Part  ONE
When i relaxed the structure of TiO2 using IBRION = 2 and  EDIFFG = -1E-3 , some errors as following appears at the end of some ionic steps in the log file:
" ZBRENT: increasing intervall"
" ZBRENT: bracketing found"
" ZBRENT: interpolating"
" ZBRENT: can't locate minimum, use default step"
Then i googled the message and found the following helpful answer
(http://cms.mpi.univie.ac.at/vasp-forum/forum_viewtopic.php?3.1856) :

" ZBRENT is an algorithm search for a root of a function by Brent's method: Numerical recipies, Section 9.3
The problem in your case might be that the Conjugate Gradient algorithm (IBRION=2) is not suitable for very small corrections of the atomic positions if your system has almost reached equilibrium (please have a look at XDATCAR to check the size of the relaxation steps done before the ZBRENT warnings show up). Usually it is sufficient to converge a system up to maximum remaining forces of about 0.01eV/A (EDIFFG=-0.01).
please try one of the following:
1) choose a different algorithm for ionic optimization (IBRION=1:quasi-Newton Method)
2) set ADDGRID=.True. in INCAR (only for vasp releases 4.4.5 and newer) "

yes.it's right!
                                                              Part     TWO
After I modifying as above, then another message appeared:
    N       E                     dE             d eps       ncg     rms          rms(c)
DAV:   1    -0.494411541193E+02   -0.49441E+02   -0.10469E-06  1950   0.105E-02    0.126E-03
DAV:   2    -0.494411538954E+02    0.22387E-06   -0.23223E-07  1950   0.601E-03
  32 F= -.49441154E+02 E0= -.49441154E+02  d E =0.754070E-06
 BRION: g(F)=  0.000E+00 g(S)=  0.204E-08 BRIONS problems: POTIM should be increased
 retain N=  1 mean eig=*****
 eig: *******

the related explanation and  solution is:

the asterisks *** simply indicate that the number which has to be written (mean eig) does not fit into the format of the write statement, which is f5.2 in that case (i.e., just 2 digits before the comma).
Please note the "BRIONS problems: POTIM should be increased",
which indicates that your ionic relaxation might get stuck. -->
as the total energies are not converged yet, please follow that instruction and increase POTIM in your INCAR file (default=0.5)
or choose a different relaxation algorithm (IBRION)