral Analysis Results and Discussion

We use Ivan Hubeny's FORTRAN suite of codes TLUSTY and SYNSPEC (Hubeny 1988; Hubeny & Lanz 1995) to generate synthetic spectra for high-gravity stellar atmosphere models. We first generate a one-dimensional vertical stellar atmosphere structure with TLUSTY for a given surface gravity log(g), effective surface temperature T_eff, and surface composition of the star. We treat hydrogen and helium explicitly, and treat nitrogen, carbon, and oxygen implicitly (Hubeny & Lanz 1995). We then run SYNSPEC using the output stellar atmosphere model from TLUSTY as an input, and generate a synthetic stellar spectrum over a given (input) wavelength range, from 900 to 3200 Å here. The code SYNSPEC derives the detailed radiation and flux distribution of the continuum and lines, and generates the output spectrum. For temperatures above 35,000 K, the approximate NLTE treatment of the lines is turned on in SYNSPEC (Hubeny & Lanz 1995). Rotational and instrumental broadening as well as limb darkening are reproduced using the routine ROTIN. For full details of the codes, see Hubeny & Lanz (2017a, 2017b, 2017c).

Following this procedure, we generated photospheric models for U Gem in quiescence with an effective WD temperature ranging from 25,000 to 50,000 K in increments of 1000 K, 500 K, and ∼250 K, as needed for the convergence of the model fit. The effective surface gravity was set to log(g) = 8.8 to match the large mass of the WD ∼ 1.1–1.2 M_⊙. We varied the projected stellar rotational velocity ${V}_{\mathrm{rot}}\sin (i)$ from 100 to 250 km s⁻¹ to match the known WD velocity of about 100–150 km s⁻¹, and in order to fit the absorption features of the spectrum, we varied the abundances of the elements, one at a time, from 0.01× solar to 20× solar.

The spectral analysis to derive the effective WD temperature and the abundance of the elements was carried out iteratively. Namely, we first assessed the WD temperature using solar abundances, then for that temperature we varied the abundances one element at a time; we then fine-tuned the WD temperature using the best-fit abundances.

3.1. Abundances

Except for hydrogen and helium, which were set to solar abundances, we started by setting all of the metals (i.e., Z > 2) to a very low abundance such that no absorption lines formed in the theoretical spectra, except for the lines of the hydrogen Lyman and helium Balmer series. We then increased the abundance of one metal at a time, starting at 0.01× solar, up to 20× solar, taking all of the discrete values: 0.01, 0.02, 0.03, 0.04, 0.05, 0.08, 0.10, 0.12, 0.15, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9. 1, 2, 3, 4, 5, 7, 10, 15, 20. In this manner, we generated theoretical spectra with varied abundances for C, N, O, P, S, Si, Al, Ar, Fe, Mg, Na, and Ne one at a time. In all of the models, we choose log(g) = 8.8, and for the first epoch spectra, the WD temperature was set (see the next subsection) to 41,500 K; for the second epoch, T = 39,850 K; for the third epoch, T = 37,650 K; and for the fourth and last epoch, T = 36,350 K. We then fitted each line in each single sub-exposure with the varied abundances WD spectra until a match was found.

The fitting did not reveal abundances for O, P, Mg, Na, or Ne (there were no matching lines).

In Table 4, we list the abundances we found for C, N, S, Si, and Al, and in Figure 7, we illustrate the abundance fits for these ions. Some spectral regions (e.g., N iii ∼990; Figure 7(a)) are noisier than others (e.g., Si iv 1400, Figure 7(l)), and some line complexes (e.g., Si iii 1110, Figure 7(h)) exhibit different line shifts within the complex. If one considers only the vicinity of a given line (or complex), then in some cases the continuum is difficult to identify (e.g., Figures 7(a) and (d)). For that reason, the abundances were checked visually for each line in each sub-exposure. For the N iii (∼990) line (Figure 7(a)), in spite of the fact that the region is very noisy, the data do agree with a very large nitrogen abundance [N] = 20× solar or larger. For the N iii (∼1085) absorption feature (Figure 7, panel b), the depth of the line barely changes from [N] = 0.01 (solid red line) to [N] = 20.0 (solid green line), and consequently, one cannot rely on this feature to assess the nitrogen abundance. In addition, the exact wavelength of the line is also difficult to assess: an [N] = 1 model redshifted by more than 1 Å is also shown for comparison (solid blue line), and it fits only the right wing of the line, while the two other models fit only the left wing of the line. As for the the N iii (1184) absorption feature (panel c), it seems to be more reliable; however, the fitting of the adjacent C iii (1175) line complex brings up the question of how to assess the continuum flux level. For the nitrogen abundance derived from the N iii (1184) absorption feature, as well as for most of the other abundances derived here, we used the continuum flux level obtained from fitting the entire theoretical spectrum to the observed spectrum. For the C iii (1175) absorption feature (panel d), we found that the left wing (including the two additional smaller carbon absorption lines to its left) and the right wing cannot be fitted simultaneously. The fitting of the continuum on the left side of the feature (as shown in panel d, giving [C] ≈ 0.4) yields a higher continuum flux level than that derived from the global fit. The fitting of the continuum on the right side of the feature, while giving the same continuum flux level as derived from the global fit, does not agree at all with the left side of the feature and produces a line that is far too narrow ([C] = 0.01). This could be due to detector effects near the edge of the spectral segment, where the flux seems to drop. This wavelength region is only covered for one of the four detector positions used with COS, i.e., only one in four sub-exposures; we therefore decided not to rely on this carbon absorption feature nor on the N iii 1184 feature. For the carbon abundance, we are left with the C iv doublet (1400; shown in panel e), which is covered in all of the sub-exposures. Here, the abundance is assessed mainly by fitting the wings of the absorption feature. The aluminum abundance is derived by fitting the Al iii doublet near 1860 Å, as shown in panel f. For the sub-exposure displayed, the aluminum abundance is in excess of 20× solar. The sulfur abundance was derived by fitting the S iv (1063 & 1073) lines (shown in panel g). This region of the spectral segment is very noisy, and we found that the fitting of the 1063 line, which is narrower than the 1073 line, gave a lower abundance than the 1073 line, possibly due to the noise. Nevertheless, the sulfur abundance listed in Table 4 is an average taken from fitting the two lines. Silicon abundances were derived from fitting five absorption line complexes. Si iii (∼1112), displayed in panel h, shows another example of a complex with lines that are shifted relative to one another. Si iv (∼1126), displayed in panel i, shows how the right side of the absorption feature (at λ > 1132 Å) has a reduced flux in the vicinity of the edge of the spectral segment (followed by a gradual drop to zero at 1140 Å). The Si iii (1143) absorption feature (panel j) is not very pronounced or deep, but still provides a relatively good assessment of the silicon abundance, contrary to the Si iii (1160) absorption feature (panel h), which is unreliable as it does not match the expected lines. The Si iv (1400) doublet (panel l) provides a more reliable assessement of the silicon abundance, in particular with the additional small line seen around 1419 Å. Accordingly, in Table 4, we marked all of the absorption features and abundances that are unreliable with an asterisk.

