Metropolis Algorithm Þ "Slow" versus "Efficient" Þ Avoid greed: If (P1 > P0) always go If (P1 < P0) go only P1/P0 of the time Þ Is the search "too hot" or "too cold" ? Þ Use local covariance matrix i. e. parameter space topology to "mold" step probability matrix Þ Convergence criterion Þ What is this mysterious NSAMP parameter !!?? Þ The meaning of the confidence intervals (all random error) GIM2D Logfiles Þ Every iteration is stored! Þ Two lines are stored per iteration: line 1 shows parameter values and line 2 shows Metropolis temperatures Þ Each line starts with a letter code: F - ICF iteration I - Best ICF model R - Iteration rejected by Metropolis algorithm A - Iteration accepted by Metropolis algorithm before convergence E - Iteration accepted by Metropolis algorithm after convergence C - Rows of final covariance matrix P - Best-fit parameter, 99% lower bound, 99% upper bound Half-Light Radius Þ Computed by integrating best-fit bulge+disk profile out to infinity Þ This is a semi-major axis radius NOT a geometric mean
| Previous slide | Next slide | Back to the index | View Graphic Version |
No notes for slide 6