Dynamic observation of manganese adatom mobility at perovskite oxide catalyst interfaces with water

Real time in-situ microscopy imaging of surface structure and atom dynamics of heterogeneous catalysts is an important step for understanding reaction mechanisms. Here, using in-situ environmental transmission electron microscopy (ETEM), we directly visualize surface atom dynamics at manganite perovskite catalyst surfaces for oxygen evolution reaction (OER), which are ≥20 times faster in water than in other ambients. Comparing (001) surfaces of La0.6Sr0.4MnO3 and Pr0.67Ca0.33MnO3 with similar initial manganese valence state and OER activity, but very different OER stability, allows us to distinguish between reversible surface adatom dynamics and irreversible surface defect chemical reactions. We observe enhanced reversible manganese adatom dynamics due to partial solvation in adsorbed water for the highly active and stable La0.6Sr0.4MnO3 system, suggesting that aspects of homogeneous catalysis must be included for understanding the OER mechanism in heterogeneous catalysis. Atomic processes on the surface of a catalyst control reaction mechanisms. Here, in-situ imaging of manganese adatom dynamics on La0.6Sr0.4MnO3 and Pr0.67Ca0.33MnO3 surfaces used in the oxygen evolution reaction reveals the importance of partial surface solvation of active metal sites on reaction mechanism.

T he efficiency of electrochemical water splitting for the sustainable production of clean fuels from renewable energy sources is limited by a large overpotential of the anodic oxygen evolution reaction (OER) [1][2][3] . Understanding of its origin requires knowledge about the pathway of the formation of di-oxygen O 2 out of 2 H 2 O molecules by a proton-coupled electron transfer. Even for the best electrocatalysts, i.e., transition metal (M) oxides (O), the reaction path for O-O formation is not well understood. For heterogeneous systems, the reaction is catalyzed at active surface sites, where different sites and mechanisms are suggested. This includes single 4 or dual [5][6][7] transition metal (M) sites with flexible valence states, concerted proton-electron transfer 8,9 as well as redox active lattice O 10-12, including oxygen vacancy formation and annihilation 13 . Despite remarkable progress in application of density functional theory (DFT) to the oxide-water interface, the theoretical understanding of the dynamic interface structure during OER is still limited since it requires serious approximations 14 . Among of the farreaching are the frozen surface approximation and that solvation effects are, if at all, only considered as a static average. This underestimates the flexibility of M surface coordination due to interface solvent effects that is on the other hand known to be crucial in homogeneous catalysis for the formation of high-or hetero-valent M-O moieties that precede O-O bond formation 15 . There are controversial conclusions whether the predicted unsuitable scaling relation between metal M-OH and M-O adsorption energies 9,16,17 as the origin of large overpotentials for single surface sites can be broken by bifunctional mechanism at two distinct M and M' sites 18 or is restored by surface solvation 19 . Recent experiments show that ternary alloying of two different transition metals with a highly flexible coordination metals indeed lead to a remarkable reduction of OER overpotential in gelled oxyhydroxides 20 . An improved theoretical understanding thus requires a comprehensive atomic scale understanding of the dynamic electrolyte-catalyst interface structure under OER conditions as an essential ingredient for the knowledge driven further improvement of electrocatalysts by material design.
