1: \begin{abstract}

2: In the numerical renormalization group (NRG) calculations of spectral

3: functions of quantum impurity models, the results are always affected

4: by discretization and truncation errors. The discretization errors can

5: be alleviated by averaging over different discretization meshes

6: (``z-averaging''), but since each partial calculation is performed for

7: a finite discrete system, there are always some residual

8: discretization and finite-size errors. The truncation errors affect

9: the energies of the states and result in the displacement of the delta

10: peak spectral contributions from their correct positions. The two

11: types of errors are interrelated: for coarser discretization, the

12: discretization errors increase, but the truncation errors decrease

13: since the separation of energy scales is enhanced. In this work, it is

14: shown that by calculating a series of spectral functions for a range

15: of the total number of states kept in the NRG truncation, it is

16: possible to estimate the errors and determine the error-bars for

17: spectral functions, which is important when making accurate comparison

18: to the results obtained by other methods and for determining the

19: errors in the extracted quantities (such as peak positions, heights,

20: and widths). The closely related problem of spectral broadening is

21: also discussed: it is shown that the overbroadening contorts the

22: results without, surprisingly, reducing the variance of the curves. It

23: is thus important to determine the results in the limit of zero

24: broadening. The method is applied to determine the error bounds for

25: the Kondo peak splitting in external magnetic field. For moderately

26: large fields, the results are consistent with the Bethe Ansatz study

27: by Moore and Wen. We also discuss the regime of large $U/\Gamma$

28: ratio. It is shown that in the high-field limit, a spectral step is

29: observed in the spectrum precisely at the Zeeman frequency until the

30: field becomes so large that the step merges with the atomic spectral

31: peak.

32: \end{abstract}

33: