Final Technical Report: “Frameworks, Algorithms and Scalable Technologies for Mathematics (FASTMath) SciDAC Institute” Author Daniel R. Reynolds dreynolds@umbc.edu 410-455-2429 ORCID ID: 0000-0002-0911-7841 DOE Grant Number DE-SC0021354 Sponsor Office of Advanced Scientific Computing Research Project period September 23, 2020 through September 22, 2025 Date prepared January 12, 2026 Organization Southern Methodist University PO Box 750156 Dallas, TX 75275-0156 Organization Point of Contact Peter Moore, Professor and Chair pmoore@smu.edu 214-768-4485 Department of Mathematics Southern Methodist University Abstract SMU personnel formed a portion of the overall “Time Integration” team within the FASTMath SciDAC-5 Institute, and we interacted very closely with team members from collaborating institutions. The major goals of our team within the FASTMath institute may be categorized into two groups: the development of advanced and application-aware time integration methods and software, and close interactions with DOE application scientists to facilitate their use of these new methods and software. The two goals are intimately linked, since our research and development of novel tools is informed by the needs of our application partners, who in turn benefit from subsequent mathematical and software advances. While the SMU personnel collaborated with the rest of the Time Integration FASTMath team on most of our shared deliverables within the larger FASTMath institute, SMU personnel primarily contributed to the following subset of those goals: ●​ expanding capabilities for higher-order and solve-decoupled multirate methods, ●​ enhancing support for temporal adaptivity within multirate methods, and ●​ adding structure-aware time integration methods. Each of these goals focused both on intellectual contributions through journal articles or research presentations, and on enhancements to the open-source SUNDIALS library of time integrators and nonlinear solvers, of which the SMU PI Reynolds is a core developer. This report is structured as follows. The next section describes our accomplishments toward the above objectives, as well as other related work that was funded by this project. We then describe the training and professional development opportunities that were funded by this project, followed by a section outlining how we disseminated our results to communities of interest. Project Accomplishments We group our accomplishments under this project topically within the following subsections. In general, we organize these so that our intellectual contributions (via publications and research presentations) are first, followed by software contributions to the SUNDIALS library, and ending with contributions in collaboration with both applications scientists and with other sub-teams within the FASTMath institute. We note that many of the accomplishments below do not fit neatly within this organizational structure, in that some accomplishments combine both publications and software contributions, while other publications were with outside collaborators, and so we group these thematically in lieu of overly dissecting each effort into separate areas. Higher-order and solve-decoupled implicit multirate methods Under previous FASTMath funding, we completed an initial implementation of both third- and fourth-order solve-decoupled implicit multirate infinitesimal (MRI) methods within the SUNDIALS ARKODE library. These developments built off of two previously-existing components in ARKODE: (1) the ARKStep module that supports implicit, explicit and mixed implicit-explicit (IMEX) time integration using single-rate additive Runge--Kutta (ARK) methods [Cooper and Sayfy, Math. Comp., 1983], and (2) the MRIStep module that supported up to third-order explicit multirate infinitesimal step (MIS) methods [Wensch, Knoth & Galant, BIT Numer. Math., 2009]. Under that previous FASTMath funding, we began extending the MRIStep module to support a new class of multirate infinitesimal GARK (MRI-GARK) methods [Sandu, SIAM J. Numer. Anal., 2019], that allow up to fourth-order temporal accuracy, as well as either explicit or solve-decoupled implicit treatment of the slow time scale. Under the current project, we completed implementation of the MRI-GARK, and have released them in SUNDIALS v5.7.0 (January 2021). We then extended MRIStep to support a novel class of MRI-GARK methods that allow IMEX treatment of the slow time scale, with so-called implicit-explicit MRI-GARK (IMEX-MRI-GARK) methods. These methods combine the strengths of ARK IMEX methods and MRI-GARK methods, allowing either third- or fourth-order multirate methods with extreme flexibility for large-scale scientific applications. In addition to releasing these methods as part of SUNDIALS v6.0.0 in December 2021, we published an article describing these methods [Chinomona and Reynolds, SISC, 2021] Lastly, under funding under both the previous FASTMath institute and this project, we have created a new class of very high order multirate methods. These approaches are rooted in exponential Rosenbrock method theory, and as