Direct Atomic‐Level Imaging of Zeolites: Oxygen, Sodium in Na‐LTA and Iron in Fe‐MFI

Abstract Zeolites are becoming more versatile in their chemical functions through rational design of their frameworks. Therefore, direct imaging of all atoms at the atomic scale, basic units (Si, Al, and O), heteroatoms in the framework, and extra‐framework cations, is needed. TEM provides local information at the atomic level, but the serious problem of electron‐beam damage needs to be overcome. Herein, all framework atoms, including oxygen and most of the extra‐framework Na cations, are successfully observed in one of the most electron‐beam‐sensitive and lowest framework density zeolites, Na‐LTA. Zeolite performance, for instance in catalysis, is highly dependent on the location of incorporated heteroatoms. Fe single atomic sites in the MFI framework have been imaged for the first time. The approach presented here, combining image analysis, electron diffraction, and DFT calculations, can provide essential structural keys for tuning catalytically active sites at the atomic level.

The samples were synthesized in fluoride medium to avoid to the maximum any such Si-dislocations. To avoid the presence of any extra-framework Fe-species several acidic washing steps were undertaken to remove them at best.

Electron microscopy observation
STEM data were collected at 300 kV with the recently installed JEM-ARM300F microscope equipped with cold FEG and double C s correctors at the Centre for High-resolution Electron Microscopy (CℏEM), ShanghaiTech University. Prior any observations the STEM corrector for the electron probe was corrected using amorphous carbon, assuring a spatial point resolution of 0.7 Å.
The samples were prepared by deeply crushing a small amount using mortar and pestle for several minutes. The powder was then dispersed in ethanol to create a suspension with the finely divided crystallites. Few drops of the suspension were placed onto holey carbon copper grids.
Before analyses, the beam current was carefully measured using a faraday cup installed at the projection screen.
For both Na-LTA and Fe-MFI, ABF conditions were satisfied by introducing a dedicated ABF aperture and by using the BF detector as a data collector. For ADF in both cases the inner collection angle was 30 mrad and the hybrid ADF detector from JEOL was used.
For all data 1024 × 1024 pixel images were recorded and exposure time was varied from 3µs up to 12 µs not exceeding an electron dose of ≈ 3000 e -/Å 2 for LTA and ≈ 4000 e -/Å 2 .
For every case the convergence semiangle selected was 17 mrad.
EM simulations were performed using QSTEM software [1] based on the multislice method. For the LTA a supercell was built with the parameters 24.55 × 24.55 × 294.66 Å 3 while for the MFI structure the unit cell was 39.886 x 26.048 x 198.710 Å 3 . Experimental data was treated by using Digital Micrograph software and CRISP software [2] . Selected area electron diffraction (SAED) and 3D ED tomography (3D-EDT) of Na-LTA and Fe-MFI were conducted using JEM-2100Plus in TEM mode operated at 200kV. The angle range varied between -60.4° to 62.2° using a step size 0.2°. Data collection was acquired through EDT-Collect and processed using EDT-Process software.

FTIR (DRIFTS).
Infrared spectra were collected in DRIFTS mode using 2-5 mg sample and a modified Harrick DRIFTS cell with KBr windows. Spectra were measured in the 4000-1000 cm −1 range with a 4 cm −1 resolution and averaging 250 scans using a Nicolet Magna-IR 560 spectrometer equipped with a MCT detector (Thermo Nicolet, Waltham, MA, USA). Samples were dried at 523 K under flowing dry nitrogen (10 mL min −1 ) for 30 min. Thereafter, the collection of spectra started at 523 K. All spectra were normalized by the intensity of the Si-O-Si overtones (2100-1750 cm −1 ). IR data were treated with Opus software and finally plotted using Origin software.

Figure S1
Deduction of the plane group symmetry  Point group symmetry deduced p4mm Figure S2. C s -STEM ADF and ABF images of Na-LTA of [001] incidence. a) and b) Raw experimental ADF and ABF images, respectively. c) and d) Simulated ADF and ABF images that take into account similar electron dose of 3000 e -/Å 2 .
In general, simulated EM images tend to have much better quality (even for the same experimental conditions) than those obtained in the electron microscope. This is because factors such as scanning distortion, a small miss-tilt of the crystal, beam damage and the noise introduced by low doses are commonly neglected. Figure S4a and S4b show the experimental raw ADF and ABF images. In order to make a closer comparison between experimental and simulated data, the dose of electrons per area (3000 e -/Å 2 ) was introduced into the simulated ADF and ABF images ( Figure S4c and S4d). Although the SNR is very low, this analysis illustrates how much information can be extracted even from highly noisy data, indicating that it is possible to obtain atomic resolution information even at such low dose conditions. Further analysis that corroborates this result is presented in Figure S3, where the same image analyses carried out for the experimental micrographs was also applied to the simulated (noised) images. Noised images together with the Fourier filter and symmetry averaged images are shown for both detectors in Figure S3.

