How long should a lockdown-relaxation cycle last?

epidemic-modelingoptimizationcovid-19adaptive-triggering

Abstraction: Mathematical optimization of lockdown-relaxation cycle triggers and lengths

Key points:

Imperial College adaptive triggering policy: alternate between suppression (cases decline) and relaxation (cases grow) using ICU demand thresholds of 100 and 50 per week
Counterintuitive result 1: for fixed trigger ratio, dividing both triggers by 2 decreases infection damage without increasing lockdown damage — triggers should be set as low as possible
Counterintuitive result 2: for fixed trigger ratio, cycles should be kept as short as reasonably possible (model breaks down at timescales shorter than the exponential growth/decay turnaround time)
Optimal policy uses at most one measure for each phase (relaxation and suppression); more phases add no benefit under constant-damage assumption
If multiple measures used per phase, relaxation measures should grow progressively more lenient; suppression measures should start strict and ease
Author argues result 1 is a reductio ad absurdum of adaptive triggering, supporting "Hammer and Dance" approach (suppress to near-zero, then use targeted testing/tracing)

Connections: Tim Gowers · Imperial College London · Epidemic Modeling · Optimization · Dynamical Systems