Prediction of the Tensor-Form Refractive Index in a Cyclic Olefin Copolymer: A Multiscale Computer Simulation Approach

We developed an efficient approach using an equivariant graph neural network (EGNN) to predict the α tensor of the copolymer, thus enabling accurate calculation of the tensor-form refractive index (n) directly from snapshots generated by tailored-potential molecular dynamics (MD) simulations, with <3% error for ρ and <1% error for scalar form n versus experiment. The deviatoric components of n and the orientation tensor S exhibit a robust linear relationship governed by two independent parameters; the temperature derivative of the deviatoric n is more sensitive to orientation state than to chain sequence, while the trace part depends primarily on monomer structure. We demonstrate that monomer geometric sphericity (⊖) serves as a universal descriptor for the upper limit of intrinsic birefringence (β) in cyclic olefin copolymers (COCs), with R2 = 0.82 across 1684 candidate monomers.