Observation of thermal hysteresis in atomic clocks, and its impact on timekeeping

The definition of the temperature sensitivity of a clock implicitly assumes thermal equilibrium. Test protocols consistent with that definition require long enough soaks at each nominal measurement temperature to enable thermal stabilization prior to each frequency measurement. In many instances, this requirement is not satisfied, either because the required test time becomes too long or because the test protocol is supposed to mirror the actual clock operational environment, in which temperature is continuously changing. Flight Integrity Test (FIT) data obtained for a Milstar rubidium atomic under those conditions indicate that its output frequency cycled over a stable hysteresis loop. We show that such a loop will introduce an additive time error over each thermal cycle.