Fig. 4: Characterization of the interaction energy landscape in LPTEM using LEONARDO.
From: Learning the diffusion of nanoparticles in liquid phase TEM via physics-informed generative AI
a Probability densities of the values of 〈μ1, μ5〉 for experimental trajectories collected at electron beam dose rates of 20 and 35 e−/Å2 ⋅ s for particle size of 60 nm. b Probability densities of the values of μ4 for the same experimental trajectories as in (a). c Probability densities of the values of 〈μ1, μ5〉 for experimental trajectories collected at particle sizes of 40 nm and 60 nm at an electron beam dose rate of 35 e−/Å2 ⋅ s. d Probability densities of the values of μ4 for the same experimental trajectories as in (c). e Probability densities of the values of non-Gaussianity (averaged over the x and y components) at τ = 1 for the same experimental trajectories as in (a). f Probability densities of the values of velocity autocorrelation (averaged over the x and y components) at τ = 1 for the same experimental trajectories as in (b). g Probability densities of the values of non-Gaussianity at τ = 1 for the same experimental trajectories as in (c). h Probability densities of the values of velocity autocorrelation at τ = 1 for the same experimental trajectories as in (d). i UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (a) and (e), color-coded by dose rate. j UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (a), (e), and (i), color-coded by non-Gaussianity (τ = 1) on a symmetric logarithmic scale (SymLogNorm) k UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (a), (e), and (i), color-coded by velocity autocorrelation (τ = 1). l UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (c) and (g), color-coded by particle size. m UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (c), (g), and (l), color-coded by non-Gaussianity (τ = 1) on a symmetric logarithmic scale (SymLogNorm) n UMAP embeddings of all twelve latent variables for the same experimental trajectories as in (c), (g), and (l), color-coded by velocity autocorrelation (τ = 1).
