CDT-NGCM Seminar with Dr Jonathan Leliaert – What makes scientific software successful? A case study in micromagnetic software.
Blog Posts, Seminars

CDT-NGCM Seminar with Dr Jonathan Leliaert – What makes scientific software successful? A case study in micromagnetic software.

This seminar took place on 13 October 2021 at 1pm.

Dr Leliaert is a research physicist in the DyNaMat research group at Ghent University in Belgium specialising on the development of multi-physics computational packages. He is also an experienced software developer. In particular, he develops Vinamax and MuMax physics simulation software packages. These packages are used broadly by the research community and industry to simulate magnetic materials, magnetic devices, or study various fundamental aspects of magnetism (Figure 1).

Figure 1 – Examples of micromagnetic simulations performed by MuMax3 software [1]. (a) Simulation of magnetic vortex in a permalloy sample of dimensions 750 nm x 750 nm x 10 nm. (b) and (c) Simulation of magnetic force microscopy images in AC mode at different lift heights.

In his seminar talk, he first briefly explained the basic background of the physics of magnetism and introduced the mathematical theory called micromagnetics. Micromagnetics is the classical continuum theory of magnetism, analogous to Navier-Stokes equations in fluid mechanics and computational fluid dynamics, for example, and is at the heart of the MuMax magnetic simulation package. Dr Leliaert then went on to discuss the historical development and explained the various functionalities and performance of the MuMax code, especially its most current main-stream version MuMax3 [1]. This part included a range of topics such as the overall user interface, validation of the accuracy of computation, or hardware requirements, for example. One important aspect of MuMax3 is its GPU capability, which drastically enhances the computational performance as illustrated in Figure 2.

Figure 2 – MuMax3 throughput as a function of GPU release date. The throughput is defined as the number of finite difference cells in a micromagnetic finite difference simulation evaluated per second. On average, the throughput increases roughly by 30% every year [2].

A substantial part of the talk was devoted to the discussion of the software engineering issues and challenges behind the development of the MuMax3 software. This included addressing the open-source licensing, organisation and maintenance of user forums and code website [3], routine testing of the code, issues in maintaining the cross-platform software development, and development of cloud services to host and run MuMax3 for those who do not have access to machines with suitable CUDA enabled graphics cards.

Dr Leliaert concluded his talk by providing a lot of useful advice to researchers who are planning to develop and distribute a code as part of their research work. He said that one of the challenges is to maintain a high-profile publication profile alongside the software development activities. It takes time and resources to develop and maintain the high standards of a code such that it can be used by the community effectively. As Dr Leliaert emphasised during his concluding remarks: “to develop robust academic carrier, it is crucial to invest time also in your own research, alongside your software development activities”.

The seminar talk was well attended (nearly 50 participants) and followed by vigorous discussion afterwards and received excellent feedback from the audience.

References:

  1. Vansteenkiste, Arne, Jonathan Leliaert, Mykola Dvornik, Mathias Helsen, Felipe Garcia-Sanchez, and Bartel Van Waeyenberge. “The Design and Verification of MuMax3”. AIP Advances 4, no. 10 (October 2014): 107133.
  2. Leliaert, J., and J. Mulkers. “Tomorrow’s Micromagnetic Simulations”. Journal of Applied Physics 125, no. 18 (14 May 2019): 180901.
  3. https://mumax.ugent.be

Written by Ondrej Hovorka