Correlative In Situ Spectro-Microscopy of Supported Single CuO Nanoparticles: Unveiling the Relationships between Morphology and Chemical State during Thermal Reduction

The activity, selectivity, and lifetime of nanocatalysts critically depend on parameters such as their morphology, support, chemical composition, and oxidation state. Thus, correlating these parameters with their final catalytic properties is essential. However, heterogeneity across nanoparticles (NPs) is generally expected. Moreover, their nature can also change during catalytic reactions. Therefore, investigating these catalysts in situ at the single-particle level provides insights into how these tunable parameters affect their efficiency. To unravel this question, we applied spectro-microscopy to investigate the thermal reduction of SiO2-supported copper oxide NPs in ultrahigh vacuum. Copper was selected since its oxidation state and morphological transformations strongly impact the product selectivity of many catalytic reactions. Here, the evolution of the NPs’ chemical state was monitored in situ during annealing and correlated with their morphology in situ. A reaction front was observed during the reduction of CuO to Cu2O. From the temperature dependence of this front, the activation energy was extracted. Two parameters were found to strongly influence the NP reduction: the initial nanoparticle size and the chemical state of the SiO2. substrate. The CuOx reduction was found to be completed first on smaller NPs and was also favored over partially reduced SiOx regions that resulted from X-ray beam irradiation. This methodology with single-particle level spectro-microscopy resolution provides a way of isolating the influence of diverse morphologic, electronic, and chemical influences on a chemical reaction. The knowledge gained is crucial for the future design of more complex multimetallic catalytic systems.


