🔢 mathematics

Implicit-adjoint finite-volume topology optimization of two-dimensional conjugate heat transfer

This paper presents a transparent, self-contained, and numerically stable two-dimensional finite-volume topology optimization framework for conjugate heat transfer that utilizes a staggered grid, a divergence-consistent advection operator, and exact discrete adjoints via the implicit function theorem to ensure reproducible and verifiable thermofluidic designs.

Original authors: Sam Yang

Published 2026-08-21

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Sam Yang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to cool a tiny, powerful computer chip or design a heat exchanger for a spacecraft. The challenge is not just moving heat away, but deciding exactly where to place the solid metal that conducts the heat and where to carve out the channels for the cooling fluid to flow. These two materials must work together perfectly within a very small, fixed space. If the metal is in the wrong place, heat gets trapped; if the fluid path is blocked or too winding, the coolant cannot reach the hot spots. For decades, engineers have tried to solve this by drawing designs by hand or using simple rules, but the best solution often looks nothing like what a human would intuitively sketch. It requires finding a balance between the solid's ability to conduct heat and the fluid's ability to carry it away, all while fighting against the resistance the fluid feels as it squeezes through narrow gaps.

This is the problem tackled by a new study that uses a powerful mathematical tool called topology optimization. Instead of starting with a shape and trying to improve it, this method starts with a blank block of material and asks a computer to figure out which parts should remain solid and which should become empty space for the fluid. The computer does this by testing millions of tiny variations, constantly adjusting the layout to find the most efficient path for heat and flow. However, for these computer-generated designs to be trusted, the underlying math must be flawless. If the computer makes a tiny error in how it calculates the flow or the heat, the final design could look perfect on screen but fail completely in the real world. The researchers in this study set out to build a completely transparent and error-free version of this process, creating a digital laboratory where every step can be checked and verified.

The team developed a new way to run these simulations on a two-dimensional grid, which is like a digital checkerboard where each square holds a piece of the puzzle. They focused on a specific type of fluid flow that moves slowly and smoothly, often found in micro-scale devices, and coupled it with the way heat moves through both the solid and the fluid. A major hurdle in previous attempts was that the computer's way of calculating how heat moves with the fluid sometimes created fake, artificial heat sources that didn't exist in reality. These ghostly errors could trick the optimization algorithm into designing a shape that looked good but was physically impossible. The researchers solved this by carefully constructing their math so that if the temperature was the same everywhere, the computer would correctly calculate that no heat was moving at all, even if the fluid was flowing. This ensured that the "ghost heat" never appeared, allowing the computer to find the true best design.

To make sure their new method was working correctly, the team did not just run the optimization and hope for the best. They built a rigorous system of checks, similar to a safety inspection for a bridge. Before accepting any design as a final answer, the computer had to prove that the fluid was flowing correctly, that the heat was moving as expected, and that the total amount of solid material used matched the strict limits set by the designer. If the numbers were even slightly off, the design was rejected, and the computer tried again. They tested this system on four different types of problems: one where heat had to travel from a single point through a tree-like structure of metal, another where fluid flowed through a porous sponge-like material, a third where solid metal and fluid worked together to cool a hot surface, and a fourth where heat was pulled from a hot wall to a cold one. In every case, the computer found designs that were significantly better than standard, uniform layouts.

The results showed that the new method could create highly efficient, complex shapes that a human engineer would likely never imagine. For the heat-conducting tree, the computer grew a branching structure that moved heat away much faster than a solid block of metal. For the fluid flow problems, it carved out channels that minimized resistance while maximizing cooling. The study confirmed that these designs were not just lucky guesses but were the result of a mathematically precise process. The researchers also noted that while their method was incredibly accurate for two-dimensional designs, it was limited to slow, smooth flows and did not yet handle the chaotic turbulence found in faster-moving fluids or three-dimensional objects. Nevertheless, by providing a completely open and verified code, they have given other scientists a reliable foundation to build upon. This work does not just offer a new design; it offers a new way to trust the designs that computers create, ensuring that when engineers build the next generation of compact, high-efficiency thermal systems, they are building on a solid, error-free mathematical truth.

Technical Summary: Implicit-Adjoint Finite-Volume Topology Optimization of Two-Dimensional Conjugate Heat Transfer

Problem Statement
Designing compact, high-efficiency thermal architectures requires balancing solid conduction, fluid convection, and flow resistance within constrained physical envelopes. While density-based topology optimization (TO) offers a systematic framework for synthesizing these coupled layouts, the reproducibility and numerical stability of resulting designs are often compromised by the underlying discrete solvers and adjoint sensitivity mechanics. Existing literature frequently relies on complex three-dimensional Navier–Stokes formulations, turbulence closures, and proprietary or non-auditable implementations, leaving a gap for transparent, verifiable reference stacks for conjugate heat transfer (CHT) problems.

