# Is burn-in necessary for MCMC/Gibbs sampling if I have samples from the true distribution already?

Say I have some samples from a distribution $p$, and I want to get more samples using MCMC/Gibbs sampling.

Since the existing samples are known from the equilibrium distribution $p$, if I use them as starting points of Markov chain sampling, it seems the burn-in period will be no longer needed, is that right?

Thanks.

• That is right, there is no need of burn-in then. All you then need to figure out is how many samples you need. – Greenparker Apr 3 '18 at 10:42