Table 4. Abundance of Elements in U Gem COS Spectra in Quiescence

Ion	C iii	C iv	N iii	N iii	N iii	S iv	Si iv	Si iii	Si iv	Si iii	Si iii	Si iv	Al iii	Orb.Phase
λ(Å)	1175*	1550	990	1085*	1184*	1063–73	1067	∼1112	∼1126	∼1143	∼1160*	∼1400	1856–64	ϕ
Exp.

1-1-1	0.4	0.75	20	0.01	0.4	0.5	0.1	1.0	0.2	0.3	0.7	0.8	1.0	0.51
2	...	2.0	20	5	...	0.2	0.5	2.0	0.2	0.3	0.7	1.5	1.0	0.54
3	...	0.75	20	10	...	0.5	0.4	1.0	0.1	0.3	...	1.0	...	0.57
4	...	0.70	20	20	...	3.0	2.0	3.0	1.0	...	...	1.0	20	0.78
2-1	0.4	1.0	20	20	0.1	1.0	0.1	5.0	0.6	0.4	0.5	1.0	20	0.81
2	...	0.9	20	1	...	0.8	0.1	3.0	0.3	0.3	0.5	0.5	0.8	0.86
3	...	0.5	20	20	...	0.6	0.05	2.0	0.1	0.3	...	0.9	0.5	0.91
4	...	1.0	20	10	...	0.4	0.5	2.0	0.4	...	...	1.0	10	0.16
3-1	0.6	1.0	20	20	0.05	1.0	0.1	...	1.0	0.3	0.5	3.0	10	0.21
2	...	1.0	20	20	...	1.5	0.05	...	1.0	0.3	0.5	1.0	20	0.25
3	...	1.0	20	20	...	0.8	0.05				...	2.0	20	0.30
4	...	2.0	20	10	...	0.5	0.05	1.0	0.2	...	...	2.0	1.0	0.53
2-1-1	0.2	1.5	≫1	5	0.7	1	0.4	3.0*	0.4*	0.3*	0.5*	1.0	7	0.80
2	...	...	...	...	...	...	...	...	...	...	...	...	...	0.83
3	...	0.05	10	0.5	...	...	...	0.6*	0.08*	0.2*	...	0.7	1	0.85
4	...	0.3	1.0	0.5	...	1	0.2	0.6	0.15	...	...	0.5	...	0.88
3-1-1	0.1	1.0	1.0	10	1.0	1	0.3	0.9	0.40	0.2	...	1.0	3	0.98
2	...	1.0	...	3	...	2	0.2	1.0	0.2	0.2	1.0	0.6	2	0.00
3	...	0.7	...	1.0	...	0.4	...	0.7*	0.1	0.2	...	0.5	0.2	0.03
4	...	0.1	...	0.1	...	0.4	0.5	0.4*	0.15*	...	...	0.6	1	0.24
4-1-1	0.5	2.5	20	20	0.5	1	0.2	1*	0.2	0.2	...	2.0	3	0.66
2	...	2.5	...	10	...	1	0.2	1*	0.15	0.15	0.4	1.0	1	0.69
3	...	1.5	10	10	...	0	...	1	0.08	0.3	...	0.7	1	0.96
4	...	3.0	10	10	...	...	...	1	0.2	...	...	1.0	3	0.99

Note. The lines were fitted with a rotational broadening velocity of 150 km s⁻¹. The values listed are accurate to the next adjacent discrete values (see text). The abundances derived from the C iii 1175, N iii 1185, N iii 1184, and Si iii 1160 line complexes were deemed unreliable and are marked with an asterisk "*" (top of the table). Unreliable abundances within the table are also marked with an asterisk "*". No value is given for lines in the detector gap or in very noisy regions of the sub-exposure. Sub-exposure 2 of epoch #2 has no signal at all.

