Fig. 4.

Time history of the sum of the squared residuals [summed across all data points, see numerator of (5)] is shown for a representative model output (thumbnail x-coordinate) for the 36-D test case 6. Each trace corresponds to the performance of a single Markov chain. The burn-in iteration is represented by the vertical line at iteration 56 400. The six-chain subset that located the true posterior distribution had values on the order of 2E–4m². Four other chains had larger residuals ranging from 4E–2 to 8E–2m², suggesting that they had found local minima.