id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
673af1a55a82cea2fa7f7cf6 | 6 | Diagnosis of asthma typically involves a combination of clinical assessment, pulmonary function testing, and response to bronchodilators. Spirometry is the most common diagnostic tool, measuring the forced expiratory volume (FEV1) and forced vital capacity (FVC). A significant improvement in FEV1 after administration of a bronchodilator indicates reversible airway obstruction, a hallmark of asthma (Global Initiative for Asthma, 2023). |
673af1a55a82cea2fa7f7cf6 | 7 | Asthma is a growing public health concern in Africa, with significant variation in prevalence across different regions. Historically underdiagnosed and undertreated, asthma in African populations is influenced by a combination of genetic, environmental, and socioeconomic factors. The continent's rapid urbanization, changing environmental conditions, and disparities in healthcare access have contributed to a rise in asthma cases, particularly among children and young adults. |
673af1a55a82cea2fa7f7cf6 | 8 | Urbanization has been closely linked with an increase in asthma prevalence across Africa. Studies have consistently shown that urban residents are more likely to develop asthma than their rural counterparts. This disparity is largely attributed to environmental factors prevalent in cities, including higher levels of pollution, vehicular emissions, and industrial activity. Urban environments also expose individuals to indoor allergens such as dust mites, cockroaches, and mold, which are less prevalent in rural areas . |
673af1a55a82cea2fa7f7cf6 | 9 | In rural areas, asthma is less common, but when present, it is often underdiagnosed and undertreated due to limited healthcare infrastructure and resources. Rural populations are more likely to rely on traditional medicine, and respiratory symptoms may not be recognized as asthma . However, with increasing rural-to-urban migration and lifestyle changes, the prevalence of asthma is also on the rise in rural regions. |
673af1a55a82cea2fa7f7cf6 | 10 | Asthma affects individuals of all ages, but the prevalence varies between age groups and genders. In Africa, asthma is more prevalent in children and adolescents, with many cases beginning in early childhood. A study by found that asthma prevalence in children aged 6-7 years in African cities ranged from 5% to 23%, depending on the region. |
673af1a55a82cea2fa7f7cf6 | 11 | Despite this increase, there is still a large gap in accurate data collection and diagnosis, which makes it difficult to assess the full scope of asthma prevalence in many regions. Limited healthcare infrastructure, especially in rural areas, contributes to underreporting of cases . Nonetheless, the overall trend indicates a growing burden of asthma in African populations. |
673af1a55a82cea2fa7f7cf6 | 12 | Environmental factors play a significant role in asthma development and exacerbation in Africa. Air pollution, both outdoor and indoor, is a major contributor. In urban areas, pollutants such as particulate matter (PM10, PM2.5), nitrogen dioxide (NO 2 ), and sulfur dioxide (SO 2 ) from vehicles, industries, and fossil fuel combustion are common asthma triggers . |
673af1a55a82cea2fa7f7cf6 | 13 | Socioeconomic disparities have a profound impact on asthma prevalence and outcomes in Africa. People living in poverty are more likely to reside in substandard housing with poor ventilation, dampness, and mold, which can trigger asthma symptoms. Low-income individuals also have limited access to clean fuels, increasing their exposure to indoor air pollution from biomass . |
673af1a55a82cea2fa7f7cf6 | 14 | Healthcare access is a critical determinant of asthma outcomes in Africa. In many regions, particularly rural areas, there is a shortage of healthcare professionals trained to diagnose and manage asthma. Additionally, the availability of essential medications, such as inhaled corticosteroids and bronchodilators, is limited in some countries, and affordability remains a challenge for many . |
673af1a55a82cea2fa7f7cf6 | 15 | Asthma is influenced by various risk factors that contribute to its development and exacerbation. In addition to these risk factors, numerous comorbidities often accompany asthma, complicating management and exacerbating symptoms. Understanding these elements is crucial for effective prevention and treatment strategies, particularly in the African context, where environmental, socioeconomic, and healthcare factors intersect. |
