ALMO-EDA/COVP SCF Convergence Failure for Open-Shell Anion Radical System

Hridita_Purba_Saha · June 3, 2026, 10:58pm

Hello, I am running an ALMO-EDA calculation (EDA2 = FALSE ) with COVP analysis in Q-Chem 6.4.0 on an intramolecular donor–acceptor system. The system is an anion radical with overall charge = −1 and multiplicity = 2. The fragments are defined as:

Fragment 1 (biphenyl donor): charge = −1, multiplicity = 1
Fragment 2 (cyclohexane-bridged naphthalene acceptor): charge = 0, multiplicity = 2

The fragment calculations appear to complete successfully, but the combined ALMO/SCF step fails to converge. Even after 500 SCF cycles, the energy oscillates substantially and does not approach convergence. Are there any recommended strategies (e.g., different SCF algorithms, level shifting, MOM, fragment spin assignments, or alternative initial guesses) that might improve convergence?

Any guidance would be greatly appreciated.
I am sharing my input file:

$molecule
-1 2
--
-1 1
C         -1.06129       -1.38834        0.86231
C         -1.98190       -1.06148        1.86571
C         -3.34703       -1.22181        1.72370
C         -3.93191       -1.74700        0.52666
C         -2.98229       -2.07465       -0.49436
C         -1.62636       -1.89482       -0.31843
C         -5.33982       -1.94315        0.36842
C         -6.30528       -1.42856        1.30030
C         -7.65989       -1.63008        1.14386
C         -8.18078       -2.36085        0.06121
C         -7.25877       -2.87675       -0.86780
C         -5.90213       -2.67817       -0.73052
H         -1.58706       -0.68542        2.80563
H         -3.98310       -0.97744        2.56764
H         -3.33094       -2.45875       -1.44683
H         -0.94728       -2.14944       -1.13083
H         -5.97167       -0.83359        2.14390
H         -8.34045       -1.20360        1.87851
H         -9.24747       -2.51960       -0.05461
H         -7.61867       -3.45740       -1.71521
H         -5.24359       -3.12524       -1.46764
--
0 2
C          2.32828        0.55327        0.73990
C          3.14022       -0.66283        1.21121
C          2.49201       -1.30986        2.44774
C          0.95274       -1.19338        2.44486
C          0.43638       -1.18507        0.99897
C          0.89676        0.12823        0.31291
H          2.24035        1.22638        1.60094
H          4.16842       -0.36129        1.43352
H          3.20608       -1.38841        0.39265
H          2.78032       -2.36536        2.48920
H          2.88997       -0.84538        3.35602
H          0.65697       -0.26329        2.94853
H          0.52707       -2.01959        3.02108
H          0.94193       -2.01895        0.48127
H          0.85056        0.00134       -0.77325
H          0.20723        0.94229        0.55601
C          5.42748        5.25454       -1.92543
C          5.63946        4.63545       -3.17796
C          5.16711        3.37069       -3.40762
C          4.46289        2.66504       -2.40068
C          4.24634        3.28626       -1.14358
C          4.74769        4.59645       -0.93527
C          3.96033        1.35356       -2.60118
C          3.28298        0.70376       -1.60967
C          3.05491        1.31293       -0.34547
C          3.53656        2.57970       -0.13803
H          5.80357        6.25702       -1.75229
H          6.17540        5.16709       -3.95648
H          5.32350        2.89085       -4.36882
H          4.58268        5.07229        0.02641
H          4.11772        0.87232       -3.56151
H          2.90545       -0.29759       -1.78928
H          3.37548        3.06629        0.82024
$end

$rem
   JOBTYPE          EDA
   EDA2             FALSE
   METHOD           wB97X-D
   BASIS            6-31G**
   PURECART         1112
   FRGM_METHOD      STOLL
   FRGM_LPCORR      ARS
   SCF_PRINT_FRGM   TRUE
   EDA_COVP         TRUE
   EDA_PRINT_COVP   TRUE
   UNRESTRICTED     TRUE
   IANLTY           200
   MAX_SCF_CYCLES   500
   SCF_CONVERGENCE  8
   SCF_ALGORITHM    GDM
   THRESH           14
$end

$plots
MOs
   120  -13.2  10.6
    91   -7.9  10.3
    80   -8.4   7.4
   12  0  0  0
   96  97  98  99  100  101  95  96  97  98  99  100
$end

jherbert · June 4, 2026, 1:47pm

You could try SCF_ALGORITHM = ROBUST.

