Molecular Modeling of Raney Nickel Catalyst using Molecular Dynamics Simulations
Catalysts play a crucial role in a large variety of chemical processes and are of major commercial importance. A prominent example is Raney nickel (Ni)  which is a nanostructured, sponge-like catalyst used in many industrial processes and organic syntheses such as the reduction of benzene, the conversion of nitro compounds to amines, or for the desulfurization of thioacteals to hydrocarbons. Raney Ni is typically prepared by quenching a molten mixture of metallic nickel and aluminum (Al). The alloy structure obtained from the quenching process contains different NiAl alloy phases, from which aluminum is finally removed by the use of sodium hydroxide solution. The leaching process leads to the characteristic porous structures of the final catalyst. Apart from the preparation itself, the initial alloy composition can affect the performance of Raney Ni. Commercially, a 50-50 wt % nickel aluminum alloy is widely used.
In this case study, we will show how molecular modeling can be leveraged for studying the influence of the initial alloy composition on the final structure of Raney Ni and thus on its catalytic activity and performance. While previous studies [2-4] mainly focus on modeling the final catalyst structure, the workflow which will be presented in the following, allows to take the initial alloy composition into account. It will be shown how realistic model structures with random pore distribution can be created using molecular dynamics (MD) simulations.
NiAl3 is one of the NiAl phases formed during the quenching process and will be used as starting point for building an initial alloy structure. The crystal structure of NiAl3 was built according to the parameters as published in Ref.  using building tools of Scienomics MAPS software suit . Based on this unit cell, a 2x2x2 supercell was created and Al atoms were replaced randomly by Ni atoms in order to obtain an initial composition of 50 wt% Ni and 50 wt% Al (see Figure 1).
Since the experimentally observed pore diameter ranges between 20‑100 Å [7,8], it has to be ensured that the periodic unit cell is large enough to cover this pore size. Therefore, a 8x8x8 supercell with the cell parameters a = 106 Å, b = 118 Å, and c = 77 Å was generated containing more than 65,000 atoms. In order to avoid possible artifacts and to guarantee a statistically relevant sampling, five different initial structures have been created in this way. These structures were then subjected to a set of molecular dynamics simulations using the LAMMPS  module of Scienomics MAPS platform together with an alloy potential developed by Mishin  which was obtained from NIST Interatomic Potentials Repository . First, a 50 ps MD simulation in the NVT ensemble was performed at room temperature. Next, the models were equilibrated over 200 ps at 2000 K in the NPT ensemble which is well above the melting temperature of the intermetallic NiAl phases. Afterwards, the system was cooled down from 2000 K to room temperature over 1 ns using an NPT ensemble.
After the quenching step, the structural parameters of the five alloy models were analyzed and compared. A density of 4.4 g/cm3 was found for all five alloy structures which is within the experimental range of 3.95-4.76 g/cm3 for NiAl3 and Ni2Al3. Like for the density, the cell parameters were in a similar range for all structure. Therefore, only one representative model was used for the subsequent studies.
In the next step, aluminum was removed. Since it can be assumed that experimentally not all of the aluminum is leached out, two different fractions of remaining aluminum content have been considered for studying its influence on the pore formation: (a) with 0% aluminum remaining in the catalyst and (b) with 5% aluminum remaining. Since the removal of aluminum leads to a comparably large void volume, a stepwise procedure has been applied to allow for a decent equilibration of the porous structure. A short geometry relaxation over 5000 steps was carried out followed by a 200 ps equilibration in the NVT ensemble. From this trajectory, 10 snapshots were extracted and geometry optimized. These structures were used as starting points for 10 ns NVT MD simulations followed by additional 20 ns NPT MD simulations at 300 K. While the equilibration in the NVT ensemble enables pore formation, the volume was allowed to adjust to experimental conditions during the subsequent NPT run. In contrast to what one might have expected, the cell did not collapse during the NPT simulation despite the large void volume which clearly indicates that structures with stable pores have been generated.
The final structures have been analyzed with respect to the density and pore size and the average values over the 10 sets of simulations are summarized in Table 1.
Since the volume had been kept constant after removing aluminum, it can be expected that the density in the simulated structures will be smaller than the experimental one which is within a range of 3.5 to 7 g/cm3. After allowing the volume to adjust during the NPT sampling, a density of 4.1 and 3.9 g/cm3 has been obtained for the 0% and 5% Al structure, respectively, which is well in line with experimental data. The largest pore size diameter was determined using Zeo++ . For the 0% and 5% Al structure, a maximum pore size diameter of 43 and 49 Å, respectively was found after NPT equilibration which lies thus well within the experimental range. Though the 5% Al system contains more atoms, the density is lower compared to the 0% Al system which could indicate a stabilizing influence of Al on the porosity of the final catalyst.
In order to study the influence of the system size on the pore size, additional MD simulations were performed on 2x2x2 supercells. The supercell was built based on the unit cells obtained after the quenching step for both the 0% and 5% Al structure and consisted of more than 160,000 and 180,000 atoms, respectively. The supercells were first equilibrated over 10 ns in the NVT ensemble, and then for 10 ns in the NPT ensemble. The final structures are illustrated in Fig. 2.
The largest pore size diameter of the final supercell was determined as 87 Å and 69 Å for the 0% and 5% Al structure, respectively. The values are somewhat larger compared to the average values obtained for the original cell. The values doubled at most, though the volume of the supercell is eightfold larger compared to the original cell confirming the pursued approach and demonstrating that stable and realistic models of Raney Ni can be created.
In this study, we have presented a simulation approach for modeling the porous structure of Raney Ni catalytic systems using molecular dynamics simulations. The analyzed structural parameters agree well with experimental data demonstrating that realistic model structures can generated starting from the alloy precursor. More details and further information are provided in Ref. . The presented approach is not limited to Raney Ni-type systems, but is generally applicable for studying nanoporous systems.
Scienomics gratefully acknowledges the computing resources provided on Blues and Fusion, high-performance computing cluster operated by the Laboratory Computing Resource Center at Argonne National Laboratory (ANL). We would like to thank our collaborator John J. Low (ANL) and Raymond Bair (ANL) at Argonne National Lab for providing HPC resources for this work.
- Raney, M. 1925, US Patent 1,563,787.
- Krivtsov, A.M. Meccanica 2003, 38, 61-70.
- Mrozek, M. Computer Methods In Materials Science 2014, 14, 27-36.
- Shi, Y. J. Chem. Phys. 2008, 128, 234707.
- Bradley, A.J.; Taylor, A. Philos. Mag. Series 7 1937, 23, 1049-1067.
- Scienomics; MAPS platform, Version 3.4.1, 2014, France.
- Freel, J.; Pieters, W.J.M.; Anderson, R.B. J. Catal. 1969, 14, 247-256.
- Zeifert, B.; Blásquez, J.S.; Moreno, G.C.; Calderón, H.A. Rev. Adv. Mater. Sci. 2008, 18, 632-638.
- Plimpton, S. J. Comput. Phys. 1995, 117, 1-19.
- Mishin, Y. Acta Mater. 2004, 52, 1451-1467.
- Becker, C.A.; Tavazza, F.; Trautt, Z.T.; Buarque de Macedo, R.A. Curr. Opin. Solid State Mater. Sci. 2013, 17, 277-283.
- Martin, R.L.; Smit, B.; Haranczyk, M. J. Chem. Inf. Model. 2012, 52, 308-318.
- Schweizer, S.; Chaudret, R.; Low, J.; Subramanian, L. Comput. Mater. Sci., 2015, 99, 336-342.