A research poster from the School of Continuous Improvement trims the memory of a standard numerical method for fractional equations and finds that the answer drifts as soon as any of the past is dropped.
A research poster asks how much of its own history a fractional relaxation solver can set aside, and finds the answer is almost none
How much of the past does a calculation need to keep? For one equation at the centre of fractional calculus, a new Slop University research poster gives a plain reply: very nearly all of it.
Nothing Is Safe to Forget is led by Dr Thandiwe Solberg, Senior Lecturer and Convenor of Demo Quarter in the School of Continuous Improvement, with Dr Joost Nwosu, Postdoctoral Fellow, and Eskil Nwabueze, Doctoral Candidate. It brings the School’s steady attention to what a process carries forward into the territory of numerical analysis.
Most of my work is about what gets left out before anyone sees it. This time we left something out on purpose, and the equation noticed at once. I think the School has learned as much from that as the solver has.
— Dr Thandiwe Solberg, Senior Lecturer and Convenor, Demo Quarter, School of Continuous Improvement
Across four fractional orders and a thousand time steps, the team set a widely used solver against an exact reference, shortening its memory stage by stage and tracking how far the answer moved. Keeping only the most recent half of the history left the result between 65 and 95 per cent too small. Even setting aside the ten oldest steps, one per cent of the run, cost up to a fifth of the answer.
The poster takes its place in the University’s growing body of work on systems whose present answers to their whole past. Dr Nwosu said the team would be glad “to compare notes with anyone whose own calculations have been asked to remember a long way back.”
The full poster is available from the University’s research repository under an open licence, doi:10.5555/slop.9y1i65.