********> bugfix.61
Correction Author: David A. Pearlman
Correction Date: 03/24/94

Programs: GIBBS

Severity: Serious

Problem: 

       A free energy calculation restart can result in erroneous
       results if ALL the following conditions are met:

       1) The free energy calculation is run in more than one piece,
          resulting in a "restart" run (IREST=1).

       2) Dynamically Modified Windows are being used where the
          the width of each window can vary (if DMW is requested,
          but the ability to change width at each window is defeated
          by setting DLMIN=DLMAX, no problem will occur).

       3) Double-wide sampling is being used (IDWIDE=0).

       4) The calculation is a "windows" type simulation, NOT
          a Thermodynamic Integration calculation.

       In any case, this bug will only affect the energy calculated in the
       direction OPPOSITE to the direction in which the simulation
       was run (see "Affects" below).


Affects: 

       If all the above conditions are met, then the free energy
       calculated in the direction OPPOSITE to that in which the
       simulation is run may be incorrect. That is, if the simulation
       is run 0->1, the total energy reported as "reverse" may be
       incorrect. If the simulation is run 1->0, the total energy
       reported as "forward" may be incorrect. In either case, the
       energy reported for the same direction as the simulation
       ("forward" for 0->1 or "reverse" for 1->0) will be CORRECT.

       The degree of error in the reported free energy in the opposite
       direction will depend on the difference in d_lambda in the
       direction of the simulation over the last window before the restart
       and the first window after the restart. If these two values of
       d_lambda are close, the error in the reported free energy will
       probably be small (and if they are equal, it will be zero).

Cause: 

       The value of delta_lambda over the window just calculated
       was being incorrectly written to the restart file. Instead
       of using the true value, the value of delta_lambda over the
       subsequent window was being written.

Fix:  Make the following changes to slwadj.f

---To routine WRTDYS in file slwadj.f -----------------------------------------

*** OLD slwadj.f
--- NEW slwadj.f
***************
*** 785,790 ****
--- 785,803 ----
        DATA FLAG/'SLOWDYN'/
        VERSON = 4.001D0
  C
+ C Bugfix.61:
+ C Following fixes bug in value of "DLOLD" written. The value written
+ C should reflect d_lambda in the direction opposite to that of the
+ C simulation. Previously, it was generally the same as lambda in the
+ C forward direction -- dap (3/94).
+ C
+       IF (IFTIME.GT.0) THEN
+          DLOLD2 = ALM(1)-ALM(5)
+       ELSE
+          DLOLD2 = ALM(3)-ALM(1)
+       END IF
+ 
+ C
  C -----------------
  C UNFORMATTED WRITES
  C -----------------
***************
*** 802,808 ****
  C Write variables for the SLWLOC block:
  C
            WRITE(NF) IONCE1,INRG,IDEL,IAVSL1,IAVDL1,IPTSSL,
!      *                     IPTSDL,IENUF,DLOLD,ALMDEL,IONCE5
            WRITE(NF) DELDS,DAFOLD,DAROLD,ETOT,CTIME
            WRITE(NF) ((DELA(I,J),I=1,2),J=1,IPTSSL)
            WRITE(NF) ((DELDA(I,J),I=1,2),J=1,IPTSDL)
--- 815,821 ----
  C Write variables for the SLWLOC block:
  C
            WRITE(NF) IONCE1,INRG,IDEL,IAVSL1,IAVDL1,IPTSSL,
!      *                     IPTSDL,IENUF,DLOLD2,ALMDEL,IONCE5
            WRITE(NF) DELDS,DAFOLD,DAROLD,ETOT,CTIME
            WRITE(NF) ((DELA(I,J),I=1,2),J=1,IPTSSL)
            WRITE(NF) ((DELDA(I,J),I=1,2),J=1,IPTSDL)
***************
*** 834,840 ****
  C Read variables for the SLWLOC block:
  C
            WRITE(NF,9100) IONCE1,INRG,IDEL,IAVSL1,IAVDL1,IPTSSL,
!      *                     IPTSDL,IENUF,DLOLD,ALMDEL,IONCE5
            WRITE(NF,9110) DELDS,DAFOLD,DAROLD,ETOT,CTIME
            WRITE(NF,9110) ((DELA(I,J),I=1,2),J=1,IPTSSL)
            WRITE(NF,9110) ((DELDA(I,J),I=1,2),J=1,IPTSDL)
--- 847,853 ----
  C Read variables for the SLWLOC block:
  C
            WRITE(NF,9100) IONCE1,INRG,IDEL,IAVSL1,IAVDL1,IPTSSL,
!      *                     IPTSDL,IENUF,DLOLD2,ALMDEL,IONCE5
            WRITE(NF,9110) DELDS,DAFOLD,DAROLD,ETOT,CTIME
            WRITE(NF,9110) ((DELA(I,J),I=1,2),J=1,IPTSSL)
            WRITE(NF,9110) ((DELDA(I,J),I=1,2),J=1,IPTSDL)

------------------------------------------------------------------------------

Temporary workarounds: 

     The results reported in the first section of the simulation 
     (pre-restart) will be entirely correct. If you wish to use these
     results as input to a restart with corrected code, you must edit
     the restart file created by the first simulation. Search for the
     line starting with the keyword "SLOWDYN" in this file. The NEXT
     line will look something like:

    1    2    0    2    0    8    0    1  0.3267250E-01  0.3267250E-01 9899
                                          ^^^^^^^^^^^^^

     where the two real values will be the same. Change the underlined value
     to the actual value of delta_lambda(old) over the last COMPLETED window
     in the first simulation. This can be ascertained by examination of
     the output file from the simulation. Find the LAST point at which
     the free energy was updated (lambda changed). Then, if the simulation
     was run 0->1, delta_lambda(old) = Lam+d_lam - Lambda. If the simulation
     was run 1->0, delta_lambda(old) = Lambda - Lam-d_lam. (The values
     of Lambda, Lam+d_lam & Lam-d_lam are clearly given in the output
     file). After making this change, the first restart leg of the
     simulation can be run. 

     Remember that in any case, this bug does not affect free energy
     results reported for the direction of the simulation--these will
     be correct even in old code. (See "Affects" above).


Routines affected:

     GIBBS   Routine WRTDYS in file ...amber4/src/gibbs/slwadj.f

----