In addition to OER activity, the stability is strongly affected by the interfacial electrolyte interactions. A distinction between reversible dynamics that emerges in a stationary state and irreversible processes that change the activity is essential. Such insights are promoted by the progress of operando studies that can address different types of dynamical changes. A first type of rather slow dynamics is the change of electrode surface during potential cycling. The important role of solvate interactions has been e.g., established for the noble metal Pt, where Pt surface oxide that is formed under anodic polarization adsorbs water that dissociates and forms hydrated Pt 21 . Metal oxides such as CoO x can form a disordered surface during OER 22,23 . A second type of rather fast dynamics is due to thermally activated mobility of surface atoms 24 or adatoms 25 that is hardly accessible in operando experiments. Clearly, surface adatom dynamics depends on temperature and surface barriers that can be modified due to H 2 O solvation or dissociation 26 and is always present, even in thermal equilibrium. Driving the OER, the emerging non-equilibrium surface should show a reversible dynamics in a stationary state that depends on the formed atomic and chemical surface structure. Non-equilibrium dynamics of catalyst atoms may be even involved in OER mechanism 27 . A third type of surface dynamics is irreversible defect reactions, typically the formation of oxygen vacancies 13,28 or the dissolution of metallic species 29 . Such processes are even gradually present at the highest performance system RuO 2 30 . Metal dissolution can depend on overpotential 31,32 surface symmetry 33 , pH value and particle size 34 . In order to avoid such corrosive processes, approaches for self-repair have been pursued 35 . Despite a huge progress in application of different operando techniques 36 , comprehensive atomic scale understanding of all three classes of surface dynamics are far of reach at present, but are urgently required to establish the intimate connections between surface stability, the nature of the active sites and reaction pathways 37 .
In situ environmental transmission electron microscopy (ETEM) in combination with aberration correction and image simulation has developed to a level where it can provide atomic resolution real space information about the equilibrium surface reconstruction in various ambient conditions [38][39][40] including strategies for controlling the impact of electron beam 41,42 . The recent development of atomic scale dynamic studies show surface mobility of Ce atoms on CeO 2 facets in high vacuum 43 and a transition of a TiO 2 surface reconstruction in H 2 O from static to dynamic after adding CO 44 . Steps towards the control of electron beam-induced and applied electric potentials at TEM samples enable in situ studies of electrode surfaces in ETEM under anodic polarization, approaching OER relevant potentials 28,45 .
In this study, two perovskite manganites AMnO 3 were selected, where the Mn valence state can be controlled by A-site doping without changing the crystal symmetry. Specifically, La 1-x Sr x MnO 3 at x = 0.4 (LSMO) is a metal-like electrode with high and stable oxygen evolution activity that is close to IrO 2 29 . Moreover, Pr 1-x Ca x MnO 3 at x = 0.33 (PCMO) is a semiconducting small polaron material with high initial OER activity, yet less stability, due to oxygen vacancy formation and fast leaching of the resulting reduced Mn species. Thus LSMO is suitable for the study of reversible surface dynamics, whereas PCMO shows a pronounced irreversible defect reaction. Both systems have almost the same initial Mn valence but differ in Mn-O covalency and charge localization. This differences allows us to correlate in situ ETEM studies of surface structure and dynamics with ex situ analysis of electrochemical OER. We find that the (001) surfaces of both manganites show an at least 20 times higher Mn-adatom mobility in an adsorbed water layer compared to other ambients.

Results and discussion
Structure, activity, and stability of LSMO and PCMO (001). Rotating ring disk electrode (RRDE) cyclovoltammetry (CV) was carried out at epitaxial (001) oriented thin film electrodes in 0.1 M aqueous KOH solution (Fig. 1). Both materials initially show good electrocatalytic oxygen evolution performance (for oxygen detection and Tafel analysis see Supplementary Figs. 1-3). Most striking is the different behavior just below the onset of the OER, where a redox wave is visible in PCMO as a negative current peak at E = 1.3 V vers reversible hydrogen electrode (RHE), complemented by a positive shoulder at the onset of OER. This feature is absent in LSMO. The redox couple is related to reversible formation and annihilation of oxygen vacancies, which are electron donors and thus induce a surface reduction of the transition metal 13 . The defect reaction is given by For neutral pH values in an unbuffered solution, which is more relevant for the comparison to in situ XANES and ETEM studies, a similar behavior of the CV is observed 13 . The easy formation of V ÁÁ o in PCMO indicates that lattice oxygen is redox active, such that transfer of holes between Mn3d and O2p states can oxidize O 2− to form O − and O 2 (g); see e.g., refs. 46,47 . For LSMO such a surface redox process is only observed when the cycling is extended to more negative potentials below E = 0.6 V vers RHE 48 .
