id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
60c74984bb8c1a03413dae4d | 45 | FDE-diab calculations of isolated inner pair models from PSI predicted about 96% of spin density being localized at P A . Although the exact spin-density ratio for the inner pair of PSI is unknown, various experimental measurements (for example, see Refs. ) predict from 75% to 100% of the spin density to be localized at the P B cofactor. Therefore, the FDE-diab calculations for isolated inner pair models contradict experimental results. This, however, is caused by a complete neglect of the co-factorprotein interactions. Inclusion of the protein environment into FDE-diab calculations changed the spin-density distribution considerably and resulted in the ratio of 24%/76% in favor of P B , which is in good agreement with available experimental results. We also managed to reproduce most of experimentally measured P B spin populations using the large protein binding pocket model. However, it turned out to be difficult to reliably calculate some small atomic spin contributions. |
60c74984bb8c1a03413dae4d | 46 | However, FDE-diab calculations for a binding pocket model show a somewhat larger degree of spin-density localization with the ratio of 93%/7%. It is also interesting to note that the calculated spin populations for the isolated BChl a and the D A co-factor from the binding pocket are very similar. This could indicate that the SP in the BRC is essentially unperturbed by the protein environment. |
60c74984bb8c1a03413dae4d | 47 | Summarizing the results presented in this work, we conclude that the FDE-diab methodology is a valuable and robust tool for spin-density calculations of large molecular systems. Avoiding the consequences of the DFT overdelocalization error in the intermolecular regime, it allowed us to obtain reliable spin-density distributions for a number of RC models and to gain a deep insight into the asymmetry of the corresponding SPs. Moreover, the spin-density distribution was found to be highly sensitive to structural arrangements of co-factors. Spin-density distributions of RC models as calculated in this work were not accessible before with existing computational techniques. Further extensions of the FDE-diab approach will soon allow us to go beyond the simplistic two-state model and |
64eb2db63fdae147faf1e1d3 | 0 | These disparities suggest that the reaction mechanisms go beyond the previously known "zippingoff" mechanism, where the removal of terminal functional groups (i.e., -COO -, -SO3 -, and -CH2CH2-SO3 -) exposes the RF-CF2• for stepwise defluorination of the two C-F bonds and oxidation of the carbon into CO2 (Fig. ). The odd/even number of -CF2-in PFCAs appears to have interesting effects on the gap from 100% defluorination (<10% for n=2,4,6 versus >20% for n=3,5,7 CnF2n+1-COO -). It is also very interesting to observe the much lower defluorination from all three n=8 structures than their n=2 and 4 analogs. Elucidating the underlying mechanisms go beyond the scope of this work but definitely warrants further study. PFSA and FTS by EO treatment and the previously known "zipping-off" pathway. Reaction conditions: individual PFAS (25 µM, except 1000 µM for TFA for the ease of F -measurement) spiked in 20 mL water with 100 mM Na2SO4 as electrolyte; current density of 15 mA/cm 2 applied to a 16 cm 2 BDD anode. Data are presented as mean values of triplicates ± standard deviation. |
64eb2db63fdae147faf1e1d3 | 1 | For the degradation of individual PFAS structures, EO has an overwhelming advantage over UV/S. The strongly oxidative environment rapidly destroyed n=4 FTS and achieved >95% defluorination. But this compound is highly recalcitrant under UV/S because the short C4F9- moiety segregated by the -CH2CH2-linker does not have a weak C-F bond for easy defluorination by eaq -. This experimental finding corroborates the insights from the DFT calculation discussed in the earlier section. Moreover, in comparison to UV/S, EO achieved much faster (10 h versus >24 h) and deeper (~100% versus 78%) defluorination of PFBS. Hence, EO has higher tolerance with short-chain PFAS than two other heterogeneous technologies-plasma and sonication-both encountered challenges from PFBS (C4F9-SO3 -) and PFBA (C3F7-COO -). Based on the above experimental findings, we hypothesized that EO could be placed after UV/S to obtain the best treatment result for multiple reasons. To validate the hypotheses above, we developed a primitive UV/S-EO layout to treat PFOS, PFOA, and PFBA, all of which are representative PFAS and could not be 100% defluorinated by EO (Fig. ). In the first stage of UV/S treatment, PFOA and PFBA were completely removed within 30 min (Fig. ). After the parent PFCAs quickly disappeared, the defluorination from the transformation products continued. In previous studies, the maximum defluorination from PFCAs under the same UV/S condition took 4-8 h. However, for the UV/S-EO layout, we arbitrarily stopped the UV/S treatment after 2-3 h when the increase of defluorination became sluggish. For the more recalcitrant PFOS under UV/S treatment, the defluorination accompanied the parent compound removal. We stopped UV/S when most parent PFOS disappeared at 5 h (Fig. ). The following EO treatment increased defluorination to 100% for all three PFAS. Transfomration product (TP) analyses verified our mechanistic hypotheses. The UV/S treatment of n=7 PFOA generated a series of shorter-chain n=1-6 PFCAs (Fig. ) as quantified by a triple-quadrupole mass spectrometer (QQQ MS/MS). These PFCAs are attributed to the wellknown decarboxylation and a recently identified C-C bond cleavage mechanism. The UV/S treatment removed most of the PFCA TPs within 3 h. Quadruple time-of-flight high-resolution mass spectrometer (Q-ToF-HRMS) found a series of hydrodelfuorination products (Fig. ) from the parent PFOA (C8F15O2 -) and the chain-shortened PFHpA (C7F13O2 -). The MS peaks for C8HF14O2 -and PFHpA showed similar abundance, indicating that the two transformation pathways proceeded in parallel and were equally significant (Fig. ). The UV/S degradation of hydrofluorinated TPs, such as C8HF14O2 -and C8H2F13O2 -, were much slower than the perfluorinated PFOA (Fig. versus Fig. ). The detection of deeply hydrodefluorinated TPs (e.g., C8H12F3O2 -and C7H10F3O2 -) is consistent with the previous study using a different photoreactor setting and a quadrupole Orbitrap HRMS instrument. The three residual C-F bonds with high recalcitrance against UV/S were most probably on the terminal CF3-. The switch to EO mode generated short-chain PFCAs again (Fig. ) from various hydrodefluorinated TPs. The sharp increase of TFA suggested that hydrodefluorination by UV/S occurred on carbon atoms near the terminal CF3-. With the extension of EO treatment, all PFCA TPs (Fig. ) and hydrodefluorinated TPs (Fig. ) were destroyed to negligible concentrations, as evidenced by the defluorination to ~100% (Fig. ). The near-quantitative defluorination of individual PFAS structures motivated us to apply UV/S-EO for AFFF treatment at ambient conditions. For fire suppression, the original AFFF liquid was typically diluted about 100-fold. It was further diluted after entering the water environment. To date, only a few studies have reported treating diluted AFFF (total fluorine 0.16-27 mg L -1 ) by individual EO, UV/S, and plasma technologies. None of these nonthermal methods realized ~100% defluorination (Table ). |
64eb2db63fdae147faf1e1d3 | 2 | The total fluorine in the original AFFF was measured as 10 g L -1 by combustion ion chromatography (Table ). Nineteen of the 30 targeted PFAS structures were detected in AFFF by QQQ MS/MS (Table ). The three most abundant targeted PFAS were 6:2 FTS (139 mg L -1 ), 8:2 FTS (7.85 mg L -1 ), and PFOA (3.54 mg L -1 ). However, F elements from all targeted PFAS only accounted for 2% of the total fluorine. F nuclear magnetic resonance (NMR) analysis found the dominant species in AFFF as n=6 FT surfactants (i.e., C6F13-(CH2)m-RO, Fig. ), but the structure of the organic moiety (RO) was unknown. We hypothesized that the surfactants could be defluorinated via similar mechanisms as for individual PFAS with the same RF building blocks (Fig. ). Hence, to effectively monitor the treatment process, we kept tracking the concentrations of FTSs, PFSAs, PFCAs, select surfactant molecules, and F -ion (Fig. ). |
64eb2db63fdae147faf1e1d3 | 3 | Under UV/S treatment of the 100-fold diluted AFFF (total fluorine at 100 mg L -1 ), the concentration of 6:2 FTS increased in the first 8 hours and then slowly decreased (Fig. ). PFSAs such as the C6 PFHxS, although in low concentrations, showed a similar generation-degradation profile (Fig. ). It exhibited higher recalcitrance than that observed in previous studies using pure PFHxS in the deionized water matrix. The slow apparent degradation of these species can be attributed to (1) competing species in the organic matrix of diluted AFFF and (2) the continuous generation of PFHxS from n=6 sulfonamide surfactant precursors. This reasoning is further supported by the rather consistent concentration of PFOS, which has higher reactivity than PFHxS in previous UV/S studies. The sustained PFOS throughout the 24 h is most probably attributed to the conversion of n=8 sulfonamide precursors. PFCAs also showed generation-degradation patterns under UV/S treatment (Fig. ). Because the initial concentrations of all PFCAs were negligible, the generated PFCAs could be attributed to the conversion of fluorotelomeric and sulfonamide precursors. A series of n=4-7 surfactant molecules (detected by Q-ToF-HRMS following literature ) demonstrated high recalcitrance or even net increase (Fig. ). The UV/S module resulted in 40% of overall defluorination after 24 h (Fig. ). Extended reaction beyond 24 h did not further increase defluorination (Fig. ). |
64eb2db63fdae147faf1e1d3 | 4 | After switching to EO mode, all surfactant molecules degraded to non-detected after 40 h (i.e., 16 h under EO, Fig. ). In comparison, most targeted PFAS structures showed concentration increases sooner or later (Figs. ), and eventually became non-detected after 44 h (i.e., 20 h under EO). In particular, elevated PFCAs showed the generation-degradation profiles in a wide time window (Fig. versus Fig. ), indicating the oxidative transformation of the abundant FT surfactants. The early generation of n=5 PFHxA, n=4 PFPeA, and n=3 PFBA in high concentrations suggest the oxidative conversion of the dominant n=6 FT precursors, as revealed by F NMR (Fig. ). The oxidation of pure n=6 FTS using HO• radicals yielded similar PFCA product distributions (i.e., "n-2 dominance" rule). The second wave of PFCA generation started after 32 h, with the most significant increase for n=6 PFHpA, followed by n=5 PFHxA and n=7 PFOA, suggesting a slower oxidative conversion of n=8 FT precursors. The increase of n=6 and 8 FTSs during EO treatment (Fig. ) suggested the oxidation of organic moieties. The very short time window for PFSAs (Fig. ) further confirmed that sulfonamide precursors were minor components in the studied AFFF, and all degraded within a few hours. After the EO treatment, all targeted PFAS were below the detection limits shown in Table . The F -ion release reached ~100% of overall defluorination (Fig. ). F NMR analysis of the residual also found no other F resonance beside F -(Fig. ), which is another evidence for the near-quantitative defluorination. |
64eb2db63fdae147faf1e1d3 | 5 | Engineering considerations for AFFF treatment by UV/S-EO. TOC removal. Besides the 10 g L -1 of organic fluorine, AFFF contained heavy amounts of hydrocarbon surfactants. Total organic carbon (TOC) analysis of the 100-fold diluted AFFF found 396 mg L -1 of organic carbon (Fig. ). However, after UV/S treatment, the measured TOC increased to 2005 mg L -1 . Notably, the combustion temperature of the TOC analyzer by default setting (680 °C) cannot thoroughly oxidize all carbons, especially the fluorinated carbons, into CO2. Hence, UV/S treatment converted the "combustion-proof" mixed surfactants into more thermally oxidizable structures. After EO treatment, TOC was drastically reduced to only 13 mg L -1 . Assuming the value of 2005 mg L -1 was similar to or still lower than the actual TOC of the 100-fold diluted AFFF, the TOC removal by EO was ≥99.4%. Because fluorinated carbon that accommodates 100 mg L -1 of organic F as CF2 and CF3 was only a small portion of TOC, we concluded that EO treatment allows very deep mineralization of most hydrocarbon surfactants. Therefore, if organic removal is needed for AFFF treatment at ambient conditions, EO is a highly competitive technology option. |
64eb2db63fdae147faf1e1d3 | 6 | Foam suppression. Although EO provides a strong capability of mineralizing both organic and fluorinated carbons in AFFF, direct EO treatment encountered a serious foaming issue due to the vigorous gas evolution from water-splitting reactions (Fig. ). To quantitatively describe the foaming, we arbitrarily define the "foaming potential" as the ratio between the height of foam and the depth of liquid under air purging at 100 mL min -1 . Before treatment, the 100-fold diluted AFFF had a foaming potential of 6 (Fig. ). After UV/S treatment, the value decreased to 1.4 (Fig. ), allowing an easy operation of EO treatment. We only observed a thin foam layer with a height of less than 8% of the liquid in the first 4 h and no foaming thereafter. As expected from the ≥99.4% |
