ChangeLog.c24
| | |
|
CHARMM 24 |
| c24a1 |
Developmental |
February 15, 1994 |
| c24x1 |
Evaluation |
February 15, 1994 |
| c24a2 |
Developmental |
August 15, 1994 |
| c24a3 |
Developmental |
March 15, 1995 |
| c24b1 |
Release |
August 15, 1995 |
| c24b2 |
Release |
February 15, 1996 |
| c24g1 |
Release |
August 15, 1996 |
| c24g2 |
Release |
February 15, 1997 |
|
| | |
|
About CHARMM24 Development |
|
|
|
As an on-going project, CHARMM development will be carried out with
CHARMM version 24 series. We set up the base revision c24a1 out of
c23f3. The following modifications and new features are planed as the
firsst phase development.
The base revision from CHARMM c23f3 (February 1, 1994) to
c24a1 (February 15, 1994)
1.1 File Organization and Style
- We have made a few stylistic changes on the base revision.
(1) version specific identifiles are replaced with version independent
identifier, e.g., "c23" to "chm".
Note that ##INCLUDE statements are changed and the fcm directory
is denoted as ~/charmm_fcm/ in the source code.
(2) The directory name "main" is changed to "charmm" and the main
driver source file name "charmm.src" is changed to "charmm_main.src"
(3) flexfort is removed and new version of prefx.src is introcued as
of c23f3.
(4) ##FORTRAN preprocessor key and all remaining DEC/CMS header lines
are removed. The first line of source and documentation files is
modified to read as "CHARMM Element {module}/{filename} 1.1".
This line should not be disturbed.
(5) install.com, setmk.com, and test.com are modified to meet the
changes in file structure. setmk.com can now handle single module
makefile, usage is 'setmk.com {machine_type} {module_name}'.
The modification of test.com (former test23.com) enables testing
individual testsuite, e.g, c20test, etc.
(6) charmm/charmm_main.src and fcm/version.fcm are updated to reflect
the current version identification.
1.2 Bugfixes
(1) Infinite loop in PRINHP
Bug940209 by Ryszard Czerminski (ryszard@msi.com)
In situation when CHCKHP calls WRNDIE infinit loop is created.
WRNDIE calls DIE, DIE calls PRINHP which call CHCKHP... So, CHCKHP
call is removed in PRINHP.
File modified:
util/util.src
(2) Parallel Restart and SHAKE with LEAP
Bug940210 by Bernard R. Brooks (brbrooks@helix.nih.com)
The parallel code was accidentally deleted from dynio.src and
needs to be reinstated.
The DYNA LEAP when running an MD step with SHAKE and with
equilibration/heating (or with saving a restart file) determined by
NCYCLE would take an incorrect (extra) step.
Files modified:
dynamc/dynio.src
dynamc/dynamc.src
CHARMM c24a1 (February 15, 1994) to c24a2 (August 15, 1994)
2.1 New Dihedral / Improper Dihedral Energy Routines
- Arnaud Blondel
Email Address : blondel@tammy.harvard.edu
Institution : Harvard University
Date : February 1994
The previous energy routines used the derivatives d(cos(phi))/dr
to calculate the forces and the second derivatives. This choice
introduced an artificial singularity at sin(phi)=0.
The new routines use the derivative d(phi)/dr and thus have no
singularities. This removes the tests to avoid numerical overflow or
the switch functions in the vector improper routines.
The new dihedral routines now support cases where planar conformation
is not an extremum. Thus a value other than 0 or 180 can be specified
in the dihedral parameters. The dihedral constraints can also use the
dihedral functional form using the key word PERIod and giving a
non-zero number.
New data structures have been introduced. The sine and cosine of
the dihedral reference angle Phi0 (E=K( 1+cos( (n*Phi) - Phi0 ) ) or
E=K(Phi-Phi0)**2) are held in new fcm(s).
..............................................................................
File Modified: 35
File Added : 2
Energy routines: 7 - New formulae introduced in the respective
dihedral/improper routines.
cenerf.src Changes have been made to preserve the
eintern.src previous routine structure as much as
enefscal.src possible.
enefst2.src
enefst2q.src
enefvect.src
enefvp.src
Data structures: 5 - Trigonometric table for the reference
angles. (functional form)
cnst.fcm : Dihedral constrains (dihe/impr)
intcr.fcm : Internal coordinate constrains (impr)
param.fcm : Dihedral and improper (resp)
pert.fcm : Copy of the dihedral constrains
vphi.fcm : Parvect energy routine tables (dihe)
Generation and - Generate and update the tables according
Manipulation : 11 to the changes.
genpsf.src : Internal coordinates
makphi.src : Parvect energy routine tables
modpsf.src : IC
replica.src : IC
images.src : IC
parmio.src : Dihedral and improper
cstran.src : Dihedral constrains + PERIod
intcor.src : IC
intcor2.src : IC
pert.src : Pert dihedral constrains
vibsub.src : IC
Calls to Energy - Introduce the trigonometric table in the
routines: 9 call parameters.
anacor.src
ecnstr.src
energy.src
intere.src
eimage.src
hbuild.src
epert.src
icpert.src
tsme.src
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Documentation : 2
cons.doc - The PERIod Key word is added to the
dihedral constrain command
io.doc - No more restriction on the reference
dihedral angle. Unlimited periodicity
for slow and scalar routines.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Testcases : 1 modified, 2 added
constest.inp - A test section had been added for the
keyword PERIod (CDIH).
- Old version complains (XTRAN).
dihetest1.inp - Simple test on a pair of nucleotide.
Tests with faster off/on and all
combination BLOCK/NOBLOCK, FIX/NOFIX,
and in block FORCE/NOFORCE.
- Old version works well (with bug fix in block).
dihetest2.inp - Special testcase for the non-planar
dihedral terms. Period 1 to 6 tested
with non planar reference angle.
FASTer OFF/ON.
TEST SECOnd command is tested here.
- Old version crashes.
------------------------------------------------------------------------------
2.2 TEST command additions
- Arnaud Blondel
Email Address : blondel@tammy.harvard.edu
Institution : Harvard University
Date : February 1994
(1) Test routine for the second derivatives.
A simple second derivatives tester. It is based on a finite
difference of the first derivatives, so first derivatives should be
tested first. Parser inputs and outputs are similar to the ones of
'Test First'. TEST SECOnd is tested in dihetest2.inp.
(2) Test of the trigonometric tables.
A test param command has been introduced for development needs.
So far it includes the following possibilities under the TRIGonometry
key word.
- DIHEdral : prints the internal list of dihedral parameters with:
Number, ForceCte, Period, Reference angle, cosine and
sine.
- IMPRoper : same as previous for improper.
- CDIHedr : same as previous for Constrain dihedral.
- VDIHedr : prints the list of parameters for parvect dihedral
routines. Number, Index in the dihedral list, ForceCte,
period, angle, cosine and sine are given.
File modified:
/misc/testch.src
------------------------------------------------------------------------------
2.3 NIH Developments
- Bernard R. Brooks
Email Address : brbrooks@helix.nih.gov
Institution : NIH
Date : June 28, 1994
(1) ANAL command extension
This command computes energy partitioned based on atoms. For free
energy perturbation simulation, it provides the the free energy
partition by atoms. It also allow the printing of selected terms in
the PSF when computing the energy which lists current geometric and
energetic information on a term-by-term basis.
File added:
/energy/anal.src
Files modified:
/charmm/charmm_main.src
/energy/eintern.src
/energy/eutil.src
/nbonds/enbond.src
/fcm/econt.fcm
..............................................................................
(2) GAMESS code added
The CHARMM-GAMMES interface is now complete. There were a few
routine name conflicts. The names were changed on the CHARMM side
where the routines differed.
Files modified:
/charmm/charmm_main.src
/charmm/iniall.src
/gamint/gamini.src
/gener/update.src
/minmiz/tnpack.src
/molvib/gmat.src
/molvib/molvut.src
/rxncor/path.src
/vibran/thermo.src
/vibran/vibran.src
/fcm/energy.fcm
/fcm/gamess.fcm
/io/mainio.src
/nbonds/nbondg.src
/nbonds/nbonds.src
/quantum/addlnat.src
..............................................................................
(3) Replica Path method created
Files added;
/energy/epath.src
/fcm/path.fcm
Files modified:
/charmm/charmm_main.src
/energy/energy.src
/energy/eutil.src
/energy/printe.src
/fcm/energy.fcm
..............................................................................
(4) Langevin Piston Code added
The constant pressure code has been overhauled. The old method
based on Berendsen's method has been replaced with a Langevin Piston
Method. When no friction is applied, this method becomes the standard
method based on Nose and Klein (adapted from Andersen). At the limit
of infinite friction with no random force, this reverts to the
Berendsen method.
Files modified:
/dynamc/dcntrl.src
/dynamc/dynamc.src
/dynamc/dynamcv.src
/dynamc/dynamvv.src
/dynamc/dynio.src
/dynamc/prssre.src ** major changes
/energy/energy.src
/energy/eutil.src
/energy/printe.src
/fcm/energy.fcm
/fcm/image.fcm
/fcm/reawri.fcm
/image/crystal.src
/image/eimage.src
/image/xtlfrq.src
/nbonds/egrad1.src
/misc/testch.src
/pert/epert.src
..............................................................................
(5) Constant pressure stochatic boundary code added.
The variable DROFF used in the MMFP has been coupled to the
pressure in the new integrator for constant pressure simulations using
a langevin piston. This allows for a langevin stochatic boundary
force to be used.
Files modified:
/dynamc/prssre.src
/energy/printe.src
/fcm/energy.fcm
/pert/epert.src
..............................................................................
(6) Anisotropic harmonic restraints added.
The global scale factors: "XSCAle", "YSCAle", and "ZSCAle" have
been added to the "CONS HARM" command. This allows using the CONS
HARM to enforce a planar or linear restraint. This feature is also
useful for use in conjunction with our COORPLAS program (for
generating 3-D coordinates from plastic models).
Files modified:
/energy/ecnstr.src
/energy/energy.src
/energy/intere.src
/fcm/cnst.fcm
/manip/cstran.src
/pert/epert.src
/pert/icpert.src
..............................................................................
(7) Flexible SHAKE code added
Files modified:
/fcm/shake.fcm
/minmiz/abner.src
..............................................................................
(8) New "ADD" option for the "COOR SEARCH" command
The "ADD" option for the COOR SEARCH command has been added to
allow the calculation of partial occupancy factors. This allow holes
in proteins to be analyzed for flexibility and variability.
Files modified:
/manip/corman.src
/manip/corman2.src
..............................................................................
(9) Enhancements to PERT code.
The energy terms calls for ASP energies and the MMFP code have
been added to EPERT by copying the calls in ENERGY.
Files modified:
/pert/epert.src
..............................................................................
(10) Format for command substitution enhanced for accuarcy
Files modified:
/util/string.src
..............................................................................
(11) Normal modes by finite difference for small system added
Files modified:
/vibran/vibcom.src
/vibran/vibran.src
/vibran/vibsub.src
..............................................................................
(12) Long line processing modified.
The commands "SHORt" and "LONG" have been added which gives users
the ability to control then length of output lines in CHARMM. Several
output sections support dual length formats. The selection of the
desired format may now be specified at run time.
Files modified:
/charmm/iniall.src
/charmm/miscom.src
/fcm/stream.fcm
/pert/pert.src
..............................................................................
(13) CORRECTED BUGS AND PROBLEMS FIXED
(13.1) Default IHBFRQ set to zero
This avoids some script problems and makes the output less confusing.
File modified:
/charmm/iniall.src
(13.2) The ENVIronment command is only valid for UNIX systems
A PREFLX if-block is added so that this command is only parsed on
UNIX systems.
File modified:
/charmm/miscom.src
(13.3) Missing code
There is missing code copying COMP to MAIN coordinates which
causes the integrator to have problems when used with SHAKE.
DO I=ATFRST,ATLAST block needs to be added after the main loop.
File modified:
/dynamc/dynamc.src
(13.4) Corrections to dynamics I/O routines
The dynamics I/O routines have been modified so that;
a) When reading a trajectory frame (READCV) which contains crystal
information, the unit cell parameters are read. Image atom
coordinates are generated if invoked through the "TRAJectory READ"
command.
b) The GTICNT routine can be used in parallel runs.
c) The dynamics restart file has been modified so that long time
pressure statistics are saved and reread. The version number
(IVERS.GT.23) specifies the new format with additional data.
d) The dynamics restart has been modified so that it runs in parallel.
Several escapes (RETURN) in the routine have been replaced with
goto's (GOTO 900).
File modified:
/dynamc/dynio.src
(13.5) Parallel code for the virial and pressure is added
The virial summation across processors done correctly by adding
GCOMB calls.
File modified:
/dynamc/prssre.src
(13.6) Fixes in graphics declarations.
Some of the declaration in the Apollo graphics were incorrectly
modified. These variables have been reverted to INTEGER*2.
File modified:
/graphics/apodraw.src
/graphics/apograph.src
/fcm/graphdep.fcm
(13.7) Z table code isolated
The Z table code has been isolated with "##IF ZTBL" precompiler
blocks. This facility was consuming memory resources that were
needed. The Z table code is only used in NBNDGC. The extra print was
cryptic and annoying.
Files modified:
/image/upimag.src
/nbonds/nbexcl.src
/nbonds/nbndgc.src
/nbonds/nbonds.src
/util/ztbl.src
/fcm/psf.fcm
(13.8) WRNLEV to PRNLEV fix
Instances of WRNLEV has been replaced with PRNLEV where appropriate.
File modified:
/gener/modpsf.src
(13.9) Crystal coordinate reading of trajectory files
The routine CREAD in file io/coorio.src has been updated to allow
the reading of trajectory files with crystal data. Also, coordinates
with bad fields will be ignored (instead of causing a fatal error).
File modified:
/io/coorio.src
(13.10) Parsing command line options fixed
The parsing of command lines was broken when multi-character
variables were introduced. These modifications to machdep/startup.src
fix the problem. Code has also been added to allow command line
variables to be used when running a cluster of workstations.
File modified:
/machdep/startup.src
(13.11) COOR ORIEnt command for parallel version
There was a bug in the routine ORINTC which caused CHARMM to fail
on parallel systems when the COOR ORIEnt command was executed.
File modified:
/manip/corman2.src
(13.12) Printing dihedral restraints
Atom identifier was wrong in printing dihedral restraints. Calls
to ATOMID are fixed. Only the printout was affected.
File modified:
/manip/cstran.src
(13.13) Bug in COOR ORIE RMS command : Bad trace in FROTU
There was a bug in COOR ORIE RMS command in computing the trace
of the coordinate Lagrangian matrix in FROTU. TRACE=W(1)+W(4)+W(6) is
correct.
File modified:
/manip/rotlsq.src
(13.14) Declaration in nbndgc modified for SGI machines.
It appears that SGI's new compiler has a nonstandard definition of
what EXTERNAL means. EXTERNAL XDIST is encapsulated by ##IFN IRIS and
##ENDIF.
File modified:
/nbonds/nbndgc.src
(13.15) Bug in parsing the IF command
The if test that was used to set the FFOUR variable was put in the
wrong place.
File modified:
/util/parse.src:
(13.16) Correcting problems in the old QM/MM code
Fix initial variable declarations. Overhaul of FINDEL.
Files modified:
/quantum/qmpac.src
/quantum/qmset.src
(13.17) Redundant FAST commands dropped
The commands "FAST CONV", "FAST IRIS", "FAST SGI", and "FAST SUN"
have been dropped. Use "FAST ON", "FAST OFF", or "FAST DEFAULT" to
control the efficiency level.
File modified:
/charmm/miscom.src
(13.18) Bug in temperature time series fixed
A pointer was not properly updates. This command was also updated
to allow the specification of the number of degrees of freedom for
when a subset of atoms is selected (keyword "NDEGf").
Files modified:
/correl/anacor.src
/correl/correl.src
(13.19) ##IF NOCORREL pre-compiler keyword
added to cluster command.
Files modified:
/correl/clustr.src
(13.20) Control code in DCNTRL modified
The code to decide which integrator to use by default has been
modified. Warning message for some ill-advised combinations have been
added. Some warning messages have been modified to make the meaning
more clear.
Files modified:
/dynamc/dcntrl.src
/dynamc/dynutil.src
(13.21) Bug in printing the internal pressure on a restart with SHAKE
The internal pressure was incorrectly computed on a dynamic
restart when SHAKE was in use. The "force" required to maintain the
bond geometries has been added to the calculation of the virial.
File modified:
/dynamc/dynamc.src
(13.22) Coordinate I/O routines updated for reading crystal parameters
READCV has been modified to work correctly with the crystal and
parallel code. CREAD modified to read crystal parameters.
Files modified:
/dynamc/dynio.src
/io/coorio.src
(13.23) MERGE command enhancement
The ability to both select and fix atoms in the MERGE command is
added.
File modified:
/dynamc/dynsub.src
(13.24) Bugs in the fortran common file fixed
SAVE statements are corrected.
File modified:
/fcm/cmdpar.fcm
(13.25) Error handling for the HPUX machines
Traceback feature is added in NIH versions
File modified:
/machdep/machutil.src
------------------------------------------------------------------------------
2.4 Addition of solvent boundary potential for water in MMFP
- Benoit Roux and Dmitrii Beglov
Email Address : rouxb@ERE.UMontreal.CA
beglovd@ERE.UMontreal.CA
Institution : Department of Chemistry, University of Montreal
Date : February 1, 1994
A solvent boundary potential for the simulation of water at
constant pressure will be added to the MMFP module. The boundary
potential is an approximation but follows from a rigorous statistical
mechanical treatment of the boundary. In light of the difficulties
raised by the previous treatments, a different route was chosen to
formulate and develop the solvent boundary potential for computer
simulations of a finite representation of an infinite bulk system.
The present theoretical formulation is based on a separation of the
multidimensional solute-solvent configurational integral in terms of n
``inner'' solvent molecules nearest to an arbitrary solute, and the
remaining ``outer'' bulk solvent molecules. This formulation, which
differs significantly from previous treatments, provides further
insight into the statistical mechanical basis of the solvent boundary
potential and is helpful in constructing useful approximations for
computer simulations in dense liquids. An approximation to the
solvent boundary potential is constructed for simulations of bulk
water at constant pressure, including the influence of van der Waals
(done with RISM) and electrostatic interactions (done with a
Kirkwood-like multipole expansion). The approach has been tested with
success on several typical systems (water, ions, n-butane and alanine
dipeptide). A paper describing the method has been submited.
Files Added:
/misc/ssbp.src
/fcm/ssbp.fcm
Files Modified:
/energy/energy.src
/energy/eutil.src
/energy/printe.src
/misc/mmfp.src
/fcm/energy.fcm
New/Modified Testcase:
/test/c23test/mmfptest2.inp
------------------------------------------------------------------------------
2.5 Bugfixes
Bugfixes reported to charmm-bugs@tammy during the development cycle
are incorporated. The bug-reports are collected in ~ftp/charmm/Bugs.
(1) Stack allocation errors in ANACOR
- Mary E. Karpen (B940425.mek and B940526.mek)
Several errors in stack allocation in the call to TRNPHI from
ANACOR are fixed. Stack allocation for TRNPHI arrays are moved from
correl to ANACOR. Since only DIHE time series require these arrays,
the stack space in ANACOR is only allocated for these time series.
Also, rather than have dihedral angle transitions calculated as the
default, it makes more sense to calculate them by request. This is
done by adding the option TRANsition on the ENTE DIHE command line.
(doc/correl.doc is updated)
Files Modified:
/correl/correl.src
/anacor.src
(2) SOLANA bugs
- Ryszard Czerminski (B940523.rcz)
Calling argument list is corrected for the RDNEXCO call, and a
warning message is coorected.
- Jay Banks (B940317.jb)
Crystal test corrected from QCRYS=(IC11.EQ.1) to QCRYS=(IC11.NE.0)
File Modified:
/correl/solana.src
(3) Parallel code addition/correction
- Bernard R. Brooks (B940210.brb and B940601.brb)
Parallel code added.
- Leo Caves and Arnaud Blondel
A bug in the addition is fixed by ##INCLUDE '~/charmm_23/stream.fcm'
File Modified:
/dynamc/dynio.src
(4) FREHP calls added
- Mary E. Karpen (B940712.mek #3)
Heap allocated in SHKCOM is not released on program exit. This
does not affect charmm job output in any way, except in the "SPACE
CURRENTLY IN USE" report at the end of the job. The space is freed,
so SPACE USED is zero at job end, allowing other potential problems
with heap allocation to be detected. The IF (QSHAKE) condition to
freeing the cray shake heap (RSCSHK) is added for consistency, though
this makes no difference in the current code logic.
Files Modified:
/main/iniall.src
/manip/shake.src
(5) PHASE value fix
- Arnaud Blondel (B940601.ab)
Wrong values (sign flip) can be returned if PseudoPhase is close
to 90 or -90 due to infinity of tan(PsPhase). For unknown reason the
same phenomenon occurs around 180.
File Modified:
/manip/pucker.src
(6) ING14P fix
- Jay Banks (B940603.jb)
In the IF (QPERT) block, the index into ING14P array should be
NXIP not NXI. Two lines are corrected; ING14P(NXI) to ING14P(NXIP).
File Modified:
/nbonds/nbondg.src
(7) BLOCK code cleaned
- Arnaud Blondel (B940302.ab and B940303.ab)
Coefficient matrix print out is corrected by checking QBLOCK upon
the BLOCK END command. BLOCK usage flags are removed upon the BLOCK
CLEAR command.
File Modified:
/pert/block.src
(8) Initialize ALLC and ALLZ arrays
- Ryszard Czerminski (B940514.rcz)
Floating point exception error is avoided by initializing ALLC and
ALLZ arrays.
File Modified:
/quantum/qmpac.src
(9) REAL/REAL*8 usage fix
- Ryszard Czerminski (B940412.rcz)
An inconsistent use of memory (real vs. real*8) is corrected.
real*8 wt(n) to real wt(n).
FIle Modified:
/rxbcor/rxnene.src
(10) Division by zero
- Bernard R. Brooks and Georgios Archontis
Division by zero problem is fixed in ADIE=(1.0+CDIE2*(3.0*YDIE-1.0))/
(3*YDIE*(CDIE2-1.0)).
File Modified:
/solvation/state.src
(11) Infinite loop problem
- Ryszard Czerminski (B940209.rcz)
An infinite loop is created when CHCKHP calls WRNDIE. Call to
CHCKHP in PRINHP should be commented out.
File Modified:
/util/util.src
(12) Vectorization Error
- Mary E. Karpen (B940425.mek #5)
AMASST in subroutine VIBFLU is incorrectly calculated on the cray
c90. In the testcase vibran.inp, the command
FLUCT ATOM MODE 1 THRU 57 TEMP 300.0 SELE TYPE CA END
results in the variable AMASST in the vibran/vibsub.src routine VIBFLU
to be assigned 19 (= NATOM) rather than the correct amount (1). By
removing vectorization from the loop, AMASST is calculated correctly.
There is no obvious reason why vectorization would break this loop,
so there is potentially another bug lurking, which was not found.
File Modified:
/vibran/vibsub.src
(13) Declaration fix in REPLICA
- Mary E. Karpen (B940425.mek #1a)
Incorrect private/shared declarations for replica code are corrected.
File Modified:
/nbonds/cnbnd.src
(14) Preprocessor keys for MTS code
- Ryszard Czerminski (B940223.rcz)
- Mary E. Karpen (B940425.mek #1b)
Problems with preprocessor directives for MTS is fixed.
Files Modified:
/dynamc/mts.src:
/image/eimage.src:
(15) Vector dependency
- Mary E. Karpen (B940425.mek #2)
Vector dependencies exist in TRANSI routine but CDIR$ IVDEP
specified on line 126. Comment out.
File Modified:
/image/images.src:
(16) Non-standard quote
- Michael E. Holder (B940531.mh)
Non-standard use of " fixed to '.
File Modified:
/util/cmdpar.src
(17) IMAGE/CRYSTAL keyword parsing bug
- Ryszard Czerminski
Nonbonded list is not set up properly upon NBOND command. LIMAGE
should be set .TRUE. in the argument list of the UPDATE call.
File modified:
/charmm/cahrmm_main.src
(18) Version number updated
- CHARMM Manager
Never used VERNMR is removed and VERNMC is updated.
File Modified:
/fcm/version.fcm
CHARMM c24a2 (August 15, 1994) to c24a3 (March 15, 1995)
3.1 Convex Exemplar SPP-100 and generic PVM Ports
Developer : Charles L. Brooks, III and Stephen H. Fleischman
Email Address : brooks@scripps.edu
Institution : The Scripps Research Institute
Date : January 5, 1995
CHARMM Version: c24a2
Proj. Report : D950105.clb
A port of CHARMM version 24a2 to general PVM based parallelism using
existing parallel code as well as a port to the Convex parallel machine are
included. The original port is the work of Stephen H. Fleischman from
Convex Computer Corp. Charles L. Brooks, III simply implemented this port
in other versions of CHARMM.
Files Added:
build/UNX/Makefile_cspp
Files Modified:
install.com
doc/parallel.doc
source/charmm/iniall.src
fcm/parllel.fcm
machdep/machutil.src
parallel.src
parcomm.src
startup.src
util/string.src
------------------------------------------------------------------------------
3.2 Cray T3D Port
Name : Charles L. Brooks, III and Barry C. Bolding
Email Address : brooks@scripps.edu
Institution : The Scripps Research Institute
Date : January 6, 1995
CHARMM Version: c24a2
Proj. Report : D950107.clb
A port of CHARMM version 24a2 to the Cray T3D parallel computer using
existing parallel code. The original port is the work of Barry C. Bolding
from Cray Research Inc. in collaboration with Charles L. Brooks, III, who
implemented this port in other versions of CHARMM.
Files Added:
build/UNX/Make_t3d
Files Modified:
install.com
doc/parallel.doc
source/charmm/charmm_main.src
machdep/cstuff.c
machio.src
machutil.src
parallel.src
space.src
startup.src
fcm/dimens.fcm
heap.fcm
number.fcm
stack.fcm
------------------------------------------------------------------------------
3.3 Thinking Machine's CM5 Port
Name : Robert Nagle
Email Address : nagle@tammy.harvard.edu
Institution : Harvard
Date : August 24, 1994
CHARMM Version: c24a2
OS : CMOST 7.3_final.1-rev.3
Proj. Report : D940825.rn
1) Many I/O operations on the CM5 resulted in a system wide hang.
Since I/O was set up to be synchronous broadcast, all processors
should have been making the same I/O requests. However, in some
instances, branches were made based on either PRNLEV (more frequently)
or IOLEV. This resulted in node 0 doing an open, say, and all other
nodes continuing on their merry way. Result - deadlock.
2) Previous communication scheme was based on a simple send and
receive model. By using TMC's active message layer, communication
bandwith can be increased by anywhere from 50% to 5X.
3) DIE on one node could leave the others merrily trundling along.
4) Other minor improvements
Files Added:
build/UNX/Makefile_cm5
source/fcm/cm5.fcm
Files Modified:
install.com
doc/parallel.doc
source/machdep/machutil.src
parallel.src
startup.src
string.src
------------------------------------------------------------------------------
3.4 OS/2 Port
Name : Stefan Boresch
Email Address : boresch@tammy.harvard.edu
Institution : Harvard University (Karplus group)
Date : November 1994
CHARMM Version: c23f4
OS : S/2 2.x and higher
CHARMM (c23f4) has been ported to the OS/2 operating system,
version 2.x and higher. The Watcom Fortran compiler (v. 9.5,
patch-level (c)) has been used. The port is fairly complete, a
detailed list of limitations and known problems is given in the 'Other
Comments' section. The modifications should also be applicable to
c24a2 by applying the changes described below to the respective files.
(I can't hold two versions of CHARMM on my hard-drive at the same time;
I'm happy to work / check the modifications of c24a2/c24a3 once they
become available.)
A new pre-processor keyword, OS2, has been introduced, and all
OS/2 related changes hide behind the OS2 keyword. The port has revealed
several bugs of the compiler, one of them is annoying, and I hope it
will be fixed soon. On the other hand, as OS/2 / Watcom FORTRAN is not
yet another Unix platform, a few quirks of the code have become apparent
as well; one might consider changing the basic code in these instances...
While most Unix utilities have been ported to OS/2, I do not
want to rely on the user having these installed. Therefore, I shall
write an install script in REXX (the OS/2 built in scripting
language); I will also provide utilities to automatically generate
makefiles. (to be done a.s.a.p!) Unfortunately, WATCOM's make
utility doesn't support a number of features used in the Unix
makefiles for CHARMM, so the makefiles differ significantly. Most
likely one can write them much more elegant, but this needs to be done
as well...
Watcom FORTRAN allows direct access to any OS/2 system call.
Therefore, there is no need for C routines; all .c files are superfluous
for OS/2!
Currently, there is no free FORTRAN compiler for OS/2 (asides from
f2c, and I really didn't want to use that...) I'll watch out for the
GNU FORTRAN compiler which apparently is in alpha testing. The WATCOM
compiler has proven quite reliable, and to the best of my knowledge is
the cheapest option for academic institutions. The same package
also supports 32bit extended DOS, Windows 3.1 and Windows NT; i.e. if
someone is so inclined, porting to these platforms should be more
or less painless. However, I used the HPFS filesystem, so I didn't
worry about shortening filenames to DOS' 8.3 convention.
Files Modified (module/file.src):
source/charmm/iniall.src
/miscom.src
dynamc/dynio.src
fcm/impnon.fcm
/number.fcm
machdep/machio.src
/machutil.src
/space.src
/startup.src
manip/corman.src
rxncor/path.src
solvation/fft.src
Other Comments:
The user needs the Watcom Fortran compiler v. 9.5. Patchlevel
(c) was used for the port. The installation will work only on a HPFS
partition. It should be considered to include an OS/2
executable in the distribution package. (The executable itself should
run on both HPFS and FAT partitions)
As described above, patchlevel (c) contains an annoying bug
that requires a somewhat clumsy workaround and Charmm to be called through
a script file. I will provide patches as soon as the compiler is fixed.
A set of makefiles that will work for c23f4 is provided. I'm
working on a general install utility written in REXX (the built in
scripting language of OS/2). I shall also provide REXX scripts to generate
makefiles solely based on the list of source files.
The executable has been currently tested only when compiled
without optimization. I'll test which optimization levels are
possible a.s.a.p. I'm confident, however, that no major problems will
arise (I never had any optimization related problems with the c22gx
series of CHARMM).
KNOWN LIMITATIONS/PROBLEMS:
* Compiler bug related to REWIND of STDIN (clumsy workaround exists)
* QUANTUM code not supported: I didn't even try as I don't think that
a PC will be used for QM calculations anyways.
* Aside from the quantum testcase, two other test-cases fail:
rism.inp
xtlala2.inp
* OS/2 doesn't have an OS funtion that reports the CPU time consumed
by a program/process. Hence dummy routines are introduced, a CPU time
of 0.00 is reported.
------------------------------------------------------------------------------
3.5 Fast Multipole Code for Electrostatic interactions
Name : Robert Nagle
Email Address : nagle@tammy.harvard.edu
Institution : Harvard University
Date : December 1994
CHARMM Version: c23f4
This is an initial implementation of a fast multipole method, based on
John Board's work. A more complete description can be found in fma.doc.
A new non-bond option (FMA) has been added. This replaces cut-off
parameters with a no cut-off hierarchical technique. The advantages
of this method are that you can control the error and that it is
amenable to parallelization. FMA is an order(N) technique but the
constant is large and so FMA will, in general, be slower for systems
of less that 5000 atoms, for the same accuracy.
Two options, LEVEL and TERMS, govern how many hierachical levels are
used and how many terms are retained in the expansion, respectively.
In the method, each box at every level is subdivided into 8 sub-boxes
- you should select LEVEL so that the boxes at the lowest (i.e.
finest) level contain 10-20 atoms on average: 3 or 4 will be typical
choices. You then select TERMS to control the accuracy that you
require: 4 will often suffice but I would generally recommend 6 or
even 8. [See the references in fma.doc for a detailed description of
the error bounds].
NOFMA is the nonbond option which turns off the multipole method.
Compilation of FMA is controlled by the flag, FMA, in pref.dat.
[Note] FAST ON is required for this initial implementation.
This implementation is not yet parallelized.
Files Added:
doc/fma.doc (to be added in c24a4)
source/nbonds/fma.src
fcm/fma.fcm
/fmacons.fcm
Files Modified:
doc/nbonds.doc (to be modified in c24a4)
source/charmm/iniall.src
nbonds/enbfast.src
/enbond.src
/nbutil.src
/inbnd.fcm
New Testcase:
test/c24test/fma.inp (to be added in c24a4)
------------------------------------------------------------------------------
3.6 New Options to CORREL and CORMAN
Name : Lennart Nilsson
Email Address : Lennart.Nilsson@csb.ki.se
Institution : Karolinska institutet
Date : January 4, 1995
CHARMM Version: c23f4
3.6.1 Two new mantime options have been added to CORREL
MANTIME time-series-name
{ DOTProdcut name } ! Q(T) x-comp=Q(T).Q2(T)
Q2(T)x-comp=angle Q(T) vs Q2(T) degrees
{ CROSproduct name } ! Q(T) = Q(T)xQ2(T)
CROSsprod name Q(T) = Q(T) x Q2(T); the 3D crossproduct of the two
3D vectors formed by the selected and named timeseries
DOTProd name Q(T) = x-comp of Q(T)= Q(T) . Q2(T)
x-comp of Q2(T) angle in degrees between the two vectors
NOTE! Modifies Q2 as well as Q
to get just the x-comp you may then edit the selected series:
EDIT series VECCOD 1
Files Modified:
doc/correl.doc
source/correl/correl.src
/mantim.src
3.6.2 The HBONd command
An option for the analysis of H-bond patterns from trajectories has been
added to corman.
COORdinates HBONd 2X(atom-selection) [CUT ] [CUTA ] [IUNIt ]
[BRIDge ]
[FIRSt int] [NUNIts int] [NSKIp int] [BEGIn int] [STOP int]
The HBONd command analyses a trajectory for hydrogen bonding patterns.
For each acceptor/donor in the first selection the average number and average
lifetime of hydrogen bonds to any atom in the second selection is calculated.
A hydrogen bond is assumed to exist when two candidate atoms are closer than
the value specified by CUT (default 2.4A,(reasonable criterion, DeLoof et al.
(1992) JACS 114,4028), and if a value for CUTAngle is given the angle formed
by D-H..A is greater than this CUTAngle (in degrees, 180 is a linear H-bond);
the default is to allow all angles. The current implementation assumes that
hbonding hydrogens are present in the PSF and also uses ACCEptor and DONOr
information from the PSF to determine what pairs are possible. If output is
wanted to a separate file the IUNIt option can be used. If the BRIDge option
is used the routine calculates average number and lifetime of bridges formed
between all pairs of atoms in the two selections; a bridge is counted a
residue of the type specicified with the BRIDge hydrogen bonds
(using same criteria as for direct hbonding) to at least one atom in each
selection. The typical use of this would be to find water bridges.
Here again, results are presented for each atom in the first selection.
NOTE: In order not to find hbonds between bonded atoms UPDATE is called,
which requires coordinates to be present when invoking this module. Since
this is done just to get the non-bond exclusion lists, the cut-offs are
set to very small values, and could influence subsequent energy evaluations
if the non-bond cutoffs are not then respecified.
Files Added:
source/manip/hbanal.src
Files Modified:
doc/corman.doc
source/manip/corman.src
------------------------------------------------------------------------------
3.7 Energy Embedding by the Addition of a Higher Spatial Dimension
Name : Elan Z. Eisenmesser / Carol Post
Email Address : eeisenme@bragg.bio.purdue.edu
Institution : Purdue University
Date : January 4, 1995
CHARMM Version: c23f3
The energy embedding technique entails placing a molecule into a
higher spatial dimension {Crippen,G.M. & Havel,T.F. (1990)
J.Chem.Inf.Comput.Sci. Vol 30, 222-227}. The possibility of surmounting
energy barriers with these added degrees of freedom may lead to lower
energy minima.
With the recent success of using four dimensions in the GROMOS
force field {Van Schaik,R.C., Berendsen,H.J.C., Torda,A.E., & Van
Gunsteren,W.F. (1993) J.Mol.Biol., Vol 234, 751-762}, creating a similar
option in CHARMM should also prove advantageous. Specifically, another
cartesian coordinate was added to the usuall X, Y, and Z coordinates and
was appropriately named FDIM for Fourth DIMension. This implementation
has led to altertations in some existing code along with the addition of
several algorithms.
Files Added:
doc/fourd.doc
source/dynamc/dynamc4.src
energy/efour.src
/inertia.src
fcm/fourd.fcm
Files Modified:
doc/dynamc.doc
source/dynamc/dcntrl.src
/dynio.src
/dynutil.src
energy/eintern.src
/energy.src
/printe.src
io/coorio.src
misc/noe.src
nbonds/enbond.src
/nbondg.src
fcm/noe.fcm
/energy.fcm
Other Comments:
The variable VIOL was included into noe.src and is not restricted to
use in 4D. It is the summation in angstroms of all the violations above the
cutoff parameter CUT (discussed in cons.doc). It's declared in noe.fcm, but
could just as easily be declared in only noe.src if its need remains local to
that source file.
The parameters INC4D and DEC4D are step numbers that define when the
projection back to 3d will occur. The counter that keeps track of this is
called FCOUNT and is initialized every restart or start. This may or may not
be a good idea. The variable DIM4, DIMension 4, is the flag for 4D.
Dynamc4.src is essentially dynamc.src in 4 dimensions. Note that
even though qeuler and shake still exist in dynamc4.src they are not as yet
supported.
------------------------------------------------------------------------------
3.8 DIMB (Diagonalization In a Mixed Basis) Method
Name : David Perahia, Liliane Mouawad, Herman van Vlijmen
Email Address : vlijmen@tammy.harvard.edu
Institution : Harvard University, Chemistry Department
Date : December 16, 1994
CHARMM Version: c24a2
The DIMB (Diagonalization In a Mixed Basis) method (see L.Mouawad
and D.Perahia (1993), Biopolymers, 33, 599) is an iterative method to
calculate the N lowest normal modes of molecules. It is especially
targeted to do large molecules, since it does not require the full
Hessian to be stored in memory or on disk. In short, the method
does repetitive reduced-basis diagonalizations in bases that consist
partially of the approximate eigenvectors, and partially of Cartesian
coordinates. Eigenvectors are saved to file during the process. Before
that is done, a new basis is again created, which consists of the
approximate eigenvectors at that point + the residual vectors (Lanczos
vectors). This accelerates the convergence. A very good property of
this method is that the final eigenvectors are as accurate as the user
wants them to be, so the results are no different from a full-blown
diagonalization.
Because the method is iterative, it takes longer to converge than
a regular diagonalization. Sizewise it can handle almost anything on
a moderately sized computer. David Perahia calculated a few dozen modes
of Hemoglobin (~600 residues = ~6000 atoms = ~18000 d.o.f.) on a
SGI workstation with 90 Mb memory. I have done several calculations
on 900 residue systems. The actual time to reach convergence depends
on the available memory, the desired accuracy, and the number of
requested normal modes.
One other area where the method saves memory is in the storage of the
original Hessian. Since this matrix is usually sparse for large systems,
a compressed Hessian is set up, which contains all non-zero elements.
In addition, I added the option to used this compressed Hessian in the
reduced-basis diagonalization option of VIBRAN. Before, the same size
limits applied to full diagonalizations and reduced-basis diagonalizations.
This should not be: people usually want to do reduced-basis calculations
because the molecule is too big for the Hessian to be stored in memory.
The option VIBRAn REDUce CMPAct will fill the compact Hessian and
form the reduced-basis Hessian from this compact Hessian. Overall, this
is a big saving on memory space.