Comparing the cycle dependence of the OER activity for LSMO and PCMO in Fig. 1b, there is a strong difference in stability between both electrodes. Initially, the current density j of both systems at E = 1.75 V vers. RHE is similar, i.e., j = 1.03 mA/cm 2 for LSMO and j = 1.05 mA/cm 2 for PCMO. These values are close to the activity of IrO 2 (110) films 49 . However, upon cycling and chronoamperometric OER between cycle 3 and 4, the decrease for PCMO is much more pronounced than for LSMO. Comparing the pristine and post electrolysis structure of LSMO and PCMO after 50 electrochemical cycles (Supplementary Figs. 4-7) reveals significant differences. Whereas for LSMO the crystalline surface structure with flat terraces and unit cell height steps is fully preserved, PCMO displays an increase of surface roughness. Postmortem high-resolution scanning transmission electron microscopy (HRSTEM) combined with electron energy loss spectroscopy (EELS) of film surfaces shows a strong surface depletion of Mn for PCMO, leading to a 2-3 nm thick Prenriched surface. Emerging small peaks in XRD after 1 h of electrolysis are consistent with a thin cubic PrO 2±x layer. In contrast, the Mn concentration at the LSMO surface is only slightly reduced.
Surface termination of the manganite films in H 2 O. The study of surface termination of epitaxial perovskite films using environmental high-resolution transmission electron microscopy (ETEM) was demonstrated by Roddatis et al. 42  The determination of surface termination requires the comparison of the HRTEM images with simulated images (Fig. 2d, h). In these simulations, the lamella thickness t at the surface is estimated from measurement of the inelastic scattering intensity by EELS and fine-tuned by a fit of contrast of simulated to experimental images using a Monte-Carlo-based simulated least-squares optimization, yielding t = 4.2 (4.3) nm for LSMO (PCMO). The obtained electron optical parameters, such as defocus, spherical aberration, astigmatism, and axial coma, fit reasonably well to those estimated from tilt-tableau measurements via the Cs-corrector software. The different surface terminations can be clearly distinguished by the weaker contrast of the MnO columns. In the subsurface, A-site and B-site columns are separable from contrast due to different atomic numbers and octahedral tilt patterns of MnO 6 octahedra. Details of the image fitting and parameters used can be found in the Supplementary Fig. 12 and Supplementary Table 1.
Reversible Mn-adatom dynamics at the LSMO-H 2 O interface. Figure 3 shows a time sequence of HRTEM images for the LSMO x = 0.4 (001) surface in 0.5 Pa of H 2 O. The full image series is compiled in Movie M01. The surface is well ordered and shows stable A-termination over more than 30 min. However, mobile Mn is, in fact, present on the surface. This is indicated by the sudden appearance and disappearance of atom contrast at B-layer surface positions on flat stable A layers (Fig. 3a). We define the detection limit of Mn atoms by the amplitude of the signal above the noise level 3σ, where σ is the standard deviation of the background fluctuations of the CCD signal at an area above the surface in the presence of the experimental gas. Image simulation and contrast quantification show that signal to noise ratio at the used frame rate of 4.4 fps does allow for detection of triple or higher occupancy of B-columns by Mn at a 4.2-nm thick lamella surface (Fig. 3b, c).
Closer inspection of image contrast in Fig. 3a reveal adatom contrast mainly appears at interstitial surface positions and rarely at the nominal octahedral positions of an ideal non-reconstructed surface. Contrast quantification of the columns of Mn adatoms is obtained by calibration of the experimental A-site contrast at the lamella surface to the simulated contrast, as shown in the Supplementary Figs. 13-15. Based on contrast simulations, the interpretation of dynamic adatom contrast other than Mn on a static A-layer is unlikely: O and OH species are below the contrast detection limit. La/Sr occupancy on B-terminated layers can be excluded from contrast. Fast La/Sr adatom dynamics on a stable A-terminated layer can be not entirely excluded from contrast. However, this would lead to an A-enriched surface with a Mn depleted layer, which is contrary to our observations for LSMO but in agreement to Mn surface leaching in PCMO reported below. Since all experiments in different environments reveal that surface Mn is more mobile than A-type cations, our overall conclusion from image contrast and electrochemical behavior is that Mn adatoms are the highly mobile species on stable LSMO surfaces in H 2 O.