64eb2db63fdae147faf1e1d3 | 7 | An imminent application scenario is the cleaning of hanger fire-fighting pipelines and fire trucks that used PFAS-based AFFF in the past decades. This time, we used tap water for the 100-fold dilution of AFFF (Table ). The UV/S-EO treatment resulted in very similar evolution/degradation kinetics for all individual PFAS and F - release (Fig. ) to the DI water diluted AFFF (Fig. ). We also observed very similar reaction kinetics for all species at the dilution factors of 50 (Fig. ) and 500 (Fig. ), except that the more diluted (i.e., less concentrated) AFFF needed less time to achieve 100% defluorination. For the UV/S module, the chemical and energy consumption appeared proportional to the dilution factor (Fig. ). The treatment of 500-fold diluted AFFF needed 10 mM sulfite and 12 h to reach the maximum defluorination of 46%. For the 50-fold diluted AFFF, 100 mM sulfite and 120 h were needed to reach the maximum defluorination of 48%. In comparison, the EO module is less sensitive to the dilution factor. The time required to achieve the 100% overall defluorination for 50-and 500-fold diluted AFFF was 24 and 12 h, respectively. It is important to highlight that the 50-fold diluted AFFF had a record-high TOC > 4000 mg L -1 and TOF at 200 mg L -1 compared with those samples treated in the previous studies (Table ). Hence, the UV/S-EO has demonstrated great promise to destroy concentrated PFAS in wastewater, particularly for the major challenges in fire-fighting system cleaning and AFFF disposal (after adequate dilution). |
64eb2db63fdae147faf1e1d3 | 8 | Energy consumptions. We calculated the energy efficiency of UV/S and EO modules based on the slopes of the quasi-linear segments of the defluorination profiles (Figs.4d, S7d, S8d, and S9d) as the required energy input (kWh) to convert per gram of the organic fluorine to F -(Figs. and). The light-adsorbing water matrices are usually expected to limit the efficacy of photochemical systems, but the UV/S system exhibited a consistent energy efficiency for the 50-, 100-, and 500-fold diluted AFFF. In particular, the UV/S treatment further reduced the absorbance at 254 nm in the 50-fold diluted AFFF from 1.36 to 0.36 (Table ). This "self-sharpening" feature makes UV/S suitable for treating concentrated AFFF. The lowest dilution factor of 1:50 in this work is three orders of magnitude lower (i.e., three orders of magnitude more concentrated) than the previous UV/S demonstration, which diluted AFFF 60,000-fold and operated at pH 9.5. The limited dilution substantially reduced the water volume to be treated, thus substantially saving the electrical energy for UV irradiation. |
64eb2db63fdae147faf1e1d3 | 9 | For EO treatment, the energy consumption decreased with the lower dilution factor. This observation aligns with the principle of heterogeneous catalysis: the higher bulk concentration creates a steeper concentration gradient at the water/electrode interface, thus enhancing the mass transfer of PFAS to the BDD surface and the subsequent oxidation by direct electron transfer. |
64eb2db63fdae147faf1e1d3 | 10 | The process design was built on the state-of-the-art understanding of the complementary capabilities of the two modules: 1) UV/S is highly effective for defluorinating long-chain PFAS that EO could not defluorinate to 100%; 2) EO is highly effective in mineralizing short-chain PFAS and H-rich TPs from UV/S treatment, both of which are recalcitrant under UV/S; and 3) UV/S treatment effectively suppressed foaming that could cause operational issues for EO. Moreover, both UV/S and EO exhibited high energy efficiency in treating AFFF with limited dilution. EO also enabled the near-complete removal of TOC in AFFF. All reactor components are commercially viable at full-scale. The integration only requires conveying the treated effluents without retrofitting the reaction units. We expect this treatment strategy to be also effective toward novel PFAS structures in various practical scenarios under ambient conditions. Lastly, we emphasize that UV/S-EO was developed for the non-potable treatment of obsolete AFFF stockpiles and fire-fighting system cleaning solutions. Therefore, the concern about the disinfection byproducts, which are only regulated in the drinking water supply, should not constrain the improvement and deployment of the process. Besides, technologies for removing halogenated byproducts and oxyanions are widely available and can be adopted as post-treatment add-ons. We are developing various engineering processes with pre-and post-treatment that can further expand the application scope of UV/S-EO in even more challenging water matrices. |
64eb2db63fdae147faf1e1d3 | 11 | Analysis. Targeted analysis of PFAS was conducted on ultra-high-performance liquid chromatography (UPLC, Thermo Vanquish) coupled to a triple quadrupole mass spectrometer (QQQ MS/MS, Thermo Altis). The analytical method includes 30 PFAS. Details of instrument setup were described in our previous publication. Nontargeted analysis of PFAS transformation products was performed on high-performance liquid chromatography-quadrupole time-of-flight mass spectrometry (HPLC/Q-ToF-MS, SCIEX). The instrument setup was described in Text S2. |
64eb2db63fdae147faf1e1d3 | 12 | The search and match of unknown fluorocarbon structures follow the protocol developed previously. The F -was quantified by ion chromatography (IC). The TF of AFFF was analyzed by combustion ion chromatography (CIC; Metrohm), with the principle of decomposing AFFF samples at 1050 °C and using an IC to measure the F -released. Details were described previously. |
64eb2db63fdae147faf1e1d3 | 13 | The pH was adjusted to 12 by 1 M NaOH to achieve the highest photo-reductive treatment efficiency. As for the UV/S treatment of AFFF, AFFF samples diluted by DI water or tap water at ratios of 1 to 50, 1 to 100, and 1 to 500 were amended with Na2SO3 at 100, 100, and 10 mM, respectively. The reactor was sealed from the air without inert gas protection in all tests. In order to establish the proof-of-concept UV/S-EO tandem treatment train, we adopted a BDD flow cell for a larger treatment capability. The BDD flow cell reactor provided by Element Six contains two BDD disks (∅ 4.4 cm each at an interspace of 0.8 cm) that serve as anode and cathodes (Fig. ). The flow cell has a chamber volume of 95 mL. In the tandem treatment process, 750 mL of diluted AFFF will be first subjected to UV/S reductive treatment; the 750 mL treated water will then be circulated through the flow cell at a flow rate of 100 mL/min. It is important to note that the batch EO tests using plate-type BDD have a current-to-volume ratio of |
63581f0bac45c7dd1c97dd17 | 0 | Many useful properties of materials manifest from the precise structure of a given composition. Traditional structure determination techniques such as X-ray diffraction (XRD) is suitable for atoms of moderate-to-high atomic number; however, it can lead to ambiguous structure assignments for materials containing atoms with low atomic number. In addition, XRD relies heavily on long-range order for correct measurement, but such long-range order is often lacking in many classes of materials, e.g. nanostructures, amorphous materials and materials with a tetrahedral network, making structural characterization of such materials via XRD difficult. Such as a class of materials are exemplified by silicates which consist of a tetrahedral structure of silicons and oxygens (which have low atomic number and are difficult to observe via XRD). |
63581f0bac45c7dd1c97dd17 | 1 | Nuclear magnetic resonance (NMR) spectroscopy has become a reliable tool for structural investigations in such materials. As a spectroscopy technique, NMR is highly sensitive to the electron density about an atom and relies on local structure rather than any longrange order. NMR measurements are typically combined with powder XRD measurements and ab initio simulations to obtain refined crystal structures in a technique termed NMR Crystallography. These refinement procedures, however, often take an expensive iterative approach as many ab initio NMR calculations are repeated until the results converge. Despite advances in computational power and algorithmic efficiencies, the NMR calculation is still expensive and time consuming. Machine learning (ML) has increasingly been shown to be useful for providing high quality predictions for material properties but with orders of magnitude less computational demand. ML techniques have recently been applied to NMR, with most applications focused on organic molecules and a few focused on Si chemical shift prediction. While chemical shifts are useful as they correlate to the average electronic environment about an atom, shifts are only one piece of the spectrum. The line-shape observed in an NMR measurement is described by a tensor, of which the chemical shift is the isotropic part of the tensor. By ignoring the tensorial nature of the NMR measurement, a myriad of structural information is lost, which is an issue for many previous ML models as they are only capable of predicting scalar quantities. |
63581f0bac45c7dd1c97dd17 | 2 | In the first part of this work we will assess the use of more traditional ML methods to attempt to learn NMR tensor parameters to demonstrate that a symmetry-invariant model is insufficient for NMR tensor parameter prediction. We show that all models trained herein fall short of predicting tensor elements, and instead a symmetry-equivariant model is needed to respect the symmetries of the tensor, as evidenced by a 47% (3.05 ppm vs. 6.44 ppm) decrease in mean absolute error when compared to invariant models. In the second part of this work we comprehensively assess the performance of our equivariant model. We show that it can learn the full NMR chemical shift tensor to a mean absolute error (MAE) of 1.05 ppm. When converted to the scalar isotropic chemical shift (which previous symmetryinvariant models can predict), our MatTEN model outperforms state-of-the-art model by a large margin, with MAEs of 2.82 ppm vs. 5.87 ppm. |
63581f0bac45c7dd1c97dd17 | 3 | We also assess the predicted tensors to show that the MatTEN model is capable of learning both tensor magnitude and shape as well as the tensor orientation in a diverse set of silicon local structures. The Si NMR dataset used in this study is a subset of ab initio NMR chemical shift tensors of relaxed structures calculated by Sun et al. The dataset is composed of oxygencoordinated silicon tetrahedral networks consisting of SiO 2 along with silicates containing group 1 and 2 cations (Li, Na, Mg, etc. cations). It contains a wide variety of structures with different number of bridging oxygen atoms, n, commonly referred to as Q n , as shown in Fig. . Each Q n species has a different chemical environment and local point group symmetry due to the differing bond lengths to bridging oxygen (BO) and non-bridging oxygen (NBO), which in effect results in different chemical shift tensor symmetries. In total there are 421 unique silicate structures, consisting of 1387 unique silicon sites. The silicon sites consist of 874 Q 4 sites, 174 Q 3 sites, 172 Q 2 sites, 32 Q 1 sites, and 97 Q 0 sites. From each site, the raw calculated rank-two asymmetric chemical shift tensor was extracted and processed in accordance with each tensor space or tensor convention used in the training of the ML models outlined below. |
63581f0bac45c7dd1c97dd17 | 4 | For each ML model, a train, validation, and test split of 8:1:1 was used. As the dataset consists of multiple types of sites (i.e. each Q n species), an attempt was made to stratify the data such that there is an approximately equal weighting of each Q n species in each of the training, validation, and test sets. To remove any opportunity for data leakage, stratification was done at the structure-level rather than the site-level. Due to structures containing multiple sites, often of different Q n 's, each structure was given a label based on whichever n was least common in the dataset. For example, in a structure with both a Q 1 and Q 2 site the structure would be labeled as Q 1 because the dataset consists of smaller number of Q 1 sites (32) than Q 2 sites (172). Structures were then randomly stratified to give roughly equal proportions of each n type in each set. |
63581f0bac45c7dd1c97dd17 | 5 | Often in the context of NMR, the quantity of interest is a scalar value, the isotropic chemical shift, δ iso , but it is important to keep in mind, that the chemical shift is a tensor quantity, formally an antisymmetric second rank tensor. Typically, only the symmetric part of the tensor is used as the symmetric tensor influences the lineshapes seen in the NMR spectrum. |
63581f0bac45c7dd1c97dd17 | 6 | It should also be noted the distinction between nuclear shielding and chemical shift. The nuclear shielding describes the relative change in magnetic field about a nuclear position with respect to the external field, and is the quantity calculated during an ab initio calculation. In NMR experiments, however, the shielding is not measured directly, and instead the common practice is to measure the chemical shift as the difference in resonant frequencies between the nucleus of interest and a reference compound. |
