edsparc% python
Python 2.0 (#3, Sep  3 2001, 11:50:37) [C] on sunos5
Type "copyright", "credits" or "license" for more information.
>>> import endyne
>>> endyne.version()
(5, 'A', 3, 1, 6, 'generic', 'Tue Jul 30 12:52:28 EDT 2002')
>>> 

Python is also embedded in the endyne executable, so that you can run endyne with a .inp file as argument to make a calculation as in version 2:

edsparc% endyne test.inp
ENDyne version 5.A.3.1 test.inp

                             *****************
                             *  E N D y n e  *
                             *****************

                   Executable version  5 alpha development phase  3
                           Restart file format  6

Or you can run endyne with a .rsl file as argument to get the functionality of evolve:

edsparc% endyne test.rsl
ENDyne version 5.A.3.1 test.rsl
RSL file test.rsl opened with the command:
    rslfile = RSLopen('test.rsl','r')
Recorded state log (RSL) file
 name:    test.rsl
 mode:    r
 format:   6
 version:      5.0603
 begin:    0.0E+0 1
 end:      0.1 29
 current:  0.0E+0 -1
 records:  11
 flags:   ownself       T
          seekdone      T
          readonly      T
          flushbuf      F
          isopen        T
          rstisopen     F

endyne>>> 

All the functionality of ENDyne can be accessed from Python and many of the new features are only available through the Python interface. Documentation of any command is available with the standard Python method:

endyne>>> print thing.__doc__

The use of the parameter RCNDLM is much more sensitive in the new algorithm. Therefore users who have set this parameter, should remove it from their input files to ensure the use of the default. A new parameter RCNDAC has been introduced, giving the relative accuracy of the condition number returned by LAPACK routines. The version 2 algorithm considered a new state bad if

new_condition_number > RCNDLM * NDIM * old_condition_number
independent of how close in time the previous condition number was obtained. The version 5 algorithm uses the inequality
d/dt ln condition_number > RCNDLM * NDIM
within the accuracy RCNDAC. Setting RCNDAC=1. and RCNDLM large ensures that the test is never satisfied. The resulting trajectory may then inaccurate and eventually wrong.

The use of the much more stringent backstep algorithm implemented in version 5 is essential for wave functions with free electrons, as the numerical integrator can produce nonsense much more easily than for wave functions with only atomic centers.

The use of coordinates for the center of free moving centers of basis functions introduces a duplication with the basis functions exapansion coefficients. To obtain a singularity free coordinate system for the phase space it is essential to perform a transformation. A new way to solve the dynamical equations has been implemented which constructs the complete symplectic form. The TDVP equations were solved by substitution in version 2, to save the space for the full symplectic form. However, the transformation is only meaningful on the entire symplectic form. The solution method with the full symplectic form is used when LUSPFM=T, and the singularity removing transformation is applied when LUSPTR=T. The Hamiltonian equation is solved with a general linear eqaution solver by default; with LUSPVD=T the equation is solved by singular value decomposition.

By turning on tracing LTRACE=T and LTRSPD=T detailed information about the solution of the equations is printed to the log file and, with LUSPVD=T, the singular values and right singular vectors are written to the RSL file in TRVEC1 and TRMAT1, respectively. The components of the singular vectors are in the order of the internal representation of the symplectic equation: real part of the Thouless coefficients, center positions, imaginary part of the Thouless coefficients, center momenta. TRVEC2 and TRVEC3 contain the velocities as computed by the substitution method (version 2) and by solving the symplectic form (version 5, LUSPFM=T). TRVEC4 holds the solution with the symplectic transformation LUSPTR=T. The velocity components are in the following order: Thouless coefficients, center positions, center momenta. TRVEC5 contains the force of the TDVP equation. The force components are in the same order as the velocities.

There is an extra singularity that can develop: the free center can get zero coefficients for all its basis functions. The algorithm to avoid this singularity in the nuclear metric uses an expanded state phase space. This feature can be turned on by setting LUSEXS=T.

Rules for choosing and constructing the basis set for free centers are still being developed. So far the only guideline is that there should be only basis functions of even angular momentum: s,d,g,j,… and not p,f,h,….