No Universal Virtual Temperature
April 11, 2016 – Under the electron beam, graphene changes as if it were thousands of degrees hot. Researchers at Ulm University counted how often a divacancy takes each of its three shapes in 935 frames at 80 kV and found the highest-energy shape the most common, so no temperature fits, and the idea of a universal virtual temperature does not hold.
Under the electron beam of a transmission electron microscope (TEM), graphene does things it would otherwise only do in a furnace. Carbon atoms swap partners, defects wander, flakes curl up into fullerenes[1]. A single bond rotation, in which two neighbouring carbon atoms turn by 90° around their common centre, has an activation barrier of 5–10 eV[2]. For it to happen about once per second through thermal vibrations alone, the sheet would have to be at roughly 2000–4000 K. Yet the sample stays near room temperature: a strong beam heats a thin carbon film locally by only a few kelvin[1]. Instead, single electrons hit single atoms, and occasionally one of these collisions, or an excitation of the electrons in the material, pushes the lattice over the barrier[3],[4].
Because the results often look as if the sample had been heated, the idea of an electron-beam-induced “virtual temperature” has gained ground. Researchers at Ulm University put this idea to a direct test. With an aberration-corrected FEI Titan 80-300 at 80 kV, they counted how often a divacancy in graphene, a defect formed by two missing carbon atoms, appeared in each of its three shapes, and compared the counts with the Boltzmann distribution that governs any system in thermal equilibrium. The answer was clear: no temperature at all fits the measured populations.
Furnace or beam?
The similarities between beam-driven and heat-driven behaviour are striking. Under the beam, a graphitic flake transformed into a fullerene[5],[6], a sulphur-terminated graphene nanoribbon assembled inside a carbon nanotube[7],[8], grain boundaries in graphene migrated[9],[10], and edge dislocations moved and annihilated[11]–[13]. In each case the system was driven towards energetically favourable states, as it would be in a hot sample. A parallel heating experiment on the nanoribbon showed comparable behaviour at about 1300 K as under 80 keV electron irradiation[7]. A group at the University of Vienna followed a divacancy with scanning TEM (STEM) at 60 kV and derived a virtual temperature of 3050 K from how often it jumped[14],[15].
Two questions remained open, the authors write: is the behaviour under the beam really the same as at an elevated temperature, and if so, how high is that temperature? They also point out that under the beam, the clock no longer ticks with ordinary time. What drives the sample forward is the arrival of electrons, so the time variable becomes the electron flux.
Three shapes of one defect
The divacancy is an ideal test object. It comes in three frequently observed shapes[16] (Fig. 1): dv1, with two pentagons and an octagon (585); dv2, with three pentagons and three heptagons (555-777); and dv3, with four pentagons, one hexagon and four heptagons (5555-6-7777). A single bond rotation converts one shape into the next, and the electron beam can trigger it[3],[17]. In the honeycomb lattice, dv1 and dv3 can sit in three orientations and dv2 in two, so the three states have degeneracies of 3, 2 and 3.
In thermal equilibrium, each state is occupied with a probability proportional to its degeneracy times exp(−Ef/kBT), where Ef is its formation energy. The team used the ratio of two such probabilities, weighted by the degeneracies. In this ratio, the partition function cancels, so neither it nor any other possible configuration needs to be known. By the principle of detailed balance, one defect that switches many times can stand in for many defects observed at one moment. The task was therefore to count, frame by frame, in which state a divacancy sits, and to find the temperature at which the ratios match the calculated energies.
935 frames of ten divacancies
The sample was single-layer graphene grown by chemical vapour deposition, transferred onto a TEM grid and dry-cleaned[18]. The image-side Cs-corrected FEI Titan 80-300 (today Thermo Fisher Scientific) ran at 80 kV in high-resolution mode, with the spherical aberration tuned below 20 µm so that atoms appear dark at Scherzer focus, and with the extraction voltage lowered to 2 kV to narrow the energy spread. The team searched the sheet for divacancies, whether already present or created by the beam[19],[20], and recorded image sequences of ten of them, 935 frames in total, each exposed for 1 s at a dose rate of 1–2 × 106 e/(nm2 s). On average the defect changed shape every third frame (Fig. 1a–h, Video 1). Some changes, for example between Fig. 1a and b, cannot be explained by a single bond rotation.
