Philipp Grete

I am a postdoctoral research associate in the Department of Physics & Astronomy at Michigan State University .
In general, my research covers (astrophysical) plasma modeling including magnetohydrodynamic processes, their role in (astro)physical systems, and related numerical methods. Currently, I am working on analyzing and characterizing energy transfer in compressible MHD turbulence, understanding driving mechanisms in astrophysical systems, investigating the role of turbulent magnetic fields in the formation of supermassive black hole seeds, and performance portable programming models for exascale computing.
Contact information
Office BPS 3248
Email grete [at] pa.msu.edu
Phone +1 (517) 884-5622
Address Department of Physics & Astronomy
Michigan State University
Biomedical Physical Sciences
567 Wilson Road, Room 3248
East Lansing, MI 48824


News and upcoming events

upcoming
Oct 2019 The Enzo code paper has been published in the The Journal of Open Source Software (JOSS).
This version includes the subgrid-scale model for compressible MHD turbulence I developed.

JOSS is a developer friendly, open access journal for research software packages. (about JOSS)
In addition to a code peer review (mostly focussing on improve the quality of the software submitted, e.g., with respect to documentation, functionality, or testing) each accepted submission is assigned a Crossref DOI that can be referenced to.
This supports the recognition of the software developers and the overall significant (but still all too often underappreciated) contribution of open source software to science.
Sep 2019 During the German Astronomical Society Meeting in Stuttgart I had the chance to tour SOFIA and see this impressive technical achievement in person.
SOFIA (Stratospheric Observatory for Infrared Astronomy) is a 2.7m telescope mounted on a plane so that observations can be made from high altitude (above most of the water vapor in the atmosphere, which absorbs infrared light).



In addition, I organized a Splinter Meeting on Computational Astrophysics during the conference.
Many different topic were presented ranging from galaxy mergers, to star formation, to radiative transport methods.
On top of the science presentations we also used the meeting to discuss current trends and developments in computational (astro)physics including GPU computing and FPGAs.
Aug 2019 Research update
In addition to the work on K-Athena/performance portability several more physics focused project have reached maturity over the past months.
  • We (W. Schmidt and P. Grete) extended the energy transfer framework previously applied to compressible MHD to also include internal energy dynamics. This extension allows the detailed study of energy dynamics within and between kinetic and thermal energy budget, and we applied it to study compressible hydrodynamic turbulence at various Mach numbers. One main result is an indication of a Mach number dependent natural forcing composition. Preprint available at arXiv:1906.12228.
  • The article on Intermittent fragmentation and statistical variations during gas collapse in magnetized atomic cooling haloes by P. Grete, M. A. Latif, D. R. G. Schleicher, and W. Schmidt has been publiblished in MNRAS. We used subgrid-scale model for unresolved MHD turbulence in cosmological simulations targeting supermassive black hole seed formation within the direct collapse scenario. We find a turbulent supersonic and super-Alfvénic collapse with an active small-scale dynamo amplifying magnetic fields to dynamically relevant strengths and only intermittent fragmentation making the direct collapse a viable scenario for the formation of massive objects.
  • Finally, we (P. Grete, B. W. O'Shea, and K. Beckwith) studied the effects in departing from an isothermal equation of state in compressible MHD turbulence. We find that several statistics (e.g., the strong anticorrelation between density and magnetic field strength or the scaling of density variations with Mach number) depend on the adiabatic index while other statistics exhibit no systematics -- this may help to break degeneracies in different observables. Preprint available at arXiv:1908.03989.
May 2019 K-Athena is public!
Last year at the Performance Portability with Kokkos training we started to experiment with combining Kokkos with Athena++ in order to allow for GPU-accelerated simulations. The experiment is a complete success and we are happy to make the resulting K-Athena (= Kokkos + Athena++) code publicly available on GitLab.
Details on the porting experience, roofline models on CPUs and GPUs, and scaling on many different architectures are presented in more detail in the accompanying code paper (arXiv preprint), which is currently under review.
In the paper, we demonstrate that K-Athena achieves about 2 trillion (1012) total cell-updates per second for double precision MHD on Summit — currently the fastest supercomputer in the world. This translates to updating an MHD simulation with a resolution of 10,0003 twice per second! At about 80% parallel efficiency using 24,576 Nvidia V100 Volta GPUs on 4,096 nodes, K-Athena achieves a speedup of more than 30 compared to using all available 172,032 CPU cores.
The code release has also been featured on Tech Xplore, and we encourage feedback and/or contribution to the code via the GitLab repository.

Image credit: Carlos Jones/ORNL CC-BY-2.0
Apr 2019 After I took part in the Professional Development Program of the Institute for Scientist & Engineer Educators last year for the first time, I returned this year as a design team leader.
Our design team (consisting of Rachel Frisbie and Forrest Glines — two MSU graduate students) will develop an inquiry activity on Sustainable Software Practices with Collaborative Version Control. We will teach the activity for two REU (Research Experiences for Undergraduates) programs at MSU in the summer.
If you are interested in the teaching material, please contact me. We are happy to share the material and experience.
Mar 2019 Our review paper covering Kris Beckwith's invited talk at ICOPS 2018 on Correlations and Cascades in Magnetized Turbulence (Beckwith, Grete, and O'Shea 2019) has been published in IEEE Transactions on Plasma Science last month.
Similarly, I also gave several talks, e.g., in Budapest, Las Vegas, Berkeley, and UC Santa Cruz, covering our recent results on correlations, energy transfer, and statistics in adiabatic, compressible MHD turbulence.
In addition, I presented first results on our GPU-enabled version of Athena++ using Kokkos. Stay tuned — more information will follow shortly.
Dec 2018 Last year when I gave a talk at the MIPSE (Michigan Institute for Plasma Science and Engineering) seminar in Ann Arbor I was also interviewed. The interview is part of a MIPSE sponsored outreach program to capture some of the importance (and excitement) of plasmas for students and the general public.
The short 3 minute clip is now online on YouTube.



