id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
64526d5527fccdb3ea6e13c8 | 5 | Non-invasive analyses were carried out to probe porosity insights of the cryogels from the macroscale to the nanoscale. Image analysis using Local Thickness was performed on SEM images of the cryogel scaffolds to obtain accurate estimates of minimum macropore sizes. "Local Thickness" is an algorithm for characterizing the minimum pore sizes of 2D or 3D pore structures by measuring the maximum diameter of a sphere that would fit the minor axis of visible pore spaces, while disregarding pore shapes . Dynamic vapor sorption (DVS) was performed to analyze mesopore data through equilibrium moisture (EMC) measurements. In DVS, a full cycle sorption desorption was conducted on every cryogel sample, and precise vapor pressure and gravimetric measurements were obtained. Vapor pressure and gravimetric measurements allow for the determination of EMC and the estimation of the volume distribution of mesopores. |
64526d5527fccdb3ea6e13c8 | 6 | X-ray has the ability to reveal intrinsic sub-nanoscale patterns as the polydispersity of particles in a cluster affects scattering . SAXS measures the X-ray scattering intensity I(q) as a function of the scattering vector q, i.e., I(q) ∝ q -τ , where τ is the scattering exponent. By combining the analyses done by SEM-local thickness, SAXS, and DVS, this experiment aims to establish a reliable method to probe porosity, morphology, and to further the understanding of the porosity of CNC-based cryogels as renewable, biocompatible, and biodegradable porous materials. The CNC suspension used in this experiment was a carboxylated entity obtained from Anomera Inc., produced by oxidation with dilute hydrogen peroxide of dissolving pulp from black spruce. The CNC was used without further purification. |
64526d5527fccdb3ea6e13c8 | 7 | The material CNC-MNP was synthesized in situ with a co-precipitation method based on the work of Hassan and coworkers . Briefly, 20 mL of the CNC suspension was thoroughly mixed in 100 mL of MiliQ water (18.2 MOhm) until fully dispersed. A separate 10 mL aqueous solution containing 1.0 mmol of FeCl2 and two equivalents of FeCl3 was prepared and was slowly injected into the stirred CNC suspension under argon. The mixture was heated to 70 ℃ for one hour before magnetite nanoparticles were co-precipitated using 1:10 diluted NH3OH(aq). The product was washed many times using DI water until the wash reached a pH of 7. Excess water was removed by centrifuge at 500 rpm, higher frequencies can lead to phase separation between the nanoparticles and CNC. The compositions of the CNC and CNC-MNP particles obtained were examined by XPS, and the attachment of the MNP onto CNC was verified using TEM (Section 1 in Supporting Information). |
64526d5527fccdb3ea6e13c8 | 8 | CNC cryogels were prepared directly with the as-obtained CNC suspension, which has a solid concentration of 3.8 %wt in water. MNP-incorporated CNC cryogels were made from a mixture of 20 % v/v of the as-synthesized MNP decorated CNCs (CNC-MNP) and 80 % v/v of asobtained CNC suspension to achieve gelation similar to that of the CNC suspension. The mixed CNC-MNP suspension has a solid concentration of 3.9 %wt in water. |
64526d5527fccdb3ea6e13c8 | 9 | Gelation of the suspensions was done through three different types of freezing. "Hightemperature" freezing of the suspension was performed in a freezer (-12 ℃), while "lowtemperature" freezing was achieved by submerging the suspension in liquid nitrogen (-196 ℃). Two-step freezing was conducted by first freezing the sample at -12 ℃ for 8 hours, allowing the sample to thaw, then re-freezing the sample at liquid nitrogen temperature. An external static magnetic field (SMF) was applied to some gelation processes using permanent magnets. The porous structure of the cryogel was revealed by freeze-drying for 48 hours. Specificities of sample preparation are detailed in Section 2 of Supporting Information. |
64526d5527fccdb3ea6e13c8 | 10 | The macroscopic structural images of the cryogels were taken using SEM. The qualitative SEM images were graphically fitted using the algorithm "Local Thickness" which provides lowerbound estimates of pore sizes . DVS and SAXS were performed to probe the microscopic structural details of the cryogels to assess their meso-and microporosities. Dynamic vapour sorption data were collected on DVS Resolution from Surface Measurement Systems. Small angle X-ray Scattering measurements were conducted on a SAXSpoint 2.0. The decay of the scattering intensity I(q) as a power law was fitted using the power law function and the surface fractal model, both available in the analysis program SasView. More details on the experimental methods on SEM-local thickness, SAXS, and DVS are available in Section 3 of Supporting Information. |
64526d5527fccdb3ea6e13c8 | 11 | Figure shows SEM images of the various microporous structures obtained under different freezing conditions. Figure in Supporting Information details the Local Thickness estimation of macropore distributions. The most prominent morphological change is observed between the cryogel scaffold frozen at -12 ℃ when compared with that frozen rapidly in liquid nitrogen. The slow freezing process associated with high-temperature freezing resulted in polydisperse pore structures with large pore sizes (Figure ), while rapid, low-temperature freezing produced closely aligned and layered structures, and narrow pore diameters (Figure ). A sample macropore size calculation of CNC cryogel scaffold frozen at -12 ℃ using the Local Thickness method is shown in Figure . |
64526d5527fccdb3ea6e13c8 | 12 | Studies on ice nucleation and crystal growth have found that lower freezing temperatures lead to a higher degree of supercooling . A higher degree of supercooling means a greater number of ice crystals are nucleated and more water is instantly frozen . As a result, fewer but larger ice crystals are expected when the freezing temperature is high, and numerous but smaller ice crystals are expected when the freezing temperature is low . |
64526d5527fccdb3ea6e13c8 | 13 | The ordered layering of elongated pores produced by low-temperature freezing is very similar to that produced by directional freezing in other studies . In those studies, directional freezing was achieved by slowly lowering the sample into liquid nitrogen, creating a temperature gradient within the suspension, where the direction of freezing was guided from the cold end to the warm end along the vertical axis. Although directional freezing was not carried out in this experiment, the alignment of the pores with tilted horizontal orientations was observed, and the cryogel scaffold exhibited radial symmetry around the center axis. This tilted horizontal ordering was likely produced by the temperature gradient created as the suspension cooled from the outer edge of the container toward the center. Furthermore, the growth of ice crystals could be affected by the inter-particle interactions and arrangements of the CNCs . The CNCs are known to exhibit nematic ordering at high enough concentrations where layers of aligned CNCs are arranged (1). This arrangement could serve as an initial template to assist with the growth of lamellar ice sheets in between nematic planes. Unlike low-temperature freezing, high-temperature freezing produced rounded pore structures in scaffolds, indicating aggregated CNCs around individual unoriented ice crystals . |
64526d5527fccdb3ea6e13c8 | 14 | The introduction of CNC-MNPs appeared to decrease the smoothness of the scaffold, creating visible clusters of entangled fibers. This observation suggested an inhomogeneous distribution of CNC-MNPs within the CNC suspension. The cause of this inhomogeneity could be the aggregation of the uncoated MNPs, as observed in a previous ferrogel study . The MNPs are also known to be effective ice nucleation sites to induce changes in the formation of ice crystals. The effect of MNP-induced surface roughness is more pronounced when high-temperature freezing is involved (Figure |
64526d5527fccdb3ea6e13c8 | 15 | The SEM images of CNC and CNC-MNP cryogel scaffolds made with two-step freezing (Figure ) show that the additional cryogelation step disrupted the elongated pore shapes observed in scaffolds made with N2(l) freezing alone. The cryogelation step introduced an aggregated network of CNCs, most likely caused by van de Waals forces and hydrogen bonding, which inhibited the redispersion of CNCs even after thawing . This network then became a template that disrupted ice crystal growth during the follow-up low-temperature freezing step. The resulting pores showed mixed characteristics of high and low-temperature freezing, with some degrees of elongation but overall less oriented pore patterns. When MNPs are involved, two-step freezing results in exposed fibers (Figure ), indicating destabilization of the CNC-MNP network between the two freezing steps. |
64526d5527fccdb3ea6e13c8 | 16 | The effect of SMF on freezing may be related to hydrogen bonding. Hydrogen bonding was found to be critical in ice nucleation . When water enters the supercooling state, hydrogen bonds become longer-lasting. Ice nucleation occurs when enough long-lived hydrogen bonds appear simultaneously at the same location . Applying SMF may lead to the alignment and thermal motions of water molecules , which reduces ice nucleation temperature and ice crystal sizes, contributing to a weakening of the hydrogen bonds. It was observed that an applied SMF lowers the ice nucleation temperature in pork and beef and reduces the ice crystal sizes in cherries . For cryogels, an applied SMF may have disrupted the formation of ice crystals during the high-temperature freezing step, reducing the aggregation of CNCs in the intercrystalline space between ice crystals. The resulting hydrogels thaw without retaining strong cryogelation networks. The stronger the cryogelation networks, the more structural restrictions they impose . Weak cryogelation networks led to the formation of cryogel scaffolds with increased lamellar pore characteristics after low-temperature freezing. |
64526d5527fccdb3ea6e13c8 | 17 | SAXS curves for CNC and CNC-MNP cryogel scaffolds are shown in Figures and, respectively. The x-axis of the SAXS curves is the scattering vector q, which relates to a length scale by 2𝜋 𝑞 . As such, the noise-excluding q range between qmin =0.065 and q max = 1.3 nm -1 in this experiment reveals nanoscale insights of approximately 96.7 -4.83 nm. Two q ranges of interest were examined, namely the low-q region (Figure , region 1) of 96.7 -27.3 nm, and the mid-q region (Figure , region 2) of 27.3 -4.83 nm. In the low-q region, the slope of the scattering curve of all CNC and CNC-MNP cryogel scaffolds displayed 𝐼(𝑞) ∝ 𝑞 -4 power-law decay. The power law q -4 suggests that the CNCs aggregated to form distinctive interfaces and well-defined shapes , which is indeed observed by the SEM of aggregated CNC sheets making up the walls of the macroporous scaffolds (Figure , Figure ). Moving towards the mid-q region, the power law relations deviated from q -4 . The mass fractal dimensions calculated for the mid-q region are 2.9 for all pure CNC samples. For samples with incorporated MNPs, the mass fractal dimensions notably reduced to 2.1 -2.3 (Table ), revealing a decrease in the compactness of the network structure (39, 50) likely due to the accommodation of nanoparticles. |
64526d5527fccdb3ea6e13c8 | 18 | Mesopore differential plots were calculated for all scaffolds based on precise water vapor pressure and gravimetric measurements are shown in Figure . The plots revealed the volume occupied by mesopores with radii ranging from 2 -36 nm during water sorption. In general, most water sorption takes place in small pores as the volume occupied increase with decreasing pore sizes. It was found that the freezing methods or an applied SMF have little effect on water sorption mesoporosity, while the incorporation of MNPs induced a change to the mesoporosity pattern. |