Video 1: AC-HRTEM image sequence at 80 kV of a divacancy in graphene switching between its states, without the noise filter of Fig. 1 (Supplemental Movie 1 of the paper).
About 14 % of the frames showed a blurred or distorted defect, because it had switched during the exposure; these frames were discarded. In the rest, dv1 appeared in 474 frames, dv2 in 198 and dv3 in 263 (Table 1). No statistically significant difference showed up between the sequences, so no dependence on the dose rate was found within the narrow range used.
| Rings | Orientations g | Frames | Probability | Ef (eV) | |
|---|---|---|---|---|---|
| dv1 | 585 | 3 | 474 | 0.51 ± 0.04 | 7.7 |
| dv2 | 555-777 | 2 | 198 | 0.21 ± 0.03 | 6.8 |
| dv3 | 5555-6-7777 | 3 | 263 | 0.28 ± 0.03 | 7.1 |
Table 1: Frames in which each divacancy state was seen in 80 kV AC-HRTEM, the resulting probabilities (binomial confidence intervals, about 95 %) and the formation energies from the analytical force field in a cell of about 10 000 atoms (this work). The highest-energy state is the most frequent.
The highest energy wins
To compare the counts with Boltzmann statistics, the formation energies of the three states had to be known precisely. The team calculated them for a sheet of about 10 000 carbon atoms, large enough to capture long-range strain and corrugation, which smaller calculations miss; how this was done and why it matters is explained in the box below. The result: dv2 has the lowest formation energy, 6.8 eV, followed by dv3 with 7.1 eV and dv1 with 7.7 eV.
Boltzmann statistics would make dv2, the lowest state, the most common one and dv1, the highest, the rarest. The experiment showed the opposite: dv1 appeared in 51 % of the frames, more often than the two others together. Per orientation, the state 0.9 eV above dv2 was seen 1.6 times as often, and the state 0.6 eV above dv3 1.8 times as often (Fig. 2). In Boltzmann terms this would mean a negative temperature. “It becomes immediately clear that no meaningful fit of an exponential curve can be made to the data points,” the authors write; a least-squares fit gives a negative R2. Even in the extreme case that all discarded frames had shown the lowest state, dv1 would still have been the most frequent. An earlier Ulm experiment on self-interstitial dimers had shown similar behaviour: there, the lowest-energy configuration was the one seen most rarely under the beam[23],[24].
An independent measurement of the same defect with another microscope, another voltage and another imaging mode found almost the same populations. The box below compares both.
Why the beam is no furnace
A population inversion, the textbook departure from Boltzmann statistics, is known from lasers, where a narrow band of energy favours one specific transition. Elastic collisions with beam electrons offer no such band: the probability of transferring energy to an atom falls steadily with the amount transferred[25]. The plasmon peaks in the energy-loss spectrum of graphene, however, lie between 4.7 and 14.5 eV[26], close to the 5–10 eV barriers of the bond rotations, and their positions do not depend on the beam energy. That would fit the near-identical statistics at 80 and 60 kV and point to electronic excitations as the driver, but the authors call this “quite speculative at this point”.
Keeping an Arrhenius picture with barriers that differ in the two directions would not help either, the authors argue, because such asymmetric barriers could be fitted to reproduce almost any population. The obvious way out is to abandon the picture of a fixed energy landscape. Molecular dynamics simulations had shown that a bond rotation under the beam starts with an atom displaced from its site that cannot escape; before it falls back, its neighbours have rearranged[3]. This non-equilibrium process depends delicately on the direction of the initial kick and on the vibrations of all atoms involved, and there is no reason why the window for a switch from a low- to a high-energy state should be narrower than for the reverse, as the Boltzmann distribution would require.
Challenges named by the authors
The authors point out several limits. Under the beam, new divacancies form and existing ones grow, so the situation is not a strict equilibrium; it was treated as one because each divacancy switched many times before it disappeared or grew. A switch during a 1-s exposure can blur the image, and a short-lived state near the start or end of an exposure can vanish behind the longer-lived one, which leaves some uncertainty in the counts, although the ranking stays the same if all states are affected alike. The dose rate varied only between 1 and 2 × 106 e/(nm2 s), and the data on pairs of nearby divacancies were sparse. The role of plasmons remains speculative, and an experiment on a tilted sample, which would change the directions of the electron kicks, could test the mechanism but is difficult to image.
