Large diffeomorphisms in 2+1 quantum gravity on the torus

The issue of how to deal with the modular transformations, large diffeomorphisms, in 2+1 quantum gravity on the torus is discussed. I study the Chern-Simons or connection representation and show that the behavior of the modular transformations on the reduced configuration space is so bad that it is possible to rule out all finite dimensional unitary representations of the modular group on the Hilbert space of $L^2$ functions on the reduced configuration space. Furthermore, by assuming piecewise continuity for a dense subset of the vectors in any Hilbert space based on the space of complex valued functions on the reduced configuration space, it is shown that finite dimensional representations are excluded no matter what inner product we define in this vector space. A brief discussion of the loop and ADM representations is also included.