TOFLUX: A Differentiable Topology Optimization Framework for Multiphysics Fluidic Problems
Rahul Kumar Padhy, Krishnan Suresh, Aaditya Chandrasekhar · 2025
Read TOFLUX: A Differentiable Topology Optimization Framework for Multiphysics Fluidic Problems by Rahul Kumar Padhy, Krishnan Suresh, Aaditya Chandrasekhar (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-17564.txt.
Introduction
Rahul Kumar Padhy, Krishnan Suresh, and Aaditya Chandrasekhar release TOFLUX, a design framework that lets algorithms sculpt fluid devices automatically, in this 2025 paper. Topology optimization can dream up channels and chambers no human would sketch, ideal for compact, efficient fluid hardware. The barrier is math: grading a design means computing sensitivities, and hand-deriving them for tangled multiphysics is slow and error-prone.
TOFLUX leans on automatic differentiation instead, through the JAX library, so gradients come free with the simulation. Designers can swap objectives, constraints, and material models quickly and watch the optimizer respond. Demonstrations take on thermo-fluidic coupling, fluid-structure interaction, and non-Newtonian flows, the hard cases. The framework also plugs into neural networks, joining classical optimization with modern machine learning for scientific computing.
Catalog record
| CREATOR | Rahul Kumar Padhy, Krishnan Suresh, Aaditya Chandrasekhar |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 36 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-17564
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
TOFLUX: A Differentiable Topology Optimization Framework for Multiphysics Fluidic Problems Rahul Kumar Padhy, Krishnan Suresh, Aaditya Chandrasekhar Abstract Topology Optimization (TO) holds the promise of designing next-generation compact and efficient fluidic devices. However, the inherent complexity of fluid-based TO systems, characterized by multiphysics nonlinear interactions, poses substantial barriers to entry for researchers. Beyond the inherent intricacies of forward simulation models, design optimization is further complicated by the difficulty of computing sensitivities, i.e., gradients. Manual derivation and implementation of sensitivities are often laborious and prone to errors, particularly for non-trivial objectives, constraints, and material models. An alternative solution is automatic differentiation (AD). Although AD has been previously demonstrated for simpler TO problems, extending its use to complex nonlinear multiphysics systems, specifically in fluidic optimization, is key to reducing the entry barrier. To this end, we introduce TOFLUX, a TO framework for flu id devices leveraging the JA X library for high-performance automatic differentiation. The flexibility afforded by AD enables the rapid exploration and evaluation of various objectives and constraints. We illustrate this capability through challenging examples encompassing thermo-fluidic coupling, fluid-structure interaction, and non-Newtonian flows. Additionally, we demonstrate the seamless integration of our framework with neural networks and machine learning methodologies, enabling modern approaches to scientific computing. Ultimately, the framework aims to provide a foundational resource to accelerate research and innovation in fluid-based TO. The software accompanying this educational paper can be accessed at github.com/UW-ERSL/TOFLUX . K eywords Topology Optimization \cdot Fluid \cdot Differentiable Simulation \cdot JAX \cdot Machine Learning 1 Introduction The simulation, design, and optimization of fluidic systems are integral to many engineering and scientific applications [ 1 ] , including aerospace [ 2 ] , biomedical systems [ 3 ] , HVAC [ 4 ] , and automotive industries [ 5 ] . These systems are governed primarily by the Navier-Stokes equations for fluid dynamics, which are often coupled with structural [ 6 ] and thermal interactions [ 7 ] . The complexity of these multi-physics systems makes the design challenging. Thus, designers have increasingly turned to topology optimization (TO) for the design of fluidic systems [ 1 ] . Fundamentally, fluid TO poses the design task as an optimization problem to compute the optimal flow path given specific objectives and constraints [ 1 ] . TO has made the design process more automated and efficient. Topology optimization for fluids was first introduced in the seminal paper by Borrvall and Petersson [ 8 ] for minimizing dissipated power in Stokes flow. Since then, fluid TO has been adapted for a variety of fluid problems, including fluid-structure interaction (FSI) [ 6 ] , conjugate heat transfer (CHT) [ 7 ] , and non-Newtonian fluid flows [ 9 , 3 ] . However, inherent to these advances is a substantial investment of time by researchers in building the necessary simulation and