concepts · updated 2026-09-25
Topology-Optimized Cooling Geometry — Letting the Algorithm Grow the Form
confidence: medium volatility: warm verified: 2026-09-25fresh
Density-based topology optimization grows cooling geometry from physics instead of designing it: solving Navier–Stokes plus thermal convection over 20–330 million degrees of freedom yields branching, tree-like heat-sink forms that beat conventional fins — with more branches at higher Grashof numbers in 3D, the reverse of the 2D result. Proven at component scale (Alexandersen et al. 2016), cheapened to 5–20 % of the compute by reduced-order methods (Pollini 2020), extended to generative cooling channels (Liu 2026), and never yet applied to building-scale passive cooling: a genuine open frontier and the computational form of Fuller's ephemeralization.
Instead of choosing a cooling geometry and simulating it, topology optimization inverts the workflow: state the physics and the material budget, and let the optimizer grow the form. For natural-convection cooling the grown forms are branching trees — and they outperform every fin array a human would have drawn. The method is proven at heat-sink scale and conspicuously absent at building scale. This article rests on a single ingested source and should be read as a well-grounded pointer rather than a survey.
The foundational result
Alexandersen, Sigmund & Aage (2016, International Journal of Heat and Mass Transfer 100) ran density-based topology optimization on three-dimensional heat sinks cooled purely by natural convection: steady incompressible Navier–Stokes coupled to thermal convection–diffusion with the Boussinesq buoyancy approximation, solved in a parallel framework at 20–330 million state degrees of freedom for laminar flow at Grashof numbers 10³–10⁶. Findings:
- The optimized high-conductivity material distributions are tree-like branching structures, with primary limbs reaching toward the far corners of the domain — nature-inspired form emerging spontaneously from the physics, not imposed by the designer.
- In 3D, branch count increases with Grashof number — the opposite of the earlier 2D results. Larger, more buoyancy-driven systems want more branching, which bodes well for scaling the idea up.
- The authors’ own conclusion: “carefully tailored and complex geometries may improve cooling behaviour considerably compared to simple heat fin geometries.” Growing the design from four seed locations instead of one improved thermal performance by up to 20 % under a 5 % volume constraint.
Follow-on work removed the cost barrier and extended the reach. Pollini et al. (2020) cut the computational cost to 5–20 % of full Navier–Stokes, making desktop-scale 3D topology optimization of natural convection feasible. Liu (2026) coupled generative algorithms — a constrained reaction–diffusion process — to physics-based thermal models to grow novel cooling-channel topologies, achieving >5 °C lower average and >35 °C lower peak temperature than parallel-channel baselines: the method works for channels at intermediate scale, not only for fins.
The building-scale gap
What exists at building scale is parametric optimization and evaluation, not generation. Genetic algorithms tune predefined knobs — window ratios, orientation, wall thickness — for 23–28 % energy reductions (Yang 2025 survey); He et al. (2023) applied a genetic algorithm to the pipe-network topology of a high-rise thermally activated building system for an 11.2 % energy-cost reduction, but that optimizes graph connectivity (which pipes join where), not physical channel geometry; and CFD evaluates given designs without proposing new ones. What does not exist is density-based or level-set topology optimization generating the physical geometry of building ventilation channels, façade cooling forms, or passive-cooling architecture. The methods exist, the need exists, and the connection has not been made.
It is the direct computational continuation of Fuller’s unfinished program of vent-geometry optimization on the cooling dome, and the algorithms’ convergence on minimal-material branching forms is ephemeralization — more cooling with less material — made literal, echoing Fuller’s conviction that nature’s coordinate system holds the key to optimal design.
Where it touches this wiki
The nearest existing practice is Adsorbent Bed Engineering: 3D-printed TPMS lattices are chosen periodic geometries balancing heat against mass transfer — exactly the kind of coupled transport problem a topology optimizer could grow a bed for instead of selecting from a catalogue (where, notably, one of the five lattices loses to plain fins). The wiki’s fractal-bed thesis — hierarchical transport structure wins, surface-area fractals don’t — reads as a prediction of what such an optimizer would find. And like Geometric Blackness, the underlying claim is the same: shape, not material, does the thermal work.
See also
- Adsorbent Bed Engineering — TPMS lattices and the heat-vs-mass-transfer tradeoff; the adsorption-side candidate for grown geometry
- Dome Cooling — Fuller’s unfinished vent-optimization agenda, the building-scale application waiting for these methods
- Geometric Blackness — the optical sibling: geometry substituting for material properties
- Cooling Technologies Index
- Thesis: Fractal/hierarchical bed geometry → SCP — the transport-structure verdict that a bed-scale topology optimizer would be expected to reproduce
Sources
- Topology Optimization for Natural Convection Cooling — Alexandersen et al. 2016 method and branching result, Pollini 2020 cost reduction, Liu 2026 generative channels, the building-scale gap, and the Fuller connection