Fast-ADT: A fast and automated electron diffraction tomography setup for structure determination and refinement

Journal Pre-proof

Fast-ADT: a fast and automated electron diffraction tomography setup for structure determination and refinement S Plana-Ruiz , Y Krysiak , J Portillo , E Alig , S Estrade´ , F Peiro´ , U Kolb PII: DOI: Reference:

S0304-3991(19)30366-3 https://doi.org/10.1016/j.ultramic.2020.112951 ULTRAM 112951

To appear in:

Ultramicroscopy

Received date: Revised date: Accepted date:

28 October 2019 20 January 2020 26 January 2020

Please cite this article as: S Plana-Ruiz , Y Krysiak , J Portillo , E Alig , S Estrade´ , F Peiro´ , U Kolb , Fast-ADT: a fast and automated electron diffraction tomography setup for structure determination and refinement, Ultramicroscopy (2020), doi: https://doi.org/10.1016/j.ultramic.2020.112951

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2020 Published by Elsevier B.V.

Highlights    

New EDT acquisition routine for fast and systematic crystallography analyses. Module designed for universal application both in TEM or STEM mode. External optical CCD camera used for analysis of disordered structures. New polymorph of DRED1 organic dye solved ab initio in ̅ .

Fast-ADT: a fast and automated electron diffraction tomography setup for structure determination and refinement S. Plana-Ruiz1, 2, 3, Y. Krysiak1, J. Portillo4, 5, E. Alig6, S. Estradé2, 3, F. Peiró2, 3, U. Kolb1, 7, * 1

Institut für Angewandte Geowissenschaften, Technische Universität Darmstadt, Petersenstrasse 23, 64287 Darmstadt, Germany. LENS, MIND/IN2UB, Departament d’Enginyeria Electrònica i Biomèdica, Universitat de Barcelona, Martí i Franquès 1, 08028 Barcelona, Catalonia. 3 Institute of Nanoscience and Nanotechnology (IN2UB), Universitat de Barcelona, 08028 Barcelona, Catalonia. 4 Centres Científics I Tecnològics, Universitat de Barcelona, Lluís Solé i Sabarís 1-3, 08028 Barcelona, Catalonia. 5 NanoMegas SPRL, Blvd Edmond Machtens 79, B-1080 Brussels, Belgium. 6 Institut für Anorganische und Analytische Chemie, Johann Wolfgang Goethe-Universität, Max-von-Laue 7, 60438 Frankfurt am Main, Germany. 7 Institut für Anorganische Chemie und Analytische Chemie, Johannes Guttenberg-Universität Mainz, Duesbergweg 10-14, 55128 Mainz, Germany. *corresponding author: [email protected] 2

Abstract Electron crystallography has focused in the last few years on the analyses of microcrystals, mainly organic compounds, triggered by recent publications on acquisition methods based on direct detection cameras and continuous stage tilting. However, the main capability of a transmission electron microscope is the access to features at the nanometre scale. In this context, a new acquisition method, called fast and automated diffraction tomography (Fast-ADT), has been developed in form of a general application in order to get the most of the diffraction space from a TEM. It consists of two subsequent tilt scans of the goniometric stage; one to obtain a crystal tracking file and a second one to acquire an electron diffraction tomography. This setup has been implemented on both TEM and STEM modes of the microscope, thus it can be installed on any TEM regardless of the availability of a scanning unit. BaSO4 crystals have been measured to demonstrate the validity of the technique for structure determination and refinement. A recently solved layered silicate, RUB-5, has been used to prove the method advantages for fine identification of disorder details. Last, a new polymorph of a DRED1 organic molecule has been solved ab initio and refined by X-ray powder diffraction to show the full application of the presented method.

1. Introduction Novel materials are investigated to develop new technologies but also to increase the performance of the already available ones. The key point of their characterization is the determination of their structure. In other words, which are the elements present in the material, which is the ordering of the atoms and, if there is one, the basic unit-cell and the atomic positions within. The use of such information allows the description a priori of material properties in fields like optics, mechanics or electromagnetism, which foresees the possible applications in which they may be best fitted. Several techniques are available nowadays for structure determination depending on the characteristics of the investigated material. X-ray diffraction was and still is the preferred method to determine crystal structures since its discovery by Max von Laue, Paul Knipping and Walter Friedrich in 1912 and, subsequently in 1913, the investigations of Sir William Lawrence Bragg [1]. X-ray methods, firmly established as standard tools and used in a systematic way, bear one main intrinsic disadvantage. They cannot acquire diffraction information from individual nanometre-sized crystals. This inconvenience may become a problem when, for instance, phase mixtures have to be used to determine the structure of an unknown phase [2], or nanocrystals embedded in a different phase matrix have to be crystallographically studied [3,4]. In this context, electron diffraction appears as a solution and it becomes an important and valuable tool [5]. The short wavelength related to electrons in transmission electron microscopes (TEM) allows the acquisition of diffraction data from nanometric volumes. Almost plane slices of the diffraction space measured and transferred to single images allow the partial reconstruction of a diffraction volume using different diffraction patterns (DPs) at different crystal orientations. On the other hand, electrons interact more strongly with matter than X-rays -both elastically and inelastically-, which results in a

higher probability of multiple scattering events, so-called dynamical effects. Thus, the proper handling and interpretation of this diffraction data may become difficult, yet not impossible [6,7]. The use of precession electron diffraction (PED) helps to diminish the effects of dynamical scattering and makes the crystallographic analysis easier [8–11]. PED, based on the precession of the electron beam with the pivot point of such rotation at the specimen plane, integrates the reflections by the Ewald sphere wobbling. However, non-kinematical effects are not completely eliminated and structure refinement on sub-micrometer crystals is still usually done with the Rietveld method [11] on X-ray powder diffractograms. During last years, structure refinement using the dynamical scattering theory – called dynamical refinement- has popped up as a final step for a complete and accurate structure characterization using electrons only. The main inconvenience is its time-consuming calculations for low symmetry crystals [12,13]. Nevertheless, the development of electron diffraction methods is at the point to compete with X-ray techniques in terms of reliable and precise crystal structure analysis.

2. Motivation Several acquisition methods to obtain 3D reconstructions of the experimental diffraction data, proposed since the invention of the technique about one decade ago [14,15], are all based on the same idea; namely the reconstruction of the observed diffraction space (intensity-weighted reciprocal lattice) from a set of electron DPs acquired at different tilt angles of the TEM goniometric stage. This technique was initially called automated diffraction tomography (ADT) [14,16], but it is generally termed as electron diffraction tomography (EDT). The flowchart in Fig. 1 summarizes the steps that the ADT routine follows. It automatically acquires a scanning transmission electron microscopy (STEM) image between each DP acquisition for position checking. A cross-correlation is calculated with a reference to determine the crystal displacement and, subsequently, to shift the beam accordingly, which can be manually modified as well. The usual tilt step between DP acquisitions using this methodology is 0.5: or 1:. One of the advantages of this serial slicing of the diffraction space is that crystals have not to be previously oriented in a specific way, therefore off zone-axis DPs are mostly recorded with less dynamic effects [17].

Figure 1. Steps followed by the ADT routine. Focused beam refers to the smallest probe size in order to acquire a focused STEM image. Defocused beam refers to a probe with the size previously selected for the DP acquisition.

However, a missing wedge between frames is present and its crystallographic information is lost, the socalled missing step wedges. For this reason, other acquisition methods were developed to sample more from the diffraction space and obtain better structure solutions and refinements. Some of these acquisition methods are: precession electron diffraction tomography (PEDT) [18], rotation electron diffraction (RED) tomography [19] and integrated electron diffraction tomography (IEDT) [20], which in some other works is called microED [21] or continuous rotation electron diffraction [22,23]. Here is noted that the use of the word rotation for tilt of the goniometric stage may be misleading as rotation in a TEM is traditionally referred to the azimuthal angle instead of the polar angle [24]. This conventional notation is the one used in this work.

PEDT couples the electron beam precession with the EDT technique. The precession of the beam (precession angles usually set between 0.5° and 1.5°) induces the Ewald sphere to sweep the missing step wedge and integrate the reflections to increase the probability to measure them at their maxima. In this way, averaged information of the missing step wedge is retrieved, which provides more accurate structure solutions using the same number of tilt steps as ADT [25]. RED performs the crystal tilt in two steps; fine beam tilt steps of 0.01-0.20° combined with coarse goniometer tilt steps of 2-4°. The fine slicing of the diffraction space results in finer rocking curves to determine the maximum intensity or integrate (software-wise) each reflection. Although the approach becomes better for unit-cell determination, dynamical artefacts such as Kikuchi lines, double-diffraction or intensity redistributions cannot be minimized as in PEDT. Finally, IEDT performs the acquisition while the sample is continuously tilted, resulting in a reduction of the sample beam exposure as well as a fine slicing of the space. This technique does not apply a beam shift during the acquisition, thus relying on the crystal staying in the illuminated or selected area. For this reason, the stage stability is the most important condition, especially in higher angular ranges. PEDT and IEDT techniques can be combined (PED-IEDT) in order to integrate reflections that are not covered when the camera is in stand-by (the time when the electronics are reading the data from the chip and saving it to the disk, i.e., the read-out time). If suitable stage velocity and precession angle are properly set, the sampled diffraction space can be maximized [20]. DPs are usually acquired with post-column charge-coupled device (CCD) cameras of high dynamic range. Although reflection intensity quality is key for successful dynamical refinements, the unit-cell determination as well as the structure solution do not need such number of intensity levels if the averaged structure is needed. While the former needs the geometrical spot positions, the latter only needs a classification or ranking of the reflection intensities as “small”, “medium” and “strong”, which was the methodology first implemented in X-ray techniques [26–28]. The disadvantages of these kinds of cameras are that they do not allow a high acquisition frame rate due to their high read-out time, and an overexposure of the transmitted beam may damage the camera scintillator. Furthermore, higher exposure times are needed for good signal-to-noise diffraction data, a main inconvenience for the investigation of beam sensitive materials. For these reasons, new cameras were launched with higher frame rate and sensitivity, to increase the number of DPs before these crystals are fully degraded. Some examples of the performance of such cameras can be found elsewhere [29,30]. Nevertheless, diffraction data even from beam sensitive crystals like organic compounds or metal-organic frameworks (MOFs) can be acquired with CCDs, and structure solutions can be successfully obtained [31–33]. The recent IEDT acquisition methods totally focus on operating the microscope in TEM mode and leave behind the advantages of STEM mode, e.g., low-dose imaging for crystal tracking, easy pinpoint of crystal check or avoiding the hysteresis behaviour of the projector system [34]. For this reason, this work shows a new developed acquisition method implemented in a setup designed for both TEM and STEM microscope modes, which provides an all-in-one platform for JEOL and ThermoFischer (former FEI) TEMs using Gatan cameras. The new approach has been called fast and automated diffraction tomography (Fast-ADT). It also shows how Fast-ADT using an optical CCD camera is capable to solve and refine beam stable structures and assess fine details of crystal structures such as disorder features.

