Scf convergence issue

Questions
#1

Hi
I am facing “SCF failed to converge” issue
Here is my input:

$molecule
0 3
O -0.8711101148 0.2979839273 -0.1724742885
C -0.2757733286 0.1196776575 0.9961011957
O 0.9147658241 0.0212806814 1.1648111133
C -0.0074170753 0.3783049723 -1.3219254857
H -0.6871587789 0.5163838946 -2.1684843364
H 0.5546652521 -0.5582060925 -1.4359910626
H 0.6760228314 1.2303119641 -1.2190790880
C -1.2838728278 0.0585306019 2.1192734994
C -1.9258769446 1.1566928188 2.5119568314
H -2.6046103887 1.0844352900 3.3704026844
C -1.8666868627 2.4945219453 1.8740695454
C -0.8958008663 3.4189697158 2.2858093322
H -0.1563170806 3.1600106157 3.0432201390
C -0.8225502398 4.6792743889 1.6523474690
H -0.0295251199 5.3645992183 1.9395660752
C -2.7961480756 2.8337767238 0.9176033311
H -3.5409344424 2.1058173785 0.5985995865
C -2.7333431858 4.0890444900 0.2795534508
H -3.4350810611 4.2981546017 -0.5227728451
C -1.7449172113 5.0287553536 0.6634837864
N -1.6916390966 6.2590010285 0.0780531639
C -2.6651400324 6.6409603254 -0.9252462510
H -2.4329529487 7.6511282442 -1.2814926057
H -2.6605760008 5.9681264439 -1.8005619423
H -3.6949621358 6.6525580018 -0.5251006900
C -0.6557726369 7.1910905376 0.4617753160
H -0.7559655186 8.1074217791 -0.1312754755
H 0.3551954934 6.7829836304 0.2863704948
H -0.7216401864 7.4690146161 1.5286918056
C -1.4774505950 -1.2256526457 2.8374786089
O -1.9105200377 -1.3213096513 3.9632432516
O -1.0785931161 -2.2492445032 2.0930985218
C -1.2112678576 -3.5526403714 2.6847268359
H -0.8192973872 -4.2554480732 1.9437426770
H -2.2661568617 -3.7634015170 2.9020128686
H -0.6256606405 -3.6050699517 3.6125825878
$end

$rem
BASIS = cc-pVDZ
EXCHANGE = wB97XD
JOB_TYPE = optimization
geom_opt_max_cycles = 200
MAX_CIS_CYCLES = 400
SCF_CONVERGENCE = 8
MAX_SCF_CYCLES = 400
THRESH = 14
UNRESTRICTED = true
SPIN_FLIP = 1
CIS_N_ROOTS = 4
sts_mom = true
SYMMETRY_IGNORE = true
SYMMETRY = false
CIS_STATE_DERIVATIVE = 1
SOLVENT_METHOD = PCM
$end

$solvent
DIELECTRIC 32.613000
OPTICALDIELECTRIC 1.765709
$end

$opt
CONSTRAINT
tors 12 11 9 8 -90.00
ENDCONSTRAINT
$end

output:

389 -898.9941242689 1.95e-05
390 -898.9943753106 3.76e-06
391 -898.9943935124 4.72e-06
392 -898.9944184020 5.05e-06
393 -898.9942854386 7.82e-06
394 -898.9940479034 2.02e-05
395 -898.9941076506 2.31e-05
396 -898.9941624585 2.36e-05
397 -898.9942047901 2.31e-05
398 -898.9942312291 2.15e-05
399 -898.9942239616 1.94e-05
400 -898.9942148762 2.05e-05
gen_scfman_exception: SCF failed to converge

Q-Chem fatal error occurred in module libgscf/gen_scfman/gen_scfman_main.C, line 244:

Error in gen_scfman

Please submit a crash report at Q-Chem Crash Reporter

Can anyone help me with this?

#2

Assuming that the constrained geometry optimization in your SF calculation is giving robust structures, for starters, you may consider switching to a different SCF algorithm at the last geometry for which the default SCF algorithm (DIIS) failed. Please refer to the Q-Chem User Manual (4.5.2 Basic Convergence Control Options‣ 4.5 Converging SCF Calculations ‣ Chapter 4 Self-Consistent Field Ground-State Methods ‣ Q-Chem 6.2 User’s Manual) for all available SCF algorithms that you can use through the SCF_ALGORITHM $rem variable. I would start with the hybrid DIIS_GDM algorithm in your case as the DIIS seems to be approaching convergence but fails. Also, a MAX_SCF_CYCLES of 400 seems to be too large in my experience. In general, if SCF does not converge within, say, a 100 cycles, I would stop and look at its behavior before proceeding to increase MAX_SCF_CYCLES or change SCF_ALGORITHM or use some other trick.

#3

this looks like a very funny (and expensive) way to optimize the geometry of the ground state, which is what you are doing in this calculation.

#4

Dear @jherbert
I want to do ground-state PES at different angles. I am doing constraint optimization at different angles. Am I wrong here if I wish to do this geometry?

#5

You are using the first excited state of spin-flip TDDFT with a triplet reference, which should in fact be the singlet ground state. If it’s your desire to optimize (either with or without constraints) on the singlet ground state, why not do that as a ground-state geometry optimization? That is significantly cheaper and generally less fraught.

#6

If the ground state is a triplet with a multi-configurational wave function, then optimizing the appropriate SF state would be the better approach. If, at different angles, the ground state is predominantly a closed-shell singlet, then optimization without SF, would be cheaper and straightforward as Prof. Herbert explained.

If there is bond breaking involved upon twisting the dihedral (in other words, if the diradical character increases upon twisting), I prefer to stick to SF approaches. Your dihedral angle suggests that you are not twisting the dihedral around a double bond between atoms 8 and 9 though, so you might not require the SF method after all.

#7

Thanks @kaushik and @jherbert
I also twisted around double bonds (i.e., atoms 8 and 9), and I did not face any convergence problem at this dihedral.
I also want to do a single bond rotation using the same method (i.e., SF-TD-DFT).
I tried this geometry with the DIID-GDM algorithm. The output is running and keeps going on, but I do not think it will converge.
So, Can you suggest anything that will work?
With the particular dihedral (12 11 9 8), I am facing issues at -90, -100, -110 degrees. Before and after degrees were smoothly optimized, and even at cis_state_derivative = 3 (which I am considering S1), I did not face any convergence issues.

#8

The (12 11 9 8) dihedral is not likely to modify the electronic structure so I still contend that ground-state optimization is the better choice.

If it converges at nearby angles you can use those as a guess.

#9

Hi @jherbert @kaushik
Using the DFT method, I tried to optimize my molecule with (12 11 9 8) dihedral at 90, 100, and 110 degrees at the ground. I did not face any issues; it optimized smoothly.
Can anyone tell me why I am facing issues with the SF-TDDFT method?

#10

excited-state optimizations are inherently more difficult, lot of moving parts. not to mention that SF-TDDFT is certainly not perfect.