Known Errors in all Releases of TNT

Table of Contents

• Release 1
• Release 2
• Release 2A
• Release 2B
• Release 3
• Release 3A
• Release 3B
• Release 3C
• Release 3D
• Release 3E
• Release 4
• Release 4A
• Release 4B
• Release 5A
• Release 5B
• Release 5C
• Release 5D
• Release 5E
• Release 5Ea
• Release 5Eb
• Release 5F

Release 1

• A problem is known to exist in ADERIV when calculating B gradients for heavy atoms (ZN, etc.) with low B values. In extreme cases the gradient can be of the wrong sign and the B value will attempt to drop to negative infinity. SHIFT can prevent the B's from dropping below 1.0. There are two possible patches to avoid this problem until I figure out how to fix it properly. The first is to prevent such atoms from dropping below about 5.0 or so where the problem first becomes bad. This can be done with a RANGE card in SHIFT (which will affect all atoms) or the equivalent effect can be achieved with a special purpose program to affect only the required atoms. The second solution is to use the SHIFTS card in ADERIV and not the GRADIENT card. This does not correct the problem but prevents the oscillations that cause large atoms to temporarily drop to low B values.
• If the map extent is longer than the unit cell edge in any direction the program will always exit with an error message saying that not enough map has been given. This can only be averted by a program modification in the routine GENDEN.

• The atomic coordinate shifts are calculated wrong when rigid body (COMBINE XYZ) refinement is being done. Release 01 works but does not converge to the answer as rapidly as it could.

Release 2

• The problem mentioned for Release 1 persists. It seems to be caused by the failure of the Agarwal method to consider series termination in it's calculation and affects the B gradients more than the other types of gradients because the modified electron density for the B gradients is more strongly affected by the high resolution terms that should not be there. A work-around has been created by writing a program that uses the original Aragwal method which is not as seriously affected by the problem. This program only calculates B gradients and is called BDERIV. It requires a modified difference map, calculated using difference coefficients generated by RFACTOR using that program's BGRAD option. The command files in EXAMPLES: demonstrate this procedure.

• This program will not handle REMARK cards within a coordinate file properly. The program closes the section within the map file that contains the remarks before reading the coordinates. When it reaches the comments it places them in the map file and this corrupts the file. To avoid the problem don't put comments in coordinate files. To fix the problem remove the line in MODEL that writes the coordinate file remark into the map file.

• This program will bomb when either Fo or Fc is zero and that reflection is to be used. To avoid the problem set a reflection rejection limit for Fo and Fc that is at least very small. This limit will cause the zero reflections to be ignored, which is a better behavior anyway.

• An error occurs when a perfect plane is discovered whose normal does not lie parallel to a cell axis. If this problem arrises move one of the atoms slightly out of the plane to make that plane imperfect.
• When doing a REPORT GEOMETRY with a multichain structure the values for the number of occurrences of each residue, or linkage type is incorrect. This has been fixed for the next release.

• One garbage structure factor is produced for each scratch file in the initial sort. This introduced a variable error around 0.1% in the maps the program produced. There is no known work-around except to fix the program.

Release 2A

• The average weights applied to Fo and Fc that is published in the log file is calculated incorrectly. The weights themselves are correct only the averages are in error. This problem has been fixed for the next release.

• This program does not work. It will work in the next release.

• This program does not work. It handles Friedel's law incorrectly and required extensive modification to the routine DEAL SF to correct. This problem has been fixed for the next release.

• The statistics printed into the log file describing the map are incorrect. The map produced is correct. This problem arrises only for VMS 4.0. The FORTRAN complier constructs incorrect code when the optimizer is on. To correct the problem compile the library TRICLIB with the FORTRAN flag /NOOPTIMIZE.

• The PUNCH LABELS command will enter an infinite loop. The variable "j" must be changed to "rss1" in the LABELS code of PUNCRD.
• Under VMS 4.0 this program will infinite loop when calculating gradients unless compiled with /NOOPTOMIZE.
• In the calculation of gradients of planarity a term was left out. This omission causes the gradients to be inaccurate but not totally wrong. To correct multiply each gradient by the variable PART1 as it is calculated in routine DOPLAN.
• The program will occasionally fail when calculating gradients for a perfect plane when the coordinates for the atoms in that plane have not been entered on ATOMC cards. To avoid the problem reformat the coordinates as ATOMC cards.

