Introduction

The performance of a functional material is often determined by only a small percentage of its surface sites and their atomic configuration. This phenomenon, known as structure sensitivity, describes the relationship between the fraction of exposed crystal surfaces and the rate of conversion and is well known in the field of heterogeneous catalysis1,2,3. For example, in supported metal nanoparticle catalysts, the variation of metal nanoparticle size results in the exposure of a different fraction of active surface sites, leading to size-dependent performance4.

Analogously, it is often observed for porous functional materials, such as metal-organic frameworks (MOFs), that their functionality is strongly dependent on their outer surface, despite their inner porosity5. As a result, evidence for the structure sensitivity of porous functional materials can be found in the literature6,7. For example, Pang and coworkers showed by ex situ scanning electron microscopy (SEM) analysis that when ZIF-8 crystals were exposed to acidic SO2 gas, the external surface of the {100} facet was more stable than the {110} facet8. However, such an ex situ technique is not able to describe the guest–host interaction during gas exposure. Additionally, the performance of undercoordinated crystal surface edges and corners was not considered8, despite the fact that these high-energy crystal planes often present highly important sorption, and/or conversion, sites on functional materials9,10.

Undercoordinated sites are purposely introduced in defect engineering approaches, for example, through the incorporation of defective linkers11. The resulting undercoordinated metal centers can behave as Lewis acid sites with potentially increased activity towards gas sorption and/or conversion12. The localized integration of defects can be characterized by highly sensitive techniques, such as high-resolution transmission electron microscopy (HRTEM), scanning electron diffraction (SED), or atom probe tomography (APT)13,14,20,21.

Infrared spectroscopy is an ideal technique to describe guest–host interactions22, but it suffers from low spatial resolution. Moreover, the low ratio of crystallite surface vs. bulk atoms may result in a loss of information on the external surface performance of porous functional materials. Tip-enhanced Raman spectroscopy (TERS) and photo-induced force microscopy (PiFM), are AFM-based vibrational techniques able to circumvent these limitations, with a spatial resolution down to the nanoscale (e.g., 5 nm for PiFM)23,24. Both techniques are non-destructive, label-free, can be performed in situ, and can be used to study a broad range of materials25. However, TERS suffers from arduous tip design- and preparation protocols, as well as measurement stability issues, which are not encountered when using PiFM. We recently showcased how PiFM can be used to study guest–host interactions in situ, looking at water sorption and induced defect formation on the surface of archetypical MOF thin films26. However, studying structure sensitivity requires the use of well-defined systems, where different crystal surface terminations can be imaged simultaneously under the same conditions.

In this work, we used in situ PiFM, in combination with density functional theory (DFT) calculations, to unravel structure sensitivity in gas sorption and conversion on the external surface of a microcrystalline MOF material in unprecedented detail (Fig. 1). Here, we show that with this toolbox we can purposely study highly functional, external surfaces of MOF crystal planes during gas exposure, which is not possible through conventional, bulk techniques. We have used surface-anchored crystals of ZIF-8, a MOF with exceptional stability and potential applications in catalysis27,Full size image

The nanoislands observed in pristine ZIF-8 microcrystals may be explained in three ways: (i) as an artifact induced by overall lower PiFM intensity; (ii) as a result of heterogeneity of ZIF-8 (e.g., a presence of different crystal surface terminations, vide infra); or (iii) as coverage dependent patterns of FA over homogeneous surfaces as a result of interactions between adsorbed FA molecules, similarly to what is observed for e.g., CO adsorption on metal surfaces41. To rule out factor (i), we took point spectra in and out of the nanoislands, which showed comparable ZIF-8 vibration intensities, but differences in the FA bands (Fig. 4c and S20). Furthermore, we performed IR map** of the same area, but with the laser tuned to an aromatic vibration of the ZIF-8 framework (νC=N 1590 cm−1, Fig. 4b and Supplementary Fig. 21), which showed an inverted image. This showed both the surface sensitivity and validity of the PiFM method and that the external surface is intrinsically heterogeneous, thus supporting hypothesis (ii) over the others (which assume homogeneous surfaces).

