This talk examines intrinsic fluctuations in chemical reactions in a large volume, using the Schlögl model as an example with two stable states. The deterministic rate equation works only for large, well-stirred systems, does not detect intrinsic fluctuations, and does not allow switches between stable equilibria. The talk therefore starts from the discrete one-step Markov jump process and its Master equation, and studies the large volume regime. Expanding the Master equation in the small parameter leads to the Fokker-Planck equation, which is valid only up to moderate deviations. A large deviation (rare event) ansatz for the exponent instead leads to a Hamilton-Jacobi equation with a classical mechanics interpretation, in which rare events correspond to heteroclinic connections. The talk concludes that stochastic differential or Fokker-Planck approximations are not always good enough for rare events in Markov jump processes, and that the relative stability of the two states can be determined from mean times of jumps between them, calculated by the correct expansion of the backward equations.