According to the fundamental postulate of statistical mechanics (as far as I know it), we take the (classical) probability for a system to be in any of its microstates to be equal (if the system is in equilibrium and isolated). My question is, how do we count the number of microstates? From what I can see, the number of microstates is usually taken to be the number of linearly independant energy eigenstates of the Hamiltonian for that particular value of energy. However, I don't see why this is the 'number' of microstates. Are we supposed to imagine the microcanonical ensemble as a macroscopic quantum system for which we have measured the energy? In that case the system could be in any one of an (uncountably) infinite number of linear superpositions of its degenerate energy eigenstates. In other words the state vector could be be any vector at all in the degenerate energy eigenspace.
So why are we allowed to count the number of states as the number of linearly independant energy eigenstates, when we don't know that the system would even be in an eigenstate (but rather only in some linear combination)? And as a direct consequence of this method of counting, it seems we assign a non-zero probability only for the system to be in energy eigenstates, and ignore the possibility that it could be in a superposition.
Answer
Summary: the question "what is the probability of state $|\psi \rangle$" is the wrong question to ask, because it's experimentally unobservable. What you really care about is the results of measurements, and the number of possible different measurement outcomes is equal to the number of microstates.
Consider a bunch of noninteracting spin 1/2 particles in thermal equilibrium, and look at a single spin inside.
On the classical level, at thermal equilibrium, the spin of this particle should be a vector with random direction. Now we might ask, on the quantum level, what's the state of this particle?
However, this is the wrong question to ask. Given any spinor $|\psi \rangle$, it's possible to find an axis $\mathbf{n}$ so that $|\psi\rangle$ is the positive eigenstate of $S_{\mathbf{n}}$, i.e. the spin is definitely up along $\mathbf{n}$. So individual elements of the Hilbert space are insufficient to represent a thermal state.
Instead, we need to do the same thing we did in the classical case: replace the state of a system (i.e. $(\mathbf{x}, \mathbf{p})$ classically and $|\psi \rangle$ in quantum) with a probability distribution over those states. One such probability distribution is $$50\% \text{ chance } |\uparrow \rangle, \quad 50\% \text{ chance } |\downarrow \rangle.$$ The miraculous thing is that this distribution is rotationally invariant! That is, the distributions $$50\% \text{ chance } | \mathbf{n} \rangle, \quad 50\% \text{ chance } |-\mathbf{n}\rangle$$ are indistinguishable by measurement for any direction $\mathbf{n}$. Every spin measurement of any $\mathbf{n}$ ensemble, along any direction $\mathbf{m}$, gives a 50/50 result.
This tells us that the correct description of a thermal ensemble of quantum particles can't be an assignment of probabilities to quantum states, because such a construction isn't unique: there are many probability distributions that are experimentally indistinguishable. (The quantity that corresponds to probability distributions "up to distinguishability" is called the density matrix $\rho$.)
With this setup, the answers to your questions are:
- Why don't we consider superpositions? We do! You can't directly ask "what is the probability the system is in state $|\psi \rangle$", because this is not an experimentally observable quantity, as explained above. But you can perform a measurement of an operator $\hat{A}$ with eigenvector $|\psi \rangle$ and eigenvalue $\lambda$, and there's some probability you get $\lambda$.
- Why do we count microstates? This is the dimension of our Hilbert space, or more physically, the number of different numbers you can get while measuring $\hat{A}$. In the example above, the probability of getting $\lambda$ is $1/2$ where $2$ is the number of microstates. We aren't privileging the spin up or spin down states, because you get a 50/50 chance for any spin operator $S_{\mathbf{n}}$.
- Why do we count specifically energy eigenstates? This is just for convenience. What we really want is the dimension of the Hilbert space, and $\hat{H}$ is often easy to deal with/reason about.
No comments:
Post a Comment