Files Added:
build/UNX/dimb.mk
source/dimb/compact.src
dimb/dimbcray.src
dimb/dimbsub.src
dimb/dimbutil.src
dimb/eispack.src
dimb/enecmp.src
dimb/nblist.src
dimb/nmdimb.src
fcm/dimb.fcm
Files Modified:
doc/vibran.doc
source/charmm/iniall.src
energy/ecnstr.src
/ehbond.src
/eintern.src
misc/noe.src
/sbound.src
nbonds/enbond.src
/etable.src
vibran/redbas.src
/vibio.src
/vibran.src
New/Modified Testcase:
test/c24test/dimb.inp
/redu_cmp.inp
------------------------------------------------------------------------------
3.9 Make BLOCK work with IMAGE/CRYSTAL and vice versa
Name : Stefan Boresch
Email Address : boresch@tammy.harvard.edu
Institution : Harvard University
Date : November 1994
CHARMM Version: c23f4
(1) BLOCk CLEAr command now also frees memory of IBLCKP/IBLOCK
data-structure
(2) Intermediate values printed out by the EAVG command with CONT +n
were wrong, the program flow/logic of this subroutine has been
overhauled.
(3) Post-processing routines FREE, EAVG and COMP will actually
print IMAGE terms if present; changes were necessary in EAVG
COMP.
(4) If INBFRQ, IMGFRQ (and IHBFRQ) are correctly specified, then
FREE, EAVG and COMP will update nonbonded list while reading
trajectories. (THIS IS IMPERATIVE WITH IMAGEs!)
Real/Image atoms leave/enter the simulation box/system dynamically.
Therefore, the nonbonded/image interaction lists have to be updated during
post-processing. The hooks were already in the program, subroutine bluplst.
The real changes hide in this routine, most changes in blfree, bleavg and
blcomp are either cosmetic or ensure proper printout (cf. (3)). The routine
(bluplst) is a sibling of routine updeci in heurist.src. The heuristic
update scheme itself is removed, as I feel that one should update the lists
at every frame. Also, the CRYSTAL specific section of UPDECI is not present
in BLUPLST as I don't understand it. Therefore, care should be exercised
when using BLOCK with CRYSTAL! Negative values of INBFRQ/IMGFRQ are trapped,
in this case they are set to 1; Printout from the update / list generation
routines is suppressed by temporarily raising the PRNLEV to 1 (Can this be
done more elegantly?)
(5) So far, if IMAGES (CRYSTAL) and BLOCK were invoked at the
same time, one would simply bomb because of a memory access
violation. This has been fixed.
BLOCK uses two data-structures, one containing the interaction matrix
between blocks, and one containing the block number for each atom (IBLCKP).
This array was allocated so far as INTEG4(NATOM) on the heap. However, when
IMAGE atoms are present, the energy routines attempt to find out to which
block an IMAGE atom belongs. This at one point or the other causes a memory
access violation. The solution consists out of two parts. (i) The IBLCKP
data-structure is now allocated as INTEG4(MAXAIM) on the heap; therefore
there is always enough space provided. (ii) The entries for the IMAGE atoms
have to be initialized, and this has to be done at EVERY image update.
However, similar things are already done for a number of other quantities
like masses, vdW params, charges etc. All this is done among a number of
other things in subroutine MKIMAT in upimag.src, where I have added an
appropriate statement.
Files Modified:
source/image/upimag.src
pert/block.src
New/Modified Documentation:
The current BLOCK documentation is very poor, and in my
opinion highly misleading in a few instances. Following this great
tradition, the following paragraph contains the bare minimum of what
is necessary to make BLOCK and IMAGE work together: (PLEASE, make sure
that it gets included in the c24 documentation!) I do not attempt to
improve the documentation as a whole.
While running dynamics or calculating an energy nothing has
changed for the user; BLOCK now simply coexists with IMAGE (CRYSTAL?)
(and even gives apparently correct results). However, if IMAGEs are
present, then POST-PROCESSING has to be carried out in a slightly
different fashion. (i) The only commands which I support are FREE,
EAVG and COMP. AVER might work, but I didn't check that. (ii) BLOCK
now supports list updates while post-processing (any of the three
commands, FREE, EAVG and COMP will accept (and use) the IMGFRQ, INBFRQ
and IHBFRQ keywords. If one doesn't use IMAGE, the old tricks will
still work: [1] FIX everything that was in BLOCK 1 during the
simulation, then [2] generate a list with CUTNB 99. and don't update
during post-processing. What that does is to generate a nonbonded
list which contains all possible interaction pairs; hence, this list
never has to be updated. HOWEVER, such a strategy CAN'T work when
IMAGES are present!!! You can (and should) still FIX all atoms that
were in BLOCK 1 during the simulation (otherwise post-processing will
consume huge amounts of CPU time unnecessarily). The further strategy
is as follows (I use EAVG, the same applies to FREE and COMP)
! set up system; parameters, psf, coordinates, IMAGE
! transformations, centering
energy [correct non-bonded and IMAGE options, cutoffs]
cons fix sele [all that was in block 1]
open trajectories
block
! partition the system as appropriate for what you want to do
EAVG [usual eavg options] INBFRQ 1 IMGFRQ 1
end
The post-processing commands will still default to inbfrq 0 and
imgfrq 0. This will give NONSENSE with IMAGEs!!! I recommend
an update frequency of 1 (as the frames in the trajectory will actually
be several MD steps apart!). Negative values (i.e. heuristic updates)
are not supported, you'll get a low priority warning and I took the
liberty to default to inbfrq=imgfrq=1 in this case. Check out
block3.inp (the new test-case) It doesn't make a lot of physical sense,
but both production and post-processing is (formally) carried out
correctly.
New/Modified Testcase:
test/c24test/block3.inp
Other Comments:
Most of the testing has been done with IMAGE. CRYSTAL, when
used to build a periodic solute/solvent box also worked. However, I
don't know CRYSTAL well enough to really thoroughly test it; could an
expert check this out? Could we come up with a realistic
CRYSTAL/BLOCK test-case?
In upimag.src (subroutine mkimat), I operate directly on the
heap. This should probably be done somehow nicer. In the meantime,
can someone check whether the code runs on a 64bit architecture at
all?
------------------------------------------------------------------------------
3.10 Arithmetic Expression Interpreter
Name : Benoit Roux
Email Address : rouxb@ERE.UMontreal.CA
Institution : University of Montreal
Date : January 1995
CHARMM Version: c24a1
An intepretor of arithmetic expression has been added to the CHARMM
command parser. It is called at the level of the miscelaneous command
handling using simply by the word CALC (for calculator). It can be used to
evaluate algebraic numerical expression. The command supports all
mathematical numerical expression with arbitrary number of nesting of
recursive parentheses, e.g., exp(1.0-cos(2*(log(2*pi))**2)/.5).
The parsing is actually very crude since the expression is translated back
and forth between character string and a real variable to handle the logic
(there is no real subroutine recursion).
Files Added:
source/misc/calc.src
Files Modified:
source/charmm/miscom.src
Modified Documentation:
doc/miscom.doc
New Testcase:
test/c24test/calc.inp
Other Comments:
A suggestion for the future? The command could be even more useful
is all input of any real number was always processed with CALC (one level
above the string-number conversion in GTRMF). One could use the CALC
subroutine inside GTRMF to process any expression in any command (on the fly).
For example, the input lines:
CALC angle = asin(0.75742)
coor rotate phi @angle xdir 1.0 select end
could be directly processed as:
coor rotate phi asin(0.75742) xdir 1.0 select end
That could make lots of input lines more compact if needed.
------------------------------------------------------------------------------
3.11 Upgrade of the NMR module
Name : Benoit Roux
Email Address : rouxb@ERE.UMontreal.CA
Institution : University of Montreal
Date : January 1995
CHARMM Version: c24a1
The NMR module is upgraded to have better output style. The old version
used the value of PRNLEV to choose the printed quantities. Since this was
a non-standard style in CHARMM, a series of logical flags have been
included in the command calls to print some chosen quantities. In addition,
the chemical shift anisotropy (CSA, used in solid state NMR of membrane
proteins in oriented samples) has been redefined in term of a
zmatrix to prevent confusion. The deuterium quadrupolar splittings (DQS)
command is also upgraded. A bug in a call to NORMAL was fixed.
Files Modified:
source/misc/nmr.src
Modified Documentation:
doc/nmr.doc
Modified Testcase:
test/c23test/nmrtest1.inp
Other Comments:
The DQS and CSA code did not work properly in the previous CHARMM version
because the normalization of some unit vectors was performed by a call to
the standard subroutine NORMAL and one argument was missing in the
modified call. The modification was done after the nmr.src file was
checked in.
There has been some discussion with people involved with the NOE module
to somehow connect it to the NMR facility. For example, the full relaxation
matrix could be obtained and compared with the list of NOE restraints that
is used. Some specific project along those line should be formulated.
------------------------------------------------------------------------------
3.12 Other Bugfixes and Enhancements
(1) Initialize R2 in ehbond.src
- Ryszard Czerminski (B940822.rcz)
There is a logic flow in energy/ehbond.src (subr. EHBOND) resulting in
computation with undetermined values and in "floating point problems"...
If S.GT.C2OFHB and ANALYS.ne.3 R2 will not be calculated but used later on...
Files involved:
source/energy/ehbond.src
..............................................................................
(2) Bug in INQUIRE Function on ALPHA Machine
- Ryszard Czerminski (B940824.rcz)
Machine : DECALPHA
Operating System : OSF/1 V2.0 240
As a work around I have increased MAXLEN from 80 to 256.
At the same time I have eliminated MXFLNM and replaced it with MAXLEN.
Files involved:
source/util/title.src
..............................................................................
(3) Uses of CHARMM Variables in Parallel Environment
- Lennart Nilsson (B940825.ln)
The last argument is left off when invoking the slave processes.
No problem as long as only the master process uses the variable,
but hangs the whole thing if the slaves need to have the correct value.
In parcomm.src, two loops having the number of program arguments as upper
bound should use NPRARG, not NPRARG-1.
Files involved:
source/machdep/parcomm.src
..............................................................................
(4) Write Commands in Parallel CHARMM with Slave Proccesses
- Lennart Nilsson (B940907.ln)
The command "write ... name ..." is incorrectly parsed by the slave
processes in a parallel environment and they crash (or hang).
Files involved:
source/io/mainio.src
..............................................................................
(5) ENCODF of ".0"
- Arnaud Blondel (B941220.ab)
On some machines' Zero is written .0. This causes ENDCOF to return
empty parameter.
Files Modified:
source/util/string.src
..............................................................................
(6) NOE Bugfixes
- Lennart Nilsson (B941221.ln)
(6.1) The RESET command in the NOE module does not zero the variable NOENM2
so add
NOENM2=0
to the RESEt handling
(6.2) When reading old format NOE input files the following two lines are
needed in the OLD section of the ASSI parser
RNEXP=THREE
RNTCN=ZERO
Files Modified:
source/misc/noe.src
..............................................................................
(7) Verlet Dynamics Coordinate Update
- Lennart Nilsson (B941222.ln)
- Ryszard Czerminski (B941221.rcz)
Coordinates after dynamics (verlet) are propagated one extra step.
When using the standard verlet integrator (ORIG keyword to dynamics command)
the coordinates in charmm after a dynamics command are not the same as the
coordinates for at the last step of dynamics, as seen if an energy command is
issued immediately after the dynamics. This does not happen with the leap
integrator.
This is because in dynamcv x,y,z are updated after the energy call in
the main dynamics loop and before the ENDDO, whereas in dynamc x,y,z are
updated before the energy call, and thus after the loopback decision is made.
DYNAMC (leap):
loop
x=xcomp+xold
writec x
updeci x
energy x
printe x
xold=....
xcomp=x
endloop ! x is still what it was at the energy call in step i
DYNAMCV (verlet):
loop
writec x
updeci x
energy x
printe x
xold=x
x=xnew
endloop ! x is NOT what it was in step i, rather what it will be in i+1
In order to fix this, Ryszard added following lines at the very end of
DCNTRL routine
...
C
##IF OLDDYN
IF(ORIG) THEN ! bug#92242xfsq01
DO I=1,NATOM
X(I)=XOLD(I)
Y(I)=YOLD(I)
Z(I)=ZOLD(I)
ENDDO
ENDIF
##ENDIF ! OLDDYN
NOTE: This fix results in differences in dynamics benchmarks.
Files Modified:
source/dynamc/dcntrl.src
..............................................................................
(8) Reading Lower Case PDB Files
- Lennart Nilsson (B950104.ln)
Lower case characters in charmm/pdb coordinate files are not understood.
Call CNVTUC after reading lines from charmm/pdb formatted files. Also change
name of varaible LEN (conflict with intrinsic function) to SLEN.
Files Involved:
source/io/coorio.src
..............................................................................
(9) Cray Dependent Coding Fixed
- Ryszard Czerminski & Mike Mirman (B950105.rcz)
FORTRAN version : CRAY FORTRAN CFT77 6.0.4.2
The construct does not compile properly on CRAY
##IF IBM IBMRS CRAY
FUNCTION FUN(A)
REAL*8 FUN
##ELSE
REAL*8 FUNCTION FUN(A)
##ENDIF
and the CRAY key is removed in ##IF
Files Inveloved:
source/dimb/eispack.src
image/crystal.src
minmiz/tnpack.src
quantum/qmpac.src
util/random.src
string.src
util.src
vector.src
..............................................................................