• Under VMS 4.0 this program will access violate if it is not compiled /NOOPTIMIZE.

• These programs will only function correctly when the map contains the same number of grid points in both the X and Y direction. The error is in the code of routine READ MP T of TRICLIB.FLX. This routine checks the extent of the map incorrectly.

Release 2B

• All large programs written in Flecs must be compiled /NOOPTIMIZE when using the VAX FORTRAN compilers of version 4.0 and 4.1.

• The curvature calculation routine is incorrect and will produce only zeros for the curvature for all parameters. The variables A, B, C must be replaced by CELL(1), CELL(2), CELL(3) in the routine GENHX.
• The B gradient calculation was missing several terms that are nonzero only when the coordinate system has non-90 degree angles. This oversight has been corrected.
• If a portion of the map along a cell axis is all that is required and all of the map along that direction is given the program will bomb with an inappropriate error message.

• An access violation will occur when a FIX card is given a residue range. The problem lies in the FIXCRD routine. Replace the line "J = RSGRSS(RSG)" with "J = RSGRSS(RSSCHS(J), RSG)".

• RFACTOR will not reject reflections with no phase when calculating difference coefficients. These reflections will be placed in the difference coef- ficient file with a phase of 1000.

Release 3

• When an atom has a bad contact with more than one symmetry image of another atom, and that contact is EXCLUDEd, only one (or maybe two) of the contacts will be ignored. The other contacts will be restrained regardless.
• When calculating only the function value for the present model with NEWGEO the CONTACT term is always zero. As a work-around the GRADIENT card should be used even in short loops. The error occurs in the routine CONCRD where a variable is called both FNCVAL and FUNCVAL. This problem is corrected by changing all references in this routine to either one name or the other. I prefer FUNCVAL.
• If a collection of atoms are being restrained to a plane whose normal is perpendicular to the Y axis, the gradients will be incorrectly calculated and will be too large by many orders of magnitude. A fix has been incorporated into the new release.
• A message will not be printed if an atom has a bad contact with a symmetry related image of itself and it has bonds to something. The contact is properly ignored. If an atom has a bad contact with the symmetry image of an atom to which it is bonded that contact is improperly ignored and no message is printed. This problem has been fixed in release 3A.

• If too few sections are given to these programs an incorrect error message is produced. This problem has been fixed.

Release 3A

• An error in the calculation of the gradient of the planarity function was discovered. This error affects the rate of convergence. It has been corrected.

Release 3B

• The RENAME command will not operate when the "chain with no name" is renamed and then the newly named chain is renamed again. The first renaming does work. The problem has been corrected.

Release 3C

• The handling of linkage definitions on CHAIN cards is incorrect. This error causes a failure in all cases. It can be corrected by editing the file ATOMS:ATOMS.FLX and changing "g1" to "g" in the line: chltyp(i) = g1
• Also "c1nam" must be changed to "cs1nam" in line:


r1 = ABS(GETRSS(cs1nam, r1nam)) ! We don't care if the node is new

Both of these lines are the the routine CHNCRD. When you have made the change all programs which read and use atoms must be recompiled.

• The SCREEN output of NEWGEO occasionally prints values for torsion angles which are outside the range of the torsion angle function (-180<theta<=180). This problem does not affect refinement, it only affects the SCREEN listing.
• The derivatives for a torsion angle with a negative value are produced with the wrong sign. To correct the problem replace DegPerRad with SGN in the line from DOTORS (GEOMETRY.FLX):


k1 = -r1*q1*SQRT(1.0-cth*cth)/DegPerRad

Release 3D

• Contrary to the documentation, this program will not work for space groups: P4122, P41212, P4322, and P43212. Also any space group which reduces to one of these space groups will also fail.

Release 3E

• A TRANSLATION card must be preceeded by a MATRIX card for the transformation to be valid. This restriction has been removed.

• An error in the main program caused all keywords longer than 4 letters to be unrecognizable. This problem can be corrected by replacing CHARACTER*4 with CHARACTER*8.
• This program does not use the proper convention for Psi angles. It used the carbonyl oxygen of residue i instead of the nitrogen of residue i+1. The program has been corrected.