such we name them Multirate Exponential Rosenbrock (MERB) methods. Notably, these constitute the first-ever sixth order accurate infinitesimal multirate methods. We published an article describing these methods and demonstrating their performance on a suite of benchmark applications [Luan, Chinomona and Reynolds, SISC, 2022]. Higher-order and time-step adaptive multirate methods Our most significant contributions under the current project have focused on extensions to allow temporal adaptivity within MRI methods. While temporal adaptivity for single-rate time integration methods has been well studied, prior to this project there existed no published work on related techniques to adapt the multiple nested step sizes that arise within MRI time integration methods. However, historically all of the time integration methods released as part of SUNDIALS provided this feature, ensuring that initial-value problems were solved to requisite accuracy instead of merely with a given number of time steps using fixed step sizes. Under this project, we first identified a range of candidate multirate methods that supply both a time-evolved solution and an embedded estimate of the local temporal error, however the quality of these embedded error estimates can vary widely between methods. At that time, we could only identify two families: multirate GARK (MrGARK) methods [Sarshar, Roberts and Sandu, SISC, 2019], MRI-GARK methods [Sandu, SINUM, 2019], and multirate exponential Runge--Kutta (MERK) methods [Luan, Chinomona & Reynolds, SISC, 2020], and of those, only MRI-GARK methods supported an “infinitesmal” structure as expected within SUNDIALS. Notably, none of the IMEX-MRI-GARK methods from our previous publication included embeddings to estimate temporal error, and as such, there did not exist IMEX MRI methods that could be used for time adaptivity. To fill this gap, we derived a new class of up to fourth order IMEX MRI methods that include embeddings. The mathematical theory for this family, called IMEX-MRI-SR, also provided much-needed understanding of the better-than-expected behavior of our previous “MERK” multirate methods for nonlinear applications. This work was published in [Fish, Reynolds, and Roberts, JCAM, 2024]. In addition to identifying candidate base methods, we derived a range of algorithms for adaptively selecting the various time steps that comprise a multirate method. These methods seek to choose both the “slow” step size H and the “fast” step size h << H such that both are as large as possible (to shorten the total integration time) while simultaneously satisfying user-provided accuracy requirements. During this reporting period, we completed derivations of novel “simultaneous” controllers for both H and h that are rooted in optimal feedback control theory, and that rely on piecewise polynomial (constant, linear and quadratic) approximations of the feedback control. Our proposed methods show considerable promise in meeting the dual goals of maximal efficiency and acceptable accuracy, and was published in [Fish and Reynolds, SISC, 2023]. Temporally adaptive MRI methods require propagation of information between the time integrator used at the fast time scale and the enclosing MRI method; however, the initial implementation of MRIStep did not provide sufficient flexibility to support this use case. We thus thoroughly overhauled the entire time adaptivity infrastructure within ARKODE to utilize an object-oriented approach. The new “SUNAdaptController” infrastructure is available to all SUNDIALS integrators. It supports all ARKODE’s previous temporal controllers, facilitates the development of new multirate controllers, and allows user-supplied adaptivity algorithms. This was released to users in Fall 2023 in SUNDIALS v6.7.0. With this infrastructure in place, we examined the performance of our existing suite of single-rate adaptivity controllers in SUNDIALS, and found that although they worked well for challenging, very stiff, problems, they under-performed when tasked with simpler, non-stiff applications. We thus developed a suite of test problems with a wide range of difficulty, and ran exhaustive testing to compare ARKODE’s existing single-rate controllers with others from the literature. Through this investigation, we identified a number of published single-rate controllers that performed remarkably well. We thus added new SUNAdaptController modules to implement these, and updated the default ARKODE controller and its parameters to the combination that performed best across the testing suite. These were released with SUNDIALS v7.3.0 in April 2025. Also with this updated adaptivity infrastructure in place, we extended the MRIStep module to support a wide range of embedded MRI methods. These include extensions in our support for MRI-GARK methods to correctly handle their final “embedding” stage, as well as new support for both IMEX-MRI-SR and MERK methods. Of these, we note that our previous work on MERK methods did not include embeddings for temporal error estimation, so as part of this work we derived novel embedding coefficients to