[CPMD-list] Convergence is still too bad !!

[Du Aijun]

Dear all, 

When I performed a geomertry optimization on methane(CH4) molecule using 
Vanderbilt ultrasoft pseudopotential and PCG MINIMIZE, I found the maximum 
gradient of wavefunction grows bigger at first serveral steps, but keep 
constant later. The problems are as following:

LINE SEARCH : LAMBDA=.281           ENERGY =          -7.9000653
   1  2.752E-02   4.714E-03      -7.897058   -7.897E+00     25.17
 LINE SEARCH : LAMBDA=.422           ENERGY =          -7.9589492
   2  2.887E-02   4.662E-03      -7.900065   -3.008E-03     25.24
 LINE SEARCH : LAMBDA=.107           ENERGY =          -7.9773182
   3  6.313E-02   3.395E-03      -7.958926   -5.886E-02     25.22
 LINE SEARCH : LAMBDA=.176E-02       ENERGY =          -7.9760463
   4  9.486E-02   2.432E-03      -7.977438   -1.851E-02     25.18
 LINE SEARCH : LAMBDA=.330E-03       ENERGY =          -7.9757286
   5  9.847E-02   2.356E-03      -7.976233    1.205E-03     25.21
 LINE SEARCH : LAMBDA=.618E-04       ENERGY =          -7.9755948
   6  9.922E-02   2.343E-03      -7.975730    5.031E-04     25.21
 LINE SEARCH : LAMBDA=.116E-04       ENERGY =          -7.9755623
   7  9.937E-02   2.340E-03      -7.975595    1.354E-04     25.14
 LINE SEARCH : LAMBDA=.217E-05       ENERGY =          -7.9755550
   8  9.940E-02   2.339E-03      -7.975562    3.244E-05     25.11
 LINE SEARCH : LAMBDA=.407E-06       ENERGY =          -7.9755533
   9  9.941E-02   2.339E-03      -7.975555    7.374E-06     25.26
 LINE SEARCH : LAMBDA=.764E-07       ENERGY =          -7.9755530
  10  9.941E-02   2.339E-03      -7.975553    1.623E-06     25.10
.................................................................
.................................................................
 997  9.941E-02   2.339E-03      -7.975552   -8.882E-16     25.23
 LINE SEARCH : LAMBDA=.525-157       ENERGY =          -7.9755518
 998  9.941E-02   2.339E-03      -7.975552    8.882E-16     25.21
 LINE SEARCH : LAMBDA=.262-157       ENERGY =          -7.9755518
 999  9.941E-02   2.339E-03      -7.975552   -8.882E-16     25.15
 LINE SEARCH : LAMBDA=.590-157       ENERGY =          -7.9755518
1000  9.941E-02   2.339E-03      -7.975552    8.882E-16     25.17
 ================================================================
 =              END OF GEOMETRY OPTIMIZATION                    =
 ================================================================

The following is my first time input file:

&CPMD
  OPTIMIZE GEOMETRY
  HESSIAN UNIT
  BFGS
  PCG MINIMIZE
  CONVERGENCE
  1.0D-4 5.0D-4
  CENTER MOLECULE
  MAXSTEP
  1000
&END


&DFT
  NEWCODE
  FUNCTIONAL PBE
  GC_CUTOFF
  0.2E-07
&END

&SYSTEM
  ANGSTROM
  pOSSION SOLVER TUCKERMAN
  SYMMETRY
  0
  CELL
  13.0 1.0  1.0  0  0  0
  CUTOFF
  30.0
  STATES
  5
  CHARGE
  0.0
  TESR
  2
&END
&ATOMS
*006-C-gpbe--bm.uspp NEWF BINARY
 LMAX=P
   1
 -4.3985  1.5414  0.0000
*001-H-gpbe--bm.uspp BINARY NEWF
 LMAX=S
   4
 -3.3985  1.5414  0.0000
 -4.7323  2.4840  0.0000
 -4.7323  1.0701  0.8164
 -4.7323  1.0701 -0.8164
&END

Then I made some changes on my input file according to the suggestion from 
others:

Adding INITIALIZE WAVEFUNCTION RANDOM in the section of $CPMD..... $END!
Increasing GC-CUTOFF to 5.D-5. But problem still exists and the only 
difference is that the gradient become a new constant!

I also changed the Convergence standard to 1.D-3 and 1.D-3 to pass several 
Geometry steps first.Then I restart the calculation with 1.D-5 and 5.D-5. 
But the same Problems still met! In addition, bad convergence were also 
observed when using another Vanderbilt potential(pw91 not PBE). All the 
uspps inputfile are from inputfile 
http://www.physics.rutgers.edu/~dhv/uspp/uspp-cur/Work/ 

When I use other pseudopotential in the PP_LIBRARY provide by www.cpmd.org, 
such as BLYP. It is very fast to reach convergence.

Could anyone help me solve this problem? Any comments and suggestions would 
be highly appreciated!

With regards,
A.J.Du
  


_________________________________________________________________
Ãâ·ÑÏÂÔØ MSN Explorer:   http://explorer.msn.com/lccn/