(10) Non-integer Charge for Quantum Molecules
- XL Charlene Liang (B950104.xl)
The quantum inplementation (MOPAC not GAMESS) only works with
integer (0,1,-1,...) charged QM system, while it is not always the case.
"quantum charge -0.35" is allowed in the fix.
Files involved:
source/quantum/qmset.src
qmpac.src
fcm/quantm.fcm
..............................................................................
(11) Density Matrix in QM/MM Dynamics Restart File
- XL Charlene Liang (B950105.xl)
Add the density matrix at the end of the restart file and read it back.
Files involved:
source/dynamc/dynio.src
..............................................................................
(12) Subroutine EQUIV replaced by COPYR8
- Ryszard Czerminski (D941228.rcz)
The subroutine EQUIV is removed because it is identical to COPYR8 except
the calling sequence. EQUIV calls are replaced with COPYR8 calls.
Files Involved:
source/util/vector.src
dimb/dimsub.src
dimbutil.src
vibran/raise.src
redbas.src
vibio.src
vibsub.src
vibutil.src
..............................................................................
(13) Branch into IF/DO Block Problem Fixed
- P950105.elm
Files Involved:
source/dynamc/dynamvv.src
misc/aspener.src
/surfac.src
vibran/vibran.src
CHARMM c24a3 (March 15, 1995) to c24b1 (August 15, 1995)
4.1 Constraint correction for PERT
Name : Stefan Boresch
Email Address : boresch@tammy.harvard.edu
Institution : Harvard University
Date : April 1995
CHARMM Version : c24a3
The current version of PERT cannot handle situations where SHAKE is
applied to bonds which change in length due to an alchemical mutation
as SHAKE and PERT do not "communicate". Furthermore, in such cases a
constraint correction has to be computed and added to the free energy
difference. Two steps are required to fix this problem:
(1) The constraint list needs to be updated as a function of the
coupling parameter lambda.
(2) The constraint correction has to be calculated.
In the following, the more obscure details of the current
implementation are briefly explained.
Only thermodynamic integration (both for slow-growth and windowing) is
supported; the exponential formula will give nonsense results. (If
someone wants to fix this, please look at Pearlman/Kollman, JCP 1991,
94, 4532 and Severance et al. J. Comput. Chem. 1995, 16, 311.)
The method to calculate the constraint corrections is based on
extracting the respective Lagrangian multipliers from the SHAKe
routine; this approach is briefly described in van Gunsteren et al.
Computer Simulation of Biomolecular Systems: Theoretical and
Experimental Applications; ESCOM: Leiden 1994; Vol. 2, pp 315-348.
The approach fully includes inertial contributions, it is left to the
user to account for those correctly in the context of the problem.
The code assumes that the SHAKE command is issued after PERT has been
activated and all modifications of the PSF have been carried out. At
this point the current PSF corresponds to lamb=1 and the original PSF
(saved) to lamb=0. The SHAKE constraint list is built as usual;
however, it is checked whether one (or both) of the atoms are affected
by the perturbation (selection in the PERT command). All such bonds
could theoretically change their equilibrium bond-length, which is
extracted from both PSFs (I explicitly verify that I have
corresponding bonds, but I'm not sure that this is necessary or could
not be done much more elegantly). If the lamb=0 and lamb=1
bond-lengths are different, then this constraint is specially marked
and a counter is incremented. If any constraint is affected by the
perturbation, a logical flag QPSHAKE is set to true. From this
description, it is easy to see that only the PARAmeter option can be
meaningfully supported. PCONSTR is a logical array of size MAXSHK
which indicates whether a certain constraint is a function of lambda.
NPCONST is the number of these constraints. An auxiliary array
PSHAUX(MAXSHK,2) exists as well. The first entry is used to
accumulate the respective Lagrangian multiplier during a call to
SHAKE, the second contains the derivative of the constrained
bond-length with respect to lambda.
If QPSHAKE=TRUE, every time SHAKEA or SHAKEF (the actual SHAKE
routines) are called, they check whether lambda has changed since the
previous call. If this is the case, then the constraint list is
updated first. The Lagrangian multiplier L_i for the i-th constraint
can be obtained from the variable ACOR used in SHAKEA, of course it
has to be accumulated over all the iterations in the SHAKE algorithm.
L_i = - (1/dt^2) * ACOR, where dt is the MD timestep
In those stages of a dynamics run where PERT accumulates the
quantities required to compute free energy differences, the Lagrangian
multipliers are used to calculate the constraint correction:
dU(cons.)/dlambda = Sum (L_i * sqrt(constr(i) * pshaux(i,2)),
where the i's are all constraints affected by PERT
The one problem is that dU(cons.)/dlambda is not calculated at the
same point as all the other energy terms. This causes problems with
the standard way PERT handles bookkeeping in EPERT. The regular
energy terms are handled as follows:
call energy (which calls epert)
call shake
calculate corrected pressure, kinetic energy (if NVT dynamics)
accumulate all energy (potential, kinetic, pressure terms)
However, the free energy related quantities are calculated AND
accumulated already at the energy step, before SHAKE is called and the
correction can be calculated. I have circumvented the problem by
taking the accumulation routines out of epert and making them into a
separate subroutine EPSUM, which is called (where appropriate) later
from the integrator. This allows to treat dU(cons.)/dl just as all
the other dU/dl terms.
Finally, I have changed the calling sequence for SHAKE in the Nose
integrator as included in dynamcv.src. THIS LEADS TO DIFFERENT
RESULTS IN THE TESTCASES (block3, which uses Nose; nose1, nose2),
although both approaches are in principle equivalent. The current
Nose sequence is more efficient, but I require this calling sequence
for my method to work. Masa Watanabe and I are thinking about ways
how we could solve this dilemma more elegantly.
Files Added:
fcm/pshake.fcm
Files Modified:
dynamc/dynamc.src
dynamc/dynamcv.src
manip/shake.src
pert/pert.src
pert/epert.src
New/Modified Documentation:
The new code is mostly transparent and does not really require
additional documentation. However, the following information should
be added to pert.doc:
If SHAKE is applied to bond terms which are changed as the
result of an alchemical mutation a constraint correction is calculated
where required in slow-growth mode and TI in windowing mode. The
exponential formula in windowing mode is not supported. The user has
to beware of subtle problems regarding a possible "moment of inertia"
term that may be or may be not included in this correction (S. Boresch
& M. Karplus, to be published) In order for the constraint correction
to work properly, attention has to be given to the following points:
(1) SHAKE must not be applied to angle terms
(2) the PARA option has to be used (it's not clear, how to support
reference coordinates in the context of an alchemical mutation)
(3) the SHAKE command has to issued after the PERT command. (This
is similar to setting up IMAGEs in connection with PERT). A
typical input will look something like
PERT SELE ... END
!change psf; after ALL changes have been made
SHAKe BOND PARA
DYNA ... ! carry out MD simulations etc.
PERT OFF
Items (2) and (3) can lead to error messages in situations where there
is actually no problem, e.g. you just want to apply SHAKe to your
solvent which is not affected by the mutation, so you specify SHAKe
before PERT and "bomb". Nevertheless, I thought better safe than
sorry and if wanted one can override the warnings with a BOMLEV -3.
New Testcase:
c24test/pert2.inp (self-contained)
Other Comments:
While adding the PERT/SHAKE communication and implementing the
constraint correction, I ran into a couple of minor bugs which
I have reported elsewhere. There is one bug, however, which may
lead to problems in connection with a moment of inertia term. I have
reported it separately, but I repeat the description here:
In the DYNAmics comand, the NOXR NOYR NOZR NOXT NOYT NOZT options,
which are undocumented and, admittedly, should only be used by
advanced users, do not work. The following logical statement in
dynamc/dcntrl.src has to be changed from
IF(QNORT(I)) QALWRT(I)=.FALSE.
to
IF(QNORT(I)) QALWRT(I)=.TRUE.
otherwise the avove options have no effect whatsoever.
Limitations:
* As mentioned, the exponential formula is not supported; maybe
one should suppress output in cases where the constraint correction
is used (as it is nonsense anyways)
* I currently report the constraint term in the user energy term;
maybe should change that to the constraint energy term.
* Finally, suggestions how to make the interaction between PERT and
SHAKE (and the DYNAMICS commands) less ugly are welcome. However, at
least things seem to work...
------------------------------------------------------------------------------
4.2 MMFP/SSBP Upgrade
Name : Benoit Roux and Dmitrii Beglov
Email Address : rouxb@ERE.UMontreal.CA, beglovd@ERE.UMontreal.CA
Institution : University of Montreal
Date : Jan-1995
CHARMM Version : c24a1
The Miscelaneous Mean-Field Potentials (MMFP) has been upgraded.
The spherical solvent boundary potential (SSBP) has also been
incorporated into EPERT. A new "membrane-like" planar potential
has been introduced using Gaussians to provide a smooth free energy
function based on hydropathy profile of individual amino acids
and solvent exposure. This is usuful to orient membrane proteins.
A new primary shell of hydration has been added to the MMFP facility
to provide one layer of solvent around a flexible polypeptide.
For more information, see Beglov & Roux, Biopolymers 35: 171-178 (1995).
Files Added:
misc/primsh.src
Files Modified (module/file.src):
misc/mmfp.src
misc/ssbp.src
pert/epert.src (introduce the two calls to SSBP)
energy/energy.src (introduce the call to SSBP)
energy/printe.src (just to include the SSBP energy term in the print-out)
energy/eutil.src
fcm/mmfp.fcm
fcm/ssbp.fcm
fcm/energy.fcm (just include the SSBP energy term)
Modified Documentation:
mmfp.doc
New/Modified Testcase:
mmfptest.inp (modified)
mmfptest3.inp (new)
Other Comments:
Warning! In using the SSBP it is quite important to avoid truncating the
interactions within the spherical region. Extended electrostatics is
the method of choice for that. There is an imbalance of the Kirkwood term
(reaction field) with the a system that has truncated electrostatics.
This leads to crazy behavior of the water sphere. A peculiar problem
was also noticed by a few people when using SSBP. First, it should
be emphasized that no-cutoff (or extended electrostatics) simulations
are needed. Second, it seems that spuriously bad results can be
obtained (non-spherical water balls) when the number of multipoles in
the Kirkwood expansion is too low (e.g., 3). The default is 15 and no
problems were experienced with this number. Right now, the SSBP is
implemented for a finite spherical region. The treatment for new
"active site solvation", in which a large part of the protein lies
outside the spherical region, is begin developed.
------------------------------------------------------------------------------
4.3 TNPACK Update
Note: The truncated-Newton minimizer (TNPACK) was introduced into
CHARMM 23a2 (October 23, 1992) by Berard R. Brooks and Tamar Schlick.
It is updated by Youngdo Won upon receiving the following check-in
suite from Tamar Schlick, Phillipe Derreumaux and Eric Barth.
Name : Tamar Schlick and Phillipe Derreumaux
Email Address : schlick@acf4.nyu.edu
Institution : New York University
Date : March 16, 1994
CHARMM Version : c24a1
The truncated-Newton minimization package TNPACK, developed by
T. Schlick and A. Fogelson, has been incorporated into CHARMM and
adopted for biomolecular energy minimization. TNPACK is based on the
preconditioned linear conjugate-gradient technique for solving the
Newton equations. The structure of the problem --- sparsity of the
Hessian --- is exploited for preconditioning.
Thorough experience with the new version of TNPACK in CHARMM has
been described in a paper now in press in the Journal of Computational
Chemistry: Applications are reported for a series of molecular systems
including Alanine Dipeptide (N-Methyl-Alanyl-Acetamide), a dimer of
N-Methyl-Acetamide, Deca-Alanine, Mellitin (26 residues), Avian
Pancreatic Polypeptide (36 residues), Rubredoxin (52 residues), Bovine
Pancreatic Trypsin Inhibitor (58 residues), a dimer of Insulin (99
residues), and Lysozyme (130 residues). Through comparisons among the
minimization algorithms available in CHARMM, we find that TNPACK
performs significantly better than ABNR in terms of CPU time when
curvature information is calculated by a finite-difference of
gradients (the "numeric" option of TNPACK). The CPU gain is 50% or
more (speedup factors of 1.5 to 2.5) for the largest molecular systems
tested and even greater for smaller systems (CPU factors of 1 to 4 for
small systems and 1 to 5 for medium systems). With the analytic
option, TNPACK converges more rapidly than ABNR for small and medium
systems (up to 400 atoms) as well as large molecules that have
reasonably good starting conformations; for large systems that are
poorly relaxed (i.e., the initial Brookhaven Protein Data Bank
structures are poor approximations to the minimum), TNPACK performs
similarly to ABNR.
TNPACK uses curvature information to escape from undesired
configurational regions and to ensure the identification of true local
minima. It converges rapidly once a convex region is reached and
achieves very low final gradient norms, such as of order 10E-8, with
little additional work. Even greater overall CPU gains are expected
for large-scale minimization problems by making the architectures of
CHARMM and TNPACK more compatible with respect to the
second-derivative calculations.
This work should be the focus of future developments. Such work
involves sparse storage of the Hessian, efficient sparse
Hessian/vector multiplications, and separation of the gradient and
Hessian calculations.
Files Involved:
minmiz/minmiz.src
egrad1.src
tnpack.src
tndriv.src
nbonds/enbond.src
fcm/euler.fcm
Documentation Updated:
doc/minmiz.doc
New Testcases:
c24test/tntest1.inp
------------------------------------------------------------------------------
4.4 NIH Developments
Name : Milan Hodoscek, Scott Feller, Yuhong Zhang,
Richard M. Venable, Jin Zhou and Bernard R. Brooks
Email Address : brbrooks@helix.nih.gov
Institution : Laboratory of Structural Biology, NIH
Date : June 28, 1994 - July 25, 1995
CHARMM Version : c24a3
..............................................................................