Image analysis and image simulation
EM image simulations were performed using QSTEM software. [1] Experimental data were treated by using Digital Micrograph software and CRISP software. [2] For Na-LTA, a supercell of 24.55 Å × 24.55 Å × 294.66 Å was created. Simulation parameters were set the same as the experimental ones, keeping the spherical aberration below 1 micron. Figure S2 compares the annular dark field (ADF) experimental data for one unit cell (24.55 Å) ( Figure S2a), with the simulated data ( Figure S2b, for a given thickness of 20 nm). An experimental ABF image is presented in Figure S2c, while the simulated one appears in Figure  S2d. The low amount of electrons used in the experimental and to simulate the data, gave as a result very noisy images that, however; still contain lots of information. Figure S3 show the ADF and ABF images of several unit cells obtained by introducing the same electron dose as the experimental conditions. After Fourier filtering, as it can be clearly appreciated, the amount of information was significantly enhanced. Symmetry averaging analysis provided a much higher degree of information which allowed the visualization of the T atoms and the Na at the 6Rs, and of all atoms, including O bridges and different Na cations in the ABF data. For image simulations the model constructed was based on the previous work by Pluth & Smith [3] , thus Na cations at coordinates 0.0996, 0.0996, 0.0996 with an occupancy of 1 as well as Na cations with ¼ of occupancy at the 8Rs were incorporated. The differences in the treatments for increasing the visibility of the original ABF image are provided here. Each treatment for original image, two filter methods, Fourier-and Wienerfiltered images (a and b) and images obtained via, two projected symmetry averaging (planesymmetry) for the image with p1 and p4mm (c and d) are shown here. MFI zeolite has an orthorhombic structure with space group Pnma and lattice parameters; a = 20.09 Å, b = 19.73 Å and c = 13.14 Å. The projected framework structure along the b-axis has plane group symmetry of p2gg ( Figure S6a) and contains two different 5-membered rings (5Rs) in size, large 5Ra and small 5Rb which are alternately arranged along c-axis; single 6membered rings (S6Rs) and the largest pores with 10-membered rings (10Rs), Figure S6b. This projection is the best one in order to visualize the MFI framework; however, some different T atoms are on the overlapped along the same columns and therefore are not visible. By looking along the c-axis, it is possible to locate all T-sites which has been labeled consistently with the IZA database nomenclature ( Figure S6c). Figure S6. Basic structure of MFI and different T-sites. a) and b) projected framework structure along the b-axis with symmetry elements of plane group p2gg, and four different rings 10R, S6R, 5Ra and 5Rb. c) all different Tsites viewed along the c-axis.

Fe-MFI simulations of different thickness
In order to corroborate the possiblity of identifying Fe that would be incorporated into the framework, only one Fe atom per unit cell was incorporated into the T2 and another into the T5 sites. See Figure S8a. Simulations for different thicknesses using an inner angle of of 30 mrad, a convergence ange of 17 mrad and probe size of 1Å are depicted in Figure S8b. For thicknesses over 135 Å no substantial diffferences were observed. Addtitional simulations, Figure S8c, presents the influence of the Fe location algong the axes of observation, observing no substantial changes in the intensity produced by the Fe atoms. For comparison, exprimental data on bare MFI zeolite, without Fe is also presented together with the correspondent iamge simulations.

