DFT Software: Who, What, and How Much Brad Hubartt Introduction: was constructed [11]. KS-DFT takes the Doing pretty much anything in original energy functional: quantum mechanics is going to involve the E[ρ]=T [ρ]+ Vext(r)ρ(r)dr + VH [ρ]+Exc[ρ] wave equation of the system. Using the wave ! equation in a large system is going to be very where T is the kinetic energy, Vext is the difficult or impossible, if the wave equation is external potential, VH is the Hartree energy, even known. This essentially rules out using Exc is the exchange correlation energy, and ρ quantum mechanics in biophysics/chemistry. is the charge density. ρ is what Kohn and A new quantum model is needed for these Sham modified. They derived the following systems. Density Functional Theory (DFT) three equations: [4][6][9][12] is such a model. It involves N using the probability density functions of the ρ(r)= ψ (r, s) 2 electrons in a molecule to determine other | i | i s 2 ! ! properties of the molecule. Instead of solving ! 2ψ (r, s)+v (r)ψ (r, s)=# ψ (r, s) problems with wave equations for each − 2m∇ i eff i i i electron it uses the probabilities of electrons 2 ρ(r!) δExc[ρ] veff (r)=Vext(r)+e dr! + being at different locations. If the wave r r! δρ function is not known many implementations " | − | of DFT will allow for Gaussians to be used in The first one is exactly what anyone would their place. expect, the second is Schrödinger’s equation, DFT works by minimizing functions and the third is the actual potential at a given of functions, functionals, to determine the location based upon the externally applied ground state of a molecule. Instead of using potential, the charge density, and the functionals based upon each individual correlation energy. ρ can be solved for by electron, a pseudopotential is used to simplify integrating over these three equations with an the calculation. Additionally it can calculate RK4 [10] or something equivalent. Often the forces and has been expanded to include time- wave equation is replaced with a Gaussian to dependence [13]. Time-dependent DFT simplify things. makes simulations of molecular dynamics Another technique used to simplify possible. Molecular dynamics are of DFT is replacing all-electron systems with a particular interest as nothing in a biological pseudopotential. A pseudopotential is an system is static; proteins are floating around approximation of the Coulombic potential in a solution and constantly interacting with from the non-valence electrons. In chemistry different ions. the valence electrons are the ones that actually form bonds and cause reactions; the Methods: non-valence electrons are just there for the This is a seemingly wonderful theory, potential field (obviously a gross but how is it implemented? It is not oversimplification). Pseudopotentials are used particularly useful if every equation has to be when generating a self-consistent field. Pulay solved by hand at each iteration. First, where Mixing [3] (also called Direct Inversion of does the electron density come from? Kohn Iterative Subspace) is a non-linear mixing and Sham came up with an approach for this technique used for large SCF problems. which is how Kohn-Sham DFT (KS-DFT) Hubartt 1 Instead of having to store a giant hessian for unit cell, nband is the number of bands to be the system this algorithm works with an considered, and iscf determines which mixing amount n (user specified integer) of charge algorithm is used to find a self-consistent density vectors. Through a few linear field. The iscf variable by default is set to use operations a final charge density vector is Pulay Mixing. Other mixing algorithms are computed. The computational benefits of built into Abinit that are simpler than Pulay Pulay Mixing are increasingly noticeable the which could give mixed results based upon large the system becomes as the hessian the size and complexity of the system. Other grows as nelectrons squared versus needing only more advanced input variables are built into a handful of vectors for Pulay mixing. The Abinit including the Bravais lattice type, how actual algorithm for Pulay mixing is in to calculate the Berry phase, units for energy, reference [3] and [4]. various dielectric properties, etc... ABINIT No one wants to solve this by hand, has two different pseudopotentials built into it the natural place for this theory is in computer for each element. Additionally, it has multiple simulations. A quick internet search reveals a algorithms to calculate a new pseudopotential few dozen pieces of software for solving if one is desired. these systems. Each one comes with its own What does it output? Besides a quirks, learning curves, availability of confirmation of all the input