Eta Carinae's historical light curve: evidence for cyclic Roche lobe overflow from the primary star

Abstract: The large amount of ground-based photometric measurements of $η$ Carinae obtained since 1940 have remained problematic for quantitative modeling due to the blending of flux from the stellar core and the surrounding circumstellar nebula. In the era of the Hubble Space Telescope, spatially resolved imaging & spectrophotometry have enabled disentanglement of these components, allowing recovery of the stellar core $V$-band brightness from ground-based observations. We isolate the $V$-band fluxes of the stellar core and nebula using 1999--2020 HST (ACS and STIS) observations, and use these to calibrate coeval ground-based photometry. The main finding is an orbital light curve with an amplitude $Δm \approx \pm 0.2\,$mag, many times higher than that modeled by ellipsoidal deformation of the primary. The observations suggest that Roche lobe overflow starts at $- 75$ days before periastron in coincidence with the start of rising in the orbital light curve, and remains for 150 days. An expanding (and afterwards dissipating) gas cloud reflecting the light from the primary would explain the observed large amplitude of the orbital light curve. A sharp periodic photometric peak occurs at $\sim -18$ days from the periastron. It is followed by a broad minimum around the superior conjunction of the secondary (T$_0+5.2$ days), which we interpret as a partial eclipse of the ejected material, in coincidence with the \emph{shallow minimum} in X-rays, which also has been attributed to an eclipse.

• The paper reconstructs Eta Carinae’s stellar-core light curve from ground-based photometry and HST observations, separating the nearly steady nebular contribution from stellar variability across six decades.
• The recovered orbital modulation reaches about ±0.2 mag—roughly ten times larger than ellipsoidal models predict—and indicates Roche lobe overflow beginning 75 days before periastron and lasting about 150 days.
• The proposed scenario links reflected light from ejected material and an eclipse of extended accreting gas to the observed photometric minimum, while declining peak amplitudes and uncertain stellar-radius estimates remain key tests.

Motivation and scope

The photometric evolution of $\eta$ Carinae since 1940 has been documented by extensive ground-based monitoring, yet it has never received a comprehensive quantitative treatment because the unresolved $V$-band flux blends light from the stellar core with the surrounding Homunculus Nebula. This paper addresses that deficit by combining six decades of ground-based photometry (1963–2026) with spatially resolved HST (ACS/HRC and STIS) imaging and spectrophotometry from 1999–2020 to isolate the stellar core's brightness. The central result is an orbital light curve of the core with amplitude $\Delta m \approx \pm 0.2$ mag — roughly an order of magnitude larger than what ellipsoidal deformation of the primary can produce — which the authors interpret as evidence for periodic Roche lobe overflow (RLOF) beginning about 75 days before periastron and lasting 150 days.

Disentangling core and nebula

The key methodological step is the recovery of the core magnitude from seeing-limited data. The nebular component is nearly constant at $V \approx 5.5$ mag over 1999–2026 ($V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$), reflecting that the secular brightening is restricted to the authors' line of sight rather than being seen in all directions of the ejecta. Subtracting this stable nebular reference from total-object magnitudes yields $V_{core}$ via a simple two-component decomposition. For the pre-HST era (1963–1999), the authors bridge between the photographic core magnitude $m_V = 7.6 \pm 0.3$ mag in 1946 (O'Connell; Thackeray) — which carries generous uncertainty — and the HST-measured $V_{core} = 6.9 \pm 0.07$ at 1999.45, giving an average core brightening rate of $-0.006$ mag yr$^{-1}$.

Two distinct brightening regimes emerge: the whole object brightened at $V$0 mag yr$V$1 during 1963–1999, then jumped by a factor of $V$2 to $V$3 mag yr$V$4 after 1999, when the stellar-core brightening rate itself increased twentyfold. The post-1999 event is attributed not to any change in the star but to dissipation of a coronagraph-like dust occulter at $V$5–1700 AU along the line of sight (2608.16818), consistent with earlier suggestions based on H$V$6 equivalent widths measured directly on the star versus in reflected nebular light.

