ATK version history/changelog

Versions: 2008.10 | 2008.02 | 2.3 | 2.2 | 2.1 | 2.0

New in 2008.10

• Substantial performance improvements (2-5 times faster)
  • Intel Math Kernel Library used for matrix operations
  • ATK now threads on multi-cores
  • Real matrices used for Gamma-point calculations
  • Algorithm improvements
• Significantly reduced memory consumption (up to 50%)
• Time required to restore a NetCDF file reduced dramatically (fixing the longer time encountered in 2008.02)
• The two-probe algorithm is now read properly from the NetCDF file

New in 2008.02

• Improved DFT accuracy by
  
  • fine-grained identification of interacting atom pairs
  • new default value of the radial sampling distance for basis functions (0.001 Bohr); also gives improved performance of relaxations through smoother total energy curves
• Improved performance through new convergence criterion based on the total energy
  • Fewer SCF steps to convergence, can mean a factor 2 in calculation time
  • New default tolerance for convergence (1e-5)
• Improved support for VNL files
  • All computational results can be stored in VNL files using a user-defined label
  • The function VNLFile.addToSample() now overwrites existing data (if any) on the VNL file
• Two-probe relaxation by means of the equivalent bulk system enabled through the function calculateOptimizedAtomicGeometry()
• Upgrade from MPICH to MPICH2 for parallel execution
• NetCDF files only generated upon explicit request
• 3D grids removed from NetCDF files implying less I/O overhead and smaller files, at the expense of additional CPU time when restoring self-consistent calculations
• Memory demand of the function restoreSelfConsistentCalculation() has been reduced by a factor of two
• Fast linear response estimation of I-V curves
• Calculate collinear spin-torque and spin current (requires the MSMC license feature) for two-probes.

New in 2.3

• Transmission eigenchannels and eigenvalues
• Quasi-Newton and Steepest Descent algorithms for relaxations
• MPI message size problems solved (fixes infrequent but annoying crashes when running in parallel)
• LDOS is now non-negative
• It is now possible to restore GGA_revPBE
• New features to facilitate faster (and easier) convergence of two-probe calculations
• Two-probe relaxation by means of the equivalent bulk system
• The function calculateOptimizedAtomicGeometry() disabled for two-probe systems
• Formatted printing of NanoLanguage objects to stdout via the function nlPrint()
• Improved tips section of manual with guidance on how to converge two-probe systems

New in 2.2

• Interactive mode; ATK now behaves like IPython
• Several improved pseudopotentials
  • LDA: He, Na, Rb, Rh, Sb, Cs, Ba, Ir
  • GGA: He, K, Sc, Ti, Cr, Cu, Zn, Ga, As, Rb, Sr, Y, Zr, Nb, Tc, Ru, Rh, Pd, Ag, Cd, Sb, I, Cs, Ba, Hf, Ta, W, Re, Os, Ir, Pt, Au, Hg
• Performance improvements, in particular for transmission spectra and two-probe systems (for transmissions 2-3 times in performance, while it is 1.5-2 times for 3D two-probe systems)
  
  • transmission and DOS calculations separated
  • improved grid evaluation technique
  • algorithmic improvements in the exchange-correlation functionals
• Important bug fixes which remove jumps in the equation of state for bulk materials:
  • Corrected pair-finding algorithm
  • Corrected long-range behavior of basis functions
• Re-introduced support for initializing a new calculation based on the density matrix of a previously converged calculation (corresponding to the -i option in ATK 2.0 and TranSIESTA-C).
• Re-introduced support for calculating the MPSH spectrum (as in ATK 2.0)
• Fixed incompatibility between 32 and 64 bit platforms regarding VNL files
• New method: convertTo() for units
• The methods for obtaining the atomic configuration coordinates and forces of a two-probe system now only returns the atoms in the central region
• Introduced a check prohibiting the SZ basis set to be used with closed-shell systems (noble gases), as this leads to failed calculations
• Improved checking of the allowed range, type, etc of some input parameters to avoid unnecessary error messages and to introduce some other, proper, error messages
• Substantially improved and extended manual (both format and contents)
• General improvement of the MPI code, and upgraded MPICH version to 1.2.7p1 on all platforms
• Modernized FLEXlm component
  • FLEXlm binaries now shipped with the packages

New in 2.1

• Entirely new scripting interface (NanoLanguage), based on Python
  • Users can add their own functionality
  • Automated loops for convergence checks
  • etc etc
• Possibility to set the initial spin on each atom individually (allows to calculate spin-dependent magneto-tunneling junctions with electrodes polarized in the opposite directions)
• Constrain atoms now works properly
• The self-consistent loop now also takes dEbs into account when using the "Strict" criterion
• The convergence information in the self-consistent loop now displays useful information also with very strict convergence criteria (below 1e-6)
• Command line option -i not supported any more

New in 2.0, compared to TranSIESTA 1.3.0.4

• Support for spin-polarized calculations
• Support for heterogeneous two-probe systems (where the electrodes are physically different)