LATEST NEWS
The rapid growth of machine learning and other data-driven modeling approaches is creating new opportunities for accelerating scientific and engineering simulations and representing complex physical processes that are difficult or expensive to model using conventional approaches. Data-driven surrogate models, reduced-order models, neural operators, learned closures, and related techniques have demonstrated impressive performance across a range of applications. However, their use as standalone replacements for conventional high-fidelity models remains challenging for many complex multiscale and multiphysics applications, particularly when simulations must operate in predictive regimes involving parameter ranges, physical phenomena, or nonlinear behaviors that are inadequately represented in the training data.
An emerging alternative is to move beyond the choice between conventional and data-driven modeling and instead develop rigorous hybrid computational approaches that integrate physics-based and data-driven models within a unified simulation framework. Here, hybrid simulation refers to approaches in which data-driven models are coupled with or embedded within conventional physics-based models, allowing the complementary capabilities of both modeling paradigms to be exploited. Such approaches can serve two complementary purposes. First, they provide systematic and rigorous mechanisms for introducing data-driven models into established high-fidelity modeling and simulation toolchains without requiring the entire simulation to be replaced by a learned surrogate. Second, coupling data-driven models with trusted physics-based models can improve their predictive capability, robustness, and trustworthiness by restricting their use to regimes or regions in which they are appropriate and relying on conventional models when they are not. Importantly, effective hybridization requires more than simply connecting independently developed models: naive coupling can introduce inaccuracies, instabilities, or violations of important physical and mathematical properties. Developing coupling strategies that preserve relevant structure—such as conservation laws, stability, constraints, symmetries, or other problem-specific properties—is therefore an important component of reliable hybrid simulation.
This special issue of Computing in Science & Engineering will explore emerging methods, software, and applications for the rigorous integration of conventional physics-based and data-driven models in scientific simulation. Of particular interest are approaches that enable data-driven models to function as components of larger predictive simulations rather than as standalone surrogates. The issue will highlight both methodological advances and practical experiences integrating such approaches into scientific computing workflows and production-scale simulation environments.
Topics of interest include, but are not limited to:
The special issue welcomes both research articles presenting new hybrid coupling methodologies, algorithms, software, and applications and perspective or position articles examining the opportunities, limitations, and open challenges associated with hybrid computational modeling. Position articles might address questions such as when hybrid modeling is preferable to standalone data-driven or conventional simulation, what is required to establish confidence in hybrid predictions, and what algorithmic and software capabilities are needed to make hybrid approaches practical for large-scale scientific computing.
Consistent with the mission of Computing in Science & Engineering, submissions should emphasize the broader computational ideas, implications, and lessons of the work and should be accessible to a general computational science and engineering audience. Research articles should therefore focus on the key methodological innovations and computational insights rather than providing the level of specialized mathematical or algorithmic detail typically expected in a disciplinary journal.
For author information and guidelines on submission criteria, visit the Author’s Information page. Articles should be between 2,400 and 6,250 words, including all main body, abstract, keyword, bibliography (25 references or less), and biography text. Each table and figure counts for 250 words.
Please submit papers through the IEEE Author Portal and be sure to select the special issue or special section name. Manuscripts should not be published or currently submitted for publication elsewhere. Please submit only full papers intended for review, not abstracts. If requested, abstracts should be sent by email to the guest editors directly.
