In a pilot calculation, a molecule usually modeled as flat appeared unstable. We stopped to check the calculation before trusting it with a harder question.
Why this molecule?
Some formaldehyde sensors also respond to ethanol, a common ingredient in cleaning products and hand sanitizer. The size of that interference depends on the sensor and conditions; our literature review examines the evidence. In study R02, we are investigating whether a carbon-based prefilter could help separate the two gases, and how humidity may affect that question.
Before comparing possible filter materials, we test the calculation on a simpler reference: ethanol near coronene, a small, flat model of a carbon surface. A published study (opens in a new tab) measured how ethanol sticks to graphene and used coronene in its calculations. That gives us a point of comparison, though coronene is not a working filter.
The unexpected warning
Coronene has 36 atoms: 24 carbon and 12 hydrogen. The carbon atoms form seven joined rings. In our first full vibration check of the isolated molecule, the software printed three imaginary frequencies. That phrase sounds mysterious, but it has a precise meaning: in this particular calculation, moving atoms along those directions appeared to lower the energy. A true minimum should not have such downhill directions. The ORCA frequency guide (opens in a new tab) explains the test.
An imaginary frequency describes the calculated energy landscape at a chosen geometry. It does not prove that a free coronene molecule really folds. Earlier research on coronene (opens in a new tab) has found that some computational choices can produce misleading out-of-plane vibrations.
To estimate how strongly ethanol sticks, we compare the pair with ethanol and coronene calculated separately. If the isolated coronene is not a confirmed minimum, that comparison is harder to interpret.
Pilot setup
3
imaginary internal vibrations
Changed setup
0
imaginary internal vibrations
Following the clues
At the original calculation level, we checked the suspect direction with a finer numerical grid, different starting points for the electron calculation, and a rotation of the molecule's coordinate frame. A small drop in calculated energy remained in each check. Separate comparisons using a larger electron basis—the mathematical building blocks used to describe electrons—gave a different energy profile.
At a revised level, we used a larger electron basis and left out two corrections used earlier: one for weak attractions between atoms (dispersion) and one for a basis-set error. A complete vibration calculation at the same earlier geometry found no imaginary frequencies. The warning therefore depends on the computational setup. Because several settings changed together, we cannot say that the basis alone caused the difference.
There was one more catch: at the revised calculation level, the atoms were still being pulled away from the earlier geometry. A later geometry search reduced the largest leftover force by about 70,000 times. At our 24 September 2026 reporting cutoff, the new frequency check at the updated geometry was still under way, so a minimum had not been confirmed.
What did we learn?
A precise number can still answer the wrong question if the model or its settings are unsuitable. This check caught a method-sensitive warning on a reference molecule, where we could investigate it before moving to the target materials.
Our earlier ethanol–coronene candidate has a separate open question: ten very soft vibrations remain unclassified. The numerical comparison used there tests those vibrations for that structure. It is not a general error bar for future binding energies. The broader R02 pilot remains in progress; this story reports a check on the method, not a measured filter performance or a ranking of materials.
For the technical trail, see our notes on rules set before a run and repeatability.