The camera frame rate of around 4 fps is selected as a compromise between achieving sufficient signal to noise ratio and capturing some of the surface dynamics. Double Mn B-layer occupancy corresponds to the detection limit at 3σ of the noise level. Both triple Mn as well as higher Mn B-layer occupancy contrast appears and disappears with the frame rate used. We can thus only determine a lower limit of the surface hopping rate, which is given by the frame rate of the images. If triple Mn occupation appears and disappears on the timescale of the frame rate, it may be attributed to one of the following events: (i) A single Mn atom hops to another site and the remaining double occupation contrast becomes invisible. (ii) Two or three Mn atoms coincidentally hop to another site and the B-site contrast on the original site disappears. Since (ii) has a lower probability  Lamella thickness is 4.2 nm. c Line profiles of simulated (red) and experimental (black with symbols) images of B-columns with 3, 4, and 5 Mn atom occupation. The 3σ threshold indicates the detection limit of Mn columns above noise level, i.e., triple Mn occupation at a B-layer column. The time sequence gives a lower limit of the Mn-adatom hopping rate of r ≥ 4 s −1 . than (i) and represents the slowest process, it defines the minimum detectable hopping rate of r ≥ 4 s −1 .
The dynamic state with highly mobile Mn atoms on top of a more or less static A-terminated layer can be strongly affected by the presence of a surface step. Since the Mn surface mobility depends on the surface activation barrier, it can be strongly reduced at step edges due to an Erich-Schwöbel barrier.  Fig. 20). Furthermore, the Mn valence state remains quite stable, indicating only a slight Mn reduction in increasing H 2 O pressure and ruling out pronounced formation of oxygen vacancies.
Irreversible Mn-adatom dynamics at the PCMO-H 2 O interface. Figure 5 shows a time series of HRTEM images of the (001) PCMO surface in 0.5 Pa H 2 O, taken from Movie M06. After fully ordering in O 2 (Fig. 2f), the addition of H 2 O causes the surface to form a disordered layer of highly mobile Mn atoms, visible as rapid fluctuation of B-site contrast. Similar to LSMO, the hopping rate of Mn atoms is above the frame rate of the camera, i.e., r ≤ 4 s −1 . In contrast to the disordered B-terminated surface, the A-subsurface layer remains highly ordered and static on the timescale of several seconds (Movie M06). In addition, an increase of contrast dynamics is observed at the B-site columns at the subsurface. Figure 5b In contrast, the Mn concentration at LSMO surfaces remains unmodified. This is fully consistent with the observation of different electrode behavior in electrochemical experiments described above.
Enhanced Mn mobility at (001) surface due to presence of adsorbed H 2 O layer. The formation of a few monolayer thick water layer on oxide surfaces down to pressures of~10 −5 Pa much below the equilibrium vapor pressure of water of p vap = 3.17 kPa at T = 25°C has been previously demonstrated [50][51][52] . It is driven by the large adsorption enthalpy of H 2 O on oxide surfaces of the order of ΔH ad,ox -160 kJ/mol 51,53 . Although this value might vary between different oxides, ΔH ad,ox < ΔH ad = −44 kJ/mol for H 2 O on bulk water reflects exothermic water adsorption on oxide surface at much reduced pressures. In the HV of the ETEM (p ≥ 10 −5 Pa), the effect of H 2 O condensation on manganite surfaces can be only avoided by using a cold trap that is reducing the partial pressure of H 2 O in HV by one order of magnitude 42 . The thickness of the adsorbed water layer in the ETEM experiments as a function of p H 2 O is calculated by using the energy balance between surface energy γ s of the oxide in HV, the surface energy γ H 2 O of liquid H 2 O to vapor and ΔH ad (see Supplementary Fig. 22 and associated text). The liquid state of 2-4 monolayer H 2 O is consistent with our observation of enhanced mobility of Mn on manganite (001) surfaces in a pressure range between 0.01 and 5 Pa of H 2 O compared to HV.
