This document discusses the formation of an institute called S2I2C2M2 that aims to promote more sustainable software development practices for computational chemistry. It outlines challenges with current software, including complexity, lack of parallelism, and outdated development practices. The institute will focus on developing portable parallel infrastructures, general tensor algebra algorithms, standards for code and data sharing, and education initiatives to train the next generation of computational chemists. The goal is to overcome barriers and change the culture around computational chemistry software to better support collaborative open science.

1. SUSTAINABLE SOFTWARE FOR
COMPUTATIONAL CHEMISTRY
AND MATERIALS MODELING
Beverly Sanders, University of Florida

2. Outline
 Overview
 Challenges and Current State
 Overcoming the Barriers—technical and cultural

3. Computational Chemistry
 Long history – over 50 years
 Underpins broad array of scientific applications—Grand
Challenges
 Efficient combustion systems
 Drug design
 Understanding biological systems
 Semiconductor design
 Water sustainability
 CO2 Sequestration
 Efficient lighting (quantum dots)
…
 Full partner with experiments—”computational experiments”
may be more reliable than lab measurements

4. Scientific Software Innovation Institute for Computational
Chemistry and Materials Modeling -- S2I2C2M2
 Collaboration between computational chemists,
computer scientists, applied mathematicians, and
computer engineers
 Goal: overcome obstacles of algorithms and culture
and change the nature of computational chemistry
software development.
 Year long conceptualization phase has been funded by
NSF
 First meeting scheduled for January 2013

5. People
Daniel Crawford (Virginia Tech) Vijay Pande (Stanford)
Robert Harrison (Tennessee, ORNL) Manish Parashar (Rutgers)
Anna I. Krylov (U.S.C.) Ram Ramanujam (LSU)
Theresa Windus (Iowa State) Beverly Sanders (Florida)
Emily Carter (Princeton) Bernhard Schlegel (Wayne State)
Edmund Chow (Georgia Tech) David Sherrill (Georgia Tech)
Erik Deumens (Florida) Lyudmila Slipchenko (Purdue)
Mark Gordon (Iowa State) Masha Sosonkina (Iowa State)
Martin Head-Gordon (Berkeley) Edward Valeev (Virginia Tech)
Todd Martinez (Stanford), Ross Walker (San Diego
David McDowell (Georgia Tech) Supercomputing Center)
+ others

6. Current State of Computational
Chemistry
 Long history--legacies of modern molecular
dynamics and quantum chemistry packages span
decades
 Both open source and commercial
 Amalgam of programming languages
 Domain specific methods
 Multi-dimensional integral engines
 General purpose
 Davidson method for computing eigenvalues of large
matrices (ranks in tens of billions)

7. Software is extremely complex
 Example: Modern ab initio quantum chemistry
simulations
 Computations scale as O(N7) or higher
 Where N represents size of molecular system (number of atoms,
electrons, or basis functions)
 Code complexity arises naturally from problems, but
 is an obstacle to long-term sustainability
 is an obstacle to exploitation of (ever changing) HPC
hardware
 hinders education of next generation of scientists
 only a handful of very senior students can make a contribution

8.  Much recent software development focused on
exploiting parallel architectures
 Varying degrees of success
 With a few exceptions, still not fully exploiting
available systems
 Utilizing exascale will require rethinking of approach
 Desperate need for tools to generate high
performance massively parallel code from high
level specifications

9. Developers
 Mostly grad students and post-docs
 Training in software engineering left to individual
research groups
 Extent to which this is done varies
 Large burden for small groups: community approach
has potential benefits for both software and the
students
 Education tends to be narrow: students learn about
software their advisors are involved with

10. Science Drivers
 Catalysis
 Catalysts facilitate control of chemical reactions by raising rates
that chemical bonds are formed or broken
 Improve selectivity and control over unwanted byproducts
 Decreased energy consumption
 Reduction of waste stream
 Rational design of catalysts for a specific application is one of
the Holy Grails of of modern chemistry and chemical engineering
 Requires quantitative information about transition states
 Intermediates low concentration and short lifetimes—thwart
experimental evaluation
 Will require state-of-the-art computation combined with
experiements