64526d5527fccdb3ea6e13c8 | 19 | While the water sorption volume generally increased with decreasing pore sizes within the probed range of mesoporosity, a sorption peak occurred at mesopore radius of 5 -6 nm for scaffolds made with only CNC, which the CNC-MNP scaffolds do not share. The MNPs synthesized in situ were polydispersed nanoparticles approximately 4 -12 nm in diameter (Figure ). The lack of sorption peak in CNC-MNP scaffolds may indicate space occupied by MNPs. |
64526d5527fccdb3ea6e13c8 | 20 | The percent change in sample mass by adsorption and desorption of water vapor over time reveals the equilibrium moisture content (EMC) and the sorption kinetics pattern of the cryogel scaffolds. Despite the difference in compositions (CNC vs. CNC-MNP) and macroporosity, the cryogel scaffolds show similar adsorption kinetics (Figure ). The EMC for different cryogel samples falls within the water uptake range of 17.6% -20.0%. CNC cryogel scaffolds show slightly higher EMC than CNC-MNP cryogel scaffolds, as shown in Figure . |
64526d5527fccdb3ea6e13c8 | 21 | Water sorption is greatly affected by the crystallinity of the material . For crystalline solids, water adsorbs on the surface as monolayers and multilayers (up to 3 -6 layers) on a crystal surface . In a study using Stöber silica spheres, it was estimated about 4 layers of water can adsorb on the surfaces at 25℃ and p/p0 = 0.89 . Water also penetrates pore structures and causes swelling, likely due to interactions with hydroxyl groups at the interface . The swelling, in turn, opens new sites for water sorption, including increasing the accessibility to micropores 2 nm or less in size . |
64526d5527fccdb3ea6e13c8 | 22 | The effects of crystallinity and structural swelling are visible in the hysteresis of water sorption. The lower the degree of cellulose crystallinity, the broader the hysteresis . The broad hysteresis is due to delayed desorption attributed to the slow breaking of hydrogen bonds at the interface, difficulty evacuating different pore sizes, and structural deformations . |
636e417ebef5d44ffb5176d3 | 0 | Analyzing the reactivity of transitional metal nitrides, researchers concluded that nitrides can be produced by the direct action of Nitrogen (N) on the metal when the small N atoms, which satisfy the empirical Hagg rule of 0.59 ratio between non-metal/metal atom radii , will occupy the octahedral spaces in the body-centered cubic (BCC) Chromium (Cr) lattice and that the formed interstitial compound is influenced by the electrons in the metal's outer shell . As such, having five electrons in the three-dimensional shell, Cr will form two nitrides: Chromium nitride (CrN) and Dichromium nitride (Cr2N). The resulting alloy inherits most of the base metal properties. Moreover, metal-nonmetal bonds are formed due to the presence of N atoms in the Cr lattice and the connection is partly ionic due to the difference in electro-negativity between the two elements . The resulting CrN has a higher melting point than the base metal, increased hardness, and retains the chemical stability of the base metal . |
636e417ebef5d44ffb5176d3 | 1 | CrN was proven as a great option for protective coatings applications due to its hardness, low coefficient of friction and great corrosion resistance . Moreover, CrN exhibits good heat resistance and great thermal stability . These exceptional properties recommend CrN as a good candidate for microelectronics applications , especially in the domain of protection against corrosion, oxidation, and heat . A remarkable property of CrN thin films is the tunability of resistivity with the orientation of crystallites , which opens a large palette of electrical applications of CrN. |
636e417ebef5d44ffb5176d3 | 2 | Supercapacitor applications represent another research area in which massive progress has been made in the last years . Due to their high specific capacitance and cycling stability, metal nitrides hold great potential as supercapacitor electrode materials. In the authors prepare CrN thin films by direct current magnetron (DC) sputtering on polished Si wafers. They found that electrochemical performance can be easily tuned by varying the deposition conditions. |
636e417ebef5d44ffb5176d3 | 3 | In this article, we deposit CrN thin films on a silicon substrate by two different techniques and three deposition durations for each technique. Both techniques are methods of physical vapor deposition of thin films onto a substrate, the main difference is the type of power supply used in the deposition process. The first method, direct current sputtering, (DC) uses a direct current power supply while the second method, high-power impulse magnetron sputtering (HiPIMS) uses an alternating current power supply. Another difference between the two methods is the quantity that they are able to process, DC sputtering is preferred when dealing with large quantities of large substrates while HiPIMS sputtering is preferred for smaller size substrates. Both are used as industrial scales DC sputtering is used for coating the edge of cutting tools and HiPIMS sputtering is used for the coating of electronic circuits that also act as structural elements. For this reason, we chose to study the structural and mechanical properties of ultrathin CrN depositions, providing insights into the best deposition conditions to a obtain uniform, highly protective coatings. In order to obtain ultrathin coatings, we chose the deposition times of 5, 10, and 15 minutes taking in account that in the case of the industrial applications, the deposition times typically start from 180 minutes, for the layers having more than microns thicknesses. |
636e417ebef5d44ffb5176d3 | 4 | CrN films were deposited using a Leybold-Heraeus Z-550-S sputtering system. A 6-inch Cr cathode was placed in the top plate of the vacuum chamber and the Silicon (Si) substrates were placed in the bottom holder which is rotating with a speed of 6 rot/min and is kept at a constant temperature of 25 °C. Prior to the deposition, the Si samples were polished and cleaned in Argon plasma for 10 minutes. |
636e417ebef5d44ffb5176d3 | 5 | In the first batch, the power supply for the sputtering system was an SSV-2.5 kW DC power supply, with a constant power of 1000 W, a current of 2.89 A, and a voltage of 392 V. The Si substrate was biased at 100 V with a frequency of 900 Hz and an impulse duration of 195 μs. |
636e417ebef5d44ffb5176d3 | 6 | In the second batch, the power supply for the sputtering system was an Ionautics high-power impulse magnetron sputtering (HiPIMS) power supply, with an average power of 1000 W, frequency of 900 Hz, with an impulse duration of 65 μs. The substrate was biased at 100V, with a frequency of 900 Hz and an impulse duration of 195 μs. |
636e417ebef5d44ffb5176d3 | 7 | For the investigations of surface topography, an NX10 Park Atomic Force Microscope (AFM) was used, equipped with a non-contact probe (NCHR, NanoAndMore) with a resonance frequency of 300 kHz and a sub-10 nm tip radius. We acquired 512x512 pixel images on 10x10 µm 2 areas on the deposited films and we further used the images for the quantitative measurements. |
636e417ebef5d44ffb5176d3 | 8 | Shallow nanoindentation was performed as well by using the AFM system with a TD26706 probe (Park System) having a nominal force constant of 154 N/m, resonant frequency of 40 kHz, and a tip radius smaller than 25 nm. Analysis of the approach-retract curves in the shallow regime was used to extract the material stiffness. |
636e417ebef5d44ffb5176d3 | 9 | Water contact angle tests were performed using a simple imaging system built on table composed by a CCD camera and an adjustable microscope sample holder. For each sample a drop of 10 μL demineralized water was placed on the surface and imaged by our system. The acquired images were analyzed using ImageJ to measure the contact angle. |
636e417ebef5d44ffb5176d3 | 10 | We used both acquired SEM and AFM images for quantitative and qualitative measurements. Being one of the crucial technical parameters of the manufactured parts, especially in metallic deposition applications, surface roughness is one of the quantities evaluated from the AFM investigations. As a measure of surface roughness, we use the surface area roughness (Sq) present in the AFM surface topography. We evaluate the surface roughness by using Gwyddion image analysis software . |
636e417ebef5d44ffb5176d3 | 11 | The fractal dimension (FD) is another measure to quantify the organization of structures appearing in a microscopic image. FDs values close to 2 indicate smoother surfaces, and values close to 3 indicate a high degree of roughness. Fractal analysis can be performed either on grayscale images or binary (black and white) images. In the case of the analysis carried out on samples within the experiment, we considered the case of grayscale images. We performed fractal analysis by ImageJ's FracLac plugin, which computes the fractal dimension using the "box counting" method. With this method, the image is covered by successively smaller squares. At each step, the number of squares containing the image elements is counted. Such image elements are different in pixel intensity in the area occupied by the squares. The fractal dimension, which is a measure of the complexity of the image, is calculated as the slope of the regression line of the log-log plot of the difference in pixel intensity in the area covered by the squares and the size of the squares. |
636e417ebef5d44ffb5176d3 | 12 | The other two measures of uniformity in an image are entropy (S) and autocorrelation length (L). The entropy reflects the uniformity of the height distribution of the topographical profile , while the autocorrelation length offers a measure of the uniformity of the surface texture . We compute these two parameters using Gwyddion. |
636e417ebef5d44ffb5176d3 | 13 | Another set of quantitative measurements can be obtained from the gray-level co-occurrence matrix (GLCM) which represents a second-order statistical modality that provides information related to the spatial relationship between pixel intensities in each image . The GLCM matrix is built based on the number of occurrences of a certain pair of pixels, at a specific distance between pixels. Each result is divided by the total number of elements to calculate a probability. Such a matrix can be calculated for any distance between pixels either horizontally (0°), vertically (90°), or diagonally (45°,135°). Because we observed only small variations between the values calculated according to different orientations, we report here the average value for the results obtained for the four orientations. We extract two features of the GLCM: energy (or the angular second moment) and correlation. |
636e417ebef5d44ffb5176d3 | 14 | EDAX elemental mapping obtained for the case of DC power supply deposition are presented in Figure . For each time duration of the deposition, a composite image is displayed containing SEM cross section and Si, Cr, and N elemental maps, respectively. The distribution of Cr and N in the deposited layer (Figs. and) is homogeneous and uniform with no noticeable defects in the deposition layer. It is therefore expected that the thin CrN layer will provide corrosion resistance for the Si surface beneath. The interaction (diffusion) at the deposition/substrate interface can also be noticed. |
636e417ebef5d44ffb5176d3 | 15 | Apart from a qualitative inspection, EDAX spectra allow the measurement of the weight distribution of each element: Si (substrate), Cr, and N. The EDAX spectra for each sample was acquired in the region of the red boxes represented in Figs. and (R1.2). The corresponding spectra (number of counts vs. energy levels) for the CrN films deposited by DC power supply for each time duration are presented in Figure . |
