I am a computational engineering researcher, using numerical methods and high-performance computing to study physical problems in energy, transportation, and aerospace.

I work with a diverse and exciting group of students on these topics; publishing in combustion, energy, and computational physics communities; and then presenting and teaching our work in the world.

We study mainly fluid dynamics phenomena using computer simulations, particularly fluid flows involving chemical reactions such as combustion. Our work extends from analyzing and simplifying chemical kinetic reaction models for transportation fuels, to developing surrogate models for predicting properties of fuels and chemicals, and even to computationally expensive, direct numerical simulations of turbulent combustion. We develop and apply methods to simulate combustion and other fluid-flow problems, including in wildfires and the ocean. We also write a lot of code!

Want to do research with me? Read about my lab.

Contributions

My lab and I have made many contributions since since I started doing research in 2008. Here are some of the highlights from our work. How I describe these is always evolving as we learn more.

A clip from the paper's text or figure
We can predict complex properties of liquid aviation fuels, including sustainable bio-based alternative fuels, using machine learning methods. (2016 — 2024)
But this requires carefully selecting features when training the models.

Kyle E. Niemeyer headshot Vi Rapp headshot Shane Daly headshot Khang Tran headshot Christopher Hagen headshot

feedstock to functionmachine learningsustainable aviation fuels  📄papers

A clip from the paper's text or figure
By incorporating key components in models, we can capture the ignition behavior of live fuels in simulations of wildfire. (2019 — 2024)
Models for live vegetative fuels (like trees and shrubs) need to include smaller components like sugars, lipids, proteins, phenols, and minerals, and not just moisture content.

Diba Behnoudfar headshot Kyle E. Niemeyer headshot Peter Hamlington headshot

live fuelswildfire  📄papers

A clip from the paper's text or figure
We can achieve exascale performance in Monte Carlo neutron transport solvers written in Python, using just-in-time compilers. (2022 — 2024)
Unlike most traditional compiled codes, this approach allows portable performance on modern architectures.

Joanna P. Morgan headshot Todd S. Palmer headshot Kyle E. Niemeyer headshot

Monte Carloneutron transportnuclear reactorsexascale  📄papers

A clip from the paper's text or figure
We can predict ignition and propagation of smoldering combustion, including effects of composition and environmental conditions, in woody fuels by representing the amounts of major components. (2019 — 2024)
Varying fuel composition in terms of cellulose, hemicellulose, and lignin proportions can noticeably change smoldering behavior, including critical moisture content.

W. Jayani Jayasuriya headshot Kyle E. Niemeyer headshot David Blunck headshot Tejas C. Mulky headshot

smolderingwildfire  📄papers

A clip from the paper's text or figure
Predicting the amount of carbon taken up by the ocean requires accurately capturing the interactions between ocean turbulence and biogeochemistry. (2019 — 2024)
Simulating the interaction between Langmuir turbulence and ocean biogeochemistry requires applying model reduction methods from combustion.

Peter Hamlington headshot Kyle E. Niemeyer headshot Emily Klee headshot

biogeochemistryoceanturbulence  📄papers

A clip from the paper's text or figure
We can significantly accelerate integration of chemical kinetics by providing some system knowledge to the algorithm and using efficient sparse linear algebra operations. (2023)
Implemented in the open-source library Cantera, benefits appear for even very small models, and we see performance gains of up to 1000 times for large models.

Anthony S. Walker headshot Raymond L. Speth headshot Kyle E. Niemeyer headshot

ODEschemical kineticslinear algebraCanteraintegrators  📄papers

A clip from the paper's text or figure
Appropriately chosen integration algorithms can allow GPUs to efficiently integrate stiff chemistry needed in reacting flow simulations. (2014 — 2023)
Particular algorithms parallelize more efficiently on GPUs.

Kyle E. Niemeyer headshot Chih-Jen Sung headshot Nicholas Curtis headshot

GPUsODEschemical kineticsintegrators  📄papers

A clip from the paper's text or figure
How we model molecular diffusion in turbulent flames can impact our predictions of turbulent flame speed and structure. (2020 — 2022)
This is true even for larger hydrocarbon fuels; fortunately, our efficient algorithm for implementing the accurate multicomponent diffusion model enables its use.

Aaron J. Fillo headshot Guillaume Blanquart headshot Kyle E. Niemeyer headshot Peter Hamlington headshot

diffusionturbulent flamesnumerical methodsdirect numerical simulation  📄papers

A clip from the paper's text or figure
Vectorized operations are a path to speeding up the integration of chemical kinetics in reacting-flow simulations. (2018 — 2022)
But it requires careful selection of integration algorithm, and may lead to new load-balancing issues.

Nicholas Curtis headshot Kyle E. Niemeyer headshot Chih-Jen Sung headshot Andrew T. Alferman headshot

ODEsvectorizationchemical kineticsintegrators  📄papers

A clip from the paper's text or figure
Advancing computations inside domains of time and space dependence can improve performance by reducing network communication. (2018 — 2021)
Getting a performance benefit in heterogeneous combinations of CPU and GPU systems depends strongly on configuration details and you can see drops in performance instead.

Anthony S. Walker headshot Kyle E. Niemeyer headshot Dan Magee headshot

domain decompositioncomputational fluid dynamicsGPUsswept rule  📄papers

A clip from the paper's text or figure
Model reduction methods developed for combustion chemical kinetics can be successfully extended to atmospheric and ocean biogeochemical models. (2019 — 2021)
But algorithms need to be adapted to handle the unique characteristics of different systems, such as non-elementary reactions.

Kyle E. Niemeyer headshot Peter Hamlington headshot Emily Klee headshot

model reductionbiogeochemistryatmospheric chemistry  📄papers

A clip from the paper's text or figure
Large chemical kinetic models can be automatically reduced using a strategy with multiple reduction stages. (2010 — 2019)
We can significantly reduce the size and complexity of detailed kinetic models, but only so far, before removing components introduces unacceptable error.

Kyle E. Niemeyer headshot Chih-Jen Sung headshot Phillip Mestas headshot Parker Clayton headshot Peter Hamlington headshot

model reductionchemical kineticsreduction algorithms  📄papers

A clip from the paper's text or figure
A human- and machine-readable format allows us to easily describe measurements from fundamental experiments in combustion. (2018)
The ChemKED format allows describing fundamental experiments in combustion in a human- and machine-readable way, including descriptions of uncertainty.

Bryan Weber headshot Kyle E. Niemeyer headshot Morgan Mayer headshot Richard West headshot

ChemKEDparameter databasesexperimental measurements  📄papers

A clip from the paper's text or figure
Electrical arcs in vacuum remelting furnaces can be accurately located using magnetic field sensors paired with a physics-based model. (2018)
Vertical misalignment can increase error, which adding multiple sensors can counteract.

Miguel Soler headshot Kyle E. Niemeyer headshot

manufacturingvacuum arc remeltingfinite element analysis  📄papers


Copyright Kyle E. Niemeyer. See this site's GitHub repository to view source and provide feedback.