Download table as: ASCII Typeset image

For argon (Ar iii ∼ 1669.5; see Figure 9), we found that the abundance would have to be 100× solar or larger. It is possible that the ∼1669 absorption feature belongs to a different element, which we could not identify.

The iron abundance was derived by fitting the sawtooth pattern in the wavelength range ∼1500–1600 Å, as shown in Figure 8.

Overall, we found that carbon and sulfur are both consistent with subsolar to solar abundances; nitrogen, aluminum, and argon have all suprasolar abundances. Silicon is subsolar except for the Si iii (∼1112 Å) and Si iv (1400 Å) lines, which match about solar abundances, confirming the claim of Long et al. (2006) that the Si iii lines at ∼1110 Å require abundances that are much larger than required at 1140 and 1155 Å. We note, however, that the silicon lines at ∼1112 Å can be fit with a lower temperature, namely, T_eff = 25,000–35,000 K. We will come back to this in Section 3.4. The suprasolar nitrogen abundance with the subsolar carbon abundance further confirms the anomalously high N/C ratio observed in previous spectra of U Gem in quiescence, which has been attributed to CNO processing (Sion et al. 1998; Long et al. 2006).

At solar abundance, the He ii 1640 absorption feature needs a higher rotational velocity of 200–250 km s⁻¹ in some of the sub-exposure spectra. We also found that the Si iv doublet (∼1400 Å) appears to be off (redshifted) by about one angstrom compared to other lines, as noted by Long et al. (2006), who suggested that they do not always form in the stellar photosphere. Some of the absorption lines appear to be transient (see the next subsection). We also found that the width, shape, shift, and depth of many lines are not uniform and vary even in a single sub-exposure, indicating that the lines do not form in the same medium. At first glance (e.g., Table 4), the abundances do not seem to change as a function of time since outburst. We, therefore, decided to analyze the absorption lines as a function of the orbital phase.

3.2. Orbital Phase Variability of the Absorption Lines

We use the ephemeris of U Gem as given by Ecchevarría et al. (2007) to find the phase at which each sub-exposure was obtained, where we take the phase ϕ = 0 as the secondary conjunction. The phase of the start of each exposure is indicated in Figure 3 as well as in the last column in Table 4 to within an accuracy of 0.01 (∼150 s).

By visual inspection of the sub-exposure spectra, we found that many absorption lines appear stronger/deeper at some orbital phases, regardless of the epoch at which the spectra were taken. We, therefore, decided to examine the sub-exposure spectra in order of increasing orbital phase from 0.0 to 1.0 and to combine them near the same orbital phases when possible. Following this procedure, we generated nine spectra at orbital phases 0.00 (made up of four sub-exposures), 0.16 (one sub-exposure), 0.25 (3), 0.54 (3), 0.67 (2), 0.78 (1), 0.81 (2), 0.88 (4) 0.98 (1), and 1.00 (=0.00). We found that the Al iii (1854.7, 1862.8), Ar iii (∼1669.5), and Si ii (1526.7, 1533.4) absorption lines are particularly sharp and deep at orbital phases ϕ = 0.25 and 0.78, and are much weaker (almost absent) around orbital phases ϕ ∼ 0.54 and 0.00 (1.00; see Figure 9). Al iii appears to be stronger at phases 0.78, 0.81, 0.16, and 0.25, namely, over a wider range of phases.

**Figure 9.** Variation of the strength of the absorption lines as a function of the orbital phase. The spectra were generated by co-adding a number of sub-exposures in the immediate vicinity of the orbital phase as shown on the right. The aluminum (Al iii 1854.7 and 1862.8) and argon (Ar iii ∼1669.5) absorption lines (upper panels) are stronger around phases ϕ ∼ 0.25 and 0.75, and are much weaker (almost unnoticeable) around phases ϕ ∼ 0.0 and 0.5. The carbon (C iv ∼ 1550) and helium (He ii 1640) lines (lower panels) do now show such a strong phase variability. Note the strengthening of the Si ii (1526.7 and 1533.4) lines around phase ϕ ∼ 0.25 and especially at phase ϕ = 0.78 (to the left of the C iv lines in the lower-left panel). The spectrum at phase 1.00 (0.00) was obtained by combining the sub-exposure spectra from phases 0.96, 0.99, 0.00, and 0.03; the spectrum at phase 0.88 from phases 0.85, 0.86, 0.88, and 0.91; the spectrum at phase 0.81 from phases 0.80 and 0.81; the spectrum at phase 0.67 from phases 0.66 and 0.69; the spectrum at phase 0.54 from phases 0.51, 0.54, and 0.57; and the spectrum at phase 0.25 from phases 0.21, 0.25, and 0.30.
Download figure:
Standard image High-resolution image

Most of the other metal absorption lines (especially the silicon lines in the range ∼1110 to ∼1130 Å) seem to exhibit some change with the orbital phase, where the depth of the absorption lines decreases from phase ∼0.80 (deepest) to phase ∼0.25 (intermediate), and phases ∼0.0 and ∼0.54 (shallowest; see Figure 10). The Si iv (1400) and C iv (1550) absorption lines are only slightly deeper around phases 0.25 and 0.80, while the He ii (1640) absorption line does not exhibit such a variation. Due to the wavelength coverage of the sub-exposure spectra, we could not carry out the same analysis for the C iii (1175) and N iii (1183) absorption lines. In addition, the low S/N below 1100 Å prevented us from obtaining robust results for absorption lines in that wavelength region.

