We test statistically the hypothesis that radio pulsar glitches result from an avalanche process , in which angular momentum is transferred erratically from the flywheel-like superfluid in the star to the slowly decelerating , solid crust via spatially connected chains of local , impulsive , threshold-activated events , so that the system fluctuates around a self-organised critical state .
Analysis of the glitch population ( currently 285 events from 101 pulsars ) demonstrates that the size distribution in individual pulsars is consistent with being scale invariant , as expected for an avalanche process .
The measured power-law exponents fall in the range -0.13 \leq a \leq 2.4 , with a \approx 1.2 for the youngest pulsars .
The waiting-time distribution is consistent with being exponential in seven out of nine pulsars where it can be measured reliably , after adjusting for observational limits on the minimum waiting time , as for a constant-rate Poisson process .
PSR J0537 - 6910 and PSR J0835 - 4510 are the exceptions ; their waiting-time distributions show evidence of quasiperiodicity .
In each object , stationarity requires that the rate \lambda equals - \epsilon \dot { \nu } / \langle \Delta \nu \rangle , where \dot { \nu } is the angular acceleration of the crust , \langle \Delta \nu \rangle is the mean glitch size , and \epsilon \dot { \nu } is the relative angular acceleration of the crust and superfluid .
Measurements yield \epsilon \leq 7 \times 10 ^ { -5 } for PSR J0358 + 5413 and \epsilon \leq 1 ( trivially ) for the other eight objects , which have a < 2 .
There is no evidence that \lambda changes monotonically with spin-down age .
The rate distribution itself is fitted reasonably well by an exponential for \lambda \geq 0.25 { yr ^ { -1 } } , with \langle \lambda \rangle = 1.3 ^ { +0.7 } _ { -0.6 } { yr ^ { -1 } } .
For \lambda < 0.25 { yr ^ { -1 } } , its exact form is unknown ; the exponential overestimates the number of glitching pulsars observed at low \lambda , where the limited total observation time exercises a selection bias .
In order to reproduce the aggregate waiting-time distribution of the glitch population as a whole , the fraction of pulsars with \lambda > 0.25 { yr ^ { -1 } } must exceed \sim 70 per cent .