3. Fast-ADT: Principle and Acquisition Routine There are two ways to acquire electron DPs: selected area electron diffraction (SAED), which uses an aperture positioned at the image plane (selected aperture) to select the feature to be illuminated, or nano-beam electron diffraction (NBED), which uses a nanometre-sized beam that can be placed at the region of interest (ROI) through the beam shift coils. The IEDT acquisition methods published up to now are based on SAED or NBED with beam sizes similar to the illuminated region of a selected aperture. Although it may be enough for micro-crystals, the use of such beam settings have problems when, for instance, individual nanoparticles in agglomerated nanoparticles have to be measured [35], or singlecrystal tips of around 30 nm in diameter are only available for analyses [36]. Fast-ADT takes into account these possibilities using the NBED setting with beam sizes down to the diffraction limit. This allows the acquisition of diffraction data from nanometre-sized volumes and avoids the illumination of the grid background, which minimizes the amorphous carbon contribution.

Figure 2. Steps followed by the Fast-ADT routine in TEM or STEM mode.

The routine followed by Fast-ADT is based on two complete stage tilts through the desired angular range. The first tilt scan is used to check and save the crystal position with respect to the tilt angle while the second one is the acquisition of the diffraction tomography. Gemmi et al. first suggested this idea, although it was not fully developed neither implemented [20]. The main steps of the Fast-ADT routine are summarized in Fig. 2. If a sequential tilting approach is preferred (left hand side of the scheme), each image is acquired with 5° of tilt step using software binning of the camera in order to be fast and efficient, and, subsequently, the crystal positions for each tilt of the tomographic acquisition are interpolated. The x and y movement of a feature of interest in a TEM grid with respect to the tilt angle is almost linear or non-existent when small angular ranges such as 60° (±30°) or 80° (±40°) are considered, but it becomes non-linear when angular ranges are as high as 120° (±60°) [37]. An important step here is to find the eucentricity for the targeted crystal. This ensures that the crystal is moving as little as possible while it is kept in focus, otherwise it might not be possible to track it, even with the smallest magnification available in the selected-area mode of the projector system. Needless to say that the stage should be optimized for tomography experiments by the TEM manufacturer engineers. The crystal displacement depends on the position of the crystal on the grid as well, thus the eucentricity and focus has to be checked for different positions. In case that the Z-height has to be modified, a new set of reference images has to be acquired in order to retrieve the new crystal movement. The Fast-ADT module has been developed to automatically acquire these images in TEM or STEM mode. After the acquisition of the crystal-tracking images, a cross-correlation calculation is performed using the middle one of the angular range as reference. Instead of using the whole image for the crosscorrelation, which is not working properly due to the change of the crystal projection, a sub-image selected by a ROI is used in order to have more accuracy on the part of the crystal that the beam should follow. Sobel together with smooth filters are applied to the images to highlight the crystal edges and get more precise crystal positions. Other methods have been tested but they do not result in a better

location of the crystal (Fig. S1). Nevertheless, the position of the ROI in all the crystal-tracking images can be carefully checked and, if necessary, modified after the cross-correlation. Once the crystal positions have been checked, a crystal-tracking file is generated and loaded on the module. Then, the stage is tilted to the initial angle, a (S)TEM image is acquired to position the beam at the desired initial place and the diffraction acquisition starts. Binning and exposure time of the camera have to be previously set in order to avoid saturated spots and scintillator damage. Besides this sequential discrete tilt of the stage a continuous tilt option is also available (right hind side of Fig. 2. It is worth to mention that both approaches can be successfully assisted with a precession of the beam. Goniometric stages unfortunately have hysteresis behaviours produced by stage motors and the holder design and mechanical condition, e.g., bending of the tip or holder axis not optimized for eucentric height at low values of (middle of the pole piece gap). Therefore, the x and y crystal movements for crystal-tracking file generation and diffraction tomography acquisition are never the same. This is the main limiting factor of the beam size because it determines when the beam will not illuminate the targeted crystal. For this reason, tilt-scan reproducibility tests have to be performed in order to assert the minimum beam size for a given TEM and holder ensuring the acquisition of any DP from the crystal. A tilt-scan reproducibility plot shows the difference of the and position of a crystal between two consecutive tilt scans. Firstly, a reference feature is located on one of the tilt scan images and the positions of this feature on all the images are saved; and . Then the x0 and y0 position of the feature at the initial angle is subtracted to all the positions to obtain displacement values referred to the initial one ( , ). Finally these displacements are compared between the two tilt scans ( , ) and the hysteresis shift is calculated ( √ ). In this context, the minimum beam size depends on the maximum hysteresis shift, , and the size of the investigated crystal, . is understood as the distance between the selected position for measurement of the crystal and its end along the maximum shift direction. There are two beam diameter conditions sketched in Fig. 3. If a targeted crystal has a higher than , the minimum beam size is the diffraction limit. If is smaller than , the minimum beam diameter is limited to the difference between the and . This can be summarized in the following equations:

Figure 3. Schematics for the two conditions that could occur to determine the minimum beam size to illuminate a crystal during the Fast-ADT routine. The red crossed-circle is the position selected on the first tilt scan and the blue crossed-circle is the same crystal position on the second tilt scan.

{

Where λ is the electron wavelength, α is the convergence angle and is the diameter of the beam. is the diffraction limit calculated as the full-width at half-maximum of the Airy disc [38].

Figure 4. Tilt-scan reproducibility plots from A) the FEI tomography holder on a Tecnai Spirit TEM, and B) the FEI single-tilt and C) Fischione tomography holder on the FEI Tecnai F30 TEM. The first tilt scan was performed with 5 degrees of tilt step and the second one with 1 degree of tilt step. The same tilt angle positions were only selected for these plots.

Fig. 4 shows some examples of such tilt-scan reproducibility plots. Fig. 4A corresponds to the FEI tomography holder on a Tecnai Spirit LaB6 TEM and it has a of about 30 nm, which means, for instance, that a beam size bigger than 40 nm will be needed if a crystal with a lateral size of 10 nm has to be analysed. Fig. 4B corresponds to a Tecnai F30 TEM using a relatively new FEI single-tilt. In this case, with a around 20 nm, the beam size needs to be bigger than 20 nm. Finally, Fig. 4C correspond to an intensively used Fischione tomography holder. In this case is of 120 nm due to a nonreproducible jump of the x-position. Thus, to analyse the same crystal size of 10 nm the beam size needs to be bigger than 220 nm. However, this beam size could be unsuitable for some challenging samples. One way to overcome this size-limitation is to position the center of the beam at the edge of the crystal before starting the DP acquisition, but the crystal-tracking file is created selecting the desired position of the crystal. In this way, the minimum beam size would be , thus 200 nm for the tomography holder following the example above. In order to reduce this size limitation even more, other solutions have to be applied. One possibility is to acquire more than one DP per tilt angle; the beam is shifted to pre-selected regions on the crystal-reference images, restraining the minimum beam shift to the beam diameter in order to avoid the illumination of already exposed areas. Furthermore, this idea allows the acquisition of diffraction datasets from different crystals or different parts of a “big” crystal with only one tilt scan, given that they are properly visible in all the reference images. Another possibility is to use the ADT method, which allows a precise positioning of the beam for each tilt angle. The main disadvantage for both possibilities is the acquisition time, which is multiplied according to the number of DPs per tilt angle for the former one, and the STEM image acquisitions between DPs for the latter one.

