Supramolecular Organization of Functional Organic Materials in the Bulk and at Organic/Organic Interfaces: A Modeling and Computer Simulation Approach

Abstract

The molecular organization of functional organic materials is one of the research areas where the combination of theoretical modeling and experimental determinations is most fruitful. Here we present a brief summary of the simulation approaches used to investigate the inner structure of organic materials with semiconducting behavior, paying special attention to applications in organic photovoltaics and clarifying the often obscure jargon hindering the access of newcomers to the literature of the field. Special attention is paid to the choice of the computational “engine” (Monte Carlo or Molecular Dynamics) used to generate equilibrium configurations of the molecular system under investigation and, more importantly, to the choice of the chemical details in describing the molecular interactions. Recent literature dealing with the simulation of organic semiconductors is critically reviewed in order of increasing complexity of the system studied, from low molecular weight molecules to semiflexible polymers, including the challenging problem of determining the morphology of heterojunctions between two different materials.

Appendix: Simulation Packages

This Appendix is intended to serve as an overview of the more common computer simulation packages suitable for molecular simulation studies of organic materials. Actually the landscape of available packages is essentially dominated by MD-based codes. In fact, MC and KMC codes are commonly developed in-house by researchers to handle specific problems requiring ad hoc sampling algorithms, and hence have a much less wide user base than MD ones. In contrast, many computer codes are available to the end-user willing to carry out MD simulations. Some of them come with an extensive support including manuals, tutorials, and online discussion boards such as forums and mailing lists. Here we report a selection of packages which are suitable for simulating large systems of organic molecules and which are used by a large community of users. This is of course not a measure of the quality of the program, but rather it ensures that a large body of published works are available as reference, therefore constituting a valuable starting point for those approaching the field of MD for the first time. Moreover, the more popular codes are continuously maintained and updated, an important aspect to take into account as machine architectures change very often. It is worth noting that while the standard packages provide MD trajectories (as sequences of instantaneous configurations), the calculation of many observables of interest have to be specifically added by the end-user and suitable algorithms often have to be devised. The available MD packages are characterized by various factors:

• 1. Features and capabilities, e.g., multiple-timestep integration algorithms [96, 216], physical representation of simulated objects (coarse-grained or all-atoms), constrained dynamics (e.g., SHAKE [217]) with many codes allowing for multiple options.

• 2. License and cost, i.e., free-academic, open source, commercial.

• 3. Portability, i.e., the code can easily be compiled and run on many platforms, from common (e.g., computer desktops) to specialized hardware (e.g., IBM’s BlueGene).

• 4. Performance and parallelization, i.e., the ability to run faster and simulate bigger systems by splitting work among multiple processors. Nowadays support of GPU boards is widely available and allows the deployment of graphics cards alongside traditional CPUs.

• 5. Extensibility of the code in order to tailor specific problems or to implement new force-fields/simulation algorithms/computation of observables, which were not available in the original code.

The choice of a particular MD package depends on the first instance on the system being studied, particularly on the model used to describe the interactions between the particles and on its scalability (i.e., the computational efficiency as the number of processing units increases), which can be a limiting factor when the system size exceed tens or even hundreds of thousands of particles. A necessarily partial list of computer programs to carry out MD simulations is reported in Table 1.

Table 1 Partial list of MD simulation packages

Two well-established computer programs for MD simulations are CHARMM [24] and AMBER [219], which implement a similar functional form of the atomistic force field and include a large number of tools for setting up the files required to run an MD simulation.

NAMD [227] is a more recent program which works with AMBER and CHARMM potential functions, parameters, and file formats, and it is specifically designed for high-performance simulation of large systems. The current version (2.9) is able to run on heterogeneous architectures made up of multiple CPUs and GPUs.

LAMMPS [225] is a classical MD program implementing potentials for soft materials (biomolecules, polymers), solid-state materials (metals, semiconductors), and coarse-grained or mesoscopic systems. The code is designed to be easy to modify or extend with new functionalities. The comprehensive manual compensates for the somewhat clumsy input script syntax. Most of its model potentials have been parallelized and run on systems with multiple CPUs and GPUs, granting very good speedups, especially for the most complicated pair potential styles, like the Gay–Berne and other CG potentials.

GROMACS [231] is conceived to carry out simulations with millions of particles. The syntax of input files is user-friendly and a major advantage is that the program comes with a large selection of tools for trajectory analysis. In version 4.5 only single GPU support is present, but from version 4.6 multiple GPUs will also be enabled.

The output of an MD simulation typically includes a trajectory file containing the position of every particle, saved with a given time increment. Trajectory files can be visualized with specific programs such as VMD [232–234], which also offers basic tools for data analysis, Jmol [235], a powerful and highly portable program written in Java, Mercury [236], and GDIS [237], two programs particularly well suited to visualize and manipulate crystal structures, and Avogadro [238, 239], which offers advanced molecular modeling tools. Among the other visualizers available, we mention OVITO [240, 241], PyMOL [242], Molekel [243], V_Sim (http://www-drfmc.cea.fr/L_Sim/V_Sim/index.en.html), Ras-Mol [(http://rasmol.org/), 244], FOX [245, 246], QMGA [247, 248], UCSF Chimera [249, 250], and BALLView [251, 252].

The vast choice of available packages might be a bit intimidating at first, so here are listed some of our (objectionable) views on the key issues. Given the high quality of many open source packages, they are probably to be preferred over commercial ones with closed sources, both for their plentiful features (as everybody can contribute to further development) and because the sources can be directly inspected to check how algorithms are implemented and can be customized to suit particular needs. As many algorithms work well only under specific conditions and many compromises are usually made, this is in some cases the most reliable way to check the validity of the simulation results.

As long as small samples and/or short timescales are needed and the package provides the required features, it really does not matter which code is used, and the one easier to run is to be preferred. If instead the problem at hand requires what can be accomplished currently to be pushed to the limits, a well optimized code becomes the only choice. Given the typical speedups of GPU systems, codes able to run on heterogeneous architectures (mixed CPU/GPU) environments are the best performers. Note: most of the codes running today on GPU use CUDA (http://www.nvidia.com/object/cuda_home_new.html) instead of OpenCL (http://www.khronos.org/opencl), which means they will run only on NVidia cards.