Computational electrodynamics.
We need to design better geometries for reactors so that we can channel the plasma flow to better improve containment while quickly removing wastes from the highly compressed energetic region where fusion mostly occurs. If we could do that, then that mostly solves the issues from reactor degradation and significantly improves efficiency.
The trouble is that at such high energy levels the plasma flow is relatively chaotic so high fidelity models are required to predict the behavior accurately.