Thermally induced versus electron beam-induced hopping. In order to separate out the effect of the electron beam on adatom motion from the thermally induced motion, the momentum transfer from scattering of the high energy primary electron beam on the adatoms is calculated. As pointed out by Egerton 54 , for energies of primary electrons E 0 ≥ 10 keV, the interaction is dominated by elastic scattering of the incident electrons at the atomic nuclei, and momentum transfer due to inelastic scattering at the electron shell can be disregarded. This allows calculation of the momentum transfer using the Rutherford scattering equation. We consider the maximum momentum transfer perpendicular to the primary electron beam, which gives rise to transversal movement of adatoms on a surface parallel to the electron beam. Figure 6 compares the beam-induced hopping rates for O, Mn, Sr and La adatoms to the thermally activated hopping rate as a function of the surface activation energy E a for an upper limit of used dose rates of 12.000 e − /Å 2 s (see method section). For E a ≥ 0.2 eV, the calculated beam-induced hopping rate for Mn adatoms is below the experimentally observed hopping rate in HV, indicating that the slow Mn motion is thermally activated.  LSMO (001) do not show surface loss of oxygen 29 . As summarized in Fig. 7, these observations suggest that stabilization of high oxidation states of surface Mn is essential for a high reversibility of Mn-adatom movement and thus for the stability on the surface. Surface reduction and Mn 2+ formation lead to irreversible Mn surface dynamics due to the higher solubility of Mn 2+ in H 2 O compared to Mn 3+ . Leaching of Mn 2+ in H 2 O is also predicted by a Pourbaix diagram of binary Mn-O compounds.
We attribute the different behavior of LSMO and PCMO to their different electronic properties, in particular covalency and charge localization (see Supplementary Figs. 23 and 24). Metallic LSMO exhibits spatially delocalized charge carriers, so called large polarons due to a larger Mn-O covalency and an electronphonon coupling constant α ≈ 1 56 . As long as the metallic state is preserved at the electrode surface, an extra charge acquired from water oxidation cannot change the valence state of a single atomic surface species. In contrast, PCMO x = 0.33 displays localized charge carriers, so called small polarons due a smaller Mn-O covalency and larger α ≈ 3 (see ref. 57  Note that the different charge carriers may be also the origin of the different Tafel slopes b of both manganites. For LSMO, b ≈ 60 mV/dec fits to the theoretical prediction for a chemical rate-limiting step that follows an electrochemical pre-equilibrium, probably O-O bond formation. For PCMO, the higher b ≈ 90-120 mV/dec is consistent with electron transfer as the rate-limiting step and thus may reflect the lower intrinsic electric conductivity as well as limitations due to surface reduction.
Solvated dynamic Mn adatoms and impact on reaction mechanisms of oxygen evolution. The demonstrated parallel trends in in situ ETEM and electrolysis studies underpins that the observation of enhanced surface dynamics of Mn adatoms in H 2 O is highly relevant for the real-world electrochemical behavior of the manganite electrodes. Clearly, there are differences in the properties of a Helmholtz layer in a liquid electrolyte with ion concentration of 10 −1 M and the thin water layer adsorbed on top of oxide materials in ETEM. In particular, the ion concentration is much lower, with typically 10 −6 -10 −7 M in the vapor 45 that is close to the ion concentration at neutral pH. However, the composition of ions generated by electron impact is different since positively charged ions such as H 2 O + dominate over anions such as OH − . Our ETEM studies are performed under positive electric potentials of U = +2 V with respect to ground and thus under anodic polarization. Due to the high electric conductivity of the manganite lamella on highly conductive Nb-doped SrTiO 3 , beam-induced potentials can be disregarded 45 . The comparison of hopping rates in different ambients clearly demonstrate that the observed ≥20× increased Mn-adatom mobility in H 2 O while other conditions of the experiment remain the same is not due to the electron beam. This is fully understandable from the calculation of beam-induced momentum transfer in Fig. 6. Therefore, the enhancement of thermally activated adatom motion in water is fully relevant for OER catalysis in ex situ electrochemical conditions. It can be attributed to a partial solvation of surface Mn in liquid H 2 O. Such solvation effects thus do not only affect the adsorption energies of the reactants but also strongly modify the coordination and electronic properties of active sites.
