A method is introduced for studying thermal relaxation in multiminima
energy landscapes. All the configurations connected to a given energy
minimum by paths never exceeding a chosen <<energy lid>> are found, ea
ch equipped with a set of pointers to its neighbours. This information
defines a phase space pocket around the minimum, in which the master
equation for the relaxation process is directly solved. As an example
we analyse some instances of the Travelling-Salesman Problem. We find
that i) the number of configurations accessible from a given suboptima
l tour grows exponentially with the energy lid, ii) the density of sta
tes within the pocket also shows exponential growth, iii) the low-temp
erature dynamical behaviour is characterized by a sequence of local eq
uilibrations in increasingly larger regions of phase space and finally
iv) the propagator decays algebraically with a temperature-dependent
exponent. These observations are related to both theoretical models an
d experimental findings on relaxation in complex systems.