Adam Paxton

profile image

Adam Paxton

Postdoctoral Researcher

I am working with the Predictability of Weather and Climate group. Our group is developing methods of inexact computing to improve the efficiency of mathematical models for weather and climate forecasting, by making use of new hardware for low-precision arithmetics which is now becoming available (in large part due to demand from the machine learning and smart-phone industries). Our work is part of an industry-wide push to develop faster and more accurate weather and climate models, to be run on next generation, exascale computers. Currently, my research is focused on the efficacy of forecasting over long time-scales---such as those relevant for climate prediction---using low precision arithmetic. In particular, I am interested in the stability of numerical algorithms for climatological modelling when run in low precision---i.e. on the effects of rounding error propagation in large codes---and on ways to mitigate adverse effects. This gives rise to some interesting theory from the standpoint of numerical analysis/modelling of nonlinear PDEs, because the way in which rounding errors are propagated in a numerical scheme is intricately related to the underlying dynamics of the system in question. My background is in pure mathematics. For some maths, visit my webpage here https://adampaxton9973.github.io/my-page/ :)