MC-QUANTITIES:  Quantities from log files relating to Monte Carlo simulation.

The quantities below relating to Markov chain Monte Carlo can be
obtained from log files (eg, for use in xxx-plt).  Note that the
generic quantities documented in quantities.doc will also be
available, as will quantities specific to the particular Markov chain
application being used.

    T   Temperature used (see xxx-mc.doc, NOT from the tempering schedule)
    E   Potential energy at end of iteration
    E0  Potential energy at end of iteration at inverse temperature of 0
    E1  Equal to E - E0
    K   Kinetic energy at end of iteration
    H   Total energy at end of iteration (sum of E and K)
    D[n] Change in total energy for last state proposed (up to max of n)

    i   Inverse temperature being used (from the tempering schedule)
    I   Index of current inverse temperature value in schedule
    j   Direction of change for inverse temperature
    J   Higher temperature for attempted transition (meaningful only if
        last operation in iteration was sim-temp)
    Q[n] Importance weight, or log of importance weight (see below)
    F[n] Factor for estimating ratio of normalizing constants using
         tempered transitions (see below).

    d   Heatbath decay factor used in this iteration
    f   Stepsize factor used in this iteration
    m   Point last moved to (zero is starting point)
    r   Rejection rate for this iteration
    e   Average number of evaluations per slice sampling update this iteration
    k   Cumulative cpu usage in minutes.

    qn  Component n of position
    pn  Component n of momentum
    sn  Stepsize selected by application for component n

None of these quantities can be used with a range; 'q', 'p', and 's'
must have a modifier, 'D' and 'Q' may have a modifier, 'E' may only
have a modifier of '0' or '1', the others may not have a modifier.

The 'Q' quantity is the importance weight, which will always be one
except during annealed importance sampling.  Note that these weights
might sometimes be so extreme that this value overflows or underflows.
If a modifier of 0 is given to 'Q', the value is the log of the
importance weight, which is much less likely to be out of bounds.  If
a modifier n greater than zero is given for 'Q', the value is the log
of the importance weight unless that is less than -n, in which case
the value is -n.

The 'F' quantities are factors obtained from the last tempered
transition in the iterations, which can be averaged to estimate the
ratio of normalizing constants for the distribution at temperature one
to that of the distribution at the highest temperature.  The 'F1'
quantity is based on only the first half of the tempered transition.
The 'F2' quantity is based on both halves if the tempered transition
is accepted, and is otherwise based on just the first half.

Momenta of zero are assumed if none exist.

            Copyright (c) 1995, 1998 by Radford M. Neal