To study these heterogeneous planes and to link gas sorption behavior to specific external surface sites within these planes, we modeled three representative ZIF-8 crystallite surfaces using DFT calculations, namely the {100} and {110} facet planes, and a high-index {310} corner plane (Fig. 3d–h). The {100} and {110} planes were chosen because of their high relative crystal surface coverage, and because of the nanoislands observed within these facets. The {310} plane was selected as representative of minority, high-energy terminations showing alternate behavior to formaldehyde vapor. It is crucial to realize that for each of the {100}, {110}, and {310} planes, different surfaces can be constructed by varying the slicing height (Fig. 3d). These surfaces can be grouped into motifs according to the coordination environment of the Zn atom as well as the orientation of terminal ligands (Fig. 3e and Supplementary Fig. 4). These motifs differ in the density of Zn atoms per surface area as well as in the number of Zn-N bonds cleaved, giving rise to various undercoordinated sites. The computed surface energies for the most stable {100}, {110}, and {310} terminations were found to be −3.21, −4.91, and −2.67 meV/Å2, respectively, confirming the higher-energy nature of higher index planes42.

However, high-energy cuts of, for example, the {100} and {110} planes with energies of 3.45 and 0.62 meV/Å2, respectively, were also found. Because of the moderate differences in surface energies between low and high energy cuts of the {100} and {110} planes, multiple terminations may co-exist on the facet surfaces, for example, as a result of entropic factors. Therefore, the formation of high-energy planes can be favored during synthesis conditions (rather than the 0 K modeling conditions), leading to the expression of the observed nanoislands43. Note that negative values for the most stable surface energies were found as a result of the use of a saturated surface model (Supplementary Methods). Using unsaturated surface models did not change our conclusions regarding the relative stabilities of different surface orientations (Supplementary Table 2).

We subsequently modeled FA adsorption on these representative ZIF-8 terminations. In general, we found that the formaldehyde molecules chemisorb on ZIF-8 by forming a carbon adduct with a nitrogen atom from the linker, while the oxygen atom adsorbs on a Zn center, i.e., via the formation of covalent OFA-ZnZIF and CFA-NZIF bonds5,44. Such chemisorption is consistent with the absence of molecularly adsorbed formaldehyde in the infrared spectra (Fig. 2d). Additionally, we found that (at 0 K) FA adsorption is an endothermic process (SI). Overall, the trend in FA binding on the most stable cleavages is consistent with experimental findings showing the order {310} > {110} > {100} (Fig. 3f–h and Supplementary Fig. 5). As the most reactive surface termination, we identify motif ІІІ, which expresses an isolated Zn site with two terminal linkers, one protonated and one unprotonated (Fig. 3e and Supplementary Fig. 5), which can be found on {310} edges as well as on high-energy cleavages of {100}. Further inspection of the high-energy cleavages showed their stronger FA binding behavior compared to their low-energy counterparts (e.g., 26.8 versus 74.1 kJ/mol for {100} and 47.1 versus 57.7 kJ/mol for {110}, (Supplementary Fig. 5). Overall, our calculations supported the experimental observations of inter- and intra-plane structure-sensitive sorption, dictated by the formation of nanodomains exposing high surface energy terminations, as described by hypothesis (ii).

Structure-sensitive defect engineering

To further explore site-specific FA adsorption, we purposefully incorporated 10% of deuterated pyrrole defect linker in the ZIF-8 crystals to enrich the surface chemistry with undersaturated metal sites (Fig. 1a). Deuterated pyrrole linkers were used because of their distinct spectrum compared to the 2-methylimidazole linkers, which allows one to observe the incorporation of defects with PiFM. Atomic Force Microscopy (AFM) images of the defective crystals showed a change in the expression of its crystal planes: an increase in {100}:{110} facet size ratio was observed (Fig. 4a and Supplementary Figs. 9, 10). This suggested a change in relative plane energies upon defect engineering, which we attribute to strain relaxation of the external surface terminations45. These findings are supported by DFT results, which show that an exchange of imidazole linker with pyrrole is thermodynamically most favorable on the high energy cut of the {100} plane (Supplementary Table 3), thereby effectively lowering a difference in surface energies between {100} and {110} planes, resulting in favored {100} expression.