673af1a55a82cea2fa7f7cf6 | 16 | Allergens and environmental pollutants are significant contributors to asthma prevalence and severity in Africa. Common indoor allergens include dust mites, mold, animal dander, and cockroach droppings. Studies indicate that exposure to these allergens is prevalent in many households, especially in urban areas where indoor environments can be poorly ventilated and damp . |
673af1a55a82cea2fa7f7cf6 | 17 | Outdoor air pollution, primarily from vehicular emissions, industrial discharges, and biomass burning, significantly impacts respiratory health. High levels of particulate matter (PM2.5 and PM10), nitrogen oxides (NO x ), and sulfur dioxide (SO 2 ) have been linked to increased asthma symptoms and hospitalizations . In Africa, rapid urbanization and population growth contribute to |
673af1a55a82cea2fa7f7cf6 | 18 | Infectious diseases, particularly respiratory infections, play a crucial role in asthma exacerbations, especially in children. Viral infections such as respiratory syncytial virus (RSV) and rhinovirus are common triggers for asthma symptoms and can lead to severe asthma attacks . In Africa, where respiratory infections are endemic, the burden of asthma can be compounded by high rates of pneumonia and other respiratory illnesses. |
673af1a55a82cea2fa7f7cf6 | 19 | Lifestyle factors, including diet and physical activity, significantly influence asthma prevalence and management. Diets high in processed foods, sugars, and unhealthy fats are associated with an increased risk of asthma and poorer control of existing asthma symptoms . Conversely, diets rich in fruits, vegetables, and omega-3 fatty acids may have a protective effect against asthma. |
673af1a55a82cea2fa7f7cf6 | 20 | Physical activity also plays a critical role in asthma management. Regular exercise can improve lung function and overall health, reducing the frequency and severity of asthma symptoms. However, exercise-induced bronchoconstriction (EIB) can occur in some individuals, particularly those with poorly controlled asthma . Therefore, tailored exercise programs that consider individual asthma profiles are essential. |
673af1a55a82cea2fa7f7cf6 | 21 | Current treatment guidelines for asthma, such as those provided by the Global Initiative for Asthma (GINA), offer evidence-based recommendations for the management of asthma. These guidelines emphasize the importance of a stepwise approach to treatment, which includes the use of bronchodilators, inhaled corticosteroids, and other medications based on the severity of the disease (GINA, |
673af1a55a82cea2fa7f7cf6 | 22 | However, the applicability of these guidelines in African settings may be limited due to factors such as medication availability, cultural beliefs about treatment, and varying healthcare practices. In many African countries, access to essential medications, including inhalers and corticosteroids, is inconsistent. Consequently, health professionals may need to adapt these guidelines to account for local resources and patient needs, ensuring that treatment plans are feasible and effective within the context of the healthcare system . |
673af1a55a82cea2fa7f7cf6 | 23 | Community health programs play a crucial role in improving asthma management and outcomes in Africa. These programs focus on raising awareness about asthma, providing education, and enhancing access to care at the community level. Initiatives such as community health worker programs can bridge gaps in healthcare access, particularly in underserved areas . |
673af1a55a82cea2fa7f7cf6 | 24 | Self-management strategies are vital for individuals with asthma, particularly in resource-limited settings. Patients should be encouraged to develop asthma action plans that outline steps to take during exacerbations and strategies to avoid triggers . Monitoring symptoms and using peak flow meters can also help patients assess their asthma control and respond proactively to worsening symptoms. |
673af1a55a82cea2fa7f7cf6 | 25 | Understanding asthma in Africa requires a robust body of research that addresses the unique challenges and contexts of the continent. Despite some progress, significant data gaps remain, limiting the ability to develop effective interventions and policies. Identifying these gaps and recommending targeted research initiatives is essential for advancing asthma care and management in African populations. |
673af1a55a82cea2fa7f7cf6 | 26 | Current research on asthma in Africa has focused on various aspects, including epidemiology, risk factors, and management strategies. Studies have highlighted the prevalence of asthma and its significant impact on morbidity and healthcare systems . Research efforts have also examined the role of environmental factors, such as air pollution and allergens, in exacerbating asthma symptoms, as well |