• Occasionally when given a low symmetry space group based on P21 or P21A GENERALSF will decide to sample the model map along the Z direction a number of times which is not divisible by 4. This causes the program to crash.
• The data set written for space groups P312, P3112, and P3212 is incomplete. It is missing the data where h=k and l<0.

• Occasionally EXPANDER will lose all coefficients with H equal to HMAX. When this occurs it will also fail to delete the last of it's scratch files.

• There is an error in the routine NEWSTP which degrades convergence of the short loops when the program encounters the "inverted parabola" condition. The statement:


SELECT (BEST)

(FUNCZER) SOLD = 0.0

SOLD = S

FIN

should be replaced by:

SELECT (BEST)

(FUNCZER) SOLD = 0.0

SOLD = S

FIN

• This program will generate incorrect results for any space group which contains a direct 3 or 4 fold and has perpen- dicular two folds. This class of space groups includes P321, P422, P622, and P6322. It also includes any space group which contains one of these space groups as a subgroup.
• This program failed to properly expand the coefficients with h=k in space groups P312, P3112, and P3212, causing these data to be corrupted.

Release 4

• When the option INTENSITY is set, the reflection rejection counts are still printed in resolution shells.

• Because this program does not have access to the sigmas for the F's the curvature calculation cannot be calculated correctly when the data are to be weighted by sigmas. When weighting by sigmas you either must not use the gradient/curvature method, or accept the extra short loops is will take for the cycle to converge.

• The limit on the number of atomic matches between the target and the overlay object is not checked. If too many matches are found the data structure will be corrupted and the results unpredictable.
• If more than 30 atoms are listed on SELECT cards an array will overflow and data will be corrupted. No error messages will be produced.

• If a residue range is being traced and a residue linked to itself is encountered, OVERLAY will continue adding atom matches until something bad happens (see above) and SHIFT will go into an infinite loop.

• If the space group is P1 and the coefficients given to one of these programs contain data to a higher resolution limit than the limit given to the program, the program will bomb with a 'REFLECTION OUT OF ORDER' error. The work-around is to always ensure that the data do not extend beyond the resolution limits given to the program.

• Non-bonded contacts are incorrectly handled. Only one atom of each bad contact pair is actually moved because of the contact.

• The program will bomb with an "integer overflow" error whenever a space group with a 2 or 21 axis is along k and there are reflections with negative l.

Release 4A

• The structure factors are not properly multiplied by the number of centering operators in this space group. If the map was calculated by ignoring the centering operators the structure factors will be too small by this factor.

• When there is a two-fold along the Z axis the program internally generates an extra row of map points for each section. This problem has no effect unless the map sections almost exactly fill the transform buffer (X). In that case the program will bomb with an "access violation" message.
• The program would produce incorrect results for any space group which contained a translation in x or y not 0 or 1/2. This only occurs for space groups in their standard settings for P4132 and P4332.

• If NX is much larger than NY an inappropriate "symmetry expansion" error could be generated. This problem is caused by a typo in the code for checking the symmetry expanded coefficients against the array bounds of the transform buffer.

• This library will not work correctly on systems where the byte order is swapped relative to a VAX, or when the varables of a subroutine are not saved between calls. The type of the symbols defined in CNSYMBOLS.INC must be changed to BYTE and the varable START in NXTCON must be SAVEd to correct the problems.

• The common block /SpaceGroup/ was not SAVEd in the main program of either of these programs. This would cause serious errors on non-VAX computers.

• When one of these programs is given the symmetry operator for P622 in the order given in the International tables it will incorrectly analyse the space group and generate completely incorrect results. Even the asymmetric unit of data will be incorrect. The workaround is to use the order:


x, y, z; -x, -y, z; x-y, x, z; -y, x-y, z; y-x, -x, z; y, y-x, z;

x-y, -y, -z; y-x, y, -z; -y, -x, -z; -x, y-x, -z; y, x, -z; x, x-y, -z.

• The temperature factor components of the form factor approximation can not be negative. The parameters listed for U+4 contain a negative number. If the sum of your blur factor and the B of the uranium atom is less than this number any program using form factors will bomb. By the way, the form factor parameters for U+4 were also incorrect and should never have contained a negative number to begin with.

