[drostie]

It's technically not 100% defined. See my comment below for more details about degrees of freedom and energy.

Suppose you have a system with 100 degrees of freedom and 2 units of energy, spread out as (0.01, 0.01, ..., 0.01, 1.01). A bunch of its energy is in one of those hundred degrees of freedom. You can assign it two different temperatures: the temperature 0.01, which would describe how energy will right now flow into the system if you connect it to another system with a bunch of degrees of freedom with their own thermal energy (assuming that the 1.01 degree of freedom is "internal" and doesn't interact directly with the outside world), and the temperature 0.02, which would describe how energy will eventually be spread out and hence how it would eventually share freedom with the outside world.

Temperature is ultimately defined in terms of how our uncertainty about the microscopic state a system is in changes as we add energy to that system. The higher this rate of change of uncertainty, the lower the temperature is -- this is why when you connect two systems of different temperatures, in the process of us becoming more uncertain about the fundamental state of the world, energy "spontaneously" flows from the higher temperature to the lower temperature: the certainty gained from stealing energy from the higher-T one is more than compensated by uncertainty created from pouring that same energy into the lower-T one. (In fact there is a family of systems of "negative temperature" which become less uncertain as you add more energy to them: they are "hotter than the hottest possible temperature" because they will gladly give their energy to any "normal" system in the process of us becoming more uncertain about the world.)

The problem is that if we're certain that some degree of freedom has a given amount of energy that's "special", we have a bunch of different definitions of "temperature" depending on how "adding energy to the system" distributes between the "special" degree of freedom and the "thermal" degrees of freedom.

So the usual process is to just totally separate those degrees of freedom as separate systems, the "thermal" ones have a temperature, the "special" ones do not.

[trentlott]

> there is a family of systems of "negative temperature" which become less uncertain as you add more energy to them . . .

I'm no physicist, just a chemist. What are they?

[drostie]

I mean it's not just one system, but the idea is what I just said.

The classical example is if you have a bunch of magnetic moments in a magnetic field and they do not interact with each other: then stuffing energy into the system requires aligning them against the magnetic field, and this makes the state more ordered. The problem is that these moments are generally in thermal contact with some apparatus that keeps them in place or vibrational degrees of freedom of their centers of mass or so. But you can get this thing to happen in magnetic resonance setups.

Negative temperature states pop up in a lot of strange places, the two that I know more closely are that lasing has this property of "as I dump more energy into the system I get more bosons in the lasing state" and Onsager in 1949 published a little article called “Statistical Hydrodynamics” which sort of went viral for the time, it points out that there is a way to view the instability of turbulent systems as due to negative temperature regimes of the vortices in those systems.