Temp(K)
Table S1.Contrast stretching details of Figure 2. Because no signs of Cu2O were observed in the first two annealing steps, only the green channel was used to make the XPEEM image.Figure S6.Normalization of the NEXAFS spectra.To use the intensity of the NEXAFS spectra as a parameter that quantifies the content of each species, we need to take into account the differences in the overall signal across different NPs.These arise from the beam intensity not being homogeneous across the field of view (FoV).In other words, the photon flux is different for the middle of the image than the sides, for instance.Normalizing by dividing the spectra by the pre-edge intensity (red-dottedline) removes this effect.From all spectra a factor 1 was subtracted to leave them spectra starting at y = 0, for practicality purposes.   =  (+)  (+) +  (2+)  (e 1)  (2+)  (+)  =  (+) +  (2+)    (2+) ~(2+)  (+)  Figure S10.CuO signal, and dividing by the sum of the intensities at both CuO and Cu2O edges.If this value v is more than 95% we considered that no conversion happened, and considered the peak intensity at this annealing step valid for the calculation of the maximum   (+) .In a) the spectra obtained at RT, and after 420K, show no Cu2O, therefore are counted for the calculation of   (+) .The spectrum after 480K, had a value of 0.87, which can be observed in a larger than noise peak appearance at the Cu2O region (933.8eV).Therefore, this spectrum didn't count for   (+) , since it also contains a Cu2O contribution.Analogously, a full conversion is defined when v is lower than 5%.As can be seen, the normalized intensity of a given NP fluctuates little across the different annealing steps, which is a further indication that the NEXAFS intensity is correlated with the morphology of this NP.In fact, when taking in account only the NPs which did not convert in the first stages, 37 NPs, the average of the standard deviation of those is equal to 0.07.This low value shows that the intensity of the NEXAFS edge remains fairly constant across the annealing steps for a given NP.b), and c) show how the intensities of CuO (  (+) ), and Cu2O(  (2+) ) are obtained.First, the peak position is determined for every NP.Then, the mode is calculated, which is necessary, because not all the NPs have the same peak position, due to a gradient in photon energy in the same FoV (3.11 µm) of the order of 1 photon-step-size (x axis).This means that the spectrum is shifted in the x axis when you compare a NP in the very extreme right of the FoV to the one in the other extreme.After the mode is determined (marked as a green dot), we check for the neighbor dots (red dots), inside 1 photon-step-size range.We average the value of the mode, with the highest intensity neighbor dot.This serves to counter the photon energy gradient effect, and the general noise expected in measuring a single NP spectra.While the intensity of CuO(  (+) ) is averaged between the intensity of up to the first three annealing steps, the c) Cu2O(  (2+) ) is obtained from the last annealing stop only (after 650 K), where most NPs achieved full reduction.The  value for a single NP is finally calculated by dividing the   (+) by the   (2+) .We than average these values for the 50 NPs which had a full conversion, and the final  was 0.84.
(eq 4)  =  (+)  (+) +   (2+)  ( 5)     The NP can be divided in any number of points, defined by the user, in this case 3000 for a), and 30000 for b).The higher the number, the more statistically sound is the final calculation.The schematic in a) presents the two different options to distribute the points (infinitesimals), either with a quasi-random (Sobol) or random simulation (Monte-Carlo).The plot in b) shows that the program can also calculate in 3D space.For easy of visualization, the calculations in the rest of this paper are in 2D, using a Sobol simulation with 500000 points.In the 2D case, the z axis is perpendicular to the substrate, in other words, normal to the surface, or parallel to the detector axis.The data in b) are displayed with an equal ratio, therefore y, and z axis also are showing -1.0 to 1.0 length in radius.
Table S2.Description of the bins used to group the NPs by size.The mean diameter D is the mean value of every NP diameter inside a given bin, and not the proper center of the bin.We opted to group the last two bigger NPs in a single bin to increase the reliability of the reading, given that a single NP is prone to display big errors.Every size is given in nm.
(+) +  (+)  (2+)  (2+)   (eq 7) Figure S17.Geometry influence on the front velocity calculations.While the front velocity can be easily calculated in the case of nanocubes, given the constant front area in a), for nanospheres the calculation is more complex, because the area of the front depends not only on the NP size, but also in the conversion f.This area can be calculated either through simulation (such as using our program), or through an approximation.
Figure S18.Ratio V/A dependence on f.In red the analytic fitting function, that is within 2% the ratio V/A.
. Effect of X-ray exposure in different conditions.In the initial minutes in UHV pressure, a) the insertion device (ID) aperture, measured in mm, is at the lowest size with an estimated X-ray flux of 1%, and the Si chemical species (Si and SiOx) are stable, when opening the ID, thus increasing the flux, the SiOx starts to decay, and the signal of Si increases with time, with a clear dependence with the total X-ray flux (a,b,c,d,g).When e) dosing O2 in the 10 -7 mbar pressure with an ID = 0.2 mm, the silica reduction stopped.However, by further f) opening the ID, the reduction continued, while the O2 pressure (10 -7 mbar) was maintained.The reduction process persisted when returning to g) UHV pressure.h) Shows the XPS at h = 630 eV of the Si 2p line, normalized by the Si peak, and i) shows an estimative for the total X-ray flux for each ID value.Flat surface (object height:width < 1:5)

Sample environment
Pressure range up to 1 bar (gas) or aqueous environment (cells with windows) Pressure range up to 1 bar (gas) or aqueous environment (cells with windows) Pressure range up to 1 bar (gas) or aqueous environment (cell with windows) Pressure range from 10 -10 mbar up to 10 -5 mbar Table S3.Comparison of LEEM/XPEEM with other spectro-microscopes.The strengths of LEEM/XPEEM are the high surface sensitivity, lower beam damage than the other microscopes and the accessibility of XPS core levels of catalytically relevant elements such as Cu, Pd, Fe, Ni, O, C and N, and their correspondent L, or K-edges for NEXAFS.Also, the standard base pressure of 10 -10 mbar allows for the study of outermost clean sample surfaces.The strength of TEM, SEM and STXM is the possibility to study samples in gaseous and liquid environment, because special reaction cells with windows for the probing and imaged beam can be used due to the high mean free path length of the used electrons and/or light.Furthermore, TEM and SEM excel in outermost spatial resolution even down to the atomic scale.

Figure
Figure S1.XPS survey after O2 annealing treatment.This treatment removed the adventitious carbon and any residual polymer from the NP synthesis.The photon energy used was 400 eV which provides a surface-sensitive probing of the sample.C 1s and Cu 3p regions are shown with rescaled intensity in the inset.

Figure S2 .
Figure S2.Histogram and contrast stretching of the same raw data.i) are data displayed in Figure 2.

