Hi, in v5.4 during simple freq calcs in vacuum and pcm, TDDFT is being called to compute analytical hessian. Ive also noticed xc_grid tag triggers the call but if not used, the calcs run into convergence issues. Ive tested freq calc for vac, pcm, smd and cosmo ,and have noticed TDDFT trigger only for vac and pcm. Moreover, for GGA (BP86) basis, im running into memory issues regardless of how large MEM_TOTAL is set to because TDDFT is called. For hybrid (-xd) I dont run into memory issue even though here also TDDFT is called for vac and pcm. This seems to be a bug?
Here is a sample input script that triggers TDDFT for me with a call of “Full subspace dimension requested for CIS/TDDFT”
$molecule
0 1
H 0 0 0
H 0 0 1
$end
$rem
job_type = freq
basis = def2-svpd
max_scf_cycles = 100
xc_grid = 3
gen_scfman = true
method = wb97x-d
geom_opt_max_cycles = 200
$end
Hi Hetal,
I just ran your input with a pre-release version of Q-Chem 6.1 (latest trunk), and I do not see the behavior you are referring to. Can you copy specific section the output? It’s possible that this behavior was fixed, but it seems weird to me that I have never seen it before because your input file is not very exotic.
Hi, thanks for your reply. This seems to be happening for v5.4, at the moment I am restricted to using this version until the HPC upgrades to qchem 6. For that exact input file, here is the section of the output with TDDFT call.
Cartesian Multipole Moments
Charge (ESU x 10^10)
0.0000
Dipole Moment (Debye)
X 0.0000 Y 0.0000 Z -0.0000
Tot 0.0000
Quadrupole Moments (Debye-Ang)
XX -2.3457 XY 0.0000 YY -2.3457
XZ 0.0000 YZ -0.0000 ZZ -1.4380
Octopole Moments (Debye-Ang^2)
XXX 0.0000 XXY 0.0000 XYY 0.0000
YYY 0.0000 XXZ -0.0000 XYZ 0.0000
YYZ -0.0000 XZZ 0.0000 YZZ 0.0000
ZZZ -0.0000
Hexadecapole Moments (Debye-Ang^3)
XXXX -2.5090 XXXY 0.0000 XXYY -0.8363
XYYY 0.0000 YYYY -2.5090 XXXZ 0.0000
XXYZ -0.0000 XYYZ 0.0000 YYYZ -0.0000
XXZZ -1.1667 XYZZ 0.0000 YYZZ -1.1667
XZZZ 0.0000 YZZZ -0.0000 ZZZZ -4.1977
Calculating MO derivatives via CPSCF
Full subspace dimension requested for CIS/TDDFT
MAX_CIS_SUBSPACE fits in available memory
1 0 9 0.1022744
2 6 3 0.0089006
3 8 1 0.0003430
4 9 0 0.0000000 Converged
Time for AOints: 0.0 s (CPU) 0.0 s (wall)
For me, this problem remain unnoticed until i was computing larger molecules for which the MAX_CIS_SUBSPACE did not fit in the available memory.
Thanks,
Hetal
Current structure of the source code definitely bypasses that for frequency jobs. A look at the comments and commits suggests I may have fixed this in Sept. 2021, which would have showed up in 5.4.2 bug-fix release (however, I’m not 100% certain… I fix a lot of bugs). Because I don’t routinely keep older versions of Q-Chem compiled, I suggest that you send email to Q-Chem support, perhaps there is a workaround.
By the way, the reason that you perhaps don’t seem this for COSMO or SMD is that I don’t think analytic frequencies (Hessians) are available for those methods. Assuming I’m correct, there should be a warning near the top of the Q-Chem output file about finite-difference derivatives. We do have analytic Hessians for gas-phase (TD)DFT and for (TD)DFT with C-PCM, which is the default PCM.
Hi, you are correct, that is definitely what i have noticed as well. It tries calling TDDFT whenever analytical Hessians are possible which happens to be for vacuum and PCM. I did run it on both 5.4 and 5.4.2, and 5.4.2 gave a warning at the top that analytical Hessian is not possible and will use a numerical approach. Currently, i am using ideriv=1 for all my calcs but do you know if it’s possible to use analytical Hessian for 5.4 in vac and pcm while avoiding TDDFT, ideriv =2 calls TDDFT. I will email Q-Chem support otherwise. Thank you so much for your help, greatly appreciated!
In principle there is not reason to call TDDFT, so this looks like a bug (which appears to be fixed in most recent version). Please contact Q-Chem support to ask about a workaround.