Fig. 3

Predictions of large crowd distributions in three new environments. a Experimental observations of dense crowds (124, 219, and 189 flies) in three chambers with different geometries, two with applications of heat creating temperature differences of up to 20 °C. b Measured single-fly probability distributions, N_Ave/N_Tot. c, d DFFT protocol applied to the stair-case geometry. c Measurement of the density for 3 flies is used to determine the vexation, v_b. d Combining this vexation with the extracted quasi 1D frustration from Fig. 2 leads to the high density DFFT prediction. e Comparison of single-fly probabilities for the sparse and dense populations shows significant population shifts as indicated by a correlation coefficient r = 0.73 and a σ_mean = 3.8. f DFFT analysis that incorporates interactions predicts the measured dense population distribution within statistical uncertainty (r = 0.96 with a σ_mean = 1.0). Vertical error bars correspond to s.d. of bin-occupation distributions and horizontal error bars correspond to s.e.m. of the observed density within a given bin