Noura Dawass

5 100 P ROPERTIES OF U REA -C HOLINE C HLORIDE M IXTURES Table 5.1: Themole fractions, number of molecules, and initial box lengths of all simulated systems Small System Large System N ChCl / N Urea x Urea N ChCl N Urea L S / [Å] N ChCl N Urea L L / [Å] 2/1 0.20 160 80 34.7 800 400 59.4 3/2 0.25 144 96 34.0 720 480 58.1 1/1 0.33 120 120 32.8 600 600 56.2 5/7 0.41 100 140 31.8 500 700 54.4 1/2 0.50 80 160 30.7 400 800 52.6 1/3 0.60 60 180 29.6 300 900 50.6 1/4 0.67 48 192 28.8 240 960 49.3 1/5 0.71 40 200 28.3 200 1000 48.4 each mole fraction, we considered a small and a large system consisting of 240 and 1200 molecules, respectively. The former was used for computing transport properties (i.e. viscosity, self-diffusivity, MS diffusion), while the latter was used for computing RDFs required for KBIs. It is important to note that larger system sizes were essential to obtain a sufficiently large linear regime of the scaling of KBIs [63] . Table 5.1 shows the number of molecules and the size of the simula- tion box at eachmolar ratio of ChCl to urea for all systems simulated in this work. Initial configurations were generated by randomly inserting molecules in a cubic simulation box using the PACKMOL package [165] . The generated sim- ulation boxes were first relaxed using the conjugate gradient method for 10000 steps. The energy minimization was followed by MD runs in the isothermal- isobaric ( NPT ) ensemble at 343.15 K and 1 atm for 10 ns. In the NPT ensem- ble, average volumes and densities were computed. Starting from the average density obtained from NPT runs, each system was then equilibrated at 343.15 K and 1 atm for 1 ns in the canonical ( NV T ) ensemble. Consecutively, transport properties were computed in the next 120 ns. For small systems (see Table 5.1) , a 120 ns run typically takes 96 hours using 24 CPUs. During production runs, the OCTP (On-the-fly Computation of Transport Properties) plugin in LAMMPS was used to compute transport properties [154] .The OCTP plugin uses Einstein relations combined with the order- n algorithm [22, 166] . For more details about the OCTP plugin, readers are referred to the original study by Jamali et al. [154] . RDFs were computed from separate MD simulations of 80 ns of the large systems in the NPT ensemble. For large systems (see Table 5.1) , a 80 ns run typically takes 720 hours using 24 CPUs. Finite-size effects of the reported RDFs are cor-