DFT calculations for Na-LTA and Fe-MFI:
DFT calculations for Na-LTA were carried out to gain structural insight into Na96Si96Al96O384. The starting structure model for dehydrated Na-LTA zeolite, with Fm-3c (#226), a = b = c = 25.554 Å, α = β = γ = 90 ∘ , was obtained from the International Zeolite Association (IZA) website ( Figure S9). Assuming that the resultant model has cubic symmetry, two models with cubic Pm-3n (#223) symmetry containing 88 Na + ions (64 at S6R sites and 24 at 8R sites) were obtained (see Figure S10). It is impossible to distribute 8 Na + ions near D4Rs while keeping cubic symmetry of the unit cell ( Figure S11); however, 11 distributions with tetragonal symmetry are possible ( Figure S12). These 11 4R-Na + models were combined with the abovementioned Models 1 and 2, so that 22 different models were produced (11 for Model 1 and 11 for Model 2). Calculated energies of the two different Na-LTA models are summarized in Table S1. Relative energies spanned up to 18 eV. After close inspection of DFT-calculated energies, Model-2-2 was identified as the most stable structure for Na-LTA.
(a) Na-LTA A CIF file (Linde_Type_A_Dehydrated.cif) of the Na-LTA zeolite (Na96Si96Al96O384) was obtained from the International Zeolite Association (IZA) website. The lattice parameters of the unit cell in this CIF file were a = b = c = 25.554 Å, and α = β = γ = 90 ∘ (see Figure S9). There are three distinct types of binding sites for Na + ions in the LTA zeolite, namely, those featuring 6-membered rings (6SR), 8-membered rings (8SR), and double 4-membered rings (D4R) formed by oxygen atoms. Structural models of Na-LTA were built via the following two-step procedure. In the first step, only the Na + ions residing at the 6MR and 8MR sites were included in the model. There are four possible bindings sites in each of the 8MR sites, and thus there is uncertainty in the positions of Na + ions at the 8MR sites. However, by assuming that the resultant model has cubic symmetry, the following two models with cubic (Pm-3n) symmetry have been obtained. These two models contain 88 Na + ions (64 at the 6MR sites and 24 at the 8MR sites, see Figure S10). We next considered how to distribute the remaining 8 Na + ions at the 4MR sites. As shown in Figure S11, in total, there are 96 4MR sites in the unit cell of Na-LTA. Figure S11. Remaining Na + . 96 4MR sites in the unit cell of Na-LTA.
No matter how 8 Na + positions are selected from the 4MR sites, it is impossible for the resultant unit cel of Na96Si96Al96O384, or even for the 8 Na + ions at the 4MR sites alone, to have cubic symmetry. As such, we considered reasonable positions of Na + ions at the 4MR sites. Even though it is not possible to find cubic symmetry for the 8 Na + ions, we could obtain the following 11 distributions that satisfy tetoragonal symmetry ( Figure S12). These 11 4MR Na + models were combined with the above-mentioned Models 1 and 2. Consequently, 22 different models in total were produced (11 for Model 1 and 11 for Model 2). In what follows, different models are designated by two numbers; for example, Model 1-2 refers to Model 1 combined with the second 4R-Na + model.
We performed periodic density functional theory (DFT) calculations on all these models, using VASP 5.4.1 [4] . The projector-augmented wave (PAW) method and the PBE exchangecorrelation functional were employed [5] , while the kinetic energy cutoff was set to 400 eV. The energy thresholds for self-consistent-field was set to 1 × 10 -5 eV. The "LREAL=AUTO" option (for real-space projection), Γ-point-only k-point sampling, and a first-order Methfessel-Paxton scheme with a σ value of 0.2 were employed for DFT calculations. Because of the large size of the unit cell, geometry optimization was not performed. The reported energies are the energies corresponding to the extrapolated σ→0 limit. VESTA was used to visualize zeolite structures [6] . Calculated energies of the two different Na-LTA models are summarized in Table S1. Relative energies spanned up to 18 eV, and the results show that Models 1-9, 1-10, 2-1, and 2-2 are relatively stable. Close inspection further shows that Models 2-1 and 2-2 are more stable than Models 1-10 and 1-11, and Model 2-2 is most stable. DFT calculations thus, suggest that distributions 1, 2, 9, and 10 in Figure S12 have higher populations than others. It should be noted here that by rotating each of these models about an axis parallel to the a, b, and c axes, we can obtain many other equivalent distributions for the Na + ions at the 4MR ions. For example, if distribution 1 is rotated by 90 degrees about the axis that is parallel to the a axis and passes through (a, b, c) = (0, 1/2, 1/2), we can get another distribution of 8 Na + ions with tetragonal symmetry, and many other equivalent distributions can be created. We infer that, although each of these distributions of the 8 Na + ions have only tetragonal symmetry, in real unit cells, many such equivalent, yet differently oriented distributions of 4MR Na + ions would be adopted, which may finally give the structure solution with cubic symmetry.