11. Science Drivers
 Organic photovoltaic cells
 Potential applications: thin-film transistors, LEDs, solar
cells, optical switches
 Advantages
 Devicescan be flexible
 Inexpensive to produce
 Limitations
 Reduced power conversion energy
 But,process leading to current generation not well
understood

12. Overcoming the Barriers
 1 year conceptualization phase funded by NSF
 First
meeting Jan 2013
 3 working groups
 Highest priority
 Portable parallel infrastructure
 General-purpose tensor algebra algorithms
 Protocols for information exchange and code
interoperability
 Education and training

13. Portable parallel infrastructure
 Technology trends
 Massive concurrency on a chip
 Massive number of sockets in largest supercomputers
 Heterogeneity (CPU + GPU)
 Deep, complex memory hierarchies
 Memory and communication bandwidth limited
 Bleeding edge applications may need to
 Coordinate over 109 threads
 Tolerate faults
 Explicitly manage energy consumption

14. Sustainability of large and widely distributed
chemistry codes
 Enable most computations in chemistry
 Likely will run on leadership class machines
 Working group will include computational chemists,
parallel programming experts, and reps from major
tech providers (NVIDIA, Intel, IBM)

15. Sustainability of software developed by smaller
research groups
 Need to understand programming models and tools
 Need to understand how both community and
software can be better organized
 Accelerate testing of new ideas at sufficient scale to
determine their worth
 Key: being able to write code and integrate into
existing software.
 Currently, new developments take months or years
to migrate from developers software to other
packages

16. General-purpose tensor algebra
algorithms
 Tensor algebra ubiquitous in science and
engineering
 Need new approaches for computing with high
dimensional tensors
 Currentsoftware—8 or fewer
 Emerging methods require 3N dimensions where N (the
number of electrons) may be O(100) or more.
 Need common framework of reusable software
elements.

17. Challenges for high-rank tensors
 Challenges
 Develop robust implementations of algorithms
 Standardize data structures and algorithms, APIs,
software elements, and frameworks
 Automate the derivation, transformation, and
implementation of tensor expressions.
 Will require cross-disciplinary collaborations
 Infrastructure will include DSL, runtimes, compilers as
well as static and dynamic algorithm analyses

18. Protocols for Information Exchange and
Code Interoperability
 Historically culture has been competitive rather than
collaborative
 Sharing of code and data limited
 Theoretical methods driving code towards greater
size and complexity
 Currently, progress may require substantial
duplication of effort—code that could be reused is
not due to lack of standards
 Impedes new science, wastes human labor

19. Information Sharing
 Standards for data shared between codes
 Standards (or methods to convert between
standards) for stored data to facilitate mining.
 Data provenance
 Cannot expect all code to use the same format, but
transformation leads to errors, computational
inefficiencies, complex interfaces

20. Code Sharing
 Need to establish level of interoperability
 Coarse-grained
 Hartree-Fock code from one app, MP2 code from another
 Fine-grained
 Calculate most of one electron contributions to a Fock matrix
in one program, relativistic and solvent terms in another
 Need architecture

21. Education and Training
 Mastering existing codes is daunting task for grad students
 Chemical education culture worse than many STEM fields
 PhD only requires modest coursework
 OK for most fields of chemistry where undergrad training is
adequate preparation for hands-on lab research
 Most students unprepared for research in computational chemistry
 Ad hoc training by individual research groups is innefficient
 How should students be prepared for 2020 and beyond?
 Programming models and tools
 Multidisciplinary foundation to computational science
 Reasoning about software
 Manipulating software with confidence and facility

22. Summer school
 Summer school for grad students supported by
community
 Fundamental algorithms of computational chemistry
 Software best practices

23. If S 2I2C2M2 is successful
 Open access to new software tools and infrastructure
 Training and educational opportunities for grad
students and post-docs in
 Algorithms
 Code standards
 Software best practices
 NOT the goal to produce monolithic computational
chemistry package to replace existing ones
 Healthy competition is good
 Sets of robust and properly validated software
components that can be shared will benefit the entire
community