[This text is taken from the Introduction of RS' Ph.D. thesis]
General relativitythe
fascinating theory of space, time and gravitation as formulated by A. Einstein 85
years agodramatically differs from all other field theories. Spacetime is
no longer given beforehand but rather it is described by a 4dimensional
manifold together with a Lorentzian metric, which itself is subject to
field equations. More precisely, the curvature of the metric is related to
the energy momentum content of the spacetime via the socalled Einstein
equations, whichwritten in coordinatesform a complicated hyperbolic
system of 10 nonlinear partial differential equations of second order for
the coefficients of the metric tensor; hence one has to assume the metric
to satisfy sufficient differentiability conditions. Usually one supposes
the metric to be smooth; C^{2} suffices to do classical differential
geometry, whereas C^{3} is needed to ensure energy conservation. C^{2}
(i.e., the first derivatives Lipschitz continuous) guarantees at least
unique solvability of the geodesic equations and local boundedness of
the Riemann curvature tensor.
A singularity in a
general relativistic spacetime intuitively is a
``place'' where curvature blows up or some other ``pathological
behavior'' of the metric occurs. There are two main obstacles, however,
to make this notion precise. The first originates from the fact that
we only may speak of an event at all (i.e., of a point in spacetime)
ifaccording to the abovethe metric is, say, C^{2} ``there.''
Hence a singularity has to be viewed as a
``singular boundary point'' rather than a point in spacetime. Since a Lorentzian
metriccontrary to a Riemannian onedoes not give rise to a distance
function, the construction of a topological boundary is a nontrivial matter.
In fact, no fully satisfactory general notion of a singular boundary of
spacetime exists.
The second difficulty in defining a singularity in general relativity
is deeply connected with another of the theory's main principles: diffeomorphism
invariance. The fact that the components of, say, the Riemann tensor blow up along a curve may simply
be due to a bad choice of coordinates in the following sense. It might be possible to find
a different coordinate system which allows to extend the spacetime beyond
the ``critical point'' with the Riemann tensor perfectly wellbehaved.
Moreover, there is a big variety of examples showing that the blowup of some
curvature quantity is not an adequate tool to detect singularities.
From the above it should become clear why it is tempting
to characterize singularities in general relativity
by their geometrical properties rather than by their analytical ones.
In fact, by the standard definition a spacetime is called
singular if there exist incomplete geodesics, i.e., geodesics of finite affine
parameter length which may not be extended. Obviously designed to capture
the intuitive notion of a ``hole in spacetime,'' there are, however,
also some problems associated with this ``geometric'' approach. First, it does not
provide an ultimate answer to the question what a singularity actually is; note
that we have only defined the notion of a singular spacetime. Instead
there is a quite lengthy catalogue of possible ways in which a
spacetime could ``break down'' (nonsmoothness, unboundedness or
local nonintegrability of the Riemann tensor, spacetimes ``created'' with a
primordial singularity and the like).
Moreover, by the singularity theorems of Penrose and Hawking (see
e.g. [4], chap. 8) many
physically reasonable spacetimes (in particular, all realistic
models of an expanding universe and of gravitational collapse) are
singular with respect to this definition.
Consequently, the recent development of the study of spacetime singularities
has focussed more upon a study of the field equations.
General relativity
as a physical theory is governed by particular physical equations; what is of
primarily interest is the breakdown of physics which may,
or may not, result in a breakdown of geometry.
Unfortunately, there is somehow a conflict between the mathematical
contexts appropriate to, on the one hand, partial differential equations
and, on the other hand, geometry. In the differential geometric study of singularities
one deals with geodesic equations which are uniquely solvable provided the
metric is C^{2} (as already remarked above); beyond this, the differenatibility
of the metric is of little geometrical significance. By contrast, in the study
of hyperbolic PDEs the question of differentiability is crucial; the differentiability
chosen determines the character of the solutions allowed. By choosing
low differentiability one admits solutions like shock waves or impulsive waves, which,
on the other hand, are ruled out as ``singular'' when insisting on high differentiability.