• The scale factors are not applied to the first solvent structure factor in the file. SHIFT The logical function NXTATM was not declared as such in the routine FIXCRD. This error did not affect the routine under VMS but did cause serious problems under UNIX.

• If the coefficient file illegally contains reflections with negative h an "Internal Error" will be flagged, not a sorting error.

• An "Access violation" error will occur whenever a potential bad contant is found but no standard minimum distance has been given for those two atom types. To avoid the problem include the required CONNECT cards and prevent the TNT error message from being printed.

Release 4B

• These programs print out an unnessasary line of diagonstic information. The line describes the progress of the program during the final transform. This data is of no interest to the general user.

• When either Alpha or Beta are not 90 degrees the PDB file produced by the PUNCH BROOKHAVEN command or read by the BROOKHAVEN command will be incorrect. The Z value will be corrupted. In addition PUNCHing out a file and reading it back in are not inverse operations.

• The parameter DEGRAD is incorrectly defined as type integer. It should be real. This error causes the phases of the HKL file produced to be slightly incorrect.

• The space group symmetry is incorrectly handled in this module. It does not work for any space group execpt P1, where it is useless anyway. If you need to perform rotation function refinement contact me for a prerelease of the next version of TNT.

• The values for the minimum and maximum density values of the map sections, which is written to the log file, are often in error. The map sections themselves are correct.

• When a curvature method is used for refinement and COMBINE cards are issued the resulting shifts are too small in magnitude and many short loops are required to find the best stepsize. The direction of shift is also affected.

Release 4C

• The form factor approximations for CA++ and MN++ are incorrect. The order of the entries in FORMB defined in FORMFACTOR.FOR need to be switched.

• If the peptide chain cannot be traced from the source to the destination of either the target or 'overlay' chain the error message will not write out the name of the residue at the site of the problem. Only blanks will be written.

• A COMBINE card with one of its atom codes specifying a single atom will be misinterpreted as implying that all atoms in that residue are to be combined. The word WILD must be replaced with RS2N in the call to NXTATM in the routine COMCRD.

• These programs will produce incorrect results when presented with a space group which contains a center of symmetry or a mirror.

• Incorrect results will occur when the symmetry operators of the space group are INCLUDEd twice (or more times). The identity operator is ignored the second time but the others are not.

Release 5A

• The SCALE and BLUR options on FILE cards are nonfunctional.

• The curvature for an atom on a special position is incorrectly calculated. The shift for a parameter unaffected by the axis of symmetry is overshifted by the multiplicity of the axis while a parameter affected by the axis is undershifted. The code has been corrected but because of the lack of consideration of the off-diagonal elements these atoms remain illbehaved in refinement.

Release 5B

• A fatal error (Access violation) occurs when a PDB or RDI file is being read and one atom in a residue has a different residue type than other's in that same group. The error occurs while the warning message is being generated.

• In space groups with a C3i point group symmetry (P3, P31, and P32 most notably) the (100) and (010) reflections are lost when calculating maps from fourier coefficients.

• All attempts to add links on CHAIN cards will fail. One must add the definition or CHNCRD


Character*4 atmnam, grpnam

Release 5C

• The General library initialization routine is not called. The only consiquence of this is that the number of current options is not set to zero. The effect of this is unclear. To correct the problem edit the file TNT_IO:io.flx and add the line


call Init General

to the routine Init IO and recompile everything.

• DSN6 files cannot be produced for cubic space groups. To correct the problem edit the file TNT_TRANSFORM:transform.flx and change the line


IF (cysgrp .eq. cNOVALUE) call classify ops

to

call classify ops

• DSN6 files cannot be produced on most Unix systems. A correction for this problem will be included in 5D.
• When a map is calculated from coefficient in a space group with octohedral point group symmetry (such as P432) the hhl reflections are lost. This error affects all map calculations and gradient calculations in the RFACTOR module.
• When calculating structure factors for a space group with a 21 axis along Z with either large cell constants or a small transform buffer size incorrect results can be produced. All reflections with h greater than some value will be wrong while all those <= that value will be correct. The solution is to incease the buffer size in TNT_TRANSFORM:FFT_BUFFER.INC.

