The CRM has had four directors in forty-two years. On Monday 19 October, the fourth, Carme Cascante, opened the 2026–2027 academic year before the centre’s community, and after the first research talk the portraits of the other three were unveiled in the new Directors’ Gallery.

Cascante’s welcome went over the centre’s figures for 2025. The CRM counted 171 members last year, and its output included 140 scientific papers and 24 doctoral theses. The centre organised 54 scientific activities, eight of them conferences and 15 advanced courses, with 1,397 participants from 66 countries. It also passed the CERCA evaluation of the 2021–2024 period successfully, and the external evaluators praised both its scientific results and its working environment.

Cascante also highlighted two activities for the forthcoming year. A lecture series on the mathematics of machine learning opens on 27 October at the Universitat de Barcelona, with Francis Bach (INRIA and École Normale Supérieure) as its first speaker. The other is a BGSMath course in November on formalising mathematics with the proof assistant Lean and AI tools, taught by Michail Karatarakis (Radboud University Nijmegen and Axiomatic AI). The CRM Faculty Colloquium, where researchers affiliated with the centre present their work to the whole mathematical community, started in February and has a 2027 session planned with Jezabel Curbelo, Joana Cirici and Eva Miranda. Maria Bruna, from the University of Oxford, visits the CRM this month through the María Wonenburger call, shared by five SOMMa centres.

The last part of her talk dealt with research integrity: the European Code of Conduct for Research Integrity, the CERCA Code of Conduct and the CRM’s rules on signatures and institutional affiliation. An open round of comments and questions followed.

Pau Varela, the CRM’s Communication Coordinator, then presented the Communication Plan 2026–2030. Born from the CRM Strategic Plan 2026–2030, it gives the centre a defined communication strategy for the first time in more than forty years, with eight strategic objectives and 13 actions. Varela also reviewed what the Communication & Outreach Department has built since early 2022, when it became a stable unit with the start of the María de Maeztu programme. Its projects include the outreach blog El Gegant del π, the video interview series Mathematics in Conversation (43 people interviewed so far) and the podcast Ments Meravelloses, which passed 70,000 views in its first year on a YouTube channel that until 2023 was an archive of recorded talks. He closed with an invitation to take part.

Diego Vidaurre, who joined the CRM’s Computational and Mathematical Neuroscience group through the ATRAE programme, gave the first research talk of the morning. Under the title “Machine learning and AI models for explaining perception”, he started from the difference between predicting and explaining. A falling barometer predicts a storm without causing it (both follow from the same drop in pressure), and a model trained on enough data can predict well on correlations alone. The field’s two standard tools (encoding models, which predict brain activity from a stimulus, and decoding models, which do the reverse) both have limits as explanations, and decoding results in particular have no mechanistic counterpart. Vidaurre’s answer is Genephys, published in eLife in 2024, a generative model of the brain’s electrophysiological signal that describes the mechanisms producing it. That brings the analysis closer to explanation as the philosopher James Woodward defined it: saying what would happen to an outcome if its causes were different. As one of his final slides acknowledged, it is still based on correlations, “but the gap is smaller”.

Between the first and second talks, the CRM inaugurated its Directors’ Gallery (Galeria de Direcció), conceived as a tribute to the centre’s history and leadership. It brings together the portraits of the three directors who preceded Cascante: Manuel Castellet, founder of the CRM and its director from 1984 to 2007; Joaquim Bruna, from 2007 to 2016; and Lluís Alsedà, from 2016 to 2024.

After the coffee break, Sergey Tikhonov, ICREA research professor at the CRM, spoke on “Uncertainty principles”. The best-known member of the family is Heisenberg’s: the position and momentum of a particle can’t both be known exactly. In Fourier analysis, a field that goes back to Joseph Fourier’s Théorie analytique de la chaleur (1822), the principle takes a precise form: a function and its Fourier transform cannot both be sharply concentrated. A police radar gun and a speeding ticket also made an appearance, under the question of whether the principle really works.

Simple, careful models outperform complicated, inconsistent ones.

Tim Myers

Tim Myers (CRM) gave the last talk, “Mathematical modelling of adsorption: From clean air to corrosion protection”. Adsorption, the process by which molecules stick to the surface of a solid, is used in water purification and carbon capture. Myers showed how a careful reduction of the equations for an adsorption column yields a simple formula, with a single fitted parameter, that predicts when the pollutant will emerge at the outlet. The standard Bohart–Adams model, from 1920, only fits the data by treating known quantities as free parameters. In steel pipelines the adsorbed molecules are, for once, wanted: corrosion inhibitors form a protective film on the metal, and a multilayer model explains why it keeps working for a while after dosing stops.

The event closed, as the programme put it, with a dinar de germanor: a standing lunch for the whole community.