1: \begin{abstract}

2: We consider the inexact Newton methods

3: $$

4: x_{n+1}^\d=x_n^\d-g_{\a_n}\left(F'(x_n^\d)^* F'(x_n^\d)\right) F'(x_n^\d)^* \left(F(x_n^\d)-y^\d\right)

5: $$

6: for solving nonlinear ill-posed inverse problems $F(x)=y$ using the only available noise data

7: $y^\d$ satisfying $\|y^\d-y\|\le \d$ with a given small noise level $\d>0$. We terminate the iteration

8: by the discrepancy principle

9: $$

10: \|F(x_{n_\d}^\d)-y^\d\|\le \tau \d<\|F(x_n^\d)-y^\d\|, \qquad 0\le n<n_\d

11: $$

12: with a given number $\tau>1$. Under certain conditions on $\{\a_n\}$ and $F$, we prove for a large class

13: of spectral filter functions $\{g_\a\}$ the convergence of $x_{n_\d}^\d$ to a true solution as

14: $\d\rightarrow 0$. Moreover, we derive the order optimal rates of convergence

15: when certain H\"{o}lder source conditions hold. Numerical examples are given to test the

16: theoretical results.

17: \end{abstract}

18: