This summer, among many other things, have read the famous book of Susskind and Lindesay about black hole information paradox. Casually just after finishing ir's reading I have to teach in private tuition (hope I am saying it right ) statistical mechanics (a conventional introduction with the main average topics). In doing that I have revised the foundations of it from the viewpoint of what I had read in the Susskind, and also some other aspects that one must face when triying to apply statistical mechanic reasoning to non physical problems (for example ecology (I collaborated with a guy working in mathematical ecology for a while).
There are a few things that I am thinking about, but for the present entry I will concentrate in an easy mental experiment that, at least apparently, violates the second law of thermodynamics. The key of the violation is the lack of a proper definition of energy in general relativity, but for the present case I don't even need to go into mathematical details about it. My idea was clear, take a situation where that problem in the energy definition rises a paradox that violates the second law. I tried a few strategies that possible also work, but in the end I found a really easy one that I think is simple and representative.
I have found that the easier way of attack is to use the Clausius enunciate: No physical process can transfer heat from a cold body to a warm one. The way to circunvate the law is as follows: Take two "black bodies", for example the canonical ones consisting of a cavity with photons in equilibrium with the walls that are at different temperatures. Now in the warmer one open the also canonical small hole that allow a photon, or a few ones, to scape. This is made in an expanding universe, the photons that have exit form the body travel in that space time loosing energy. Them they arrive at the coldest body. The photons of the warmer body initially had more energy that the ones in the coldest body, but now, after travelling in the expanding space time, arrive to the second with less energy that the photons in the cold body. That means that the cold body becomes coldest after it gets in equilibrium with that photons. Now we make the reverse procedure, we send some photons from the cold (now coldest) body to the warmest one. But, as this is an imaginary experiment, choose to do so when the universe is contracting (for example we make the experiment in the edge of the time when a FRW goes for the expanding to the contracting phase). In the travel on this contracting universe the photons gain energy and when they arrive to the warmest body they can be, if we wait enough, be more energetic than the ones in equilibrium with the warm body and so they actually drive it hotter. As far as we can make the experiment as far as we want from anything else we are in a closed system and in that closed system we have effectively transferred heat from a cold body to a hot one, breaking the second law.
Of course we have not counted the entropy of spacetime, but how could we do so? In the Hawking laws of black holes we learned that classical gravity worked as entropy, the area of a black hole playing the role of entropy. And it is well known how the Hawking radiation, a semiclassical effect (so taking quantum mechanics into play) gave a further argument. String theory (and ulterior works using only geometry and CFT, the kerr/CFT correspondence) gave microscopic support to that correspondence among gravity and thermodinamics. And also are were known the, probably wrong, ideas about gravity as entropy, that have gained a rebirth with the paper of Verlinde about "gravity as en entropic force". But in this mind experiment I don't use nothing special in GR, something like an horizon, or quantum effects. Neither is any claim about saying that gravity is entropy. The whole point is that, if there is no mistake, if you don't know how to count the entropy of the spacetime, in this nonstationary case, you can violate the second law of thermodynamics, and that looks very unfunny, isn't it? ;)
The most similar situation that I know is the famous case of Hawkings telling once (and later felling shame about the idea) that in a contracting universe entropy would go in the opposite direction, and the worries of Sean Carrol and others about the thermodynamic arrow. But as far as I know none made such an explicit case as the one I am presenting here.
Just to avoid some trivial criticisms I clarify that in GR the energy of a body (or a system of bodies) is the time component of a cuadrivector, so it is not an invariant. As E=Q-W (first law) and \[ \Delta S= \Delta Q /T \] entropy also should be some time component of some cuadrivector and that makes it's precise definition somewhat tricky, but as far as I see this experiment could be suited in a single reference system so we don't need to care about that things.
Well, this is the idea, and probably I am making some very trivial mistake, or this simply has already been considered and discarded, but as I don't know for sure than that is the case I present the idea here so anyone can blame me if necessary ;).