• All page references in the index are too large by 4 pages. To find the topic of interest look up the keyword in the index and subtract 4.

• One entry in the geometry definition for BME is called BMD by mistake. Change the BMD to BME to correct the problem.

• When reading DSN2 format coordinates a warning message is generated if the first letter of the atom name does not match the atom type. In the message the name of the atom is omitted.

• The GRADIENT command produces a file which causes the B correlation restraints to be ignored.

Release 5D

• When calculating a map from coefficients it is possible that the input routine could attempt to read the coef- ficents twice. This causes the routine to fail with an error message indicating that the coefficients are not sorted properly. Since this is a low probability error the simplest correction is to increase the size of the parameter SIZE in FFT_BUFFER.INC and recompile with TNT_COMMAND:RECOMPILE.

• When using the CORRECT option under Unix Fourier will fail. To correct this error you should edit the file tnt/util/fourier.flx. Find every instance of the line


Save rsqrsq, rsqloh, rsqhih, rsq mz, rsqnhk, rsqlol

and add this line right after it.

Save buf1, buf2

There will be two places where this has to be done. Once in the routine NORMALIZE SF and again in SORT SF.

You will then have to compile Fourier. cd to $tntbin and type "make fourier".

• Convert will write several junk records when PUNCHing a BROOKHAVEN file on a MIPS processor system. This is caused by an error in the MIPS compiler. The workaround is to replace the line in tnt/util/convert.flx subroutine PUNPDB


DO (chs = 1, chscnt)

with

chs = 0

WHILE (chs .lt. chscnt)

chs = chs + 1

This problem will occur on older DECstations, SGI's, and ESV's.

• Convert will bomb when reading a PDB file with the mes- sage "Too many generic chains". To correct the problem find the lines in subroutine SETATM in tnt/util/convert.flx

Data rsgusd / rsgmax*.FALSE. /

Data rsgflg / rsgmax*.FALSE. /

and insert the line

Save cnamold, rnamold, rsgusd, rsgflg

directly afterwards. Then recompile Convert by cd'ing to $tntbin and typing 'make convert'.

This problem will occur only of SGI machines with IRIX 5.<something>.

• On IRIX 4.0.5 on SGI computers the CORRECT operator will fail (sometimes) in that it will only return a fraction of the data on the input file. To correct the problem edit the file tnt/util/fourier.flx and find the routine SYM EXPAND. After the line


Save rsqsig, rsqfom, rsqf , rsqrsq, rsqhkl, rsqop

insert the line

Save GotOne

• If one has an overall temperature factor correction (B) which is negative curvatures will be calculated incorrectly. Since B should always be zero in refinement this should not be a big problem.


To correct the problem replace the lines

IF (Expon1 .ne. 0.0)

I3 = MIN(1 + NINT(100.0*S2*Expon1), GauSiz)

Solvent = Solvent*GAUSS(I3)

FIN

with

IF (Expon1 .ne. 0.0) Solvent = Solvent*EXP(-Expon1*S2)

The change will have to be made in two places.

Release 5E

• Structure factors which have their phase set to 1000 will not be recognized as not having a valid phase. Such phases should also have their FOM set to 0.0

• Fourier maps will be incorrectly calculated in space groups I41 and I4132.

• This command will not work unless a TNT control file exists in the current directory. The file does not need to contain anything, it simply must exist.

• The "install" identifier will not work with the GNU make program used at some sites. The identifier refroot must be defined twice with GNU make.

Release 5Ea

• NCS does not work on Unix systems. To correct the problem edit the file tnt/ncs/ncs.flx and add the variable "currtyp" to the line


Save currng, curjflg, currnam, curanam

Release 5Eb

• Convert will not properly read PDB format files which contain disordered atoms whose atom names are one letter long. The name written to the ATOMC file will contain a blank, which is illegal.  To correct the problem replace the lines in tnt/util/convert.flx


C     TNT does not allow embedded blanks in names but PDB does.  Here
C      we remove any such blanks.  Any gap must occur in the first
C      four letters because of the special nature to the last two.
C      This operation also left justifies the name.