**Figure 10.** Silicon lines in the wavelength range 1100–1130 Å are deeper around phase 0.80, and shallower around phase 0.54 (and 1.00—not shown for clarity). The increase in the depth of the absorption lines indicates the presence of absorbing material in front of the WD at phases ∼0.25 and ∼0.80. The spectrum around phase 0.54 was created by combining together the sub-exposure spectra from phases 0.51, 0.53, 0.54, and 0.57; the phase 0.25 spectrum was obtained from combining phases 0.21, 0.24, 0.25, and 0.30; and that of phase 0.80 was from combining phases 0.78, 0.80, and 0.81.
Download figure:
Standard image High-resolution image

In spite of the fact that the sub-exposure spectra were obtained at four different epochs, the results show a consistent increase in the depth of the absorption lines at orbital phases 0.16–0.30 and 0.67–0.81, indicating the presence of an absorbing material with high metal content (i.e., Si, Al, Ar) in front of the WD at those phases. This is in complete agreement with the results of Froning et al. (2001) and Long et al. (2006), who found an increase in depths of the lines and the sudden appearance of low-ionization lines at orbital phases between 0.53 and 0.85. These are the same orbital phases where the X-ray and EUV light curves dip, as seen in U Gem in both outburst and quiescence (Mason et al. 1988; Long et al. 1996; Szkody et al. 1996). The increase in the depth of the absorption lines and the EUV/X-ray light-curve dips possibly have the same origin, i.e., a (rim) disk bulge with mass stream overflow material located at large disk radii, to agree with the observed low radial velocity (Froning et al. 2001; Long et al. 2006). The increase in the depth of the lines around phases 0.16–0.30 can be attributed to the stream material overflowing the disk rim and bouncing off the disk surface near phase 0.5 where it is then redirected toward phases ∼0.10–0.30 (Kunze et al. 2001).

In order to try and further identify the location of the formation of the absorption lines in the COS spectrum of U Gem, we measured the velocity shift of the lines in each sub-exposure spectra. We match each line individually with the corresponding line in a synthetic stellar atmosphere spectrum generated with TLUSTY/SYNSPEC, and we (red- or blue-) shifted the spectrum until the observed line became superposed on the theoretical line. In this manner, we obtained a spectral shift accurate within ∼0.1 Å, which we then translated into a radial velocity shift. The lines for which the velocity was measured were C iv (1550), Si iv (1400, 1122.5, 1128.3), Si iii (1113, 1144, 1500), and Si ii (1530). For the C iii (1175), N iii (1183), and Al iii (∼1860) lines, only a limited number of data sets were available. The velocity shift of the absorption lines is displayed as a function of the orbital phase in Figure 11, together with the expected velocity shift of the WD, taking into account a 122 km s⁻¹ orbital velocity (Long et al. 2006), a 80.4 km s⁻¹ gravitational redshift (Sion et al. 1998), and a recessional velocity of 84 km s⁻¹ (Wade 1981). The first thing that is striking in Figure 11 is the large scatter observed at all phases, but this scatter has been observed before (Froning et al. 2001). The carbon lines (in black) appear to follow the WD velocity more closely than the other lines, and the Si iv lines seem to have the largest velocity offset, followed closely by the Si iii lines. Long et al. (2006) also found the Si iv doublet (∼1400 Å) to be redshifted by about one angstrom compared to the other lines, suggesting that they do not always form in the stellar photosphere. For the most part, the velocity offset of the silicon is positive, indicative of material falling toward the WD. The average velocity offset is the largest (positive) at phases 0.78–0.91, which is consistent with the location of the L1-stream material overflowing the disk's edge and falling toward smaller radii.

**Figure 11.** Absorption line velocity shift (in km s⁻¹) in the COS spectrum of U Gem as a function of the orbital phase ϕ. The velocity offsets were measured from the rest wavelengths for the 23 sub-exposures using a synthetic WD photosphere spectrum. The solid black line is the orbital motion of the WD assuming an orbital velocity amplitude of ∼122 km s⁻¹ (Long et al. 2006), a WD gravitational redshift of ∼80.4 km s⁻¹ (Sion et al. 1998), and a proper motion radial velocity of ∼84 km s⁻¹ for U Gem (Wade 1981). The carbon lines (black circles and dots) follow more closely the WD orbital velocity. The Si iv (λλ1122.5, 1128.3, ∼1400) lines (red stars) exhibit the largest offset from the WD orbital velocity, followed by the Si iii (λλ ∼1113, ∼1144, 1500) lines (red circles) and Si ii (λλ1530) lines (red triangles). Only a small number of data points could be obtained for nitrogen (green diamonds) and aluminum (blue squares).
Download figure:
Standard image High-resolution image