636e417ebef5d44ffb5176d3 | 16 | The surface morphology imaged by SEM for the case of DC power supply is represented below (Figure ), for the three durations: 5 (Figure ), 10 (Figure ), and 15 (Figure ) minutes. Similarly, the morphology analysis results for the case of HiPIMS power supply are represented in Figure , for the same time durations of the deposition process (5, 10, and 15 minutes, respectively). All the images in Figures and are obtained with the same magnification. |
636e417ebef5d44ffb5176d3 | 17 | For the DC power supply (Figure ), the deposition is rough, and a fine columnar growth pattern can be seen even after a deposition time of 5 minutes. The growth pattern for the DC method first develops as small blobs, and as the deposition time increases those blobs act as nuclei for other blobs to form near and on top, thus increasing in size. Finally, because of the distribution of these blobs, a columnar growth pattern begins to form. Our findings are confirmed by other results in literature, as DC sputtering is known to create fine and columnar depositions, a fact which we have shown holds true even for ultrathin deposition as in our case. On the contrary, in the case of HiPIMS power supply (Figure ), no noticeable growth pattern can be distinguished. This shows that the deposition layer is smooth and homogeneous as there are no noticeable growth patterns. This kind of deposition mechanism is also representative, as the HiPIMS deposition is known to create dense and glassy depositions. This seems to hold even for ultrathin depositions as in our case. |
636e417ebef5d44ffb5176d3 | 18 | The above morphology images obtained by SEM investigations were further on analyzed to compute the FD parameter for each sample. Table summarizes the results of this analysis. The fractal dimension values in Table show the common behavior of the samples irrespective of the power supply or the deposition time it varies between 2.696 for the CrNHi5 sample and 2.832 for the CrNDC5 sample. |
636e417ebef5d44ffb5176d3 | 19 | The same morphology images were analyzed as well from a statistical point of view. After computing the GLCM matrix for each image, the energy and correlation parameters were extracted for different pixel distances (1, 10, and 100 pixels) in each image. The results are summarized in Table . The energy values of the surface morphology GLCM matrix (Table ) do not have great variations based on the pixel distance, but they do vary based on power supply type and deposition time. For the DC power supply, the energy increases with deposition time and for the HiPIMS power supply, the energy decreases with deposition time. This means that the layer deposited using the DC power supply becomes smoother as the deposition time increases while the layer deposited using the HiPIMS power supply becomes rougher. |
636e417ebef5d44ffb5176d3 | 20 | For the correlation value extracted from the surface morphology GLCM matrix (Table ), we notice a dependency on three factors: pixel distance, power supply type, and deposition time. Correlation greatly decreases with pixel distance. For both the DC and HiPIMS power supplies the correlation increases with deposition time. An interesting aspect is the rate at which correlation increases, the correlation for the deposition done using the DC power supply starts at 0.033 for the distance of 1 pixel and after 15 minutes it ends at 0.085 while the correlation for the deposition done using the HiPIMS power supply starts at a greater value of 0.064 for the distance of 1 pixel and after 15 minutes it ends at 0.067. So even if the correlation parameter for the HiPIMS power supply starts at a greater value because the correlation for the DC power supply increases at a greater rate, it ends up being smaller than the correlation for the DC power supply. |
636e417ebef5d44ffb5176d3 | 21 | Measurements for the thickness of the CrN layer performed by SEM are presented below for the case of DC power supply (Figure ). Similarly, the corresponding images for thickness measurements of the deposited film for the case of HiPIMS power supply are represented in Figure . The same magnification factor was used in all displayed SEM investigations to aid the visual perception. and one can notice that the main difference in the films deposited by DC sputtering and HiPIMS sputtering consists greatly in the structure of the morphology. In Figure , the deposited layer when using the DC power supply seems fine and has a columnar structure, while in Figure , the deposited layer in the case of HiPIMS power supply seems dense and uniform. Moreover, the film thickness can be extracted from the figures mentioned before. When using the DC power supply, the film thickness increases at approximately a constant rate but when using the HiPIMS power supply the film thickness for the deposition times of 5 and 10 minutes are approximately the same while for the deposition time of 15 minutes it sees a sudden jump of 200%. |
636e417ebef5d44ffb5176d3 | 22 | The topography images obtained by AFM investigations on samples containing CrN films deposited by DC power supply are presented in Figure . Analysis of the topography images acquired on samples obtained by the DC sputtering method (Figure ) reveals that the defects tend to be visible at reduced deposition times, with the characteristic structures of the substrate fading out after a deposition time of 10 minutes. Another noticeable observation is that in DC sputtering, a characteristic columnar growth structure is observed. Even after 15 minutes of deposition, the CrN layer does not become uniform. |
636e417ebef5d44ffb5176d3 | 23 | The results obtained by AFM imaging of the CrN layer deposited through HiPIMS sputtering (Figure ) indicate that at reduced deposition times the microscopic scratches that were on the Si substrate took place are still visible after the deposition process, indicating an ultrathin layer (which is confirmed by the results from Table ). As the deposition time increases to 10 minutes, the microscopic scratches that were on the Si substrate are still visible, indicating the formation of a layer that has a thickness comparable to the layer deposited in 5 minutes. Lastly, for the deposition time of 15 minutes, the HiPIMS sputtering method creates a layer that has a more uniform appearance than those obtained at shorter deposition times. The same conclusion is supported by the statistical analysis discussed below. |
636e417ebef5d44ffb5176d3 | 24 | We compute Young's modulus from force-distance curve analysis performed in 15 different locations for each sample. Similarly, we determine the surface stiffness as the slope of the force-separation curve in the contact region from approach-retract curves resulting from nanoindentation processes. The results can be found in Table . For both DC and HiPIMS methods, there is a decreasing trend of Young's modulus with the deposition time. This trend appears due to the measurement method, which is highly dependent on the tip-surface interaction. Consequently, the decrease of the Young's modulus has different explanations depending on the power supply. For DC power supply, the columnar deposition which is gradually formed with the increase of the time deposition strongly influences the tip-sample interaction due to holes created between the columns. Therefore, the measured Young's modulus decreases because of this columnar character of the films. On the other hand, in the case of the HiPIMS power supply the films are very thin and the interaction between tip and the sample surface is strongly influenced by the substrate. |
636e417ebef5d44ffb5176d3 | 25 | The statistical parameters extracted from the surface topography images were roughness (Rq), fractal dimension (FD), entropy (S), and autocorrelation length (L). The values resulting from this analysis of all the samples are summarized in Table . The surface roughness analysis (Table ) shows that although both DC and HiPIMS generally result in sub-1 nm surface roughness, in the HiPIMS method the highest sur-face roughness recorded was 0.603 nm, which is 1.7 times less than the highest surface roughness recorded with the DC method. It is worth discussing here the increased rate of surface roughness depending on the power supply method. For both methods, it can be observed a non-linear increase of roughness with time deposition. For DC method the roughness tends to increase rapidly after 5 to 10 minutes of deposition time (29% increase) and slower after another 5 minutes (13% increase in the interval between 10 to 15 minutes deposition time). On the other hand, for the HiPIMS method, we observe a slow increase of the roughness between 5-and 10-minutes deposition time (13%) and a rapid increase of the roughness between 10-and 15-minutes deposition time (24%). |
636e417ebef5d44ffb5176d3 | 26 | For the fractal dimension, there is an increasing trend with deposition time regardless of the power supply type used. While the increase of the fractal dimension is non-linear with time deposition, the fractal dimension increases slowly between 5 to 10 minutes and much faster between 10 to 15 minutes of deposition time, for both methods. Regarding the entropy values (Table ) there is no easily discernable dependency in either deposition time or power supply used. This means that the uniformity of the height distribution remains nearly constant regardless of the power supply or time deposition. For the autocorrelation length, there is a decreasing trend with the increase of deposition time, which means that the uniformity of surface texture is highly dependent on the time deposition. |
636e417ebef5d44ffb5176d3 | 27 | After GLCM analysis the interest parameters (energy and correlation) were extracted and summarized in Table . Both energy and correlation were extracted at three different pixel distances (1, 10, and 100). In the data summarized in Table , we notice a decreasing trend both with the pixel distance and with deposition time. This means that as deposition time and pixel distance increase the texture of the deposited layer becomes less uniform regardless of power supply type used. |
636e417ebef5d44ffb5176d3 | 28 | The surface wettability of the samples was tested by performing water contact angle measurements as descripted into methods section. The purpose was to evaluate the protective capability of the deposited CrN layers, which means that the surface must exhibit hydrophobic properties (contact angles > 90°). Such surface will prevent water infiltrations and will have self-cleaning behavior as well. Figure shows the results obtained on the samples deposited by DC power method for each deposition time. It can be observed that for the 5-and 10-minutes deposition the surface is hydrophobic (contact angles of 99.5° and 98.2° respectively), while for 15 minutes deposition the surface is almost hydrophobic (contact angle of 88.9°). Interestingly, the contact angle decreases with the increase of the CrN layer thickness. For the HiPIMS power supply, similar results are obtained. For 5-and 10-minutes deposition the contact angle is 98.5° and 95.2° respectively. For the case of 15 minutes deposition time the contact angle is 89.5°, which is a limit case, close to the hydrophobicity condition. The results are displayed in Figure . Similar to the DC power supply case, the contact angle decreases with the increase of the deposited layer thickness. This means that the interface between Si substrate and the CrN layer plays a key role in the resulted wettability property of the surface. |
636e417ebef5d44ffb5176d3 | 29 | We investigated the CrN thin films deposited on silicon by two different methods and for three durations of time deposition and we found important qualitative and quantitative changes in the properties of the deposited layer. Columnar and fine structures are noticed in the cross sections depending on the type of power supply used. The surface of the deposited films changes and becomes smooth and homogeneous for the HiPIMS power supply, but rough and with noticeable growth patterns for the DC power supply. |