enhance our previously-introduced MERK methods, and implemented them within MRIStep. Following our publication on “simultaneous” MRI adaptivity controllers above, and their implementation in MRIStep, we tested them on a range of target applications. Those results showed that although these simultaneous controllers could effectively control multirate temporal error, they did not achieve as significant a time scale separation as expected for the underlying applications (i.e., the ratio H/h that came from the controller was significantly smaller than the applications warranted, thus resulting in less computational efficiency than expected. We thus developed two new classes of MRI adaptivity controllers that should be usable within “telescopic” multirate applications (i.e., applications with 3 or more time scales). We included these controllers in the SUNDIALS v7.2.0 release (December 2024), and submitted a journal article detailing their derivation and performance in October 2025 [Reynolds, Amihere, Mitchell, and Luan, arXiv, 2025]. Solve-coupled multirate methods Although solve-coupled infinitesimal multirate methods have been theoretically viable for some time, and offer the hope of absolute linear stability for any combination of slow and fast time step sizes (H and h) within a multirate method, none have been proposed in the literature due to a perception of overly high computational cost. This cost arises due to the “solve-coupled” nature: the slow stage solution must serve dual roles as a forcing term to, and the eventual solution of, a modified differential equation at the fast time scale. In order to compute a slow stage solution that satisfies both of these competing criteria, one must solve a highly challenging system of nonlinear equations. Under this project, we benchmarked novel approaches that could be used to solve this system of equations, and developed the mathematical theory for the achievable linear stability that could be afforded by solve-coupled infinitesimal multirate methods. Unfortunately, all approaches that we devised required very significant computational costs. After discussing these options with our scientific collaborators, we have found that they have little/no “appetite” for these costlier (but potentially more stable) multirate algorithms. A summary of their feedback is that they only consider using multirate methods for problems where they can perform as much of the application as possible using explicit methods, and that they would reserve any use of fully-coupled implicit algorithms for a single time scale, choosing to “step over” faster dynamics using an L-stable implicit integrator. We thus halted work on this topic. ML tools for optimizing solver parameters Adaptive time integration methods, particularly those that include implicit algebraic solvers, include a multitude of parameters that may be tuned to optimize performance for a given application. Traditionally, these options are modified manually by experts, who leverage extensive experience in solving similar applications. In order to help remove this “expert” bottleneck for new applications, we wanted to expose these SUNDIALS parameters so that they could be modified at the command-line, enabling application-specific optimization through automated parameter tuning software (e.g., GPTune). Thus under this project, we added utilities to all SUNDIALS integrators (ARKODE, CVODE, CVODES, IDA, IDAS, KINSOL) and shared modules (SUNLinearSolver, SUNNonlinearSolver, SUNAdaptController, etc.) to support command-line options for all scalar-valued “set” routines. By adding a small number of function calls to a user code to enable command-line control, this infrastructure allows users to adjust solver parameters without modifying source code or recompiling their applications. This support was released as part of SUNDIALS v7.5.0 in September 2025. General SUNDIALS support SUNDIALS has undergone numerous improvements since the first “method” paper was published in 2005. In this project, we published a follow-on method paper that describes its new flexibility for supporting application-specific data structures and algebraic solvers, as well as numerous new options [Gardner, Reynolds, Woodward and Balos, ACM TOMS, 2022]. Additionally, we published a second method paper that focused exclusively on the new ARKODE solver module in SUNDIALS, that has been primarily authored by the SMU PI Reynolds [Reynolds, Gardner, Woodward, and Chinomona, ACM TOMS, 2023]. This publication described ARKODE’s role as an infrastructure to support one-step methods for initial-value problems, the current time-stepping algorithms that it includes, and examples highlighting both its flexibility and parallel scalability. We contributed to a final method paper that describes more recent updates to SUNDIALS from releases 6.2.0 through 7.3.0 (i.e., April 2022 through April 2025); which we submitted for publication in Spring 2025 [Roberts, Aggul, Reynolds, Balos, Gardner, and Woodward, arXiv, 2025]. Also within this project, we worked with our LLNL collaborators to completely rewrite the SUNDIALS