This document summarizes the changes between successive versions of the ADF package (including BAND) and of the utility programs used in combination with the main programs.
The Chapters are in reverse time order: the most recent ones come first.
This document focuses on reporting error-repairs and changes in functionality or performance. Technical changes in the code may not be mentioned if they
[Sigma] Have only very small effects on numerical results and/or the program's performance,
[Sigma] Solve bugs that let the program crash in certain circumstances, but which do not affect the outcomes in calculations that terminate normally.
Explanations of new keywords, options, etc. are brief: for more details see the User manuals. The same holds for installation stuff. Generally: read the documentation! This write-up is intended to draw your attention to important changes, not to explain every detail.
17 July, 2000
ADF2000.02 is a very minor upgrade of ADF2000.01, primarily to include versions for more platforms.
Versions for the following platforms have been added:
[Sigma] Compaq Alpha running Linux
[Sigma] NEC SX5 (should also run on SX4, which we have not yet verified)
[Sigma] SGI running IRIX64 6.4
A complete list of currently supported platforms is available on our web site.
On some systems the program may find that the license file is `CORRUPT'. The Installation manual discusses the possible causes. A possible reason that was recently found is that the line breaks between the different records in the license file may contain (invisible) <CR> (carriage-return) characters. These can be removed by the (new) utility fix_license, which will be put into your $ADFBIN directory by the installation procedure. See the Installation manual for details.
The PW91 correlation implementation contained an error that may have affected some calculations using the PW91 functional, in particular for calculations with BAND.
17 May, 2000
ADF2000.01 is an upgrade of ADF1999.02. It cures a few problems in ADF1999 and new functionality has been added. The fact that the new release is coined "2000" does not signal a major difference with respect to the ADF1999.x series. We simply name the versions after the year of release.
The PVM "ARCH" on SGI systems to use with the ADF package has been changed from "SGI64" to "SGI6". Please note carefully, that this is different from the usual (default) settings that may be present on your system (often: SGI64 or SGIMP64). This implies that you may need to install another PVM version.
At the time of writing we have prepared only executables of ADF2000.01 for a few platforms. Others will be added successively. At all times, the web site provides a survey of currently available versions.
We have extended the Examples document to include more sample runs. The new functionality QMMM (in ADF) is described in a completely separate (and new) QMMM manual. The corresponding description in the ADF User's Guide is brief and mainly points to the QMMM manual.
The organization of sample runs, in the $ADFHOME/examples/ directory has been changed. Each sample run has now its own (sub) directory, containing the run script and the output file.
A few new basis sets have been added for Cm with 5d core, in the database directories II/ and IV/
The QM/MM (Quantum Mechanics / Molecular Mechanics) functionality is available to treat large systems by handling a `core' part by the first-principles and accurate QM approach, while the remainder is treated by a force-field based approximation (MM). QMMM cannot yet be used in combination with a Frequencies run. In a Linear Transit run with QMMM, only the QM atoms may define the LT parameters of the path to scan.
NMR chemical shifts and the advanced ZORA relativistic methodology can now be combined. ZORA-NMR may give a marked improvement over the traditional Pauli relativistic NMR results.
BAND has acquired a new Fragments feature to analyze results, in particular Density-of-States data, in terms of fragment orbitals rather than elementary basis functions.
No new utility programs yet. Some of them are in development and parts (or all) of their code may already be contained in the package, but without documentation and without support. We're working on them and will inform you when they are ready for usage.
In some cases we found that the accuracy of the ADF numerical integration grid fell somewhat below expectation, sometimes resulting in a warning or error abort "INACCURATE ETA INTEGRALS" or "BAD ETA INTEGRALS". To solve this, a few technical parameters in the grid generator have been adjusted. For a given accuracy parameter (the input key INTEGRATION) this will produce more points (roughly 10%) than was the case in ADF1999, with a corresponding increase in computing time but yielding more stability, robustness and reliability.
ADF calculations with large numbers of point charges - most of them far away from the real molecule - were very slow, due to a very large number of (mostly redundant) numerical integration grid points. This has been cleaned up, resulting in a significant gain in speed for such calculations.
In the SFO / MO analysis output section a list of MOs has been inserted with a population breakdown in terms of the SFOs.
The Compaq release of ADF1999 failed on any substantial calculation, due to an overflow of the stack. Although this was not due to an error in the program, being a Compaq specific problem, we have solved it by reprogramming the involved routines.
The COSMO implementation in ADF1999 contained several major errors. Usually they `only' caused the program to crash. Occasionally, the problems affected also the numerical results in a calculation that terminated normally. All problems in this functionality that we have encountered have been repaired; all related errors that were reported to us now run correctly.
IRC calculations often failed to proceed correctly, either by quickly going astray or by convergence problems with the point-by-point geometry optimizations. Another error that could show up was a sudden swap of coordinates, resulting of course in an energy jump and failure to produce anything meaningful.
Volume maps generated with Cerius2 were (almost) a factor 2 wrong in size. (The program mixed up bohr and angstrom units of length). The error was, in fact, not in the Cerius2 software but in the densf utility program in ADF, so to repair it you have to download the new version of densf.
ESR calculations with LDA for the XC functional failed to compute the A-tensor (hyperfine splitting) in a scalar-relativistic spin-unrestricted calculation.
There were a few deviations from the correct symmetry in NMR tensor results. Most of the discrepancies have been solved; the remaining deviations are a few orders of magnitude smaller than before and (probably) related to insignificant numerical issues.