3.3. Orbital Phase Variability of the Continuum

The 12 sub-exposures of the first epoch are covering a complete binary orbit and were all obtained within that time frame. The sub-exposures obtained at epochs #2, #3, and #4 have a lower flux since they were obtained as the system was declining. Consequently, in order to check the variability of the continuum in a self-consistent manner, we used only the 12 sub-exposures from the first epoch. For each of the 12 sub-exposures, we integrated the flux from 915 to 2000 Å, and normalized it. The resulting COS FUV light curve is presented in Figure 12 and exhibits an orbital phase variability with a maximum amplitude of only ±5%, which is relatively small compared to +20% and −30% observed in FUSE observations obtained in deep quiescence (Long et al. 2006). Even though the light curve is made of only 12 data points covering the entire binary orbit, we notice that the two lowest points are at ϕ = 0.25 (local minimum) and ϕ = 0.78 (global minimum), with maximum at phases ϕ = 0.81–0.91.

**Figure 12.** UV light curve as a function of the binary phase generated by integrating the 12 sub-exposures from epoch #1 over the spectral region 915–2000 Å, from the *HST*/COS observation of U Gem, obtained 15 days after the peak of a "wide" outburst. There is a dip at ϕ = 0.25 and a second deeper dip at ϕ = 0.78, and a maximum toward ϕ ∼0.8–0.9. The total flux (915–2000 Å) varies by only ±5%.
Download figure:
Standard image High-resolution image

In order to make a better comparison with the FUSE and optical light curves, we generated two light curves, one covering the shorter wavelengths 915–1100 Å, and the other one covering the 1380–2000 Å region (Figure 13). The two dips observed in Figure 12 appear mainly as features in the short-wavelength light curve only (in blue in Figure 13). The two dips correspond to the veiling of the WD by the same L1-stream material that has overflowed the disk rim (near ϕ ∼ 0.9): first, as it overflows higher above the disk toward smaller radii near ϕ = 0.78, then as it is redirected toward phase ϕ ∼ 0.25 after it has bounced off the disk surface near phase 0.5 (Kunze et al. 2001). This is completely consistent with the FUSE light curve revealing dips at orbital phases 0.6–0.85 as well as at orbital phases 0.2–0.35. Note that in the FUSE light curve the flux below 970 Å decreases substantially during the ∼0.7 phase dip (Long et al. 2006).

The longer-wavelength light curve (Figure 13) exhibits an increase of flux near phases 0.8–0.9. This is similar to the optical (blue and red) light curves revealing a strong hump at phases 0.8–0.9 due to the contribution of the hot spot facing the observer and before it is occulted by the secondary near phase 0.0 (Unda-Sanzana et al. 2006; Ecchevarría et al. 2007). A secondary broad hump around phase 0.25 is also observed in the Hα and Hβ light curves (Marsh et al. 1990; Unda-Sanzana et al. 2006; Ecchevarría et al. 2007).

3.4. The White Dwarf Temperature

Since many absorption lines in the spectrum of U Gem are either forming in an absorbing medium in front of the WD or are being affected by that absorbing medium, we decided to model the WD spectrum using the continuum and not concentrate on the metal absorption lines in the model fit.

The continuum is essentially characterized by its slope and by the broad absorption features of the hydrogen Lyman and helium Balmer series. The advantage of the COS spectrum is that it covers the spectral range down to the Lyman limit (though the Lyα feature falls into the detector gap) and consequently allows us to derive the WD temperature by fitting the Lyman series. Although the continuum slope can also be used to derive the WD temperature, it is not as sensitive to the temperature as the Lyman series (for the WD temperatures considered here).

For our WD temperature analysis, we combined the sub-exposure spectra by epoch to generate four spectra, one for each epoch. Since the first epoch has 12 sub-exposures, its spectrum has a much higher S/N than the second, third, and fourth exposure spectra. Also, since the lines have a "scattered" velocity broadening of about 300 km s⁻¹ (see Figure 11; reaching almost 400 km s⁻¹ at ϕ = 0.25), we shifted only the first epoch sub-exposure spectra by the WD orbital velocity before combining them. As a result, the absorption lines have a velocity broadening larger than the known WD projected rotational velocity of 100–150 km s⁻¹. In order to approximately match the absorption lines, we set up solar abundances for all elements, except for nitrogen and aluminum, which were both set to 20× solar. We then modeled the four epoch spectra with synthetic WD (with log(g) = 8.8) model spectra generated in the manner described at the very beginning of this section.

3.4.1. The First Epoch

As we fitted synthetic WD model spectra to the first epoch spectrum, we found that the slope of the COS continuum in the longer wavelengths down to 2000 Å agreed with a temperature as low as 25,000 K, which gives an extremely short distance (43 pc) but did not provide enough flux below 1200 Å and basically no flux below 1000 Å. As we increased the temperature to 35,000 K, the slope of the theoretical continuum decreased very slowly, while the short-wavelength region was still deficient in flux, and the profile of the Lyman series was not fitted properly. In longer wavelengths, the slope of the continuum agreed down to about λ ∼ 1800 Å, but the model was very slightly deficient in flux for λ > 1800 Å.