The orbital light curve and its excess amplitude

Detrending the recovered core light curve with a second-order polynomial exposes a coherent orbital modulation across four cycles (2003–2025). The $V$7 color index of the core remains essentially constant through the cycle, superficially consistent with ellipsoidal variation of the reddened primary. However, Wilson-Devinney modeling constrained by the Keplerian elements ($V$8 d, $V$9, $\Delta m \approx \pm 0.2$0, $\Delta m \approx \pm 0.2$1, $\Delta m \approx \pm 0.2$2) produces a synthetic curve that is nearly flat except for a $\Delta m \approx \pm 0.2$3 mag rise followed by a $\Delta m \approx \pm 0.2$4 mag dip near periastron — far smaller than observed. Even adopting the interferometric pseudo-photospheric radius $\Delta m \approx \pm 0.2$5 (VLTI-MATISSE) and rescaling the orbit so the Roche lobe accommodates it leaves the synthetic curve unchanged, as expected from the scale-free nature of Roche geometry at fixed mass ratio. The authors therefore conclude that geometric deformation alone cannot account for the observations, regardless of assumed stellar scale.

Cyclic Roche lobe overflow scenario

Using the Eggleton approximation, the primary's Roche lobe shrinks by a factor of $\Delta m \approx \pm 0.2$6 from apastron to periastron. With $\Delta m \approx \pm 0.2$7, overflow begins 75 days before periastron — coincident with the onset of the light-curve rise at phase $\Delta m \approx \pm 0.2$8 — and persists for 150 days. The proposed mechanism is reflection: an expanding cloud of ejected material scatters the primary's light, inheriting its constant $\Delta m \approx \pm 0.2$9 color and producing the large orbital amplitude. Cycle-to-cycle decorrelation supports a contribution from ejected matter rather than purely geometric effects.

A sharp photometric peak occurs at $V \approx 5.5$0 days before periastron, followed by a broad minimum centered on $V \approx 5.5$1 days, exactly at the superior conjunction of the secondary derived from radial-velocity solutions. Because the luminosity ratio $V \approx 5.5$2 implies that complete occultation of the secondary would produce only $V \approx 5.5$3 mag, whereas the observed minimum reaches $V \approx 5.5$4 mag, the authors argue the eclipsed source must be extended accreting material around the secondary rather than the star itself. This minimum coincides with the shallow X-ray minimum, independently attributed to an eclipse (2608.16818). The peak amplitude also declines secularly, from $V \approx 5.5$5 mag in 2003/2009 to $V \approx 5.5$6 mag in 2020/2025, and the peak may simply be the left edge of the post-periastron eclipse rather than a discrete ejection event.

Dust grain evolution in the occulter

During 1999–2012 the core brightened by $V \approx 5.5$7 mag while bluing by $V \approx 5.5$8 mag, implying $V \approx 5.5$9 and grain sizes $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$0m, $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$1m under standard grain models. The constancy of $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$2 indicates neutral extinction at shorter wavelengths. After 2015 all color indices flatten while brightening continues, implying only large grains ($V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$3m) survive, still being destroyed. The authors note $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$4 may be overestimated if part of the $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$5-band change reflects gray extinction, in which case the inferred grains are larger.

Limitations and open questions

Several caveats bear directly on the interpretation. The pre-1946 core zero point rests on uncertain photographic magnitudes ($V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$6 mag); the simplified core/nebula subtraction ignores possible local nebular fluctuations, though these appear minor post-1999; and the nature of the pre-periastron peak remains explicitly uncertain. The RLOF conclusion depends on adopting $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$7, which the authors themselves note likely represents the optically thick wind pseudo-photosphere rather than the hydrostatic stellar radius — the physical radius entering the overflow condition is therefore model-dependent. The binary-merger origin of the system sits uneasily with the projected final magnitude $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$8 (near the pre-eruption brightness), since merger products should be brighter by $V_{neb} = 5.5 - 0.017(\mathrm{yr}-2003.87)$9 mag, and no fast rotation consistent with a merger remnant has yet been demonstrated. If periastron mass ejection is real, its impact on apsidal precession and orbital evolution remains unquantified.

Conclusion

By calibrating ground-based photometry against HST-resolved measurements, this work recovers the first quantitatively modeled $V_{core}$0-band light curve of $V_{core}$1 Carinae's stellar core spanning more than six decades. The orbital modulation's amplitude exceeds ellipsoidal-model predictions by an order of magnitude, and its phasing — overflow onset 75 days before periastron, eclipse-like minimum at secondary superior conjunction matching the shallow X-ray dip — supports a picture of cyclic Roche lobe overflow with episodic mass transfer toward the companion. Confirmation will require testing whether the predicted apsidal-precession signature and continued decline of the peak amplitude persist in subsequent orbital cycles.