63581f0bac45c7dd1c97dd17 | 7 | While nuclear shielding tensors are the typical quantities calculated by ab initio methods, we elect to instead use the absolute chemical shift tensor, which is the quantity calculated by VASP and the form originally reported for the data-base of which the data used herein is obtained. The formal relationship between the absolute chemical shift tensor, δ, and the nuclear shielding tensor, σ is given by: 42 |
63581f0bac45c7dd1c97dd17 | 8 | For more information on NMR conventions, the reader is directed to the numerous reviews and textbooks on the topics. In symmetry-invariant ML models, an assortment of tensor conventions are used as the training targets. The targets will, in each case, consist of three parameters: the isotropic chemical shift, and two additional parameters used to describe the shape of the tensor. |
63581f0bac45c7dd1c97dd17 | 9 | Of the conventions used, the Maryland (Ωκ) and Haeberlen (ζη) are the most commonly found conventions as both are recommended for reporting by IUPAC (Note that Haeberlen is occasionally presented as (∆δη), however the (ζη) definition is far more common). Two less practiced conventions are used as well. The Axiality/Rhombicity (AxRh) convention is sometimes used in spin dynamics and spin relaxation theory as the parameters come from the irreducible spherical tensor expansion of an interaction Hamiltonian. In addition to the four conventions listed in Table , we also investigate learning directly on the principal axes (i.e. δ 11 , δ 22 , and δ 33 ) in the standard convention to give a total of five conventions investigated for the invariant property predictions. |
63581f0bac45c7dd1c97dd17 | 10 | A symmetric second rank tensor has six independent parameters which will need to each be assessed. The spherical tensor elements are ideal but are difficult to interpret, as are the tensor indicies themselves. The IUPAC recommended conventions may not be optimal either. The Maryland convention is a descriptor of the lineshape as a statistical distribution and is only applicable in specific cases and lacks generalization. The Haeberlen convention is based on the tensor itself and is generally true to the chemical shift tensor; however, the convention requires the ζ parameter to be sign invariant at η = 1, which creates a degeneracy. |
63581f0bac45c7dd1c97dd17 | 11 | In chemical graph neural networks (GNNs), crystalline structures can be represented as graphs in which each atom is represented by a node, v, and relationships between nodes are represented by an edge, e, which are commonly thought of as chemical bonds, coulombic interactions, etc. Often the notion of a chemical bond is ill-defined in a crystalline structure, so edges are frequently constructed as all pair-wise node-connections within some cutoff radius, r cut , about each atom, taking into account the periodic boundary conditions of the system. All together, the set of nodes, V = {v 1 , v 2 , ..., v N }, and the set of edges, E = {e 1 , e 2 , ..., e M }, make up a graph, G(V, E). To make the graph G amenable to machine learning, each node is assigned a feature vector (information of the atomic number and the Cartesian coordinates of the atom in this work). The node data may also be processed to create edge features, e ij , which encode positional information between the two nodes, i and j. |
63581f0bac45c7dd1c97dd17 | 12 | The GNNs used in the present work follow the message passing neural networks (MPNN) paradigm in which node features are updated from neighboring nodes in a message passing phase, and then the updated features on a node is mapped to a property of interest in a readout phase. The objective of the message passing phase is to learn an embedding for each node, h, such that unique structural fingerprints for the node is encoded. The message passing typically occurs over a certain number of iterations. During iteration t, pairwise interactions between atom i and neighboring atoms j are summed and processed to produce a message |
63581f0bac45c7dd1c97dd17 | 13 | where N i is the neighborhood of all atoms surrounding atom i within a distance cutoff r cut , M t is a learnable function that takes as input the embeddings h t i and h t j of atoms i and j, as well as their edge data, e ij . The embedding of atom i is then updated using the message from Eq. ( ), |
63581f0bac45c7dd1c97dd17 | 14 | We explored rotation-invariant GNNs as example ML models designed for scalar properties. Specifically, the DGL (deep graph library) implementation of the DimeNet++ invariant GNN is selected (we note that DimeNet++ is a strong baseline model in terms of accuracy). The model is customized to allow for the prediction of node properties. A hyperparameter grid sweep was performed to optimize the DimeNet++ model to predict the shift tensor eigenvalues. A total of six separate models are created, each trained on a different shift tensor convention (Eq. (1) -Eq. ( )), and the standard convention eigenvalues) using the three parameters of the convention as the target. See the Supplemental Information for further implementation details and the optimal hyperparameters. |
63581f0bac45c7dd1c97dd17 | 15 | The DimeNet++ model itself is limited to a rotation-invariant mapping from the input structure to the target chemical shift tensor parameters. While the basis functions used in Eq. ( ) and Eq. ( ) themselves are rotation-equivariant, the message passing framework is rotation-invariant. Specifically, the coordinate information on each node is only used to initialize distances and angles which are invariant geometric properties. As a result, DimeNet++ (and similar frameworks) are limited to predictions of scalar targets, and in our case multiple, uncorrelated scalar values. The target shift tensor parameters, henceforth refereed to as scalar NMR parameters, fall under the assumption that the tensor parameters are independent of crystal orientation-as the powder pattern may be used to obtain the scalar NMR parameters-and therefore the parameters are invariant to rotations. |
63581f0bac45c7dd1c97dd17 | 16 | A rotation-equivariant model was created using the Tensor Field Network (TFN) and e3nn 56 frameworks, as implemented in the MatTEN 57 package. In addition to the scalar NMR shift tensor parameters discussed above, this model can directly predict the full chemical shift tensors. An initial MatTEN model was implemented (details in Supplemental Information) to determine the optimal target to train on (symmetric vs. asymmetric and spherical vs. Cartesian tensors). A symmetric spherical tensor target was found to yield the best loss and a hyperparameter grid search was performed to optimize the MatTEN model for symmetric spherical tensor. To yield a useful model, the symmetric spherical tensor is then converted to a Cartesian tensor which may be processed as a shift tensor. |
63581f0bac45c7dd1c97dd17 | 17 | An additional rotation-invariant MatTEN model was similarly created, however, the model was trained on the shift tensor eigenvalues. Internally, this model still does message passing using equivariant embeddings (explained in the following paragraph), however, in this case, the target is set to the (scalar) eigenvalues instead of the full chemical shift tensor. |
63581f0bac45c7dd1c97dd17 | 18 | where R(r) is a learnable function of the distance between the two nodes and Y m l (r) are spherical harmonics taking in the orientation between the nodes. The W matrix has the form of a block diagonal matrix where the blocks correspond to the irreps selected for the network. Additionally, the embedding vectors used in the TFN framework contain blocks corresponding to the irreps. The W matrix along with the embedding vectors may then be convolved according to Clebsh-Grodan tensor products to ensure the symmetries of each irrep is preserved. So, the message passing phase of the TFN model uses equivariant embeddings. |
63581f0bac45c7dd1c97dd17 | 19 | To the best of our knowledge, no previous model has been proposed to predict full shift tensors, thus benchmarking will take place in two steps. The current state-of-the-art model for Si scalar NMR parameter prediction was introduced by Chaker et al., who use linear ridge regression (LRR) over the Smooth Overlap of Atomic Positions (SOAP) features to predict the Si isotropic chemical shift. We reimplemented this approach using the SOAP features generated by DScribe 59 and the LRR in scikit-learn. During the invariant-target benchmark an LRR-SOAP model will be trained to predict the three eigenvalues of the chemical shift tensor and all models will be compared on their predictions of the eigenvalues. |
63581f0bac45c7dd1c97dd17 | 20 | The models are categorized according to the symmetries of its internal embeddings and its final target, labeled as "Embedding Symmetry" and "Target Symmetry", respectively, in Table . Of the invariant target models, DimeNet++ performs the best, and LRR-SOAP and MatTEN are on par with each other. However, if a fully equivariant MatTEN model (equivarant embedding and equivariant target) is used to predict the full shift tensor and then diagonalized to yield the eignevalues, significant improvement over the invariant target models can be achieved. For example, the total MAE reduces to 3.05 ppm, which is less than half of that from the DimeNet++ model (6.44 ppm). Comparing the MatTEN models trained using invariant versus equivariant target symmetry, it is clear that the boost in performance is due to the additional constraint afforded by learning a second rank tensor rather than three independent scalars. The fully equivariant model will be further discussed in the next section, and here we focus on the invariant models. |
63581f0bac45c7dd1c97dd17 | 21 | The eigenvalues of the shift tensor are not the only invariant targets to consider; one can train ML models to directly predict the NMR parameters in different conventions outlined in Table . For this purpose, we have selected the DimeNet++ model based on its superior performance on the eigenvalue prediction. Five DimeNet++ models are trained (one for each convention), and, for the ease of comparison, their predictions are converted to the eigenvalues in the standard convention and the two IUPAC recommended conventions (i.e. |
63581f0bac45c7dd1c97dd17 | 22 | the Haeberlen (ζη) convention and the Maryland (Ωκ) convention) using the equations in Table . The results are listed in Table . We observe that there is no single optimal model, only a 'best-in-class' per NMR parameter. For example, the Axiality/Rhombicity convention has the best overall performance but still under performs on isotropic shift and Maryland-Ω values. Furthermore, tensor conventions should be inter-convertible. While this is typically true for experimental spectra, we find that it is not the case for ML models trained on individual NMR parameters. For example, the model trained in the Maryland convention performs well when predicting Maryland convention values, but when converted to Haeberlen (ζη), the model ranks the lowest. |
63581f0bac45c7dd1c97dd17 | 23 | provide inferior results as compared to a fully equivariant model. As noted earlier, some conventions are ill-defined and discontinuous for certain values. Additionally, all tensor conventions based on the Cartesian tensor have an issue of explicitly defined axes which can cause confusion when parameters are predicted outside their range, which result in a change of the order of the eigenvalues. A full discussion of the numerical issues that arise when fitting with the tensor conventions can be found in the supplemental information. |
63581f0bac45c7dd1c97dd17 | 24 | We now turn our attention towards the rotation equivariant MatTEN model, which was shown to significantly outperform the invariant models (see Table ). Similar to the case of invariant models, there are a variety of output targets available for the equivariant model, of which there is not an a priori optimal choice. We focus on using an asymmetric Cartesian tensor, symmetric Cartesian tensor, asymmetric irreps (E, A ik , S ik ), or symmetric irreps (E, S ik ) as the target (refer to Eq. ( )). Additionally, the question of which loss function is suitable for learning chemical shift tensors has not, to our knowledge, been investigated for training ML models. There has, however, been substantial work by the MRI diffusion tensor community on optimal tensor metrics for diffusion tensors. The l n -norms offer a good balance of optimizing the shape, magnitude, and orientation of a tensor, and in our case we adopted the l 1 norm as the loss function. We found there is a small benefit to learning on a symmetric tensor versus an asymmetric tensor before symmetrizing. Additionally, there is a minor decrease in both epoch time and the loss when training on an irreps tensor versus a Cartesian tensor. Thus, the optimal space was chosen to be symmetric irreps tensor using an l 1 -norm loss function, and all subsequent results are obtained from models trained using this optimal space. The best performing equivariant MatTEN model exhibits an MAE of 1.05 ppm over the entire chemical shift Cartesian tensor. However, we admit that a MAE calculated for all the tensor components is challenging to interpret as asymmetric second rank tensor has six independent parameters which must be assessed in order to evaluate performance without loss of information. Therefore, we compare the predicted and DFT-calculated isotropic chemical shift, X, Y, and Euler angles (α, β, γ) (refer to Section for their definition). These parame-ters are chosen as they provide an intuitive view of the magnitude, shape, and orientation of the tensor, as described above. Additionally, because the shift tensor is very closely linked to the structural point group, 71 the results are grouped by Q n into three clusters reflecting the broad symmetry point group: T d , C 3 and C 2 . The results are summarized in Table . as shown in Fig. . The isotropic shift, however, performs considerably worse than the tetrahedral case as is clearly seen in Table (3.28 ppm and 8.58 ppm in Q 3 and Q 1 respectively vs. 1.52 ppm and 1.46 ppm in Q 4 and Q 0 respectively). Barring a cluster of 3 data points, the Q 3 isotropic shift tends to be over-predicted by MatTEN. The sites with over-predicted isotropic shift also correspond to the sites with poorly predicted Euler angles in Fig. . |
63581f0bac45c7dd1c97dd17 | 25 | We speculate that the relatively poor performance of the model is due to the lack of Q 3 sites in the training data combined with the significant increase in structural diversity compared to the Q 4 and Q 0 sites. Furthermore, the structures with anomalous Q 3 sites trends as outliers in the training set as well and are not well sampled, which results in poor learning of their structural correlations. In some cases the formula of the material was not seen in the training set nor were there any similar formulas seen which resulted in a poor extrapolation by the MatTEN model. In other cases, the same formula was seen but with minor structural variation, which led to an unfortunate case of MatTEN memorizing the solution poorly. |
63581f0bac45c7dd1c97dd17 | 26 | The Q 1 case is, however, fortunate in that despite the poor isotropic shift prediction, the remaining 5 parameters all show good correlation. Overall, the MatTEN model is able to learn to predict the tensor of silicates with the best performance being on sites with high symmetry, as summarized in Table . Even in cases where the model struggles with isotropic shift or tensor orientation, the shape of the tensor is well predicted. |
63581f0bac45c7dd1c97dd17 | 27 | It is also instructive to benchmark our MatTEN model to historic models and previous state-of-the-art models to ensure that our model constitutes an advance in the field, especially for the domains where previous models were successful. For should be noted that one drawback of the SOAP descriptor is that its size scales with number of species in the dataset, N , as N (N -1), thus for our entire dataset the SOAP descriptor encoding has a size of 10980, and while this is not an issue for LRR, the dimensionality may be an issue for other methods especially as the descriptor size is far greater than the dataset size. It should also be noted that the LRR model can only handle scalar values, whereas the main benefit of the MatTEN model is in providing the full shift tensor. |
63581f0bac45c7dd1c97dd17 | 28 | Machine learning approaches are increasingly employed to predict a variety of physical properties, accelerating and expanding access to material data. However, many of those physical properties adhere to inherent constraints, such as symmetry relationships or limits. In the case of tensorial properties, each eigenvalue may be predicted as an independent scalar, however, such treatment effectively ignores the underlying symmetry information of the tensor. |
63581f0bac45c7dd1c97dd17 | 29 | We find that the NMR tensor parameters cannot be easily learned via symmetry invariant processes and often contain algebraic structure that make the learning process more difficult, independent of the tensor convention. By imposing symmetry equivariance, our MatTEN model is able to outperform by 47% the symmetry invariant models, demonstrating that handling the tensorial nature of the target is the key to accurately modeling the system. |
63581f0bac45c7dd1c97dd17 | 30 | Examining the results of the equivariant MatTEN model, we observe that the model is able to accurately predict the tensor, not just in terms of shape and magnitude, but in most cases in the orientation as well. Closer inspection of the cases where the orientation seems to fail shows that these are often the cases of highly spherically symmetric tensors where an orientation is not meaningful. Most surprisingly, the model is able to capture the shape (anisotropy and asymmetry) of the tensor, even in cases where the tensor exhibits very little anisotropy, for example for Q 4 and Q 0 sites. Despite the successes of the model, there are still cases where it fails; however, these failures are likely associated with a lack of data in the training set. Future work will be focused towards expanding the dataset, particularly for the Q n species, which are less well sampled. |
63581f0bac45c7dd1c97dd17 | 31 | Notably, through the demonstrated work on silicates, it is feasible to predict the full NMR tensor, with reasonable accuracy, in seconds rather than hours to days that were required for ab initio calculations. This opens the realm of possibilities, from high throughput screening of materials via comparison to experimental NMR spectra, to expediating NMR crystallography refinement procedures. |
677e020b81d2151a0245838e | 0 | Materials exhibiting actuation properties in response to external stimuli have been extensively studied owing to their wide range of applications, including soft robotics, the medical industry (artificial muscles and limbs), and energy storage and conversion. Sophisticated material design has led to the development of soft polymer composites exhibiting autonomous locomotion similar to that of living organisms. Liquid crystal elastomers exhibit continuous rolling and oscillation when exposed to consistent light and temperature. In these examples, mechanical energy is extracted through a nearly constant external field, eliminating the need for the on-off cycling of the light or heat source. Polymer composites are materials that function as external field-response actuators owing to their high flexibility, which enables facile deformation. Recent developments in crystal engineering have revealed the possibility of endowing organic crystals, which were previously considered rigid, with actuation properties. Elastic crystals that can be reversibly bent like elastomers while maintaining their crystallinity have also been reported; these materials are considered promising materials for highly durable and flexible optoelectronic and magnetic devices. Reports on elastic crystals exhibiting actuation in response to light, chemicals, and temperature underscore the diversity of the operating principles of actuators prepared from molecular crystals. However, the extraction of energy from a constant external field, as demonstrated in polymer composites, is rarely observed in organic crystals and remains a challenging endeavour. Most reported crystal-based actuators require fluctuating external fields, such as the on-off cycling of a light source or heater, for continuous motion. The conversion of heat, the lowest grade of energy, into kinetic energy is notably more challenging than the conversion of light owing to the lack of established design criteria for thermal conversion. |
677e020b81d2151a0245838e | 1 | To address the gaps in the literature, in this study, we present the first example of a thermal engine based on elastic crystals. When the crystal is loaded with a weight and placed between high (34-37 °C)-and low (~0 °C)-temperature heat sources, it continuously oscillates, indicating that it can extract kinetic energy from the heat sources. The oscillations continued for at least 160 h, corresponding to over 3.9 million cycles of bending and stretching, demonstrating the high durability of the elastic crystal. To the best of our knowledge, this study is the first to report the continuous operation of an organic crystal-based thermal engine under only static temperature differences. Our results highlight the potential applications of organic elastic crystals as actuators owing to their high-speed responsiveness to external fields and remarkable durability. |
677e020b81d2151a0245838e | 2 | The dodecylated porphyrin molecule was synthesised by the ester-exchange reaction of 10,15,20tetrakis(4-methoxycarbonylphenyl) porphyrin H2(TMCPP) in 1-dodecanol in the presence of a base catalyst (DBU), as depicted in Figure . Crystallisation of the deacylated porphyrin using dichloromethane and ethanol at 40 °C afforded the needle-like elastic crystals (1), with a minor amount of brittle block-shaped crystals (2). 1 could be bent under a mechanical force and returned to its original shape when the force is released, indicating its elasticity (Figure and Movie 1). |
677e020b81d2151a0245838e | 3 | The nanoindentation test results for 1 (Figures and) revealed a Young's modulus (Er) and hardness (H) of 1.8(2) and 0.072(3) GPa, respectively. Furthermore, the Young's modulus of 1 upon bending was determined to be 0.13 GPa (Figures ). These values are lower than those of most organic compounds , confirming the high flexibility of 1. Figure summarises the crystal packing of 1 at 298 K. 1 crystallised in the space group P-1, and the entire molecule was crystallographically independent. The two dodecyl chains of 1 exhibited heavy disorder at 298 K, as indicated in green, orange, and purple. The porphyrin molecules stacked along the a-axis to form a columnar structure surrounded by dodecyl chains. The a-axis coincided with the direction of crystal growth. Needle-like crystals with such a one-dimensional molecular arrangement are commonly elastic. The intracolumnar distances between porphyrin π-planes (3.67 and 3.82 Å) are longer than that expected for a π-π stacked system (3.3 Å), indicating that the van der Waals interactions between dodecyl chains are the dominant driving force behind the one-dimensional packing of the crystals (Figure ). Figure summarises the temperature dependence of the a-axis length as a representative parameter. When the temperature was swept from approximately 275 to 223 K, the a-axis length gradually increased with the gradual suppression of the disorder on the dodecyl chains (Figure ). The crystalline parameters exhibited a similar temperature dependence during the corresponding heating process, indicating that this process is reversible. Adiabatic and differential scanning calorimetry indicated that the gradual structural changes observed between 275 and 223 K correspond to a second-or higher-order phase transition from the high-temperature (HT) to the low-temperature (LT) phase (see the thermal analysis section in the Supporting Information (SI)). |
677e020b81d2151a0245838e | 4 | In this paper, the 275-223 K region is referred to as the medium-temperature (MT) state for convenience. The a-axis length of 1 increases with decreasing temperature, whereas its b-axis length, c-axis length, and cell volume decrease. In other words, the needle-like crystals become long and slim upon cooling. The elongation ratio of the crystals associated with the decrease in temperature was observed using optical microscopy (Figure and Table ). The increase in a-axis length correlated with the magnitude of disorder of the dodecyl groups. The heavy thermal motion of the dodecyl chains in the HT phase prevented the porphyrin columns from approaching each other, whereas the suppressed thermal motion of the dodecyl chains upon cooling decreased the intercolumnar distances (Figure ). Owing to the chemical pressure from adjacent columns, the intracolumnar molecular distances increased, resulting in an increase in a-axis length. thermal motion, its flexibility should depend on the temperature. To briefly examine the temperature dependence of the flexibility of 1, we attached a weight (1.168 mg of Al) to the tip of 1 (length: 5.7 mm) and cooled the crystal in a refrigerant-free cooling device sealed with an insulated window. Figure summarises the deformation of the crystal with the weight load upon cooling and heating (Movie 2). Surprisingly, the crystal bent during cooling from 298 to 265 K, indicating that the crystal softens as the temperature decreases. This behaviour is unusual because typical solids harden with decreasing temperature. A further decrease in temperature from 265 to 251 K straightened the crystal owing to curing. The heating process induced a crystal deformation identical to that observed during the cooling process, indicating that the deformation of the crystal with changing temperature is reversible. |
677e020b81d2151a0245838e | 5 | When 1 is loaded with a weight (1.168 mg of Al) and sandwiched between a high-temperature (34-37 ℃) heat source and a low-temperature (~0 ℃) heat source, a thermal engine that can extract kinetic energy from thermal energy is constructed. The crystal is ~5.2 mm long and ~43 μm thick (Figure ). As shown in Figure , the weight attached to the crystal moves with a large oscillation at a frequency of 6.84 (7) Hz. When the Peltier cooler is switched on, the crystal begins to vibrate, and the vibrations are gradually amplified until a large oscillating motion is achieved (Movie 3). This movement continues as long as the temperatures of the heat sources are maintained, and was confirmed to last for at least 160 h, corresponding to over 3.9 million cycles of bending and stretching. Unlike a previous example of molecular crystals that could extract kinetic energy from heat, 10 our crystal does not require periodic changes in temperature for actuation. Thus, to the best of our knowledge, our molecular crystal-based heat engine is the first of its kind. Figures and summarise the time dependence of the x and y coordinates as well as the velocity (v) of the weight, clearly demonstrating the high reversibility of the lifting and descending cycles (Movie 4). The average time interval for the descending and lifting processes over 70 cycles was 0.146(1) s. The maximum v (vmax) of the weight was 174(9) mm s -1 for descending and 179( ) mm s -1 for lifting (Figure ); these values are 5-6 times faster than the reported v of a glass bead (0.15 mg) pushed by a thermoelastic organic crystal (29 mm s -1 ). Figure summarises the thermodynamic cycle of the heat engine. As the crystal approaches the low-temperature heat source (273 K; state A), it deforms from its equilibrium position by Δq = ~5 mm (state B and Figure ) owing to softening. Using an elastic constant k of 2.94 mN m -1 (as summarised in Figure ), we determined the elastic energy (EEl) to be k(Δq) 2 /2 = 37 nJ. k rapidly increased to 5.48 mN m -1 when 1 was heated by the high-temperature heat source (298 K; state C), and EEl increased to 69 nJ. In other words, 32 nJ of EEl was supplied from the thermal energy. Based on the v and y (height) of the weight, the time dependence of its kinetic energy (EK) and potential energy owing to gravity (EG) can be calculated as mv 2 /2 and mgy, respectively, where g is the gravitational acceleration and m is the mass of the weight. Figure summarises EEl, EG, EK, and their sum (ET). The maximum EEl (~69 nJ) is greater than the maximum EG (~45 nJ), indicating that the EEl stored via the temperature change is sufficient to lift the weight to its highest position (y = ~3.8 mm). In other words, the EEl supplied from the heat source is sufficiently large for the fast and continuous motion of the weight. Based on the thermal energy absorbed by 1 as the temperature is increased from 273 to 298 K (1.0 mJ, as summarised in the caption of Figure ) and the supplied EEl (32 nJ), the energy conversion efficiency of our system was estimated to be 0.003%, which is similar to that of a photothermally driven crystal actuator. This heatengine behaviour is highly reproducible, and large vibrations could be observed even in crystals of low quality (bent or chipped) (Figures S25-S27 and Movies 5-8). |
677e020b81d2151a0245838e | 6 | The thermal-engine behaviour of 1 is driven by changes in k with temperature. Notably, the temperature at which the crystal becomes most bendable (265 K in Figure ) is close to the boundary between the HT phase and MT state (~270 K in Figure ), indicating that softening is correlated with its structural changes. The following bending mechanism can be proposed based on the elongation of the crystal upon cooling (Figure ). ( ) The weight causes the elastic crystal to arc. The outer arc is stretched, whereas the inner arc is compressed. That is, the mechanical force elongates the a-axis in the outer arc and compresses the a-axis in the inner arc. (2) When the crystal is cooled to approximately 270 K, the outer-arc region preferentially changes to the MT state with a preference for a long a-axis, whereas the inner-arc region remains in the HT phase with a preference for a short a-axis. Such an inhomogeneous structural phase transition renders the outer arc longer and the inner arc shorter, resulting in the bending of the crystal, which appears as softening, upon cooling. (3) Further decreases in temperature convert the crystal into an entirely |
677e020b81d2151a0245838e | 7 | MT state, causing it to straighten (251 K in Figure ). The mechanism proposed here is identical to the operating principle of bimetals, and a similar type of curing by decreases in temperature has been reported for composite materials composed of elastic crystals, metals, and polymers. To confirm the inhomogeneous structural phase transition of the crystal arc near the phasetransition temperature, we performed pinpoint X-ray structural analyses of the crystal arc. Based on the crystal mount pin used to stretch crystals reported by Shi et al., we created a 3D model of a mount pin that could bend the crystal by adjusting a screw (Figure and 3D data in the SI). |
677e020b81d2151a0245838e | 8 | The crystal was adhered to a rectangular piece of Kapton tape to prevent it from moving and fluctuating under the N2 gas stream used for temperature control. The crystal had a straight configuration before the screw was tightened. The position of the microfocus X-ray beam (height: structure. The ω-scan range was restricted to -45° to 45° to avoid the X-ray irradiation of positions other than the centre of the crystal. As summarised in Figure , the a-axis length of the straight crystal was nearly independent of the beam position and similar to that in the MT state. |
677e020b81d2151a0245838e | 9 | Tightening of the screw caused the crystal to arc. Structural mapping from the lower to the upper regions across the centre of the arc revealed that the a-axis length of the outer arc was approximately 1.7% longer than that of the inner arc. The longer and shorter a-axis lengths correspond to the MT state and HT phase, respectively, indicating the outer-and inner-arc regions adopt the MT state and HT phase, respectively. The z dependence of the a-axis length of a bent crystal is not expressed as a linear line but is fitted with a sigmoidal curve, indicating an inhomogeneous structural phase transition. These results are consistent with the proposed mechanism of softening upon cooling, where the outer-and inner-arc regions adopt the MT state and HT phase, respectively (Figure ). Figure summarises the porphyrin core structures in the inner (small z)and outer (large z)-arc regions. The molecular structures of the outer-and inner-arc regions of the straight and bent configurations were identical, and no clear difference in the magnitudes of the thermal factors on the dodecyl chains was observed. In contrast to the molecular structures, the intracolumnar molecular distances depended on the position of the arc. In the straight configuration, the intermolecular centroid-centroid distances in the outer-and inner-arc regions were similar (Figure , left). However, in the bent configuration, the intermolecular distances changed from 5.87 to 6.01 Å in the inner arc and from 5.67 to 5.81 Å in the outer arc, corresponding to a ~2.3% difference in molecular distance (Figure , right). This deformation indicates that a 60 μm-thick crystal can be bent to form an arc with a diameter of 5.2 mm, indicating high flexibility. |
677e020b81d2151a0245838e | 10 | In conclusion, we successfully constructed a molecular crystal-based thermal engine that can extract kinetic energy from ambient-temperature differences. The flexible conversion of the kinetic energy of the weight to the potential energy of crystal deformation and the rapid deformation of the crystal to accelerate the weight are crucial factors for continuous motion. The bending of the weight-loaded crystal as the temperature decreases correlates with the straininduced phase transition associated with crystal elongation, which occurs preferentially in the outer-arc region. The mechanical properties and energy efficiency of our porphyrin-based system can be modified by changing the alkyl chain length, inserting metals, and constructing solid solutions. |
6762d904fa469535b9fc7a90 | 0 | Semi-artificial photosynthesis bridges the fields of materials science and chemical biology for the sustainable synthesis of solar fuels . It benefits from the evolution and bioengineering of enzymes, enabling complex chemical conversions with high selectivity and efficiency . Materials chemistry provides an avenue to rationally wire these enzymes to semiconducting light-harvesters, adopting a modular approach where individual components can be optimised separately for photoelectrochemical (PEC) applications . However, a challenge in the assembly of biohybrid PEC systems remains the frequent need for kinetically-fast buffers (to offset pH gradients) and diffusional mediators (to transfer charge from light absorber to biocatalyst) that prevent sustainable and stable semi-artificial photosynthesis. These components are not only expensive , but are also often noninnocent as they can be easily oxidised at low overpotentials by the photoanode , which precludes the coupling of fuel production with stoichiometric O2 evolution. Moreover, existing enzymatic biohybrid PEC devices are based on photoactive materials such as silicon , perovskite and Cuchalcogenide , often limited by either insufficient photovoltages for constructing bias-free devices, moisture degradation, material toxicity or high costs. |
6762d904fa469535b9fc7a90 | 1 | Organic semiconductors (OSCs) composed of earth-abundant elements are promising biocompatible materials for assembling light-driven enzymatic biohybrid systems. Solution processable π-conjugated OSCs exhibit bandgap tuneability at the molecular level , and the bulk heterojunction (BHJ) design supported by a built-in electric field for organic photovoltaic (OPV) applications facilitates efficient charge separation at the nanoscale. Hence, the solar energy conversion efficiency of OPV devices has increased sixfold, reaching 20% over the past two decades . Their early onset potentials and high photocurrents , especially in neutral pH solutions , have attracted substantial interest towards solar fuel production, with state-of-the-art synthetic catalyst systems sustaining H2 and syngas production for days. However, effectively interfacing OSCs with enzymes remains challenging . Early works on the electrochemical wiring of enzymes to electrodes have focused on (photo-inactive) redox polymer mediators comprising of viologen-or osmium-based moieties tethered to a non-π-conjugated backbone . Despite being non-diffusional, they remain limited by slow electron hopping and the polymeric mediator that controls the driving force (potential) available for the enzyme . Direct enzymatic electrochemistry is thus an appealing strategy for optimising electron transfer and maximising efficiency at the bioticabiotic interface of biohybrid PEC devices. |
6762d904fa469535b9fc7a90 | 2 | Here, we introduce the rational design of OPVs for sustainable and direct semi-artificial photosynthesis, driving unassisted fuel synthesis for up to 1 day. To this end, BHJ-based OPVs are interfaced with hierarchically nanostructured inverse opal TiO2 (IO-TiO2) electrodes hosting [NiFeSe]hydrogenase (H2ase) and [W]-formate dehydrogenase (FDH). The enzymes used in this study have been selected for solar-driven H2 evolution or CO2-to-formate conversion because of their capability of interfacing with electrodes through a direct-electron transfer (DET) mechanism. Simulations and experiments highlight the benefit of co-immobilising the enzyme carbonic anhydrase (CA), for tuning the local chemical equilibrium and enhancing electrochemical performance. Electrochemical impedance spectroscopy (EIS) also resolves electronic signatures for each biotic-abiotic interface, revealing new insights on interfacial charge transfer mechanisms. Our approach avoids the use of external and non-innocent components such as Good's buffers or viologen-based mediators, allowing for the construction of a nature-inspired artificial leaf, driving H2 or formate production directly coupled to H2O oxidation completely powered by sunlight (Fig. ). |
6762d904fa469535b9fc7a90 | 3 | Hierarchically structured IO-TiO2 electrodes were fabricated on Ti foil via a co-assembly method adapted for thick films of 37 µm (geometric surface area = 0.19 cm 2 ). Polystyrene beads (750 nm) served as a sacrificial template, forming post-annealing macropores (~660 nm) supported by a mesoporous TiO2 nanoparticle (~20 nm) skeleton (Fig. ,e and Fig. ). The IO-TiO2 electrode has a roughness factor of ~130 (Fig. ), forms a robust biocompatible interface (see successive protein film voltammetry, PFV, in Fig. ) and provides a high loading of enzymes with good electrochemical stability at cathodic potentials (see Fig. for comparison with less stable mesoporous ITO). H2ase and FDH from Desulfovibrio vulgaris Hildenborough (DvH) were chosen for their excellent H + /CO2 reduction activities and significant tolerance to small amounts of O2 , suitable for application in a semi-artificial leaf coupled to H2O oxidation. |