(b) Fe-MFI
For computational analyses of MFI, the structure of silicalite-1 (Si96O192), with Pnma (#62), a = 20.0900 Å, b = 19.7380 Å, c = 13.1420 Å, α = β = γ = 90 ∘ , was obtained from the International Zeolite Association (IZA) website. Silicalite-1 has 12 crystallographically distinct T-sites at 8(d) positions within the unit cell, and thus 12 models of Fe-substituted MFI were built by replacing one of the 8 equivalent sites of each T site by an Fe atom ( Figure S13). For the current calculations the oxidation state of Fe was assumed to be +3, which is deemed chemically more plausible than the +4 state [7] . To ensure charge neutrality of the unit cell, one of the 4 neighboring O atoms around the introduced Fe atom was protonated. Consequently, the unit cell of each of the 12 different T-site models consists of 1 Fe atom, 95 Si atoms, 192 O atoms, and one H atom. As each T-site model has 4 possible protonation sites, we examined 48 models in total, table S2. This is in agreement with IR data obtained, which clearly show an OH stretching signal at 3617 cm -1 (Al-MFI samples show an OH stretching signal at 3610 cm -1 ) confirming the existence of Si-(OH)-Fe sites ( Figure S7 and Figure S14).
The DFT-calculated energies for the models obtained are presented in Table S3. Although some specific sites were seen to have somewhat lower energies than others, small energy differences did not allow us to draw a definitive conclusion on the Fe location at specific sites.
Periodic Kohn-Sham density functional theory (DFT) calculations were performed in the same way as above, but here, geometry optimization was performed. The energy threshold for geometry optimization cycles was set to 1 × 10 -4 eV. The difference between the spin-up and spin-down electrons was set to 5. During geometry optimization, atomic positions were relaxed, while cell parameters were held constant.   Table S3 summarizes the DFT-calculated energies obtained for Fe-MFI models. According to our results, the relative energy increases in the order T4 < T7 < T10 < T1 < T8 < T3 < T11 < T5 < T2 < T12 < T6 < T9. Therefore, if the atom displacement process is thermodynamically controlled, preferred positions will have correlations with this order. However, in case the process is kinetically controlled, the DFT-calculated relative energies will not tell which position is likely occupied by Fe. As such, at the current stage, our DFT results only have a qualitative meaning. The low energy of the T4 model is attributed in part to the H-bonding interaction formed between the O-H bond and a nearby O atom ( Figure S14). A few authors have reported periodic quantum mechanical calculations on Ti-doped MFI zeolites (TS-1).
Gale performed periodic DFT calculations on TS-1, and energy differences of different models were small [8] . Pirc and Stare employed periodic Hartree-Fock calculations to compare the relative stability of different TS-1 models, while imposing symmetry [9] . Here again, energy differences were small. It should be noted that in the case of Fe-substitution, the introduced iron will have a +3 oxidation state.

Structural study of Na-LTA and Fe-MFI by electron diffraction
LTA: Structure solution was obtained with the highest symmetry Fm3 ̅ c (#226). Modern direct method implemented in SIR2014 software was used to solve the structure. Chemical formula (NaSiO2)96 was used in the solution process. A reasonable framework structure could be obtained and was consistent with the structure reported by Pluth and Smith. However, considering hhl reflection (for example 131) and possibility of Laue class m-3, then lower symmetries such as F4 ̅ 3c (#219), F4 ̅ 3m (# 216) and Fm3 ̅ (# 202) could also give reasonable results. Therefore, the possibilities of space groups with lower symmetry than Fm3 ̅ c are not completely discarded yet. Na (1) and Na (2) in Pluth and Smith reports were detected and located after the refinement, while Na (3) was not observed because of the low occupancy. Atomic coordinates are summarized in Table S4. Figure S15 corresponds to the three-dimensional electron diffraction tomography (3D-EDT) data and to selected area electron diffraction (SAED) patterns recorded for nano-single crystals of Na-LTA in the volume range of (100 ~ 400 nm) 3    In order to check the possibility that globally averaged symmetry was higher despite local one could be lower, additional EDT data was collected from very small single crystals with dimensions of 100-150 nm. For smaller crystals, the superlattice reflections became much weaker (and close to the noise level) and they were thus hard to distinguish directly in the reconstructed 3D diffraction pattern. However, by comparing the intensity of odd number spots (h, k, l = 2n + 1), (135) reflections are distinct and relatively strong. No other significant violations to the reflection condition of Fm3 ̅ c (No. 226) could be observed, Figure S16. SAED always yields relatively strong multiple scattering, which may generate extra spots in a diffraction pattern. Nonetheless, when the incidence direction is slightly tilted away from the zone axis, these extra spots caused by multiple scattering should decrease and disappear, Figure  S17. Figure S17. a) SAED pattern at zone axis and b) SAED pattern obtained with an incidence direction that was slightly tilted away from the zone axis. The extra spots are marked by yellow circles In addition to tilting away from the zone axis, precession electron diffraction (PED) was also used to examine the space group. In the PED patterns, no evidence of violating the reflection rules of Fm3 ̅ c (No. 226) could be found. Figure S18. Two sets of PED data for large and nano crystals along the [001] direction.

Fe-MFI
The structure of Fe-MFI was determined through 3D-EDT, SG Pnma with lattice parameters of a = 20.6 Å, b = 20.6 Å, c = 13.9 Å, α = 90.3˚, β = 90.7˚, γ = 90.3˚ from the observed reflections and the extinction rules. The structure was solved and all framework atoms were identified. From the electrostatic potential map, no preferential occupancies of Fe on different T-sites were observed (see also Figure S20). The structure of Fe-MFI was determined by 3D-EDT. From the observed reflections and extinction rules, Pnma with lattice parameters: a = 20.6 Å, b = 20.6 Å, c = 13.9 Å, α = 90.3˚, β = 90.7˚, γ = 90.3˚were used for structure solution. There were no preferential occupancies of Fe on different T-sites from the high-quality ED data obtained from rather large volume, in agreement with the STEM analyses.