reposted from: http://www.fourmilab.ch/hackdiet/e4/feedback.html
Hackers Diet discusses feedback (positive and negative) and control (bang-bang and proportioanal).
The problem isn't losing weight, it's gaining weight. If you hadn't already gained more weight than you're happy toting around, you wouldn't be reading a diet book. Anybody can lose weight…for a while. Many of us have lost weight on numerous occasions, only to put it right back on within a short time. The problem isn't weight loss, it's weight control, and the insights into feedback and control from the thermostat were necessary to allow us, finally, get a handle on our problems with weight.
Rather than temperature, what's being controlled is the number of calories of food eaten. Rather than switching on a furnace or air conditioner, control is accomplished through a signal called appetite. The effect is the same.
It isn't necessary to exactly balance calories eaten with calories burned. If you eat a little too much, the body cranks up the burn rate a little: you feel warmer and more inclined to run up a flight of stairs rather than walk. If you eat a tad less than ideal, you may feel chilly and inclined to curl up with a book under the blankets and get a little extra sleep.
This is the lowest level of proportional weight control, and it works for all of us. It means that there's no need to become obsessed with matching calories eaten and burned; it's only when the balance shifts decisively and consistently to one side or the other of the ideal calorie intake that we get into trouble.
Staying the same weight requires proportional negative feedback control.
But if the proportional appetite control is 'out' a little a life time of dieting is the result.
If the 'i'm stuffed signal is weak then you get really obese.
The temperature charts used as illustrations in this chapter were made with a mathematical model of the thermostat system implemented in Excel. Playing around with this model, entering different values that control the feedback, will give you an excellent feel for the kinds of behaviour a system exhibits in the presence of various kinds of feedback.To experiment with this model, load the Excel worksheet FEEDLAB.XLS.