6762d904fa469535b9fc7a90 | 4 | While the model describes thermodynamics, kinetics and mass transport phenomena, slight deviations between experimental and simulated CPE traces can be attributed to H2 bubble growth and break-off (Fig. ), contributing to film loss through enzyme desorption, inactivation and reorientation . While DvH [NiFeSe]-H2ase is less susceptible to H2 inhibition than standard [NiFe]-H2ase , it could lose catalytic activity in the presence of high concentrations within the confined space of the macropores. COMSOL simulations revealed a local H2 concentration of 35 mM sufficient to inhibit H2ase (Fig. ) . |
6762d904fa469535b9fc7a90 | 5 | In contrast, the performance of IO-TiO2|FDH (100 pmol) electrodes depends on the CO2 hydration equilibrium through both the local H + (or pH) and CO2 concentration, as CO2 (not HCO3 -) is also consumed during catalysis (Fig. ) . Co-immobilisation of CA increased current densities from of 95% and TOFFDH of 6.2×10 4 h -1 (Fig. ). When FDH reduces CO2 to formate, it is thermodynamically favourable for CA to buffer the local pH increase by converting CO2 and H2O into HCO3 -and H + . Despite a corresponding reduction in FDH activity from a lower local CO2 concentration (from 24.5 mM to 2.9 mM, Fig. ), it is still above its Michaelis-Menten constant (KM = 0.42 mM) , and as such this factor is outweighed by an improved local pH (from 8.74 to 7.76, Fig. and Fig. ) closer to the enzyme's optimum (pH 7.1) . |
6762d904fa469535b9fc7a90 | 6 | The confined macroporous structure also creates a unique chemical environment that controls the maximum amount of co-immobilised CA, in contrast to reported mesoporous ITO electrodes . Doubling the CA loading (200 pmol) led to a low FY of 36% over 10 h (123 µmolformate cm -2 ), whereas shortening the CPE duration to 2 h improved FY to 100% (65 µmolformate cm -2 ) (Fig. ). Due to the non-redox active nature of CA, and when there is sufficient CA such that CO2, H + and HCO3 |
6762d904fa469535b9fc7a90 | 7 | As PFV and CPE primarily show the effect of CA on net catalytic electron flow, EIS provides new insights on the potential-and frequency-dependent charge carrier behaviours for IO-TiO2|enzyme electrodes. This complements two previous reports investigating [FeFe]-H2ase 40 and [W]-FDH immobilised on carbon electrodes, as detailed EIS fittings have not been conducted for [NiFeSe]-H2ase and [W]-FDH. Quantitative analysis was undertaken by fitting the Nyquist plots with equivalent circuits , comprising of a series resistor (Rs) to describe cell resistance; a double layer capacitor (Cdl) connected in parallel with a charge transport resistor (Rct) and a Warburg element (Zw) in series to represent the electrical double layer; and an RC circuit (Re, Ce) to depict electron transfer in the faradaic process (Fig. and detailed plots in Fig. ). Specifically, the top, middle and bottom circuits were employed to fit the bare IO-TiO2, IO-TiO2|enzyme (pre-onset) and IO-TiO2|enzymes (post-onset) electrodes, respectively. |
6762d904fa469535b9fc7a90 | 8 | Representative Nyquist plots measured at 0.2 V vs. RHE (pre-onset) show that enzyme-coated electrodes exhibit larger semi-circles, indicative of the resistive nature of TiO2 at positive applied potentials and the absence of catalysis (Fig. ). Conversely, moving into the conductive TiO2 regime at -0.2 V vs RHE, two semi-circles were identified. The low frequency arc of bare IO-TiO2 electrodes displayed a linear response indicative of Warburg impedance (Fig. ), while its size diminishes upon the establishment of a faradaic current by adding H2ase or FDH, which further decreases in the presence of CA. In contrast, the high frequency arcs remain unchanged, indicating non-faradaic behaviours. |
6762d904fa469535b9fc7a90 | 9 | Electrodes immobilised with H2ase demonstrate distinct behaviours between Rct and Re as a function of applied potential (Fig. ). The potential-and enzyme-independent Rct suggests that the resistivity of the electrical double layer is determined by the bare IO-TiO2 electrode and is decoupled from the faradaic process. Upon the introduction of CA, a significant decrease in Re is observed, consistent with PFV traces in Fig. . Regarding capacitance, Ce exhibits enzyme-independent behaviour and increases with cathodic potential, consistent with reported observations for TiO2-based electrodes (Fig. ) . The value of Cdl depends on the electrode area and local ion concentrations. |
6762d904fa469535b9fc7a90 | 10 | The Re values decrease with more negative potentials and exhibit even smaller values in the presence of CA. Notably, the Cdl of IO-TiO2|FDH electrodes increases from 0.2 V vs. RHE to -0.4 V vs. RHE, indicating a rise in local ion concentration due to CO2-to-formate conversion. The introduction of CA, which consumes H + and HCO3 -to compensate for CO2 depletion during formate production, effectively balances Cdl values across the voltage range. |
6762d904fa469535b9fc7a90 | 11 | Organic biohybrid photocathodes were next assembled by interfacing the IO-TiO2 network with conventional structure OPVs using conductive graphite epoxy (GE) paste as encapsulant (Fig. , see Fig. for energy levels) . The BHJ PCE10:EH-IDTBR was selected for its promising PV open circuit voltage (Voc) and operational stability . The 23 devices with active areas of ~0.25 cm 2 displayed similar performance as our previous work , averaging 1.04±0.01 V Voc, -19.0±0.9 mA cm -2 short-circuit current density (Jsc), 56.1±1.2% fill factor (FF) and 11.1±0.6% photovoltaic cell efficiency (PCE) in the reverse scan direction (Fig. ). The champion device attained a PCE of 11.8%, comparable to reported benchmark devices (Fig. ) . |
6762d904fa469535b9fc7a90 | 12 | These encapsulated organic photoelectrodes maintain their performance even under benign aqueous conditions , making them suitable for biocatalytic applications. Chopped PFVs of OPV|IO-TiO2|H2ase photocathodes displayed an onset potential of 1 V vs. RHE, consistent with opencircuit potential measurements (Fig. ). The addition of CA (100 pmol) enhanced photocurrent densities from -4 mA cm -2 to -8 mA cm -2 (Fig. ), which stands out compared to previous H2ase PEC reports , especially in the absence of external buffer components (Table ). Accordingly, OPV|IO-TiO2|H2ase+CA photoelectrodes produced 239±8 µmolH 2 cm -2 (FY = 101±1%) over 10 h at 0.6 V vs. RHE (Fig. , Fig. and Table ). As an OPV provides a Voc of 1.0 V, this effectively corresponds to an applied potential of -0.4 V vs. RHE on the IO-TiO2|H2ase+CA biohybrid, consistent with electrochemical data above (Fig. ). An exclusion control experiment in the absence of H2ase did not produce H2, validating the selective catalytic nature of H2ase, and that chopped photocurrents observed for bare OPV|IO-TiO2 electrodes correspond to the capacitive charging of TiO2 (Fig. and trumpet-shaped PFV in Fig. ). |
6762d904fa469535b9fc7a90 | 13 | Benchmark photocurrents of -5 mA cm -2 were also observed for OPV|IO-TiO2|FDH+CA photocathodes at 0 V vs. RHE, showing improved photocurrent generation and charge extraction over state-of-the-art devices 7,10,14 (Fig. , see Table for literature comparison). While bare OPV|IO-TiO2 photocathodes displayed a transient photocurrent spike upon illumination, this was notably absent in OPV|IO-TiO2|FDH+CA photoelectrodes (Fig. ). The photocathodes also maintained reproducible CO2 reduction over 10 h CPE at 0.6 V vs. RHE, producing 156±8 µmolformate cm -2 with a FY of 98±2% (Fig. , Fig. and Table ). Degradation of FDH and the buried PV device contributed to a gradual decrease in performance, supported by declining onset potentials and photocurrent densities after 10 h (Fig. ) . 1 H NMR spectra of the C-labelled product mixture further confirmed that C-formate was solely derived from CO2 reduction (Fig. and Fig. ). |
6762d904fa469535b9fc7a90 | 14 | External quantum efficiency (EQE) spectra of the corresponding photoelectrodes at 0 V vs. RHE showed a plateau between 550 and 700 nm, consistent with that of the underlying OPV (Figs. S23, 24) . Integrating the EQE spectra yielded ideal Jsc values of 13 mA cm -2 (H2ase+CA) and 16 mA cm -2 (FDH+CA), which differ from the respective -8 mA cm -2 and -5 mA cm -2 by PFV measurements. This indicates that electrode kinetics play a more substantial role at higher photocurrent densities obtained under 1 sun irradiation, compared to lower light intensities of EQE measurements . H2 bubbles produced also adhere to the electrode surface, affecting the mass transport of dissolved species such as H + and CO2 into the porous IO-TiO2 matrix . Likewise, EQE measurements at 0.6 V vs. RHE yielded low Jsc values of ~ 5.5 mA cm -2 for both photoelectrodes, further complicated by increased bimolecular recombination within the OPV under forward bias . |
6762d904fa469535b9fc7a90 | 15 | To construct a bias-free tandem device, OPV|IO-TiO2|enzyme photocathodes were coupled to BiVO4 photoanodes in a back-to-back configuration to achieve overall unassisted H2O splitting and CO2 conversion (Fig. ) . The complementary light absorption of BiVO4 (< 500 nm) and PCE10:EH-IDTBR (< 800 nm) supported good utilisation of the solar spectrum . Amorphous TiCoOx (TiCo) was spin-coated onto BiVO4, serving as a water oxidation catalyst . To further improve electrocatalytic current densities, H2ase and FDH loadings were increased to 500 pmol, while keeping the CA loading constant at 100 pmol (electrochemical optimisation in Figs. S25,26 and Table ). |
6762d904fa469535b9fc7a90 | 16 | These unassisted tandem devices solely driven by sunlight showcase the potential of OPV-based PEC biohybrids as a sustainable approach for the direct synthesis of industrially important platform chemicals. Our system takes advantage of a biocompatible organic photocathode design that maintains its performance even under neutral pH conditions , simultaneously achieving high onset potentials, unprecedented photocurrent densities and near-unity FYs in a benign bicarbonate solution (see literature comparisons in Tables ). The absence of non-innocent buffer components endows greater flexibility in the synthetic application of our biohybrid devices, including the direct coupling to water oxidation. The versatile nature of our PEC biohybrid design also allows for future applications with enzymatic cascades in a one-pot reaction , driving the synthesis of fuels such as methanol or the C-H activation of inert hydrocarbons . Whole cells could also be integrated in the IO-TiO2 matrix and biologically engineered to achieve more complex biochemical transformations. Our computational and spectroscopic methods provide the essential tools to further understand fundamental charge transfer mechanisms and catalytic reaction kinetics in 3-dimensional inverse opal electrodes, which may be extended to other porous structures such as micropillars and reactor designs . |
6762d904fa469535b9fc7a90 | 17 | We overcame the necessity of non-innocent buffer components in traditional enzymatic systems, by co-immobilising CA with H2ase or FDH in an IO-TiO2 matrix. The dynamic interplay between CA and the unique local chemical environment within IO-TiO2 enhanced PEC performance, supported by COMSOL simulations and EIS insights on charge carrier behaviour at the abiotic-biotic interface. |
6762d904fa469535b9fc7a90 | 18 | Accordingly, the optimised tandem devices with BiVO4 demonstrated an unprecedented low onset bias of -0.8 V, enabling bias-free enzymatic CO2 reduction coupled to direct water oxidation over 24 h. This fundamental understanding of biohybrid devices opens opportunities to synergistically combine the strengths of synthetic materials and chemical biology, as we ultimately target natureinspired chemical refineries. |
6762d904fa469535b9fc7a90 | 19 | IO-TiO2 electrodes were fabricated by adapting a previously reported procedure . Titanium foil was pre-cut into pieces of dimensions 1 × 2 cm 2 and were thoroughly cleaned by sonication in ethanol for 30 minutes prior to use. TiO2 nanoparticles (30 mg) were first homogeneously dispersed by sonication in a water/methanol mixture (300 µL, 4:1 volume ratio) for 2 h. A suspension of polystyrene beads (1 mL) was then centrifuged at 10000 r.p.m., followed by supernatant removal to obtain a compact polystyrene bead pellet. The pellet was further re-dispersed in 1 mL of MeOH and subjected to centrifugation at 10000 r.p.m. for removal of the supernatant. The TiO2 nanoparticle suspension (300 µL) was subsequently added to the polystyrene bead pellet and sonicated for 5 minutes in ice cold water (< 5 °C) to obtain a homogeneous mixture. The resulting dispersion was drop-cast onto the Ti foil over a well-defined geometrical surface area of 0.19 cm 2 , as demarcated by a Parafilm ring. The electrodes were then left to dry prior to removal of the Parafilm ring template. All electrodes were subsequently annealed in an oven at 1 °C min -1 ramp rate from room temperature and sintered at 500 °C for 20 minutes, removing the polystyrene beads to yield an inverse opal structure. The IO-TiO2 electrodes were then wired to metal rods using copper tape, and all electrical connections were protected from water infiltration via layers of Parafilm and Teflon tape. Kapton tape with pre-cut circular holes were also attached onto the Titanium foil, defining an active area of 0.19 cm 2 for each electrode. |
