Szczepanik Research Group

Department of Theoretical Chemistry
Faculty of Chemistry, Jagiellonian University
Gronostajowa 2, 30-387 Krakow, Poland
Tel: (+48) 12 686 23 90
E-mail: dariusz.szczepanik@uj.edu.pl

HOME      PROJECTS      PUBLICATIONS      runEDDB      X

About EDDB       Download       Tutorials       Manual


Manual (2026-Nov-03)

The practical side: what has been tested, the known limits, how to make large runs fast, what to cite, and whom to write to. Each method page of this manual ends with its own table of methods and references; this page collects the general ones.

1. Something looks wrong

SymptomWhere to look
a post-HF file gives the same numbers as HFthe correlated density was not written (the keyword of each program: Input files); -v shows the Reference type
-a selects too many or too few NOBDsa non-planar molecule: check --table nobd and select with -s (EDDB analysis)
NAO orbitals look delocalized in Avogadro2 or IQmola viewer limitation at large basis sets: use the default NAO algorithm or orbital cubes (EDDB analysis)
NMR tables say APPROXIMATE, or a WARNING says “NMR data missing”the Gaussian CSGT .log must sit next to the .fchk or be given with --nmr-log (NMR)
a differential map shows the molecule twicethe two geometries are not in one frame: --cube-align (Maps & files)
the GFN2-xTB SCC does not convergeraise --xtb-etemp, --xtb-maxiter (GFN2-xTB)
a large run is slower with more coresPerformance below
the last digit differs between two runsa different -n (Reproducibility)

2. Tested programs and formats

3. Known limits

4. Warnings and errors

Problems that do not stop the run are printed as WARNING after the stage that found them and counted on the closing line, also in quiet mode (-q), where the text itself is not shown:

terminal
$ runEDDB -i @rings/cyclobutadiene --nics-rings 4 --xtb-coupled -q
▶ Calculation completed (with 1 warning!) on 2026-10-28. Total time: 0s.

Run the same command without -q to read the warning. Problems that would make the result wrong (for example inconsistent ECP data in a Molden or Q-Chem file) are errors: the program stops and says how to fix the input.

5. Performance

OptionEffect
-n NOpenMP threads for the NAO, BOP, cube, NMR and xTB stages (default 1)
-n 0every core the job may use; the count is printed under the banner (Auto-detected 128 available CPU cores). On Linux it follows the CPU affinity, so a SLURM or taskset allocation of 8 cores gives 8.
-b nvb (default)the fastest analysis basis; -b nmb costs 10–30 % more, -b nao much more at large basis sets
-f, -p, -rrestrict the analysis, the cubes and the bond-order tables to the part you study
More cores is not always faster. The dense linear algebra of the NAO and BOP stages stops scaling after a few tens of threads; beyond that, threads wait for each other, worst on multi-socket machines. On a dual-socket 128-core AMD EPYC an aug-cc-pV5Z benchmark took 2.2 s with -n 8 and 41 s with -n 0. Use -n 0 on laptops, desktops and single-socket servers; on large nodes set -n 8–32 and time a short run at two or three settings. Cube grids, ICSS and current maps are the exception: they scale almost perfectly with cores.
Large NUMA machines: when to interleave memory

Linux places memory on the NUMA node of the thread that first touches it, which for runEDDB is the input stage. That is ideal while all threads stay on that node and harmful when a job spans several nodes. Compare -n with the cores per node (numactl --hardware):

terminal
$ numactl --hardware                                   # nodes and cores per node
$ numactl --interleave=all runEDDB -i large.fchk -n 128 # only when -n spans several nodes

On a dual-socket EPYC 9554 (8 NUMA nodes of 16 cores) a large system took 15.9 s at -n 128 by default and 9.4 s with --interleave=all; at -n 16, inside one node, interleaving was 4 % slower. Binding a large job to one socket does not help either: the work is limited by memory bandwidth, and one socket has half the memory controllers.

6. Citing runEDDB

Let the program choose: --cite both writes the exact, de-duplicated set of papers for that run to mol.bib (BibTeX) and mol.refs.txt; the set depends on the methods and options used, and -v shows the papers of every stage on screen. The papers of the method itself:

TopicReference
Bond-order projection (BOP)D.W. Szczepanik, E.J. Zak, K. Dyduch, J. Mrozek, Chem. Phys. Lett. 593 (2014) 154, doi:10.1016/j.cplett.2014.01.006
NOBDs, multicentre bondingD.W. Szczepanik, M. Andrzejak, K. Dyduch, E.J. Zak, M. Makowski, G. Mazur, J. Mrozek, Phys. Chem. Chem. Phys. 16 (2014) 20514, doi:10.1039/c4cp02932a
EDDB / EDLBD.W. Szczepanik, Comput. Theor. Chem. 1080 (2016) 33, doi:10.1016/j.comptc.2016.02.003
Three-centre orbital projectionD.W. Szczepanik, Comput. Theor. Chem. 1100 (2017) 13, doi:10.1016/j.comptc.2016.12.003
EDDB for aromaticity; fragments, pathways, CDID.W. Szczepanik, M. Andrzejak, J. Dominikowska, B. Pawełek, T.M. Krygowski, H. Szatylowicz, M. Solà, Phys. Chem. Chem. Phys. 19 (2017) 28970, doi:10.1039/c7cp06114e
Orbital-resolved EDDBD.W. Szczepanik, M. Solà, ChemistryOpen 8 (2019) 219, doi:10.1002/open.201900014
Review chapterD.W. Szczepanik, M. Solà, in Aromaticity: Modern Computational Methods and Applications, ed. I. Fernández, Elsevier (2021) 259, doi:10.1016/b978-0-12-822723-7.00008-x
Bond-delocalization field (BDF)D.W. Szczepanik, P.A. Wieczorkiewicz, ACS Phys. Chem. Au (2026), doi:10.1021/acsphyschemau.6c00031
Frozen-NMB NAOs (default)D.W. Szczepanik, J. Comput. Chem. (2026), in preparation

Methods by others that a run uses (NAO/NPA, GFN2-xTB, GIAO, CSGT, nc-AFM, ...) are listed at the end of the page that describes them, and --cite includes them automatically.

7. Contact and licences

← Options  |  Back to the start →

HOME      PROJECTS      PUBLICATIONS      runEDDB      X

About EDDB       Download       Tutorials       Manual