673af1a55a82cea2fa7f7cf6 | 27 | Additionally, some research initiatives have explored the effectiveness of community health programs in improving asthma management and raising awareness about the disease . However, many studies are limited in scope, often focusing on specific regions or populations, leading to an incomplete understanding of asthma as a public health issue across the continent. |
673af1a55a82cea2fa7f7cf6 | 28 | To address the existing knowledge gaps and improve asthma management in Africa, the following research initiatives are recommended: Healthcare and Management: Access to asthma care is limited in many parts of Africa, with significant disparities between urban and rural settings. Although global guidelines like GINA exist, their applicability in Africa can be hindered by resource constraints. |
673af1a55a82cea2fa7f7cf6 | 29 | Research and Data Gaps: While some progress has been made, there remain substantial gaps in our understanding of asthma in Africa, particularly regarding its epidemiology, Community Health Leaders: Develop and expand grassroots health programs that empower patients and families to understand and manage asthma, fostering a culture of preventive healthcare and self-management. |
66afd279c9c6a5c07a0101fd | 0 | Life has been defined as a self-sustaining chemical system that undergoes Darwinian evolution. In addition, life is considered self-sustaining, self-contained, self-regulating, self-replicating, and self-organizing. As a result of these properties, living systems can undergo Darwinian evolution, and a constant increase in fitness through trial-and-error is achieved. To gain deeper insights into the origins and mechanisms of life, scientists have turned to designing model systems, such as synthetic cells, which mimic these essential life traits. An ideal synthetic cell contains all the traits above and can undergo open-ended Darwinian evolution. Complex coacervate droplets and liposomes are promising synthetic cells since they form compartments and enrich different molecules, serving as nutrients, catalysts, or building blocks. This work focuses on the self-sustainment and division of complex coacervate-based synthetic cells. Several approaches to dividing coacervate protocells have been described, frequently using external factors such as shear forces , mechanical pressure , or temperature. Alternatively, it has been demonstrated that droplets can be divided by molecules produced within the coacervate droplet, such as proteins that form bundles that deform the coacervate phase until it is split into two droplets. These systems use static coacervates, meaning they are close to equilibrium and do not consume energy to sustain themselves. We propose to use active coacervate droplets to get closer to living cells. In this context, active droplets do not exist in equilibrium but require a chemical reaction to sustain themselves. Theory on such droplets has predicted that active coacervate droplets can undergo division due to shape deformation despite the surface tension. However, the number of examples of self-division in active coacervates is limited. We recently found an example of active coacervate-based droplets that can self-divide. These active complex coacervate droplets comprise the homo-polymeric, polydisperse RNA (polyuridylic acid, poly-U) as a polyanion and a peptide as a polycation. They are active droplets because a chemically fueled reaction cycle regulates the peptide's affinity for the polyanion. In response to fuel, these droplets emerge, and as they consume the fuel, they decay again. Excitingly, we found that towards the end of such a fuelingstarvation cycle, the droplets could self-divide by ripping apart into tiny daughter droplet fragments. Eventually, these fragments also dissolved. As the mechanism of the division of these droplets was not understood, we could not control it. Moreover, poly-U is a critical component for the division, and as it is derived from a biological source, it suffers from batch-to-batch variations, making the self-division unpredictable, further complicating the understanding of the division. |
66afd279c9c6a5c07a0101fd | 1 | In this work, we unveiled the underlying mechanism of the selfdivision of active coacervate droplets, which is related to the maturation of droplets, leading to small, glassy domains within the original droplet. Balancing the maturation rate with the lifetime of the droplets leads to tiny speckles that dissolve roughly a minute later than the original droplets. We can tune the number of offspring each droplet produces and for how long they survive. We show that additional molecules are passed from the original droplet to the offspring. Finally, the daughters can be rescued and become the next generation when additional fuel is supplied after division. We envision the self-division mechanism as a critical development toward synthetic life. For example, when the droplets are combined with self-replicating molecules, they are passed on to the next generation through our new self-division mechanism. |
