I'm not sure whether the scientific tests cover the essential parameters. In fact we have small temperature jumps from the PWM frequency, and possibly much larger jumps from one stable temperature to another one. Since PWM frequency doesn't cost money with an Arduino, and the high frequencies allow for small flattening inductors, this problem is almost academic. More attention should be paid on the start of cooling, and on temperature changes causing mechanical strain. The Peltier should not be stressed by too high temperature jumps at either side, i.e. the temperatures should be given some time for settling on a new setpoint. The PWM duty cycle should not change abruptly, but should increase or decrease with an acceptable slope.