Running a single chain is heavily discouraged precisely because it makes the Rhat much less sensitive. Debugging big models ishard and it is usually sensible to start with a smaller model and only a subset of the data - hard to say anythingmorespecific without seeing your model and/or your data. Some additional pointers can be found in the “Bayesian workflow” preprint by Gelman et al. and at Divergent transitions - a primer
Best of luck with your model!