documentation so that it could be hosted online (https://sundials.readthedocs.io/en/latest/). This web documentation is at least as complete and well-curated as our previous PDF-based documentation, and has proven to be an invaluable resource for our research collaborators. To improve support for applications that use complex arithmetic within time-adaptive simulations or nonlinear root-finding problems, we have worked to upgrade the entire SUNDIALS library to natively support complex solution data. While it has always been possible for users to solve problems posed in complex arithmetic, users could only accomplish this by providing custom vector interfaces and/or packing both real and imaginary coefficients into a double-length real-valued array, and for implicit calculations they also had to provide custom algebraic solver modules. Partially in support of a SciDAC-FES partnership project (Frontiers in leadership gyrokinetic simulation), that funded extension of our Krylov linear solver modules (PCG, GMRES, FGMRES, BiCGStab, and TFQMR) to complex arithmetic, under this project we implemented additional infrastructure to support complex arithmetic throughout the rest of SUNDIALS. While this implementation is now complete, and is accessible in the public GitHub repository, it has not yet been released. Collaborations with FES SciDAC projects In anticipation of future SciDAC partnership proposals, we dedicated a portion of our effort on this project to initiate collaborations with other SciDAC projects, particularly those funded under FES. To that end, we collaborated closely with a team from the Partnership for Multiscale Gyrokinetic Turbulence (MGK) SciDAC project, on the construction of a novel reduced-order code for studies of coupled electron and ion scale turbulence in magnetic fusion devices. Our MGK collaborators created a reduced model (named MuSHrooM) to simulate interacting electron-scale and ion-scale turbulence, and we worked to explore optimal time integration methods for such simulations. To this end, we have guided our collaborators through the construction of a fully featured but lightweight interoperability layer between MuSHrooM and ARKODE, and explored the use of single-rate, adaptive, IMEX time integration methods for the problem. These results were highly promising, resulting in ion-scale simulations that reproduce plasma instabilities at less than 1/18th the cost of full-order 5D simulations. We then constructed a MuSHrooM interface to MRIStep, to enable multirate simulations that could subcycle the faster electron scale components. The most challenging aspect of this, however, was determination of an effective splitting of the application into “fast” and “slow” components. This task was not resolved before the MGK project ended, at which time the project splintered and work on MuSHrooM on their end was halted. However, the connections forged through this collaboration resulted in our inclusion in multiple FES SciDAC partnership project proposals, and thus the FASTMath-funded efforts planted strong seeds for subsequent collaborations. Collaborations with AMReX team on reactive flow applications In addition to inter-project collaboration, we collaborated extensively with the AMReX team in FASTMath throughout the duration of the project, resulting in two publications and multiple training opportunities. We worked with members of the AMReX team to explore the benefits from using explicit and mixed implicit-explicit multirate methods from MRIStep for structured grid simulations of reactive flow. In previous work, the AMReX team achieved significant efficiency gains through using multirate spectral deferred correction (MRSDC) methods on such applications [Emmett, Zhang and Bell, Combustion Theory and Modeling, 2014]. Under previous FASTMath funding, we converted their MRSDC codes for reactive flow to instead utilize the MIS, MRI-GARK and IMEX-MRI-GARK infinitesimal multirate methods in ARKODE. Initial results showed that these MRI methods resulted in dramatic performance gains over MRSDC for such applications. Our article detailing this research was published in [Loffeld, Nonaka, Reynolds, Gardner, and Woodward, Int. J. High Perf. Comput. Appl., 2024]. We additionally worked with members of the AMReX team who contribute to the Pele and Nyx codes, that perform large-scale simulations of combustion and primordial cosmology, respectively. Both codes use SUNDIALS similarly, to perform time integration of stiff reaction networks within each finite-volume cell, but where these cells are decoupled from one another. We published an article describing our collaborations with these teams on using SUNDIALS in these contexts within exascale applications in [Balos, Day, Esclapez, Felden, Gardner, Hassanaly, Reynolds, Rood, Sexton, Wimer, & Woodward, Int. J. High Perf. Comput. Appl., 2024]. Training and Development Opportunities This project directly funded the efforts of one full time graduate student, and the partial efforts of two postdoctoral researchers. Through this support, it provided professional development of the skills