In order to match the profile of the Lyman series (down to ∼940 Å), we had to increase the temperature to 41,500 K, while the continuum of the model in the longer wavelengths slightly decreased. The Lyman series is much more sensitive to the temperature than the continuum slope in the longer wavelengths, and therefore provides a much more reliable way to determine the temperature of the WD. This model presents a satisfactory fit to the overall spectrum and is presented in Figure 14.

**Figure 14.** COS/G140L (center wavelength 1280 Å) spectrum of U Gem (in red) obtained at epoch #1, 15 days after the peak of the outburst, modeled with a WD photosphere spectrum (in black). The observed spectrum extends from ∼915 Å (the hydrogen Lyman limit) to ∼2150 Å, while the theoretical spectrum extends from 900 Å to 3200 Å. The spectral regions of the observed spectrum that are matched to the theoretical spectrum are in red, the excluded regions are in blue, namely, λ < 915 Å, λ > 2000 Å, and the gap between the two spectral segments overlapping the hydrogen Lyα region ∼1190–1268 Å. The model has log(g) = 8.8 and is scaled to the distance of ∼100 pc, giving a temperature of 41,500 K, with a projected rotational velocity of 250 km s⁻¹. The abundances are all solar, except for nitrogen and aluminum, which had to be set to 20× solar to fit the absorption features. Note in the very long wavelengths, the model is slightly deficient in flux, which could be due to the small contribution from a very weak component, but could also be due to some detector effect, as shown in the lower panel of Figure 2. This spectrum is a combination of 12 sub-exposures and has the best S/N, but consequently it also suffers from line broadening.
Download figure:
Standard image High-resolution image

Our grid of models for epoch 1 is spaced every 250 K, and we find that the best fit spans three models, from 40,250 K to 40,750 K, giving limiting values of 40,000 and 41,000 K. We therefore take a 500 K error bar for the WD temperature, namely, for the first epoch spectrum we find a WD temperature of 41,500 K ± 500 K, 15 days after the peak of the outburst. For comparison, a 40,500 K model has a flux that is 3% too low at 1000 Å when scaled from the flux at 1400 Å, but the calibration errors of COS are ∼1.5% at 1400 Å.

With a distance of 100.4 pc, the scaled WD radius is R_wd =5,029 km, corresponding to a WD mass of M_wd = 1.111 M_⊙ at a temperature of 41,500 K (Wood 1995; Panei et al. 2000). The WD mass and radius give an effective surface gravity of log(g) = 8.765, which is consistent with our assumption of log(g) = 8.8. This model is a little deficient in flux at very short wavelengths (<950 Å) and long wavelengths (>1725 Å; see Figure 14). A possible explanation for the discrepancy in the very long-wavelength region could be detector edges effects as apparent from Figure 2.

There are other possible emitting components in the system, such as the hot spot (L1-stream impacting the rim of the disk), the boundary layer, or the very low mass accretion rate disk, which could contribute additional flux. It has also been suggested that the WD itself could have a hot accretion belt remaining in its equatorial region after outburst, or that the WD could have a non-uniform temperature (Long et al. 1993; Sion et al. 1998; Long & Gilliland 1999; Long et al. 2006). As mentioned previously, not all of the silicon lines can be fitted at the same time with the same abundance; however, this could also be due to two emitting components with different temperatures. For example, the Si ii (1530) and Si iii (∼1112) lines are best fitted at ∼30,000 K (assuming log(g) = 8.8), while the other silicon lines give a better fit at 35,000–45,000 K. We therefore tried two-temperature WD models, with a hot equatorial region on the surface of a cooler WD.

In order to fit the longer-wavelength region, we started by taking a low WD temperature (∼25,000 K) model (with a corresponding larger radius) that did not provide any flux in the short-wavelength region, and added to it the contribution from a hotter region (T ∼ 45,000 K and higher) to provide the flux in the shorter-wavelength region. Such models did not fit the COS spectrum better than the single-WD models. We varied the size and temperature of the hot equatorial region and the temperature of the WD, but this combined model did not provide a qualitatively better fit to the single-WD model and was not required by the data. The number of two-temperature models that could fit the data is very large (divergence of the solution). Therefore, in our analysis below, we only considered single-WD models.

The WD temperature depends on the assumed effective surface gravity (log(g)) of the model as well as on the addition of a hot second component (Long et al. 1995, 2006). The second component is not always detected (Sion et al. 1998), either because of the wavelength coverage (i.e., λ > 1000 Å) or maybe simply because it is not always there. The nature of the second component is still a matter of debate, as different physical models for the second component fit the data just as well (Long et al. 2006). Nevertheless, the WD temperature we obtained two weeks after the peak of the outburst is in good agreement with previous spectral analyses of the system in quiescence and decline (Sion et al. 1998; Long & Gilliland 1999; Froning et al. 2001; Long et al. 2006).

3.4.2. The Second Epoch

With only three sub-exposures, the second epoch spectrum is noisy, especially in the short-wavelength region. Our single-WD model fit gave a temperature of 39,850 ± 500 K and a scaled WD radius of 5,022 km, nearly equal to that obtained for epoch #1. However, due to the slightly lower temperature, this radius corresponds to a 1.109 M_⊙ WD mass or a log(g) of 8.766. This model fit is presented in Figure 15. This spectrum was obtained 21 days after the peak of the outburst, exhibiting a cooling of 1,650 K in six days. The discrepancy in the long-wavelength region is barely noticeable, but it is still noticeable in the very short-wavelength region. The broadening of the lines (200 km s⁻¹) includes some orbital velocity broadening as well as lines forming in different regions, as the projected WD rotational velocity is known to be about 100–150 km s⁻¹.

