Investigator Updates
Matthew Turk
DDD Investigator Symposium 2019
mjturk@illinois.edu
Collaborative Effort
Before beginning, I want to acknowledge collaborators (students, postdocs, research scientists and faculty members), whose work I am going to present. Wherever possible I have identified the collaborating individual on the appropriate slides, but I acknowledge I have likely passed some over.
Additionally, this work is enabled by a host of open-source software, without which it would have been completely impossible. This includes the python ecosystem as well as a wealth of scientific software.
Collaborators
- Students
- Alex Gagliano
- Colleen Heinemann
- Wei-Ting Liao
- Xinyang Lu
- Kwok-Sun Tang
- Cassidy Wagner
- Samantha Walkow
- Postdocs
- Jared Coughlin
- Madicken Munk
- Research Scientists
- Nathan Goldbaum
- Kacper Kowalik
- Meagan Lang
- Faculty
- Amy Marshall-Colon
- Hsi-Yu Schive
The Prompt
- Your field: What are the important questions? What are you doing to move the field forward?
- New discovery: What new science are you exploring?
- Tool/method: What new tool or method are you working on that might be useful for others?
- Challenge: What's a challenge you are facing?
Important Questions
- How can we improve fidelity of our training data in astrophysics and crop sciences?
- What techniques can we apply to data-rich, dimensionally-reduced datasets?
- Which conceptual frameworks can be applied to spatially-organized data analysis?
- How can data sharing be accelerated?
Improving Fidelity in Astrophysics
The Big Picture
what can we understand about the origins of the elements that make up our world?
Improving Fidelity in Astrophysics
The Medium Picture
Can we improve our simulations to put observations of the universe into context?
Improving Fidelity in Astrophysics
The Pixel Picture
- Improving chemical solution
- Improving fluid structure solution
- Identifying parameters of compressible turbulence
Astro Collaborators
This work was done in collaboration with Hsi-Yu Schive, Kwok-Sun Tang, and Wei-Ting Liao.
Chemistry
In the early universe, we have these species:
- Hydrogen: ${\rm H}$, ${\rm H}^{+}$, ${\rm H}^{-}$
- Helium: ${\rm He}$, ${\rm He}^{+}$, ${\rm He}^{++}$
- Electrons: ${\rm e}^{-}$
- Molecular Hydrogen: ${\rm H}_{2}$, ${\rm H}_{2}^{+}$
Within these, there are dozens of reactions operating on different time scales and with different dependencies on density and temperature.
This is the most time-intensive part of conducting simulations of primordial star formation.
Chemistry
We utilize symbolic mathematics and templating engines to generate networks of ODEs usable within conventional ODE solvers.
Over the past year, utilizing this technique, we have:
- Moved RHS generation to GPU, reducing memory transfer and improving speed
- Identified problematic phase spaces
- Begun to explore perturbation against equilibrium solutions
The flexibility afforded during the DDD award has enabled an entirely new class of chemical solvers to be implemented and utilized.
Improving Fidelity in Astrophysics
Our interaction with astrophysical fluids is mediated by telescopes, necessarily reducing the dimensionality of our perception.
To understand the properties of turbulent gas, we have begun training on our observations in order to determine the initial conditions.
Our first results have been somewhat promising. Our process:
- Generate turbulent boxes with variety of spectral indices for the density ($\gamma$) and velocity ($\beta$).
- (Soon) Simulate observation from multiple viewpoints.
- [training montage]
Improving Fidelity in Crop Sciences
Climate change models predict unprecedented warming by the end of this century, resulting in a temperature-related yield loss of 15% for corn and soybean.
How can we utilize computational biology and molecular breeding technologies to develop crops that are highly productive under challenging environmental conditions such as heat, water, and nutrient stress.
Crop Collaborators
This work was done in collaboration with the Crops in Silico team, including Meagan Lang and Amy Marshall-Colon.
Improving Fidelity in Crop Sciences
(Disclaimer: I am not a biologist)
In the biological literature there are genuinely thousands of plant and plant component models that can be applied to different relevant processes.
For the most part, making these work together is really, really tricky. The
yggdrasil framework, developed by Meagan Lang, is designed to provide
information-aware connections between different models.
WT: Data Sharing
The Whole Tale project is an open source implementation of computational environment sharing. Without question, my favorite part of Whole Tale has finally been deployed.
The Whole Tale filesystem is able to synthesize data from many different sources (including DataVerse, HTTP, Globus, local, WebDAV, and institutional repositories), dynamically update accessible data during the course of a session, and enable direct collaboration across projects and individuals.
Even in light of the massive advances in computational environments, and the rich ecosystem (including Binder), this is the key enabling component that will help to reduce the technical friction to so many projects.
(Also it is totally usable outside of Whole Tale)
WT: Demo
Haha, I'm kidding, I'm not going to demo this now.
It's all online at wholetale.org and documented at wholetale.readthedocs.io
yt: Viz
yt: Viz
yt: Domain Contexts
- Jargon
- Data formats
- Operations
This last year we have engaged with the Advanced Radar Toolkit community, as well as other familiar domains such as oceanography, seismology and weather.
yt: Declarative Analysis
(This is the one I'm currently most excited about.)
Inspired by vega-lite and its family of projects, we have begun work on developing a schema (informed by previous efforts) for declarative analysis of volume-organized data.
{
'projection': {
'dataSource': 'allData',
'operation': {
'max': {'field': 'density'}
},
'axis': 'x'
}
(plus wasm, plus data repositories, ...)
Some Challenges
People leave.
It's always obvious what leads to success, but it's easy to convince yourself you know (in advance)
Engagement can be really emotionally taxing.
Book Recommendations
I always get lots out of book recommendations from you folks, so here are mine:
- Radical Candor by Kim Scott (h/t Karthik Ram and Tracy Teal)
- Bossfight: ZZT by Anna Anthropy
- Lovecraft Country by Matt Ruff
- Planetfall (series) by Emma Newman
Thank you
Matthew Turk
mjturk@illinois.edu