**2. A Checkered History**

The origins of this topic are summarized in an appendix in [1]. Since only 2 to 4 giant eruptions or supernova impostors were known before 2000, they were conflated with LBV outbursts. Early discoveries followed two paths, and today we are not ye<sup>t</sup> sure whether those paths really intersect. First, the examples of *η* Car, P Cyg, SN 1961v, and SN 1954J were known before 1970 [2]. Their eruptions produced supernova-like amounts of radiation, but with longer durations than a supernova and the stars survived. (SN 1961v may have been a true supernova [3,4], but, ironically, that doesn't alter its historical role.) The second path began with the recognition in 1979 of an upper boundary in the empirical HR diagram, the diagonal line in the middle of Figure 1 [5]. Almost no stars are found above and to the right of that line, and the rare exceptions are temporary. Since massive stars evolve almost horizontally across the diagram, this boundary indicates some sort of barrier to the outer-layer evolution of stars with *M* > 50 *M*. The probable explanation involves episodic mass loss as follows.

Note the various zones and boundaries in Figure 1, though in reality they are not so well defined. If a massive star loses a considerable fraction of its mass, then it cannot evolve far toward the right in the HR diagram. Thus a good way to explain the HRD boundary is to suppose that stars above 50 *M* lose mass in some process that exceeds their line-driven winds. The S Doradus class of variable stars occurs in the "LBV1" and "LBV2" zones in Figure 1, to the left of the empirical boundary. They are remarkably close to the Eddington Limit (Section 3.1 below), and they exhibit sporadic outbursts which expel more material than their normal winds. If every star above 50 *M* behaves in that way after it evolves into the LBV1 zone, then the boundary is an obvious consequence. This scenario was proposed as soon as the empirical limit was recognized [5]. In a variant idea noted in Section 3.1 below, the decisive mass loss occurs just before the stars become LBV's; but both hypotheses invoke eruptions in that part of the HR diagram. No better alternative has appeared in the decades since they were proposed.

**Figure 1.** The empirical upper boundary and LBV instability strip in the Hertzsprung-Russell Diagram. In reality they are ill-defined and may depend on rotation and chemical composition. The interval between the LBV strip and the boundary is very uncertain. The zero-age main sequence on the left side shows initial masses, and most of a very massive star's evolution occurs at roughly twice the initial luminosity.

Today, S Dor variables are usually called LBV's, an acronym for "Luminous Blue Variables." Rightly or wrongly, they are frequently mentioned in connection with giant eruptions and supernova impostors. Many of them are easier to observe than eruptions in distant galaxies, and they probably offer hints to the relevant physics, Section 5 below.

Thus the key facts—episodic mass loss, and the existence of giant eruptions—were well recognized before 1995, and credible mechanisms had been noted; see many references in [1]. A decade later, when the mass loss rates of normal line-driven winds were revised downward [6], the same concepts were proposed again as a way to rescue the published evolution tracks (e.g., [7]). Unrelated to that development, extragalactic giant eruptions attracted attention after 2003 and some of them were aptly called "supernova impostors" because their stars survived [8]. Modern SN surveys found many examples [9], often classed among the Type IIn SNae. Some giant eruptions preceded real supernova events, the most notorious being SN 2009ip where the real SN explosion did not occur until 2012 [10,11]. Theoretical explanations continue to be diverse and highly speculative.