6762d904fa469535b9fc7a90 | 20 | [NiFeSe]-hydrogenase and [W]-formate dehydrogenase (denoted as H2ase and FDH thereafter) from Desulfovibrio vulgaris Hildenborough (DvH) were expressed, characterised and purified using previously published methods . Stock solutions of H2ase (20 µM in 20 mM TRIS-HCl, pH 7.6) and FDH (50 µM in 20 mM TRIS-HCl, 10% glycerol, 10 mM NaNO3, pH 7.6) were stored at -40 °C in an anaerobic glovebox (MBraun, N2 atmosphere, <0.1 ppm O2) and thawed immediately prior to use. While H2ase could be used directly for electrochemical experiments, FDH required an additional activation step via pre-incubation with DTT for 20 minutes . For experiments requiring 100 pmol loadings of H2ase and FDH each, 5 µL and 2 µL of the respective stock solutions were drop-cast onto the IO-TiO2 electrodes and left to incubate for 2-3 minutes before full immersion in the buffer solution. Proportional volumes were used for higher enzyme loadings of 250 and 500 pmol. Experiments with carbonic anhydrase (CA, 100 pmol) involved an additional but similar drop-casting step on IO-TiO2|H2ase or IO-TiO2|FDH electrodes before electrochemical characterisation. |
6762d904fa469535b9fc7a90 | 21 | Protein film voltammetry (PFV) and controlled potential electrolysis (CPE) traces were recorded on an Ivium CompactStat potentiostat in a three-electrode configuration, consisting of an IO-TiO2 working electrode, a Ag/AgCl reference electrode (in saturated NaCl, BASi MW-2030) and a Platinum mesh counter electrode. The experiments were performed using a gas-tight two-compartment H-cell separated by a Selemion or Nafion ion exchange membrane, and in a pH 6.45 buffer solution containing 50 mM NaHCO3 and 50 mM KCl. The buffer solutions were purged with CO2 (containing 2% CH4 as an internal standard) prior to each experiment. PFV scans were collected at a scan rate of 5 mV s -1 within the range -0.4 to +0.2 V vs. RHE and in 3 successive cycles to verify the stability of the enzyme-IO-TiO2 interface. Potentials measured against Ag/AgCl were converted to the RHE scale using the equation: E (V vs. RHE) = E (V vs. Ag/AgCl) + 0.059 × pH + 0.197 V (at 298 K). CPE experiments conducted over a 10 h duration were performed at -0.4 V vs. RHE under continuous stirring. All reported current densities (in mA cm -2 ) are based on the geometrical surface area (0.19 cm 2 ) of the IO-TiO2 electrodes. TON and TOF values were calculated based on total H2ase or FDH drop-cast onto the IO-TiO2 scaffold. |
6762d904fa469535b9fc7a90 | 22 | EIS measurements were carried out in the same electrochemical cell under identical conditions as PFV and CPE experiments. Impedance response was recorded with a potentiostat (IviumStat) with frequency ranges from 500 kHz to 0.1 Hz and a 25 mV sinusoidal amplitude. Impedance data were fitted with equivalent circuits using the modelling software ZView2 (Scribner Associates). |
6762d904fa469535b9fc7a90 | 23 | The finite element model was constructed in COMSOL 6.1, based on mathematical descriptions of thermodynamics, enzyme kinetics and mass transport for H2ase, FDH and CA in IO-TiO2 electrodes. A three-domain model accounting for bulk solution, electrode diffusion layer and IO-TiO2 was used to predict trends in the current densities, local pH, and local CO2 concentrations of the systems (Detailed information in Supplementary Notes Sections 2-4). |
6762d904fa469535b9fc7a90 | 24 | Solution-processed organic solar cells based on the bulk heterojunction PCE10:EH-IDTBR were fabricated (ITO|PEDOT:PSS|PCE10:EH-IDTBR|ZnO|Ag) . ITO-coated glass substrates (1.3 × 1.3 cm 2 ) were selectively etched with zinc dust and 2 M HCl, followed by thorough sonication in acetone, ethanol, isopropyl alcohol and deionised water. The cleaned substrates were then subjected to UV-Ozone treatment for 40 minutes (BioForce Nanosciences UV/Ozone ProCleaner). PEDOT:PSS was subsequently filtered through a Millex-GP 0.22 µm PES filter and spin-coated onto the ITO substrates at 4000 r.p.m. Further annealing on a hot plate in air at 383 K for 40 minutes yielded the final holetransport layer. To form the light-harvesting active layer, a solution of PCE10:EH-IDTBR (1:2 weight ratio, 24 mg ml -1 ) was prepared in chlorobenzene and spin-coated in a N2-filled glovebox at 3000 r.p.m. ZnO nanoparticles were then spin coated at 4000 r.p.m. to obtain the electron-transport layer. Finally, Ag (100 nm) was thermally evaporated through a patterned mask under vacuum (~10 -5 mbar), defining an active area of ~0.5 × 0.5 cm 2 for each device. |
6762d904fa469535b9fc7a90 | 25 | The performance of all OPVs were measured using a Sun 2000 Solar Simulator (Abet Technologies) without any additional masking at room temperature. Calibration for 1 sun illumination (AM1.5G, 100 mW cm -2 ) was performed with a certified RS-OD4 reference silicon diode. Reverse and forward J-V sweeps between -0.1 V and 1.1 V were collected at a scan rate of 100 mV s -1 over 20 mV steps, with a Keithley 2635 source meter. The shutter was switched on for dark J-V measurements. The active area of each device (~0.5 × 0.5 cm 2 ) was measured manually. |
6762d904fa469535b9fc7a90 | 26 | Graphite epoxy (GE) paste was prepared by thoroughly mixing Araldite Standard 2-part epoxy and graphite powder (4:3 weight ratio) . The GE paste was subsequently doctor-bladed onto the Ag contact of the OPV, forming a conductive, waterproof and adhesive layer that is directly interfaced with the IO-TiO2 electrode. A copper wire was then attached onto the electrode, followed by encapsulation of the entire device with Araldite 5-Minute Rapid 2-part epoxy to protect moisturesensitive components. The whole device was left to dry overnight under ambient conditions. Tandem devices were constructed in a similar manner, further wiring the organic photocathodes to BiVO4 photoanodes (defined active area of ~0.5 × 0.5 cm 2 , matching that of OPV) spin-coated with a TiCoOx water oxidation catalyst . |
6762d904fa469535b9fc7a90 | 27 | The PEC performance of encapsulated organic photocathodes was evaluated on a Newport Oriel 67005 solar light simulator equipped with an AM 1.5G optical filter, calibrated to 1 sun (100 mW cm -2 ) using a certified Newport 843-R optical power meter. The two-compartment PEC H-cell used contained either a Selemion or Nafion ion exchange membrane. A three-electrode set-up was adopted and comprised of an OPV|IO-TiO2 photocathode, a Ag/AgCl reference electrode (BASi MW-2030, stored in saturated NaCl) and a Platinum mesh counter electrode. The corresponding volumes of H2ase (100 pmol) or FDH (100 pmol) and CA (100 pmol) were drop-cast onto the OPV|IO-TiO2 electrode and incubated for 2-3 minutes prior to immersion in the buffer solution (50 mM NaHCO3 + 50 mM KCl, pH 6.45). Protein film voltammetry (PFV) and CPE scans were recorded on an Ivium CompactStat potentiostat under continuous stirring at room temperature. The buffer solution was purged with CO2 containing 2% CH4 as an internal standard before each experiment, and the entire cell was kept gas tight by sealing the septa with Loctite superglue. PFV scans were recorded between -0.1 V vs. RHE and 1.1 V vs. RHE at a scan rate of 10 mV s -1 and CPE experiments were conducted at 0.6 V vs. RHE over 10 h under chopped illumination (cycles of 50 min light on, 10 min light off). Experimentally measured potentials against Ag/AgCl were converted to the RHE scale using the equation: E (V vs. RHE) = E (V vs. Ag/AgCl) + 0.059 × pH + 0.197 V (at 298 K). Bias-free PEC tandems were evaluated in a single compartment PEC cell using a two-electrode set-up under 1 sun irradiation (AM1.5G, 100 mW cm -2 ). The BiVO4 photoanode was positioned in front of the OPV|IO-TiO2 photocathode in a back-to-back configuration and the irradiation window was defined by black masking tape on the BiVO4 photoanode, preventing illumination of the OPV|IO-TiO2 photocathode by unfiltered light. All data reported are the average of triplicate samples, with error bars referring to the corresponding standard deviation at each data point. EQE measurements were conducted using a monochromator coupled to a 300 W Xe light source (LOT-Quantum Design MSH-300) and an Ivium CompactStat potentiostat. The light intensity at each wavelength was measured using a Thorlabs PM100D power meter connected to a Thorlabs S302C thermal power sensor. The wavelength (full-width at half-maximum of 15 nm) was increased in 25 nm steps from 300 nm to 800 nm every 30 s. EQE was calculated using the equation: EQE (%) = hcJ/(eλPλ) × 100%, where h is the Planck constant, c is the speed of light, J is the photocurrent density, e is the elementary charge, λ is the wavelength and Pλ is the wavelength-dependant light intensity flux. |
6762d904fa469535b9fc7a90 | 28 | The amount of H2 produced by H2ase with time was monitored using a gas chromatograph (Agilent 7890A). Aliquots of gas (50 µL) withdrawn from the headspace of electrochemical or PEC cells were injected manually at periodic intervals, and quantified based the 2% CH4 internal standard. Formate produced by FDH was quantified using a Metrohm 882 Compact IC Plus ion chromatograph equipped with a conductivity detector (eluent comprising of 3.2 mM Na2CO3 and 1 mM NaHCO3). Reaction samples were diluted 10-fold with ultrapure deionised water prior to IC injection. O2 evolution was monitored in a N2-filled glovebox (Belle Technology) using a NeoFox-GT fluorometer and Fospor-R fluorescence oxygen sensor probe (Ocean Optics). Henry's law was used to determine the amount of dissolved H2 and O2 in the aqueous buffer solution. The faradaic yield (FY) was calculated using the equation: FY (%) = nZF/Q × 100%, where n is the number of moles of product (H2, formate or O2) produced, Z is the number of electrons needed per molecule of product (Z = 2 for both H2 and formate production, Z = 4 for oxygen production), F is the Faraday constant (96485 C mol -1 ), and Q is the total charge passed. The total charge Q was determined by integrating the current trace over a defined period. TON and TOF values for H2 and formate production were calculated based on the amount of enzyme drop-cast onto the IO-TiO2 electrodes, representing the lower limit assuming that all enzymes are immobilised and electrochemically active. Solar-to-fuel efficiencies were determined using the equation: STF (%) = (J × ΔU × FY)/P × 100%, where J is the absolute value of the operating photocurrent density, FY is the faradaic yield for the solar fuel produced and P is the incident solar power. As all tandem PEC experiments were performed at zero applied bias, ΔU = 1.23 for both overall water splitting and CO2 reduction coupled to O2 evolution. |
6762d904fa469535b9fc7a90 | 29 | The surface morphology of the IO-TiO2 electrodes was studied by field emission scanning electron microscope imaging (TESCAN MIRA3 FEG-SEM) at an electron beam accelerating voltage of 5 kV (In-beam Secondary Electron detector). The roughness factor was determined by an organic dye adsorption experiment adapted from that for nanostructured metal oxide electrodes . The dye has been shown to form a monolayer on the electrode surface. An IO-TiO2 electrode with a defined geometrical area of 0.19 cm 2 was first immersed in a 1.5 mM aqueous solution of Orange II for 30 minutes. The adsorbed Orange II dye molecules were then desorbed in 1 M NaOH (3 mL) and the amount was quantified using a UV-Visible spectrophotometer (Agilent Cary 60) through the construction of a calibration curve with known concentrations (1.5 µM, 3 µM and 6 µM). Calculations assumed that each Orange II dye molecule occupied an area of 0.4 nm 2 . |
67554cbbf9980725cf68e03a | 0 | and Fmoc-Phe-Phe-Arg-NH2,15 and ferrocenyl-Phe-Phe-His. The introduction of D-amino acids can confer further advantages, such as increased resistance against enzymatic degradation. For this reason, we designed L-His-D-Phe-D-Phe that is catalytic only in its assembled state.17 Subsequent studies demonstrated that C-terminal amidation enables a five-fold improvement in catalytic activity under analogous conditions.18 Conversely, other modifications, such as N-acetylation or peptide elongation to include Ser to serve as nucleophile, resulted in detrimental effects in catalytic performance and/or gelation ability, thus demonstrating that the design of such minimalistic supramolecular catalysts is not trivial. In light of these results, we selected the best-performing catalytic gelator identified from our previous research endeavours, i.e. L-His-D-Phe-D-Phe-NH2 (Hff), and studied its equimolar co-assembly with its enantiomer, D-His-L-Phe-L-Phe-NH2 (hFF) (Fig. ), as a different strategy to further improve catalytic activity. We took inspiration from studies led by Schneider that reported racemic peptide assemblies with synergistic assembling behaviour leading to more rigid hydrogels, relative to each enantiomer alone.20 Subsequently, the same group demonstrated the molecular basis for such an effect, revealing that the two enantiomers were alternating in the co-assembled rippled β-sheet,21 which is a structure that was predicted by Pauling and Corey in 1953. The racemic co-assembly is held together by hydrogen bonds within the sheet, thus, creating nested hydrophobic interactions between enantiomers in the dry fibrils' interior that do not occur in the case of enantiopure assemblies. We reasoned that the maximization of such hydrophobic regions could enhance hydrophobic substrate binding for catalysis. The two enantiomers were synthesised, purified by HPLC, freeze-dried, and their identity and purity were confirmed by 1H-and 13C-NMR, and ESI-MS spectra (ESI, S1-S4). Their circular dichroism (CD) spectra (Fig. ) were mirror-imaged, while their equimolar mixture was featureless, as expected. The CD signatures were reminiscent of those of other heterochiral tripeptides forming amphipathic β-sheets upon assembly, and characterised by two main peaks at 200 and 220 nm.23 Upon assembly conditions at higher concentrations, the CD spectra could be acquired only above 215 nm due to scattering, and maintained the same features (ESI, Section S5). Fourier-transformed infrared spectra (FT-IR, Fig. ) revealed the typical signals of β-sheets in the amide I and II regions, at 1632 and 1542 cm-1, respectively. Microscopy analyses were also performed. Atomic force microscopy (AFM) and scanning electron microscopy (SEM) images revealed anisotropic structures. No significant difference was found in the fibers' width across samples (Fig. , bottom), corresponding to a median of 105 ± 18 nm for hFF, 104 ± 17 nm for Hff, and 107 ± 18 nm for the racemic mixture (n = 100 counts). |
