what was the starting position like?
what type of no penetration contact are you using? how have you told it to handle gaps?
what does the mesh look like? if it is too coarse, the contact may not behave well
Thanks Jared. I was running the analysis with a coarse mesh just to check convergence and overall deflection, but the separating contact faces wasn't making sense. Decreasing the element size seems to keep the faces touching in the deflected model. The initial tangent contact likely requires more elements to behave correctly.