Luminous Blue Variable

Stars living on the edge

Most stars are boringly stable. The Sun has shone within a fraction of a percent of its current brightness for billions of years, held in equilibrium by a self-correcting thermostat: if the core compresses, it heats, fusion speeds up, pressure rises, and it expands back. The most massive stars in the universe have no such comfort. A luminous blue variable (LBV) is a star so luminous — around a million times the Sun, L ≈ 10⁶ L_⊙ — that the very light it produces is on the verge of blowing it apart. It sits at the Eddington limit, the knife-edge where outward radiation pressure nearly equals inward gravity, and the result is a star that pulses, flickers, and occasionally erupts with the energy of a small supernova.

The defining property is not just brightness but instability. LBVs vary on a hierarchy of timescales: nearly imperceptible microvariations of a few hundredths of a magnitude; the characteristic S Doradus cycles of 1–2 magnitudes over years to decades; and the rare, spectacular giant eruptions in which the star brightens by several magnitudes and sheds an entire Sun's worth of mass — or ten — in a geological blink. The prototype, eta Carinae, did exactly this in the 1840s and lived to tell about it. LBVs are short-lived transitional objects, a bridge in the evolution of the most massive stars between the blue supergiant phase and the hydrogen-stripped Wolf-Rayet stage.

How it works: the Eddington limit and the iron bump

The physics that makes an LBV unstable starts with the Eddington luminosity. Photons carry momentum, and when they stream outward through a star's gas they push on it. For fully ionized hydrogen, the relevant opacity is electron (Thomson) scattering, and the luminosity at which radiation pressure exactly balances gravity is

L_Edd = 4π G M c / κ ≈ 3.2 × 10⁴ (M / M_⊙) L_⊙

The dimensionless Eddington ratio Γ = L / L_Edd measures how close a star is to the brink. For the Sun, Γ ≈ 3 × 10⁻⁵ — utterly safe. For an LBV with M ≈ 70 M_⊙ and L ≈ 10⁶ L_⊙, the limit is L_Edd ≈ 2.2 × 10⁶ L_⊙, so the global Γ ≈ 0.45. That sounds safe too, until you look at the outer layers in detail.

Opacity is not constant. At temperatures around 150,000–200,000 K, partially ionized iron group elements produce a forest of bound-bound and bound-free transitions — the iron opacity bump — that can raise the opacity by a factor of several over the electron-scattering value. Because L_Edd ∝ 1/κ, a local opacity spike drives the local Eddington ratio Γ_local above 1 in those layers. When Γ_local > 1, radiation pushes harder than gravity pulls, and that layer is no longer bound to the star. It can be lifted off bodily. This is why LBVs perch right at the Humphreys-Davidson limit and why a small perturbation — a change in radius, rotation, or ionization — can trigger runaway, super-Eddington mass loss. The star is a balloon being inflated by its own light, and the skin is thin.

Worked example: how massive a shell can eta Carinae eject?

Let's check whether eta Carinae's Great Eruption is energetically plausible. The Homunculus Nebula contains roughly M_ej ≈ 10 M_⊙ of gas moving at v ≈ 650 km/s. The kinetic energy carried away is

E_kin = ½ M_ej v²
      = ½ × (10 × 2×10³⁰ kg) × (6.5×10⁵ m/s)²
      = ½ × (2×10³¹) × (4.2×10¹¹)
      ≈ 4.2 × 10⁴² J
      ≈ 4 × 10⁴⁹ erg

That is comparable to the radiated energy of the eruption (~10⁴⁹·⁵ erg over the ~20-year event) and only about an order of magnitude below the ~10⁵¹ erg kinetic energy of a true core-collapse supernova. eta Carinae released a near-supernova's worth of energy — and survived. Now ask how long the star could sustain this. At L ≈ 10⁶ L_⊙ ≈ 3.8 × 10³² W, the eruption's total radiated energy of ~10⁴²·⁵ J divided by this luminosity gives a duration of about

t = E_rad / L ≈ (3×10⁴² J) / (3.8×10³² W) ≈ 8×10⁹ s ≈ 250 yr

— but during the eruption the luminosity was super-Eddington, several times higher, compressing the radiative output into the observed ~20 years. The mass-loss rate during the peak was a staggering dM/dt ≈ 0.5 M_⊙ per year, four to five orders of magnitude above the already-prodigious steady winds of normal hot supergiants (~10⁻⁵ M_⊙/yr). No line-driven wind can do this; the eruption must be a continuum-driven, porosity-moderated super-Eddington outflow. The numbers hang together: an LBV near the Eddington limit can plausibly unbind ~10 M_⊙ in a decade without destroying itself.