(1) New RESDistance facility
A new facility, RESD, has been created to allow general distance restraints
based on a linear combination of distances. This is useful for searching
reaction pathways.
Files modified:
charmm/charmm_main.src
/iniall.src
energy/energy.src
/eutil.src
/printe.src
fcm/dimens.fcm
/energy.fcm
io/mainio.src
Files added:
fcm/resdist.fcm
manip/resdist.src
..............................................................................
(2) New READ PARAm APPEnd option.
An append option has been added to the READ PARAM CARD command. This allows
just a few parameters to be modified without editing an entire parameter file.
A modification to the binary parameter file format was necessary.
Old binary files may not be appended, but they are still supported.
Files modified:
charmm/iniall.src
fcm/dimens.fcm
/param.fcm
io/mainio.src
/parmio.src
..............................................................................
(3) New READ PSF APPEnd option.
An append option has been added to the READ PSF command. This allows PSFs
to be easily merged to make a larger PSF. No modification to the binary
parameter file format was necessary. This option works with both FILE
and CARD options.
Files modified:
io/mainio.src
/psfres.src
fcm/dimens.fcm
..............................................................................
(4) Best fit operation added as option to CORREL TRAJectory command.
The TRAJectory command in correl now accepts an ORIENt keyword with an
optional [MASS] qualifier in conjunction with a second atom selection
that will best fit selected atoms with respect to the rms deviation
from the reference structure (in the comparison coordinate set). This
operation is done prior to the determinination of any time series
value. This operation will not affect any time series value that is
based only on relative distances and angles.
Files modified:
charmm/usersb.src
correl/anacor.src
/correl.src
..............................................................................
(5) Modification of ECHECK processing
When dynamics is terminated prematurly due to the ECHECK tolerance,
DCNTRL now resets PRNLEV and frees used stack space. This makes it
easier for error processing when CHARMM continues (with a low
specified BOMLEV value).
Files modified:
dynamc/dcntrl.src
..............................................................................
(6) Energy printout modified.
The energy printout fields have been modified based on new developments.
Special formats for numbers that are too large have been developed.
Files modified:
energy/printe.src
..............................................................................
(7) QM/MM exclude group option added
An option EXGRoup has been added which causes all atoms in the group
of the link atom host to be excluded from the QM/MM electrostatic
interaction terms. Code for specifying the charge of link atoms and
their placement has also been added.
Files modified:
fcm/gamess.fcm
gamint/gamini.src
quantum/addlnat.src
..............................................................................
(8) X-window graphics extensively modified.
Several new features have been added to the X-window version of CHARMM
graphics. This code has also been tested on a wider variety of
hardware platforms (for example: SGI).
Changes include: double-buffering, clipping, StaticColor, symbol fonts,
window title, modified colormap calls, and a misc. Bug fixes in the
labeling of the X axis. A NODISPLAY compile option has been added to
the X windows version of CHARMM graphics in which only derivative
files are produced. The GRAPhics NOWIndow option can be used to
generate the same ffect at run time.
Files modified:
fcm/graph.fcm
graphics/apodraw.src
/apograph.src
/drawit.src
/graphx.src
/psdraw.src
machdep/cstuff.c
/socket.c
/xdisp.c
..............................................................................
(9) Overhaul of the image update code.
Due to the cost involved in the image update (when running CHARMM in
parallel), the routine MKIMAT was gutted by the removal of all atom
based loops. These were only used to compute the minimum distance for
an image to the primary atoms for the prinout in the image update.
This shortest distance is no longer printed as an accurate value.
Image atom positions are now computed with a call to TRANSO.
Files modified:
image/upimag.src
energy/eutil.src
..............................................................................
(10) Langevin Piston extended pressure system and surface tension ensembles
New Langevin Piston method for contant pressure implemented. The unit
cell information has been added to the trajectory file format. This
implementation required an update to the image and crystal code which
cleaned up some ancient problems. Options for including the surface
tension (gamma-Area) term is also completed and tested. This has been
developed for the accurate simulation of interfacial systems.
Files modified:
fcm/reawri.fcm
/energy.fcm
/image.fcm
dynamc/dcntrl.src
/dynamc.src
/dynamcv.src
/dynamvv.src
/dynio.src
/prssre.src
energy/energy.src
/printe.src
image/crystal.src
/eimage.src
/imnbf2p.src
/xtlfrq.src
minmiz/egrad1.src
misc/testch.src
pert/epert.src
..............................................................................
(11) Enhancements to the Ewald code
The EWALD electrostatic option now runs efficiently for parallel
architectures. Also, the maximum K-space values can be specified
independently for each direction. Several bugs were fixed.
Files modified:
fcm/ewald.fcm
/image.fcm
/inbnd.fcm
/inbnd.fcm
nbonds/ewald.src
/ewaldf.src
/nbutil.src
..............................................................................
(12) Enhancement to parallel code.
There has been continued development of the parallel code for CHARMM.
This includes new features run in parallel, new machine types supported,
new parallelization methods, and code made to run more efficiently.
Due to conflict in routine names with library routines, the subroutines:
WRITEC and READC had to be renamed.
Initial code to allow the use of the Terra parallel computer has been added.
Added preflx keyword SGIMP for multiprocessor SG machines using PVM
massage passing library. The difference between PVM and (SGIMP,PVM)
is that all the processes are spawned on one host and some communication
parameters are not supported on MP machines. It can be used on a single
processor SG for testing purpose. Use PVM only on a cluster of any type of
workstation.
Files modified:
charmm/iniall.src
/miscom.src
correl/solana.src
dynamc/dynamc.src
/dynamcv.src
/dynamvv.src
/dynio.src
/dynsub.src
fcm/dimens.fcm
/exfunc.fcm
/impnon.fcm
/iochan.fcm
/mmfp.fcm
/number.fcm
/parallel.fcm
image/crystal.src
io/mainio.src
/rtfio.src
/trajio.src
machdep/cstuff.c
/machio.src
/machutil.src
/parallel.src
/socket.c
/space.src
/startup.src
/xdisp.c
minmiz/tnpack.src
misc/nmr.src
nbonds/nbndgc.src
quantum/qmpac.src
rxncor/travel.src
util/random.src
/string.src
/util.src
/vector.src
vibran/vibio.src
Files added:
machdep/parset.src (replace parcomm.src)
Files removed:
machdep/parcomm.src
------------------------------------------------------------------------------
4.5 Bugs and Problems Fixed at NIH
Name : Milan Hodoscek, Scott Feller, Yuhong Zhang,
Richard M. Venable, Jin Zhou and Bernard R. Brooks
Email Address : brbrooks@helix.nih.gov
Institution : Laboratory of Structural Biology, NIH
Date : June 28, 1994 - July 25, 1995
CHARMM Version : c24a3
..............................................................................
(1) Commands and features removed
The FAST command subcommands CONVex, IRIS, SGI, and SUN have been removed.
The equivalent function can be invoked by using "FAST ON" instead.
Files modified:
charmm/miscom.src
..............................................................................
(2) The CORREL SHOW command is re-enabled.
The CORREL SHOW command was inadvertantly disabled due to enhances in the
SHOW (parameter) command. Additional code has been added which recreates
the command line if the SHOW (parameter) command has unrecognized features.
Files modified:
charmm/miscom.src
..............................................................................
(3) Bug in parsing the piston masses
There was a stupid bug in the parsing of the piston mass for
off-diagonal piston elements (for non-orthogonal space groups).
Files modified:
dynamc/prssre.src
..............................................................................
(4) Bug in the printout of bond distances in the analysis facility.
The wrong variable was printed in EBOND when the analysis facility was
enabled (and the appropriate atoms selected).
Files modified:
energy/eintern.src
nbonds/enbond.src
..............................................................................
(5) Bug an GAMESS alignment of common block CGAMES
There was an alignment bug in the common block CGAMES. A logical variable
(QGMREM) was moved to another common block to correct this problem.
Files modified:
fcm/gamess.fcm
gamint/gamess/int1.src
/grd1.src
/scflib.src
..............................................................................
(6) Bug in the COOR HELIX command
There was a bug in the coor helix command in which the wrong variable
was used. This was missed in early testing because it only shows up
when a helix is not near the origin or if initial memory is not zeroed.
Files modified:
manip/helix.src
..............................................................................
(7) Bug in parsing the IC DELEte command.
There was a logic bug in the parsing of the IC DELEte command.
In particular, if the [IMPR] option is used, then ONLY improper terms
can be deleted. If the [DIHE] option is specified, then ONLY normal
terms can be deleted. In the (unfixed) version, any item specified
was not subject to deletion regardless of the atom selection.
Files modified:
manip/intcor.src
..............................................................................
(8) Bug in the use of the double atom selection in the SHAKE command.
There was a parsing bug in the use of the double atom selection in the
setup of SHAKE constraints. The command was supposed to incorporate a
constraint if each atom of a pair was a member of a different atom
selection. The code only worked for the (usual) case where only one
atom selection is provided. Also, the COOR SHAKE command now checks
to see if all constrained atoms have defined positions before proceeding.
Files modified:
manip/corman.src
/shake.src
..............................................................................
(9) Bugs in the initialization of NOE restraint terms.
Several lines were missing from the initialization and parallelization
section of the NOE restraint module.
Files modified:
misc/noe.src
..............................................................................
(10) Bugs in the ewald energy routines.
There was a bug in the calculation of the vdfw forces in the ewald energy
routines. It caused CHARMM to produce bad forces. There was also a
bug in use of the cutoff distance which caused heating during dynamcs
(bug gave correct energies when using the ENERgy command).
Files modified:
nbonds/ewald.src
..............................................................................
(11) Default file numbers in DCNTRL modified.
The default file numbers have been changed to -1 in order to prevent
simple errors when multiple DYNAmics commands are used in a single
CHARMM run. With this modification, the units must be specified for
each invokation of DYNAmics.
Files modified:
dynamc/dcntrl.src
..............................................................................
(12) Bug in generate of PERT group nonbond list
There were bugs in the generation of the group-group PERT nonbond list
that has been corrected. The PRINT NBONd command has been expanded so
that it prints out all nonbond lists (PERT, IMAGE,...). This will
facilitate future debugging efforts.
Files modified:
image/nbondm.src
nbonds/nbondg.src
/nbutil.src
..............................................................................
(13) High PRNLEV Printout added in ENBOND
High PRNLEV print staements have been added to ENBOND to facilite the
testing and decoding of various nonbond options and FAST options.
Files modified:
nbonds/enbond.src
..............................................................................
(14) Incorrect order of statements corrected
DATA statements have been moved so as to follow declaration statements.
Descriptions (EXTERNAL, EQUIVALENCE,...) also follow declarations
(REAL*8,...).
..............................................................................
(15) Improved atom selection error handling
The handling of errors in the atom selection is a bit better.
Files modified:
util/select.src
..............................................................................
(16) Leading zero added to floating point conversions.
A leading zero has been added to all conversions text of floating
point numbers. This makes the comparison of output from different
compilers a bit easier.
Files modified:
util/string.src
..............................................................................
(17) Jumps into an IF block.
Bugs in the analytic surface area routines where a branch jumped into an IF
block have been fixed.
Files modified:
misc/surfac.src
------------------------------------------------------------------------------
4.6 Other Bugfixes and Enhancements
The following are the bugfixes cumulated to charmm-bugs@tammy.harvard.edu
during the period of February 1, 1995 through August 15, 1995.
..............................................................................
(1) Initialize BSTXCS in ALLHP
- Ryszard Czerminski (B950315.rcz)
Initialize BSTXCS to 0 by the DATA statement, DATA BSTXCS /0/
Files involved:
util/util.src
..............................................................................
(2) REPLICA set-up code corrected
- Charles L. Brooks, III (B950321.clb)
Replica command does not function properly when using with setup.
IC tables are corrupted. Problem manifests itself in the ICs for the
replica atoms, the IC tables for replicas grow too big with bogus
elements, thereby making their use to build and manipulate individual
replicas impossible.
Files involved:
gener/replica.src
..............................................................................
(3) CORREL HELIx timeseries returns all zero
- Lennart Nilsson (B950330a.ln)
Add the GECD=0 statement in the 'HELI' section
Files involved:
correl/correl.src
..............................................................................
(4) Parallel Code Fix on DEC ALPHA
- Lennart Nilsson (B950330b.ln)
All routines in socket.c called from fortran have to have names
ending with an underscore. tThe remote shell command is rsh, not
remsh on the dec alphas, this has to be put in correctly in the ARGV0.
Files involved:
machdep/socket.c
machdep/parcomm.src
..............................................................................
(5) READ command fix on Parallel Cluster
- Lennart Nilsson (B950330c.ln)
read ..... name .... the read command fails on parallel cluster
Files involved:
io/mainio.src
..............................................................................
(6) TESTCH.SRC Compilation
- Lennart Nilsson (B950330d.ln)
misc/testch.src doesn't compile properly. Remove unused dummy
array XN(N) from declarations in TESTPAR8.
Files involved:
misc/testch.src
..............................................................................
(7) FLOAT Overflow/Underflow Fix
- Stefan Boresch (B950501a.sb)
Variable DFDIMI potentially uninitialized in subroutine EVDW
(enbond.src), can lead to floating overflows/underflows.
