Correction author: David A. Pearlman
Correction date: 8/29/91


Problem: The description of ISLP=2 behavior in the Gibbs manual was
  misleading. As previously described, one would expect ISLP=2 to
  result in the quantity 
       [ G(forward, current window)-G(reverse, current window)]/2
  being used to calculate the free energy slope.

  In fact, the quantity used if ISLP=2 is specified is:

    [G(forward, prev. window) - G(reverse, curr. window)]/2  (if 0->1)
    [G(reverse, prev. window) - G(forward, curr. window)]/2  (if 1->0)

  in other words, the average of the two redundant values calculated
  over the interval because of double wide sampling.

Consequences: Because the average free energy calculated using ISLP=2 is
  in the direction *opposite* to the direction of the simulation, one can
  not conveniently change the width of the window used to determine this
  free energy (the width was chosen at the previous lambda; the width
  must remain the same at the current lambda to maintain correct double
  wide sampling). This means that if AMXRST is used with ISLP, and a window
  is recalculated because the free energy exceeded the specified threshold,
  the only factor changing in the interval being used to calculate the energy
  is the number of steps of sampling, not the width. This is not expected
  to be very significant as long as AMXRST is not being exceeded very
  frequently.

Fix: The documentation of ISLP has been corrected and clarified (bugfix.3). 
  In addition, an option ISLP=3, which results in the behavior previously
  described for ISLP=2 has been added and documented.

Routines affected: slwadj, mdread, gibbs.me (documentation, see bugfix.3)