A few minor inaccuracies and (small) numerical errors in relativistic NMR calculations have been corrected. The effects are believed to be small and often insignificant.
The list of atomic coordinates in Z-matrix format in the logfile could have incorrect atom numbering indices, depending on the order of atoms in the input file.
ZORA relativistic calculations sometimes failed with error abort "SORXR error"
The following error aborts have also been repaired:
"Error 4 Polyhe",
"no convergence bounds"
"TOO MANY TRIES. HULL32"
"inconsistent merge of edges"
"point edge. HULL32"
We have treated the error abort "BAD ETA INTEGRALS" and the related warning "INACCURATE ETA INTEGRALS" by tuning the internal precision control in the numerical grid generator. Most of the cases should be cured. Please report any persisting ones.
Relativistic calculations involving ghost atoms in the molecule failed, for instance in a Basis Set Superposition Error calculation.
There was an error in the computation of IR intensities for certain point group symmetries. This resulted in the following message in the logfile: "WARNING: INFRARED INTENSITIES WRONG"
A computation with an Icosahedral system with an atom in the symmetry point
produced garbage results, due to an error in the numerical integration grid
around the central atom.
Note: ADF does not support I and I(h) symmetry, but the system may itself have that symmetry (in which case you may run it in, for instance, D(5h) symmetry). The error did not depend on what subgroup symmetry you actually used, only on the fact that the true symmetry is Icosahedral and that it has an atom in the center.
In a spin-unrestricted calculation the printout of one-electron energies for the second spin were (often) wrong.
A particular numerical instability occasionally caused a further slowdown of SCF convergence - or even failure - in BAND for systems with very slow convergence anyway.
Computations on very large molecules sometimes failed with the message "TOO MANY POINTS. PLGOUT"
RESTARTDOS in BAND sometimes failed (due to non-initialized variables)
Due to a complicated numerical problem, resulting in loss of precision for certain integrals, calculations with very steep functions (typically, using ZORA basis sets) may fail on Cray systems with the error message "singular sfit matrix". We have analyzed and understood the cause and are working on a repair.
Cerius2 fails to display orbitals computed in a relativistic Spin-Orbit calculation. This is not really a bug: it has simply not been implemented (for good reasons) yet. We're considering to work on this for a future release.
Much more of the ADF sources have been made available (for those who have a license including source code). Released are now major parts of the Time-Dependent DFT, NMR, and Spin-Orbit routines, as well as scattered others.
A few data arrays related to the frozen cores in ADF were written to TAPE21 with one (redundant) element too many. This has been corrected. Consequently, ADF1999, which also reads the redundant array element will fail to read an ADF2000 TAPE21 file. It will abort upon an attempt to read this redundant element.
With the DEFINE key you can assign values (or functions) to identifiers and then use these identifiers rather than numerical data for, say atomic coordinates. Usage of one-character identifiers is not recommended and easily leads to unintended (and sometimes unnoticed) mistakes in the input file. To `protect' against such situations, `loud' warnings are issued in standard output as well as in the logfile as soon as you "DEFINE" one-character variables.
30 June, 1999
ADF1999.02 is a minor upgrade of ADF1999, primarily to solve a few installation problems and to extend the number of supported platforms. For several platforms already supported by ADF1999 no executables have been replaced. The source code at the web site corresponds to ADF1999.02
There is also new functionality: ESR properties. However, to use this you don't need to download ADF1999.02, as the functionality already exists within ADF1999. It is documented now in the ADF User's Guide. If your current ADF1999 license file denies you access to the ESR feature, contact us for an upgrade of the license file.
ADF1999.02 supports Compaq (DEC) platforms. Note carefully that default settings for the maximum stack size on your Compaq station may be insufficient to run ADF1999. See the README at the download page.
The ADF1999 version for the IBM RS6K has been replaced to cure a few installation problems. Contrary to the first release of ADF1999, we now have only one RS6K version, which should be usable for all users with Fortran compilers 3.2.5 or higher. Unfortunately we cannot support yet older versions of the Fortran compiler.
In addition, we have placed at the web site the MPI executable for the IBM SP. The script to properly run an MPI version is rather system dependent, and you will probably have to modify the standard run script ($ADFBIN/start) for the MPI version.
Description of TAPE21 contents in the User's Guide updated: was rather incomplete. Description of TAPE13 (checkpoint file) added.
ESR properties can be computed with ADF1999. See the ADF User's Guide. Keywords: ESR and QTENS.
18 May, 1999
ADF 1999 involves a significant enhancement of functionality to include solvent effects, excitation energies, (hyper) polarizabilities, NMR chemical shifts and the ZORA relativistic formalism, to mention a few major extensions. Technically, the program has been `upgraded' from the Fortran77 to the Fortran90 language. This implies that, in order to compile the program, you must have a Fortran90 compiler. However, we have also started to distribute complete executables, so the need to compile the program by your self is reduced.
The extended functionality, new language features and other technical matters have made it unavoidable that the format of the TAPE21 output file had to be changed. This implies that the TAPE21 files of ADF2.3 and ADF1999 are technically not compatible, i.e. you cannot hook up a version 2.3 TAPE21 file as a fragment file to a calculation with ADF1999. However, the logical content is largely identical. You may verify this by making an ASCII dump of the files, using the KF utilities. The logical compatibility means that privately developed tools that access TAPE21 files through the KF library procedures should work with the 1999 version as well (using the KF library of the 1999 release, of course), or may need only minor adjustments.