4. Fast-ADT: Experimental Setups First of all, the NBED setting has to be available and aligned. If TEM mode is considered, both main TEM manufacturers (JEOL and ThermoFischer) have default NBED illumination modes, called NBD and Microprobe, respectively. If STEM mode is preferred, ThermoFischer TEMs have the same Microprobe mode available, but JEOL microscopes fix it to one highly convergent beam mode, thus the angular resolution is considerably decreased even using the smallest condenser aperture (disk-like DPs). Nevertheless, a quasi-parallel STEM can be aligned and saved if a free lens control module is available and the condenser system has a minimum of 3 lenses. Detailed alignment methods and analyses of electron probe diameter against convergence angle can be found elsewhere [34,39,40]. The Fast-ADT module is based on the scripting language of the Digital Micrograph™ (DM) environment. It takes control of Gatan cameras, produces the beam shifts for the crystal tracking and tilts the alpha angle of the stage in the sequential approach. STEM images are generated via the DigiScan™ unit provided by Gatan. In the case of the continuous approach, external software tools to control the tilt velocity of the stage (Temspy program for ThermoFischer microscopes and GonioTool program for JEOL microscopes) are synchronized with the Fast-ADT module. The resulting DM plug-in has been successfully tested in a Tecnai F30 FEG operated at 300 kV, a JEOL 2100F operated at 200 kV and a Tecnai Spirit LaB6 operated at 120 kV. The Gatan cameras tested in this work are UltraScan4000 (US4000, 4096 x 4096 pixels) and UltraScan1000 (US1000, 2048 x 2048 pixels). The advantage of this camera series is its good quality reflection intensities for diffraction, but its main disadvantage is its high read-out time (≈ 3.6 seconds for binning 2 of US4000 and ≈ 1.7 seconds for binning 1 of US1000). Although these times are feasible for the continuous approach, there is no real benefit on comparison to the sequential one in terms of experimental time and data quality. The tilt velocity has to be reduced down to ≈ 0.09 °/s for 4 s of exposure and binning 2 on the US4000, which is the usual camera setting for low-dose experiments [41]. For this reason, a second module was developed in Matlab-environment in order to take control of other fast frame-rate cameras for an optimum continuous approach. This second design is based on the setup used by the precession-assisted crystal orientation and phase mapping technique [42,43] -commercially available as ASTAR™ and distributed by NanoMegas SPRL. It uses an external and optical CCD camera (Allied Vision Stingray F-145B) mounted on the binocular position of the TEM to capture the DPs that are projected at the small or big fluorescent screen. The major advantage of such cameras is the high frame-rate to acquire DPs, which means that acquisitions

of a PED-IEDT experiment between ± 60° can be performed in less than 1 minute. The signal generator unit of the ASTAR™ system is used to produce the beam shift and a modified tilt controller -following the design of Shi et al. [29]- is used to trigger the start of the continuous tilt. The use of external equipment specifically designed for this approach allows the synchronization of the camera acquisition, crystal tracking and stage tilting. In this case, the module was successfully tested in a Tecnai F30 FEG operated at 300 kV and a Tecnai Spirit LaB6 operated at 120 kV. Details on the frame processing to correct the geometrical distortions at the tilted fluorescent screen can be found in Appendix I. Finally, the ADT is re-programmed as a DM plug-in that works in TEM and STEM mode in order to meet the most challenging nanocrystal problems. All the DM modules are free available (https://www.akkolb.chemistry.uni-mainz.de). Further technical details on the different modules can be found in Appendix II and the graphical user interfaces are shown in Fig. S2.

5. Data analyses The aim of this section is to show how diffraction data acquired with the Fast-ADT technique provides good reflections quality for structure determination and refinement. All TEM measurements were carried out on a Tecnai F30 S-TWIN operated at 300 kV with the Microprobe illumination mode. STEM images were acquired with a Fischione high-angle annular dark field (HAADF) detector mounted at the side port of the TEM column. The precession signal was generated through the DigiStar processing unit provided by NanoMegas SPRL. First of all, BaSO4 crystals were used as a case example to demonstrate that structure solution and dynamical refinement is suitable from Fast-ADT, even with an optical and external CCD camera. Secondly, the use of this external CCD camera is shown for the fine slicing of the diffraction space to assess the disorder features of a layered structure [44]. Finally, an unknown polymorph of a commercially available organic red dye, disperse red 1 (DRED1), is solved ab initio using Fast-ADT data and refined via X-ray powder diffraction (XRPD). 5.1 Structure solution & dynamical refinement: BaSO4 case example Fine powder of Barite (BaSO4) [45–47] was purchased from Merck. Barite is an inorganic salt with an orthorhombic crystal system (Pnma) mostly used in oil and gas exploration. The sample was dispersed in ethanol and sprayed on a carbon-coated copper grip with a UIS250v Hielscher sonifier. Three crystals were selected and two Fast-ADT datasets were acquired on each one with 1 degree of precession angle; one using the US4000 post-column CCD camera and the other with the Stingray F-145B external CCD camera. Spot size 6, gun lens 8 and a 10-μm condenser aperture were used to produce a quasi-parallel beam for the US4000 dataset. Spot Size 6, gun lens 4 and a 10- μm condenser aperture were selected for the F-145B dataset. A 200-nm beam was chosen for both cases. Table 1 summarize the acquisition and processing details for all the Fast-ADT datasets. The post-column CCD dataset was acquired through the sequential approach with 1: of tilt step. An exposure time of 0.5 s was selected, which resulted on a diffraction acquisition time of around 10 minutes. The external CCD dataset was acquired with the continuous approach. The tilt velocity was set to 1.77:/s for the first crystal and 1.5:/s for the other two. The exposure time was set to 100 ms, which translated to around 1 minute for the whole diffraction data acquisitions. In this second case, special care was taken to avoid saturated reflections because the exposure time and the gain of the camera can be coarsely changed without any danger to damage the sensor. The use of such low velocity and exposure time resulted in a reflection integration of 0.15-0.18: per DP. If we consider the integration due to 1 degree of precession movement and the reflection integration by the continuous tilt of the stage, the dataset can be later analysed as one acquired with the sequential approach. One of the main key points for reflection intensity extraction is the background of the DPs [48–50]. The US4000 provides a background with almost all pixels at 0 intensity, thus any processing of the DPs is not necessary. On the other hand, the data from Stingray F-145B contains a lot of noise in the background that reach a 2000 intensity level on average, even if the TEM room is kept fully dark. In order to take into account this huge noise, a histogram from a background region near the edge of the frame without reflections is analysed to find the best constant value to subtract on the DPs. This value is selected to supress most of the noise but still keeps intensity for weak reflections. Fig. S3 shows an example of such

histogram. Independent background fitting and subtraction for each reflection in each DP was done later on through the different data processing programs used in this work. The unit-cell determination and intensity extraction was done with the eADT program [49] followed by structure solution using direct methods with the Sir2014 program [51]. The reconstruction of the observed diffraction space shows that only the b-axis is covered for crystal 1, while b-axis and c-axis are covered for crystal 2 and 3. Structure solutions show all the atomic positions except for one oxygen at a special position in the F-145B dataset of crystal 2. Nevertheless, similar and positive atomic displacement parameters (ADPs) as well as peak heights are obtained for both datasets of each crystal. Table S1 shows the atomic parameters for all the datasets analysed. After successful structure solutions, dynamical refinements were carried out. PETS2 program [50] was used to extract the reflection intensities and a specific module in the Jana2006 program [12,52] was used for the refinement (Table S2 shows the parameters used for the refinement). This refinement was also successful even using anisotropic ADPs, which resulted on positive values for all the principal tensor components. The comparison with the obtained refined structures and refined by X-ray [47] shows that atomic position differences are as high as 2 pm for the US4000 data and 6 pm for the F-145B data. More details on the dynamically refined structures can be found in Table S3 and Table S4. ADPs and R1 values are significantly higher for the F-145B data, although this could be understood because of the worse reflections quality compared to US4000 data. Nevertheless, the refinement converges and provides an accurate description of the crystal.

Crystal 1

Crystal 1

Crystal 2

Crystal 2

Crystal 3

Crystal 3

US4000

F-145B

US4000

F-145B

US4000

F-145B

Post-column

External

Post-column

External

Post-column

External

Tilt range (⁰)

-45/60

-45/60

-60/60

-60/58

-60/60

-60/59

Acquired DPs

106

580

121

796

121

802

Num. of sampled reflections

4651

1728

4826

5304

4464

5684

Num. of ind. reflections

480

398

480

498

319

494

Used resolution (Å)

0.7

0.7

0.7

0.7

0.8

0.7

Ind. reflection coverage (%)

82.2

68.2

82.2

85.2

82.0

84.6

Reflections/Parameter ratio

12.43

10.75

12.71

12.46

8.46

13.36

2

Overall B (Å )

1.493

2.754

1.424

2.054

1.318

2.542

Rsym (Sir2014)

0.185

0.137

0.138

0.163

0.128

0.148

Residual R(F) (Sir2014)

0.152

0.165

0.118

0.158

0.126

0.160

Num. of used reflections (obs/all) - Dyn.

3161/2667

11923/9542

3033/2944

10574/8590

1963/1931

11071/8899

Rint (obs) (PETS2)

0.159

0.101

0.170

0.126

0.185

0.128

0.075/0.085

0.183/0.229

0.074/0.076

0.173/0.220

0.079/0.080

0.180/0.229

a (Å)

8.880

8.841

8.883

8.882

8.899

8.842

b (Å)

5.474

5.454

5.460

5.434

5.471

5.483

c (Å)

7.127

7.185

7.143

7.179

7.116

7.145

α (⁰)

90.6

90.2

90.4

89.3

89.7

89.7

β (⁰)

89.7

89.4

90.2

89.7

90.2

90.2

γ (⁰)

90.3

89.3

90.0

90.1

89.7

90.4

R1 (obs/all) - Dyn. (1)

Unit-cell determination :

(1)

The unit-cell parameters were scaled by the averaged factor from the unit-cell parameters of Jacobsen et al. [47] (8.879 Å, 5.454 Å and

7.154 Å). Table 1. Acquisition and processing details for the different Fast-ADT datasets of BaSO4 (space group Pnma).

5.2 Disorder analyses: RUB-5, layered silicate example RUB-5 is a recently solved new zeolite-like structure that exhibits a new silica polymorph in a monoclinic crystal system [44]. Due to its layer-like building units, the resulting crystal exhibits a high degree of stacking disorder with intergrowths of different polymorphs. Moreover, the plate-like morphology of the crystals make the disorder analyses difficult because of the strong preferred orientation. In other words, these layer-like building units lay always parallel to the carbon film of the Cu-grid, thus most of the streaks did not appear on single frames, and the main critical experimental parameter for resolving them was the tilt step itself. Fast-ADT datasets were acquired using the sequential approach with the US4000 and the continuous approach with the Stingray F-145B. While the US4000 data was taken in around 20 minutes (121 DPs acquired with 4 s of exposure time and a tilt step of 1 : between -60: and 60:) the F-145B data took about 1 minute (1018 DPs acquired with 30 ms of exposure time and a tilt velocity of 1.75 :/s between 50: and 60:). For comparison, although ADT worked for the characterization of such crystallographic fine details [53], in this case the need of a tilt step smaller than 0.5: and higher exposure times for good signal-to-noise ratio would resulted in 2-3 hours of data acquisition for a single particle (± 60° of angular range). Prior to the reconstruction of the F-145B diffraction data, single frames were background corrected with the modified rolling ball option of the eADT software [49]. Fig. 5 shows the same 2D section from the reconstruction of the observed diffraction space by the two setups, and the zoom-in of 3 selected regions that contain diffuse scattering. The use of smaller tiltsteps provides a better way to sample the streaks, what is much clearly seen the further the reflections are from the non-scattered beam (see Fig. 5C). It also demonstrates that the intensity coming from the fluorescent screen is enough to detect such features. Although the beam dose was higher in the F-145B case to obtain enough intensity from the screen, it was not strong enough to deteriorate the crystallinity of the particle. One problem that could prevent the hampering of the analysis is the reminiscence of the phosphor. For this reason, the screen should be coated with the fast-decay phosphor and special care should be taken on the beam settings. Nevertheless, in this case we can confirm that the low intensities between reflections come from diffuse scattering as other intense reflections do not have such reminiscence effect. Further examples and details on the quantitative analyses of the disorder can be found on the paper of Krysiak et al. [44] as this is not the scope of the present work [49].

Figure 5. A) and B) are 2D sections from the reconstruction of the observed diffraction space that correspond to the crystallographic plane hhl of RUB-5. A) comes from the data acquired with the continuous approach from -50: to 60: and Stingray F-145B, and B) with the sequential approach and UltraScan4000 (US) from -60: to 60:. C) shows zoomed images of the reflections from the red rectangles of A) and B) together with the respective indexes.

5.3 New polymorph of DRED1

DRED1 is an azobenzene derivate from the family of dye molecules (see Fig. 6A). These organic dyes are well known for their photochromatic properties and, large optical nonlinearities and electro-optic properties as dopants or side groups in various polymeric films [54–57]. DRED1 has been intensively characterized with spectroscopic tools (such as FT-IR, vis-UV, Raman and NMR) in different chemical environments, but few papers are found regarding its crystalline structure. It was firstly reported in a hydrated form by Lacroix et al. [58] with a triclinic crystal system (P ̅ ) and Z = 4. Subsequently, Nath et al. [59] showed an unhydrated form with a monoclinic crystal system of unique c-axis (P1121/n) and Z = 8. Technical details on the preparation and measurements of the DRED1 powder can be found in Appendix III. eADT program [49] was used for the unit-cell determination on six DRED1 datasets revealing a triclinic crystal system with parameters a = 7.73 Å, b = 11.06 Å, c = 19.69 Å, α = 74.5:, β = 82.9: and γ = 71.0:. The XRPD diffractogram was indexed with DICVOL [60] using the cell from eADT, and it was followed by a Pawley refinement [61] to accurately obtain the lattice parameters (a = 7.7217 Å, b = 11.1397 Å, c = 19.5753 Å, α = 73.80:, β = 83.04: and γ = 70.50:). Projections from the reconstruction of the observed diffraction space along the main reciprocal axes are shown in Fig. S4. Afterwards, reflection intensities were extracted based on the maximum value of the rocking curves with the eADT software. Using a composition of 4 molecules per unit-cell obtained by the method of the Hofmann’s volume increments [62], ab initio structure solutions in P [51,63] delivered a scattering density distribution approximately providing the position of the molecules. As the independent reflections acquired by each dataset was low (between 50 % and 65 %), one data set covering the a* and b* axes was completed with a second one including the c*-axis in order to reach a completeness of 85%. Simulated annealing (implemented in Sir2014 [51]) was used and the position of the azobenzenes were localized, using both P1 and P . The small difference between both space groups indicated that the compound could be described with P . The correct orientation of the nitro, ethyl and ethanol groups was more difficult to retrieve as they have more freedom to move. A Rietveld refinement [11] from the program TOPAS [64] based on the structure from simulated annealing showed some significant discrepancies. A new approach implemented in PETS2 [50] based on the fitting of a pseudo-Voigt function on the rocking curves was used for intensity extraction. Following the method explained by Brázda et al. [65], the orientation of the frames were refined by fitting the experimentally determined rocking curves by least-squares methods. The unit cell determined from XRPD was used and constrained to prevent the triclinic lattice to partially compensate the misorientation of the frames. Using the new frame orientations, the reflection intensities were re-extracted from both complementing datasets and a scale factor for each frame was refined on the rocking curves. After merging the resulting hkl files, Sir2014 2 and Superflip [63] retrieved a full structure solution. The Biso was constrained to 2 Å and the chemical composition was fixed to 4 molecules in the unit-cell for the execution of Superflip. Jana2006 [52] and Vesta [66] were used to interpret the scattering density map (shown in Fig. S5), which showed all the 46 non-hydrogen atoms of the two molecules in the asymmetric unit. Then, the elements were manually assigned to the atom positions found by the program. The parameters for structure solutions in space group P for these three different hkl files are provided in Table S5. The molecules of the solved structure were restrained and the crystal structure was refined by the Rietveld method against the XRPD data. Three isotropic ADPs were defined for each atom type and refined separately, whereas hydrogen was fixed to be 1.2 times of the carbon ADPs. Fig. 6D shows the two independent molecules inside the unit-cell and Fig. 7 shows the result of the Rietveld refinement on the XRPD pattern. The unit cell found in this work has a volume of 1523.5 Å3, which compares to half-of the cell volume of the unhydrated monoclinic phase [59] (1521 Å3), whereas the volume of the hydrated triclinic compound [58] is 1578 Å3. Additionally, differential thermal analysis (DTA) and thermogravimetric (TG) analysis proved that no H2O molecules are present (shown in Fig. S6). The final crystal structure in P space group is formed by two independent DRED1 molecules, which are not fully planar and of slightly different conformation. The two phenyl rings linked by N=N bond are almost fully planar (mean torsion angle of 0.7 : and 0.4:). Para positioned nitro group deviates slightly from the plane of the phenyl ring (21.4: and 21.1:). The biggest difference between the independent molecules lies in the ethyl group (CN-C-C = 115.7: and 71.5:) and the ethanol group (C-N-C-O = 85.5: and 78.1:) pointing in opposite directions along the a-axis. Fig. 6C shows the layering of the azobenzene groups approximately along the c-axis following a herringbone-like packing. Due to the orientation of the end-groups, the compound

exhibit four O-H···O hydrogen bonds (distance between Oxygens of 2.94 Å and 2.69 Å) from the ethanol groups, leading to a square-like network. The hydrogens were placed between the oxygens by chemical reasonable restraints. This configuration is different to the hydrated triclinic phase as the water molecules produce more hydrogen bonds and form a hexagonal-like network.

Figure 6. A) DRED1 formula, B) HAADF images of two measured crystals with red circles where the beam was placed during the data acquisition, C) the packing pattern of DRED1 with only the azobenzene groups projected approximately along the c-axis and D) two independent molecules inside the unit-cell. C) and D) were generated by VESTA [66].

Figure 7. Rietveld refinement of the new DRED1 polymorph. Black profile corresponds to the measured intensities (I0), red profile to the calculated ones (Ic) and grey profile to the difference (I 0 – Ic). Blue tick marks represent the reflection positions.

6. Discussion Fast-ADT was developed to work both in TEM and STEM modes of the microscope. However, STEM is preferred against TEM because it permits a fast and clear visualization of tiny or layered crystals using

