Transient phenomena in the cosmos present some of the greatest challenges to our understanding of Nature, and to our ability to extract information from data. They also provide windows deep into the structures of objects otherwise inaccessible to us, and serve us as surrogates for the experiments that we cannot otherwise perform on those objects.
Transients occur when systems change rapidly. Violent phenomena are characteristically transient: an explosion occurs, and the disassembly is final and irreversible. An instability results in the dumping of a large amount of energy in a relatively small place, and we observe the production of entropy that arises. There is promise of new physics in the study of gamma-ray bursts, or of supernova explosions, or of active galactic nuclei. And much old physics has been applied toward learning the detailed structure and evolution of stars by watching the way they pulsate.
Variable stars, of course, can be catalogued and studied with some regularity and predictability. Even flaring stars, or active galactic nuclei, can be selected for monitoring based on a known history of previous activity, though episodes are unpredictable. But true right-out-of-the-blue transients are the most interesting and the most challenging of sources.
The understanding of transient sources involves constructing physical and computational models of the source regions, predicting how they evolve, and comparing those predictions with the observational data. Unfortunately, the rapid evolution of transients makes both the computational side and the observational side exceedingly challenging.
Because true transients are rare, an observing program for discovering them must continually monitor the whole sky, or a large fraction of it. For every object that is seen to change rapidly there are millions that change either not at all or in a regular fashion. Sophisticated and automated data processing techniques must be employed to aid the search, and since much is learned at the threshold of detectability, these algorithms must contend with nonuniform and fluctuating background levels and low signal-to-noise ratios.
Computationally modeling rapidly changing systems is also difficult because the relevant timescales and processes are so uncertain. There are no equilibria available to start from, and assumptions that certain processes can be ignored often lead to peril. Supernova calculations have made great progress in recent years, and stand as a good example of the success of truly dynamic and sophisticated computational physics. These calculations were helped enormously by the observations of neutrinos from SN 1987A, confirming an extremely important piece of the puzzle that had long been suspected, but until then unproven. For gamma-ray bursts, or active galactic nuclei, or X-ray flaring sources, the missing piece may be one of several competing hypotheses now floating around, or something completely different, that awaits confirmation only by some as yet unforeseen data. In the meanwhile theorists have to pursue a variety of computational models, ever more sophisticated and dynamic.