63b88ff653a45aa65750cd4b | 0 | The precise construction of polyfunctional molecules remains a topic of great interest. Many targets of value are densely functionalized and are comprised of several contiguous stereocenters. The ring-opening of epoxides is an attractive method for the assembly of alcohol-containing stereoarrays. Our laboratory has a programmatic focus on the use of unusual nucleophiles for the ring-opening of both transient and stable electrophiles. As part of this line of inquiry, it occurred to us that a ring-opening of epoxides by pendant sulfamates and sulfamides would offer predictable access to vicinal amino-alcohols and would complement our previous efforts with intramolecular cleavage of aziridines by di-tert-butyl-silanol auxiliaries. There is no shortage of interesting amino-alcohols, and new methods for their construction are valuable. Unfortunately, a simple intermolecular aminolysis of epoxides often leads to intractable mixtures of regioisomeric products (Scheme 1A). Many creative investigators have developed "temporary tethering" approaches for the regioselective opening of epoxides with N-nucleophiles (Scheme 1B). Nevertheless, even with chelating Lewis acids, depending on the substrate, a mixture of regioisomers can still result. Covalent tethering offers a complementary approach (Scheme 1C). While a synthetic step must be expended to attach the tether, the subsequent cyclization is often highly regioselective and diastereoselective. To our surprise, cleaving epoxides with covalently tethered N-nucleophiles has not been extensively investigated. Sporadic reports exist with carbamate and acetamidate nucleophiles. Here, we detail our efforts to develop the first cleavage of epoxides by pendant sulfamates and sulfamides. |
63b88ff653a45aa65750cd4b | 1 | A methodology campaign cannot continue without access to the requisite test substrates. Fortunately, for homoallylic sulfamates, allylic sulfamides, and homoallylic sulfamides, standard Prilezhaev oxidation conditions allowed for reliable access to the desired epoxides (Scheme 2A). Allylic sulfamates are not stable to synthesis and isolation. Thus, allylic alcohols were first converted into the corresponding epoxides, and the sulfamate auxiliary was subsequently appended using the Johnson-Magolan protocol (Scheme 2B). Our first attempts at tethered ring-opening with a pendant sulfamate were informed by our previous work with di-tert-butylsilanol auxiliaries. In sharp contrast to our past experience, treatment of sulfamate A with either Lewis acids ( We were next interested in exploring the effects of various N-substituents on reaction performance (Scheme 3). Our optimized protocol of 1 equivalent of Bu4NOH in a biphasic solvent mixture of CF3toluene/H2O worked nicely with NH2-sulfamate 1 and N-Me-sulfamate 3 (Scheme 3, Entries 1-2). With N-Et-sulfamate 5, the reaction time had to be extended to 48 hours for full consumption of the starting material (Scheme 3, Entry 3). Bulkier substituents on the sulfamate nitrogen (Scheme 3, Entries 4-5), required us to abandon NBu4OH in favor of 1M aqueous NaOH. With N-Bn-sulfamate 7, an extended reaction time of 48 hours was required for optimal product formation using 1M aqueous NaOH in Et2O (Scheme 3, Entry 4). With N-p-methoxyphenyl-sulfamate 9, optimal product formation occurred with 1M aqueous NaOH in CF3-toluene at an elevated reaction temperature of 45 °C. Finally, the reaction invariably failed with N-cyclohexyl-sulfamate 11 over a range of conditions. From this series of experiments, we conclude that the cyclization is quite sensitive to substituents on the sulfamate nitrogen. In addition, as the steric bulk increases, switching from Bu4NOH to NaOH is required, and elevating the reaction temperature is beneficial in some cases. |
63b88ff653a45aa65750cd4b | 2 | Our optimized biphasic protocol (1 equiv. Bu4NOH, CF3-toluene/H2O, 23 °C) worked well with a range of sulfamate and sulfamide epoxide substrates (Scheme 4). In general, substrates cyclize cleanly and without observable side products. The mass balance of the reactions is good and is generally comprised of product and small amounts of unreacted starting material. Several functional groups are compatible with the reaction conditions, including aryl halides (Scheme 4, Entries 1 and 8), aryl ethers (Scheme 4, Entry 1), benzyl ethers (Scheme 4, Entry 2), and pendant sulfamates (Scheme 4, Entry 3). Both trans and cis sulfamate epoxides (Scheme 4, Entry 5) cyclized efficiently. While 6-membered rings were preferred in most cases, through judicious choice of the epoxide, 5membered heterocycles could be forced to form (Scheme 4, Entry 6). Products 13 (CCDC: 2231586), 27 (CCDC: 2231587), and 31 (CCDC: 2231588) were crystalline solids, and their x-ray structures have allowed us to confidently assign product identity and relative stereochemistry (See Supporting Information for full crystallographic details and additional structural proof). |
63b88ff653a45aa65750cd4b | 3 | Over the course of our survey, certain substrates behaved a bit differently than expected (Scheme 5A). With sulfamate epoxide 36, seven-membered ring 37 was the major product, forming in a 70% isolated yield. Here, the epoxide carbon attached to the aryl ring is highly activated for SN2 attack, and this likely underlies formation of an unusual sevenmembered ring in good yield. With tosylate substrate 39, tandem nucleophilic attacks took place to form pyrrolidine 40 in a single transformation. |
63b88ff653a45aa65750cd4b | 4 | No method is compatible with all substrates (Scheme 5B). Subjecting tert-butyldimethylsilyl ether substrate 41 to our reaction conditions was met with unproductive decomposition. We hypothesize that the instability of the TBS group to the strongly basic reaction conditions led to substrate failure. Substrates 42 and 43 also failed to provide product cleanly. In both cases, unproductive competition between exo and endo modes of nucleophilic attack likely led to substrate decomposition. |
63b88ff653a45aa65750cd4b | 5 | We were able to scale the cyclization reaction with sulfamate epoxide 1 from 0.2 mmol to 5.1 mmol without loss of yield or selectivity (Scheme 6A). The hydroxy group of 2 was converted into the corresponding TBS ether and the oxathiazinane ring was activated by appending a Cbz group (Scheme 6B). 45 served as a very effective synthon for oxathiazinane ring opening by sulfur, nitrogen, and oxygen nucleophiles (Scheme 6B). |
63b88ff653a45aa65750cd4b | 6 | In summary, we have of the first ring-opening of epoxides using pendant sulfamates and sulfamides. These reactions are promoted by base and proceed under mild conditions to afford oxathiazinanes and cyclic sulfamides with excellent diastereoselectivity and regiocontrol. The reactions scale well, and the products serve as synthons for ring-opening reactions. Given the ubiquity of stereochemical arrays in targets of value, we expect that this technology will be valuable to both academic and industrial organic chemists. |
67878e316dde43c908a7931a | 0 | Each dimer is composed of one of nine big monomers containing multiple potential binding sites and a small monomer binding to one of them. The resultant dimer arrangement is optimized at DFT level (PBE0+MBD), resulting in conformer geometries categorized as Linear, Semi-Folded, and Folded. For 16 equilibrium dimers, eight non-equilibrium conformations are designed along the dissociation of the non-covalent bond as illustrated by an example. q is a multiplicative dimensionless factor in the range of 0.9 to 2, which denotes the ratio of the inter-monomer distance to that of the equilibrium dimer. shape of the corresponding large monomer: 'Linear', in which the original chain-like geometry is mainly retained; 'Semi-Folded', in which parts of the large monomer are bent while other sections remain linear, and 'Folded', in which the big monomer encapsulates the smaller one. Thus, a variety of pockets with different packing densities is modeled by the QUID dimers. This classification is shown in terms of the radius of gyration and chemical diversity in Fig. of the Supplementary Information (SI). As a result, a wide range of interaction energies E int between the monomers is produced, ranging from -24.3 to -5.5 kcal/mol at PBE0+MBD level, with imidazole usually resulting in stronger non-covalent bonds (see Fig. of the SI). |
67878e316dde43c908a7931a | 1 | Next, a representative selection of 16 dimers is used to construct non-equilibrium conformations along the dissociation pathway of the non-covalent bond (along π-π or H-bond vector), modeling snapshots of a ligand binding to a pocket. These conformations were generated at eight distances, characterized by a multiplicative dimensionless factor q, defined as the ratio of the inter-monomer distance to that of the equilibrium dimer. The chosen values of q are 0.90, 0.95, 1.00, 1.05, 1.10, 1.25, 1.50, 1.75, and 2.00, where q = 1.00 denotes the equilibrium dimer. The structure of these non-equilibrium dimers was also optimized at PBE0+MBD level with the heavy atoms of the small monomer and the respective binding site kept frozen. The resulting systems demonstrate the varied E int spectrum for different pocket types via the structure categories in equilibrium and along the dissociation paths (see Fig. of the SI). The generation protocol for the 42 equilibrium and 128 (16x8) non-equilibrium dimers is schematically outlined in Fig. and detailed in the "Methods" section. These model systems represent a significant step forward in accurately investigating ligand-pocket interactions, characterized by robustly optimized molecular dimers that exhibit chemical diversity, larger size, and complex binding conformations. |
67878e316dde43c908a7931a | 2 | Analysis of non-covalent interaction components. A detailed characterization of the physical interactions in ligand-pocket systems is needed to both aid understanding and ensuring diversity in the coverage of interactions. To that end, a decomposition of E int of QUID systems with SAPT analysis was performed, specifically with the sSAPT0 version due to its good balance between accuracy and computational cost . The sSAPT0 E int predictions for the equilibrium dimers were found qualitatively consistent with those computed at the LNO-CCSD(T) level (MAE of 0.85 kcal/mol). The SAPT analysis provides insight into the energy components of the NCIs, namely induction, dispersion, electrostatic, and exchange contributions, which elucidates the balance of intermolecular interactions. Such a measure of the dispersion and electrostatic components has been used before to gain insight into specific protein-ligand interactions and will provide a solid basis for interpreting the differing predictions of E int from different QM methods. |
