At this point, one must remember that this formula is deceptively simple looking--that this is the case becomes clear upon remembering
that does not commute with unless is
a conserved quantity and itself commutes with the system Hamiltonian .
Before we move on to discuss the properties of the linear
response function in more detail, it is perhaps important
to first highlight the following subtlety:
When is a conserved quantity, the linear
response is predicted to be zero by the foregoing formula,
since the commutator
is zero. However, we know that a uniform external field, that couples
to the total magnetization (which is a conserved quantity), does
change the magnetization of the system by an amount proportional
to the field, i.e. within linear response.
So what goes wrong with the formalism developed here?
The answer has to do with the fact that we have assumed
that the initial state of the system at time far in the past
is drawn from an ensemble governed by the Gibbs probability
for the system with Hamiltonian . And since the quantity
to which the external perturbation couples is conserved, there
is no way for the system to change this "incorrect'' probability
distribution and replace it by a slightly perturbed probability distribution
that takes into account the linear effect of the uniform field that couples
to the conserved magnetization. This makes sense: If the magnetization were truly
conserved on the time scale of the experiment, and if the external
field was truly uniform, then it would indeed be impossible for
the external field to change the magnetisation. A better description
of the real experimental conditions is therefore to take a slightly
non-uniform external field varying in time at some low frequency , and measure the linear response at the corresponding small wavevector and frequency . Then, the linear response formalism developed here will give
a sensible result, which should be extrapolated to the uniform
d.c. limit by first sending the frequency to zero keeping the
wavevector non-zero, and then send q to zero. |