Ionic Relaxation in This allows for employing classical mechanics
DFT instead of quantum mechanics as the nuclei rather
behave like a point particle. The force acting on the Simulation of Semiconductor Properties I-th nuclei can then be expressed as When simulating quantum systems, often single ∂U 𝑭𝐼 = − . atoms are placed by hand in specified positions. ∂𝑹𝐼 The resulting constellation does not necessarily The potential 𝑈(𝑹1 , … , 𝑹𝑀 ) can be thought of a so- represent a stable system, as the atoms influence called potential energy surface as depicted in fig.2. one another, which could cause them to move. To achieve a stable system, the equilibrium positions of these atoms, hence the positions where all forces on every particle vanish completely, have to be computed. When using density functional theory, the equilibrium is first found for the electrons for specific nucleus constellation via the self- consistent loop. After convergence, the nuclei of the atoms might still experience forces, either of repelling nature from the other nuclei or of attracting nature from the electrons. If these forces are still present, the system is in need of finding new positions for the nuclei, in other words the Figure 2: Potential energy surface. system has to be relaxed. The goal is to find the positions 𝑹1 … 𝑹𝑀 such that Mathematical Foundation the energy is at a minimum or in other words the forces vanish. Searching for an atomic constellation where all In general, this can be regarded as an optimization forces vanish, can also be thought of as searching problem for which reason this procedure is often for a global minimum for the energy of the system. also called geometry optimization. The overall potential of the nuclei can be expressed By adjusting the positions of the atoms, a global as 1 𝑍𝐼 𝑍𝐽 minimum for the energy can be found. 𝑈(𝑹1 , … , 𝑹𝑀 ) = ∑ + 𝐸(𝑹1 … 𝑹𝑀 ), This, however, is not always trivial, especially 2 |𝑅𝐼 − 𝑅𝐽 | when the energy function has local minima as well. 𝐼≠𝐽 where 𝐸(𝑹1 … 𝑹𝑀 ) is the total energy of the In a local minimum, the gradient (forces) would electrons when the nuclei are held in place in the vanish, even though the global minimum is not positions 𝑹1 … 𝑹𝑀 . reached yet. Therefore, optimization is a very Another simplification can be achieved by taking a demanding task, for which a lot of approaches have look at the quantum mechanical behavior of nuclei. been developed. When looking at fig. 1, it can be noticed that the wave function of the nucleus is very localized Optimization Algorithms compared to the one of the electron. There exist several different optimization algorithms, for ionic relaxation mainly steepest descent and conjugate gradients are being used as an approach for finding the equilibrium constellation. Quantum Espresso for example mainly uses a Quasi-Newton method. The commercially available software VASP uses a Quasi-Newton approach, however of different type, when compared to Quantum Espresso. VASP features a conjugate gradient method as well. Figure 1: Ground state wave functions of an electron and nucleus in a parabolic potential. 1 Another approach for finding the equilibrium of a Example in Quantum Espresso quantum system is called Damped Molecular Dynamics. In this approach the forces for each In Quantum Espresso the relaxed structure can be particle are calculated individually. The resulting calculated by modifying the input file. A changes in positions can be calculated via corresponding input file could be the following: &control Newton’s equation of motion calculation = ’relax', 𝑑 2 𝑹𝑰 nstep=50, 𝑀 = 𝑭𝑰 etot_conv_thr = 1.d-4 𝑑𝑡 2 forc_conv_thr = 1.d-3 In order to reach an energy minimum, an additional ... friction term is added, which leads to the following / equation: ... 𝑑 2 𝑹𝑰 2𝑀 𝑑𝑹𝑰 &electrons 𝑀 2 = 𝑭𝑰 − . conv_thr = 1.d-7 𝑑𝑡 𝜏 𝑑𝑡 ... / This equation can be solved numerically for the &ions ion_dynamics = "bfgs" positions by employing the so called Verlet pot_extrapolation = "second_order", algorithm. wfc_extrapolation = "second_order", Both VASP and Quantum Espresso feature this upscale = 100, method. / ...
Integration in the DFT Algorithm
Figure 4: Example input file for Quantum Espresso to compute After computing a relaxed system, the solutions for relaxed structure. the wave functions of the electrons might have to be adapted due to the changed positions of the The parameter calculation is specified as relax, nuclei. Therefore, the self-consistent loop of the which indicates the computation of the relaxed DFT is entered once again. Now, the wave system. The next line specifies the number of functions are being updated based on the new iterations for the structural optimization with the positions of the atoms. This procedure is shown as parameter nstep. After that, the different a flow graph in fig.3: convergence thresholds for the energy and forces are defined. If the total energy of the system changes less than the value specified in etot_conv_thr and the overall forces are smaller than the value specified in forc_conv_thr the algorithm terminates. After the convergence threshold for the self-consistent loop is set by conv_thr, the optimization algorithm is specified with the parameter ion_dynamics. In this example it is set to bfgs which indicates the use of the BFGS Quasi-Newton algorithm. If set to damp, Quantum Espresso uses the previously mentioned Damped Molecular Dynamics method to find an equilibrium.