Doshi R, Markl M (2026)
Publication Type: Journal article
Publication year: 2026
Book Volume: 7
Article Number: 165
Journal Issue: 4
Constitutive models define how physical properties depend on evolving state variables, and consequently have a strong influence on computational simulations. However, material descriptions are commonly embedded within solver implementations, limiting their reusability and exchangeability. We introduce a declarative material layer that separates state-dependent material descriptions from numerical solvers and that integrates material behavior into the generated solver code. Material properties are represented symbolically, allowing constitutive relations to be defined independently of discretization methods and reused across different simulation frameworks without solver-specific modifications. A reference implementation demonstrates the compile-time integration of this approach into code-generated solvers for one reference code generation backend. Flow and thermal diffusion benchmarks show that identical constitutive descriptions can be applied consistently across different numerical methods while preserving physical behavior. Performance measurements reveal that the computational impact depends on the interaction between constitutive model complexity and solver characteristics. The proposed declarative material layer opens up the possibility of reusable and solver-independent integration of state-dependent constitutive models into high-performance simulation workflows.
APA:
Doshi, R., & Markl, M. (2026). A Solver-Independent Declarative Material Layer for Compile-Time Integration of Symbolic Constitutive Models. Modelling, 7(4). https://doi.org/10.3390/modelling7040165
MLA:
Doshi, Rahil, and Matthias Markl. "A Solver-Independent Declarative Material Layer for Compile-Time Integration of Symbolic Constitutive Models." Modelling 7.4 (2026).
BibTeX: Download