The Improved Ep-TL-Lp Diagram and a Robust Regression Method

The accuracy and reliability of gamma-ray bursts (GRBs) as distance indicators are strongly restricted by their systematic errors which are larger than statistical errors. These systematic errors might come from either intrinsic variations of GRBs, or systematic errors in observations. In this paper, we consider the possible origins of systematic errors in the following observables, (i) the spectral peak energies (Ep) estimated by Cut-off power law (CPL) function, (ii) the peak luminosities (Lp) estimated by 1 second in observer time. Removing or correcting them, we reveal the true intrinsic variation of the Ep-TL-Lp relation of GRBs. Here TL is the third parameter of GRBs defined as TL ~ Eiso / Lp. Not only the time resolution of Lp is converted from observer time to GRB rest frame time, the time resolution with the largest likelihood is sought for. After removing obvious origin of systematic errors in observation mentioned above, there seems to be still remain some outliers. For this reason, we take account another origin of the systematic error as below, (iii) the contamination of short GRBs or other populations. To estimate the best fit parameters of the Ep-TL-Lp relations from data including outliers, we develop a new method which combine robust regression and an outlier identification technique. Using our new method for 18 GRBs with {\sigma}Ep/Ep < 0.1, we detect 6 outliers and find the Ep-TL-Lp relation become the tightest around 3 second.