All interviews of the series can be found on the MIPSE YouTube channel.
Aug 2018 Last month I took part in the Performance Portability with Kokkos training at Oak Ridge National Laboratory.
The four-day training was organized by the Exascale Computing Project and covered the Kokkos programming model and library for writing performance portable code. In other words, code is only written once and can afterwards be compiled for different target architectures, e.g., CPUs or GPUs, automatically taking into account architecture specific properties such as the memory layout and hierarchy. We are currently implementing this approach and will use it for the next project.

In addition, SparPi, the Raspberry Pi supercomputer model, had another appearance this month at Astronomy on Tap.
I presented a talk on What’s so super in supercomputer? Theory & Practice including a live demonstration. After the presentation, people made ample use of the opportunity to get their hands-on model themselves and directly interact with the live simulation.
May 2018 Our latest paper on Systematic Biases in Idealized Turbulence Simulations has been accepted for publication in The Astrophysical Journal Letters, see Grete et al 2018 ApJL 858 L19.
We show how the autocorrelation time of the forcing is intrinsically linked to the amplitude of the driving field, and how this affects the presence of (unphysical) compressive modes even in the subsonic regime.
I also presented the results at
Apr 2018 The MSU's science festival was a great success. We had a lot of interested people of all ages at our booth.
The assembled supercomputer model featured all components of an actual supercomputer, e.g., power supply units, networking/interconnects, compute nodes, and a frontend. On the frontend we are running TinySPH, an interactive hydrodynamics code that allows for both illustrating high performance computing techniques such as dynamic load balancing, and illustrating physics by changing fluid parameters such as viscosity or density.
In addition, we offered multiple virtual reality experiences including one that featured a magnetohydrodynamic turbulence simulation. Visitors were able to dive into a turbulent box and experience the rich and complex fluid structure interacting with highly tangled magnetic fields from within.
Feb 2018 I am very happy that I was accepted to participate in the 2018 Professional Development Program of the Institute for Scientist & Engineer Educators. This inquiry based teaching training covers multiple workshops where the participants collaboratively design an inquiry activity that will eventually be implemented in class later this year. The first workshop is the "Inquiry Institute" from March 25-28, 2018 in Monterey, CA.

In addition, I will attend SnowCluster 2018 - The Physics of Galaxy Clusters from March 18 - 23, 2018 in Salt Lake City where I present a poster talk of our latest results on "Systematic biases in idealized turbulence simulations".
Feb 2018 Parts for a Raspberry Pi based "supercomputer" arrived.
We will use the system for outreach, e.g., to demonstrate high performance computing, and in class for hands-on tutorials. I am currently setting the system up, which consists of 8x Raspberry Pi 3 for a total of 32 cores with 8 GB main memory.
The system will premiere at MSU's science festival.
Visit our booth at the Expo on April 7 for Hands-on Supercomputing and multiple Virtual Reality experiences.
Nov 2017 I presented the latest the results on energy transfer in compressible MHD turbulence
  • on Nov 14 in the Astrophysics Seminar at Notre Dame
  • on Nov 15 in the MIPSE seminar (Michigan Institute for Plasma Science and Engineering) at the University of Michigan
A recording of the MIPSE seminar is available here.
Sep 2017 My PhD thesis on Large eddy simulations of compressible magnetohydrodynamic turbulence was awarded with this year's Doctoral Thesis Award by the German Astronomical Society.
I presented the work at the Annual Meeting of the German Astronomical Society . See official press release [pdf] (in German), and MSU press releases (in English) from the CMSE department and the College of Natural Science .
Sep 2017 Our article on Energy transfer in compressible magnetohydrodynamic turbulence has been accepted for publication. In the paper, we introduce a scale-by-scale energy transfer analysis framework that specifically takes into account energy transfer within and between kinetic and magnetic reservoirs by compressible effects. The paper appears as a featured article in the journal Physics of Plasmas. Moreover, the article is covered by an AIP Scilight.
Sep 2017 We (PI B. W. O'Shea, Co-PIs B. Côté, P. Grete, and D. Silvia) successfully obtained computing time through an XSEDE allocation. I will use the resources to study driving mechanisms in astrophysical systems.
Aug 2017 I took part in the Argonne Training Program on Extreme-Scale Computing. This intensive two-week training is funded by the DOE's Exascale Computing Project and allowed me to gain knowledge and hands-on experience on next-generation hardware, programming models, and algorithms. I can highly recommend this training to everyone involved in high-performance computing.
Jun 2017 I presented first results of our energy transfer study for compressible MHD turbulence and the method itself