      DO (i=1,4)
      IF (anam(i:i) .eq. ' ')
      anam(i:ATMLEN-1) = anam(i+1:ATMLEN)
      anam(ATMLEN:ATMLEN) = ' '
      FIN
      FIN

                with...
 

C     TNT does not allow embedded blanks in names but PDB does.  Here
C      we remove any such blanks.  Any gap must occur in the first
C      four letters because of the special nature to the last two.
C      This operation also left justifies the name.

      i = 1
      WHILE (i.le.4)
      IF (anam(i:i) .eq. ' ')
      i2 = i
      WHILE (i2.le.4 .and. anam(i2:i2).eq.' ') i2 = i2 + 1
      anam(i:ATMLEN-(i2-i)) = anam(i2:ATMLEN)
      anam(ATMLEN-(i2-i)+1:ATMLEN) = '     '
      FIN
      i = i + 1
      FIN

                Then cd to $tntbin and "make convert".

• Geometry will not recognize an atom with a disorder indicator which is a number. For example, CB.1 will not be recognized as a CB atom while CB.A will.
• Geometry will not issue a warning when the center of mass of the atoms in the model is more than one unit cell from the origin. A warning is needed because the symmetry related bad contact finder will sometimes fail to find all such contacts when the molecule is far from the origin.

• When you have a reflection whose resolution is EXACTLY equal to the outer resolution limit the HISTOGRAM command will fail with an obscure error message.
• If you are using REPORT RFACTOR with option INTENSITY set and you have a reflection whose amplitude is less than 10 the statistics reported will be incorrect.  The overall R factor and highest amplitude bin will be affected the worst.

• RMSTEP B and RMSTEP OCC are confused. If you want to specify the rms step for B you should use the RMSTEP OCC statement.

• The "from_pdb" command will not correctly identify the atom types of two specific kinds of atoms. One is the AO5* atom of Coenzyme A and the other is the AP atom of NAD. In both cases, the affected atom will end up as atom type A, which is illegal.

• This refinement script confuses the conjguate gradient (cg) and gradient/curvature (gc) methods of minimization. To use one method specify the other on the "tnt" command. Most of the time one should use the conjugate direction (cd) method, which is the default and works properly.

• These commands generate sequence files for DNA which end with a d5'END linkage instead of a d3'END. You should edit your DNA sequence files to correct the problem. An example sequence file for DNA is given in tnt/doc/examples/bdna/bdna.seq.

• The file p32.dat incorrectly contains the equivalent positions for space group p31.dat. The correct contents are

NOTE  P32    TRIGONAL PAT SYM: P3B   NEQ=3 S.G.NO:145
OPERATOR X,Y,Z
OPERATOR -Y,X-Y,Z+2/3
OPERATOR -X+Y,-X,Z+1/3

• The definition of NDP has one incorrect line and is missing another.  The line

         should be changed to

        GEOMETRY  NDP   ANGLE   109.5   3       AOP1*, AP2*, AOP2

         and the line

        GEOMETRY  NDP   ANGLE   109.5   3       AOP3*, AP2*, AOP2

         should be added.

Release 5F

• Contrary to the statement in tnt/doc/release_5f.txt, this script does not use the NO_HYD option to discard hydrogen atoms in the PDB file. You will have to set this option in your control file, which means you have to have a control file. Oh well... Convert the coordinate file from pdb, edit the prototype tnt control file to fill in the blanks and set the option, and then convert the pdb file again. The second time the hydrogen atoms will go away.

• On a CLUSTER statement the name of the cluster must be unique when considering only the first three letters. The intention was to require uniqueness over the first four letters.

• When one of these programs is requested to produce a DSN6 format map file and a particular grid and layout is specificed by the user (as opposed to asking for an asymmetric unit or molecular volume) the program will err and not produce the correct layout.

  To correct this problem go to the file tnt/library/transform/transform.flx. Find the routine Copy DSN6 (they are alphabetical). You will find a snippet of code

           ('FIXED ')
           DO (i=1,6) lay(i) = datlay(i,dat)
           FIN
  

  Replace it with

           ('FIXED ')
           DO (i=1,6) lay(i) = rsqlay(i,rsq)
           FIN
  

  Now cd to $tntbin and type "make"