Methodology
This work presents a self-contained, two-dimensional finite-volume formulation on a staggered Marker-and-Cell (MAC) grid. The approach integrates the following core components:

Governing Equations & Interpolation: The system couples design-dependent energy transport with Stokes–Brinkman or Darcy flow at a fixed solid volume. Material properties (thermal conductivity and permeability) are interpolated using the Rational Approximation of Material Properties (RAMP) scheme. A compact linear cone filter regularizes the raw design field, which is then projected to near-binary states via a hyperbolic tangent function with a continuation parameter $\beta$ .
Discrete Advection & Nullspace Preservation: A critical methodological innovation is the construction of the discrete advection operator to satisfy the identity $\mathbf{u} \cdot \nabla T = \nabla \cdot (\mathbf{u}T) - T(\nabla \cdot \mathbf{u})$ cellwise. This ensures that uniform temperature fields remain exact discrete nullspaces, preventing spurious artificial thermal sources in porous Brinkman domains where iterative continuity solvers may exhibit small non-zero residuals ( $\nabla \cdot \mathbf{u} \neq 0$ ).
Implicit Adjoint Sensitivity: Rather than unrolling iterative solver loops, reverse-mode derivatives are evaluated via the implicit function theorem. This yields exact discrete adjoints with bounded memory requirements. The adjoint system is solved using direct dense factorization for smaller grids or Schur-reduced block solvers for larger meshes.
Optimization Algorithm: Design evolution is driven by a projected-gradient scheme with $\beta$ -continuation. Volume conservation is enforced strictly via a one-dimensional bisection search on the projected density field. The algorithm employs a "keep-best" strategy within each $\beta$ stage and rejects iterates that fail rigorous predeclared evidence gates regarding residual convergence, mass conservation, and numerical finiteness.

Key Contributions
The primary contribution of this work is not the novelty of implicit differentiation or conjugate layout concepts themselves, but the provision of an inspectable, fully deterministic reference stack. Key features include:

Transparent Implementation: A complete 2D finite-volume stack comprising a nullspace-preserving energy advection scheme, operator-split Brinkman formulation, and implicit residual adjoint.
Rigorous Verification: Systematic validation through the Method of Manufactured Solutions (MMS) and directional Taylor remainder tests, confirming second-order convergence for diffusion and first-order for upwind advection.
Automated Evidence Gates: A protocol where candidate designs are only accepted and published if they satisfy strict numerical criteria (e.g., relative energy residual $\le 10^{-3}$ , volume error $\le 10^{-8}$ ), ensuring that reported results are physically admissible and numerically converged.
Reproducibility: All code, verification test suites, and machine-readable optimization histories are openly available, enabling independent researchers to regenerate every published layout and diagnostic metric.

Results
The framework was applied to four representative thermofluid benchmarks:

Volume-to-Point Conduction Tree: Optimized a hierarchical conductive structure ( $v^*=0.30$ ), achieving an objective improvement from a baseline of $-0.0571$ to $-0.0206$.
Darcy Convection: Designed a low-resistance fluid pathway ( $v^*=0.45, Pe=40$ ), improving the objective from $-0.0311$ to $-0.0295$. The design exhibited a high fraction of intermediate gray densities (72.7%), attributed to the uniform conductivity allowing diffuse transition zones to suffice for hydraulic channeling.
Stokes–Brinkman Conjugate Cooling: Coupled solid conduction and Stokes flow ( $v^*=0.45, Pe=40, \Delta p=20$ ). The optimized design achieved an objective of $-0.0767$ (vs. baseline $-7.8958$) with a nearly binary interface (1.0% gray fraction) and satisfied strict residual norms (Stokes residual $\le 5.70 \times 10^{-6}$ ).
Dirichlet Conduction Sandwich: Maximized boundary heat flux ( $v^*=0.40$ ), increasing extraction from $20.953$ to $38.307$.

Across all cases, the optimization successfully converged to designs satisfying all evidence gates, with multi-seed sensitivity studies confirming robustness.

Significance and Claims
The paper modestly claims to provide a verifiable reference stack for conjugate thermofluidic topology optimization. It emphasizes that the presence of intermediate gray densities in converged layouts is a physical consequence of the filtering and projection scheme, not a solver failure. By explicitly avoiding advanced 3D turbulence models or complex optimizers like MMA in favor of a transparent, 2D finite-volume approach with exact discrete adjoints, the work aims to establish a baseline for independent verification. The author asserts that this formulation allows researchers to "regenerate and verify every published layout and diagnostic metric," addressing the scarcity of open-source, easily auditable reference implementations in the field. The work is intentionally circumscribed to 2D, inertialess creeping flow, and first-order upwind advection, acknowledging that extensions to 3D and Navier–Stokes flows remain outside the current scope.

Technical Summary: Implicit-Adjoint Finite-Volume Topology Optimization of Two-Dimensional Conjugate Heat Transfer

More like this