The 1999 release has focused on a significant increase in functionality compared to ADF2.3 and we believe this is a major step forward. However, we appreciate and acknowledge the fact that there is room for improvement in the area of speed, memory, IO, and disk usage and we will give this aspect more attention in the near future.
With version 2.3 we introduced the run-script "start". This was convenient, if not necessary for the control of parallel calculations. The script still exists but we have now `soft-linked' each of the program names (adf, band, densf...) to this particular script in $ADFBIN. This means that you can call "adf" and thereby in fact invoke the "start" script. The program name that you actually use in the command is passed to "start" as the main argument. As a result, usage is simpler now and has the form
$ADFBIN/adf -n4 <in > out
This replaces the (still available) 2.3 variety
$ADFBIN/start -n4 adf <in >out
See also the Installation manual and the Examples document. Take a look at the sample runs for the new format used in practice.
Technical note: since the names "adf", et cetera are now used as soft-links to the start script, the program binaries have acquired other names: adf.exe, band.exe, and so on, in $ADFBIN.
We are shifting the distribution from using ftp accounts (as we did up to release 2.3) to downloading from our web site: http://www.scm.com (starting with release 1999). All users will get password-protected access to the web site with their old ftp account data (user name and password). This will, we hope, make downloading easier for you.
The ftp facility remains in place, but only to upload files for error reports.
We have introduced a few more (UNIX) environment variables that should be defined when running programs of the ADF suite, in particular - but not only - when running in parallel. We recommend that you define these variables in your login script (.cshrc if you run C-shell). Consult the Installation manual for details.
Control of parallelization features at the first line of the input file has been disabled. All parallelization is now controlled by environment variables and by the -n flag to the program execution scripts. See the Installation manual.
Due to difficulties (i.e. compiler bugs) in several Fortran90 compilers as they were distributed by hardware vendors we had a hard time to create versions of ADF 1999 for all the platforms that we wish to support. Unfortunately, we have not sorted out all the problems yet. As a consequence, we are distributing only a few (major) platform types now and will follow up as soon as possible with the other ones. Please contact us (by email to: email@example.com) if the now-distributed-platform types cause any difficulties on your machine.
Support for IBM RS6K running AIX3 has to be discontinued, unfortunately, for lack of a suitable Fortran90 compiler.
New all-electron basis set files up to Kr (type I through V in the database directory)
Several polarization functions have been changed in type III, IV and V basis sets, after extensive tests with calculations on small molecules.
For S: the exponential decay factor of the d-type polarization function changed from 1.5 to 2.1.
For Si: d-function exponent changed from 1.35 to 1.75; f-function exponent changed from 2.0 to 1.9
For Cl: d-exponent changed from 1.65 to 2.0; f-exponent from 2.5 to 2.0
For Al: d-exponent changed from 1.2 to 1.6; f-exponent from 1.8 to 1.75
A few basis sets with (too) large frozen cores have been declared obsolescent. They have been moved to a new "Old" directory in the database. This applies to heavy elements (Hf ... Hg).
Some keywords for BAND were not documented in the BAND User's Guide. The list has been completed.
The database contains also files that are used for the Cerius2-ADF interface developed by SCM (based on the older version created by MSI for ADF2.x).
Response properties can now be calculated, based on an implementation of Time Dependent DFT (TDDFT). For closed shell systems you can compute frequency dependent (hyper) polarizabilities, Van der Waals coefficients (see also the utility program DISPER), excitation energies and Raman intensities.
NMR chemical shifts are computed with a new utility program NMR. See the Utilities manual for an input description.
Solvent effects (COSMO model) are available. See the ADF User's Guide for a description of input parameters (keyword SOLVATION).
The ZORA relativistic option is available in ADF. This is a major improvement over the (still available) Pauli approach. See the ADF User's Guide for details. ZORA requires special basis sets. The ZORA basis sets are, exactly as for the non-ZORA sets, organized in subdirectories I, II, III, IV and V for increasing flexibility. All ZORA basis sets are in the $ADFHOME/atomicdata/ZORA/ directory of the ADF database.
BOB: the Basic Output Browser has been included in the package, to enable a quick scan of the (potentially large) standard output file. See the Utilities documentation.
The SlaterDeterminants key has been introduced to enable the computation of multiplet states. See the Theory document for a discussion.
Atomic coordinates can be input (ADF) with MOPAC format. ASCII output files can be generated with coordinates in MOPAC and/or XYZ and/or MOL format.
The SCF Level-Shift option has a new sub-option to turn the level shifting off as soon as the SCF error drops below a user-set threshold. This may, occasionally, improve the convergence rate when the final stages of the SCF may only slow down by the level shifting technique.
The AUFBAU option to the OCCUPATIONS key has been disabled. The KEEPORBITALS parameters have somewhat changed. ElectronSmearing is suppressed in more cases and should be understood as only a trick to achieve SCF convergence in an intermediate calculation, for instance during a geometry optimization.
NMR: computes NMR chemical shifts, using a TAPE21 file from ADF.
DISPER utility program to compute Van der Waals dispersion coefficients. It requires files from one or two preceding ADF calculations using Time Dependent DFT.
BOB: the Basic Output Browser allows you to quickly scan the standard output file for important data.
