Azobenzene is built from two benzene rings — six carbon atoms each — joined by a bridge of two nitrogen atoms, with a dangling bond at each end of the bridge. The molecule contains only carbon, nitrogen, and hydrogen.
It has a textbook light-switch behavior: after absorbing light, the two benzene rings swing from opposite sides of the bridge (trans) to the same side (cis). No atoms added, none removed — just a change of pose.
How that pose change happens was argued over for nearly fifty years:
- Rotation: the benzene rings rotate around the nitrogen bridge
- Inversion: the molecule bends within its own plane
- Hula twist: several segments twist together
- Inversion-assisted rotation: inversion drags rotation along
- trans to I1: driven purely by C–N torsion
- I1 to I2: 67% N–N rotation, 33% C–N torsion retained
- I2 to cis: 56% rotation, 39% inversion, 5% C–N torsion
All four pathways were proposed in turn, and none could claim structural evidence. The reason is hard: the process happens on picosecond timescales, intermediates vanish as soon as they form, and spectroscopy cannot resolve transient structures.
On September 30, 2026, a paper from Hyotcherl Ihee's team at South Korea's Institute for Basic Science (IBS) went online at Nature, and captured the whole thing on film.
The solvent buried over 90% of the signal
First, how unfriendly the problem was. Azobenzene has no heavy atoms, so X-ray scattering from it is intrinsically weak. It was also dissolved in methanol, whose scattering signal far exceeds that of the molecule itself: the solvent term accounts for more than 90% of the total TRXL signal. It is like trying to hear a faint voice in a noisy room.
The team's solution is called PEPC — projected extraction of the perpendicular component. It projects the scattering data onto a space orthogonal to the solvent response, so the solvent contribution goes mathematically to zero. This amounts to canceling the noise source in signal space, rather than shouting louder.
Three steps, each slower than the last
The result is a three-step sequential pathway, with distinct time constants:
| Stage | Time constant | Event | |---|---|---| | First | 0.4 ± 0.2 ps | I1 forms | | Second | 2.7 ± 1.0 ps | I1 converts to I2 | | Third | 14 ± 4 ps | I2 relaxes to cis |
0.4 picoseconds is 4×10⁻¹³ s. At this scale, 2.5 trillion units of 0.4 ps fit into one second.
Branching yields support this route. The I1-to-I2 conversion proceeds at 100 ± 19% yield within experimental uncertainty, and I2-to-cis at 91 ± 7% — both consistent with a dominant sequential pathway. The total trans-to-cis conversion time of about 17.1 ps matches the sum of the three constants (0.4 + 2.7 + 14 = 17.1).
The first move is a C–N twist
Structure refinement delivered the most important picture. I1 deviates clearly from planar trans-azobenzene: the CCNN torsion angles of the two rings reach about 87 degrees, while the CNNC dihedral and CNN bond angles barely change.
This means the reaction starts with C–N torsion, not the traditionally assumed N–N rotation. The molecule first twists on the carbon–nitrogen single-bond side.
By I2, things change: the CNNC rotation angle becomes 115.5 degrees, and N–N rotation takes over. The quantitative decomposition across the three steps:
Why this molecule prefers to twist rather than rotate
There is a self-consistent physical explanation, and it is elegant. When motion is concentrated at the center, the molecule pushes aside less of the surrounding medium. If both bulky rings rotated wholesale, the swept solvent volume would be enormous — like moving a large piece of furniture in a narrow hallway. The observed initial motion instead resembles bicycle pedals: the nitrogen atoms move in opposite phase on either side, while the rings barely move. This sweeps roughly 80% less solvent volume than moving the large rings alone.
This resolves a long-standing puzzle: why making the surrounding liquid more viscous barely changes azobenzene's reaction speed. Since early motion is concentrated at the molecular center and requires displacing little liquid, viscosity has little effect.
Theoretically, CASPT2 calculations independently revealed two distinct minima on the S1 potential energy surface, providing independent support for the two intermediates — not just self-consistent structure refinement.
The method matters more than this one molecule
The reach of this work extends beyond azobenzene. PEPC lets weak solute signals emerge from the solvent background, effectively removing the "must contain heavy atoms" prerequisite for X-ray liquidography. Previously, no successful TRXL study existed for small molecules lacking heavy atoms under such conditions.
After decades of competing proposals, the paper states that it settles the debate and captures motions that earlier spectroscopic studies missed.
Stated limitations
The paper appeared in Nature, titled *X-ray liquidography decodes complex motions in azobenzene isomerization*, online September 30, 2026. Experiments were performed at the XSS beamline of the Pohang Accelerator Laboratory's X-ray free-electron laser, using methanol as solvent. All authors belong to the KAIST Department of Chemistry and the IBS center. The 17.1 ps total appears in two places across reports — once as the measured conversion time and once as the sum of the three time constants — with consistent values. The 80% reduction in swept solvent volume is the team's own analysis. CASPT2 is theoretical support, not an independent experiment. The experiment was performed only in methanol, not repeated in the solid state or other solvents.
The paper does not make any azobenzene-based drug or material better. Its value, as the team positions it, lies elsewhere: it reveals the actual pathway the molecule takes, which can serve as a reference for designing photoresponsive materials and molecular machines. The most direct hint for designers: move the molecular center first, not the large rings.
Sources: Nature, *X-ray liquidography decodes complex motions in azobenzene isomerization*, online 2026-09-30. Experiments at the Pohang Accelerator Laboratory XFEL's XSS beamline in methanol; all authors affiliated with KAIST Department of Chemistry and the IBS center. Data from a Tencent News in-depth report (Oct 1) including the PEPC principle diagram, CCNN and CNNC angles, three time constants, and branching yields. Quotes from AJU Press and Korea Herald reports (Oct 1), citing Ihee and co-first authors Kim Jung-min and Ki Ho-sung. The DOI on nature.com was not retrieved in this search; cited here by journal name and date.