Double-diffusive convection refers to mixing where the effects of thermal and composition gradients compete to determine the stability of a fluid. In addition to the familiar fast convective instability, such fluids exhibit the slow, direct salt finger instability and the slow, overstable semiconvective instability. Previous approaches to this subject usually have been based on linear stability analyses. We develop here the nonlinear mixing-length theory (MLT) of double-diffusive convection, in analogy to the more familiar MLT for a fluid of homogeneous composition. We present approximate solutions for the mixing rate in the various regimes, and show that the familiar Schwarzschild and Ledoux stability criteria are good approximations to the precise criteria in stellar interiors. We have implemented the self-consistent computation of the temperature gradient and turbulent mixing rate in a stellar evolution code and solved a diffusion equation to mix composition at the appropriate rate. We have evolved $15\msun$ and $30\msun$ stars from the zero-age main sequence to the end of core He-burning. Semiconvective mixing is fast enough to alter stellar composition profiles on relevant time scales, but not so fast that instantaneous readjustment is appropriate.

We solve the nonlocal convection equations. The solutions for four model problems are compared with results of GSPH simulations. In each case we test two closure schemes: 1) where third moments are defined by the diffusion approximation; and 2) where the full third moment equations are used and fourth moments are defined by a modified form of the quasi-normal approximation. In overshooting models, the convective flux becomes negative shortly after the stability boundary. The negative amplitude remains small, and the temperature gradient in the overshooting zone has nearly the radiative value. Turbulent velocities decay by a factor of $e$ after $0.5$--$1.5\ell$, depending on the model, where $\ell$ is the mixing length. Turbulent viscosity is more important than negative buoyancy in decelerating overshooting fluid blobs. These predictions are consistent with helioseismology.

Possibly the only unambiguous verification that gamma-ray bursts (GRBs) are at cosmological distances would be the observation of multiple images of a gravitationally lensed burst. Each images would arrive at a different time, but exhibit identical light curves. We improve upon previous calculations of GRB lensing by using revised cosmological burst models based upon better knowledge of burst spectra, and we consider several sets of cosmological parameters. Only lensing by the known population of galaxies is considered, and we compute the lensing rate and amplification bias using three different lens models. Assuming that 800 bursts per year are above the detection threshold of the BATSE experiment on the Compton Gamma-Ray Observatory, we predict one lensing event every 1.5--25 years, with a median time delay between images of about 7 days. However, BATSE has an instrumental efficiency of about 12\% for detecting double bursts, decreasing the detection rate to fewer than one per century for some cosmological parameters. This inefficiency can be overcome partially if cases of quadruple imaging by elliptical potentials are common, but simulations show that lensing by elliptical potentials enhances the detection rate by only about 20\%. We conclude that it is unlikely that the BATSE experiment will detect a lensed burst, but that a future experiment, which is either much more sensitive or has a much more efficient duty cycle, could.

A gravitationally lensed gamma-ray burst (GRB) would appear as multiple bursts with identical light curves, separated in time and differing only by the scaling of their amplitudes. However, noise may make them difficult to identify as lensed images. Furthermore, faint, intrinsically similar, yet distinct light curves may be falsely identified as lensing events. In this paper we develop two statistical tests to distinguish noisy burst light curves. We use Fourier analysis techniques to analyze the signals for both intrinsic variability and variability due to noise. We are able to determine the noise level, and we compare the bursts only at frequency channels that are signal dominated. Utilizing these methods, we are able to make quantitative statements about whether two bursts are distinct. We apply these statistics to scaled versions of two subbursts of GRB 910503--- subbursts previously investigated by Wambsganss (1993) using a different statistical test. We find that our methods are able to distinguish these bursts at slightly smaller amplitudes than those at which Wambsganss's method succeeds. We then apply our techniques to ``candidate" lensing events taken from the BATSE catalogue, and we find that nearly all of them, except for the very shortest ones (durations $\aproxlt 0.3 ~s$), are distinguishable. We therefore expect that a majority of bursts will be disinguishable from one another.