Regimes of variability

LBV variability is best understood as a ladder of increasingly violent regimes, all driven by the same proximity to the Eddington limit:

• Microvariations. Amplitude ≲ 0.1 mag, timescales of weeks to months. Low-level pulsation and wind fluctuation, present essentially all the time.
• S Doradus cycles (normal eruptions). Amplitude 1–2 mag, timescales of years to decades. The bolometric luminosity stays roughly constant; the star swells into a cool, dense pseudo-photosphere at maximum and contracts to a hot, compact state at minimum, trading temperature for radius. This is the defining, recurrent behavior used to confirm an LBV.
• Giant eruptions. Amplitude ≳ 2–3 mag (sometimes much more), recurrence of centuries to millennia. The bolometric luminosity genuinely rises, becoming super-Eddington, and the star sheds a large fraction of a solar mass to several solar masses. eta Carinae (1840s) and P Cygni (1600) are the two historical Galactic examples.
• Supernova impostors. The extragalactic manifestation of giant eruptions — transients that briefly reach absolute magnitudes near −14, faint by supernova standards but bright enough to be discovered by surveys and initially misclassified as genuine explosions, only for the star to reappear.

There is also a divide by mass. Classical, high-luminosity LBVs (log L/L_⊙ ≳ 5.8, descended from M_initial ≳ 50 M_⊙) never become red supergiants — the Eddington barrier stops them. Lower-luminosity LBVs (log L/L_⊙ ≈ 5.4–5.6) may be post-red-supergiant objects that have already dredged up processed material.

The bridge to Wolf-Rayet stars

The reason LBVs matter for stellar evolution is that they do destructive work the rest of the lifecycle depends on. A very massive star (initial mass ≳ 25–40 M_⊙) burns hydrogen on the main sequence as an O star for only a few million years, then expands. For the most massive stars, the path runs:

O star  →  blue supergiant  →  LBV  →  Wolf-Rayet star  →  supernova (or direct collapse)

During the LBV phase — lasting only a few times 10⁴ to 10⁵ years — the combination of strong steady winds and eruptive mass loss strips away most or all of the hydrogen-rich envelope. Once the hot helium-burning core is laid bare, the object becomes a Wolf-Rayet star: a small, blistering-hot star with broad emission lines from a dense, fast (~2,000 km/s) wind. The LBV is the agent of stripping; without its eruptions, the most massive stars could not reach the Wolf-Rayet stage on the timescale observed. This is also why LBVs cluster at the upper boundary of the HR diagram and effectively enforce the Humphreys-Davidson limit — they shed mass so aggressively that no stable, very luminous red supergiant can exist.

Observational status and the supernova surprise

LBVs are genuinely rare. Only about a dozen are confirmed in the Milky Way — eta Carinae, AG Carinae, HR Carinae, P Cygni, and others — with comparable handfuls in the Large and Small Magellanic Clouds (S Doradus itself, R71, R127, the binary HD 5980). Confirmation is hard because the diagnostic S Doradus variability can take decades to reveal itself, so many objects with LBV-like spectra remain "candidate LBVs."

The biggest recent surprise concerns when LBVs die. Standard models said massive stars become Wolf-Rayet stars first and only then explode, so an LBV should never be caught as the immediate progenitor of a supernova. Yet pre-explosion Hubble imaging of SN 2005gl in NGC 266 revealed a luminous point source matching an LBV at the exact location, which then vanished in a Type IIn supernova (Gal-Yam & Leonard 2009). More broadly, the defining feature of Type IIn supernovae — narrow hydrogen emission lines from dense circumstellar gas — is naturally explained by LBV-like eruptions shedding shells in the years to centuries before collapse, the explosion then slamming into that material. So at least some LBVs explode while still LBVs, a direct challenge to textbook pre-collapse evolution. SN 2009ip provided a real-time example: an LBV that erupted repeatedly, was repeatedly called a supernova, and finally appears to have actually exploded.