parameters it has documentation, open/closed-source, and a hefty amount of items to output. The first is whether the originator of the software has the point symmetry group and the Bravais been banned for life from using it (amongst lattice type. The Bravais lattice, along with other people). The purpose of this paper is to many of the input parameters, are geared select a few of these programs to discuss how more for solid state physics than molecular they work. Featured programs were picked physics/chemistry. It has to do with the fairly randomly from a list of available ones crystal structure of the material being with little bias save one. analyzed. The point symmetry group has to do with how many rotational symmetries Abinit [2]: exist, again this is geared more for crystal This software was developed using a structures. After this comes the real output. It popular formulation of DFT known as KS- gives the atomic positions in cartesian DFT (Kohn-Sham). It is currently run by coordinates, energy gradients, forces in eV/ Xavier Gonze and written in FORTRAN. Angstrom, eigenvalues of each band for each Dozens of variables are available for k point, minimum and maximum charge customizing the input conditions of a system. densities, a list of all the different energy A few of the basic variables are: acell, components, and the stress (average force per angdeg, ecut, natom, nband, and iscf; area) tensor in GPa. essentially every property that can change from one system to the next is customizable if Gaussian: desired. Keep in mind that most of these have This company’s program has had quite fairly generic default settings. acell gives the the history. It was originally developed by size of the unit cell usually in Borh radii noble laureate John Pople in 1971 at Carnegie (other units are allowed if you specify them), Mellon in FORTRAN. In the beginning the angdeg sets the angles between the three source was open and freely distributed to any primitive vectors for the unit cell, ecut sets a university that wanted to use it. This is far maximum value for the kinetic energy of the from the present case. The program is now system, natom is the number of atoms in a somewhat closed-source and sold by Hubartt 2 Gaussian Inc. under a fairly restrictive forces calculated at a given time-step with license. Various stipulations of the license are: DFT and then integrated via a Verlet not being allowed to disclose how fast the algorithm. The timescale for these dynamics program is, work on any “competing” is small, on order of picoseconds, but it is a program, or allow anyone that is banned from start. using Gaussian to use your copy. Who exactly For visualizations the output files can would be banned from using this? Pople (the be imported to a variety of additional creator!) is banned from using it, as are programs. These allow for 3D overlays of the Princeton and CalTech along with many other electron densities with molecular structures. universities and individuals. The official list One such program, gOpenMol [8] produced of bannings is kept at Fig 1. www.bannedbygaussian.org. As it is illegal to discuss how fast this program is, any judgement as to whether Gaussian is a good program that is worth the $4500-36,000 (academic and industry prices for OS X version) for a site license is difficult to determine. It is especially hard to justify this price when there are open-source programs that are freely available. This company has established a huge barrier of entry for anyone wishing to get into the field. With the risk of being banned from using this software a clear Fig. 1 Electron density of the aspirin molecule danger, it is hard to imagine anyone using this software. Despite this, many groups do use Setting up the input for QE is not Gaussian for their DFT calculations[7]. particularly difficult. The user specifies the element(s) being used, which pseudopotential Quantum Espresso (QE) [15]: to use, the Bravais lattice, the k-point QE is another open-source, free locations of the atoms in the 1st Brillouin software suite for DFT. It is operated through zone (locations of the atoms in the inverse a collaboration between MIT, Princeton lattice), and other characteristics of their (probably why they are banned from system. Many common lattices can be found Gaussian), and multiple European on their website or on a few other databases Universities; it is coordinated by Paolo that they refer the user to. Giannozzi. It is written in a combination of C++ and FORTRAN. C++ is used to make Current Biophysics Uses: the program object-oriented, very modular, The purpose of this paper was to and easier to learn initially with bit of describe a few of the many choices in FORTRAN for speed boosts.