67878e316dde43c908a7931a | 3 | The variety in the spread of the sSAPT0 components for the QUID equilibrium dimers is shown in a stacked histogram in Fig. ). This showcases the diversity as a result of the different chemical environments and NCI types (see the associated NCI-plots in Fig. of the SI). Notably, the dispersion and electrostatic terms are strongly represented (see also Fig. )), while the induction contribution is the smallest, about 15-20% of the total value of E int . This indicates a supplemental role of the polarization effects on one monomer as a result of permanent dipoles of the other, with the imidazole-based dimers consistently having larger induction components compared to their benzene counterparts when settled in the same environment. For 9 dimers, electrostatics are the dominant term (see Fig. in SI), being particularly strong for the SF1I2 dimer. Such strong electrostatics are rare in the QUID systems as seen in Fig. ), where single outliers are seen for electrostatic contributions higher than -14 kcal/mol, while there is a skew towards smaller values in the range -8.0 to -4.0 kcal/mol. All 9 electrostatic-dominant cases involve imidazole as the small monomer. This is consistent with the presence of the two N atoms and their lone pair orbitals, capable of forming H-bonds and dipole-dipole interactions. For SF1I2, also a sulphonyl group is found near the binding site (see Fig. in SI), and the amino H atom in imidazole is interacting with the SO 2 functional group, possibly forming an H-bond, in addition to the second H-bond between the imidazole and an amino group in the large monomer. Dispersion is the dominant component for the other 33 equilibrium dimers, as expected given the choice of binding sites (pre-optimization) on accessible aromatic rings arranged for π-π stacking. The spreads of the electrostatic and dispersion components are contrasted against the E int spread in Fig. ). In considering the distribution, one should keep in mind that the positive (repulsive) exchange contributions arising from the overlap of the wavefunctions of the monomers within the S 2 approximation result in a partial offset of the negative components. E int ranges from -8 to -20 kcal/mol, clearly larger than π-π stacked or single H-bonds in small dimers, which usually contribute between -2 to -8 kcal/mol to the interaction energy . Furthermore, the dispersion component around -10 kcal/mol is larger than that obtained in a benzene dimer around -5 kcal/mol . Some dimers can also depict mixed NCI character, e.g. both an H-bond and ππ stacking in F1I3, or both sandwich ππ stacking and T-shaped interaction of aromatic groups simultaneously in F1I1. Therefore, we can conclude that the ligand binding is enhanced due to the collective interactions with the pocket with SAPT characterizing the specifics of those interactions and also revealing the complexity of the physical interactions of the QUID systems. |
67878e316dde43c908a7931a | 4 | Benchmarking the quantum-mechanical benchmarks towards "platinum standard". Reliable models for pocketligand systems rest on robust methods performing consistently and accurately in such systems. A prerequisite towards understanding and estimating the performance of current methods is the existence of reliable data, which can be challenging when results in literature obtained at "gold standard" level of computation have been found to disagree , and comprehensive and computationally expensive studies are needed to explore sources of the discrepancy . Hence, to establish a thoroughly Comparison of the interaction energies (E int ) computed using Diffusion Monte Carlo, QMC (0.025 time step) and LNO-CCSD(T) (CBS limit) for a selection of 13 equilibrium QUID dimers. Three cases are highlighted: L3I1 for which the methods are in perfect agreement, F2B1 for which the methods agree within their uncertainty estimates, and SF3I3 as the one case for which they are in slight disagreement. The NCI plots illustrating the non-covalent interactions in those molecular dimers are also shown . |
67878e316dde43c908a7931a | 5 | and compared for the two gold standard methods LNO-CCSD(T) and fixed-node diffusion Monte Carlo (QMC) to produce a "platinum standard". Within our methodology, both approaches were employed with particular care to achieve convergence, see the "Methods" Section for more details. Their results were compared for 13 of the most challenging cases [biggest disagreements between the DFT results (PBE0+MBD) and LNO-CCSD(T), see Fig. ) of the SI] and found in agreement within their uncertainty estimates in 12 of the 13 cases (i.e. 92%) as seen in Fig. . The MAE between the two methods is 0.32 kcal/mol compared to 0.38 and 0.39 kcal/mol mean absolute value of the uncertainty estimate for QMC and LNO-CCSD(T), respectively. |
67878e316dde43c908a7931a | 6 | The benchmark ab initio methods are in excellent agreement for the QUID systems, a result previously achieved for smaller and simpler systems, e.g. S66 dimers , but found unattainable for larger non-covalent systems with dispersion-dominated interactions, e.g. in the L7 dataset . In QUID, for only one case (SF3I3) out of the 13 studied, the LNO-CCSD(T) prediction with its uncertainty estimate lies outside one-sigma agreement with the QMC result but remains within two sigmas. This still means statistical consistency considering 68% and 95% assigned to one and two sigma intervals, respectively. Notably, this is the only dimer among the selected, for which two S atoms are present, both in proximity to the binding site. The sp 2 hybridization of the S atom in the 5-membered thiazole rings increases the negative charge on the N atom (as evidenced by the significant electrostatic component found in the SAPT analysis, see Fig. in the SI), thereby influencing the non-covalent bond. Unlike the L2B2 and L2B3 dimers, the NCI in SF3I3 is a N--H-N bond, where the bond exhibits a π-π stacking character with the thiazol group. Accordingly, the H-bond between the strongly charged N atom and the imidazole monomer could pose a challenge for achieving agreement between LNO-CCSD(T) and QMC results and deserves further investigation. Hence, based on the comparison between "gold standard" methods, we take LNO-CCSD(T) as a robust and reliable reference for E int of ligand-pocket NCIs in the complex QUID dimers. LNO-CCSD(T) results were subsequently obtained for all 42 equilibrium dimers and the full dissociation curves of a representative selection of 6 dimers (details in "Methods" section). |
67878e316dde43c908a7931a | 7 | Assessing the performance of DFT, semiempirical, and empirical methods. Given the "platinum standard" E int reference, we next conduct a comprehensive and reliable examination of its prediction and reliability obtained from different approximate computational methods for capturing NCIs in QUID equilibrium systems. This is done with the goal of identifying approximate methods that can be used in eventually building a trustworthy pipeline for calculating binding affinities. |
67878e316dde43c908a7931a | 8 | With the aim to provide a systematic investigation of QM and MM approximations, we include a wide selection of methods. First, we study DFT functionals with dispersion interactions, namely PBE0+MBD, PBE0+D4, ωB97X-D3, ωB97M-V, PBE0+XDM , M06-2X, PBE0+MBD-NL, and PBE0+TS. Second, among the SE methods we study the third-order Density Functional Tight Binding DFTB3+MBD and GFN2-xTB . From the available empirical classical FFs, we included GAFF2 |
67878e316dde43c908a7931a | 9 | (computed with AMBER) and CHARMM36 (computed with GROMACS) . The results are presented in Table of the SI and as a spread of E int predictions obtained with these methods with respect to the LNO-CCSD(T) reference values, ∆E int , shown in Fig. ). They are presented in ascending order of MAE, whose values can be found in the first column of Fig. ). |
67878e316dde43c908a7931a | 10 | The performance must be analyzed in the context of the intrinsic uncertainty estimate of the LNO-CCSD(T) method on the QUID dataset (mean value 0.39 kcal/mol). From this perspective, both PBE0+MBD and PBE0+D4 are within the uncertainty of LNO-CCSD(T), although a case-by-case analysis reveals deviations beyond the uncertainty of the benchmark data. |
67878e316dde43c908a7931a | 11 | Overall, it is reassuring that all recent DFT approximations yield rather accurate results. Even if some of them underestimate (PBE0+XDM, M06-2X, PBE0+MBD-NL) or overestimate (PBE0+TS) the reference interaction energies on average, the spread of deviations is rather narrow for all DFT methods. On the other hand, both empirical and semiempirical methods show a tendency to underbind, producing larger spreads and exhibiting large outliers. The most prominent outliers are found for CHARMM36 the DFTB3+MBD methods for ∆E int values in ranges from -12.5 kcal/mol to -5 kcal/mol (details in Fig. of the SI) and -7.5 kcal/mol to -4.5 kcal/mol, correspondingly. Examining the DFTB3+MBD outliers-SF1I2, SF3I1, and SF3I3 dimers-reveals that for SF1I2 the strongest interactions are driven by electrostatics, including contributions from a sulfonyl group at the binding site, while the SF3Ix dimers exhibit reactive thiazole groups. It is noticeable that the error distributions of GFN2-xTB (a SE method) and AMBER-GAFF2 (an empirical FF) are quite similar, although GFN2-xTB has a slightly lower average deviation from LNO-CCSD(T). Since GFN2-xTB was partially fitted to CCSD(T) data while AMBER-GAFF was not, this is not surprising. |
67878e316dde43c908a7931a | 12 | For AMBER-GAFF2, dimers with high electrostatic contributions result in larger errors (see Fig. of the SI) pointing to a limitation of partial charges. Contrarily, for GFN2-xTB, the higher errors appear to be associated with the local chemical environment. For example, the presence of P (in all 2Ix dimers), S (in all SF3Ix dimers) or Cl atoms (in both Folded F1I1, F1I3 |
67878e316dde43c908a7931a | 13 | show excellent agreement with each other (MAE = 0.22 kcal/mol) despite their differing functional forms and the distinct incorporation of dispersion terms, D3 and non-local correlation VV10, in ωB97X-D3 and ωB97M-V, respectively . This indicates that the critical similarity between ωB97M-V and ωB97X-D3, which sets them apart from the Minnesota functional M06-2X or PBE0+MBD, is the range-separation treatment of the DFT functional . Particularly important for QUID systems appears to be the long-range handling of the electron-electron interactions, as the short-range ones differ for the GGA and meta-GGA functionals . In the same vein, we consider the related PBE0+MBD and PBE0+MBD-NL methods (MAE of 0.47 kcal/mol) -we notice that the MBD-NL method increases the deviation compared to MBD in almost all cases except for the dimer with two S atoms and imidazole ligand, SF3I1-3 (see Fig. of the SI). We note that the MBD-NL functional was designed to achieve broad applicability to inorganic solids and molecular systems, while the original MBD (or MBD@rsSCS) method was developed for molecular systems. |
67878e316dde43c908a7931a | 14 | Let us now investigate more in-depth the performance of the two best performing methods PBE0+MBD and PBE0+D4, both of which obtain E int within the LNO-CCSD(T) uncertainty estimate (0.39 kcal/mol) at 0.33 kcal/mol and 0.37 kcal/mol, respectively. As the chemical environment and energetic balance in the NCIs proved to be a more distinguishing factor for the method than the structure types, we focus on a consideration of dispersion versus electrostatics contributions to E int . Overall, the MAE value of the 14 electrostatics-dominated dimers for PBE0+MBD is 0.26 kcal/mol, notably better than the PBE0+D4 results with an MAE of 0.56 kcal/mol. On the other hand, for the 28 dispersion-dominated dimers, PBE0+D4 yields 0.27 kcal/mol, while PBE0+MBD obtains a close MAE of 0.35 kcal/mol. This suggests that systems with stronger electrostatic interactions pose a greater challenge for the D4 dispersion correction. This could stem from the different underlying mechanisms of the two approaches for modeling long-range correlation effects . |
67878e316dde43c908a7931a | 15 | In summary, empirical and semiempirical methods have demonstrated variable performance for ligand-pocket model systems in QUID, yielding a MAE of about 1 kcal/mol or higher and exhibiting a tendency to underbind. In contrast, among the many DFT methods examined, PBE0+MBD and PBE0+D4 proved most effective in capturing the complex QM effects contributing to E int calculations, while PBE0+XDM also showed excellent performance as a pairwise dispersion method. These findings enhance our understanding of the applicability and limitations of the various investigated computational methods. |
