Per Arne Rikvold (University of Oslo): "Fluctuations and cascades in models of evolving ecosystems"
Abstract:
It is not only in geophysics and porous-media research that one observes dynamics characterized by quiet periods of variable duration, interrupted by brief, chaotic periods that leave the system significantly changed. In evolutionary biology, this kind of “stick-slip” behavior of extinctions and community reorganizations on multiple scales is known as Punctuated Equilibria.
In this brief talk, I will present results on some simple, individual-based models of ecosystems consisting of interacting species, into which new species enter randomly through mutation or invasion. In these models, time series of population size and number of species (diversity) show power-law behavior over many orders of magnitude. However, the specific powers and the resulting community structures depend on details, including the types of interactions (e.g., mutualistic or predator-prey), and whether the new species are supplied by mutation of existing ones, or by invasion from a different habitat.
Henrik Sveinsson (University of Oslo): "Predicting fracture strength in silica using machine learning"
Abstract:
Machine learning is becoming a powerful tool in materials engineering. It can be used for example to discover new materials and metamaterials and to discover new mechanisms that are hard to find by using intuition.
A particular application of machine learning in materials science is to predict the the mechanical properties of a physical system with weak zones in it. In most cases, the number of possible geometrical possibilities of such weak zones is immense, resulting in a massive design space. This massive design space must be overcome by developing a model that can evaluate systems without running the full physics simulations and a way to obtain a reasonable sample from the design space.
In this talk, I will present our attempt at predicting the fracture strength of weak layers in silica. We use simplex noise to generate a large set of interface geometries.These interface geometries are then used as weak layers in a series of molecular dynamics simulations of fracture in silica. The results from these simulations are used to train a neural network that (i) predicts the fracture strength and (ii) generates new training samples to retrain itself. We run this process iteratively and obtain much better predictions than we get from a more traditional model taking the porosity of the interface into account. We then apply the neural network to a test case where we have good intuition for the physics, to evaluate whether the model makes predictions based on sound physics.
You will find the complete schedule for Njord Seminar Series spring '22 here.
To get news, invitations to seminars and more from Njord, please go here to subscribe to our newsletter.