67554cbbf9980725cf68e03a | 1 | Oscillatory rheology was then used to assess the viscoelastic properties of the hydrogels (Fig. ). The two enantiomers displayed similar behaviour. Time sweeps revealed immediate gelation and an increasing elastic modulus (G') over one hour until 34.3 ± 5.0 kPa for hFF, and 34.9 ± 3.4 kPa for Hff. By contrast, the racemic mixture displayed a 15-min. lag time, with G' reaching a plateau after ~3h. Remarkably, G' of the racemic mixture reached 592 ± 45 kPa, corresponding to a ~17-fold increase relative to each enantiomer alone. Furthermore, stress sweeps demonstrated that the racemic gel displayed also enhanced resistance against applied stress, with gel-to-sol transition occurring above 300 Pa, in contrast with each enantiomer whose moduli dropped at 70 Pa. Overall, the enhancement in the viscoelastic properties was consistent with the expected behaviour for the co-assembled rippled β-sheet.17 Singlecrystal X-ray diffraction is the technique of choice to confirm the enantiomeric peptide alternation in the stacks,24 but unfortunately all our attempts to crystallize Hff, hFF, or their racemic mixture were unsuccessful. Finally, we tested the assemblies for catalytic hydrolysis of p-nitrophenyl acetate (pNPA) as a chromogenic substrate (Fig. ). The parent compound Hff-COOH was reported to display a kobs = 3.5•10-4 s-1 when tested in the form of fibrils (25 mM) with 0.2 mM pNPA.17 C-terminally amidated Hff performed similarly, also at two-fold concentration to yield a hydrogel (i.e., kobs = 3.6•10-4 s-1).18 In this work, when tested with 1 mM pNPA, both Hff and hFF fibrils (25 mM) displayed a kobs = 4.5•10-4 s-1. Remarkably, the racemic mixture with the same peptide total concentration (i.e. = 25 mM obtained from 12.5 mM of each enantiomer) showed a further improvement by over 50% corresponding to a kobs = 7.0•10-4 s-1. To the best of our knowledge, this is the first report of a self-assembled catalyst whereby the racemic mixture displays improved catalysis relative to each enantiomer alone, which suggests the presence of organized co-assembled peptide fibrils, in agreement with spectroscopic and microscopic data that are compatible with the presence of rippled β-sheets. The presence of more hydrophobic pockets in the co-assembled catalyst that could favor substrate binding, and lead to enhanced catalysis, was then verified by fluorescence measurements using 8-anilinonaphthalene-1sulfonate ammonium salt (ANS). The use of this fluorofore is well-established to monitor the formation of hydrophobic environments, such as those arising upon peptide self-assembly.26,27 In particular, ANS fluorescence undergoes a blue-shift and increase in intensity when it is located in more hydrophobic environments,28 and this was also our case, as shown in Fig. . ANS alone displayed a mild fluorescence with a broad peak centred at 475 nm in the 450-500 nm range. A blueshift to 455 nm occurred when ANS was embedded in peptide assemblies, and fluorescence intensity increased 2-fold and 3-fold, in case of enantiopure and racemic assemblies, respectively. We inferred the presence of more hydrophobic pockets in the latter, as expected for rippled β-sheets. In conclusion, this proof of concept work describes the spectroscopic, microscopic, and rheological behaviour of a racemic mixture composed of a minimalistic tripeptide sequence that bears His for hydrolase mimicry,6,29,30 and Phe-Phe31,32 as a self-assembling motif. All the data support the formation of a rippled β-sheet, which was already described as a useful type of assembly for racemic tripeptides,24 as well as longer peptides to yield supramolecular hydrogels with increased stiffness.16 Importantly, this is the first work that proposes the use of racemic minimalistic tripeptide co-assemblies for enhanced biocatalysis, thus opening the way to the use of this approach to develop green catalysts. Given that peptide catalysts' optimization based on expert-knowledge-guided discovery is not trivial, we anticipate that integration with machine-learning approaches will be key to speed up advances in the field. The vast progress in peptide modelling methods,35 with the concomitant generation of curated datasets for catalytic peptides36 are important steps ahead to enable rapid developments. Indeed, combination of in silico and experimental approaches is already being successful to shed new light on the structure-activity relationship of catalytic amyloids,37 and looks promising to unveil key mechanistic details to unlock their full potential in various applications. |
6617632991aefa6ce1505d28 | 0 | Agriculture has been a cornerstone of global economic development, providing sustenance, income, and employment opportunities. As we strive for enhanced agricultural productivity to alleviate hunger, it is imperative to address sustainable waste management. Low-income countries are projected to witness a 40% surge in waste generation, with agriculture accounting for a significant share of this increase. In Nigeria, approximately 70% of the population relies heavily on subsistence agriculture for their daily sustenance. With 70.8 million hectares of arable land, Nigeria's major crops encompass cassava, maize, guinea corn, beans, yam, rice, and millet. Agricultural residues, comprising various organic materials derived from crop harvesting and processing, encompass diverse components such as corn stalks, husks, fruit flesh, and seeds. Their characteristics, including particle size, moisture content, and bulk density, vary depending on geographical location and handling methods. |
6617632991aefa6ce1505d28 | 1 | Escalating prices of inorganic fertilizers, driven by factors like surging energy costs and geopolitical tensions, have led farmers to seek sustainable alternatives to maintain yields. Notably, fertilizer production consumes substantial energy, relying on fossil fuels (as seen in N-fertilizers using the Haber-Bosch process) or fossil ore deposits (e.g., phosphate rock). Synthetic fertilizers pose considerable environmental risks, including soil degradation, reduced organic matter content, leaching, pollution, and climate change exacerbation. As the global population expands, there is heightened pressure to augment food production, potentially surpassing the continent's capacity to meet demand, especially in Africa. The Circular Economy (CE) concept, introduced by the European Commission, advocates for material recovery to combat environmental and social challenges. While postharvest residues hold potential as substitutes for raw materials in fertilizer production, the fertilizer industry has yet to fully exploit this renewable resource. Encouraging the fertilizer industry to integrate biomass valorisation in their technologies could be facilitated through direct subsidies. Agriculture currently contributes to approximately 21% of greenhouse gas emissions, intensifying the impacts of climate change. Indiscriminate burning of agricultural waste exacerbates these issues, with only a fraction being utilized for fodder, erosion control, and fertilizer. The majority is disposed of through burning, which can have dire health consequences due to the inhalation of toxic fumes. This method also contributes significantly to carbon emissions, exacerbating global warming. Implementing innovative waste management techniques that prioritize environmental and human health is crucial, yet not all waste categories are addressed, and widespread adoption faces certain challenges. Various plant biomass resources have been utilized in the production of different types of manures worldwide, primarily in the form of compost or biochar. While the conversion of plant biomass to biochar or compost holds promise for carbon sequestration and soil enhancement, concerns such as high energy requirements, emissions during production, limited scalability, environmental trade-offs, nutrient loss, long processing times, potential odour and pathogen issues, and the need for significant space and infrastructure must be addressed. This study aims to develop eco-friendly and cost-effective methods for the instantaneous conversion of residual plant biomass into organic fertilizer. Additionally, it seeks to compare its impact on the growth, development, and yield of maize plants to that of NPK 20:10:5 and rabbit manure. Consideration will be given to fertilizer yield and conversion time in comparison to biochar and compost. |
6617632991aefa6ce1505d28 | 2 | Clay soil samples were collected using a shovel to a depth of more than two feet. Each sample was then placed in five sacks, each 20 cm deep and weighing 30 kg. This was carried out in the month of March, during a period of no rainfall and minimal weed growth, at Tsuanin Kura GRA Sabon Tasha, Kaduna South, Kaduna State. Maize seedlings (AS-SAMAD Agro Allied Co., PZ Kaduna State) were purchased from an agrochemical store in Kaduna Metropolis, Kaduna State, Nigeria. NPK 20:10:5 Matrix Fertilizer (KM 3 Dumbin Duste Zaria, Kaduna) was also obtained from an agrochemical store in Kaduna Metropolis. Rabbit manure was collected from a garden at Tsuanin Kura GRA. Calcium hydroxide, Sodium Hydroxide, thiourea, Nitric and Sulfuric acids were procured from chemical vendors along Kano Road, Kaduna Metropolis, Kaduna State, Nigeria. |
6617632991aefa6ce1505d28 | 3 | To prepare Fertilizer A, 50 g of the homogenized plant residue was placed in a 250 mL beaker. Subsequently, 25 mL of 50% Nitric acid was added, and the mixture was heated at 120°C for 20 minutes. After cooling to 60°C, 10 g of wood ash was added while stirring for 10 minutes. The mixture was then transferred to an aluminum foil and oven-dried at 80°C for 24 hours, as illustrated in Figure . |
6617632991aefa6ce1505d28 | 4 | For the preparation of Fertilizer B, 50 g of the homogenized plant residue was placed in a beaker containing 20 mL of 20% sodium hydroxide. The mixture was heated at 100-110°C for 10 minutes, after which 15 mL of 30% sulphuric acid was added on continual heating for 10 minutes. Following cooling to room temperature, and the mixture was oven-dried at 47°C for 1 hour, as shown in Figure . |
6617632991aefa6ce1505d28 | 5 | To produce Fertilizer C, 50 g of the homogenized plant residue was transferred into a beaker containing 30 mL of 50% Nitric acid. The mixture was heated at 100-110°C for 20 minutes, 20 g of wood ash was added. After cooling to room temperature, and the mixture was oven-dried at 47°C for 1 hour, as depicted in Figure . |
6617632991aefa6ce1505d28 | 6 | The sample fertilizer C, manure and NPK fertilizer were collected for chemical and physical analyses. Moisture content was determined as weight loss upon drying at 105 o C in an oven for 24 hrs. Electrical conductivity (EC) and pH were measured using water extract 1:5 (w/v), total nitrogen (TN) by the Kjeldhal method. Flame atomic absorption spectroscopy was used to measure total potassium (TK) on acid digested samples. Total Phosphorus (TP) was determined using the H2SO4 and H2O2 digestion method, total carbon (TC) was determined by dry combustion method. |
6617632991aefa6ce1505d28 | 7 | For the irrigation process, 700 cm 3 of water is sprinkle on each pot morning and evening in four days' interval for two months before adequate rainfall in the month of May. For the initial application, two weeks after germination, equal quantities of each fertilizer were used, with individual weights as follows: 20.0 g of Fertilizer A was applied near the roots of maize A and maize B, 32.80 g of NPK 20:10:5 fertilizers was applied to maize C, and 12.2 g of rabbit manure was applied to maize E. No fertilizer was applied to maize D, which served as the control (see Figure ). Fourteen days after the first application, a mixture of 17.2 g of Fertilizer B, calcium hydroxide (solid), and clay soil was applied to maize A, while a mixture of 21.0 g of Fertilizer C, calcium hydroxide (solid), and clay soil was applied to maize B. Maize C received 49.2 g of NPK 20:10:5 fertilizers, and maize E was given 18.3 g of rabbit manure (see Figure ). Fourteen days after the second application, maize A received a mixture of 21.8 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.