Fig. 4: Nanoscale visualization of heterogeneity within crystal planes of ZIF-8.
figure 4

a Representative AFM image of pyrrole-d5 defect-engineered SURZIF-8, showing a higher ratio of {100}/{110} facets compared to the pristine crystals (Fig. 2). b Spectroscopic evidence for defect linker incorporation was found when comparing the crystal-averaged spectra of a pristine (solid trace) and a defective (dashed trace) crystal. The defective ZIF-8 spectrum showed δCD and νC=N vibrations at 960 and 885 cm−1 corresponding to the defect linker, and δOH vibrations at 840 and 790 cm−1 corresponding to defect-induced Zn-OH sites, see inset. c Using peak ratios of the defective linker (885/1590 cm−1) and Zn-OH (790/1590 cm−1) sites, we show that the concentrations of such sites are larger for high-energy planes. Crystal-averaged peak ratios are given by solid lines. d A clustered 200 × 200 nm2 hyperspectral image made through principle component analysis (PCA) and clustering where each color refers to a cluster of similar spectral identity, and e the corresponding spectra of each cluster of defective ZIF-8 under N2 showing one {100} facet and two {110} facets. Clusters were calculated using principal component analysis (PCA) and clustering, to visualize ZIF-8 defect linker distribution (δCD and νC=N in e). f Bubble plot where bubble size corresponds to the fraction of a facet attributed to a certain cluster of the corresponding color, and the bubble position on the y-axis indicates the relative defect concentration. For clarity, the two main contributing clusters were selected for each facet. Facet-averaged defect concentrations are marked with a star. This plot shows differences in relative defect concentrations among facets with comparable surface energies. However, it also shows that strong concentration differences between each facet and their highly localized intra-facet domains can be found.

Full size image

To inspect whether defect linker incorporation can be linked to surface energies, we measured a 1 × 1 µm2 hyperspectral image of a defective crystal in N2. Crystal-averaged IR spectra showed the vibrational fingerprint of pyrrole-d5 by two IR bands at 960 and 885 cm−1, corresponding to δCD and νC=N, respectively (Fig. 4b and Supplementary Fig. 22)46. Further inspection of the spectral data also revealed the presence of infrared bands at 840 and 790 cm−1, corresponding to two δOH vibrations of Zn-OH2 sites47. These bands indicate the enrichment of the external ZIF-8 surface chemistry with Lewis acid sites (Zn2+) and weakly basic sites (Zn-OH) upon the incorporation of pyrrole. We then averaged the spectra of facet, edge, and corner pixels and calculated their 885/1590 cm−1 (pyrrole/imidazole) peak ratios. We plotted these ratios together with the crystal-averaged ratio in Fig. 4c. These ratios showed that the expression of defect sites scaled with surface energy and is thus structure sensitive, with the {310}, {210}, and {211} planes having an above-average concentration of defects.

Isotope-labeled deuterated pyrrole was used to pinpoint the distribution and clustering of pyrrole defects using their spectral fingerprint, and to correlate defect-specific guest–host interactions. To visualize pyrrole incorporation, we acquired another 200 × 200 nm2 hyperspectral images of a defective ZIF-8 crystal in N2. Subsequently, we used principal component analysis (PCA) and clustering of the hyperspectral image, to group pixels into clusters based on spectral similarities (Fig. 4d, Supplementary Information). The spectra of the clusters revealed that clustering mainly occurred based on defect concentration, as all spectra showed a largely unchanged ZIF-8 spectrum yet showed variation in relative δCD and νC=N band intensities (Fig. 4e and Supplementary Fig. 22).

