Open thread, Jan. 19 - Jan. 25, 2015

Gondolinian

Open thread, Jan. 19 - Jan. 25, 2015 — LessWrong

Comment Permalink

Whereas energy does exist 'out there'

"Out there" are fields, particles, interacting, moving, bumping into each other, turning into each other. Energy is a convenient description of some part of this process in many models. Just like with Jaynes' entropy, knowing more about the system changes its energy. For example, just like knowing about isotopes affects the calculated entropy of a mixed system, knowing about nuclear forces changes the calculated potential energy of the system.

spxtr11y60

I agree with passive_fist, and my argument hasn't changed since last time.

If we learn that energy changes in some process, then we are wrong about the laws that the system is obeying. If we learn that entropy goes down, then we can still be right about the physical laws, as Jaynes shows.

Another way: if we know the laws, then energy is a function of the individual microstate and nothing else, while entropy is a function of our probability distribution over the microstates and nothing else.

1passive_fist11y

Again, this is very different from the situation with entropy. I think you're confusing two meanings of the word 'model'. It's one thing to have an incomplete description of the physics of the system (for instance, lacking nuclear forces, as you describe). It's another to lack knowledge about the internal microstates of the system, even if all relevant physics are known. (In the statistics view, these two meanings are analogous to the 'model' and the 'parameters', respectively). Entropy measures the uncertainty in the distribution of the parameters. It measures something about our information about the system. The most vivid demonstration of this is that entropy changes the more you know about the parameters (microstates) of the system. In the limit of perfect microstate knowledge, the system has zero entropy and is at absolute zero. But energy (relative to ground state) doesn't change no matter how much information you gain about a system's internal microstates.

See in context