optimization models. Consequently, the entry barrier for new researchers into TO for fluid problems is high. In particular, fluid simulations are plagued by their nonlinearity and high computational cost [ 10 ] . As emphasized in a recent review, addressing this challenge requires educational tools that elucidate multiphysics fluid TO problems [ 11 ] . In addition, fluid TO problems have the added complexity of computing sensitivities. This process involves computing the derivatives of objectives, constraints, material models, and projection or filter operations with respect to the design variables [ 12 ] . The traditional manual sensitivity analysis process can stagnate progress, particularly given the trend toward more practical objectives and simulation models [ 12 ] . This inevitably increases the time required for researchers to build and iterate through optimization models, making the design process inefficient. Recently, with the emergence of machine learning, automatic differentiation (AD) has resurfaced as a powerful tool for sensitivity analysis [ 12 , 13 ] . While previous research has demonstrated to the use of AD for linear, single-physics TO problems [ 12 , 14 , 15 ] , its adoption and application to complex, nonlinear problems, particularly for fluid-based systems, remain scarce. Furthermore, current trends of integrating machine learning with scientific computing dictate that models be end-to-end differentiable [ 16 , 17 , 18 ] . Towards this goal, we employ JAX [ 19 ] , a high-performance Python library designed for end-to-end differentiability. Its NumPy-like syntax [ 20 ] , low memory footprint, and support for vectorized just-in-time (JIT) compilation for accelerated code performance make it an ideal candidate for this task. In addition, we employ implicit automatic differentiation [ 21 , 22 ] to handle the nonlinearities present in the system. We showcase the framework’s capabilities on various fluid-based TO problems, including fluidic devices, FSI, CHT, and non-Newtonian flows. For educational purposes, we focus on simple, introductory studies in these domains. We also demonstrate integration with machine learning for design representation, showcasing the framework’s utility for timely applications. Finally, we place a great emphasis on a well-documented, consistently styled code with well-illustrated Jupyter notebooks, which are available as supplementary material to this educational paper. The readers are urged to refer to the notebooks for an involved discussion. 2 Related Work This section surveys the field of TO for fluid-based problems. The survey is structured to reflect the central contribution of this paper: the promotion of a differentiable simulation framework for designing fluidic devices. Therefore, we organize our review into two key areas. The first area covers the foundations and applications of fluid-based TO. The second area details the principles of differentiable simulation and its role in computing design sensitivities. 2.1 Topology Optimization of Fluidic Devices Fluidic devices play a central role in systems involving material and energy transport. The optimization of such transport systems is critical in established fields such as automotive [ 23 ] , HVAC [ 24 ] , and aerospace [ 25 ] , as well as in emerging areas such as soft robotics [ 26 ] and battery technology [ 27 ] . Given their complex and multiphysics nature, TO is emerging as a method for the automated design of these fluidic systems. This review provides an overview of seminal work in fluid-based TO, with a particular emphasis on non-Newtonian fluids, conjugate heat transfer, and fluid-structure interaction. For a more comprehensive review of the field, we refer readers to Alexandersen et al. [ 1 ] . The foundation of TO for fluid flow was laid by Borrvall and Petersson, who formulated a density-based approach to minimize pressure drop [ 8 ] . Their approach utilized the Stokes equation with a Brinkman–Darcy penalization for low-Reynolds number conditions. This methodology was later extended by Gersborg-Hansen et al. [ 28 ] to applications in microfluidics and micro-electro-mechanical systems. Subsequent research has focused on refining this approach. For instance, Guest et al. [ 29 ] proposed a Darcy–Stokes finite element model in lieu of the Brinkman penalization to produce distinct void-solid topologies without artificial material regions. Furthermore, Haubner et al. [ 30 ] promoted the use of fractional-order Sobolev spaces to improve numerical stability and convergence. Another refinement involved geometric representation, where Jensen et al. proposed anisotropic mesh adaptation for higher-fidelity descriptions [ 31 ] . Alongside these numerical improvements, the physical modeling itself has also evolved. For instance, Wiker et