A warning for the analogy
For a sample to behave exactly as if it were hot, every process in it would have to follow thermally activated behaviour, without exception. The divacancy is a counterexample, and so the high-temperature analogy is not universal. Earlier observations that matched a hot sample remain valid, but they looked at selected processes and must be seen as special cases. “Our results show that great care needs to be taken when resorting to the analogy between electron beam and thermally driven transformations,” the authors conclude. For in situ TEM, this means that beam-driven changes in a 2D material need to be interpreted as what they are, a sequence of single electron impacts, rather than as a snapshot of a furnace.
Resource:
Börner, P., Kaiser, U., Lehtinen, O. (2016).
Evidence against a universal electron-beam-induced virtual temperature in graphene.
Phys. Rev. B 93(13), 134104.
https://doi.org/10.1103/PhysRevB.93.134104
Update – May 14, 2019
The news article of April 11, 2016 ends above. Later work that builds on it is listed below.
Beam heating in 2D materials
A review from the University of Vienna and the University of Tübingen on quantifying electron-beam effects in 2D materials cited the Ulm study for the statement that “direct heating by the electron beam is negligible”[27]. It also named plasmon excitations in the range of about 5–25 eV, which can be quite intense even in atomically thin materials, among the inelastic processes that need to be quantified.
Update – December 9, 2025
Real heat on a chip
When a study from Ulm University and the University of Nottingham on the solidification of liquid metal nanoparticles on graphene needed real temperatures, it set them with heating chips between 20 and 800 °C, and cited the Ulm result to state that beam-induced heating of the graphene sample is not a concern[28] (see the news article Stationary Atoms Solidification).
Download this news article as PDF – for reading on paper, taking notes or filing.
-
Egerton, R., Li, P., Malac, M. (2004). Radiation Damage in the TEM and SEM. Micron, 35, 399–409. https://doi.org/10.1016/j.micron.2004.02.003
-
Li, L., Reich, S., Robertson, J. (2005). Defect Energies of Graphite: Density-Functional Calculations. Phys. Rev. B, 72, 184109. https://doi.org/10.1103/PhysRevB.72.184109
-
Kotakoski, J., Meyer, J. C., Kurasch, S., Santos-Cottin, D., Kaiser, U., Krasheninnikov, A. V. (2011). Stone-Wales-Type Transformations in Carbon Nanostructures Driven by Electron Irradiation. Phys. Rev. B, 83, 245420. https://doi.org/10.1103/PhysRevB.83.245420
-
Algara-Siller, G., Kurasch, S., Sedighi, M., Lehtinen, O., Kaiser, U. (2013). The Pristine Atomic Structure of MoS2 Monolayer Protected From Electron Radiation Damage by Graphene. Appl. Phys. Lett., 103, 203107. https://doi.org/10.1063/1.4830036
-
Chuvilin, A., Kaiser, U., Bichoutskaia, E., Besley, N. A., Khlobystov, A. N. (2010). Direct Transformation of Graphene to Fullerene. Nat. Chem., 2, 450–453. https://doi.org/10.1038/nchem.644
-
SALVE news (2010). Fullerenes From Graphene. salve-project.de, May 9, 2010. https://salve-project.de/newspress/formation-of-fullerenes-out-of-graphene.html
-
Chuvilin, A., Bichoutskaia, E., Gimenez-Lopez, M. C., Chamberlain, T. W., Rance, G. A., Kuganathan, N., Biskupek, J., Kaiser, U., Khlobystov, A. N. (2011). Self-Assembly of a Sulphur-Terminated Graphene Nanoribbon Within a Single-Walled Carbon Nanotube. Nat. Mater., 10, 687–692. https://doi.org/10.1038/nmat3082
-
SALVE news (2011). Sulphur-Terminated Nanoribbons. salve-project.de, August 7, 2011. https://salve-project.de/newspress/sulphur-terminated-graphene-nanoribbon.html
-
Kurasch, S., Kotakoski, J., Lehtinen, O., Skákalová, V., Smet, J., Krill, C. E., Krasheninnikov, A. V., Kaiser, U. (2012). Atom-by-Atom Observation of Grain Boundary Migration in Graphene. Nano Lett., 12, 3168–3173. https://doi.org/10.1021/nl301141g