Accepting the field equations to play the primary role, they should determine the
choice of differentiability. A singularity should be an obstruction to
the existence of a solution to the field equations. Clearly, this cannot be
determined in general. On the other hand, there are existence theorems
that tell when it is possible to construct a solution, i.e., we know what is not
a singularity. However, in general it is difficult to link back a certain differentiability
condition necessary to prove (local) existence of solutions to Einstein's equations
(Sobolev conditions, usually imposed on the
Riemann tensor on specific hypersurfaces in a particular class of coordinate
systems, well suited for the formulation of the Einstein
equations as an initial value problem) to geometry,
that is the differentiability of the metric on the entire spacetime manifold.
Summing up the entire discussion so far, it would be very desirable to have a
description of singularities as internal points of the spacetime manifold
where the field equations are satisfied in a weak (probably distributional) sense
(see also [2]).
Hence one would wish to significantly lower the ``geometric
differentiability bound'' (i.e., C^{2}) on the metric.
Indeed, a recent monograph on the subject
of spacetime singularities [1] reaches the conclusion that the answer
to many of the questions raised above ``involve detailed considerations of
distributional solutions to Einstein's equations, leading into an area that is only
starting to be explored [...]''
In particular, one wants to be able to describe spacetimes containing matter
whose density function is e.g. unbounded but integrable or confined
to a submanifold in spacetime (both scenarios amounting to a finite mass per unit volume).
Important examples in this class include thin cosmic strings and impulsive gravitational
waves.
Another strong motivation for a study of distributional
spacetimes arises from the fact that by an argument of Isham [5] the latter will
substantially contribute to a path integral description of (a yet to be formulated theory of)
quantum gravity.
The big challenge in setting up a framework that might be called a
``distributional geometry'' adapted to the needs of general relativity of
course resides in the immanent nonlinearity of the latter theory;
calculating the Riemann tensor from the metric is an essentially nonlinear operation.
On the other hand, classical distribution theory as founded by L. Schwartz in his
famous book [6] is a linear theory. This deep conceptual
problem obviously is the main reason why applications of distribution theory to general
relativity have been rare and either limited to special situations or lacking
the necessary mathematical rigor. Moreover, Geroch and Traschen ([3])
have shown that a physically sensible and mathematically sound framework based on linear
distribution theory cannot handle such interesting sources of the gravitational field
as strings and point particles.
At this point the theory of algebras of generalized functions
enters the field. It provides a rigorous mathematical framework for
simultaneously treating singular (i.e., distributional) objects, nonlinear operations and differentiation. Moreover recently the
question of diffeomorphism invariance of the constructionwhich is vital in the contex of General Relativity has been
finally setteled. Hence algebras of generalized functions
provide a suitable setting for describing
general relativistic spacetimes of low differentiability; indeed there has been
a strong line of research following this approach.
For a more detailed discussion of ''singularities as internal points'' and distributional curvature in this context see this page by Jonathan Wilson.
For a recent ovierview of applications of generalized functions in General Relativity see the review article by James Vickers [7].
Bibliography

 [1]
 C. J. S. Clarke, ``The Analysis of SpaceTime Singularities,''
Cambridge Lecture Notes in Physics, (Cambridge, 1993).
 [2]
 C. J. S. Clarke, ``Singularities: boundaries or internal points,'' in
Singularities, Black Holes and Cosmic Censorship edited by P. S.
Joshi, 2432 (IUCCA, Bombay, 1996).
 [3]
 R. Geroch, ``Limits of spacetimes,'' Comm. Math. Phys. 13,
180193, (1969).
 [4]
 S. W. Hawking, G. F. R. Ellis, ``The Large Scale Structure of SpaceTime,''
(Cambridge University Press, Cambridge, 1973).
 [5]
 C. J. Isham, ``Some Quantum field theory aspects of the superspace
quantization of general relativity,''
Proc. R. Soc. A 351, 209232 (1976).
 [6]
 L. Schwartz, ``Théorie des Distributions,''
(Herman, Paris, 1966).
 [7]
 J.A. Vickers, ``Nonlinear generalised functions in
general relativity,'' in Nonlinear Theory of Generalized Functions,
Chapman & Hall/CRC Research Notes in Mathematics 401, 275290,
eds. M. Grosser, G. Hörmann, M. Kunzinger,
M. Oberguggenberger (Chapman & Hall/CRC, Boca Raton 1999).