I’m using Q-Chem 6.3 on RHEL
For this input:
$comment
Geometry optimization for CH3Cl + OH- reactant complex C-O bond at 4-5 A
$end
$molecule
-1 1
C 0.000000 0.000000 0.000000
H 0.000000 0.000000 1.090000
H 1.026719 0.000000 -0.363333
H -1.026719 0.000000 -0.363333
Cl 0.000000 1.800000 0.000000
O 0.000000 -4.500000 0.000000
H 0.000000 -5.300000 0.000000
$end
$rem
JOBTYPE opt
SYMMETRY false
SYM_IGNORE true
METHOD b3lyp
BASIS 6-31G
$end
I get this error in the output file:
** Ill-Behaving Topology Definitions (No action needed.) **
Ill-Defined Angles Removed:
5 1 6, 1 6 7, 2 6 7, 3 6 7, 4 6 7
Ill-Defined Torsions Removed:
1 2 6 7, 1 3 6 7, 1 4 6 7, 2 1 6 7, 3 1 6 7
4 1 6 7, 5 1 6 2, 5 1 6 3, 5 1 6 4, 5 1 6 7
Done Checking Topology
in topocoord::build_bmatrix returned error
And, at the command line I get:
terminate called after throwing an instance of 'int'
/work/QChem6p3/bin/qchem: line 129: 2287830 Aborted (core dumped) ${QCPROG_S} ${inp} ${scr}
Error in Q-Chem run part 1
Error in the Q-Chem run
Any suggestions how to avoid this problem?
I think that the highly stretched bond in conjunction with the high symmetry / linearity may have created something of a corner case for the internal coordinate transformation (Wilson B-matrix). I can reproduce this with Q-Chem v. 6.3 and will submit a bug ticket. If you switch to the old optimizer,
GEOM_OPT_DRIVER = OPTIMIZE
then the job fails in a “normal” way, i.e., SCF iterations don’t converge because the highly stretched bond makes this a challenging SCF problem and you will need to deal with that by varying the SCF algorithm, etc.
Thanks John,
Isn’t this a significant problem when we are looking for transition states in reactions that have bond breakage? Is there a way to cope with that, since most reactions we’re trying to characterize do have stretched bonds associated with going from reactants to products?
That’s true but this is a pathological case (4.5 A bond length) that’s much more stretched than in a typical transition state.
The problem of SCF convergence in stretched bonds is a rather general one because in such cases there is likely (almost by definition) more than one electronic state, and thus more than one SCF solution, in close energetic proximity. There are ways to cope by changing the SCF algorithm, changing the guess, maybe using MOM starting from a less highly stretched bond, increasing fraction of exact exchange (to open up the HOMO/LUMO gap) and then using those MOs as a guess for the target functional, tightening grids and thresholds, etc. You may also need to break spin symmetry. It’s not black box. That’s what makes it fun. 
But again, I don’t think this example is representative of most TS searches; rather, this particular geometry is probably well into the product channel.
Here is an input file that will converge at least the first SCF (albeit in ~150 iterations). Not sure all these options are necessary, you would probably want to test variations.
$molecule
-1 1
C 0.000000 0.000000 0.000000
H 0.000000 0.000000 1.090000
H 1.026719 0.000000 -0.363333
H -1.026719 0.000000 -0.363333
Cl 0.000000 1.800000 0.000000
O 0.000000 -4.500000 0.000000
H 0.000000 -5.300000 0.000000
$end
$rem
!JOBTYPE opt
SYMMETRY false
SYM_IGNORE true
METHOD b3lyp
BASIS 6-31G
geom_opt_driver optimize
thresh 14
s2thresh 16
max_scf_cycles 200
scf_convergence 8
xc_grid 3
unrestricted true ! UKS
scf_guess_mix 3 ! break spin symmetry of the guess
$end
I also tried SCF_ALGORITHM = ROBUST (which is new in Q-Chem 6.3), and that approaches manages to converge this calculation in far fewer cycles without any of the other modifications (i.e., just adding that one keyword to your $rem). That’s a new algorithm that’s designed to try some of these variations for you. That said, I think this is a case where you should be breaking spin symmetry so that alpha and beta electrons of the broken bond can localize on different fragments, whereas for a nominally closed-shell system like this, a spin-restricted solution leads to a very high energy stretched sigma-bonding MO.
1 Like