To deduce the location and degree of clustering of defects, we first sectioned the clustered image into one {100} and two {110} facets, based on the morphology map (Fig. 4d and Supplementary Fig. 22). Then, we calculated the fraction of facet surface that was covered by each cluster (Supplementary Figs. 23–26) which is represented by the bubble sizes in Fig. 4f. Subsequently, we quantified the relative defect linker concentration of each cluster by calculating the peak ratio between pristine and defective ZIF-8 vibrations (i.e., 885/1590 cm−1) for the five clusters (Fig. 4f, y-axis). The weighted average of defect linker concentration is also reported per facet (stars in Fig. 4f). In this way, we were able to distinguish between plane-averaged defect concentrations and nanodomains of high/low defect concentration. These results showed that defect concentrations varied more intra-facet, i.e. between parent facet and nanoislands, than inter-facet, an experimental insight that could only be gained because of the nanoscale resolution of the PiFM method. Notably, the defect-rich domains were similar in size to the high-energy termination domains formed on non-defective ZIF-8, which preferentially adsorbed FA also on non-defective ZIF-8 crystals (Fig. 3), suggesting that pyrrole is preferentially incorporated into high-energy planes. DFT calculations on defect incorporation further support this claim (Supplementary Table 3).

Structure-sensitive formaldehyde conversion

To discern between the effects of crystal plane terminations and pyrrole defects on the structure-sensitive adsorption of formaldehyde, we performed in situ PiFM FA adsorption experiments on the defective crystals (Fig. 5a and Supplementary Fig. 27). The rising intensity of 1280, 1200, and 895 cm−1 bands with FA pressure (0–480 ppm) indicated the sorption of formaldehyde, similarly to the pristine crystals. However, in addition to these bands, signals at 1580, 1380, 1320, 1150, and 1060 cm−1 were found. Such bands suggest the formation of different external surface adsorbates, such as formates, dioxymethylene (DOM) and polyoxymethylene (POM), and methoxy species (Fig. 5c, d and Supplementary Table 4)32,39. The systematic occurrence of these adsorbates showed that defect-mediated conversion of formaldehyde was taking place on the defective ZIF-8 surface32,48. The results were corroborated by bulk in situ FA adsorption experiments on commercial and synthesized ZIF-8 powders and films, using attenuated total reflection FTIR spectroscopy, in which comparable spectral features and pressure-dependent trends in signal intensity were observed (Supplementary Figs. 36–40).

Fig. 5: Structure-sensitive conversion of formaldehyde (FA) over defect-engineered ZIF-8.
figure 5

a Contour plot of the crystal surface-averaged in situ infrared spectra of a defective ZIF-8 crystal, showing the formation of formate, di-/poly-oxymethylene (DOM/POM), and methoxy surface species (relevant bands indicated by dashed lines in the contour plot). b A plot showing plane-dependent response pressures for intermediate surface species, thus showing structure-sensitive FA conversion, as intermediates are observed at lower pressures on high-index planes (note: pyrrole-induced changes in crystal facet size ratios prevented the analysis of the {111} facets). Based on the observed species and in accordance with the literature32,48, we propose two possible mechanisms for FA conversion over defective ZIF-8: a methoxy-mediated (c) and formate-mediated (d) mechanism.

Full size image

Based on our spectroscopic results, and in accordance with the literature, we propose two possible formaldehyde conversion mechanisms that operate in parallel on the defective ZIF-8 surface: a methoxy and a formate mechanism (Fig. 5c, d)49,50. In the methoxy mechanism, the chemisorbed FA undergoes a Cannizzaro-type reaction to form a methoxy species, a precursor for methanol and formic acid. In the formate mechanism, the FA is chemisorbed in the form of polymers such as dioxymethylene and polyoxymethylene, which are oxidized to form monodentate formates and converted to yield formic acid. Highly relevant for these mechanisms is the hydrolysis of the open zinc sites (vide supra), which provide an additional source of oxygen, which are in turn continuously replenished by water from the aqueous formaldehyde solution. These findings are in line with bulk IR studies on oxide-supported metal nanoparticles, where oxygen from the support is instrumental to the conversion of formaldehyde51,52.

To study the effect of plane terminations (structure sensitivity) on FA conversion, we first compared the response pressure of FA conversion over different crystal planes (Fig. 5b). This comparison showed that FA conversion, similar to FA adsorption, is structure sensitive as a function of crystal termination, with the following order: {310} > {210} > {211} > {110} ~ {100}, which is consistent with what was observed for pristine crystals (Fig. 2). However, by using PCA and clustering within a single facet (e.g., {100}), we showed that defect-rich areas did show more responsive FA (conversion) behavior than the facet average (Supplementary Figs. 28, 29), similarly to what observed for nanoislands on pristine crystals (Fig. 3).

