How conformer generation and MMFF94 workYou draw a flat structure; the site answers with three-dimensional shapes and a column of energies. This page is what happens in between — where the shapes come from, what the energies are energies of, and which questions they cannot answer. Every figure below came out of running this site.What a conformer isA drawing says which atoms are bonded to which, not where they are. Turning a single bond costs a couple of kcal/mol — the order of the thermal energy available at room temperature — so a molecule in a flask is never one shape but a population of shapes, interconverting billions of times a second.A conformer is one of them: a geometry at a minimum of the energy surface, reachable from any other by turning single bonds and never breaking one. That last clause is the whole difference from a stereoisomer. R and S are two molecules; anti and gauche butane are one molecule twice.Rotatable bonds and torsion anglesTake four bonded atoms in a row, A–B–C–D. Looking down the B→C bond, the torsion angle is how far you must turn A to bring it over D. It is the one coordinate a conformer search moves.The count above the viewer follows OpenChemLib's rule: a single, non-aromatic bond outside a ring whose both ends carry more than one non-hydrogen neighbour, and never an amide C–N. So an alcohol's O–H is not counted — its hydrogen is placed by the minimiser, never sampled — nor are cyclohexane's ring bonds. The generator is less strict: it does turn the amide C–N, but the torsion database gives that bond one populated angle, so N-methylacetamide comes back as a single conformer — six of them once 60° steps replace the database. The count sets the cost, because combinations multiply: each sp³–sp³ chain bond contributes three angles, so n-decane's seven rotatable bonds span 3⁷ = 2 187 combinations and squalene's fifteen span eighty billion.How the conformers are builtA ring shape is kept only when it is expected to be populated: above 1% of the population of the best shape of that ring at room temperature. The self-organiser first drops any shape more than twice as strained as the best one — its own overlap score, not an energy. The survivors are relaxed with MMFF94, and any that ends more than 2.72 kcal/mol above the best is dropped too. OpenChemLib takes 1.36 kcal/mol as a factor of ten in population — RT ln 10 at 25 °C — so 2.72 kcal/mol is a factor of a hundred. The threshold is fixed inside OpenChemLib; no setting here changes it.For each rotatable bond the preferred angles come from a torsion database built from crystal structures — the open data of the Crystallography Open Database: 7 382 bond environments, each with one to twenty-two preferred angles, the frequency of each, and the window it may wander in. An unknown environment is predicted from the atom types instead.60° steps throws the database away and turns every rotatable bond through a blind ladder — 0°, 60°, 120°, 180°, 240°, 300° — all six equally likely, each with a ±20° window. It is the honest option when you suspect the crystallographic prior is steering you, and it costs time: butane goes from three torsions to six, and all six minimise back onto anti and the two gauche minima, so the list still shows three.Every candidate is then built and checked for collisions: two atoms not bonded to one another may not approach closer than 0.9 times the sum of their van der Waals radii, and a set overlapping more than 3 units past the lowest strain seen so far is dropped — that strain is OpenChemLib's own overlap penalty, not a force-field energy. The strategy is told which pair collided, so it stops proposing that combination. That check, plus the random strategies giving up after 64 failures to invent a set not already delivered, is why a run can return fewer conformers than you asked for.MMFF94, and what an energy means hereA torsion set gives a plausible pose, not a relaxed one: bond lengths and angles are still whatever the rigid fragments came with. MMFF94 is an empirical force field — bond stretches, angle bends, stretch–bend coupling, out-of-plane bending, torsions, van der Waals and electrostatics, with parameters Halgren fitted to computed and measured data for organic molecules. Minimising walks the geometry downhill until the gradient flattens or the iteration budget runs out. It relaxes; it does not search, so a conformer stays in its well.What this site does not doEverything above is a geometry engine with a spring model bolted on. What it is good for: a starting point for a real calculation, a pose for a docking run, a figure, an intuition for how floppy a molecule is. What it is silent about decides whether a number off this site can be used at all.

No quantum mechanics: MMFF94 is springs and point charges. It has no electrons, so it cannot break or make a bond, cannot describe a transition state, and knows nothing about conjugation beyond what its parameter tables were fitted to reproduce.

No solvent: Every conformer here is an isolated molecule in vacuum. A shape that folds to hide a polar group is rewarded by that vacuum and would not necessarily fold in water, and an intramolecular hydrogen bond is worth far more here than it is in a hydrogen-bonding solvent.

No entropy, and therefore no populations: A relative energy is an energy, not a free energy. Turning butane’s 0.78 kcal/mol into “what fraction is gauche” needs the entropy term this site never computes — and the fact that there are two gauche wells, listed here as one, and one anti well, which no number on the page accounts for.

No Boltzmann weighting: The list is ordered by energy and nothing more. It is not a population-weighted ensemble, so averaging a property over it would weight a 5 kcal/mol conformer exactly as heavily as the lowest one.

Not an exhaustive search: Ten conformers out of the 2 187 that n-decane can reach is a sample, and a small one. Raise the conformer limit and the timeout before concluding that a shape is absent.

Minima only, never barriers: The generator visits torsion values the crystal database says are populated, and minimisation walks downhill from each. The eclipsed maximum between butane’s anti and gauche wells is never on the list, so nothing here tells you how fast the two interconvert.

One stereoisomer, chosen for you: Given an undefined stereocentre, the generator settles on one of the possible stereoisomers and every conformer of the run is that one — and which one depends on the seed: alanine drawn without a wedge comes back as (S) at seed 42 and as (R) at seed 7. Draw the wedges you mean.