The Background Story
My work on unlabeled distance geometry originates from the period where I was a postdoc at Brown University. At the time, I was working on using invariants to represent objects in order to be able to recognize them using fewer computations than by comparing them directly. I was inspired by discussions with my colleague Senem Velipasalar, who was using statistics of features to represent objects. Thus I became interested in using statistics of invariants to represent objects and began to wonder in what circumstances statistics of invariants could separate the orbits of a group. In other words, under what circumstances could statistics of invariants be a lossless representation of the object modulo the corresponding group action.
I had already done work on the problem of recognizing curves and polygons modulo a group action. To take the next step, I was interested in going from ordered sets of points on a curve to unordered sets of points. I was especially interested in one particular question: is the distribution of the pairwise distances between a set of points in R^n (e.g., in the plane) a lossless representation of the set of points modulo rotation, translation, reflection, and scaling. I discussed this question with various people, including my postdoctoral advisor David Mumford, who told me that the answer was no. Still, I could not shake that question out of my mind, and I continued to try to “make it be true,” somehow. It was in London, Ontario at a conference that I was introduced to Gregor Kemper, who showed interest in my question and brought up the algebra knowledge needed to properly answer it. Thus began a very fruitful collaboration.