The formation of a partial solvation shell of H 2 O around chemisorbed Mn adatoms in OER thus has far-reaching consequences for understanding of the reaction pathways. First  of all, Mn can develop other, more flexible coordination to OH 2 and OH compared to a static octahedral Mn-O coordination of a frozen surface (Fig. 7). This would allow new configuration for the formation of the O-O double bond. Furthermore, the question is raised whether the Mn dynamics affect the OER. The electron transfer rate per surface Mn atom derived from RRDE measurements is between 0.2 s −1 (1.6 V) and 10 s −1 (1.74 V vers RHE), assuming a Faraday efficiency of 1 and a complete Mn surface coverage. Due to partial coverage, these numbers represent a lower limit of the electron transfer rate per Mn. Comparing these numbers with the observed lower limit of Mn-adatom hopping r ≈ 4 s −1 , shows that both are of the same order of magnitude or even higher. Consequently, Mn can move over several surface sites during a full O 2 evolution cycle. In summary, our study reveals unexpectedly high Mn-adatom hoping rates implying that the understanding of OER mechanisms requires theoretical treatment beyond the frozen surface approximation. There is strong experimental evidence from ref. 21 that surface Mn species represent the catalytically active sites for LSMO. Fast Mn leaching at PCMO surfaces dramatically decreases OER activity. Although hopping rates and contrast of single Mn atoms are beyond the resolution of our experiments, the observation of dynamic multiple occupancy of surface sites by Mn moieties establishes a pronounced dynamical surface reconstruction of the catalyst-water interface with partial solvation of surface Mn atoms in H 2 O. Mn-adatom contrast dominantly appears at interstitial surface sites rather than the nominal octahedral B sites of an unreconstructed ideal surface, suggesting a modified surface coordination. The implications of these findings are quite far-reaching. First of all, leaching of reduced transition metal species is observed for many transition metal oxides 61,62 . Our results imply that the underlying transition from reversible dynamics of surface atoms to irreversible leaching of surface and subsurface metal species is controlled by charge localization and a related shift of the Fermi energy. Indeed, the recent results on self-gating of electrocatalysts evidences a modification of the Fermi level at the onset of an electrochemical reaction 63 . The finding of pronounced reversible dynamics of partially solvated adatoms in the stationary state of OER implies that for the theoretical understanding of heterogenous catalysis some aspects of homogeneous catalysis must be taken into account, such as modified metal coordination, different valence states, solvation shell reorganization during electron transfer, solvent thermal fluctuations and possibly Mn displacements during intermediate formation. Incorporating these effects into theoretical analysis will open new strategies of breaking scaling relations between adsorption energies and thus new perspectives in atomic scale design of efficient and stable electrode surfaces for OER. Backside contacts for the STNO substrate consisted of 5 nm Ti and 100 nm Pt layers prepared by IBS at room temperature to ensure an ohmic contact. Carbon tape and InGa eutectic (Sigma-Aldrich, Munich, Germany, 99.99%) served as the flexible and adjustable conductive spacer. In the final assembly, only the thin film surface is exposed to the electrolyte after fixation with chemically stable, nonconductive epoxy (Omegabond 101).

