Finite Density (Spheres) Integration and Collision Detection

Hey all! I've been wracking my brain over this and have been running into issues. I'm modelling a planar 3-body system with a few restrictions such as:

1. All bodies are spheres (radius and density can vary between them). 2. The spheres all start in contact with each in a triangular formation. 3. The spheres configuration is rotated about the center of mass until the centripetal accelerations overcome the gravitational orbits and the spheres enter into an orbit.

I might simplify this to having two spheres 'glued' together.

Example (uniform radius and density of all spheres)

http://3.bp.blogspot.com/_7Vt66UxO7os/S6kmmy1ORlI/AAAAAAAAAK0/iSxys4-1fgg/s1600/Circle%20packing%203.jpg

As the 3 spheres orbit each other, there is a minimum distance each body can approach, the sum of their radii. I'm having trouble enforcing this constraint in ODE 45. I'm thinking of using a classical impact model.

However for the cases where the spheres all start in contact, a collision model isn't really applicable (I think) and I'm wondering if there is an easier way to do this.

Perhaps by writing the equations of motion in the rotating frame rather than the inertial frame?