low electron doses, reduces the beam damage for crystal imaging, provides an easy pin-point to check the crystallinity of different parts of the sample, and avoids hysteresis lenses effects because the projector system does not need to switch between image and diffraction mode. STEM coupled with the use of NBED for diffraction acquisition speeds up the whole task of EDT acquisitions, as it avoids the mechanical insertion of the selected area aperture and everything can be controlled software-wise. Another software solution called Instamatic and based in Python [67] has been freely available during these last years, which implements the RED and the IEDT techniques in TEM mode. Although it works quite well for microcrystals [68,69], it is not the most suitable option for acquiring individual diffraction data from the most nanometre-sized crystallites. That is because the positioning of the beam at the nanometre scale in TEM mode suffers from the aberrations of the projector system [70] and an optimum tracking tool is not available for the IEDT technique. The main difference of Fast-ADT with respect to all other EDT acquisition methods is the use of two tiltscans. This decreases the experimental time required to acquire an EDT dataset but also reduces the chance of user errors, because the DP acquisition is automated and no interaction is needed once the tomography acquisition has started. From a technical point of view, the only limiting factor for a successful Fast-ADT acquisition, which is the illumination of the targeted crystal during the whole angular range, is the goniometric stage and the holder mechanical conditions, as they determine the beam sizes required to obtain diffraction data. If the stage is mechanically aligned for tomography experiments, the holder is the only variable left. Nevertheless, this work has shown that even a thoroughly used holder, which was previously bended, can be effectively operated. The described reproducibility tests give a way to assess the holder stability and quantify the limitations of the beam size for crystal tracking. It is worth to mention, that the minimum beam size obtained from these plots could give poor quality DPs for the tilt angles, in which the mismatch between beam position and the current crystal position is higher. To assert a general rule on how much a crystal needs to be illuminated to result in a good signal-to-noise DP is more difficult because it depends on multiple factors, such as crystallinity, thickness or beam influence. Therefore, every material has to be evaluated individually before starting a Fast-ADT acquisition. If the beam conditions are not suitable for the specimen characteristics in order to acquire a successful Fast-ADT dataset, the ADT method has to be used to carefully position the beam for each tilt angle. The use of the US4000 with Fast-ADT has provided very good results for both beam stable and sensitive materials, but its disadvantages are its long exposure times for good signal-to-noise data and long readout times. On the other hand, external and optical CCD cameras, such as the Stingray F-145B, have short read-out times but requires higher electron doses to excite the fluorescent screen. The results from the Barite crystal structure determinations show that the quality of reflection intensities from both cameras is high enough to retrieve accurate models. Nevertheless, R1 values, which quantify how good is the structural model by comparing the acquired and calculated reflection intensities, demonstrate that F145B data is significantly worse than the US4000 data, which was expected due to the high level of noise in the F-145B frames. Despite of the noisy DPs, the results from the disordered RUB-5 layered silicate show how frame-rate cameras has a huge advantage compared to long read-out CCDs in terms of acquisition time. At this point, the combination of a complementary metal-oxide-semiconductor (CMOS)-based camera (like the XF416 from TVIPS or the OneView from Gatan) or a direct detection camera (like the Medipix-based detectors from ASI or QD, the Quadro from Dectris or the EMPAD from ThermoFischer) with the Fast-ADT technique could be an optimum experimental setup to deal with any crystallographic analysis of the diffraction space. However, careful tests should be made to check the best pixel counting setting in order to supress most of the noise but still retain the weakest intensities. In case of direct detection cameras, additional studies on the influence of the bigger pixel sizes should be made to analyse the shape of the reflections, the sensitivity to diffuse scattering and the maximum attainable resolution before reflection overlapping. Another reason to switch to a more sensitive detector is to be able to acquire sufficient DPs from a beam sensitive material before the particle loses its crystallinity. The Fast-ADT acquisition on the DRED1 crystals finalized with an accumulated dose of about 100 e-/Å2. Last DPs from the Fast-ADT datasets reveal the significant decrease of intensity from reflections at high resolution in comparison to previous DPs. Re-check of populated DPs after the tomography acquisitions showed that the resolution for reflections with statistically meaningful intensity was reduced to about half, confirming the radiolysis

damage by the electron beam. Nevertheless, such accumulated dose number is enough to measure most of organic materials, being pure aliphatic compounds the most challenging ones [71]. In case of MOFs or proteins, few DPs could be collected as crystallinity is completely lost much before an accumulated dose of 100 e-/Å2 [72,73]. If relatively large crystals from these high sensitive materials are available, the crystal tracking file from Fast-ADT can be created to shift the beam at different positions of the same crystal through the tomography acquisition, thus increasing the number of DPs before the full degradation of the entire crystal [65,74]. In any case, the use of Fast-ADT and a high electronsensitive detector will be definitely advantageous for the acquisition of diffraction data from such compounds.

7. Conclusions Software has been developed for automatic and systematic acquisition of EDT datasets based on two tilt-scans of the goniometric stage, which has been called Fast-ADT. Two modules have been programmed in Digital Micrograph™ and Matlab environments to give access to most of the cameras used in TEMs. The new acquisition method can work in STEM mode as well as in TEM mode, although STEM is preferred. One significant difference compared to other reported methods is the use of NBD instead of SAED in order to successfully track single nanocrystals. In this way, a platform is provided that can be installed in any microscope, which can be readily used in other laboratories. BaSO4 single particles, were measured and analysed to prove, that the proposed acquisition method works for both crystal structure determination and dynamical refinement. The maximum difference of the atomic positions between the dynamically refined and Rietveld refined structures from the US4000 (2 pm) and the Stingray F-145B (6 pm) cameras show how reliable the acquisition method can be for accurate crystal description. RUB-5 layered-like crystals were used as a case example to assess disorder details. Interestingly, it has been shown how the use of an external optical CCD is suitable and, sometimes, advantageous for such crystallographic analyses. Finally, a beam sensitive material was successfully investigated, a new polymorph of the organic molecule DRED1 solved ab initio, revealing the 46 non-hydrogen independent atoms from the scattering density map. The merging of two datasets in order to obtain a high number of independent reflections as well as the fitting of a pseudo-Voigt function on the rocking curves for frame orientation refinement and reflection intensity extraction were key for this result. The determined crystal structure was finally refined by XRPD using the Rietveld method. These three material examples demonstrate the range of applicability and potential of the Fast-ADT technique.

Appendix I. Images projected at the tilted fluorescent screen are distorted. The screen and binocular position are designed in such a way that the view of the binocular is perpendicular to the screen, thus the screen is usually tilted about 45° with respect to the TEM horizontal plane. This angle introduces a √ along the vertical direction and a magnification difference effect between the top and bottom part of the image, which is enhanced by large field of views. 2D homographies (projective transformations) have to be applied in order to recover the non-distorted image. A homography is a non-singular, line preserving and projective application from an -space to itself, . It is represented by a square matrix of dimensions, δ, which has degrees of freedom. In this work the homography has to be applied to an image, thus is equal to 2 and homogeneous image coordinates need to be used. ( )

In this way, the more general form of the projective transformation is

⃗

( )

⃗

(

)( )

If a small field of view is used, like a small portion of the small fluorescent screen, the magnification difference effect is not significantly strong. This allows a simplification of the homography that has to be applied to correct the geometrical distortions; and . This is a type of homography called affine transformation that has 6 degrees of freedom, which includes 2 of scaling, 2 of rotation and 2 of translation. The simplified application for this particular case is: ( )

(

)( )

where the components correspond to the translation factors and the and scale ones.

components to the rotation

When a bigger area of the fluorescent screen is viewed, the image magnification is reduced and the projecting distortion clearly appears. This image distortion is shown as a vanishing point due to the higher difference of the image plane distance from the projector system between the top and the bottom of the acquired area. In other words, the optical system projects the image at the tilted screen with less magnification at the top of the image than at the bottom. For such case all the components of the projective transformations are needed in order to take into account two more degrees of freedom that correspond to the control of the vanishing point. The following step is to assign an appropriate intensity value to the new set of pixels created by the homography with respect to the distorted image. These intensity values are found using interpolation algorithms, which take the positions of the corrected image, calculate the theoretical positions in the original image using the applied correction factors and, finally, use their specific algorithm to assign the intensity. The interpolation algorithms in image processing are known as texture mapping and the most common ones are the nearest neighbour, bilinear and bicubic. The first one assigns the intensity value of a pixel on the corrected image with respect to the nearest pixel from the theoretical position at the distorted image. The bilinear interpolation uses the known nearer neighbour pixels for these theoretical positions. It weights each of these positions with the related computed pixel distance through two linear interpolations:

where ) is the coordinates of the distorted image transformed back to the reference system of the original image, the position , , and are coordinates of the original image that correspond to the four nearest pixels to the position ) , ) is the intensity value from the distorted image at the and position and ) is the interpolated intensity value for the corrected image position, ). In case of the bicubic interpolation, it uses the known nearer neighbour pixels from the theoretical position. Fig. A.1 shows the use of such interpolation algorithms on a non-saturated reflection acquired with the already explained experimental setup. Bilinear interpolation is preferred because it reduces the visual distortions when the scale factors are not integer values. Nearest neighbour interpolation let appear some pixels larger than others and bicubic interpolation gives smoother intensity values because of the

bigger area used to assign the intensity. Although bicubic could be interesting for smooth changes of intensity profiles, a small interpolation area is better to have a more step-like intensity profile for a reflection. Moreover, Fig. A.1C and A.1D shows that the difference between bilinear and bicubic is not so high, but bilinear is chosen because of its faster calculation and almost same result.

Figure A.1. A) Non-saturated and distorted spot from an oriented [011] Si sample, and corrected spot using B) nearest neighbour, C) bilinear and D) bicubic interpolation.

Appendix II. 

Fast-ADT: DM-based module

The advantage of this experimental setup is that it can be applied to any microscope with an available Gatan camera and the version 3 of DM. This software contains a C++-like object-oriented language that is able to take control of all camera parameters as well as the main functions of the TEM, thus a microscope control program can be developed inside DM. Although the code is interpreted instead of compiled like C++, the program is fast enough to synchronize the different routine parts. Beam settings and shifts are controlled through the functions available on the EMControl library, which may change the functions depending on the software version. For instance, the function called EMBeamShift() is present with the old program versions, which uses relative values for the coil currents, while the function called EMSetBeamShift() is in the newer versions and uses absolute ones. Our module uses the EMBeamShift() because it creates the crystal tracking file with relative values, thus it does not depend on the exact same position of the crystal to start a successful Fast-ADT acquisition. When STEM mode is considered, the DigiScan scanning generator unit provided by Gatan is used, as it allows an easy acquisition of fast and low-dose scanned images and beam positioning through a HAADF detector. In this case, the functions available on the DigiScan library are used for beam shift, because the change of the reference system between image coordinates and coils coordinates is internally calculated, which has to be calibrated in the TEM mode option. The EMSetBrightness() function is used to set the two different beam configurations; one fully spread in TEM (or fully focused in STEM) for imaging and one with the desired beam diameter for DP acquisition. Although the control of the tilt velocity is not currently available in the EMControl library, the alpha angle can be set with the function EMSetStageAlpha() or CMTiltGoniometerTo(), which as well depends on the DM version. Therefore an automatic and totally DM-integrated Fast-ADT acquisition can be performed following the sequential approach. This impressively decreases the experiment acquisition time from the traditional 1 hour to slightly more than 10 minutes, while using the same microscope tools and angular ranges. When the continuous approach is selected, the Temspy (ThermoFischer) or GonioTool (JEOL) program has to be used. Here the module is prepared to synchronize with these stage controller programs through a countdown for triggering the continuous tilt and frames acquisition (both in TEM and STEM mode).