I have noticed the problem so far only when I was running a reduced
version of c24a3 (i.e. very few options in pref.dat) under OS/2. The
problem is close to random, as the same testcase (pert.inp) ran
without problems when the default pref.dat (the one generated by
install.com) was used. However, it's NOT a ##IF preprocessor problem.
If the 4D code is not used (most of the time), there actually is an
operation in subroutine evdw (enbond.src) which involves the variable
DFDIMI, which, under these circumstances, is uninitialized. The
program crashed with a floating point overflow, further inspection
with the debugger revealed that indeed that the 'content' of DFDIMI
was 'not a number'.
As said above, the problem is the uninitialized local variable DFDIMI,
which only is used if energies are calculated in 4D space. (All other
local 4D variables are either initialized or used only in IF (4D)
blocks.
I have fixed evdw, which is the slow energy routine. There might be
similar problems in the fast energy routines. It would be great if
the people who have written the 4D integrator could check whether
there are some more of these 'time-bombs'.
Files involved:
nbonds/enbond.src
..............................................................................
(8) QUANTUM1 Testcase
- Stefan Boresch (B950501b.sb)
The quantum1.inp testcase is weird and may be a problem of our
specific compiler and/or OS version. The symptoms can only be
explained by the BLOCK DATA subprogram qmmmdata (qmdata.src) never
being called. There is a similar problem under OS/2 which I resolved
by changing the BLOCK DATA subroutine into a regular subroutine, but
one should possible first try whether HP has a bugfix for this..
I tried adding HPUX to the list of machines that need an
external qmmmdata statement, and I also tried to use a shorter name
instead of qmmmdata (block data names in strict Fortran 77 must not
be longer than 6 characters), but none of these two ideas worked.
Files involved:
quantum/qmdata.src
quantum/qmset.src
..............................................................................
(9) Using PERT with NOSE integrator
- Stefan Boresch (B950501c.sb)
Using PERT with NOSE integrator causes LINKHP error when exiting
dynamics. The error only occurs after finishing a dynamics run
(everything is calculated correctly); however, as this is a -4
warning, you cannot continue a CHARMM script. Furthermore, the
problem is a nice warning how confused the code in our integrators has
become.
At some point in subroutine dynopt (dcntrl.src) the new (leap-frog)
integrator was set to be the default (even on none NIH machines) if
PERT is used. If you actually run dynamics with the Nose integrator,
memory gets allocated as if the leapfrog algorithm were used although
eventually the Nose integrator (based on the old integrator) is
called. When the memory gets deallocated, LINKHP notices
the mixup and calls WRNDIE.
The quick fix is to comment out the following three lines in
dcntrl.src, subroutine dynopt.
##IF PERT |
IF(QPERT) ORIG = .FALSE. | <- comment these out
##ENDIF |
If that's done, then (on none NIH) machines memory is allocated for
the old integrator, which in this case is still the default. The
mismatch in memory allocation/deallocation described above does not
occur.
Files involved:
dynamc/dcntrl.src
..............................................................................
(10) Misplaced ##IF REPLICA statement
- Stefan Boresch (B950501d.sb)
Module doesn't compile (on machines with Implicit None) if REPLICA
is not in pref.dat file. Essentially, the print statement
IF(QPRINT) WRITE(OUTU,45) NREPL
45 FORMAT(' Entering EPATH: nrepl=',i5)
has to be in the ##IF REPLICA block, otherwise NREPL has not been
declared if the code is compiled without the REPLICA keyword.
Files involved:
energy/epath.src
..............................................................................
(11) DYNA (NOXR NOYR NOZR NOXT NOYT NOZT) Commands
- Stefan Boresch (B950501e.sb)
Provided that I understand correctly what these (undocumented?)
options to the dynamics keyword were originally supposed to do, then
the current code is wrong as it does not do anything.
I need the option so that overall rotation/translation is NOT removed
before starting dynamics, i.e. I want to run (gas phase) simulations
with 3n as opposed to 3n-6 degrees of freedom. It's my understanding
that this is the purpose of the NOXR NOYR NOZR etc keywords (??)
Files involved:
dynamc/dcntrl.src
..............................................................................
(12) Port of CHARMM to OS/2 2.x
- Stefan Boresch (B950501f.sb)
A few modifications were necessary to get c24a3 to run under
OS/2; with these it passes all the testcases with the exception of
icpert (some weird problem with the FMM/FMA code which shows up on
other platforms as well, so I didn't bother...). All other changes
are trivial and fall in two categories:
(1) blank SAVE statements in combinations with COMMON blocks give all
sorts of trouble with the Watcom compiler. Similar problems were
reported previously, and some new additions to c24a3, in particular
the 4D code use such blank SAVEs. Unfortunately, I think the compile
error, meaningless as it is, is supported by the language standard.
So currently there are two options. Ideally, all these blank SAVEs
get replaced by explicit SAVEs with contain only the variables which
really need to be saved. I know, I'm dreaming, so I did what I did
before and mechanically introduced explicit SAVE lists for all local
variables in the respective subroutines in ##IF OS2 brackets. That's
ugly, but it works. (Files affected are dynamc/dynamc4.src,
dynamc/dynio.src and dynamc/prssre.src)
(2) In hbanal.src (manip) there is an instance where the concatenation
operator is used on a dummy variable declared as CHARACTER*(*). The
compiler complains, so I have replaced this WRITE statement with
something that should be equivalent (IS THERE A TESTCASE so that I
could check?). Yet another of Watcom's idiosyncracies...
Files affected:
dynamc/dynamc4.src
dynamc/dynio.src
dynamc/prssre.src
manip/hbanal.src
..............................................................................
(13) PERT with IMAGE
- Stefan Boresch (B950513.sb)
Combination of PERT with IMAGE doesn't work as subroutine
LATTFD(eimage.src) is called without a crystal type defined.
Apparently, there have been some changes made regarding the
calculation of the virial/pressure. However, the changes implicitly
assume that the CRYSTAL facility is used. When IMAGE is used instead,
no crystal type is defined and eventually the calling sequence
epert -> lattfd -> lattrd causes a level -5 error/warning
The same problem must have been present in subroutine energy, as there
already a bugfix from last summer can be found.
Files involved:
pert/epert.src
..............................................................................
(14) Typo in the Code Fixed
- Arnaud Blondel (B950603.ab)
In EPCRAY,
SP=RG*RABR*(AX*HX+AY*HY+AZ*HZ)
should be
SP=-RG*RABR*(AX*RKLX+AY*RKLY+AZ*RKLZ)
and in EICRAY,
SP=RG*RABR*(AX*HX+AY*HY+AZ*HZ)
should be
SP=-RG*RABR*(AX*RKLX+AY*RKLY+AZ*RKLZ)
Files involved:
energy/cenerf.src
..............................................................................
(15) NORESET option for SHAKE
- Lennart Nilsson (B950718.ln)
The NORESET option is added to allow multiple shake commands.
It is useful to be able to define shake on bonds, bonh or so on
several different sets of atoms, with different shake options. The
NORESET keyword to shake command allows this by not zeroing counter.
Files modified:
manip/shake.src
..............................................................................
(16) Trajectory Reading
- Lennart Nilsson (B950718.ln)
READCV is modified to read coordinates at multiples of skip FROM
the actual first coordinate set in a trajectory file.
Files modified:
dynamc/dynio.src
..............................................................................
(17) Non-Cubic Crystal Building Problem Fix
- Wonpil Im (B950730.wi) and Ryszard Czerminski (B950802.rcz)
The crystal build facility uses the symmetrized rotated shape matrix
XTLABC obtained from lattice parameters. However, it does not apply
the same rotation to the unit cell moiety, which may result in bad
contacts in non-cubic crystals. The problem is fixed by calling the
subroutine ROTXTL. Some tests for the rotation are added by Ryszard.
Files modified:
image/crystal.src
New Testcases:
c24test/bcdtest.inp
..............................................................................
(18) MCSS Support
- Erik Evensen (erik@tammy.harvard.edu) (D950807.ee)
The Multiple Copy Simultaneous Search (MCSS) procedure is a method
for efficiently finding favorably binding positions and orientations for
small functional groups in the binding site of a protein with known
three-dimensional structure. The submitted changes are necessary for
CHARMM to be used for the MCSS procedure. Most pertain to redimensioning
arrays used in Leo Caves' replica code. One change works around a bug
associated with replica introduced in nbexcl.src. This fix works for
MCSS but is not general to cases where covalent replica are used.
Files Modified:
fcm/dimens.fcm - Increase MAXSEG to 10000 for LARGE, to 5000
for MEDIUM and SMALL versions
nbonds/nbexcl.src - Comment out the line 66 in the ##IF REPLICA block
as in the following
Cee...26-Jul-95 Erik Evensen, comment this out because it causes
Cee MCSS to blow up - it is not necessary for MCSS since there are
Cee no covalently bound replicas.
Cee IF (qRep) IATMXB = nRepl * IATMXB
------------------------------------------------------------------------------
4.7 Port of parallel CHARMM to Convex Exemplar SPP-1000 and generic MPI
Name : Charles L. Brooks, III and Stephen H. Fleischman
Email Address : brooks@scripps.edu
Institution : The Scripps Research Institute
Date : August 10, 1995
CHARMM Version : c24b0 (c24a3+ developmental)
A port of CHARMM version 24b0 to general MPI based parallelism
using existing parallel code as well as a port to the Convex parallel
machine are included. The original port is the work of Stephen H.
Fleischman from Convex Computer Corp. Charles L. Brooks, III simply
implemented this port in other versions of CHARMM.
Files Modified:
charmm/iniall.src - added call to MPI_FINALIZE
fcm/parallel.src - added NPARPT array for blockcounts needed by MPI
machdep/machutil.src - added CSPP to skull and cross-bones and stack
trace. Added MPI_ABORT call and included mpif.h
machdep/startup.src - added handling of -np from command
line in case of mpi.
machdep/parset.src - added MPI keyword to ##if above nnods
machdep/parallel.src - added MPI code. Affected routines:
Parset, vdgsum, vdgbr, gcomb, psync,
eclock, grec and grecsen.
install.com - added mpi installation to script.
UNX/Makefile_cspp - MPI some Convex library stuff. +FPD option.
New/Modified Documentation:
parallel.doc - added Convex MPI information.
------------------------------------------------------------------------------
4.8 Minimum Image Periodic Boundary Code
Addition of in-line minimum periodic bc for standard CHARMM
dynamics to enhance scaling of parallel jobs
Name : Charles L. Brooks, III, William A. Shirley and
Stephen H. Fleischman
Email Address : brooks@scripps.edu, wshirley@scripps.edu
Institution : The Scripps Research Institute
Date : August 10, 1995
CHARMM Version : c24b0 (c24a3+ developmental)
We have added simple minimum periodic boundary conditions for
cubic, truncated octahedra and rhomboidal (dodecahedra) periodicities
which augments the image facilty and enhances parallel scaling on
scalar parallel machines as well as significantly reducing the memory
requirements. This code is developed and fully tested for the
simulation cells described above when the cell edgelength is the same
in all dimensions. The (trivial) extension to non-identical cell
sides will be added. However, it is critical to see reasonable
performance on all scalar parallel platforms where simulations using
images are currently employed that this enhancement be added now.
Files Added:
images/pbound.src
fcm/pbound.fcm
Files Modified:
charmm/charmm_main.src - PBOUND call
charmm/iniall.src - PBOUND initialization
energy/energy.src - PBOUND control for minimum-image
image/upimag.src - PBOUND control for minimum image
nbonds/nbonds.src - PBOUND control for minimum image
nbonds/nbondg.src - PBOUND control for minimum image
nbonds/enbfast.src - PBOUND control for minimum image
build/UNX/image.mk - added rules for pbound.src
Modified Documentation:
image.doc - added PBOUND documentation
New Testcase:
c24test/pbound1.inp - added simple periodic boundary test file
------------------------------------------------------------------------------
4.9 NBFIX for SOD/MG with atom type ON1C omitted
Name : Alex MacKerell
Email Address : alex@mmiris.ab.umd.edu
Institution : University of Maryland at Baltimore
Date : August 1, 1995
CHARMM version : c22g1 and later
The NBFIX for SOD/MG with atom type ON1C was omitted from the
nucleic acid parameter file (par_all22_na.inp). This omission
incorrectly alters the interactions between the sodium and magnesium
ions and the carbonyl oxygen of the cytosine base.
To correct add the following lines to the NBFIX
section of the nucleic acid parameter file
SOD ON1C -0.1000 2.5570
MG ON1C -40.00 1.8
CHARMM c24b1 (August 15, 1995) to c24b2 (February 15, 1996)
5.1 Port to DEC Alpha parallel platforms
- W .S. Young and C. L. Brooks, III
A port to the new DEC Alpha parallel platforms using PVM was done
by W.S. Young and the Pittsburgh Supercomputing Center in
collaboration with DEC. A new machine type has been introduced.
USAGE: install.com alphamp size P
The addition of the P key queries for where the PVM libraries and deamon
is and then allows all subsequent builds to proceed after making a link
to the appropriate PVM files.
Key for the machine is ALPHAMP and PVMC is used in pref.dat. In
machdep/parset.src, catchout call is removed similar to CSPP and SGIMP.
In machdep/parallel.src, PVMDATAINPLACE calls are removed.
