Mentor: Dr. Afsaneh Doryab

Client:  Dr. Zachary Ulissi, Assistant Professor, Chemical Engineering, Carnegie Mellon University

Description: 
Rapid developments in solar and wind power are pushing a transition in the energy economy from fossil fuels to renewable energy sources.  In many areas, the cost of new solar installations is at or below the cost of electricity generated from existing fossil fuel power plants such as coal or oil. Unfortunately, de-carbonizing the rest of the economy is a major challenge.  Consumer products such as plastics, lubricants, specialty chemicals, among many others require small hydrocarbon building blocks that are primarily derived from oil. To meet this challenge, we need new ways to take cheap renewable electricity and generate chemicals with societal value, such as hydrogen (H2), oxygen (O2), CO (poisonous on its own, but valuable as a feedstock chemical), and ethanol (alcohol, as a fuel additive or as a building block). 
Electrochemistry uses an applied electric potential to drive chemical reactions in aqueous (water-based) environments.  The electricity can come from either renewable or non-renewable sources.  Examples of electrochemical devices include fuel cells for vehicles (oxygen and hydrogen form water yielding electricity), water splitting devices (water forms hydrogen and oxygen, which can be stored for electricity), and CO2 reduction (taking CO2 and converting to methane, ethanol, or higher hydrocarbons).  All of these reactions require more efficient electrochemical catalysts, materials on the electrode that make the reactions faster.  New catalysts can increase the rate of reaction, reduce the energy cost, and change the selectivity (which reactions happen when several are competing). 
Most catalysts are formed from various combinations of transition metals (silver, gold, iron, platinum, iridium, copper, etc), which are active due to their d electron orbitals.  By studying in great detail a small number of these catalysts for various reactions, we have relatively simple descriptors that can predict how active a particular surface will be.  For example, how tightly a single hydrogen atom binds to a particular metal surface is an excellent predictor of how active the surface will be for the hydrogen evolution reaction (H+ + e- → H2 gas). Our group is developing high-throughput methods to calculate these descriptors using quantum chemistry theoretical methods and add them to a central database.  This process takes roughly 1 day on a high-performance computing node.  An example of the current database of descriptors can be found at http://gilgamesh.cheme.cmu.edu:5000/.  The database is currently hosted as a Mongo database, with separate collections for the existing calculations and for a catalog of all possible surfaces that could be made (the design space).
The scientific value of this database would increase greatly with a more focused interface to expose existing calculations, online visualization of the results for various reaction classes, and predictions of the most active surfaces to be studied in more detail.  This will enable better catalyst predictions and accelerate the rate of materials discovery. You will be able to work closely with a current PhD student with knowledge of the database structure and calculation methods. The final project (and exposed data) will be open and public.

Deliverables:
1) An improved database front-end that exposes mongo-style queries to find existing surfaces with properties of interest.  Visualization of entries with a javascript applet such as jsmol.  Could be implemented with a flask/python/mongo framework such as Flamyngo.
2) For a number of standard reactions with known ideal descriptors, provide visualizations of the current best and predicted surfaces as a volcano plot. Could be implemented with a framework such as plotly or Bokeh (similar to Shiny/R).
3) Given criteria of an ideal surface from the user (for example, an ideal hydrogen adsorption energy of approximately -0.2 eV) and a trained python-based sklearn model that can predict activity based on material structure, visualize the results, including both known (calculated) data points and predicted activities. 
4) Interacting to interact with the data in all visualizations - for example, provide mouse-over details about a data point (configuration, material, structure, etc), or on clicking take the user to the database entry visualization from (1).