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: Marton Voros Applicant's institution: ANL Applicant's division: MSD Project Name: reduceco2 Project title: Kinetics of CO2 reduction of clusters Associated funding: DOE BES AC0206CH11357 Other Systems: NERSC, 1.000.000 (project on nanoparticle solar cells) Science: In this project, we will investigate the kinetics of charge transfer for the electrochemical reduction of CO2 to CO on supported clusters. Efficiently and cheaply generating fuels using reducing CO2 to hydrocarbons is a prime target in the search for renewable energy materials and processes. A variety of catalysts have been discovered, and the catalytic properties have been elucidated using atomistic density functional theory techniques. The most common theory of the atomistic mechanism behind the first catalytic steps of reducing CO2 to CO is based on coupled electron and proton transfer on the surface of the catalyst yielding COOH. This scenario ignores the kinetics of electron transfer. Indeed, this theory is unable to explain recent experimental findings on transition metal clusters that showed that CO2 reduction is about 100 times more active than those reported on MoS2 nanoflakes (Science 353, 467 (2016)). Our primary goal in this project is to use our recently developed constrained density functional theory (CDFT) code (J. Chem. Theory Comput. 13, 2581 (2017)) to compute the electron transfer kinetics from transition metal clusters to CO2 and compare the performance of transition metal clusters to other known catalysts, such as surface of Cu or edges of MoS2 flakes. Finally, using our kinetic insights, we will rationalize recent experimental findings. We expect that the insights gained by computing the charge transfer kinetics, which are inaccessible in standard DFT calculations, will be imperative to design improved CO2 electrocatalysts. Project description: We will be using first principles methods based on density functional theory. In particular, for simulating the charge transfer from the transition metal clusters to CO2 we will be using the plane wave code Quantum Espresso (QE). QE can make use of efficient linear algebra (Lapack, BLAS, Scalapack, BLACS) and FFT libraries if they are available. QE was shown to efficiently scale to a few hundreds of processors in the case of non-hybrid functional calculations. Hybrid functional calculations can scale up to a few thousands of cores, especially because of recent methodological and algorithmic development (J. Chem. Theory Comput., 2012, 2242 (2016)). NERSC staff has recently restructured the algorithm behind hybrid functionals in Quantum-Espresso and has shown that this method is capable of making use of the new Cori KNL nodes due to its hierarchical, multi-level parallelization strategy that involves using several layers of MPI and OpenMP parallelizati on (e.g. http://www.nersc.gov/users/computational-systems/cori/application-porting-an...). We expect the same or similar benefits when using the KNL nodes of the new Bebop cluster for the hybrid functional calculations, however, we know from our experience that Quantum-Espresso also runs fine on the Blues cluster and we expect it can also nicely run on the broadwell nodes. Indeed, our initial tests on small systems during the early user access program showed good performance. Since our preliminary tests showed that hybrid functional calculations are necessary to get physically meaningful results, we will plan to exclusively use 100% of our allocation on hybrid functional calculations. Further, for the time being, we will neglect the effect of the support (in the experiments it is glassy carbon or reduced graphene oxide), although we will plan to estimate its importance using our recently developed computationally efficient technique that includes a model electrode through electrostatic interactions (Chem. Mater. 29, 1255 (2017)). To simulate charge transfer processes, we will use our own implementation of constrained DFT (CDFT) to localize charges on either the catalyst or CO2 and then to compute all the parameters (including electronic coupling, reorganization energy, driving force) entering semi-empirical charge transfer models, such as Marcus-theory. Our own implementation of CDFT is integrated into a private version of the most recent release of Quantum-Espresso and it inherits its performance, although typical CDFT calculations are about one order of magnitude more expensive since they require a double self-consistent cycle instead of the regular single self-consistent cycles. We will focus on five different catalysts: W4, W3MO, W2MO2, WMo3 and Mo4, since most of these are available to our experimental collaborators. Having found the ground state geometry (this will take a negligibly small amount of time), we will then simulate charge transfer from these clusters to CO2 in the gas phase. We will investigate the charge transfer kinetics by computing the electronic coupling as a function of distance and also the angle of CO2 since it is known that reduced CO2 (anion radical) traps the excess electron by bending. We estimate to use a coarse sampling of 8 angles and 8 distances for each of the systems. Altogether we plan to run 64 constrained DFT calculations for each system, which amounts to 320 calculations. Given that each CDFT calculation requires a "donor" and "acceptor" calculation, this means 640 double self-consistent hybrid functional calculations. Furthermore, to compute the reorganization energies and driving forces we will have to optimize the structure in the presence of the constraints which amounts to about 20 regular CDFT steps per cluster per site. Our preliminary estimate suggests that we can use about 128 cores (assuming KNL nodes) per calculation and each calculation would take about two hours. Finally, we will compare our results to other known catalysts, such as Cu clusters and MoS2 nanoflakes. We will take literature structures on Cu3, Cu4 (J. Am. Chem. Soc. 137, 8676 (2015)), which will amount to 2/5 of the time estimated for the calculations on the WMo clusters. Finally, we will use a recently published cluster model of MoS2 to simulate CO2 reduction on MoS2 nanoflakes (J. Am. Chem. Soc. 137, 6692 (2015)). Since this cluster is about twice as large and our calculations scale as N^4, where N is the characteristic system size, we estimate to spend almost as much time simulating this one cluster than for all the Cu and WMo clusters together. All in all, we request 440.000 core hours for the entire year evenly distributed. For the reasons discussed above, we believe that our hybrid functional calculations can really benefit from using the KNL nodes, however, we can definitely use the broadwell nodes efficiently and we had a successful experience in using Quantum-Espresso of Blues. Industry partnership: Project URL: Requested allocation: 440000 Q1: 110000 Q2: 110000 Q3: 110000 Q4: 110000 Justification: Storage requirements: 1 TB. We will need to, at least temporarily, save wave functions and charge densities to enable efficient restarts of QE calculations. Since these quantities are saved on grids and can be several tens of GBs in size, we estimate we will need about 1 TB of storage. 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