Post-processing paths for orbital mapping of rutile by STEM-EELS

Recently, it was shown that scanning transmission electron microscopy (STEM) in combination with electron energy loss spectroscopy (EELS) allows for a real-space mapping of atomic orbitals [1]. Although state of the art electron microscopes offer the required spatial- and energy resolution, the inherently poor signal-to-noise ratio (SNR) for such experiments imposes a major challenge, which necessitates the development and application of advanced post-processing procedures.

To overcome the problems with low SNR EELS data, often multivariate analysis techniques, such as principal component analysis (PCA) are used. For high noise, however, PCA introduces artifacts or even removes faint fine structures, which prevents mapping orbital signatures directly from the raw data. As a remedy, reference dark field images can be recorded simultaneously with the EELS signal. Containing high-spatial resolution information, these references can be used to faithfully stack and re-align multiple cells and to average the corresponding electron energy-loss spectra to a signal level sufficient for PCA denoising.
New generation EELS sensors, based on direct electron detection, intrinsically offer higher SNR through their much-improved detective quantum efficiency. Noise is mostly governed by the shot noise contribution (Poisson noise), which enables the application of weighted PCA optimized to such Poisson noise dominated data.

Figure 1 exemplifies a possible procedure that is capable of imaging orbitals by mapping the eg-like states of the titanium L2 ionization edge in rutile [001]. In this case, a 1.3 eV energy window was chosen to map the states after background subtraction. Our experimental results rather well match accompanying multi-slice calculations based on mixed dynamic form factors obtained from density functional theory simulations [1].