Model-mismatch errors in least-squares 1-D centering.

The nature of "pixelation errors" in 1-D centering by means of least-squares fits is analyzed in detail. For the case of intrinsically symmetric data we propose an improved position estimator, based on a combination of the fit results obtained with an even and an odd number of fitpoints. Numerical tests on synthetic data confirm that the combined estimator reduces the pixelation errors - which might be called more aptly model mismatch errors - by roughly an order of magnitude. The influence of weak asymmetries in a feature is discussed and we indicate how, in practice, the importance of model mismatch errors (and the relevance of a correction thereof) with respect to noise-induced errors, can be assessed.