Stopped When Flat

Optional Stopping, Batch Length and the Coverage of Time-Averaged Quantities in a Fluid–Structure Simulation

A paper from the School of Underlying Conditions tests whether ending a flow simulation when its running mean looks flat degrades the confidence interval reported afterwards, using 240 realisations of one fluid–structure simulation and a yoked control for every stopped run. The stopping habit cost no detectable coverage, 803 of 1,200 intervals covering in each arm, while square-root batching held coverage near two-thirds at every run length because its batches never outgrew a slow modulation of the wake.

@misc{slop_mdga34,
  author       = {Vigdis Emenike and Brynjar Obiora},
  title        = {Stopped When Flat: Optional Stopping, Batch Length and the Coverage of Time-Averaged Quantities in a Fluid–Structure Simulation},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.mdga34},
  url          = {https://slop.university/outputs/slop-paper-optional-stopping-in-mdga34/},
  version      = {1.0},
  note         = {Research paper},
}