Since different crystal planes showed a different concentration of defects, dependent on the plane energy (Fig. 4c), we subsequently performed a control analysis to rule out a defect concentration effect on the trends observed in Fig. 5b. To this end, we used PCA and clustering to identify defect-rich and defect-poor areas spanning over the entire crystal surface (i.e., a combination of contributions from all planes). Both areas showed comparable FA adsorption and conversion behavior (Supplementary Figs. 28, 30). Therefore, we conclude that the defect concentration did not affect the trends observed in Fig. 5b and that a higher defect concentration did not result in the detection of FA conversion intermediates at lower pressures.

To rationalize these findings, we modeled FA adsorption on the defective, low energy cuts of {100}, {110}, and {310} surfaces by DFT and compared the results to the previously discussed pristine ZIF-8 (Supplementary Fig. 31). Formaldehyde reaction energies (at 0 K) were found to be 97.1, 88.9, and 1.6 kJ/mol for the low-energy cuts of {100}, {110}, and {310}, respectively, again confirming experimental FA sorption (and conversion) trends. The models further showed that the absence of the second nitrogen atom on the pyrrole linker resulted in the formation of undersaturated Zn sites exhibiting Lewis acidity, in contrast to what was observed for pristine sites. We thus confirm that such Lewis acidity was responsible for the subsequent conversion of chemisorbed formaldehyde.

Overall, the present in-depth nano-spectroscopic characterization campaign revealed that: (i) as expected, high-index crystal terminations outperform extended facets in FA adsorption due to exposure of undercoordinated moieties; (ii) seemingly extended and highly coordinated surfaces in MOFs microcrystals contain uncoordinated nanodomains which have similar behavior to high-index terminations, resulting in locally enhanced gas sorption; (iii) the inclusion of defects results in structure-sensitive FA conversion, with higher defect density in high-energy surfaces; (iv) the facet-dependent structure sensitivity of FA adsorption is still observed in defect-engineered microcrystals, meaning that synthesis efforts for better gas sorption materials should be directed to the exposure of high-index crystal planes, rather than to higher defect density. These findings conclusively showed that not all defect sites are equal, adding a layer of complexity to the structure sensitivity of FA sorption and conversion on ZIF-8, which could only be discovered using the proposed spatially resolved, in situ PiFM technique.

Using a combination of well-defined ZIF-8 microcrystals, nanoscale in situ PiFM measurements and DFT calculations, we have shown the existence and importance of structure sensitivity in formaldehyde sorption and conversion over the porous functional material ZIF-8. We have shown that for (defect-engineered) ZIF-8 formaldehyde gas sorption preferentially occurs on the sterically isolated external surface sites of high-energy planes, such as edges and corners ({310} > {111} > {210} > {211} > {110} ~ {100}), and that defect linker incorporation adheres to this surface energy-based trend as well. Furthermore, as a result of the nanoscale infrared resolution, we found the co-existence of high-energy and low-energy crystal surface terminations within single facets, where the high-energy nanodomains showed enhanced formaldehyde sorption behavior. Similarly, such nanodomains show higher affinity to defect linker incorporation than their low-energy counterparts. Additionally, the incorporation of defects was found to lead to the formation of Lewis acidity and the consequent conversion of formaldehyde on the ZIF-8 surface through both a formate and a methoxy mechanism. Importantly, the performance of these defect sites for formaldehyde conversion was found to be heavily dependent on structure sensitivity phenomena, rather than on defect concentration. We believe that in situ PiFM is a highly applicable analytical toolbox for establishing nanoscale structure-performance relationships since localized sorption behavior on the nanoscale will greatly affect the macroscale performance of (porous) functional materials (e.g., MOFs, zeolites) and guest/probe molecules such as CO2, NO, CO, etc. providing insights for the rational synthesis of improved nanostructured sorbents and catalysts. Outside the field of catalysis, we see high relevance to the understanding of nanostructured functional materials for sensing, gas separation (e.g., composite membranes), and drug delivery, and for environmental science problems such as understanding micro/nano-plastic degradation and atmospheric nanoparticulate chemistry.