Summary:  NSF support permitted our team to install a 34 node cluster in the first half-year of the project. The new resource, Hawk, has theoretical performances of 69.2 TFLOPS (CPU) and 8.1 TFLOPS (GPU); CPU performance is 59% higher than originally proposed due to market evolution; GPU performance is just above what was proposed. Collectively, there are 1752 CPUs and 32 GPUs providing 15,347,520 Service Units (SUs) annually. The seven storage nodes acquired provide 214 TB of usable Ceph storage and 11.3 TB of usable CephFS storage.  Hawk was made available to the proposal team in December 2020, and to the general Lehigh research community one month later. Access to Open Science Grid was enabled in April 2021 at the proposed sharing level (20%).  

Lehigh researchers gain access to Hawk via an allocation request. A faculty advisory committee reviews requests and interacts with the relevant user(s) to advance them to approval. The primary goal is to get researchers access quickly, so close guidance is provided as needed.  A secondary goal is to prepare researchers to make successful allocation requests to national computational facilities such as ACCESS, if local resources become insufficient for their work.  To date, approximately 100 separate Hawk allocation requests have been processed with 100% of them resulting in access.

Intellectual Merit:  Research enabled by Hawk positively impacted a range of science drivers that were described in our proposed work; some of the scholarly productivity made possible by this resource is highlighted below (also in Images).

Alkane dehydrogenation using Density Functional Theory: Computations at the quantum mechanical level of theory and resulting data obtained from Hawk elucidated structure reactivity relationships for hydrogenation/dehydrogenation reactions on molybdenum sulfide, ethylene oxidation on silver catalyst, alkane activation on sulfur-tolerant catalysts, and transfer hydrogenation. These chemistries are particularly important in converting shale gas and biomass, two carbon rich sources that are plentifully available in the US, to fuels and value-added chemicals. 

Deep Learning:  Access to Hawk made possible the advancement of machine learning, including deep learning, based studies for reaction modeling, molecular property prediction, and molecule design.  Topics included reaction pathways for polyolefin upcycling, mechanisms for oxidative coupling of methane, and identification of liquid organic hydrogen carriers.  While each of these studies had impact as evidenced by publications in their fields, it is also the collective expansion of machine learning techniques to materials exploration that promises continuing impact.

Design of Organic Solar Cells:  Hawk enabled computations assisted with identifying the mechanisms whereby different processing parameters, such as the molecular weight of semiconducting polymers, affect the efficiency of solar cells. These computations can guide re-engineering of the fabrication conditions to optimize performance and cost of organic solar cell production. This outcome represents an important impact on solar materials research. 

Singlet Fission in Solar Cells:  Hawk has enabled us to expand our models of radical photophysics to both large organic molecules and inorganic chromophores. We have developed two new models which allow for the direct calculation of charge in organic single crystals using either hopping or band transport. We are currently combining these two methods to directly predict singlet fission and propagation of charge through molecular materials that have potential to significantly improve solar cell efficiency.

Molecular Flow Sensors in Blood:  Massively parallel but independent simulations made possible by Hawk allowed us to advance time-scale bridging methods that permit us to make predictions about blood protein unfolding kinetics at times far beyond what is accessible in coarse-grained molecular simulations - time scales that are nonetheless relevant to pathological protein behavior.  This is allowing us to extend this body of research to have more clinical impact around the diagnosis and treatment of certain bleeding disorders, which affect approximately 2% of the population.

Broader Impacts:  A Hawk scholarly productivity site was created to publicize this grant's impact (a link to the site is below).  Numbers that follow are approximates, because they are only updated by project PIs when allocations end or renew.  Hawk has enabled some portion of research described in over 40 journal articles and proceedings and over 60 presentations at professional society meetings (~20% of which were invited).  Hawk supports the dissertation and thesis research for dozens of Doctoral and Masters students, including some who have now completed their degrees. Hawk partly enabled the formation of new research teams across the university, including multiple engineering disciplines, physics, chemistry, biology, management, business, mathematics, accounting, and economics; Hawk also supports research being conducted by Lehigh's Energy Research Center.  Preliminary data generated on this resource supported submission of 14 proposals, 4 of which were supported.  Five training workshops and eight courses or related academic experiences were supported by Hawk, affecting hundreds of students, staff, and faculty.  This is all in conjunction with our ongoing 20% resource share to the Open Science Grid.

https://confluence.cc.lehigh.edu/x/2oPSCQ

Last Modified: 05/03/2023
Modified by: Edmund B Webb Iii