(All utility programs are described in the Utilities manual)
The energy decomposition in a single-point run (and at the end of an optimization) and the energy calculation during optimization / Linear Transit / ... runs were not exactly identical in version 2.3, with discrepancies in the order of 0.01eV. Analysis revealed that a different handling of fit-related inaccuracies caused the difference and that, in fact, a slightly more accurate correction for the fit-error could be used in either case. Both algorithms have accordingly been improved and now give the same results.
Differences with ADF2.3 results (regarding this aspect) should be fairly small: 0.01eV or less. They may be larger when rather inaccurate fit sets are used.
Single-Point runs with single-atom fragments that were computed in ATOM symmetry now also use exact rather than fit densities for the computation of the electrostatic interaction between the fragments. This eliminates any fit-related error from that energy term.
The parallel execution sometimes failed when file permissions or machine-access settings prevented the ADF run scripts to copy (fragment) files from one host to another. We have solved this by implementing in the code an automatic transfer of (file) data through the parallel message passing facility if the kid processes do not have the same file as the parent. This will cause a few seconds delay in the very early phase of the program, because parallel message passing of a fair amount of data is relatively slow. At the other hand, it has reduced the run-script-controlled copying of data files and makes the procedure more robust.
A serious slow-down of BAND in the final stage of the calculation, usually accompanied by one of more logfile messages "BROKEN ELSTAT LOOP" (occurring during the processing of electrostatic interactions) has been cured.
The continuation code in the usage of block-type keys with arguments has been changed from the double plus character (++) to the ampersand (&). The old format is still supported, but not recommended.
User comments in the input file start with the exclamation mark (!) instead of the double colon ( :: ). The old format is still supported. The exclamation mark is recommended.
BAND cohesive energies are (by default) with respect to spin-restricted atoms. This has been different in BAND0.5; it was changed in a later version, but not stated correctly in the manual.
Energies in Linear Transit path points (ADF) included Hamiltonian terms that the user did not specify on input (in particular: relativistic). This has been changed.
ASCII output files can be generated with MOL and/or MOPAC and/or XYZ style coordinates, see the new FILE key.
The ADF logfile contains lists of Cartesian coordinates (not in a single point run). The number of displayed digits has been extended from 4 to 6 (to better preserve symmetry properties). Internal (Z-matrix) coordinates are now displayed also in the logfile if a Z-matrix structure has been given in the input file.
Bond energies (ADF) in LARGE molecules with COMPOUND fragments, or with single-atom IONIC fragments could be wrong (errors up to 1eV or more!) as a result of incorrectly ignoring electrostatic monopole interactions over larger distances. If you have done such calculations with ADF2.3 we recommend that you carefully check the results, and if feasible compare the data against calculations that use neutral single-atom fragments and/or against ADF1999. For the latter, please be aware that other improvements may also cause some (smaller) differences in results.
DENSF went wrong when higher virtual orbitals were processed while at least one lower one was skipped, producing incorrect data for the higher orbitals.
Frequency calculations (ADF) in T(D) symmetry with a large number of atoms and/or a large number of basis functions could cause overflow of data arrays, resulting in (often) an error abort, depending on what other data were overwritten exactly.
The Level Shifting procedure sometimes processed data incorrectly, leading to unintended changes in orbital energies during the SCF. Usually, this was rather counter-productive in terms of convergence speed although it did not affect the final density and related properties like the energy.
A frequency run could not start by using the TAPE21 file from a non-frequency run as a restart file (to pick up converged geometry data, for instance)
Parallel execution on SGI Power Challenge (maybe also Origin2000) failed when using more than 14 CPUs.
Mulliken populations in BAND did not always reflect the crystal symmetry correctly.
The ORBPOP key didn't work in BAND.
Symmetry breaking by an homogeneous electric field (ADF) was not handled correctly for the infinite symmetries D(lin), C(lin) and ATOM.
Linear Transit variables would change in the wrong direction when the FIRST reference to it has a negative sign in the input file.
The following error messages have been analyzed and (hopefully) cured:
- "wrong index block read". Note that this error may also show up when a fragment calculation failed and you incorrectly try to use the resulting fragment file. Usually, however, such attempt will give another error: "Wrong Superindex" or something similar.
- "no vectors in mergev". This was a numerical threshold problem.
- "Lowdin is not a linear combination of ETAS". This error occurred when the Create database file contained core orbital expansion sets for higher l-values, while no corresponding core orbitals were actually present.
Usage of PW86 or PW91 XC functionals could result in a core dump. Cause: use of uninitialized variables. Fortunately, this occurred in terms that weren't used anyway, so if the computation completed successfully, the results should be fine. On some machines, however, processing the uninitialized variables resulted in a crash.
Relativistic ADF calculations require preparatory DIRAC calculations for the involved atoms, to create a TAPE12 file with relativistic potentials. Contrary to ADF2.3 you must now also run DIRAC for atoms that don't have a frozen core at all, such as Hydrogen, because the TAPE12 file of the 1999 release contains not only the core potential and density but also the (relativistic) total atomic potential and density. This also implies that the TAPE12 files from release 2.3 must not be used in release 1999.
Moreover, relativistic ADF calculation now must use a relativistic TAPE12 file (prepared with DIRAC). In ADF2.3 you could use a non-relativistic TAPE12 file, or no a priori generated TAPE12 file at all (and get incorrect results) without even a warning from the program.
Default for the input key DIRECTIONALSCREENING (for the BAND code) has been set to "true".
Symmetry tolerance default (ADF) was documented as 1e-3 Angstrom, but was implemented as 1e-3 bohr. The documentation reflected the intention and the implementation has accordingly been corrected.
