Limit of atomic-resolution-tomography reconstruction of amorphous nanoparticles

Summary

This Nature Analysis paper uses realistic simulations of atomic-resolution electron tomography (AET) to define what AET can — and crucially cannot — determine for amorphous nanoparticles. The authors simulated projection images and reconstructions across a range of electron fluences, sampling, projection coverage and compositions to quantify the accuracy of atomic-position recovery and the limits of chemical identification in non-crystalline particles.

For single-element (monoatomic) nanoparticles, atomic positions can be recovered to accuracies on the order of tens of picometres, but only under strict conditions of high fluence, fine sampling and many projection angles. For multi-element amorphous nanoparticles, chemical identification becomes limited by noise and sampling: heavier atoms are easier to distinguish, while overlaps in atomic peak intensity and background produce large uncertainties. The study maps nanoparticle size, composition, electron dose and image-sampling requirements and supplies simulation data and code for others to use as a benchmark.

Key Points

Simulation-based study establishes the practical limits of AET for amorphous nanoparticles, not just crystals.
Monoatomic particles: atomic positions can be recovered to tens of picometres but only with stringent electron fluence, fine pixel sampling and dense projection coverage.
Multi-element particles: chemical identification is noise- and sampling-limited; heavier elements are resolved more reliably than lighter ones.
Atomic peak intensity overlap and background noise cause large chemical-assignment errors; k-means clustering approaches can fail when intensity distributions overlap.
Experimental factors that degrade reconstructions include the missing wedge, depth-of-focus effects and insufficient sampling — all are quantified in extended-data simulations.
Reconstruction algorithms (SIRT vs RESIRE) produced nearly identical tomograms in tests, indicating algorithm choice alone is not the main limit.
Authors provide data and simulation code (Zenodo and GitHub) so the community can reproduce and design experiments to meet the identified limits.

Context and relevance

AET has matured as a tool for crystalline materials, but amorphous solids are harder: no lattice symmetry means more degrees of freedom and lower signal per identifiable site. This work is important because it converts qualitative concerns about noise, dose and sampling into quantitative requirements and failure modes. For researchers using AET to study amorphous catalysts, glasses, or mixed-composition nanoparticles, the paper gives concrete guidance on when 3D atomic positions and chemistry claims are defensible and when they are not. It also directly tests the experimental claims in earlier high-profile reports, showing where low identification success can arise from unavoidable measurement limits rather than algorithmic failings alone.

Why should I read this

Short version: if you do electron tomography, nanoparticle characterisation or design experiments that claim 3D atomic maps for amorphous materials — read this. It tells you exactly what dose, sampling and projection coverage you need, warns where chemical assignments will be dodgy, and points you to the data and code so you can check your own setups. Saves you time and avoids over-claiming in papers and grant proposals.

Author’s take

Punchy and to the point: this is a benchmarking paper. It doesn’t just celebrate what AET can do — it sets the bar for responsible interpretation. If you value rigour in atomic-scale 3D imaging, this is a must-read.

Source

Source: https://www.nature.com/articles/s41586-025-09924-w

Data & code

Simulated images and models: https://zenodo.org/records/10850980. ZMULT image-simulation code: https://github.com/rtbusch/AET_CodeModRepo (links and resources listed by the authors).