**Figure 15.** COS spectrum of U Gem obtained at epoch #2, 21 days since the peak of the outburst, modeled with a WD photosphere spectrum, as in the previous figure. The temperature obtained from the fitting is 39,850 K. The abundances are all solar, except for nitrogen and aluminum, which were set to 20× solar. This spectrum is pretty noisy, especially in the very short-wavelength region, as it is a combination of only three sub-exposures (sub-exposure #2 was lost due to the COS light path being blocked).
Download figure:
Standard image High-resolution image

3.4.3. The Third Epoch

With four sub-exposure spectra obtained 33 days after the outburst peak, the third epoch spectrum yielded a WD temperature of 37,650 ± 500 K and a scaled radius of 4,999 km. This corresponds to a WD mass of 1.109 M_⊙ and an effective surface gravity of log(g) = 8.77. In 17 days (since epoch #2), the WD temperature dropped by an additional 2200 K. Compared to the two previous epoch spectra, the model fit (Figure 16) has improved; there is no longer discrepancy in the long-wavelength region, and it is barely noticeable in the short-wavelength region.

**Figure 16.** COS spectrum of U Gem obtained at epoch #3, 33 days after the peak of the outburst, modeled with a 37,650 K WD model. Except for the WD temperature, all of the other parameters are the same as in the modeling carried for epoch #2. The spectrum was generated by co-adding the four sub-exposures obtained in one (*HST*) orbit.
Download figure:
Standard image High-resolution image

3.4.4. The Fourth Epoch

Also with four sub-exposure spectra, the fourth and last epoch spectrum was fitted with a single-WD model with a temperature of 36,250 ± 500 K (Figure 17), revealing a further cooling of 1400 K in 23 days (i.e., 56 days after the outburst peak). The scaled radius obtained was 5005 km, corresponding to a WD mass of 1.107 M_⊙ and an effective surface gravity of log(g) = 8.768. Here, too, there is no longer discrepancy in the long-wavelength region, and very little discrepancy in the short-wavelength region.

**Figure 17.** COS spectrum of U Gem obtained at epoch #4, 56 days after the peak of the outburst, modeled with a 36,250 K WD model. All of the other parameters of the model are as specified for the modeling of epochs #2 and #3. The spectrum was generated by co-adding the four sub-exposures obtained in one (*HST*) orbit.
Download figure:
Standard image High-resolution image

For the third and fourth epoch spectra, the fit to the continuum is relatively good, and there is little indication of a possible second emitting component. We find that our results are self-consistent with a massive WD with an effective surface gravity of log(g) ≈ 8.77 (as there is basically no difference between theoretical spectra with log(g) = 8.77 or log(g) = 8.8, which was our initial assumption) and a radius of about 5000 km. If we take the error in the distance, about 4%, into account, we find an error in the WD radius of 200 km, giving R_wd = 5000 ± 200 km. We find that the WD cools by 5250 K in 41 days, from a temperature of 41,500 K 15 days after the peak of the outburst, down to 36,250 K 56 days after the peak of the outburst. The cooling of the WD is displayed in Figure 18. The WD radius and temperatures we obtained in our analysis are consistent with the results from the FUSE spectral analysis of Long et al. (2006), who obtained a temperature of T_eff ∼ 43,600–47,100 (for log(g) ∼ 8.5–9.0, respectively) 10 days after the peak of a wide outburst, and T_eff ∼ 30,300–31,600 (for log(g) ∼ 8.5–9.0) 135 days after the peak of a (different but similar) wide outburst.

**Figure 18.** Effective surface temperature of the WD shown as a function of time since the peak of the outburst. The temperatures were obtained assuming a massive WD with log(g) = 8.8 and a distance of ∼100 pc. The outburst was a wide outburst lasting about 15 days.
Download figure:
Standard image High-resolution image

4. Summary and Conclusion

We obtained HST/COS spectra of U Gem following a wide outburst when the WD dominated the FUV. The spectra were taken at four epochs, 15 days, 21 days, 33 days, and 56 days after the peak of the outburst, and reveal a decreasing FUV flux as a function of time, an obvious sign of the cooling of the WD.

