London Fluids Colloquium

The London Fluids Colloquium aims to bring together researchers in Fluid Dynamics in London. The Colloquium will take place once a term at a London Institution on a Friday at 2-4pm. Each colloquium will feature short early career researcher talks and one seminar length talk from an invited researcher. There will be refreshments following the talks and an opportunity to discuss and meet other fluids researchers. The series is open to researchers from across London and is free to attend.

To receive emails about upcoming London Fluids Colloquium events please subscribe to our mailing list via this link.

Upcoming colloquium

The next colloquium will be held at UCL, Friday 23rd May 2025 2pm. For catering purposes, if you plan on attending please enter your name in this form: https://forms.cloud.microsoft/e/uEss0CHyqf.

Seminar speaker abstract:

Matthew Juniper (Department of Engineering, University of Cambridge)
Information flows: probability, inference, and information in fluid mechanics.

Abstract: John von Neumann is often quoted as saying "with four parameters I can fit an elephant, and with five I can make him wiggle his trunk." The implication seems to be that physical models should contain only a handful of parameters. A century later, however, we are happy to use physics-agnostic neural networks containing millions of parameters. What would von Neumann say? How should physical modellers respond?

In this talk, I will show that von Neumann's quote is more nuanced than it sounds. I will then frame a response within a Bayesian framework, in which physical principles such as conservation of mass, momentum, and energy are treated as high quality prior information, with quantified uncertainty, expressed as PDEs or low order models. The information content of relevant data can then be quantified and the likelihood of different candidate models can be compared after the data arrives. I will show how Bayesian inference becomes computationally tractable in fluid mechancis when combined with adjoint methods. I will demonstrate this through assimilation of 3D Flow-MRI data in complex geometry into Finite Element CFD. The main message of the talk is "keep the physics in the model if you can."

Bio: Matthew Juniper is Professor of Thermofluid Mechanics at the University of Cambridge. His research interests are flow instability, adjoint-based sensitivity analysis, shape optimization, and physics-based Bayesian inference, particularly when accelerated with adjoint codes. He is PI of the UK Fluids Network and the Hub for EPSRC's National Fellowships in Fluid Dynamics.

Location

The colloquium will take place in Harrie Massey Lecture Theatre, 25 Gordon Street.

25 Gordon Street (UCL Mathematics / Students Union) is a short walk from various stations: Euston Square (3 min), Euston (5 min), Warren Street (7 min), Russell Square (13 min).

Harrie Massey Lecture Theatre is on the ground floor. From the main entrance, walk past the shop and lifts to the back of the building and turn left.

Organisers

Edwina Yeo (Department of Mathematics, University College London)

Catherine Kamal (Department of Mathematics, University College London)

Gunnar Peng (Department of Mathematics, University College London)

Ory Schnitzer (Department of Mathematics, Imperial College London)

Sponsors

We gratefully acknowledge support from the EPSRC National Fellowships in Fluid Dynamics scheme (Grant number EP/X027902/1) and the Department of Mathematics at Imperial College London.