May 7, 1997
To solve a couple of problems with the parallel execution of our programs on certain platforms, a run-control script has been included in the package, called `start', which is placed by the installation procedure in the $ADFBIN directory. We recommend using this script always, also for the serial version. For parallel (PVM) versions you have to use it. This applies to executing all programs of the ADF suite (adf, band, densf, ...). See the Installation manual and the ADF User's Guide for comments and instructions. All Example runs have been adapted to use the script, providing you with an easy-to-copy illustration of how to use it (easy!).
Please let us know if you have any problems with it. Parallel programs are by far not yet a standard thing and controlling what precisely happens on the various nodes is very machine dependent. We strive of course to make our setup as portable and robust as possible, but comments and suggestions are always welcome to help us improve things.
The new setup with the script `start' assumes the existence of an environment variable (NSCM). This is not necessary, but it makes it easier to set / adjust the defaults for parallel calculations. See the manuals (Installation, User's Guide).
Serial (non-parallel) versions could encounter problems in the linking due to missing external statements.
SUN: missing library-resolve specification could lead to failure of the installation / link.
NEC-SX4 version available. Only the serial version has been tested so far. Work on the parallel (PVM) version is on its way.
Installation parameters for BAND on the Cray-T3E led to too many open files during the calculation and hence to a program abort. We are still testing the T3E version. Please let us know your experiences, if any...
On SGI_irix64 an additional compiler flag `-64' was needed for some older levels of the operating system to avoid linking in 32-bit object code. Also, the -O3 optimization level is now completely suppressed and replaced by -O2. This should not affect efficiency very much, while it does circumvent a couple of compiler bugs (!).
The Installation Manual and User's Guide (for ADF) contain a description of and remarks about the (new) method to run the programs of the ADF suite, using the UNIX script `start', which is delivered with the ADF package.
Parsing of expressions in input, previously applicable everywhere except in the data blocks of certain keys, has been changed into: only applicable for a specified list of keys, see the User's Guide. This does not imply a lot of change in practice, but merely reflects a different approach as regards the generality of the feature. The change was triggered by a previous error, that the program incorrectly `parsed' and hence corrupted symmetry labels of double-group symmetry groups in the input file.
On DEC and SGI (irix64) platforms, a failure could occur due to wrong data types in some calling sequences (only in parallel versions).
An error abort "SPHERE TOO LARGE" could happen due to numerical round-off problems.
`ILLEGAL DET. MATGEN' error abort fixed.
`MARKER ERROR' at the end of the SCF procedure, more precisely in module COREPS, could happen for molecules with large basis sets.
Date_and_time on HP was not activated correctly, leading to `TODAY' and `NOW', rather than correct date and time information in the logfile.
Due to numerical noise, diagonalizations and vector set orthonormalizations could yield inexact results, in some cases detected by the program as a fatal error.
February 28, 1997
There were a few problems with installing ADF2.2 that we've fixed in ADF2.2.1:
A runtime error occurred on SGI running IRIX6.1, probably due to a compiler error. The problem is circumvented in the new release.
Checking the license file could fail if the domain name as picked up by the license control module was empty.
Suppressing the usage of C code altogether could fail due to a linking problem (calls to the C modules were runtime suppressed but not eliminated from the source code, so that the linker reported unsatisfied externals.
Added to the supported platforms: Pentium_Linux, cray_t3e, rs6000_aix3.
License-file info now also printed when running with a license file that not includes the machine on which the run is attempted. Previously, you had to remove the license file first; otherwise the program would abort with a message `Invalid License'.
Installation on a SUN using C-code could fail (due to a typo in one of the C-routines).
Installation on sgi 64 bits (R8000 or higher) could fail due to problems with linking in lexicographical routines from the standard library.
W recommend that before you install the package you define the environment variable TESTBIN. This is used from now on in all sample runs to specify the directory in which the executables are located. Previously the sample runs addressed ADFBIN. You may, therefore, set TESTBIN equal to ADFBIN.
The documentation has not changed significantly, the most relevant issues being the remarks about the LB94 functional (ADF User's Guide) and the memory-usage control in DOS, DENSF, CNTRS (Utilities).
Two sample runs are provided for BAND (input and output files) in $ADFHOME/examples/band.
The BAND documentation incorrectly stated the availability of standard example runs in the code: these have been disabled.
The LB94 DFT functional is not suitable for geometry optimization (generally: not for the computation of gradients, hence also not for Frequencies). This is now documented in the User's Guide. In the program an attempt to use the LB94 option in combination with energy gradients will now fail, unless the internal check is disabled with the ALLOW key.
Installation (compile)-time adjustment of (maximum) memory usage didn't function: the pertaining parameters were rigidly defined within the binary part of the distributed code. Run-time adjustments of memory parameters worked OK.
In some cases ADF wouldn't recognize the true symmetry, finding a lower one instead.
Error abort `wrong nr of points' repaired.
Calculations that used a TAPE21 (or TAPE13) restart file from a SinglePoint run in a Geometry Optimization failed (because the program assumed the restart file to contain optimization-type data).
Parallel versions on DEC or SGI machines with LARGE memory (more than 2Gb) could fail (due to incorrect memory addressing).
LinearTransit runs without degrees of freedom for optimizations in each LT point aborted with a `KF error' message.
