Hello, A new project on the LCRC cluster has been requested. Please forward the information on to the LCRC Allocation sub-committee. Applicant's name: John J. Low Applicant's institution: ANL Applicant's division: MCS Project Name: Raney_Nickel Project title: A Molecular Model of Raney Ni Associated funding: None Other Systems: None Science: Raney nickel is a nanostructured catalyst which is used in a variety of industrial processes and organic syntheses such as hydrogenation reactions. The basis forms a NiAl alloy which is transformed into the active form by leaching out aluminum which leads to the characteristic porous structure of the final catalyst. The porosity depends on the initial composition of the NiAl alloy which is typically prepared by melting nickel together with aluminum and subsequent quenching. The quenching process induces the formation of different NiAl phases which react differently with the caustic and thus affect the pore formation in the final catalyst. Experimental studies indicate that the sponge-type structure of Raney Nickel grains is formed by a collection of crystallites adhering together. Since the structure of both the precursor alloy and the active form of Raney Nickel is determining the catalytic activity and performance, it is crucial to gain a detail ed understanding of the structural properties at the atomistic level. The goal of this project is to leverage molecular modeling to provide critical insights into how the composition of the precursor alloy affects the porosity and thus activity of Raney Nickel. Project description: For this purpose, we will investigate the formation of NiAl nanostructures in the initial alloy using molecular dynamics. The initial NiAl compositions can range from 40 to 70 wt% Al 2, , . Freel at al.2 determined the mean pore diameter to be in a range between 2.5 to 10 nm. Based on these data, we will create several model structures using Scienomics MAPS platform . We will start from a NiAl3 unit cell, build a 3x3x3 super cell and create model structures for four different compositions (40/50/60/70 wt% Al) by randomly replacing Ni and Al in the corresponding fractions. In order to ensure a statistically relevant sampling, five initial starting configurations will be generated for each of these model structures. Based on these model structures, large super cells with cell lengths between 7 - 14 nm will be built and minimized using the embedded atom method (EAM) as implemented in LAMMPS. Each of these 20 model structures will be equilibrated sufficiently long at 2000 K, which is well above the melting temperature of the intermetallic NiAl phases, and then cooled down to room temperature following the procedure as described by Noya et al. The computational time needed for performing MD simulations of all 20 models will be considerable. Each model structure contains between 35,000-150,000 atoms. According to the LAMMPS Cu benchmarks , 100 steps for a system size of 32,000 atoms would take 0.3 s (fixed CPU time) on 512 cores. A 1 ns trajectory (1,000,000 steps) for 32,000 atoms would take 500 core-hours and a 150,000 atom system about 2500 core-hours. 45,000 core-hours will be needed for the simulations of the 20 model structures. We will determine how to get the best throughput for these calculations by determining the optimum scaling of LAMMPS as a function of number of MPI processes and OpenMP threads or GPU nodes on Fusion. The open source program RINGS (Rigorous Investigation of Networks Generated Using Simulations - http://rings-code.sourceforge.net) will be used to analyse local structural environments, voids and simulated X-ray and Neutron Diffraction patterns. The trajectories will be analyzed with respect to the formation of nanostructured clusters. From each model, Al will be removed and only 5-10 % Al will be left which corresponds to the experimentally observed final catalyst composition. These porous models will be minimized, which will only require a few hundred core-hours, and afterwards analyzed with regard to pore size, pore distribution, and accessibility from surface for the various compositions and initial configurations to explore trends that can be linked to the catalytic performance. In this way, we can gain a fundamental understanding about the formation of nanostructures in the precursor alloy in dependence of the initial composition. In the laboratory, the precursor is activated though treatment with sodium hydroxide leaching out the majority of the Al and leading to structural re-arrangements of the Ni bulk. This process will not be simulated as it is beyond the scope of the present project. Instead, we will focus on modeling the structure of the final catalyst. To provide a reliable description of the nanoporous structure, large system sizes will be required which will also allow to capture effects of grain boundaries that a supposed to be important for the chemistry of the final catalyst. The basis for approaching the structure of the active from are the characterization studies of Raney Nickel2, , , by Anderson’s group and others which yield a qualitative picture of Raney Nickel. The void volume of these particles ranges from 0.05 to 0.15 cc/g.1 X-ray diffraction3 and X-ray Absorption Spectra show that local structure of Raney Ni’s is similar to FCC Ni. Fouilloux gave a qualitative description of Raney Nickel as a spongy material composed of 100 nm nanoagglomerates of smaller 2.5 to 15 nm nanocrystals.3 Several modeling studies of grain boundaries use molecular dynamics to generate structures. Van Swygenhoven and Caro used a stochastic procedure that starts with randomly placed seed particles with the rest of the space filled with a Voroni construction. Keblinski randomly placed seed particles in a liquid metal and then quenched the melt to generate a model crystal with grain boundaries. These procedures yield a model of a solid composed of nanocrystalline grains. We will follow the spirit of these modeling approaches by randomly placing seeds of Ni crystallites in a gas (or liquid) of Ni atoms and use simulated annealing to generate a low energy structure. Raney Nickel, which has a void volume of 0.1 cc/g, is essentially half void by volume. A simple model would be to seed a gas with a density half that of Ni metal. The gas will condense on the seed crystallites and create an agglomerate with