67878e316dde43c908a7931a | 16 | Non-covalent bond dissociation pathways: non-equilibrium dimers. A key factor in modeling the dynamics of ligandpocket systems is the capability of a physical model to investigate systems out of equilibrium accurately. To that end, we have considered six representative dissociation curves (i.e. F2B1, F2I1, L2B3, L2I3, SF2B2, and SF2I2) and conducted an in-depth analysis of the performance of selected computational methods: PBE0+MBD, PBE0+D4, PBE0+XDM, GFN2-xTB, DFTB3+MBD, and AMBER-GAFF2. In Fig. ), we present the averaged results (over six dimers) for the 'Delta metric' ∆ that measures the agreement between the dissociation curves of a given computational method and the LNO-CCSD(T) reference (see Fig. of the SI). Indeed, it is evident that SE and classical FF methods, which are the tools of choice for biomolecular modeling, perform notably worse than DFT methods. The ∆ values per dimer are listed in Table of the SI, with the best performing equilibrium methods PBE0+MBD and PBE0+D4 achieving smaller ∆ values. These findings are confirmed by analyzing the average error of E int w.r.t. LNO-CCSD(T) at each q, see Fig. ) (corresponding six individual plots are available in Fig. of the SI). Notably, the performance of each method shows a strong dependence on the intermolecular distance. |
67878e316dde43c908a7931a | 17 | To elucidate the results, the dissociation curve profiles for all methods are presented in Fig. of the SI in aggregate and individually in Fig. of the SI. Interestingly, unlike DFT methods, AMBER-GAFF2 either underestimates or overestimates E int , depending on the dimer configuration. The discrepancies are more pronounced in configurations where dispersion components dominate the NCI, and for those dimers particularly at distances with factor q < 1.0, where dispersion interactions are stronger. On the other hand, both SE methods predominantly underestimate E int and fail to accurately capture the position of the minimum on the dissociation curve or its overall shape. This behavior changes only for GFN2-xTB at q < 1.0, where it overestimates E int in most cases. Hence, to assess the efficiency of the methods in different interaction regimes, two ranges have been defined: compressed for q ≤ 1.0 and elongated for q > 1.0. |
67878e316dde43c908a7931a | 18 | The MAE values for E int in the compressed and elongated regimes obtained using all methods are provided in The average is calculated at each multiplicative distance factor q (ranging from 0.9 to 2.0), defined as the ratio between the bond length and the equilibrium non-covalent bond length for the corresponding dimer. The average is shown for six selected molecular dimers: F2B1, F2I1, SF2B2, SF2I2, L2B3, and L2I3, using the same methods as in graph a). c) Average of the difference in the molecular polarizability of the dimer and the sum of isotropic polarizabilities of its corresponding monomers at each distance factor q. The results are shown as an average over all non-equilibrium dimers in the QUID dataset, split by structural type in Linear, Semi-Folded, and Folded. SI. As expected from previous results, PBE0+MBD, ωB97X-D3, and PBE0+D4 yield the best results in the compressed regime, while PBE0+MBD, PBE0+XDM, PBE0+D4, and ωB97M-V, perform best in the elongated regime. The SF2B2 and SF2I2 dimers proved to be the most challenging among those examined, likely due to the interaction of a 5-membered oxadiazole ring (C 2 N 2 O) via π-π stacking with the small monomer. The presence of two N and one O atoms in the aromatic ring contributes to an increase in the dipole moment and polarizability of the monomer, thereby enhancing both electrostatic and dispersion interactions. Interestingly, F2B1 and F2I1 are the easiest dimers to predict among the examined methods, as the molecular environment contributing to NCI is located within a few Å of the molecule. |
67878e316dde43c908a7931a | 19 | While the analysis of the dissociation curves has confirmed the performance of the methods for computing E int (vide supra), it has also revealed that the accuracy of SE methods and classical FF strongly depends on the distance range and the dimer configuration. This is a critical result, as both approaches are widely used to investigate intermolecular interactions in biomolecular simulations, raising questions about the reliability of the results obtained in molecular dynamics simulations carried out with empirical methods. |
67878e316dde43c908a7931a | 20 | Quantum-mechanical property space of QUID systems. To enhance our understanding of the effects of dimer configuration and intermolecular distances on the properties of pocket-ligand systems, several global and local physicochemical properties, in addition to E int , were computed for all equilibrium and non-equilibrium QUID dimers at PBE0+MBD level of theory (see "Methods" section for more details). A full list of properties, similar to those in the Aquamarine 29 dataset of large monomers, is provided in Table of the SI. Further, quantities describing the Hirschfield partitioning, i.e. Hirschfield volumes, ratios, charges, (scalar) dipole moments provides information for the electron response of an atom-in-molecule environment. |
67878e316dde43c908a7931a | 21 | To that end, analogous to E int , the difference in α between each dimer and the sum of its corresponding large and small where the radius is the difference in magnitude of the vdW forces (given as a percentage) and the angle is the arccosine between the dot product of the force vectors. Four such polar plots depict the distributions of the forces acting on all the N, O, Cl, and S atoms, respectively, in the equilibrium QUID dimers. b) Overall distribution of the magnitudes of the atomic vdW forces on all atoms for equilibrium QUID dimers using the MBD, D4, and XDM methods. c) An example of the deviation in atomic van der Waals (vdW) forces between different methods is visualized for the SF2I2 dimer (produced by the FFAST software ). The range is represented by colors varying from blue to red. monomers, ∆α, was calculated for all 128 non-equilibrium conformations. The average ∆α values for each structure type as a function of the distance factor q are plotted in Fig. ). Overall, the three structure types exhibit an almost linear behavior, with slight deviations near the equilibrium distance for the Linear and Semi-Folded structures. According to a recent concept of chemical bonding based on α, proposed by D. Hait and M. Head-Gordon , this linear behavior suggests no significant modification in the covalent bond arrangements along the dissociation curve. The variation can thus be attributed to the self-consistent screening effect between the monomers. Indeed, in both Linear and Semi-Folded structures, the small monomer affects fewer atoms of the large monomer. In contrast, in Folded structures, the small monomer remains within 5 Å of a significant number of atoms of the large monomer, substantially influencing the electrostatics and dispersion in the pocket site, resulting in a steeper curve. Moreover, no correlation between α with E int and µ emerges from the exploration QUID dimers (see Fig. of the SI). On the one hand, this could suggest flexibility for rational ligand design as observed for small molecules of up to 7 heavy atoms . At this stage, we can ascertain the interplay between electrostatic and dispersion interactions in a structurally and chemically complex local environment in the QUID pocket-ligand proxies requires exact QM models beyond the capture of a single key global property in the high-dimensional chemical compound space. |
67878e316dde43c908a7931a | 22 | Another relevant property for understanding NCIs in QUID dimers is the atomic forces, which are widely used to parameterize MLFFs for (bio)molecular systems. Indeed, the molecular conformational sampling at a given temperature strongly depends on the accuracy of the chosen computational method in adequately describing the forces acting on the atoms in the molecule. Accordingly, we have analyzed the atomic force distributions using a selection of DFT methods: the best performing ones PBE0+MBD and PBE0+D4, and the well-performing pairwise PBE0+XDM method. Since these methods share the DFT functional and our primary interest lies in NCIs, the focus will be on the vdW components of the atomic forces. |
67878e316dde43c908a7931a | 23 | The differences in the vdW atomic forces are examined in terms of their magnitude and direction, treated as two distinct yet interrelated driving factors in MD simulations. The results are presented in Fig. ) as a radial plot illustrating the difference in forces of D4 and XDM with respect to MBD. In this plot, the angular part represents the arccosine between the dot product of two force vectors, while the radius is the difference in force magnitudes, scaled by the MBD force. This analysis focuses on 9/19 the differences observed per atom type, with the results for N, O, Cl, and S atoms shown in Fig. ) (see other atom types in Fig. of the SI). |
67878e316dde43c908a7931a | 24 | The overall force analysis is provided by the plot in Fig. ), which shows that MBD yields the smallest forces on average. The vdW force magnitudes are generally not much higher than 1 kcal/mol/Å; this is expected given that equilibrium geometries are involved. Higher vdW forces would be expected for non-equilibrium geometries. Further, largest discrepancies in force magnitude were found in D4 compared to XDM. This is corroborated by the distribution of the atomic force magnitudes for all atoms. The most significant outliers in force magnitude discrepancies are associated with D4 and Fig. ) demonstrates that those outliers of up to 3-4 times the magnitude of the MBD force are found on C atoms in descending order for L2I3, L2B3, L3I1, and L3I2 (Fig. of the SI). The FFAST software 66 allows for visualization of the discrepancies between the vdW atomic forces to confirm that the C atoms were found at the binding site, an example of such a visualization can be seen in Fig. ) depicting SF2I2. The difference in force magnitudes is visible not only on the atoms of the binding site as expected but also in proximity to it showing the differing impact of the ligand interactions in the different methods. For the SF2I2, there is a notable difference between the comparisons of XDM and D4 w.r.t. MBD (the MBD depiction is available in Fig. of the SI). |
67878e316dde43c908a7931a | 25 | This also holds true in general for the vdW forces on the Cl atoms in the QUID dimers. As seen on Fig. ) XDM is in better agreement with MBD than D4 for Cl atoms, and the same is true for the other halogen element F, as well as the few P atoms (Fig. of the SI). Hence, the presence of more electronegative atoms can hint at a systematic difference between the different vdW atomic forces. In that vein, a particular outlier is seen for the S element, with a 113% gap between the magnitude of the MBD and D4 forces for the SF1I1 dimer, in the sulfonyl group of the binding site of the imidazole ligand (Fig. ). |
67878e316dde43c908a7931a | 26 | Interestingly, there is a split in the vdW force directions between 'heavier' atoms in the QUID dimers, i.e. O, F, Cl, S, and P and the 'lighter' ones i.e. H, C, and N. The 'lighter' atoms represented more in organic molecules demonstrate a larger spread of angle difference between the forces up to 120 • -180 • . By construction, for the pairwise D4 and XDM methods, the vdW force is a simple vector sum where all force vectors are aligned along the vector connecting pairs of atoms. In contrast, many-body effects that are treated to infinite order in MBD can thus substantially alter the force directions, and this difference is much more pronounced than for force magnitudes. This could have a visible effect on MD trajectories, although these implications remain to be assessed in a future study. |