LinearTransit runs print a survey of the energy along the LT path when the scan is completed. This energy included only the terms used for the computation of gradients, not any post-SCF terms for instance or any correction terms to improve the estimate of the bonding energy. This applied also to the storage of such energy data in the LT section on TAPE21.
The printout of Dipole Moment values along an LT path, at the end of the run, was incorrect. The Dipole Moment values printed in the RESULT sections at the end of each of the restricted optimizations along the LT path, were correct.
Geometry Optimization runs with a dummy atom sometimes failed with an error message `wrong nr removed'.
An error abort `ERROR1 PLGIRR' could occur due to too sloppy thresholds in an internal numerical precision test.
Calculations on a molecule that was close to a higher symmetry than actually detected by the program sometimes failed with an error message `ERROR 1. SYTABL'.
The program evaluated `mathematical expressions' also in parts of the input file where it was not supposed to do so. This led to errors when supplying, for instance, OCCUPATION numbers for double-group symmetries with labels that end with an asterisk (*).
Unrestricted LinearTransit runs failed with a `KF' error message like `non-existing variable'
Due to the allocation of unnecessarily large arrays the program could run out of memory in relatively small calculations on platforms where character-type data are not dynamically, but statically allocated. In particular: Cray platforms.
Maximum memory usage can now also be adjusted (run-time) for DOS, DENSF, CNTRS. See the Utilities manual.
DOS utility program wouldn't execute on some systems (sgi, ...) with an error message like "no atom pairs" or something similar.
Part of the package includes binary code (object files) for a few minor modules, and for license-control software. This has increased the size of the whole package since for each of the supported platforms separate object files are incorporated.
The supported platforms are each represented by a subdirectory within $ADFHOME/Install.
For UNIX platforms not available there, contact us (email: firstname.lastname@example.org). The issue here is not that the program would not be portable on other platforms, but only that we haven't yet generated the required object files for them.
The program controls license matters by reading and checking a license file. If you don't have a license file yet, don't worry: just install the program and run it. It will stop and print out information that we need to generate your license file. E-mail (or fax) this info to us.
Be aware that at installation time, although you don't need to have the license file yet, you must know where the license file will be put (the installation script will ask you for the "name" of the license file). This must be a place accessible by all people that are going to use the program.
The installation script has been revised.
READ THE INSTALLATION MANUAL!
New basis sets: W.4d, Xe.4p, Ni_d9.2p and Ni_d9.3p: in II and IV.
The Ni_d9 basis sets are tuned to the d9 configuration of the isolated Ni atom. The other Ni basis sets (without _d9 suffix) are more suitable for the d8 configuration. Both sets can be applied in Ni-containing molecules. The differences are not very large, but the more appropriate choice in an actual situation may improve the basis set quality somewhat.
New basis sets with various frozen core levels and relatively large fit sets (larger fit sets also for the old basis sets) are available now for La through Rn.
All-electron basis sets are available, in the directory $ADFHOME/atomicdata/AE. However, the data base files here have no fit sets with them so you cannot use them directly in a CREATE run. Standard fit sets, as used in the other database files, cannot be used because they would be highly inadequate and lead to incorrect results. The generation of fit sets suitable for a particular basis set is not a triviality. The all-electron basis sets are being made available only for experimentation purposes, not to be used as a "standard". We intend of course to generate fit sets for them, but other issues have higher priority.
Major parts of the ADF documentation are already available on the WorldWideWeb. This will be extended in the near future.
The documents EXAMPLES and APPLICATIONS have been combined into the new (more extensive) EXAMPLES document. A few more sample runs have been incorporated to cover new functionalities. The "old" sample runs have been renamed.
All restart facilities available, including cases such as: use the Hessian from a Frequencies run in a follow-up optimization or TS search.
LinearTransit and Frequency runs can be carried out in parts, using the restart facility to control the continuation. The number of steps to be done in one particular run is controlled with the ITERATIONS subkey (in the GEOMETRY block), which may have a second argument (for the LT run).
Frequency calculations can be carried out with constraints (for instance, to study the vibrational modes of a more-or-less rigid fragment w.r.t. a metal surface).
Spin-Orbit relativistic option available: only in SinglePoint runs.
Orbital Localization (Boys-Foster method) available. Localized orbitals can also be addressed by DENSF (preparation for plotting or 3D maps).
New XC functionals: Xonly (pure LDA exchange), LYP (Lee-Yang-Parr), BLYP (same, but combined with Becke88 exchange corrections), PW86 (Perdew-Wang formulas of their 1986 paper).
XPS: The one-electron energies of the Core Orbitals in the final SCF field are computed and printed, in the RESULTS output section. Absolute values may not be very accurate (because the core orbitals are not self-consistently relaxed in the SCF field), but differences between e.g. the 1s orbitals of distinct Carbon atoms should be fair and hence provide a good estimate of (differences in) chemical shifts.
The AUFBAU principle in geometry optimizations etc. will by default be applied in the same way as for the start-up geometry (a few SCF cycles, occupations frozen thereafter). Used to be: only applied in first geometry. This feature can be tuned with the (new) option AUFBAU2 to the key OCCUPATIONS.
ElectronSmearing has been largely suppressed: only applicable in optimizations (incl. TS searches), but not in the final geometry.
The symbol for an "Alternative Element" can also be Gh (rather than J).
The atomic mass can also be specified (to override default values) for "normal" chemical elements: this used to be possible only for J-type elements.
DFP (Davidon-Fletcher-Powell) Hessian update scheme available in TS search (was available already in optimizations).
