In an article published in Microscopy and Microanalysis recently (Jia et al., 2004), it was claimed that aberration-corrected high resolution transmission electron microscopy (HRTEM) allows the quantitative measurement of oxygen concentrations in ceramic materials with atomic resolution. Similar claims have recently appeared elsewhere, based on images obtained through aberration correction (Jia et al., 2003; Jia & Urban, 2004) or very high voltages (Zhang et al., 2003). Seeing oxygen columns is a significant achievement of great importance (Spence, 2003) that will doubtlessly allow some exciting new science; however, other models could provide a better explanation for some of the experimental data than variations in the oxygen concentration. Quantification of the oxygen concentrations was attempted by comparing experimental images with simulations in which the fractional occupancy in individual oxygen columns was reduced. The results were interpreted as representing nonstoichiometry within the bulk and at grain boundaries. This is plausible because previous studies have shown that grain boundaries can be nonstoichiometric (Kim et al., 2001), and it is indeed possible that oxygen vacancies are present at boundaries or in the bulk. However, is this the only possible interpretation? We show that for the thicknesses considered a better match to the images is obtained using a simple model of surface damage in which atoms are removed from the surface, which would usually be interpreted as surface damage or local thickness variation (from ion milling, for example).

This alternative interpretation is suggested by the observation that there are also changes to the other column intensities, which are at least as significant as the changes seen at the O columns, for example, Figure 7 of Jia et al. (2004) or Figure 1 of Jia and Urban (2004). As a specific example, consider Figures 1e and 2 of Jia et al. (2003). Those simulations reveal that the intensity measured at an oxygen-deficient oxygen column does indeed decrease, in agreement with the experimental images. However, the simulations also show that the neighboring Ti columns appear slightly brighter than in the bulk, an observation not supported by the experimental data. In fact, in some places the nearby Ti columns appear significantly dimmer. This is shown most clearly in Figure 2a of Jia et al. (2003).

We show that removing atoms from the surface explains both observations by a single mechanism. Figure 1a of this letter shows a simulation of oxygen vacancies, modeled via fractional occupancy, with parameters matching those of Jia et al. (2003), which agrees well with the simulation given in Figure 2c of that paper. Figure 1 also shows simulations for surface roughness modeled by a partial surface step (b), and a full surface step (c), as defined in (d) and the figure caption. From this figure it is apparent that reduced occupancies and surface roughness both introduce changes in the intensity of the oxygen columns. However, only the latter phenomenon introduces notable decreases in the intensities of the neighboring Ti columns. The nearby Ti columns appear slightly brighter near an O column with reduced occupancy and generally appear dimmer in the presence of surface damage. This is only intended to be a simple model for the effect of surface roughness to use in the simulation, and other models might be more appropriate, such as surface relaxations, adsorbates, or amorphous layers, which might be several tenths of a nanometer thick.

Linetrace simulations of the (HRTEM) image intensity for SrTiO3 [011] for the conditions given by Jia et al. (2003) (E = 200 kV, Cs = −0.04 mm, Z = +8 nm, aperture = 20 mrad, 4 nm thick, information limit 7.7 nm−1) for (a) fractional occupancy in two O columns as labeled, (b) partial surface step located above the brace, and (c) full surface step located above the brace. In all cases, the taller peaks represent the Ti columns and the smaller peaks are the O columns. The intensity values in stoichiometric SrTiO3 are shown with dotted lines as a visual aid. d: Projected potentials of the four even slices of the repeat distance. In a partial surface step, we omit the marked areas in slices 3 and 4 of the final repeat unit; in a full step, we omit the marked areas in all slices of the final repeat unit. The arrow points to the line from which the extracted scans are taken. e: Atom locations in the repeating unit are shown for slices 1–3 (slice 4 is identical in structure to slice 2).

Nevertheless, this simple surface roughness model clearly compares well with the experimental data in Figure 2a of Jia et al. (2003), suggesting that the observed intensity changes are better fitted by some form of surface roughness rather than oxygen vacancies in the bulk. There are many other possible models of surface damage effects from ion beam thinning or other sample preparation techniques. Preferential thinning at boundaries could, in some circumstances, concentrate these effects at grain boundaries. We would expect the presence of a thin surface layer on HRTEM specimens that would be reconstructed, amorphous, or damaged. There is also the possibility of electron beam damage or surface contamination that varies over the sample. These effects could be more random than the sharp step in our simulations, allowing a closer match to the observed intensity variation. Even if some technique were used to anneal the surface, we would still expect surface steps to be present similar to those we model.

Finally, Jia et al. (2004) echo the sentiment, stated explicitly by Jia and Urban (2004), that Z-contrast imaging in scanning transmission electron microscopy (STEM) “does not allow oxygen to be imaged because it is confined to elements with a high value of the nuclear charge Z.” This is a misconception because, although the oxygen columns could not be resolved in earlier work (Kim et al., 2001), there is no fundamental reason why this should always be so. In fact recent work has since shown that, following aberration correction, oxygen columns can be resolved in Z-contrast STEM, even in the presence of heavier columns (Chisholm et al., 2004). Further calculations (Cosgriff et al., 2005) suggest that column-by-column electron energy loss spectroscopy may allow quantitative analysis of a single column in a manner that is not so strongly dependent on the atomic number or surface damage.

In summary, we have shown that effects other than oxygen vacancies better account for contrast in recent experimental images. Unless these alternative explanations can be definitively excluded, there would seem to be no justification for making assertions about the oxygen concentration in the bulk. Given these results, we suggest that “seeing is believing” (Spence, 2003) is not necessarily true in this context.