As mentioned above, US1000 and US4000 Gatan cameras are tested for the present work. As the readout time for the 4k x 4k chip is around 7.7 seconds, binning 2 is applied to decrease it and obtain a faster retrieve of the DPs (around 3.6 seconds). In the case of the 2k chip, the read-out is lower (around 1.5 seconds) and binning 2 does not decrease it so strongly (around 1.1 seconds). Binning 4 on both cameras is used for the acquisition of the crystal tracking images, but it is not considered for DP acquisition because the spot intensity quality is then reduced. The pre-specimen shutter is active during the read-out time in order to decrease the electron dose on the sample. The advantage of such big chips is the recording of DPs at high resolution with well-resolved reflections to properly solve structures of more than 100 Å of unit-cell parameters, like pharmaceuticals or proteins.



Fast-ADT: Matlab-based module

The optical CCD camera used in this setup is the Stingray F-145B provided by Allied Vision GmbH, which is plugged to the PC through a Firewire interface of a PCI card. The sensor has 1388 x 1038 pixels with a cell size of 6.45 μm x 6.45 μm and a 14-bit analog-to-digital converter. This allows the acquisition of 14bit data at a maximum frame rate of 16 fps at full resolution. Matlab was used in this setting as it has an easy accessible user interface to program the needed features; full control of the camera as well as signal generation and triggering through the hardware unit of the ASTAR™ system. The environment is fast enough to display and acquire geometrically corrected DPs in real-time. Appendix I gives the mathematical details on how the geometrical correction is implemented in the program. The different binocular geometries by the TEM manufacturers require the design of specific and different supports in order to place the camera and acquire images from the viewing chamber. Moreover, the C-mount optical lens for the camera also changes depending on the focused screen and the desired field of view. If the camera is focused on the small fluorescent screen, an optical lens of 50 mm of focal distance is enough to cover the whole area of the screen. If the camera is focused on the big fluorescent screen and a large field of view is needed, an optical lens of 25 mm of focal distance may be used. The big screen case scenario also requires a closed diaphragm for the optical lens; otherwise the focus depth is too small that the whole frame cannot be properly focused (Fig. A.2). This is because of the relative inclination of the screen with respect to the plane of the sensor size, which on the small fluorescent screen is almost 0°. In JEOL microscopes the big fluorescent screen can be tilted 7 degrees with respect to the horizontal in order to minimize this image defocusing while in ThermoFischer microscopes is not possible to do so at the present moment. Nevertheless, the 50 mm optical lens focused on the small screen was used in this setup. If we consider a camera length that provides a resolution of 0.8 Å and a reflection size of 15 pixels for proper intensity integration, the maximum measurable cell-parameter is around 45 Å, which is suitable for most of the materials to be studied with this setup. The big screen and higher size sensors may be considered in order to deal with bigger unitcells. It is worth to say that a fast decay phosphor (Type P46; Gd2O2S:Tb) instead of the high conversion efficiency and spatial resolution one (Type P43; Y3Al5O12:Ce) should be used to vanish as fast as possible the remaining reflections when high tilt velocities are used.

Figure A.2. DPs of a tetragonal structure near [001] acquired and geometrically corrected by the Fast-ADT software with different focus depths determined by the diaphragm aperture of the Stingray F-145B on the big fluorescent screen of a Tecnai Spirit TEM.

The shift of the electron beam is generated through an external voltage output device. The Fast-ADT software sends digital signals to the USB-3101FS voltage generator from the Measurement Computing Company, which in turn sends analogue signals to the P1000 unit provided by NanoMegas. Then, this external signal generator merges the shifting signal with the precession signal and sends the mixed one to the deflector and image shift coils of the microscope. The beam tilt and beam shift alignments of the microscope need to be taken care of in order to set a beam that shifts without tilt and tilt without shift. The continuous tilt of the stage is produced by means of the continuous tilt controller shown in Fig. A.3. The electronic circuits from Shi et al. [29] are used to short-cut the resistance values produced by the pressure-sensor buttons from the control pad. However, the switcher that enables the use of the external resistances (properly calibrated for different tilt velocities) is replaced by a small electronic board with a relay that is energized through one of the available outputs of the USB-3101FS. This controller can be used on any FEI microscope that uses the same control pad as the modern Tecnai series. It could be similarly implemented in JEOL microscopes after some hardware modifications.

Figure A.3. The continuous tilt controller and the respective connections to the left control pad of the Tecnai Spirit TEM.

Appendix III. DRED1 powder was purchased from Sigma-Aldrich and recrystallized with toluene in order to improve the crystallinity. Fig. S7 shows the XRPDs of the as-made and recrystallized DRED1 for comparison. The resulting powder was dispersed in n-hexane and sprayed on a carbon-coated copper grid. Mild illumination (Spot size 8, gun lens 8 and 10-μm condenser aperture) was set in order to produce a 200nm beam with an electron dose rate of 0.21 e-/Å2·s (FWHM, full-width at half-maximum) for the - 2 acquisition of DPs. A 5-nm beam with an electron dose rate of 293 e /Å ·s was chosen for the STEM images. A US4000 CCD was used with binning 2 and an exposure time of 4 seconds for each DP. The crystal tracking file was generated from STEM images (1024 x 1024) with a dwell time of 4 μs per pixel. A final accumulated dose of 102 e-/Å2 was obtained per dataset. Initially, three datasets were acquired at room temperature using a Fischione tomography holder. Later, three other crystals were measured with a Gatan cryotransfer tomography holder (model 914) at liquid N2 temperature in order to increase the stability of the sample [75]. A large camera length of 1.2 meters was chosen to take advantage of the full CCD field-of-view as different crystals scatter up to 0.8 Å at maximum. XRPD measurements were carried on a STOE transmission powder diffraction system (STADI P) equipped with a Ge (111) monochromator (Cu-Kα1 radiation, λ = 1.54056 Å) and a linear position-sensitive director. A borosilicate glass capillary of 1 mm in diameter was filled with the microcrystalline powder, which was cooled down to 140 K during the measurement by an Oxford Cryostream 700Plus.

REFERENCES [1] W.L. Bragg, The Structure of Some Crystals as Indicated by Their Diffraction of X-rays, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences. 89 (1913) 248–277. https://doi.org/10.1098/rspa.1913.0083. [2] D. Bowden, Y. Krysiak, L. Palatinus, D. Tsivoulas, S. Plana-Ruiz, E. Sarakinou, U. Kolb, D. Stewart, M. Preuss, A high-strength silicide phase in a stainless steel alloy designed for wear-resistant applications, Nature Communications. 9 (2018) 1374. https://doi.org/10.1038/s41467-018-03875-9. [3] A.S. Eggeman, R. Krakow, P.A. Midgley, Scanning precession electron tomography for three-dimensional nanoscale orientation imaging and crystallographic analysis, Nature Communications. 6 (2015). https://doi.org/10.1038/ncomms8267. [4] S. Nicolopoulos, P.P. Das, P.J. Bereciartua, F. Karavasili, N. Zacharias, A. Gómez Pérez, A.S. Galanis, E.F. Rauch, R. Arenal, J. Portillo, J. Roqué-Rosell, M. Kollia, I. Margiolaki, Novel characterization techniques for cultural heritage using a TEM orientation imaging in combination with 3D precession diffraction tomography: a case study of green and white ancient Roman glass tesserae, Heritage Science. 6 (2018). https://doi.org/10.1186/s40494-018-0229-7. [5] E. Mugnaioli, Closing the gap between electron and X-ray crystallography, Acta Cryst B. 71 (2015) 737–739. https://doi.org/10.1107/S2052520615022441. [6] J. Gjønnes, A.F. Moodie, Extinction conditions in the dynamic theory of electron diffraction, Acta Cryst. 19 (1965) 65–67. https://doi.org/10.1107/S0365110X65002773. [7] W. Sinkler, L.D. Marks, D.D. Edwards, T.O. Mason, K.R. Poeppelmeier, Z. Hu, J.D. Jorgensen, Determination of Oxygen Atomic Positions in a Ga–In–Sn–O Ceramic Using Direct Methods and Electron Diffraction, Journal of Solid State Chemistry. 136 (1998) 145–149. https://doi.org/10.1006/jssc.1998.7804. [8] R. Vincent, P.A. Midgley, Double conical beam-rocking system for measurement of integrated electron diffraction intensities, Ultramicroscopy. 53 (1994) 271– 282. https://doi.org/10.1016/0304-3991(94)90039-6. [9] J.P. Morniroli, A. Redjaïmia, S. Nicolopoulos, Contribution of electron precession to the identification of the space group from microdiffraction patterns, Ultramicroscopy. 107 (2007) 514–522. https://doi.org/10.1016/j.ultramic.2006.03.013. [10] J.-P. Morniroli, P. Stadelmann, G. Ji, S. Nicolopoulos, The symmetry of precession electron diffraction patterns, Journal of Microscopy. 237 (2010) 511–515. https://doi.org/10.1111/j.1365-2818.2009.03311.x. [11] H.M. Rietveld, The Rietveld method, Phys. Scr. 89 (2014) 098002. https://doi.org/10.1088/0031-8949/89/9/098002. [12] L. Palatinus, V. Petříček, C.A. Corrêa, Structure refinement using precession electron diffraction tomography and dynamical diffraction: theory and implementation, Acta Crystallographica Section A Foundations and Advances. 71 (2015) 235–244. https://doi.org/10.1107/S2053273315001266. [13] L. Palatinus, C.A. Corrêa, G. Steciuk, D. Jacob, P. Roussel, P. Boullay, M. Klementová, M. Gemmi, J. Kopeček, M.C. Domeneghetti, F. Cámara, V. Petříček, Structure refinement using precession electron diffraction tomography and dynamical diffraction: tests on experimental data, Acta Crystallographica