LBVs versus related massive-star phases

Phase	T_eff (K)	log(L/L_⊙)	Wind/loss rate (M_⊙/yr)	Spectrum	Duration
O main sequence	30,000–50,000	5.0–6.0	~10⁻⁶	absorption, H + He II	~3–4 Myr
Blue supergiant	10,000–25,000	5.3–5.9	~10⁻⁶	H absorption + emission	~10⁵ yr
LBV (quiescent / S Dor)	8,000–30,000 (varies)	5.4–6.4	~10⁻⁵–10⁻⁴	P Cygni emission, Fe II	~10⁴–10⁵ yr
LBV (giant eruption)	~7,000–9,000 (cool shell)	up to 6.5+	~0.1–1 (super-Edd.)	dense, blackbody-like	years–decades
Wolf-Rayet star	30,000–200,000	5.3–6.2	~10⁻⁵	broad emission, He/N/C	~few × 10⁵ yr
Red supergiant	3,500–4,500	4.5–5.7	~10⁻⁶–10⁻⁴	cool absorption, TiO	~10⁵–10⁶ yr

The key columns are luminosity and mass-loss rate. The LBV phase is where the loss rate spikes hardest, especially in eruption, which is precisely what makes it the transformative bridge between the hydrogen-rich and hydrogen-stripped halves of a massive star's life.

Common pitfalls and misconceptions

• "The S Doradus variability means the star is changing its energy output." No — in the classic S Doradus cycle the bolometric luminosity is roughly constant. The visual brightness changes because the star redistributes its fixed energy across the spectrum as its apparent temperature and radius swing. Only in giant eruptions does the total output truly rise.
• "A giant eruption is a kind of supernova." A supernova is (for core collapse) the death of the star — its core implodes. A giant eruption leaves the star intact; eta Carinae erupted in the 1840s and is still there, brighter than ever. They look similar from afar (hence "supernova impostor"), but the underlying physics is opposite: surface instability versus core collapse.
• "LBVs are blue because they are young and hot like O stars." They are evolved, not young. The "blue" refers to their high surface temperature in quiescence, but LBVs are post-main-sequence objects nearing the end of their few-million-year lives.
• "All LBVs become Wolf-Rayet stars before exploding." This was the textbook expectation, but SN 2005gl and the Type IIn class show at least some explode while still in the LBV state, with their freshly ejected circumstellar shells lighting up.
• "eta Carinae is a single star." It is a binary — a massive primary plus a hot companion on a highly eccentric 5.5-year orbit. The companion's interaction is thought to help sculpt the bipolar Homunculus and modulate the system's behavior.
• "The Homunculus is the star's normal extent." The Homunculus is ejecta — material thrown off in the 1840s eruption, now ~0.7 light-years across and still expanding. The star itself is vastly smaller and hidden deep inside the dusty lobes.

Quantitative analysis: why the iron bump matters

Consider why the Eddington ratio can locally exceed 1 even when the star is globally sub-Eddington. The local Eddington ratio in a stellar layer is

Γ_local(r) = κ(r) L(r) / (4π G M(r) c)

Take an LBV with global Γ = 0.45 computed using electron scattering κ_es ≈ 0.34 cm²/g (for solar composition). In the deep interior κ ≈ κ_es and Γ_local ≈ 0.45, comfortably bound. Now move out to the layer near T ≈ 1.8 × 10⁵ K where the iron opacity bump peaks. There the Rosseland mean opacity can rise to κ ≈ 1–2 cm²/g — a factor of 3–6 over κ_es. Since L and the enclosed M(r) are nearly unchanged across the thin outer envelope, Γ_local scales directly with κ:

Γ_local(iron bump) ≈ 0.45 × (κ_bump / κ_es) ≈ 0.45 × (3 to 6) ≈ 1.4 to 2.7

So a star that is only 45% of the way to the Eddington limit on average is locally super-Eddington in its iron-bump layer. That layer experiences a net outward force and accelerates upward, inflating the envelope or, if the bump is strong enough, launching it as an eruption. This is the modern explanation for why LBVs sit exactly where they do in the HR diagram and why their atmospheres are perpetually on the verge of instability. The same machinery — opacity-driven, super-Eddington, continuum-rather-than-line-driven mass loss — is what powers the giant eruptions, with the eruption simply being the runaway case where the inflated layers detach entirely. The exact trigger (envelope inflation, binary interaction in eta Car, a sub-surface energy deposit from late-stage nuclear burning) is still debated, but the energetics and the location are well explained by Γ_local crossing unity at the iron bump.