This field can be roughly divided into two halves, with results from the former half being used to tackle problems from the latter half:
The estimates mentioned above are linear, however in the past ten years it has been realized (by Klainerman, Machedon, Bourgain, and others) that bilinear estimates for wave equations may be more fundamental and yield better results for applications. In particular, the problem of estimating null forms such as
Q(\phi,\psi) = \phi_t \psi_t - \nabla \phi . \nabla \psi
where \phi, \psi solve the homogeneous or inhomogeneous wave equation, are key to the low-regularity theory of non-linear wave equations. These estimates have traditionally been proven using Plancherel's theorem and were thus restricted to L^2-based spaces, but recent advances in bilinear restriction theory have opened the way to L^p null form estimates which may soon prove useful in applications. Null forms are also sometimes connected to Strichartz estimates. The bilinear theory for the cone parallels that of the sphere, and has many applications beyond non-linear wave equations.
Once one can obtain good control on solutions to the linear equation, one can often obtain local well-posedness for small perturbations of this equation, even if the data is somewhat rough. There is a standard technique to do this, which goes by various names including perturbation theory, Picard iteration, the method of power series, and the bootstrap argument. The effectiveness of this technique depends on whether the regularity of the initial data is sub-critical, critical, or super-critical. In the sub-critical regime these methods work fairly well. When the regularity is critical then the argument becomes much more delicate, and perturbation theory techniques can sometimes fail completely. Indeed, some equations are ill-posed at the critical regularity. In the super-critical regime it is expected that ill-posedness occurs for virtually all equations.
For simple equations such as the semi-linear wave equation, a satisfactory local well-posedness theory can be obtained by the Strichartz estimates, which give L^p control on the solution, although in the case of very low regularity one must use more intricate norms. However, for more interesting equations the non-linearity is more complicated and tends to involve the null forms mentioned earlier. In this case one seems forced to use null form estimates to get close to the critical regularity, and it seems one must use even more precise techniques (multilinear estimates?) to push these equations to the critical level.
Once local well-posedness has been obtained, the next step is to look for global well-posedness, and possibly scattering to a linear solution. For very smooth, rapidly decaying data this can often be achieved by conformal compactification or by showing that the solution decays and becomes more regular in certain directions as time progresses; without enough decay, these equations can blow up. For small data one can sometimes use the critical theory (which is somewhat insensitive to the local/global distinction). If the data has finite energy, then one can also exploit energy conservation to get global well posedness (but not necessarily scattering); this works best when the energy is positive definite, and the energy norm is sub-critical. When the energy norm is critical then this strategy can be threatened by the possible phenomenon of energy concentration; however, many equations have conservation laws or monotonicity formulae which prevent this concentration from occuring. Recently, Bourgain has demonstrated a general argument which shows that global existence can also be obtained slightly below the energy norm if there is an additional smoothing effect in the equation, although it seems that one can sometimes do without this smoothing.
I have worked on endpoint Strichartz estimates and their failure in low dimension for both the wave and Schrodinger equation. I've also investigated blowup for the Klein-Gordon and 1D wave-map equations under certain circumstances, and global existence below the energy norm for 1D wave maps and Maxwell-Klein-Gordon under different circumstances. This work is mostly in collaboration with Mark Keel.
This area is only a small part of the field of non-linear wave equations, which in turn is a sub-field of non-linear PDE, and yet there are an incredible number of people who are currently working in this area. There are also many other people working on analogues of these results in curved space, with obstacles, or on Lie groups, and on the closely related problems for Schrodinger and KdV.
If you are interested in learning more about this field, I suggest: