Dear user of ABINIT (in short : ABINITioner),

If this is the first time that you have access to ABINIT, or that you receive an ABINIT announcement, welcome !

On the Web site, you will find a lot of things, including installation notes for different versions of ABINIT, pseudopotentials, some utilities, benchmarks, and a list of FAQ.

If you follow carefully the installation notes on the Web site, for a particular version of the code, you will soon be able to read information files contained in the ~ABINIT/Infos directory on your local machine.

We will use the name ~ABINIT to refer to the directory that contains locally the ABINIT package, uncompressed. In practice, a version number is appended to this name, to give for example : ABINITv4.1.2. ~ABINIT contains different subdirectories. For example, the present file, as well as other descriptive files, should be found in ~ABINIT/Infos . Some of these files are written in html, and are accessible also from the Web site : e.g. the tutorial and the list of all ABINIT input variables. The best entry point, after the installation, is likely the "new user's guide". Contact us if you have still questions after having gone through the different files.

The logic of version releases is as follows:
The full name of a version has three digits (for example, 4.1.2). The first digit is the slowly varying one (in average, it is changed each other year). It indicates the major efforts and trends in that version. At the level of 1.x.y ABINIT, the major effort was placed on the "ground-state" properties (total energy, forces, geometry optimisation, molecular dynamics ...). With version 2.x.y , response-function features (phonons, dielectric response, effective charges, interatomic force constants ...) were included. The main additional characteristics of version 3.x.y were the distribution under the GNU General Public Licence, the set-up of the documentation and help to the user through the Web site in html format, and the availability of GW capabilities. The version 4.x.y should put a lot of effort in the speed of ABINIT (e.g. PAW), and its parallelisation.

When additional features are present, the second digit in incremented. So versions 1.x started with 1.0 and finished at 1.9, with more and more features added. Version 2.0 was released mid-July 1999, and version 2.1 mid-November 1999. Two versions differing by the last (third) digit have the same capabilities, but one with the largest last digit is more debugged than the other : version 2.0.4 is more debugged than 2.0.3 or 2.0.2, but no new features has been added (so likely, no additional bug!).

As mentioned above, version 3 of ABINIT (as well as the present version 4 and future versions) is distributed under the GNU General Public Licence. The GNU General Public License is intended to guarantee your freedom to share and change free software--to make sure the software is free for all its users. To protect these rights, one needs to make restrictions that forbid anyone to deny you these rights or to ask you to surrender the rights. These restrictions translate to certain responsibilities for you if you distribute copies of the software, or if you modify it.

Usually, a version of ABINIT is released only when it has been installed 'successfully' on 5 different platforms. A successful installation includes a battery of more than 300 tests, whose results are automatically analyzed (this does not garantee that the code is bug free, however, and the contribution of testers is crucial).

For those who plan to install ABINIT on a new platform :

Also the default non-local routines (nloalg=3 or 4) are optimized for superscalar architecture, You should likely use nloalg=1 or 2, theoretically optimized for vector machines.

For those who plan to develop new features : you will find the information for developpers in the subdirectory ~ABINIT/Infos/Notes_for_coding, when the code is installed on your machine.

A few information not contained in ~ABINIT/Infos are worth to mention here:

Here are a few advices about the SCF convergence. For small unit cells, there should be no problem : use iprcel=0 with iscf=3 (likely the fastest) or iscf=5 (more robust). For small molecules in a big box, use iprcel=0 with diemix on the order of 0.3 ... 0.8, and diemac=1.0d0, again with iscf=3 or iscf=5. For a larger number of atoms (>10), or long cells (more than 4 layers of a material), you might gain a lot by using iprcel=45, but then you MUST include a few conduction bands (20%-30% of the valence bands) and play also with diecut (the default 2.2d0 might be a bit low). On some systems (metallic slabs + atom or molecule), it is possible to gain a factor of three ... or simply to be able to converge !

For those who would like to compare ABINIT with other codes, especially for speed, please note the following :

Despite the FFT and non-local psp efficiency, ABINIT has likely little chance to outperform a code that uses ultra-soft pseudopotentials. However, ABINIT should have nice features that might not be found in some other codes, noticeably :

If you find a problem with ABINIT, please contact us, and try to follow the ~ABINIT/Infos/problem_report prescriptions.

Thank you for your help in this project !

Xavier and the ABINIT group.

Copyright (C) 1998-2004 ABINIT group (XG)
This file is distributed under the terms of the GNU General Public License, see ~ABINIT/Infos/copyright or .
For the initials of contributors, see ~ABINIT/Infos/contributors .

Goto : ABINIT home Page | Welcome | Suggested acknowledgments | List of input variables | Tutorial home page | Bibliography
Help files : New user's guide | Abinis (main) | Abinis (respfn) | Mrgddb | Anaddb
Other links : UCL ab initio group | Calcul Intensif

Responsable : Xavier Gonze <>
Conception : Jean-Michel Beuken <>