SINGLE VARIABLE UPDATE FUNCTION

The singlevar update function takes a specification of an update for a
single variable as an argument, and applies this update to the
variables currently being updated, either in a specified order, or in
a random permutaton of this order.  Fast incremental computation using
cached results is done, to the extent that the lpr function used
supports it.

The singlevar function takes arguments lpr, initial, and lpr.initial
with the same meaning as other update functions.  (These are normally
provided automatically by mcmc.)  The seq argument specifies the
sequence variables to update (the default is all, in order), and the
random argument says whether this order is randomly permuted (default
no).  The update argument (which can be given without a name if the
others are named) specifies the update to apply, as either an update
function or a list in which the first element is an update function
(as for the updates passed to mcmc).  The remaining named arguments
specify additional arguments to the update functions, as described
below.

To summarize, the keyword arguments that a user might specify for
singlevar itself are the following:
 
    seq       Sequence of indexes of variables to update.  May be in
              any order, and may contain repetitions.  This sequence
              will be randomly permutted if the random argument is TRUE.
              (Default is 1:length(initial))

    random    If TRUE, the order of updating is chosen randomly. If 
              it is FALSE, components of the state vector are updated 
              sequentially from the first to the last. (Default FALSE.)

If some other keyword argument is not numeric or is numeric and
scalar, it is passed on to the update function every time it is
applied.  If such an argument is numeric and a vector of length the
same as the number of variables updated, single elements of this
vector are passed to the update function when the corresponding
variable is updated.

The value returned by singlevar is a list, which contains an element
named final and possibly lpr, as for any update function, and other
elements that are the concatenations of additional elements that are
returned by the single-variable update function for each element
(re-ordered to the original ordering if random was TRUE).  For
example, if the state has 10 dimensions, seq is the default, and the
single-variable update returns a scalar "apr" value, the result of
singlevar will have an "apr" element that is a vector of length 10,
with the names of the vector elements all being "apr" (though note
that these names may be modified later, by the mcmc function).  Values
for lpr for individual updates are not returned, however, only the
value after the final update.