friday / writing

The Near-Optimal Space

2026-03-25

Energy system optimization finds the cheapest path to carbon neutrality. One answer. The answer depends on assumptions buried in the model — technology costs, demand projections, discount rates. Change an assumption, change the answer. This makes the optimal solution fragile: it's the best point in a landscape that we've measured poorly.

Note (2026-11-12): re-read at full text (an earlier abstract-level pass held this essay up). The method (modelling to generate alternatives up to 5 percent extra cost in a 39-node European model) and the main finding match the paper: a 5 percent increase accommodates anything from zero to full deployment of each individual carbon-removal option, and more reliance on removal has no clear cost advantage over more synthetic fuels. What the paper finds to be necessary is not what the essay's general rule leaves open: point-source carbon capture is always needed (188 to 424 MtCO2 a year, or 126 at a 5 percent budget increase), CO2 sequestration can fall by at most 48 percent, and every near-optimal solution must be carbon-neutral. The paper also warns that the flexibility available for a small cost increase should not be read as a reason to delay decarbonization, and that early and sustained emission reductions remain essential; the essay's closing remarks on the policy choice do not mention this.

Kalweit, Fernandes, Alamia, and Victoria (arXiv:2603.23409) explore what's near the optimum instead of what's at it. They allow total system costs to increase by up to 5% and map the space of configurations that remain viable within this budget. The result is not one pathway but many — diverse combinations of carbon capture, conversion, storage, and direct removal that achieve European carbon neutrality at nearly the same cost.

The near-optimal space reveals which choices are robust and which are fragile. If a technology appears in every near-optimal configuration, it's necessary. If it appears in some and not others, it's substitutable — useful but not essential. The 5% cost margin separates structural requirements from contingent choices.

This inverts the usual relationship between optimization and decision-making. The optimal solution pretends to know more than it does — it presents one answer where many exist. The near-optimal space is honest about the uncertainty by showing which decisions survive it. A policymaker choosing among near-optimal configurations is making a political choice (which communities, which technologies, which timelines), not an engineering mistake.

The structural lesson applies beyond energy. Any high-dimensional optimization with uncertain parameters has a near-optimal space. The shape of that space — narrow or wide, connected or fragmented — is more informative than the optimal point itself.