necessary to pursue leading computational and applied mathematics research. In addition to institutionally-provided training in these areas, the PI provides one-on-one mentorship in a variety of skills related to high performance mathematical computing. Specific activities include research-related “reading courses” focusing on the development of modern mathematical software libraries, the use of such libraries within scientific research codes, techniques for running these codes on large-scale supercomputing resources, grantsmanship (including reporting and presentations), and vertical mentoring of junior researchers in the group. In addition to directly training the SMU graduate student under this grant, the SMU PI regularly represents the FASTMath SciDAC Institute as a lecturer in the Argonne Training Program for Extreme Scale Computing (ATPESC). This is an intensive, two-week program where students (primarily graduate students and postdoctoral researchers throughout STEM fields) learn “skills, approaches, and tools to design, implement, and execute computational science and engineering applications on current high-end computing systems and the leadership-class computing systems of the future.” (from https://extremecomputingtraining.anl.gov). Research Dissemination to Communities of Interest Results from research in this project have been disseminated in three primary ways: (a) through publication in scholarly journals, (b) through presentations at disciplinary conferences, and (c) through the release of publicly-available open source software. These are listed below in separate subsections. Journal publications ●​ Rujeko Chinomona and Daniel R. Reynolds, “Implicit-explicit multirate infinitesimal GARK methods,” SIAM J. Sci. Comput., 43(5), A3082-A3113, 2021. https://doi.org/10.1137/20M1354349 ●​ David J. Gardner, Daniel R. Reynolds, Cody J. Balos and Carol S. Woodward, “Enabling new flexibility in the SUNDIALS suite of nonlinear and differential/algebraic equation solvers,” ACM Transactions on Mathematical Software, 48(3), 1–24, 2022. https://doi.org/10.1145/3539801 ●​ Vu Thai Luan, Rujeko Chinomona and Daniel R. Reynolds, “Multirate exponential Rosenbrock methods,” SIAM Journal on Scientific Computing, 44(5), A3265–A3289, 2022. https://doi.org/10.1137/21M1439481 ●​ Alex C. Fish and Daniel R. Reynolds, “Adaptive time step control for infinitesimal multirate methods,” SIAM Journal on Scientific Computing, 45(2), A958--A984, 2023. https://doi.org/10.1137/22M1479798 ●​ Daniel R. Reynolds, David J. Gardner, Carol S. Woodward and Rujeko Chinomona, “ARKODE: a flexible IVP solver infrastructure for one-step methods,” ACM Transactions on Mathematical Software, 49(2), 1--26, 2023. https://doi.org/10.1145/3594632 ●​ John Loffeld, Andy Nonaka, Daniel R. Reynolds, David J. Gardner and Carol S. Woodward, “Performance of explicit and IMEX MRI multirate methods on complex reactive flow problems within modern parallel adaptive structured grid frameworks,” International Journal of High Performance Computing Applications, 2024. https://doi.org/10.1177/10943420241227914 ●​ Alex C. Fish, Daniel R. Reynolds, and Steven B. Roberts, “Implicit-explicit multirate infinitesimal stage-restart methods,” Journal of Computational and Applied Mathematics, 438, 115534, 2024. https://doi.org/10.1016/j.cam.2023.115534 ●​ Cody J. Balos, Marc Day, Lucas Esclapez, Anne M. Felden, David J. Gardner, Malik Hassanaly, Daniel R. Reynolds, Jon Rood, Jean M. Sexton, Nicholas T. Wimer, and Carol S. Woodward, “SUNDIALS Time Integrators for Exascale Applications with Many Independent ODE Systems,” International Journal of High Performance Computing Applications, 39(1), 2024. https://doi.org/10.1177/10943420241280060 ●​ Steven B. Roberts, Mustafa Aggul, Daniel R. Reynolds, Cody J. Balos, David J. Gardner, and Carol S. Woodward, “New Time Integrators and Capabilities in SUNDIALS Versions 6.2.0-7.4.0,” arXiv:2506.14976, 2025. https://doi.org/10.48550/arXiv.2506.14976 ●​ Daniel R. Reynolds, Sylvia Amihere, Dashon Mitchell, and Vu Thai Luan, “Efficient and Flexible Multirate Temporal Adaptivity,” arXiv:2510.14964, 2025. https://doi.org/10.48550/arXiv.2510.14964 Presentations ●​ Invited presentation: Rujeko Chinomona and Daniel R. Reynolds, “High-order implicit-explicit multirate infinitesimal methods for multiphysics applications,” SIAM Conference on Computational Science and Engineering, March 5, 2021. ●​ Invited presentation: Vu Thai Luan, Daniel R. Reynolds and Rujeko Chinomona, “Multirate exponential Rosenbrock methods,” SIAM Conference on Computational Science and Engineering, March 5, 2021. ●​ Invited presentation: David J. Gardner, Daniel R. Reynolds, Rujeko Chinomona, Cody J. Balos and Carol S. Woodward, “Multirate time integrators in SUNDIALS,” SIAM Conference on Computational Science and Engineering, March 5, 2021. ●​ Invited presentation: Daniel R. Reynolds, “Time Integration (with hands-on examples using SUNDIALS),” Argonne Training Program in Extreme-Scale Computing (ATPESC),” August 10, 2021. ●​ Invited presentation: Daniel R. Reynolds, Rujeko Chinomona and Vu Thai Luan, “High-order