skip to content

Materials Theory Group


Ephemeral Data Derived Potentials

The ddp package contains a suite of tools to construct and test data derived interatomic potentials. They were originally designed to be used with the airss first principles structure prediction package. Ab initio random structure searching (AIRSS) can be used to generate data, and exploit the generated ddp potentials to potentially accelerate searches. They are referred to as ephemeral data derived potentials as they can quickly be custom-built for a particular set of structure searching parameters, discarded and regenerated as those parameters change. The methodology is introduced in Pickard, Ephemeral data derived potentials for random structure search, 2022EDDP version 0.2, which extends the use of EDDPs to high-accuracy molecular and lattice dynamics, is presented in Salzbrenner et al., Developments and further applications of ephemeral data derived potentials, 2023.


The airss package should be installed, as well as the source for the nn package, before installing the ddp and repose packages. Specifically, the SPGLIB library built by airss is reused by the repose package. Various command line tools, such as gnu parallel are used. The grace plotting package is supported. gfortran 9.3 or above is recommended, but ifort is supported.

Obtaining the eddp packages

The eddp packages (ddp, repose and nn) are released under the GPL2 licence, as is the airss package. You can download the full source code for all the packages below. Note that you will require a Unix-like environment in order to build the code. Windows users can install a Linux subsystem if required.

Download current versions of airss, ddp, repose and nn

Getting started

The airss, ddp, repose, and nn packages are compiled and installed using make && make install, and if successful you should add {airss,ddp,repose}/bin to your path. The ddp-batch suite of scripts enables convenient training of EDDPs on batch queueing systems.

Next steps

The README file for the ddp package provides some sample data for generating potentials, as well a instructions for perfoming an iterative fit.

Specific versions of EDDP:

EDDP Version 0.2

EDDP Version 0.1

Latest news

Fast and easy exploration of crystal properties using machine-learned Ephemeral Data-derived Potentials

12 January 2024

Machine learning is quickly gaining prominence in the field of computational materials science. In the Materials Theory Group, we develop so-called ‘machine learned interatomic potentials’ (MLIPs), which can describe the...

Structure and colour in nitrogen-doped lutetium hydride

19 December 2023

Superconducting materials have a wide range of applications - from efficient power transmission to the advanced electromagnetics used in MRI machines - due to their loss-free conductivity. Current practical superconductors...

Quantum-induced hydrogen hopping in high-temperature superconducting lanthanum polyhydride

14 April 2023

Figure caption : Quantum effects are essential for hydrogen to dynamically explore different configurations. On the left, we see how the hydrogen atoms cover much larger distances at all temperatures when quantum effects are...

Flat water and ice

26 September 2022

Figure Caption : Pentagonal ice – a two-dimensional form of ice predicted to form when water is squeezed between graphene sheets. Water can be found trapped in nanoscale cavities, for example in biological membranes, or in...

Congratulations Ben Shires!

2 August 2022

Ben completed his PhD viva last week, covering his work on SHEAP , and he will soon be Dr Shires. Congratulations! shires.jpg

Quicker Crystals

14 July 2022

First-principles structure prediction has enabled the computational discovery of materials with extreme, or exotic properties. For example, the dense hydrides, which following computational searches have been found to...

New carbonates uncovered

7 January 2022

A study by Joseph Nelson and Chris Pickard of the Department of Materials Science and Metallurgy, University of Cambridge and the AIMR, Tohoku University, uses structure prediction to exhaustively explore the Ti-C-O and Al-C-O ternary systems.