Figure S3 .
Figure S3.Extended view of the XPEEM image after the 593 K annealing treatment.The scale bar is 200 nm.

Figure
Figure S4.XPEEM-intensity dependence on NPs size.The NPs XPEEM intensity is taken before the annealing steps in UHV; the NEXAFS peak energy is at 931 eV.The intensity is corrected by subtracting the pre-edge baseline.The bigger the NP, the higher is their XPEEM intensity, with the Pearson correlation coefficient equal to 0.57

Figure
Figure S7.Average NEXAFS signal of every NP.Each spectrum shown is an average, derived from aggregating the individual NEXAFS spectrum of each of the NPs visible in the field of view.All spectra were generated from the same NP set, in other words, the same region, taken after each annealing step.Notice that this is different from an integral spectrum, where the entire image would contribute to the signal, here only the NPs are responsible for the signal.

Figure S8 .
Figure S8.The progression of the converted fraction f against the NP diameter D measured in LEEM.The complete set of annealing temperatures of Figure 3c.

Figure S9 .
Figure S9.Estimation of  factor.Spectra showing a NP which had a full conversion from CuO to Cu2O after the annealing steps.The peak intensity of the NEXAFS spectra decreased after the conversion, signalizing the need for a weighting factor which ensures that the weighted sum of the two peak intensities is constant during the conversion and equal to the final intensity.

Figure S11 .Figure
Figure S11.Reaction front on single Cu NPs.Composite XPEEM images were taken after cooling down from the 650 K annealing step, at two different photon energies, near the CuO L3 edge (green) and Cu2O L3 edge (magenta).Each image in the composite has a 20 s acquisition time.For this data, no correction by the pre-edge was performed.A consistent gradient in the oxidation state is observed, strongly indicating the presence of a reaction front.Different NPs have gradients in different directions.A white profile line with a length of 250 nm traced in the same position for each image shows b) three almost similar curves for each energy, highlighting how this gradient is consistent and not a result of experimental artifacts.

Figure S13 .
Figure S13.Front direction influence on the f value.The direction that the front propagates influence the measured f converted volume.The theoretical f value (fTh), which is the percentage of the geometric volume converted from the first phase (green, CuO) to the second phase (magenta, Cu2O) is fixed in these images, 25%.Depending if the front propagates parallel to the substrate, or perpendicularly, the measured f values are distinct.However, averaging two opposite moving fronts, can mitigate the deviation of f from fTh.This highlights the importance of binning, or averaging, a significant amount of NPs of the same height.

Figure S14 .
Figure S14.Information depth for different NPs.a) small NP in AFM, b) average, and c) big.We simulated the signal coming from the center of a NP, represented by the red line, more precisely between -0.1 and 0.1 x (r).The graphs on the left show the z axis of the NP in terms of radius versus the cumulative intensity.The red point represents the information depth, the depth where you have 95% of the total signal of your sample.The figure shows that for the same inelastic mean free path, but different NPs heights, you have different surface sensitivity across the NPs.While in smaller NPs a), the signal comes almost from the entirely NP, so bulk sensitive, in the bigger NPs, c) most of the signal originates from the surface.

Figure S15 .
Figure S15.Interface of the simulation program.Parameters such as IMFP, and front direction can be chosen to simulate the f (simulated detected converted fraction), and fTh (Theoretical/geometric converted fraction) values.

Figure S16 .
Figure S16.Simulation options.The NP can be divided in any number of points, defined by the user, in this case 3000 for a), and 30000 for b).The higher the number, the more statistically sound is the final calculation.The schematic in a) presents the two different options to distribute the points (infinitesimals), either with a quasi-random (Sobol) or random simulation (Monte-Carlo).The plot in b) shows that the program can also calculate in 3D space.For easy of visualization, the calculations in the rest of this paper are in 2D, using a Sobol simulation with 500000 points.In the 2D case, the z axis is perpendicular to the substrate, in other words, normal to the surface, or parallel to the detector axis.The data in b) are displayed with an equal ratio, therefore y, and z axis also are showing -1.0 to 1.0 length in radius.

Figure S20 .
Figure S20.Average f ratios for NPs binned by their size.The complete data, including the standard error, of Figure 4a.