1.
The HST/COS spectra of U Gem exhibits the usual hydrogen Lyman and helium Balmer absorption lines forming in the WD photosphere, as well as the absorption lines of low-ionization energy species of carbon (C ii iii iv), nitrogen (N iii), sulfur (S iv), and silicon (Si ii iii iv). We also identified absorption lines of iron (Fe ii), aluminum (Al iii), and possibly also argon (Ar iii). All of the metal lines could potentially form in the photosphere of a moderately hot WD. However, many of them (i.e., Si, Al, Ar) are deeper at orbital phases ϕ ∼ 0.25 and ϕ = 0.67–0.81, indicating the presence of absorbing material at these specific phases. Higher-ionization energy species of N iv and O vi seen in absorption are known to form at a much higher temperature (∼80,000 K) and are not attributed to the WD. The spectra do not show any emission features.
2.
The first epoch spectra cover an entire orbital phase and show a variation in the continuum flux level with the orbital phase that is more pronounced in the short-wavelength range. The FUV light curves display an increase in flux near phase ϕ = 0.9, a small dip near phase ϕ = 0.25, and a stronger dip near phase ϕ = 0.78. The increase in flux at ϕ = 0.9 is most likely due to the hot spot facing the observer (Unda-Sanzana et al. 2006; Ecchevarría et al. 2007). The dip around phase ϕ = 0.78 is due to L1-stream material overflowing the disk rim and falling toward smaller radii, where it has been shown to bounce back off the disk plane near phase ϕ = 0.5; there, it is redirected toward ϕ ≈ 0.2 (Kunze et al. 2001), where it is likely responsible for the small dip observed in the FUV light curve at ϕ = 0.25. The overflowing L1-stream material is also responsible for the strong orbital phase variation of the depth of the the absorption lines of silicon, aluminum, and argon.
3.
Our analysis of the absorption line velocity as a function of the binary orbital phase reveals an overall redshift in the velocity of the silicon lines, especially Si iii and iv, compared to the WD orbital velocity, indicating material falling toward the WD. The largest velocity shift in the lines is observed around phases ϕ ∼ 0.78–0.91. The carbon lines appear to follow more closely the WD orbital velocity, implying that they are probably forming in the WD photosphere.
4.
The spectra exhibit solar to subsolar abundances of carbon and sulfur, suprasolar (up to 20×) abundances of nitrogen, aluminum, and argon, while iron is only marginally suprasolar (2×). Silicon is subsolar except for Si iii (∼1112 Å) and Si iv (1400 Å), which are solar. Our analysis indicates that the silicon, aluminum, and argon lines are at least in part forming in the material above the disk veiling the WD at orbital phases ϕ ∼ 0.25 and 0.75. In spite of the fact that many lines might be affected by absorbing material in front of the WD at all orbital phases, our analysis confirms the high N/C ratio observed previously in the FUV spectra of U Gem in quiescence, a sign of possible CNO enrichment of the accreted material on the WD surface.
5.
At the time of the first epoch, 15 days after the peak of the outburst, the WD was still rather hot, reaching a temperature of 41,500 K. By the time of the fourth epoch, 41 days later, the WD had cooled down to a temperature of 36,250 K. The temperatures we obtained are consistent with previous observations of U Gem in quiescence for the log(g) considered here (Long et al. 2006) and are theoretically consistent with the compressional heating and subsequent cooling of the WD as shown by Sion (1995).
6.
The spectra are consistent with a single-temperature WD component, and two-temperature WD models did not provide any improvement to the fit. Our single-WD temperature model fits are self-consistent with a massive WD with an effective surface gravity log(g) = 8.77 and a (∼40,000 K) WD radius of 5000 ± 200 km corresponding to a WD mass M_wd ≈ 1.11 M_⊙ and are consistent with a previous spectral analysis of U Gem in quiescence (Long et al. 2006).
7.
We confirm that phase-dependent absorption is the most plausible interpretation of the time-variable absorption seen in the FUV spectra of U Gem during quiescence, and we concur with Long et al. (2006) that it complicates the analysis of the WD spectra.The picture that emerges from the orbital phase variability of the spectra is a familiar one. The L1-stream material hits the disk's rim near ϕ = 0.81–0.91, creating a peak in the UV light curve at that orbital phase. Some of the material overflows the disk's edge and moves to smaller radii, where it reduces the UV flux at phase 0.78, increases the depth of some absorption lines at ϕ = 0.78–0.81, and adds a redshifted component to absorption lines near ϕ = 0.78–0.91. The overflowing material eventually falls back onto the disk near phase 0.5 (as corroborated by Doppler mapping, Ecchevarría et al. 2007, and simulations, Kunze et al. 2001), where it bounces back off the disk and continues in a trajectory toward ϕ = 0.25. At this phase, the material is high enough above the disk to veil the WD for a second time, creating a second dip in the UV light curve as well deeper absorption lines in the spectrum.

P.G. is pleased to thank William (Bill) P. Blair at the Henry Augustus Rowland Department of Physics and Astronomy at the Johns Hopkins University, Baltimore, MD, for his kind hospitality. We wish to thank an anonymous referee for comments that helped improve the manuscript. We used some of the online data from the AAVSO, and are thankful to the AAVSO and its members worldwide for making these data public and for their constant monitoring of cataclysmic variables. This research was based on observations made with the NASA/ESA Hubble Space Telescope, obtained at the Space Telescope Science Institute, located in Baltimore, MD. Support for program #12926 was provided by NASA through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555.

Facility: HST(COS) - Hubble Space Telescope satellite.

Software: IRAF, FORTRAN, PGPLOT, TLUSTY/SYNSPEC (all installed and running under Cygwin-X).

HST/COS Far-ultraviolet Spectroscopic Analysis of U Geminorum Following a Wide Outburst^{^*}

Article metrics

Permissions

Author e-mails

Author affiliations

ORCID iDs

Dates

Abstract

1. Introduction: U Geminorium—The Prototypical Dwarf Nova