Let A and B be two Hermitian matrices. The famous Baker-Campell-Hausdorff formula says, exp(A+B)=exp(A)exp(B)exp(−[A,B]/2)exp([A,[A,B]]/6+[B,[A,B]]/3)⋯
In general, exp(A+B)≠exp(A)exp(B). Applying the trace-determinant relation detexpA=exptrA, however, we still have detexp(A+B)=detexp(A)exp(B).
In a similar fashion, the Golden-Thompson theorem states, trexp(A+B)≤trexp(A)exp(B),
the equality holds iff [A,B]=0.
As an application, let's consider −β−1A=ˆp2/(2m),−β−1B=V(ˆq), and [ˆq,ˆp]=i. −β−1(A+B)=ˆp2/(2m)+V(ˆq)=ˆH is the single-particle Hamiltonian. Apparently, Z(β)=trexp(−βˆH) is the partition function of the one-body system. Now consider the right-hand side: trexp(−βˆp2/(2m))exp(−βV(ˆq))=∫dq∫dp2π⟨q|exp(−βˆp2/(2m))|p⟩⟨p|exp(−βV(ˆq))|q⟩=∫dx∫dp2πexp(−βp2/(2m))exp(−βV(q))⟨q|p⟩⟨p|q⟩=∫dx∫dp2πexp(−β(p2/(2m)+V(q)))=∫dx∫dp2πexp(−βH)=Zcl(β).
The last line is the partition function of a classical single-particle system.
Therefore, the Golden-Thompson theorem says, the classical partition function is the upper bounds of the quantum partition function, Z(β)≤Zcl(β). Replace the single-particle operator and single-particle basis with the many-body ones, or even second quantized fields, it is straight-forward to show the general relation between quantum and classical partition functions.