Dynamic allocation of memory has been implemented. Installation-defined workspace sizes are now only defaults which can be adjusted by input-keys. No need anymore to compile differently-sized program versions. It also improves in a general way the efficiency of memory usage.
Default (automatic) values for Numerical Integration precision have been increased AND now depend on the runtype (high default value for Frequencies, for instance).
Minor modifications in the Numerical Integration grid generator, leading to higher standard precision within the atomic core regions. This may result in slightly (but not essentially) different outcomes.
The electrostatic interaction energies, and their derivatives w.r.t. nuclear displacements, were computed from the fitted (i.e.: approximate) densities. An evaluation using the exact densities has been implemented. This generally leads to somewhat different results. The so-corrected inaccuracies in the old version are larger for atoms that had relatively inaccurate fit sets in their database files.
The SCF convergence criterion is relaxed during the intermediate cycles of optimizations. This reduces (usually) the total execution time. In some cases it may lead to using one or two geometry cycles more. This feature (a secondary SCF criterion) is controllable with the (extended) CONVERGENCE option to the SCF key. The (new) key FULLSCF can be used to suppress the feature altogether.
Optimization and TS search strategies have been modified to solve stability problems and improve robustness (geometry convergence sometimes failed after initial progress towards the minimum; TS searches very often failed to converge by randomly walking around the TS point).
Calculated energy gradients are adjusted to eliminate spurious components of rigid motions. This used to be treated in a slightly different way in Optimizations versus Frequency calculations. As a result, the computation of Frequencies in a previously optimized geometry could issue warnings that "GRADIENTS IN THE EQUILIBRIUM ARE NOT SMALL".
A new key DEPENDENCY has been introduced to handle numerical problems with very large basis sets that contain diffuse functions. It should not be necessary/useful to apply it to the standard basis sets provided with the package.
Standard output has been restructured a bit to separate the distinct main parts more clearly: general characterization of the run, building info (fragments, basis sets), computational progress, and results. These main sections are controllable with PRINT keys.
PRINT keys/options EIGSFO and EIGCFSFO have been removed. A general (EPRINT) option SFO is available to address all SFO-related output switches.
For hints to suppress large parts of printed output see the User's Guide.
Printed Voronoi Charge analysis: results are printed with opposite sign wrt ADF2.0: a net total negative charge (electrons!) is now printed negative. Previously: positive (i.e.: effectively counting the number of electrons). In addition, the "Voronoi Deformation Density" is printed, in the last column of the table.
In a Frequencies run the computed Force Constants Matrix is by default printed on standard output.
Calculations on linear molecules performed in finite ("broken") symmetries used incorrect numerical integration schemes in some special cases.
SFO populations in spin-unrestricted calculations were incorrect for spin-A (the spin-B values were printed both for spin-A and for spin-B).
Constrained optimization in Cartesian coordinates could go wrong when the molecule was rotated by the program to adjust the input coordinates to symmetry-related orientation requirements.
In O(h) symmetry f-type atomic basis functions were treated incorrectly. The effects are (in all cases that we verified) minor.
LinearTransit runs failed when all non-transit coordinates were kept frozen AND non-local (GGA) XC-options were used.
The PW91 functional implementation had a few numerical problems leading in some cases to ridiculous (and obviously wrong) values. Application of PW91 in UNRESTRICTED calculations often failed.
Recognition and treatment of point group symmetries D(2h) and D(2d) often went wrong. C(s) occasionally also caused difficulties.
The calculation of Frequencies for a linear molecule specified with finite ("broken") symmetry failed (error message: WRONG NR OF RIGIDS).
The key REMOVEFRAGORBITALS failed when ALL orbitals of a fragment irrep were removed.
When the LevelShift SCF-option was used the printed one-electron orbital energies were not corrected (and hence included the LevelShift value).
The dipole moment was incorrect in an UNRESTRICTED calculation that was executed in PARALLEL mode.
The LevelShift option didn't function correctly for spin-B (in an unrestricted calculation).
On some machines and in some applications the program would abort with an error message like "STEBZ ERROR". Some (hopefully all) of these cases have been cured. Please notify us when this error still occurs: we have not been able to REPRODUCE some of the reported errors, so we are not sure that we've traced all underlying causes.
Input coordinates used to be adjusted so as to place the center-of-mass in the origin. This is applied now only as far as necessary to match symmetry-related orientation requirements.
Installation defaults for the maximum numbers of atoms, basis functions, et cetera are increased considerably and should accommodate large calculations without adjustment. See the Installation manual.
A new key INCLFILE is available to store for instance personal preferences for standard input items in a separate file and link that to an actually used input, rather than typing in all the preference values each time again.
The SAVE key options has been extended to enable saving scratch files (for instance).
The counting of free (optimization) variables has been modified so as to make them be strictly ascending along the input list of atomic coordinates. This affects the selection of a specified coordinate to focus on in the initial phase of a TS search.
We will start including a list of ADF-related publications (by users) in the standard documentation of the package. This list will (by necessity) not be complete but is meant primarily as a starting point for (other) users to access literature related to their own research. We therefore invite you to send us a reprint of publications with a brief accompanying note containing a one-line characterization of the publication and the program's relevance for it and two or three keywords.
DENSF documentation has been upgraded (was rather unclear and in some cases incomplete or incorrect). Printed output gives a summary of invoked items. Values of the electrostatic potential had the wrong sign.
Documentation of KF utilities (pkf, dmpkf) improved to clarify usage and results.
