Event
Control and optimisation of complex flows: gradients or no gradients?
Meeting of the Edinburgh Mathematical Society, with a talk by Taraneh Sayadi from CNAM Paris
Monday 16 November 2026
Tower Building
University of Dundee
Nethergate
Dundee
DD1 4HN
Optimising and controlling complex flows means searching high-dimensional design spaces under the constraints of the governing equations. Adjoint methods make this tractable by giving the exact gradient at low cost, but the advantage has a price. Adjoints are intrusive to build, they need a differentiable solver and heavy checkpointing, and in chaotic or turbulent regimes the gradient itself loses its meaning. Gradient-free methods avoid these problems but quickly become unaffordable as the number of parameters grows.
This talk presents the ensemble-variational (EnVar) method as a bridge between the two families. It keeps the efficiency of a gradient-based descent without ever solving an adjoint. Through examples ranging from simple model problems to high-dimensional and turbulent flows, we will show where the adjoint route breaks down, and how ensembles can take over in terms of parallelism, cost and robustness to chaos.
Jeremy Parker
[email protected]