Basic examples
Before going further with the details on how to compute the two-time
correlator and the power spectrum for a general system described by
the master Eq. (2.71), we can try to learn on the basic
structure and properties of these quantities through some basic
examples. For the isolated modes, the two-time correlator can be
obtained directly by solving the Heisenberg equations:
|
(2.85) |
for the creation/annihilation operators of the fields
,
,
. A free mode (
), propagates
like
and therefore:
|
(2.86) |
is conserved and equal to the initial mean number of
particles. The spectrum is therefore just a
function,
, with the pole at the energy of
the mode
. We can think of the spectrum of the Hamiltonian
as the probability of emission when nothing is allowed to exit. The
time uncertainty is very large, as we must wait an infinite time to
detect a particle. The resonant energies are, therefore, exactly
defined. They are given by the energy difference between eigenenergies
of two consecutive manifolds. The
emission spectrum is nothing
else than the
energy spectrum. We can see this with another
example, adding interactions in the bosonic case, with the
Hamiltonian of Eq. (
2.66),
. Note that
and therefore
is a constant of motion. Then, the operator
depends on the manifold as
|
(2.87) |
and so does the correlator:
|
(2.88) |
This yields a spectrum for the
operator,
|
(2.89) |
which simply weights with the occupation (
) and the
intensity
, the resonances that corresponds to transitions between
each consecutive pair of manifolds:
|
(2.90) |
When the lifetime of the particles in the system is not infinite, the
uncertainty in the energy of the emitted particle increases. This
corresponds to changing the functions by a broadened function
with a linewidth. One can think naively and simply break the
conservation of particles by adding some exponential decay to the
operators
which
results in the expected decay of particles
. This corresponds--and it is often found in the
literature--to adding an imaginary part to the energy
|
(2.91) |
in the free Hamiltonian
and solving the equation
. This procedure is in general
incorrect. This was made clear, for instance, by
Yamamoto & Imamoglu (1999). It leads to unphysical results like the
decay of the bosonic commutation relation:
. Dissipation not only empties the
system but also induces quantum noise due to fluctuations in the
reservoir. A more elaborated method is needed, such as a master
equation with Lindblad terms, that we presented in
Section
2.4. Equivalently, the Heisenberg
equations for the operators
can be upgraded to
the
quantum Langevin equations,
|
(2.92) |
where the quantum white
noise operator is
introduced. This operator is determined by the state of the bath. The
average value of its commutation relations carries the statistic
information which leads to the expected physical results. However,
depending on the system, solving the Heisenberg equations with decay
introduced as an imaginary frequency, can give rise to the same
results as solving the Langevin equations. For example, in the case of
averaged quantities like
or the two-time correlator,
, the correct
expression is obtained. Before we derive them using the proper
methods, let us just write the spectrum this yields:
|
(2.93) |
where we used
|
(2.94) |
This is the
Cauchy-Lorentz distribution with a full width at
half its maximum (FWHM) given by
, which is also the inverse
lifetime of the particles in the system. This shape is the most
commonly found in spectroscopy as it appears when the mechanisms
causing the broadening of the line affects homogeneously all the
emitters.
Elena del Valle
©2009-2010-2011-2012.