File Added:
build/UNX/Makefile_alphamp
Files Modified:
install.com
source/machdep/parset.src
/parallel.src
/util/cmdpar.src
------------------------------------------------------------------------------
5.2 New substitution variables
- Bernard R. Brooks
Box size and box angles are now available as substitution
parameters when READCV is called. The variables are: @XTLA, @XTLB,
@XTLC, @XTLALPHA, @XTLBETA, @XTLGAMMA,
Files modified:
source/dynamc/dynio.src
------------------------------------------------------------------------------
5.3 PBOUND Enhancements
- Lennart Nilsson
Periodic boxes with unequal length edges are supported.
Files modified:
doc/image.doc
source/image/pbound.src
/nbonds/enbfast.src
/nbondg.src
/fcm/pbound.fcm
------------------------------------------------------------------------------
5.4 Topology/Parameter File Updates
- Alex D. MacKerell
Files Added:
toppar/top_all22_prot_lipid.inp
/par_all22_prot_lipid.inp
/top_all22_sugar.inp
/par_all22_sugar.inp
Files Modified:
toppar/top_all22_model.inp
/top_all22_lipid.inp
/par_all22_lipid.inp
/top_all22_na.inp
/par_all22_na.inp
/top_all22_prot.inp
/par_all22_prot.inp
/top_all22_prot_na.inp
/par_all22_prot_na.inp
/toppar_all.history
doc/parmfile.doc
rtop.doc
------------------------------------------------------------------------------
5.5 Bugfixes
(1) Rename of ALP in the QM code
- Bernard R. Brooks
ALP has been renamed ALFA to avoid a naming conflict with the
variable ALP in param.fcm.
Files modified:
source/fcm/am1parm.fcm
/quantum/qmdata.src
/qmene.src
/qmset.src
..............................................................................
(2) Bug in QM/MM parsing
- Bernard R. Brooks
There was a bug in the use of the subroutine FINDEL for QM/MM
using GAMESS. The incorrect RTF array was used for element
identification that can cause errors.
Files modified:
source/gamint/gamini.src
/quantum/addlnat.src
/qmset.src
..............................................................................
(3) IATMXB for Replica
- Bernard R. Brooks
Fixed a bug in defining IATBMX when replicas are bound to
non-replicated atoms. I found a compromise that hopefully satisfies
all needs. This could be done better by counting the bonds to each
atom when replicas are in use.
Files modified:
nbonds/nbexcl.src
..............................................................................
(4) Graphics parsing improved
- Bernard R. Brooks
##IF IRIS ; ##IFN XDISPLAY is changed to ##IF GLDISPLAY
Files modified:
source/charmm/charmm.src
..............................................................................
(5) Old integrator and Parallel Compatibility
- Bernard R. Brooks
It is now possible to have a single executable support both
parallel methods and the old integrator. Note, the old integrator
will not run in parallel.
Files modified:
source/dynamc/dynamcv.src
..............................................................................
(6) Parallel bug in the ENERGY command when SHAKE is used
- Bernard R. Brooks
The call to VDGBR (global broadcast) in the ENERGY command was
poorly placed and caused errors in the force calculation when SHAKE
was turned on. A simple modification fixes the problem.
Files modified:
source/energy/eutil.src
..............................................................................
(7) Error processing in Hbonds
- Bernard R. Brooks
An error message has been added in the routine HCODES.
Files modified:
source/gener/update.src
..............................................................................
(8) Format error in parallel code
- Bernard R. Brooks
Files modified:
source/machdep/parset.src
..............................................................................
(9) Compatibility between TNPACK and GAMESS
- Bernard R. Brooks
Several routines in TNPACK were the same as those needed to GAMESS.
The duplicate routines are now removed from the TNPACK code is the GAMESS
compile flag is enabled.
Files modified:
source/minmiz/tnpack.src
..............................................................................
(10) Jump into an IF block
- Bernard R. Brooks
The line labeled 180 was placed inside of an IF block in NBONDG.
Some compilers object when code jumps into an IF block.
Files modified:
source/nbonds/nbondg.src
..............................................................................
(11) CALC error
- Bernard R. Brooks
A problem of error handling in calc is fixed.
Files modified:
source/util/calc.src
..............................................................................
(12) Non-bonded exclusions for replicas (B960206.clb)
- Charles L. Brooks, III
Non-bonded exclusions for replicas done incorrectly in cnbnd.src.
Files modified:
source/nbonds/cnbnd.src
..............................................................................
(13) Vectorization on C90 (R960129.clb)
- Charles L. Brooks, III
1. vectorized energy/cenerf.src, new dihedral calculation was not
vectorized.
2. ivdep added to image/upimag.src
3. nbonds/nbonds.src changed to use vector routine if crayvec or parvec
in pref.dat
Files modified:
source/energy/cenerf.src
/image/upimag.src
/nbonds/nbonds.src
..............................................................................
(14) T3D code fixed (R960129.clb)
- Charles L. Brooks, III
1. rewrote psndc, psnd4, psnd8 in machdep/parallel.src to be more
efficient, current routines fail on large machines.
2. rewrote t3d startup stuff in machdep/cstuff.c to use environment
variables for startup configurations.
3. increased HEAPDM in fcm/heap.fcm
4. changed MAXSHK in fcm/dimens.fcm, PARAMETER (MAXSHK = SIZE*3/4)
5. changed util/cmdpar.src to add t3d work around.
6. commented out redundant code in machdep/machutil.src, compiler won't
compile same routine twice!
7. rewrote psync in machdep/parallel.src to use hardware barriers
8. removed call to shell in graphics/apograph.src, not a supported system
call on the t3d.
9. removed synchron from pref.dat.
files modified:
source/machdep/parallel.src
/machutil.src
/cstuff.c
/fcm/heap.fcm
/dimens.fcm
/graphics/apograph.src
..............................................................................
(15) U-B term is added for TSM (R960129.clb)
- Charles L. Brooks, III
Added code to permit Urey-Bradley energy terms to be computed.
Files modified:
source/pert/tsms.src
/tsme.src
/fcm/tsmh.fcm
..............................................................................
(16) Bugfix for MPI version on SPP1000 (R960129.clb)
- Charles L. Brooks, III
Files modified:
source/machdep/parallel.src
..............................................................................
(17) FMTMX increase (R960129.clb)
- Charles L. Brooks, III
changed parameter FMTMX=8 to FMTMX=50 to accommodate a larger
number of translate commands.
Files modified:
source/fcm/univ.fcm
..............................................................................
(18) Error in allocation of tasks to specific cpu's (R960129.clb)
- William A. Shirley and Charles L. Brooks, III
Calculation of the restraint energy terms which use the normal
internal energy functions were being performed incorrectly on parallel
machines because work was being partitioned among processors both in
energy and at the loop level in the internal energy routine itself.
The problem meant that only part of the restraint energy would be
computed. The fix was to remove the processor partitioning of these
energy terms at the level of the subroutine call.
Files modified:
source/energy/energy.src
..............................................................................
(19) Parallel logic error (GRECSEN) (R960129.clb)
- William S. Young and Charles L. Brooks, III
At the top of subroutine GRECSEN the code probes for a message
and receive the message if present. later in the code, the code probes
and loops looking for a message, if the message has already been
received it can not be received a second time causing an infinite loop.
Files modified:
source/machdep/parallel.src
..............................................................................
(20) Non-functional parallel code on clusters with socket option (B950817.ln)
- Lennart Nilsson
The job immediately crashes because the slaves get an "UNRECOGNIZED
COMMAND" error in the charmm command parser. Adding ##ELIF SOCKET
fixes the problem.
Files modified:
source/machdep/parallel.src
..............................................................................
(21) Excessive f.p. underflow messages from DEXP (B950823.rv)
- Rick Venable
The RISM testcase generates numerous error messages about f.p.
underflow errors occurring within DEXP, when compiled with the +T
command line flag for the compiler. The flag suppresses core dumps
and generates a traceback instead, but also turns off the default f.p.
trap routines. Trap routines were added which handle any underflows
in code, but these don't apply to external functions such as DEXP.
Files modified:
source/machdep/startup.src
..............................................................................
(22) SHAKE crashes when PERT is being used with TIP3 (B950914.br)
- Benoit Roux
To incorporate the influence of bond length changes in PERT, the
SHAKE code was modified in c24b1 (relative to c24a3). It turns out
that problems can be encountered, even in cases in which no bond
length is being perturbed. For example, in perturbation of a solute
in the presence of TIP3. The origin of the problem has now been
traced to the subroutine REAL*8 FUNCTION PTARGET(IX,JX,DERI) (in
/source/manip/shake.src).
The reason is that a "Lambda-dependent" shake constraint is setup
even where the value of CONSTR(NCONST) (the bond length squared) is
set to
PTARGET={(LAMDAM*KOLD*ROLD+LAMDA*KNEW*RNEW)/(LAMDAM*KOLD+LAMDA*KNEW)}**2
Normally, if ROLD=RNEW, the expression should give back the good old
CONSTR(NCONST), but there are obvious problems with TIP3 because the
H1-H2 distance has a zero force constant.
Files modified:
source/manip/shake.src
..............................................................................
(23) QNOROT (B951002.clb)
- Zhongxiang Zhou, Charles L. Brooks, III
It seems that the bugfix B950501e.sb done on dcntrl.src is a bug
itself. QNORT = .TRUE. means no translation or rotation. QALWRT = .TRUE.
means the opposite. Hence, QNORT(I) = .TRUE. should leads to
QALWRT(I) = .FALSE.
Files modified:
source/dynamc/dcntrl.src
..............................................................................
(24) Dimension problem for NA-Protein energy calculations
- Jianpeng Ma, Lennart Nilsson
Program crashes or gives erroneous energies, or other errors upon
energy calculations with combined parameter files (protein:dna,
protein:lipid). The problem is due to the dimensional relations
between parameters MAXCUS, MAXCN and MAXCNB, which were not properly
observed. MAXCUS=128 is sufficient and MAXCN=20000 is also sufficient
(has to be .GE. MAXCUS*MAXCUS) but with a recent increase in MAXCUS
from 100 to 128 MAXCNB was no longer sufficient at 20000 MAXCNB=40000
works (should be 2*MAXCN, which is also stated in dimens.fcm).
Minimal checking and a error message are added in MAKITC of nbexcl.src.
Files modified:
source/fcm/dimens.fcm
/nbonds/nbexcl.src
..............................................................................
(25) CUTA option of HBALAL
- Lennart Nilsson
The angle cutoff option in the coor hbonds module is fixed.
Files modified:
source/manip/hbanal.src
..............................................................................
(26) Fixes for OS2 port
- Stefan Boresch
Problems with REWIND and DCMPLX on PCs are corrected.
Files modified:
source/dynamc/dynio.src
/machdep/machio.src
/minmiz/tndriv.src
tnpack.src
/misc/nmr.src
ssbp.src
..............................................................................
(27) Divide by zero error
- Jay L. Banks
Floating point exception (divide by zero) occurs if no atoms
selected. Add zero-checking of AMASST in the STAT command section.
Files modified:
source/manip/corman.src
..............................................................................
(28) Array name correction
- Ryszard Czerminski
iRepID = repNoA(iRepNo) is misspelled, should be repID(iRepNo).
Files modified:
source/nbonds/nbndf2p.src
CHARMM c24b2 (February 15, 1996) to c24g1 (August 15, 1996)
6.1 Parallel Code
- Charles L. Brooks, III
Fixed problems in calling sequence/logic for mpi direct gather and
scatter calls in generic mpi (non-Convex) calls. Introduced new preflx
keyword MPIDIRECT to identify use of direct (mpich supplied) global
operators versus use of CHARMM supplied global operators w/ MPI send/receive.
Files modified:
source/machdep/parallel.src
------------------------------------------------------------------------------
6.2 CRAYVEC code
- Charles L. Brooks, III
Cray fixes to clean-up CRAYVEC/UNICOS ifdefs. MAXCPU is increased
to 16 from .8
Files modified:
source/charmm/iniall.src
manip/shake.src
fcm/machdep.fcm
------------------------------------------------------------------------------
6.3 Bugfixes
(1) Calling Argument
- Matthias Buck, B960710.mb
The array FREEAT is used for reading trajectory files which contain
fixed coordinates (lines 1992-1996: RDHEAD) later handeled by
RDNEXCO (lines 1900-1912). However, program crashed at line 1903 because
FREEAT had not been transmitted to the subroutine RDNEXCO.
Files modified:
source/correl/solana.src
..............................................................................
(2) Parallel I/O
- Lennart Nilsson
Fix problem in writing trajectories on parallel machines.
Files modified:
source/dynamc/dynio.src
..............................................................................
(3) IBMSP Timing
- Lennart Nilsson
RTC is used for IBMSP timing
Files modified:
source/machdep/parallel.src
..............................................................................
(4) Hole Searching
- Boryeu Mao, B960606.bm
Hole searching in the -Z direction corrected.
Files modified:
source//manip/corman2.src
..............................................................................
(5) GNU Fix
- Milan Hodoscek
Use DCMPLX for GNU.
Files modified:
source/misc/nmr.src
/ssbp.src
..............................................................................
(6) Analys Command
- Bernard R. Brooks
SQRT(R2) is fixed to ONE/SQRT(R2) in output.
Files modified:
source/nbonds/enbond.src
..............................................................................
(7) Code Encapsulation
- Charles L Brooks, III
##IF FMA added
Files modified:
source/nbonds/nbutil.src
..............................................................................
(8) EXSG Option
- Ryszard Czerminski
Enable the EXSG option with GROUP
Files modified:
source/nbonds/nbondg.src