67878e316dde43c908a7931a | 27 | The observed difference in force directions has potential repercussions for MLFF methods, where molecular data is routinely optimized at one level of theory, method or functional or dispersion correction, and computed at a different one to serve as input or reference for energies and forces. Unfortunately, while for E int we have highly accurate LNO-CCSD(T) reference data and even the ability to achieve a "platinum standard" confidence by comparing with QMC, obtaining forces at benchmark ab initio level is prohibitively expensive. Current research in developing gradients for LNO-CCSD(T) and QMC could facilitate a reference benchmark for the vdW forces in the future and provide a clearer picture for the accuracy of the DFT methods for MD of ligands binding to a pocket. |
67878e316dde43c908a7931a | 28 | The "Quantum Interacting Dimer" (QUID) benchmarking framework was developed here to redefine the state of the art in QM-based modeling of ligand-pocket motifs. Currently, QUID contains 170 structurally and chemically diverse large molecular dimers (42 equilibrium and 128 non-equilibrium) of up to 64 atoms, including the H, N, C, O, F, P, S, and Cl chemical elements, encompassing most atom types of interest for drug discovery purposes. This diversity enables a single dimer to exhibit multiple types of steric effects and NCIs simultaneously, including, but not limited to, π-π stacking, hydrogen, and halogen bonds. |
67878e316dde43c908a7931a | 29 | Accordingly, we conducted a series of analyses that provided valuable insights into inter-and intramolecular interactions of these model protein-ligand systems from a QM perspective. Indeed, the energy decomposition of the interaction energy E int , as obtained through SAPT analysis, revealed that ligandpocket interactions are predominantly governed by dispersion and electrostatics-types of interactions often inadequately represented by MM methods. Moreover, we defined a "platinum standard" of accuracy for E int of ligand-pocket interactions by contrasting the results with the "gold standard" methods such as LNO-CCSD(T) and QMC. Notably, the previously reported disagreement between LNO-CCSD(T) and QMC for large non-covalent systems was not observed for the QUID dimers, and the overall discrepancy is small, approximately 0.3 kcal/mol. Thus, our findings demonstrated that among all studied MM and QM approaches, DFT methods such as PBE0+MBD and PBE0+D4 achieve excellent agreement with the costly LNO-CCSD(T) method in the determination of E int for equilibrium and non-equilibrium dimers. Additionally, we have identified certain limitations in widely used semiempirical (e.g., GFN2-xTB and DFTB3+MBD) and MM methods (e.g., AMBER-GAFF2 and CHARMM36) for investigating complex non-covalent motifs, which raises questions about their reliability in binding affinitny simulations. These intriguing results highlight the relevance of determining the appropriate level of theory to accurately characterize protein-ligand systems, particularly in the development of extensive QM datasets utilized in physical method benchmarking and ML-based investigations. E int (at PBE0+MBD level), enabling the electronic characterization of chemical environments within ligand-pocket motifs-a limitation of current benchmark frameworks, which primarily focus on structural and energetic features. The structural dependence of the polarizability change in dimers as a function of the monomer separation, as well as the lack of correlation between global electronic properties and E int , offers a new perspective for understanding the NCIs in protein-ligand systems. |
67878e316dde43c908a7931a | 30 | These insights can rationalize the design of drug-like molecules targeting specific pocket sites with a desired set of QM properties . The observed discrepancies in the van der Waals (vdW) component of the atomic forces using MBD, D4, and XDM methods also show the importance of investigating additional properties beyond the traditional E int in non-covalent complexes. An inaccurate description of vdW forces can strongly impact the reaction pathway and the resulting binding pose when simulating the interaction of ligands with protein pockets. Hence, QUID has the potential to revolutionize standard procedures in approaching the modeling of ligand-pocket interactions in physical and ML-based frameworks by providing global and local electronic property data for large molecular dimers and their building blocks, which are critical for a faithful incorporation of long-range effects . |
67878e316dde43c908a7931a | 31 | In summary, the QUID benchmark framework presents a novel approach for accurately analyzing ligand interactions at various protein binding sites, facilitating the development of robust QM datasets for reliable predictions of binding affinities and structural conformations. While the insights gained from this work highlight the importance of an appropriate QM description for inter-and intramolecular properties of ligand-pocket motifs, we acknowledge that further efforts should incorporate more flexible and charged pocket structures, as well as solvation effects . The sampled chemical space should ultimately encompass full pocket-ligand molecular systems, similar to those in MM datasets (e.g. PL-REX and QR50 ), to enhance the reliability of the findings. We expect this work to pave the way for a more informed use and refinement of physical and chemical models for simulating ligand-pocket interactions, offering particular value in fine-tuning MLFFs and ML-augmented semiempirical models, which are increasingly integrated into screening pipelines for drug discovery. |
67878e316dde43c908a7931a | 32 | Generation procedure of QUID systems. A schematic representation of the procedure used to generate the 170 (equilibrium and non-equilibrium) QUID dimers is outlined in Fig. . Each QUID dimer consists of one of nine chemically diverse large monomers (of ∼50 atoms) selected from the Aquamarine dataset of drug-like molecules , paired with a small monomer, either benzene C 6 H 6 or imidazole C 3 H 4 N 2 . Each large monomer features at least 2 sterically accessible aromatic rings, which serve as binding sites. The larger monomer initially has chain-like geometry, modeling a protein pocket, while the small molecule represents the ligand. The chemical composition of QUID systems reflects this purpose: each conformer contains not only C, N, O, and H atoms but also at least one of the elements F, P, S, or Cl for eight out of the nine large monomers. These elements are incorporated into the following functional groups: substituted five-or six-membered aromatic rings, aliphatic heterocycles, ketones, ethers, hydroxyls, amines, haloalkanes, and sulfonyl. Moreover, the choice of benzene and imidazole molecules enables a comparison of the effects of small aromatic rings and amphoteric compounds (imidazole can be both H-bond donor and acceptor), which is a key characteristic for many compounds of biomedical significance such as amino-acid histidine and anti-inflammatory drugs . |
67878e316dde43c908a7931a | 33 | In the initial dimer conformation, prior to structural optimization, benzene and imidazole occupy the same position relative to a given binding site, specifically at 3.55±0.05 Å parallel to the site. A similar distance was used in the generation procedure of S66x8 dataset . Here, the aromatic ring of the small monomer was aligned with that of the binding site, ensuring that the distance between the corresponding pair of heavy atoms on the two rings remained within a margin of 0.05 Å. Each dimer was then optimized using non-empirical hybrid density functional theory (DFT) with many-body dispersion approach (range-separated self-consistent screening, MBD@rsSCS), namely PBE0+MBD, in conjunction with tightly-converged numeric atom-centered orbitals ("tight settings"), as implemented in the FHI-aims 72 software. We classify the resulting structures into three categories based on their shape: 'Linear', where the original chain-like geometry is mainly preserved; 'Semi-Folded', where parts of the large monomer bend while other sections remain linear; and 'Folded', where the large monomer encases the smaller ones. Following this classification, the first letter of the dimer names corresponds to the structural category: 'F' for Folded, 'SF' for Semi-Folded, and 'L' for Linear. The small monomer is indicated by the letter 'B' for benzene or 'I' for imidazole, while the number that follows indicates the binding site (in QUID dimers as found viable post-optimization). The chemical diversity of the equilibrium QUID dimers (represented by the counts of non-H elements) is plotted in Fig. of the SI against the radius of gyration, illustrating the spread of the three structural categories. Furthermore, we acknowledge that modeling ligand-pocket systems would be incomplete without computing nonequilibrium dimer conformations, which are crucial for understanding the binding of drugs to protein pockets. To this end, we generated eight optimized out-of-equilibrium structures along the reaction pathway of the non-covalent bond for 16 representative molecular dimers (F2B1, F2I1, F2B2, F2I2, SF2B1, SF2I1, SF2B2, SF2I2, SF2B3, SF2I3, L2B1, L2I1, L2B2, L2I2, L2B3, L2I3). These non-equilibrium conformations were constructed with inter-monomer distances ranging from 3.2Å to 7.1Å, defined by a dimensionless multiplicative factor q, which represents the ratio of the inter-monomer distance in a given conformation to that in its equilibrium state. The chosen values of q are 0.90, 0.95, 1.00, 1.05, 1.10, 1.25, 1.50, 1.75, and 2.00, as shown in Fig. . For the π-π interactions between aromatic rings, the dissociation vector was defined as the distance between the plane of the aromatic ring on the large monomer and the center of mass of the small monomer. In the case of H-bonds, dissociation was measured between the proton acceptor and donor atoms. All the non-equilibrium geometries were also optimized at PBE0+MBD level with "tight" settings using FHI-Aims (version 221103). Unlike the equilibrium dimers, the heavy atoms of the binding site and small monomer were kept frozen in their respective positions during optimization. An example of the dissociation geometries for one of the non-equilibrium dimers is shown in Fig. . |
67878e316dde43c908a7931a | 34 | Counterpoise corrections were applied to PBE0+MBD and CCSD(T) single-point calculations. The basis set superposition error was negligibly small (under 1.5%) for the former and ca. 4% on the average for the latter in the complete basis set (CBS) limit (see results in Fig. of the SI). To investigate the level of agreement among QM methods for calculating E int of QUID dimers, we have considered a selection of well-performing DFT functionals, including M06-2X , ωB97X-D3 21 , ωB97M-V . Additionally, the PBE0 functional was combined with multiple two-body or many-body corrections: MBD (range-separated self-consistent screening (MBD@rsSCS) approach), MBD-NL (Non-Local), XDM (eXchange Dipole Moment), TS (Tkatchenko-Scheffler)-vdW , D4 , ωB97X-D3 21 , ωB97M-V 60 . These calculations were performed using either the FHI-Aims (version 221103 )software with "tight" settings or the Psi4 software (version 1.9.1) with the quadruple-zeta def2-QZVPPD basis set. Notably, for the XDM method, a specific parametrization for the PBE0 functional was applied and then computed using FHI-aims software on "tight" settings (a 1 =0.4710 and a 2 =2.3857) . SAPT energy decomposition calculations were carried out at the sSAPT0/jaDZ level employing the Psi4 software 75 (version 1.9.1). At the semiempirical level, E int was calculated via single-point calculations using Density Functional Tight Binding 15 DFTB3+MBD |
67878e316dde43c908a7931a | 35 | Regarding MM methods, the results for AMBER were obtained using Openbabel for molecular format conversion. The parametrization with AmberTools and GAFF2 11 required manual assignment and adjustment of bonds for more complex cases, such as ring interactions, as well as modification of the self-consistent loop limits for the F2I1 dimer. The CHARMM36 calculations were conducted using GROMACS following a CGenFF2 82 parametrization. For these calculations, manual inclusion of the dihedral angles for the flexible side chains, such as the 'C-C-N' type, was necessary. An example is the L4B1 dimer, which was assumed to exhibit relatively low flexibility due to the nature of its bonds and chemical environment. A list of the E int storage details is available in Table of the SI. |