the appropriate void volume. Since the condensation is exothermic, a Nose or Andersen thermostat will keep the temperature below the melting temperature (1728K) of Ni during the simulation. Starting at 1500K and scaling the temperature to 0K over 100 picoseconds will quench to a low energy structure through simulated annealing. Since the low energy structure will create particles which have the density of Ni metal, the voids between the particles should have half the volume of the initial volume. The number of crystallites should control t he size of the nanocrystals formed during simulated annealing. A careful choice of initial conditions should lead to the desired size of nanocrystals and void volumes. The RINGS program will allow a systematic and analytical analysis of these structures.9 A 100 nm droplet of Ni with half the density of bulk Ni is a model of Fouilloux’s description of Raney Nickel.6 The scale of this calculation would be significant. The model would be initiated by randomly placing 500 seed crystallites (Ni55 clusters) in a hundred nanometer sphere and then randomly inserting 25 million atoms into the space between the seeds. The embedded atom method as implemented in LAMMPS can handle simulated annealing of models of this size. The Cu EAM benchmark on the LAMMPS website shows that 100 steps of molecular dynamics for 16 million atoms could be done on 512 cores of a Xeon/Myrinet cluster in 21 seconds.8 A 100 picosecond trajectory (100,000 steps) would take 60 hours (30,000 core-hours). Fusion or Blues have faster processors and an InfiniBand interconnect that will yield better performance. The estimated CPU requirements derived from timings for the Xeon/Myrinet cluster is an upper bound. Our initial calculations will use a periodic model containing a million atoms and 20 seeds to verify this approach will give morphology similar to Raney Nickel. The test will involve comparing calculated properties from the predicted structure to the experimentally determined pore size distribution, heterogeneities from small angle X-ray scattering and line broadening in X-ray diffraction.11,13 If this model does not yield a satisfactory result we will implement and test a stochastic Voroni procedure.15 These calculations will require a few thousand core hours and will not exhaust our allocation before developing a working model. Once we have established a procedure that generates a model of a porous solid composed of nanocrystalline grains, we will carry out calculations on a more realistic 50 million atom droplet model of Raney Ni. 30,000 core hours will be needed for these experiments with smaller periodic models. We will use EAM to model hydrogen adsorption on our model of Raney Nickel. The model will have monolayer coverage (one hydrogen atom per surface Ni atom). The surface will be defined by Ni atoms with coordination less than twelve. Simulated annealing will be use to find a low energy distribution of hydrogen on the surface of Raney Ni. The predicted binding energies, vibrational frequencies and geometries will be used to classify the chemisorption sites to a smaller number of representative sites. 30,000 core-hours will be required for the chemisorption models. The pore size, pore distribution, and the accessibility of the droplet model and chemisorbed model will be compared with the results obtained for the precursor models. This strategy will allow to explore to what extent the nanostructural agglomerations in the precursor are reflected in the porous active form. The generalized chemisorption sites will form the basis of smaller periodic models. We will use the smaller models to model hydrogenation of benzene on Raney Nickel with an electronic structure program like abinit, bigdft, cp2k or qbox. These calculations will be done in a future project. Overall, 105,000 core-hours will be needed over a period of five months for performing the in silico experiments for the project. Three people will be members of this project. Sabine Schweizer, Lalitha Subramanina (from Scienomics, more details below) and John J. Low from Argonne. Dr. Sabine Schweizer, Senior Scientist at Scienomics has a Ph. D. in Theoretical Chemistry from University of Tuebingen, Germany. Her work experience is in the area of applying quantum chemical methods to large and small molecules. Her experience is particularly useful in the areas of homogenous and heterogenous catalysis, where energetics and dynamics play an important role in the reactions. She has experience working with multiscale hybrid methods combining quantum chemical and classical methods which makes her particularly versatile in applying Scienomics’ multiscale, multiparadigm MAPS platform to study hard materials such as surfaces, coatings, semiconductors, solar cells, etc. Dr. Lalitha Subramanian, Chief Scientific Officer and VP of Services has 18 years of experience providing insight into chemical systems and processes that are of critical interest to industry. Following her Ph. D. in Chemistry, her post-doctoral work was with Prof. Roald Hoffmann (Nobel Laureate) at Cornell University. She has been a leading architect of solutions in the areas of alternate energy, catalysis, materials design, and process optimization. She has worked on numerous proprietary projects for diverse companies and in this pursuit; she has delivered product enhancements, process optimization and cost savings to her customers. Lalitha maintains a broad range of partnerships in oil & gas, chemical, personal care, materials, semiconductors, automotive, aerospace, and pharmaceutical industries. She has co-authored a book on software techniques used in Materials Science published by CRC Press, 2005. She continues to present invited lectures and has numerous scientifi c publications. Project URL: Requested allocation: 105000 Q1: 0 Q2: 0 Q3: 55000 Q4: 50000 Justification: The requester has used undetermined amount hours of their initial startup project. In addition to approving an initial amount, please specify a Category and Subcategory for this project. For a list of the current selection of approved categories, please see: https://wiki.lcrc.anl.gov/wiki/Processes/Categories Once the Allocation committee has approved the project, please go to the Project Management page to create it: https://accounts.lcrc.anl.gov/projects.php Thank You, The LCRC Accounts System