The precision of the forces was increased from 4 to 7-8 significant digits. In the previous version, 1st order finite differences were used to calculate certain components of the forces, in the new version, central differences are used in all places (please do not panic, the precision was sufficient for all calculations we have done in the past, but ``just in case'' ...).
A new flag was introduce to control the behavior of the mixer. It is called MAXMIX, and specifies the maximum number of iterations stored in the mixer. If MAXMIX is set to a positive value (for example 40) the mixer is not reset after an update of the ions. This can reduce the number of electronic steps during MD's and ionic relaxations. Please read section 7.40 to find more information. In addition in vasp.4.4, an improved charge density prediction (based on a quadratic extrapolation of the bond charge) was implemented by Dario Alfe which also reduces the number of iterations during MD simulations.
The RMM-DIIS algorithm has been rewritten to run in a blocked mode (i.e. several bands are optimized at the same time). This allows to use matrix-matrix operations instead of matrix-vector operation for the evaluations of the non local projection operators in real space, and might speed up calculations on some machines (see section 7.39). The start up phase of the RMM-DIIS algorithm was also rewritten. In the new version eight non selfconsistent steps are performed, and in each step each band is optimized one using a steepest descent algorithm. The new version is significantly more reliable.
A new real space projection scheme was implemented in vasp. It can reduce the computational requirements of the real space projection scheme by 10-30 . This new scheme is selected by specifying LREAL=Auto or LREAL=A in the INCAR file (might not be available in the first release of vasp.4.4). Per default, the scheme also searches automatically for an optimized real space cutoff and makes the specification of the flags ROPT unnecessary (see section 7.34). (If LREAL=A is used, the ROPT line must be removed from the INCAR file to activate the automatic search).
Vasp.4.4 is also the first version that supports officially the PAW method. However data sets will not be released before the end of the year 1998 (except to selected ``long time'' vasp users, coauthorship on the first PAW paper is required). Vasp.4.4 contains also all files required for the parallel version and is thus the first official parallel version of vasp.
In Vasp.4.4 the spring constant was redefined in the nudged elastic band method. In the old version the spring constant had to be halved when the number of images was doubled. The default for the spring constant is now SPRING=-5. This is a very sensible number in most cases.
Vasp.4.4 now allows to perform (real) damped molecular dynamics, by setting the SMASS tag in the INCAR file. Although this feature was documented before (see also Sec. 7.19), it was not properly working in the previous releases. For reasons of consistency the time step (POTIM) has been redefined for damped molecular dynamics. The POTIM parameter in the INCAR files should be changed to
POTIM = old POTIM / 2
Now, for IBRION=1, the number of degrees of freedom (NFREE) can be specified allowing a better control of IBRION=1 (see Sec. 7.19).
Up to now, for MD's the temperature was defined as In fact this definition is not quite correct, and one should have used (translational invariance of the Hamiltonian). As pointed out in the VASP guide, this means that all calculations were effectivly done at too high temperatures. The correct (new) definition is used starting from vasp.4.4.3. To obtain the old behaviour one has to set
TEBEG = TEBEG(old) * NIONS / (NIONS-1)
Selective dynamics are now correctly supported even for MD's.