67878e316dde43c908a7931a | 36 | Additionally, the optimized structures of equilibrium and non-equilibrium QUID dimers were also utilized for more accurate QM single-point calculations using PBE0+MBD level of theory to compute other physicochemical properties (as detailed in Table of the SI). For these calculations, we have used the FHI-aims code together with "tight" settings for basis functions and integration grids. Energies were converged to 10 -6 eV and the accuracy of the forces was set to 10 -4 eV/Å. The convergence criteria used during self-consistent field (SCF) optimizations were 10 -3 eV for the sum of eigenvalues and 10 -6 electrons/Å 3 for the charge density. The MBD energies and MBD atomic forces were here computed using the range-separated self-consistent screening (rsSCS) approach , while the atomic C 6 coefficients, isotropic atomic polarizabilities, molecular C 6 coefficients and molecular polarizabilities (both isotropic and tensor) were obtained via the SCS approach . Here, we have also computed van der Waals forces using D4 and XDM methods. Hirshfeld ratios correspond to the Hirshfeld volumes divided by the free atom volumes. The TS dispersion energy refers to the pairwise Tkatchenko-Scheffler (TS) dispersion energy in conjunction with the PBE0 functional . The vdW radii were also obtained using the SCS approach via R vdW = α SCS /α TS 1/3 R TS vdW , where α TS and R TS vdW are the atomic polarizability and vdW radius computed according to the TS scheme, respectively. Atomization energies were obtained by subtracting the atomic PBE0 energies from the PBE0 total energy of each molecular conformation. |
67878e316dde43c908a7931a | 37 | LNO-CCSD(T) reference interaction energies. The large CCSD(T) computations well beyond the limits of conventional implementations were performed with the highly optimized local natural orbital (LNO) CCSD(T) method in the MRCC quantum chemistry suite. First, a detailed basis set and LNO approximation convergence study is performed to determine the most efficient LNO and basis set settings that provide high accuracy for the interactions relevant to QUID. To that end, the interaction energies were tested for three representative dimers, namely SF2I2 (Fig. ), F2B1, and L2B3 (Fig. of the SI), at 1×, 0.9×, and 2.0× equilibrium distances. Here, we used the systematically improving series of aug-cc-pV(X+d)Z basis sets with X = D, T and Q as well as Normal (N), Tight (T), and very Tight (vT) LNO thresholds. The convergence toward the complete basis set (CBS) limit was accelerated for the HF and correlation 87 energies via CBS extrapolation [CBS(X,X + 1)]. To accelerate the convergence toward conventional CCSD(T), that is the local approximation free (LAF) limit, the LAF extrapolations Our best estimate of CCSD(T)/CBS for the above nine dimers is given by a composite ( Ē) scheme benefiting from CBS(T,Q) and LAF T-vT extrapolations as well as counterpoise corrections : |
67878e316dde43c908a7931a | 38 | Inspecting all nine cases (Fig. and Fig. ), we find fast convergence with the LNO settings reaching ca. ) can be completed in ca. 30-70 hours for each dimer on 16 cores and at most 20 GB memory, making it ideal for high-throughput computations even on clusters with short wall-time limits. For the sake of completeness, the frozen-core approximation and conventional auxiliary basis sets were employed for all LNO-CCSD(T) interaction energies. |
67878e316dde43c908a7931a | 39 | QMC Interaction energy calculation. We performed Fixed-Node Diffusion Monte Carlo (FN-DMC) calculations for QUID dimers in which there is a discrepancy between E int values at CCSD(T) and PBE0-MBD results, corroborating the correct E int with another accurate QM method. Our FN-DMC wave function ansatz was built using a single Slater determinant in addition to a Jastrow factor with one-body, two-body, and 3/4-body terms to account for cusps conditions at the nuclei, fermionic pair correlations, and product of pair correlations at the field of the nuclei, respectively. See references for further details about the electronic wave function ansatz. The molecular orbitals of the Slater determinant were taken from a previous DFT calculation employing ORCA code with the Local-density approximation (LDA) exchange-correlation functional, a cc-pVTZ basis set for all atoms, and the ccECP pseudopotential . |
67878e316dde43c908a7931a | 40 | All variational parameters of the Jastrow factor were variationally optimized at the VMC level with the stochastic reconfiguration optimization method , while molecular orbital coefficients, basis set contraction coefficients, and exponents were kept fixed from the initial DFT calculation. The optimized VMC wave functions were taken as guiding functions in the FN-DMC calculation, in which we also employed the ccECP pseudopotentials to approximate core electrons for each atom, integrated with the determinant localization approximation (DLA) . Consequently, the FN-DMC calculations were computed at two time-steps of 0.050 and 0.025 (a.u.), using 12800 walkers divided into 300 blocks, each 100 steps long, for a total of 4 × 10 8 sampled configurations. For some systems, it was required to run an additional third calculation with a time-step of 0.015 a.u. to get statistical agreement within 1σ in the observed binding energies. In Fig. we displayed the time-step convergence against the CCSD(T) reference values. Both VMC wave function optimization and DMC calculations were performed with the QMeCha code . |
649ebcfb9ea64cc16737f53c | 0 | Charge transfer in organic fluorophores is a fundamental photophysical process that can be either beneficial, e.g., facilitating thermally activated delayed fluorescence, or detrimental, e.g., mediating emission quenching. It has been previously shown that both protonation and N-alkylation of quinoline containing compounds allows straightforward synthetic control of the charge transfer, emission energy and photoluminescence quantum yield (PLQY) (Scheme 1a), with the effect of alkylation mirroring those of protonation while being necessarily more permanent. Control of photoluminescence and photophysical properties in organic molecules has significant appeal for a range of applications from organic light-emitting diodes (OLEDs) to biological imaging and sensing. Given the importance of the charge transfer (CT) process in influencing key photophysical parameters developing structure-activity relationships (SARs) to describe CT formation will be of interest to a wide range of stakeholders. In medicinal chemistry the development of SARs is standard practice, but a similarly detailed understanding of the structural properties controlling charge transfer has yet to be fully developed. Since protonation and N-alkylation have a significant and predictable effect on CT states in quinoline-containing compounds these provide a good starting point for exploring these vital fundamental processes and developing effective SARs. |
649ebcfb9ea64cc16737f53c | 1 | One prominent set of quinoline containing compounds are the cinchona alkaloids, in particular the historically important anti-malarial quinine. However, both the chromophore and overall architecture of quinine is relatively inflexible due to its isolation from a natural source, limiting the application of such species in the development of SARs. Further, while methods for the synthesis of cinchona alkaloids exist, they are often insufficiently modular to permit significant structural variation. We therefore considered whether structural analogues of quinine might demonstrate similar PL properties while being accessible via a more modular synthetic strategy. This would provide easy access to families of compounds possessing both variable chromophores and intramolecular distances, thus enabling the unpacking of the fundamental processes of through-space charge transfer (TSCT) between the cage nitrogen lone pair and quinoline unit. Together this would enable the development of a basic SAR around these parameters. Further, getting a greater handle on the propensity of forming this TSCT would have important impact on the study of dimeric, excimeric and exciplex systems for use in OLEDs and other applications where intramolecular and intermolecular interactions are crucial. These include use in fluorescence microscopy and biological imaging, where tuneable properties would permit control of wavelengths of excitation and emission, as well as the nature of charge transfer. |
649ebcfb9ea64cc16737f53c | 2 | We recently reported that photochemically-accessed tricyclic aziridines undergo efficient Pdcatalysed cascade processes to provide cage structures 5 (Scheme 1b) which, while initially lacking a quinoline chromophore, appeared to offer similar rigidity, functionality and N-aryl distances as those observed in quinine. Importantly, the modular nature of this synthesis enables facile interchange of the components employed within it. We therefore considered whether this system could be used as a quinine surrogate, enabling modification of the aryl chromophore via modification of allylic amine 4 together with variation of both N-aryl distance and the tetracyclic core via the substitution of tether 6 and aziridine 3 respectively (Scheme 1c). We thus undertook the synthesis of a range of quinolinecontaining allylic amines 6 in order to explore their Pd-catalysed reaction with aziridines 3 and develop a series of photophysical SARs focused around the switching on and off of charge transfer. |
649ebcfb9ea64cc16737f53c | 3 | Quinoline-containing tethers 11a-c were prepared straightforwardly though standard heterocyclic synthesis as shown in Scheme 2. While compound 10a was commercially available, quinoline aldehydes 10b and 10c were prepared in moderate yield via Skraup synthesis followed by a benzylic oxidation with selenium dioxide. All three aldehydes then underwent an efficient Grignard addition/acetylation sequence to yield the desired tethers 11. All syntheses proved to be relatively scalable, enabling gram quantities to be formed in some cases. Routes to tether 15 were made somewhat more challenging by the need for the construction of a geminally-substituted alkene. While various less modular, multi-step approaches appeared possible, we observed that Kutsumura et al. recently reported 14 a one-pot bromination/cross-coupling strategy to enable the synthesis of 2-aryl substituted allylic alcohol derivatives. Such an approach would enable late-stage variation of the quinoline unit by introducing it in the penultimate cross-coupling step and would further lead to a short and efficient overall synthesis. We therefore prepared 4-methoxybenzyl (PMB)-protected allylic ether 12a. Bromination and cross-coupling proceeded as reported, furnishing the PMB-protected quinoline 14a in reasonable yield (Scheme 3). However, PMB deprotection proved troublesome, leading to either full recovery of the starting material or complete degradation. We therefore considered different protecting groups and after some experimentation found that tetrahydropyranyl (THP) worked efficiently. While a one-pot bromination/cross-coupling approach also proved possible in this case, isolation of bromide 13b gave improved results in the subsequent cross-coupling, with quinoline 14b being isolated in good yield with a reaction that could be scaled to 1 g. Deprotection proved straightforward under standard acidic hydrolysis conditions, and acetylation yielded the desired tether 15b in good yield. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.