Section B Structural Science, Crystal Engineering and Materials. 71 (2015) 740–751. https://doi.org/10.1107/S2052520615017023. [14] U. Kolb, T. Gorelik, C. Kübel, M.T. Otten, D. Hubert, Towards automated diffraction tomography: Part I—Data acquisition, Ultramicroscopy. 107 (2007) 507–513. https://doi.org/10.1016/j.ultramic.2006.10.007. [15] M. Gemmi, E. Mugnaioli, T.E. Gorelik, U. Kolb, L. Palatinus, P. Boullay, S. Hovmöller, J.P. Abrahams, 3D Electron Diffraction: The Nanocrystallography Revolution, ACS Cent. Sci. (2019). https://doi.org/10.1021/acscentsci.9b00394. [16] U. Kolb, T. Gorelik, M.T. Otten, Towards automated diffraction tomography. Part II—Cell parameter determination, Ultramicroscopy. 108 (2008) 763–772. https://doi.org/10.1016/j.ultramic.2007.12.002. [17] U. Kolb, E. Mugnaioli, T.E. Gorelik, Automated electron diffraction tomography – a new tool for nano crystal structure analysis, Crystal Research and Technology. 46 (2011) 542–554. https://doi.org/10.1002/crat.201100036. [18] E. Mugnaioli, T. Gorelik, U. Kolb, “Ab initio” structure solution from electron diffraction data obtained by a combination of automated diffraction tomography and precession technique, Ultramicroscopy. 109 (2009) 758–765. https://doi.org/10.1016/j.ultramic.2009.01.011. [19] D. Zhang, P. Oleynikov, S. Hovmöller, X. Zou, Collecting 3D electron diffraction data by the rotation method, Zeitschrift Für Kristallographie. 225 (2010). https://doi.org/10.1524/zkri.2010.1202. [20] M. Gemmi, M.G.I. La Placa, A.S. Galanis, E.F. Rauch, S. Nicolopoulos, Fast electron diffraction tomography, J Appl Cryst. 48 (2015) 718–727. https://doi.org/10.1107/S1600576715004604. [21] B.L. Nannenga, D. Shi, A.G.W. Leslie, T. Gonen, High-resolution structure determination by continuous-rotation data collection in MicroED, Nature Methods. 11 (2014) 927–930. https://doi.org/10.1038/nmeth.3043. [22] I. Nederlof, E. van Genderen, Y.-W. Li, J.P. Abrahams, A Medipix quantum area detector allows rotation electron diffraction data collection from submicrometre three-dimensional protein crystals, Acta Crystallographica Section D Biological Crystallography. 69 (2013) 1223–1230. https://doi.org/10.1107/S0907444913009700. [23] M.O. Cichocka, J. Ångström, B. Wang, X. Zou, S. Smeets, High-throughput continuous rotation electron diffraction data acquisition via software automation, Journal of Applied Crystallography. 51 (2018) 1652–1661. https://doi.org/10.1107/S1600576718015145. [24] D.B. Williams, C.B. Carter, Transmission electron microscopy: a textbook for materials science, 2nd ed, Springer, New York, 2008. [25] L. Palatinus, D. Jacob, P. Cuvillier, M. Klementová, W. Sinkler, L.D. Marks, Structure refinement from precession electron diffraction data, Acta Cryst A. 69 (2013) 171–188. https://doi.org/10.1107/S010876731204946X. [26] H. Klein, J. David, The quality of precession electron diffraction data is higher than necessary for structure solution of unknown crystalline phases, Acta Crystallographica Section A Foundations of Crystallography. 67 (2011) 297–302. https://doi.org/10.1107/S0108767311006581. [27] A.S. Eggeman, P.A. Midgley, Refining structures against reflection rank: an alternative metric for electron crystallography, Acta Crystallographica Section

A Foundations of Crystallography. 68 (2012) 352–358. https://doi.org/10.1107/S0108767312007234. [28] S. Smeets, W. Wan, Serial electron crystallography: merging diffraction data through rank aggregation, Journal of Applied Crystallography. 50 (2017) 885– 892. https://doi.org/10.1107/S1600576717005854. [29] D. Shi, B.L. Nannenga, M.J. de la Cruz, J. Liu, S. Sawtelle, G. Calero, F.E. Reyes, J. Hattne, T. Gonen, The collection of MicroED data for macromolecular crystallography, Nat Protoc. 11 (2016) 895–904. https://doi.org/10.1038/nprot.2016.046. [30] E. van Genderen, M.T.B. Clabbers, P.P. Das, A. Stewart, I. Nederlof, K.C. Barentsen, Q. Portillo, N.S. Pannu, S. Nicolopoulos, T. Gruene, J.P. Abrahams, Ab initio structure determination of nanocrystals of organic pharmaceutical compounds by electron diffraction at room temperature using a Timepix quantum area direct electron detector, Acta Cryst A. 72 (2016) 236–242. https://doi.org/10.1107/S2053273315022500. [31] U. Kolb, T.E. Gorelik, E. Mugnaioli, A. Stewart, Structural Characterization of Organics Using Manual and Automated Electron Diffraction, Polymer Reviews. 50 (2010) 385–409. https://doi.org/10.1080/15583724.2010.494238. [32] T.E. Gorelik, J. van de Streek, A.F.M. Kilbinger, G. Brunklaus, U. Kolb, Ab-initio crystal structure analysis and refinement approaches of oligo p -benzamides based on electron diffraction data, Acta Crystallographica Section B Structural Science. 68 (2012) 171–181. https://doi.org/10.1107/S0108768112003138. [33] T. Rhauderwiek, H. Zhao, P. Hirschle, M. Döblinger, B. Bueken, H. Reinsch, D.D. Vos, S. Wuttke, U. Kolb, N. Stock, Highly stable and porous porphyrinbased zirconium and hafnium phosphonates – electron crystallography as an important tool for structure elucidation, Chemical Science. 9 (2018) 5467– 5478. https://doi.org/10.1039/C8SC01533C. [34] S. Plana-Ruiz, J. Portillo, S. Estradé, F. Peiró, U. Kolb, S. Nicolopoulos, Quasiparallel precession diffraction: Alignment method for scanning transmission electron microscopes, Ultramicroscopy. 193 (2018) 39–51. https://doi.org/10.1016/j.ultramic.2018.06.005. [35] M. Kluenker, M.N. Tahir, R. Dören, M. Deuker, P. Komforth, S. Plana-Ruiz, B. Barton, S.I. Shylin, V. Ksenofontov, M. Panthöfer, N. Wiesmann, J. Herzberger, A. Möller, H. Frey, J. Brieger, U. Kolb, W. Tremel, Iron Oxide Superparticles with Enhanced MRI Performance by Solution Phase Epitaxial Growth, Chemistry of Materials. 30 (2018) 4277–4288. https://doi.org/10.1021/acs.chemmater.8b01128. [36] Z. Zou, W.J.E.M. Habraken, G. Matveeva, A.C.S. Jensen, L. Bertinetti, M.A. Hood, C. Sun, P.U.P.A. Gilbert, I. Polishchuk, B. Pokroy, J. Mahamid, Y. Politi, S. Weiner, P. Werner, S. Bette, R. Dinnebier, U. Kolb, E. Zolotoyabko, P. Fratzl, A hydrated crystalline calcium carbonate phase: Calcium carbonate hemihydrate, Science. 363 (2019) 396–400. https://doi.org/10.1126/science.aav0210. [37] U. Ziese, A.H. Janssen, J.-L. Murk, W.J.C. Geerts, T. Van der Krift, A. j. Verkleij, A. j. Koster, Automated high-throughput electron tomography by precalibration of image shifts, Journal of Microscopy. 205 (2002) 187–200. https://doi.org/10.1046/j.0022-2720.2001.00987.x. [38] A.M. Airy, On the diffraction of an object-glass with circular aperture, Transaction of the Cambridge Philosophical Society. 5 (1834) 283–291.

