I’m studying thermodynamics and statistical mechanics from the book by Kerson Huang. I’m having a conceptual difficulty with the notion of thermodynamic equilibrium.

At the beginning of the first chapter, the author states

A thermodynamic state is specified by a set of values of all the thermodynamic parameters necessary for the description of the system.

and

Thermodynamic equilibrium prevails when the thermodynamic state of the system does not change in time.

Then, after stating the laws of thermodynamics, a criterion for finding the equilibrium state is provided: for a closed system, the equilibrium state is the one that maximizes the entropy. This only makes sense if entropy is defined for non-equilibrium states, too. The same thing applies for the minimization of the free energy etc.

However, in the statistical mechanics part of the book, there is the following statement

The entropy in thermodynamics, just as $$S$$ here, is defined only for equilibrium situations.

It's not that this does not make sense: in fact, one talks about “equilibrium thermodynamics” because the subject is only concerned with equilibrium states. Thermodynamic parameters are average values over times longer than the relaxation time, and as such they require equilibrium situations in order to be defined. However, in my understanding, this makes the criterion of maximization of entropy meaningless.

I found this question, which is basically the same I'm asking, but I do not understand the answer. To be more specific, the accepted answer seems to boil down to the following example: one imagines that some “constraint” is placed on the system, that maintains the system in equilibrium. Then the constraint is relaxed, allowing the system to evolve, and then it is placed again. In this way, one can approximate the evolution of the system as a sequence of equilibrium states separated by “infinitesimal transformations”. The state that maximizes the entropy is then the one which does not require a constraint to remain in equilibrium.

I don't understand this explanation because it seems to me that the constraint invalidates the hypothesis that the system is closed. In other words, the presence of the constraint “forces” a state to be an equilibrium state, but then would no longer be described by the equilibrium parameters of the system alone.

I also have something else to add: in the book by Landau and Lifshitz (Course of theoretical physics V) the picture for the law of increase of entropy is that a non equilibrium system can be viewed as a composition of smaller system which are approximately in equilibrium, and for them an entropy can be defined. The total entropy is then the sum of the individual entropies. As the subsystem reach mutual equilibrium, the total entropy increases. This is to me a more logical picture. However, does it correspond to the thermodynamics one? i.e., can such a non equilibrium system be represented by a set of thermodynamic parameters? I guess not, but then the law of increase of entropy would have a different meaning.

In short, I'm terribly confused.

• Given everything that I said (if it is correct), what does the criterion of maximization of entropy actually mean?

• Does the law of increase of entropy require to consider non equilibrium systems?

Edit:

I now think that what I said about the hypothesis of closed system is nonsense, because my definition is wrong (see comments below) and also because the entropy increases when the system is thermally isolated, not when it is closed: from Clausius theorem,

$$\Delta S \ge \int_A^B \frac{\delta Q}{T},$$

and the right hand side is zero when $$\delta Q = 0$$. However, the questions don’t strictly depend on this.

• "I don't understand this explanation because it seems to me that the constraint invalidates the hypothesis that the system is closed." The definition of a closed system is that it cannot exchange mass with its surroundings. That doesn't preclude the possibility system constraints. Jul 20, 2023 at 19:37
• I think I got confused with a closed system in the mechanical sense? Jul 20, 2023 at 19:40
• what do you mean by"mechanical sense"? Jul 20, 2023 at 19:41
• “In nonrelativistic classical mechanics, a closed system is a physical system that doesn't exchange any matter with its surroundings, and isn't subject to any net force whose source is external to the system” by wikipedia. Jul 20, 2023 at 19:42
• can you give link? Never heard about no net force. A closed system in thermodynamics allows for the possibility of both heat and work energy transfer. Jul 20, 2023 at 19:45

Second, the professional theorists of thermodynamics show a lack of respect for experiment I find hard to admire. Experiments are being done today in all sorts of extreme conditions. My friends and colleagues tell me they measure the temperatures in polymer solutions undergoing strong normal-stress effects, they control the temperatures in explosion fronts, they infer the temperatures on the skins of artificial satellites. It would be presumptuous on my part to question the details of the work of these men, and I see no reason to deny that they know their business. The temperatures occurring as values of the temperature function $$\theta(\cdot)$$ in modern theories of materials are intended to represent the numbers these experimenters report and call "temperatures", although their systems bear little likeness to those the old books on thermodynamics describe as being "in equilibrium ", and their processes seem to have little in common with whatever it is the thermodynamicists mean by "quasi-static". I do not claim that we yet know whether or not a particular theory be borne out by any particular experiment, but I do not see why the professionals should forbid us from trying to explain the experiments by a rational theory, approached in that very spirit everyone today regards as the right one for mechanics and electromagnetism.
• Conventionally, entropy maximization can be done along the lines of Gibbs, Tisza, Callen, over purely equilibrium states. That is done by saying that, for example, for 2 variables the postulate is that $S=S(U,V)$ be a concave function of the variables. That is $S(U+\delta U, V+\delta V) + S(U-\delta U, V-\delta V) \le 2S(U, V)$ for arbitrary small deviations $\delta U, \delta V)$. Understood this way there is no need for a process, just the existence of equilibrium states. If the claim is that nothing else can be done then there is Truesdell's answer to it. Jul 20, 2023 at 21:26