multirate time integration for multiphysics PDE systems,” SIAM Southeast Atlantic Section Conference, September 18, 2021. ●​ Invited presentation: Daniel R. Reynolds, Rujeko Chinomona, David J. Gardner, Carol S. Woodward, Cody J. Balos and Vu Thai Luan, “Flexible time integration methods for multiphysics PDE systems,” ICERM Workshop on Holistic Design of Time-Dependent PDE Discretizations, January 10, 2022. ●​ Invited presentation: Daniel R. Reynolds, Rujeko Chinomona, David J. Gardner, Carol S. Woodward, and Cody J. Balos, “Flexible time integration methods for multiphysics PDE systems,” Mathematics in Computation (MiC) Seminar, Oak Ridge National Laboratory, April 7, 2022. ●​ Invited presentation: Daniel R. Reynolds, “Time Integration (with hands-on examples using SUNDIALS),” Argonne Training Program in Extreme-Scale Computing (ATPESC),” August 9, 2022. ●​ Invited presentation: Daniel R. Reynolds, Cody J. Balos, Darin R. Ernst, Manaure Francisquez, and Carol S. Woodward, “High-order multirate time integration for tokamak turbulence,” SciCA∂E International Conference on Scientific Computation and Differential Equations, University of Iceland, Reykjavik, July 27, 2022. ●​ Invited presentation: Daniel R. Reynolds, Alex C. Fish, Rujeko Chinomona, Steven B. Roberts, David J. Gardner, Carol S. Woodward, and Cody J. Balos, “Implicit-explicit multirate time integration methods,” Los Alamos Workshop on Time Integration for Multiphysics, Los Alamos, NM, August 8, 2023. ●​ Invited posters: 2023 SciDAC PI Meeting, Rockville, MD, 2023. ●​ Invited presentation: Daniel R. Reynolds, “Flexible time integration methods for multiphysics PDE systems,” SIAM Central States Section Computational and Applied Mathematics Forum, virtual, November 13, 2023. ●​ Invited presentation: Daniel R. Reynolds, “Flexible time integration methods for multiphysics models,” Department of Computer Science, University of Colorado Boulder, April 9, 2024. ●​ Invited posters: 2024 SciDAC PI Meeting, Rockville, MD, 2024. ●​ Invited presentation: Daniel R. Reynolds and David J. Gardner, “Using SUNDIALS on GPU-based HPC platforms,” Center for Advanced Computation, Princeton Plasma Physics Lab, virtual, September 25, 2024. ●​ Invited presentation: Daniel R. Reynolds, “Flexible time integration methods for multiphysics models,” Department of Mathematics and Statistics, University of Maryland Baltimore County, Baltimore, MD, November 22, 2024. ●​ Invited presentation: Daniel R. Reynolds, Sylvia Amihere, Dashon Mitchell, and Vu Thai Luan, “Adaptive multirate time integration in the ARKODE library,” SIAM Conference on Computational Science and Engineering, Fort Worth, TX, March 5, 2025. ●​ Invited presentation: Daniel R. Reynolds, “Flexible time integration methods for multiphysics PDE systems,” VIASM Workshop on Advanced Numerical Methods for Deterministic and Stochastic Differential Equations, Hanoi, Vietnam, June 13, 2025. ●​ Invited seminars: Daniel R. Reynolds, “Runge-Kutta implementation, adaptive and multirate time integration,” VIASM Summer School on Advanced Numerical Methods for Deterministic and Stochastic Differential Equations, Hanoi, Vietnam, June 9-12, 2025. PhD Theses ●​ Alex C. Fish, “Adaptive Multirate Infinitesimal Time Integration,” PhD Thesis, Southern Methodist University, May 13, 2023. Open source software releases ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v5.7.0, January 2021, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v5.8.0, September 2021, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.0.0, December 2021, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.1.0, January 2022, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.2.0, April 2022, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.3.0, August 2022, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.4.0, October 2022, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.5.0, December 2022, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.6.0, July 2023, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v6.7.0, December 2023, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.0.0, February 2024, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.1.0, June 2024, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.2.0, December 2024, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.3.0, April 2025, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.4.0, June 2025, https://github.com/LLNL/sundials. ●​ “SUNDIALS – SUite of Nonlinear and DIfferential/ALgebraic Solvers”, Open source software release, v7.5.0, September 2025, https://github.com/LLNL/sundials. Other Instructional materials and lecture videos for the 2020, 2021, and 2022 Argonne Training Programs in Extreme Scale Computing are publicly available at https://extremecomputingtraining.anl.gov/atpesc-2020, https://extremecomputingtraining.anl.gov/atpesc-2021, and https://extremecomputingtraining.anl.gov/atpesc-2022. In addition, the source code for our “hands on” coding examples that accompany the ATPESC program are available at https://github.com/AMReX-Codes/ATPESC-codes/tree/main/SUNDIALS%2BAMReX