Methods
Electrochemistry. The electrochemical measurements were carried out with two Interface 1000E potentiostats (Gamry Instruments Inc., Warminster, PA, USA) in a bipotentiostat setup and an RRDE-3A rotator (ALS Co. Ltd., Tokyo, Japan). The rotating ring electrode consists of a disk electrode of the assembled PCMO|LSMO| NSTO electrode with a diameter of 4 mm and a Pt ring electrode with an inner diameter of 5 mm and an outer diameter of 7 mm. All electrochemical measurements were performed in 0.1 M KOH electrolyte prepared by diluting KOH stock solution (Sigma-Aldrich, Munich, Germany) with deionized water (MilliQ, >16.5 MΩ). The electrolyte was saturated with Ar gas at least 30 min before measurements and continuously purged with Ar throughout the measurement. Electrode potentials were converted to the reversible hydrogen electrode (RHE) scale using E RHE = E applied + E ref , where E ref = 0.993 V vs. SCE was obtained from the hydrogen evolution using a clean Pt disk. Cyclovoltagramms are measured at 10 mV/s with rotation of 1600 turns per minute. In Fig. 1b, between third and  were performed at a beam current of 4 nA and beam radius between 80 and 100 nm, resulting in a spatially in homogeneous electron dose rate of 8000-12,000 e/Å −2 s. Local electron dose rates at the location of TEM lamella surfaces are measured by calibrated CCD contrast with 0.136 electrons/ counts, yielding~10,000 e/Å −2 s for the adatom hopping studies. This dose rate is two orders of magnitude below the damage threshold in HV. Imaging over a time of 10 min keeps the total electron dose ≤6 10 3 As/cm 2 , where no changes by electron irradiation are observed in HV, in good agreement with the studies of the intrinsic surface in ref. 41 .
Electron energy loss spectroscopy (EELS). Was performed using Gatan Quantum 965ER post-column energy filter in the same FEI Titan ETEM. Spectra of the Mn L, O K, and Pr M edges were acquired. For analysis of the Mn L edge spectra, Hartree-Slater (HS) type cross-section functions from Gatan's Digital Micrograph are used. The Mn valence state has been acquired from a C++ based LabTalk script using OriginLab software. EEL spectra are extracted from spectrum images at an area of vacuum and at different H 2 O vapor pressure for postmortem analysis. Power-law background functions are fitted to a 50 eV wide window before each Mn L-edge and 25 eV for O K-edge for background subtraction. By using dual EELS acquisition, the ZLP is used to correct the drift of the energy shift.
TEM lamella preparation. TEM lamellae were prepared from epitaxially grown (001) oriented thin films by Focused Ion-Beam lift-out technique using a noble metal free protection layer and careful low energy ion milling for the preparation of ultrathin lamella with a thickness close to the surface between 4 and 10 nm. The loss of oxygen by Ar milling is compensated by a treatment in pure oxygen. Further details are described in the Supplementary information and Movies M09-M10.
Image simulations. Atomic models and the high-resolution TEM simulated images were calculated using the multislice method as established in the QSTEM package 64 , where electron optical parameters were optimized by using a metropolis algorithm implementation to fit simulated to experimental images. Structure models for the image simulations are based on the Pbnm spacegroup for PCMO with lattice parameters a = 5.42 Å, b = 5.45 Å and c = 7.67 Å and R3′c for LSMO using a c = 3.87 Å. Details and parameters used for image simulation are given in the SI.
Calculation of electron beam-induced and thermally activated hopping. Using the analytical solution given in ref. 54 , the adatom hopping rate is where ϑ is the scattering angle of the electrons. The limits ϑ min ¼ arcsin 2 E s E max 0:5 " # and ϑ max ¼ π À arcsin 2 E s E max 0:5 " # ð2Þ are given by ratio of the captured surface energy E s of the adatom to the maximum possible energy transfer E max , which would correspond to a 180°scattering, i.e., A is the atomic mass, Z the nuclear charge, e the elementary electron charge, J is the current density of the primary electron beam, σ 0 = 1 barn = 10 −28 m 2 and where c is the velocity of light and v the velocity of the primary electrons. In order to calculate the hopping rate of an adatom on a surface, it is assumed that the kinetic energy transversal to the surface is equal to the surface activation energy E a , i.e., E s = E a . In the Mn-adatom hopping studies, E 0 = 300 keV and for the current density an upper limit of the experimental values of J = 2.0 10 5 A/m 2 is chosen, corresponding to 12,000 e/Å 2 s.
The thermally activated hopping rate is given by with adatom vibration frequency v 0 and Boltzmann constant k B . We have chosen v 0 = 10 12 s −1 , which for E a = 0 eV results in a thermal velocity for Mn atoms matching the maximum of the Maxwell-Boltzmann distribution.

Data availability
All relevant data are available from the corresponding authors on request.