[39] D. Alloyeau, C. Ricolleau, C. Mottet, T. Oikawa, C. Langlois, Y. Le Bouar, N. Braidy, A. Loiseau, Size and shape effects on the order–disorder phase transition in CoPt nanoparticles, Nature Materials. 8 (2009) 940–946. https://doi.org/10.1038/nmat2574. [40] K.J. Ganesh, M. Kawasaki, J.P. Zhou, P.J. Ferreira, D-STEM: A Parallel Electron Diffraction Technique Applied to Nanomaterials, Microscopy and Microanalysis. 16 (2010) 614–621. https://doi.org/10.1017/S1431927610000334. [41] H. Zhao, Y. Krysiak, K. Hoffmann, B. Barton, L. Molina-Luna, R.B. Neder, H.-J. Kleebe, T.M. Gesing, H. Schneider, R.X. Fischer, U. Kolb, Elucidating structural order and disorder phenomena in mullite-type Al4B2O9 by automated electron diffraction tomography, Journal of Solid State Chemistry. 249 (2017) 114–123. https://doi.org/10.1016/j.jssc.2017.02.023. [42] J. Portillo, E.F. Rauch, S. Nicolopoulos, M. Gemmi, D. Bultreys, Precession Electron Diffraction Assisted Orientation Mapping in the Transmission Electron Microscope, Materials Science Forum. (2010). https://doi.org/10.4028/www.scientific.net/MSF.644.1. [43] D. Viladot, M. Véron, M. Gemmi, F. Peiró, J. Portillo, S. Estradé, J. Mendoza, N. Llorca‐Isern, S. Nicolopoulos, Orientation and phase mapping in the transmission electron microscope using precession-assisted diffraction spot recognition: state-of-the-art results, Journal of Microscopy. 252 (2013) 23–34. https://doi.org/10.1111/jmi.12065. [44] Y. Krysiak, B. Marler, B. Barton, S. Plana-Ruiz, H. Gies, R.B. Neder, U. Kolb, New zeolite-like RUB-5 and its related hydrous layer silicate RUB-6 structurally characterised by electron microscopy, IUCrJ. (2020). [45] A.A. Colville, K. Staudhammer, A refinement of the structure of barite, Mineralogical Notes. 52 (1967) 1877. [46] R.J. Hill, A further refinement of the barite structure, The Canadian Mineralogist. 15 (1977) 522–526. [47] S.D. Jacobsen, J.R. Smyth, R.J. Swope, R.T. Downs, Rigid-body character of the SO 4 groups in celestine, anglesite and barite, The Canadian Mineralogist. 36 (1998) 1053–1060. [48] D. Waterman, G. Evans, Estimation of errors in diffraction data measured by CCD area detectors, J Appl Cryst. 43 (2010) 1356–1371. https://doi.org/10.1107/S0021889810033418. [49] U. Kolb, Y. Krysiak, S. Plana-Ruiz, Automated electron diffraction tomography – development and applications, Acta Cryst B. 75 (2019) 463–474. https://doi.org/10.1107/S2052520619006711. [50] L. Palatinus, P. Brázda, M. Jelínek, J. Hrdá, G. Steciuk, M. Klementová, Specifics of the data processing of precession electron diffraction tomography data and their implementation in the program PETS2.0, Acta Cryst B. 75 (2019) 512–522. https://doi.org/10.1107/S2052520619007534. [51] M.C. Burla, R. Caliandro, B. Carrozzini, G.L. Cascarano, C. Cuocci, C. Giacovazzo, M. Mallamo, A. Mazzone, G. Polidori, Crystal structure determination and refinement via SIR2014, J Appl Cryst. 48 (2015) 306–309. https://doi.org/10.1107/S1600576715001132. [52] V. Petříček, M. Dušek, L. Palatinus, Crystallographic Computing System JANA2006: General features, Zeitschrift Für Kristallographie - Crystalline Materials. 229 (2014) 345–352. https://doi.org/10.1515/zkri-2014-1737.

[53] Y. Krysiak, B. Barton, B. Marler, R.B. Neder, U. Kolb, Ab initio structure determination and quantitative disorder analysis on nanoparticles by electron diffraction tomography, Acta Cryst A. 74 (2018) 93–101. https://doi.org/10.1107/S2053273317018277. [54] Z. Sekkat, M. Dumont, Polarization effects in photoisomerization of azo dyes in polymeric films, Applied Physics B Photophysics and Laser Chemistry. 53 (1991) 121–123. https://doi.org/10.1007/BF00325543. [55] G. Schottner, S. Hofacker, J. Sandrock, Organic Dye-Doped Hybrid Sol-Gel Coatings and Pigments – Synthesis, Structural Elucidation and Application, MRS Proceedings. 519 (1998) 251. https://doi.org/10.1557/PROC-519-251. [56] A. Miniewicz, S. Bartkiewicz, J. Sworakowski, J.A. Giacometti, M.M. Costa, On optical phase conjugation in polystyrene films containing the azobenzene dye Disperse Red 1, Pure and Applied Optics: Journal of the European Optical Society Part A. 7 (1998) 709–721. https://doi.org/10.1088/09639659/7/4/008. [57] Y. Cui, M. Wang, L. Chen, G. Qian, Synthesis and spectroscopic characterization of an alkoxysilane dye containing C. I. Disperse Red 1, Dyes and Pigments. 62 (2004) 43–47. https://doi.org/10.1016/j.dyepig.2003.11.009. [58] P.G. Lacroix, I. Malfant, G. Iftime, A.C. Razus, K. Nakatani, J.A. Delaire, AzoAzulene Derivatives as Second-Order Nonlinear Optical Chromophores, Chemistry – A European Journal. 6 (2000) 2599–2608. https://doi.org/10.1002/1521-3765(20000717)6:14<2599::AIDCHEM2599>3.0.CO;2-G. [59] N.K. Nath, L. Pejov, S.M. Nichols, C. Hu, N. Saleh, B. Kahr, P. Naumov, Model for Photoinduced Bending of Slender Molecular Crystals, Journal of the American Chemical Society. 136 (2014) 2757–2766. https://doi.org/10.1021/ja4101497. [60] D. Louër, A. Boultif, Some further considerations in powder diffraction pattern indexing with the dichotomy method, Powder Diffraction. 29 (2014) S7–S12. https://doi.org/10.1017/S0885715614000906. [61] G.S. Pawley, Unit-cell refinement from powder diffraction scans, J Appl Cryst. 14 (1981) 357–361. https://doi.org/10.1107/S0021889881009618. [62] D.W.M. Hofmann, Fast estimation of crystal densities, Acta Cryst B. 58 (2002) 489–493. https://doi.org/10.1107/S0108768101021814. [63] L. Palatinus, G. Chapuis, SUPERFLIP – a computer program for the solution of crystal structures by charge flipping in arbitrary dimensions, J Appl Cryst. 40 (2007) 786–790. https://doi.org/10.1107/S0021889807029238. [64] A.A. Coelho, TOPAS and TOPAS-Academic: an optimization program integrating computer algebra and crystallographic objects written in C++, Journal of Applied Crystallography. 51 (2018) 210–218. https://doi.org/10.1107/S1600576718000183. [65] P. Brázda, L. Palatinus, M. Babor, Electron diffraction determines molecular absolute configuration in a pharmaceutical nanocrystal, Science. 364 (2019) 667–669. https://doi.org/10.1126/science.aaw2560. [66] K. Momma, F. Izumi, VESTA 3 for three-dimensional visualization of crystal, volumetric and morphology data, J Appl Cryst. 44 (2011) 1272–1276. https://doi.org/10.1107/S0021889811038970.

[67] Stef Smeets, Bin Wang, Magdalena O. Cichocka, Jonas Ångström, Wei Wan, Instamatic, Zenodo, 2018. https://doi.org/10.5281/zenodo.2026774. [68] B. Wang, T. Rhauderwiek, A.K. Inge, H. Xu, T. Yang, Z. Huang, N. Stock, X. Zou, A Porous Cobalt Tetraphosphonate Metal–Organic Framework: Accurate Structure and Guest Molecule Location Determined by Continuous-Rotation Electron Diffraction, Chemistry – A European Journal. 24 (2018) 17429–17433. https://doi.org/10.1002/chem.201804133. [69] C. Zhang, E. Kapaca, J. Li, Y. Liu, X. Yi, A. Zheng, X. Zou, J. Jiang, J. Yu, An ExtraLarge-Pore Zeolite with 24×8×8-Ring Channels Using a Structure-Directing Agent Derived from Traditional Chinese Medicine, Angewandte Chemie International Edition. 57 (2018) 6486–6490. https://doi.org/10.1002/anie.201801386. [70] J.S. Barnard, D.N. Johnstone, P.A. Midgley, High-resolution scanning precession electron diffraction: Alignment and spatial resolution, Ultramicroscopy. 174 (2017) 79–88. https://doi.org/10.1016/j.ultramic.2016.12.018. [71] R.F. Egerton, Control of radiation damage in the TEM, Ultramicroscopy. 127 (2013) 100–108. https://doi.org/10.1016/j.ultramic.2012.07.006. [72] Y. Zhu, J. Ciston, B. Zheng, X. Miao, C. Czarnik, Y. Pan, R. Sougrat, Z. Lai, C.-E. Hsiung, K. Yao, I. Pinnau, M. Pan, Y. Han, Unravelling surface and interfacial structures of a metal–organic framework by transmission electron microscopy, Nature Mater. 16 (2017) 532–536. https://doi.org/10.1038/nmat4852. [73] A. Lanza, E. e, E. Mugnaioli, V. Cappello, G. Garau, M. Gemmi, Nanobeam precession-assisted 3D electron diffraction reveals a new polymorph of hen egg-white lysozyme, IUCrJ. 6 (2019) 178–188. https://doi.org/10.1107/S2052252518017657. [74] L. Palatinus, P. Brázda, P. Boullay, O. Perez, M. Klementová, S. Petit, V. Eigner, M. Zaarour, S. Mintova, Hydrogen positions in single nanocrystals revealed by electron diffraction, Science. 355 (2017) 166–169. https://doi.org/10.1126/science.aak9652. [75] International Experimental Study Group, Cryoprotection in electron microscopy, Journal of Microscopy. 141 (1986) 385–391. https://doi.org/10.1111/j.1365-2818.1986.tb02731.x.

Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

☐The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: