It's incredibly valuable to think of the system as a mixture, and work on both components of the problem in parallel. (Both increasing precision to narrow the width of the gaussian, and removing the defective interaction that causes the spikes.)