New utility program DOS: for the generation of density-of-states type information: Total as well as different kinds of Partial (gross-, overlap-, ...) density-of-states data.
The UNIX Installation procedure has been modified. The script "configure", which previously had to be executed with as argument the name of the platform on which the installation takes place, is now executed without argument. It is interactive and starts a dialogue with the user about various installation issues.
Read the Installation manual.
Special attention is required for the installation of a PARALLEL version of ADF. As message passing libraries both PVM and MPI are supported, but virtually all testing has been done with PVM. We strongly recommend therefore PVM. PVM is public domain software and has to be installed separately before you install the PVM version of ADF.
Read the Installation manual carefully!!
The Installation manual has been thoroughly revised, in particular with an eye on the parallelization options.
Additional documentation has come available:
[Sigma] "Applications": a discussion of special applications of ADF the description of which would not fit well in the User's Guide. In this release "Applications" contains only a discussion of the BSSE (Basis Set Superposition Error).
[Sigma] "Utilities": description of auxiliary programs and utilities that are part of the ADF package. One of these programs s a utility to generate density, potential, orbital values on a user-specified grid (e.g. for plotting).
The User's Guide has not been changed in a very fundamental way but the modifications are so many that it is undoable to summarize them here.
Relativistic (Scalar, Pauli formalism) geometry optimization.
In geometry optimizations an initial guess of the Hessian is computed from a Force Field (previously: set to diagonal unity). This results in most cases in faster geometry convergence. More importantly: a Transition State search can now be started directly, without first computing the Hessian by way of a Frequencies run.
Hirshfeld and Voronoi analyses of atomic/fragment charges.
Atomic coordinates can be input in a new format "ZCART": Cartesian coordinates with Z-matrix structure. This can be used to apply optimization in internal coordinates while input is given in Cartesians.
Density Functional options extended: Perdew-Wang-1991 formulas for exchange and/or correlation, and the Perdew-Wang-1986 formulas for
ADF has been parallelized. For installing a parallel version read the Installation manual.
The Direct SCF feature has been extended from fit functions to (also) basis functions. This gives a considerable reduction of disk access. The keyword controlling this is DISK (used to be DIRECTSCF).
The internal handling of files has been totally revised. This has made other developments (now and in the future) much easier. One of the (negative) consequences is that the old files are incompatible with ADF2.0: any TAPE21s etc. from ADF1.1 can not be used in ADF2.0.
Better use is made of symmetry in geometry optimizations, giving significant speedups for molecules with high symmetry.
Processing of basis and fit functions and the computation of (Fock) matrix elements takes distance effects into account, so as to speed up the calculation by not computing negligible contributions. Technical details can be set with TAILS.
The DIIS (Direct Inversion in the Iterative Subspace) approach has been implemented for geometry updates in optimizations and TS searches. This supplements the usual Newton-type strategy and in most cases improves the convergence behaviour.
Numerical Integration precision is (optionally) self-adapting during geometry optimizations.
Interpolation method of frocen core data from the core tables has been improved. Same for interpolation of Electrostatic interaction terms in geometry optimizations. In both cases the improvements are minor.
fragments computed in C(lin) or D(lin) fragment symmetry could not be used in molecules with C(lin) or D(lin) symmetry: the results were wrong or the program crashed.
An error in the computation of energy gradients has been detected and repaired. This played hardly a role in most cases, but was significant in the presence of large frozen cores.
A1 frequencies were not computed in T(d) and O(h) symmetries.
A small error in the Fit-Correction terms in the bond energy has been detected and repaired. Effect on energy very system dependent, usually (much) smaller than 0.1 eV.
Labels of T1.u and T2.u subspecies in O(h) symmetry have been
interchanged to match the most often encountered assignment of labels
in the literature.
"SELECTED" optimization was not recognized (resulting in complete
variation of all coordinates) if the type of coordinates was specified
explicitly by the OPTIM subkey in the GEOMETRY block.
The LevelShift option (key: SCF) did not work correctly.
Restarts of geometry optimizations etc. have temporarily been disabled. The same holds for the computation of frequencies with restricted degrees of freedom. Both are related to the new file system and the fact that we are still fine-tuning the new file structure. The disabled options will be re-enabled in the next release.
For several keywords the units in which input is interpreted has been made absolute, independent of the units used for atomic coordinates input. This applies for instance to EFIELD (homogeneous Electric Field). Read the User's Guide carefully. Such changes in units are notified in bold face in the User's Guide.
Several keywords have been modified somewhat as regards suboptions defaults. See the User's Guide, in particular for the keywords GEOMETRY, SCF, OCCUPATIONS, CHARGE.
The key ELECTRONS has become obsolete. It can still be used, but will be removed in the next release. Its functionality is represented by OCCUPATIONS and CHARGE.
Atomic masses stored in the database have been adapted. They used to be isotope-averaged values. Now the correspond to the most abundant isotope. This affects frequencies (not much).
PRINT options have been added to completely suppress certains types of information on output. See the PRINT options FRAG and SFO.
Description of the ZORA relativistic option has been removed from the User's Guide. The ZORA formalism is still in development and the currently available basis sets in the ADF data base are not suitable for this approach: results can absolutely not be trusted.
All-electron basis sets have been added to the database. However, the files do NOT contain fit sets and hence cannot be used as such in a Create run. We do not yet have fit sets that are adequate for the all-electron basis sets. Using the standard fit sets from the other (frozen-core) data base files is NOT recommended.