Hridita_Purba_Saha · June 4, 2026, 6:08pm

Thank you for the suggestion. Unfortunately, SCF_ALGORITHM = ROBUST does not appear to be supported within the fragment subsystem calculations in Q-Chem 6.4.0. The job crashes immediately on Fragment 1 with the following error:

Q-Chem fatal error occurred in module libgen/printscf.C, line 151:
Illegal MetSCF in SCFman
Input file:

$molecule
-1 2
--
-1 1
C         -1.06129       -1.38834        0.86231
C         -1.98190       -1.06148        1.86571
C         -3.34703       -1.22181        1.72370
C         -3.93191       -1.74700        0.52666
C         -2.98229       -2.07465       -0.49436
C         -1.62636       -1.89482       -0.31843
C         -5.33982       -1.94315        0.36842
C         -6.30528       -1.42856        1.30030
C         -7.65989       -1.63008        1.14386
C         -8.18078       -2.36085        0.06121
C         -7.25877       -2.87675       -0.86780
C         -5.90213       -2.67817       -0.73052
H         -1.58706       -0.68542        2.80563
H         -3.98310       -0.97744        2.56764
H         -3.33094       -2.45875       -1.44683
H         -0.94728       -2.14944       -1.13083
H         -5.97167       -0.83359        2.14390
H         -8.34045       -1.20360        1.87851
H         -9.24747       -2.51960       -0.05461
H         -7.61867       -3.45740       -1.71521
H         -5.24359       -3.12524       -1.46764
--
0 2
C          2.32828        0.55327        0.73990
C          3.14022       -0.66283        1.21121
C          2.49201       -1.30986        2.44774
C          0.95274       -1.19338        2.44486
C          0.43638       -1.18507        0.99897
C          0.89676        0.12823        0.31291
H          2.24035        1.22638        1.60094
H          4.16842       -0.36129        1.43352
H          3.20608       -1.38841        0.39265
H          2.78032       -2.36536        2.48920
H          2.88997       -0.84538        3.35602
H          0.65697       -0.26329        2.94853
H          0.52707       -2.01959        3.02108
H          0.94193       -2.01895        0.48127
H          0.85056        0.00134       -0.77325
H          0.20723        0.94229        0.55601
C          5.42748        5.25454       -1.92543
C          5.63946        4.63545       -3.17796
C          5.16711        3.37069       -3.40762
C          4.46289        2.66504       -2.40068
C          4.24634        3.28626       -1.14358
C          4.74769        4.59645       -0.93527
C          3.96033        1.35356       -2.60118
C          3.28298        0.70376       -1.60967
C          3.05491        1.31293       -0.34547
C          3.53656        2.57970       -0.13803
H          5.80357        6.25702       -1.75229
H          6.17540        5.16709       -3.95648
H          5.32350        2.89085       -4.36882
H          4.58268        5.07229        0.02641
H          4.11772        0.87232       -3.56151
H          2.90545       -0.29759       -1.78928
H          3.37548        3.06629        0.82024
$end

$rem
   JOBTYPE          EDA
   EDA2             FALSE
   METHOD           wB97X-D
   BASIS            6-31G**
   PURECART         1112
   FRGM_METHOD      STOLL
   FRGM_LPCORR      ARS
   SCF_PRINT_FRGM   TRUE
   EDA_COVP         TRUE
   EDA_PRINT_COVP   TRUE
   UNRESTRICTED     TRUE
   IANLTY           200
   SCF_GUESS        FRAGMO
   MAX_SCF_CYCLES   500
   SCF_CONVERGENCE  8
   SCF_ALGORITHM    ROBUST
   THRESH           14
$end

$plots
MOs
   120  -13.2  10.6
    91   -7.9  10.3
    80   -8.4   7.4
   12  0  0  0
   96  97  98  99  100  101  95  96  97  98  99  100
$end

jherbert · June 4, 2026, 8:41pm

Then I guess you need to try a different SCF_ALGORITHM. For large molecules/basis sets, I would also set S2THRESH = 16 just to be safe.

Hridita_Purba_Saha · June 5, 2026, 1:08am

The fragment SCFs converge normally, but the supersystem SCF diverges immediately, producing extremely large energies and NaN orbital/coupling values. I have tested several SCF algorithms (GDM, DIIS_GDM, and ADIIS_DIIS) without success. Since the donor and acceptor are covalently connected through the bridge (D-B-A), I am wondering whether the FRAGMO approach is appropriate for such systems. Does FRAGMO assume weakly interacting or non-covalent fragments, making it unsuitable when covalent bonds cross the fragment boundary?

jherbert · June 5, 2026, 8:02am

FRAGMO doesn’t “assume” anything, it’s a a superposition of fragment density matrices. I wouldn’t use it for covalent bonds.

Yuezhi · June 13, 2026, 12:21am

I think you should remove the keyword EDA2 = FALSE. When this is turned on, it goes through the old ALMO-EDA implementation in the old (mostly deprecated) SCF code, and a lot of SCF convergence techniques (including the GDM algorithm) are not supported for ALMO-EDA calculations using that code.