-
SALVE news (2012). Grain Boundary Migration. salve-project.de, May 10, 2012. https://salve-project.de/newspress/grain-boundary-migration-in-graphene.html
-
Warner, J. H., Margine, E. R., Mukai, M., Robertson, A. W., Giustino, F., Kirkland, A. I. (2012). Dislocation-Driven Deformations in Graphene. Science, 337, 209–212. https://doi.org/10.1126/science.1217529
-
Lehtinen, O., Kurasch, S., Krasheninnikov, A., Kaiser, U. (2013). Atomic Scale Study of the Life Cycle of a Dislocation in Graphene From Birth to Annihilation. Nat. Commun., 4, 2098. https://doi.org/10.1038/ncomms3098
-
SALVE news (2013). 3D Buckling in Graphene. salve-project.de, June 28, 2013. https://salve-project.de/newspress/3d_buckling_under_the_microscope.html
-
Kotakoski, J., Mangler, C., Meyer, J. C. (2014). Imaging Atomic-Level Random Walk of a Point Defect in Graphene. Nat. Commun., 5, 3991. https://doi.org/10.1038/ncomms4991
-
SALVE news (2013). Defects in Silica and Graphene. salve-project.de, December 16, 2013. https://salve-project.de/newspress/defects-in-bilayer-silica-and-graphene-common-trends-in-hexagonal-systems.html
-
Banhart, F., Kotakoski, J., Krasheninnikov, A. V. (2010). Structural Defects in Graphene. ACS Nano, 5, 26–41. https://doi.org/10.1021/nn102598m
-
SALVE news (2011). Transformations in Carbon Nanostructures. salve-project.de, June 23, 2011. https://salve-project.de/newspress/transformation-in-carbon-nanostructures.html
-
Algara-Siller, G., Lehtinen, O., Turchanin, A., Kaiser, U. (2014). Dry-Cleaning of Graphene. Appl. Phys. Lett., 104, 153115. https://doi.org/10.1063/1.4871997
-
Meyer, J. C., Eder, F., Kurasch, S., et al. (2012). Accurate Measurement of Electron Beam Induced Displacement Cross Sections for Single-Layer Graphene. Phys. Rev. Lett., 108, 196102. https://doi.org/10.1103/PhysRevLett.108.196102
-
SALVE news (2012). Knock-on Damage in Graphene Linked to Lattice Vibrations. salve-project.de, May 7, 2012. https://salve-project.de/newspress/knock-on-damage-in-graphene-linked-to-lattice-vibrations.html
-
Lee, S., Hwang, G. S. (2011). Valence Force Field-Based Monte Carlo Bond-Rotation Method for the Determination of sp2-Bonded Carbon Structures. J. Appl. Phys., 110, 093524. https://doi.org/10.1063/1.3660383
-
Kotakoski, J., Eder, F. R., Meyer, J. C. (2014). Atomic Structure and Energetics of Large Vacancies in Graphene. Phys. Rev. B, 89, 201406. https://doi.org/10.1103/PhysRevB.89.201406
-
Lehtinen, O., Vats, N., Algara-Siller, G., Knyrim, P., Kaiser, U. (2014). Implantation and Atomic-Scale Investigation of Self-Interstitials in Graphene. Nano Lett., 15, 235–241. https://doi.org/10.1021/nl503453u
-
SALVE news (2015). First Observation of Self-Interstitials in Graphene. salve-project.de, January 1, 2015. https://salve-project.de/newspress/first-observation-of-self-interstitials-in-graphene.html
-
Reimer, L., Kohl, H. (2008). Transmission Electron Microscopy: Physics of Image Formation, 5th ed. Springer, New York. https://doi.org/10.1007/978-0-387-40093-8
-
Eberlein, T., Bangert, U., Nair, R. R., Jones, R., Gass, M., Bleloch, A. L., Novoselov, K. S., Geim, A., Briddon, P. R. (2008). Plasmon Spectroscopy of Free-Standing Graphene Films. Phys. Rev. B, 77, 233406. https://doi.org/10.1103/PhysRevB.77.233406
-
Susi, T., Meyer, J. C., Kotakoski, J. (2019). Quantifying Transmission Electron Microscopy Irradiation Effects Using Two-Dimensional Materials. Nat. Rev. Phys., 1, 397–405. https://doi.org/10.1038/s42254-019-0058-y
-
Leist, C., Ghaderzadeh, S., Kohlrausch, E. C., et al. (2025). Stationary Atoms in Liquid Metals and Their Role in Solidification Mechanisms. ACS Nano, 19, 42002–42012. https://doi.org/10.1021/acsnano.5c08201


