Internal units and conventions

Everything is a.u.

All the units used internally by i-PI are atomic units, as given below. All of them, even the temperature. By default, both input and output data are given in atomic units, but in most cases the default units can be overridden if one wishes so. For details on how to do this, see Overriding default units and Properties.

Unit

Name

S.I. Value

Length

Bohr radius

5.2917721e-11 m

Time

Atomic units

2.4188843e-17 s

Mass

Electron mass

9.1093819e-31 kg

Temperature

Hartree

315774.66 K

Energy

Hartree

4.3597438e-18 J

Pressure

Atomic units

2.9421912e13 Pa

Regarding the specification of these units in the i-PI input files, the user is able to specify units both in the i-PI input file or in the structure file. If the structure file is of the .xyz format, the units specifications should be present in the comment line. Examples of such inputs can be found in examples/pes-regtest/io-units/ . The code then behaves in the following way, depending on the user’s choice:

  • If no units are specified in the input, i-PI tries to guess from the comment line of the structure file and it nothing is present, assumes atomic units, except for the cases discussed below.

  • If units are specified in both input and structure file and they match, conversion happens just once. If they do not match, an error is raised and i-PI stops.

There are a few exceptions, concerning quantities that are not used to describe the atomic-scale system, or I/O that is constrained by existing standards:

  • Quantities that define run-time execution options, such as the simulation wall-clock time limit (specified by the flag <total_time>) or the latency of the forcefield socket (specified by the parameter <latency>), have to be provided in seconds.

  • The PDB standard defines units to be in angstrom, and so default units for PDB are angstrom.

  • When using ASE format for I/O, the units are dictated by the ASE defaults. So for example, the ASE extended xyz format expects positions in angstroms. (See for example examples/ase-io/)

Lattice parameters

The unit cell is defined internally by i-PI with the lattice vectors stored in the columns - i.e. as a \(3\times 3\) array in which h[i,j] contains the i-th Cartesian component of the j-th lattice vector. Furthermore, the cell is constrained to be oriented with the first vector along the \(x\) axis, the second vector within the \(xy\) plane, and the third in an arbitrary position (so that the cell is stored internally as an upper-triangular matrix). This is also reflected in how the cell parameters should be provided in the input file: when given in an array format, the correct format is e.g. <h units='angstrom'> [10, 1, 2, 0, 9, -1, 0, 0, 11] </h>.

The socket flag pbc=True or pbc=False

Inside the ffsocket and almost all other socket-type blocks, there is a flag called pbc, which has generated a lot of debate regarding its meaning and regarding its behaviour (see Input file tags).

The purpose of this flag is to tell i-PI whether to wrap the positions before sending them to the client or not. It bears no consequence to the type of simulation you are performing (i.e., this is not about whether the simulation is periodic or not).

Before the release of i-PI v.3.0, the default of this flag was set to True. This choice was made especifically because of the LAMMPS client (see below), but for all other clients we had experience with, they either did not care whether this flag was true or false and worked fine regardless or they would break down completely or in corner cases pbc=True.

With the release of i-PI v3.0 we set the default of this flag to be False. This has improved the user experience in most cases. However, we do recommend using the latest version of LAMMPS with this new default, as we also had to update the LAMMPS client interface, which incidentally became a lot faster (because we avoid triggering the neighbor-list calculation at each step).

If you se issues with your LAMMPS calculation (instability, missing atoms), especially for older versions of LAMMPS, setting pbc=True in your socket should fix it.