66afd279c9c6a5c07a0101fd | 2 | System design. The self-dividing, active droplets are based on complex coacervation regulated by a fuel-driven chemical reaction cycle. We used the peptide acetyl-FRGRGD-OH as polycation (peptide) in which F means phenylalanine, R means arginine, G means glycine, and D means aspartic acid. Upon adding 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide (EDC, fuel), the C-terminal aspartic acid reacts to its corresponding cyclic anhydride state-the peptide activation. The fuel is converted into its corresponding urea form (EDU, waste) in the activation. The cyclic anhydride of the activated peptide spontaneously hydrolyzes to yield the original peptide-the deactivation. The peptide is zwitterionic with a net charge of 0, whereas the activated peptide has an overall charge of +2. The cationization of the peptide enables it to form complex coacervates with polyanions through electrostatic interactions (Scheme 1). This work explores polystyrene sulfonate (PSS) of different lengths as polyanion. The transient, activated peptide is produced as long as fuel is present and the complex coacervate droplets can be sustained. However, the droplets dissolve as soon as the peptide deactivation outcompetes activation. Coupling a fuel-driven reaction cycle with droplet material formation implies that droplets emerge in response to chemical fuel. Thus, in fueling-starvation experiments, a batch of fuel is added, and the droplets have a finite lifetime. |
66afd279c9c6a5c07a0101fd | 3 | Mechanism of the self-division. In previous work, when homopolymeric poly-U was used as a polyanion, we found droplet selfdivision towards the end of their lifetime. In the last minutes of their lifetime, a large droplet would fragment into tens of smaller droplet fragments. We hypothesize that the division results from the polydispersity of the polyanion. Specifically, we suspect that during the lifetime of the droplet, the long polyanion separates from the shorter polyanion and forms solid domains, a process referred to as maturation (Fig. ). As the droplet dissolves, the solid domains are expelled and survive longer than the main droplet. |
66afd279c9c6a5c07a0101fd | 4 | To verify the maturing droplet mechanism and test its generalizability, we ordered analytical standards of PSS of 4.3, 261, 453, 622, and 976 kDa molecular weight corresponding to a range in the degree of polymerization of 21 up to close to 5000 (Supporting Table ). We refer to the 4.3 kDa PSS as the "short" PSS. All others are referred to as "long" PSS. Noteworthy, all these polymers had a polydispersity index below 1.2. We used 12.3 mM peptide and combined it with 20 mM PSS (as expressed in monomer concentration) in 200 mM MES buffered water at pH 5.3. When we supplied these solutions with 75 mM EDC as fuel, the samples became measurably turbid with all PSS independent of length (Fig. ). The turbidity decayed again with time, indicating the formation of transient assemblies that scatter light. Importantly, the samples containing long PSS regained their original transparency much later than those with short PSS, e.g., after 12 minutes for short PSS and after 19 minutes for 453 kDa PSS. Moreover, confocal microscopy revealed round, micron-sized droplets for short PSS. In contrast, for all the long PSS lengths we used, we observed rugged, non-spherical agglomerates of microns in size that dissolved much later than their short PSS counterparts (Fig. ). In conclusion, the long PSS forms more solid-like coacervates that dissolve slower than the short PSS in response to fuel, which aligns with our hypothesis of droplet division-when combined, long polyanions form solid-like coacervate-based assemblies that dissolve slower than the droplets formed by the short polyanion. Excitingly, when we used mixtures of short and long PSS as the polyanion, we observed that the droplets self-divided through a similar fragmentation mechanism observed for poly-U. Specifically, we prepared droplets like the ones above but now used 25% long PSS (622 kDa) and 75% short PSS and imaged their behavior by confocal microscopy. The droplets consisting of PSS mixtures had a lifetime measured by turbidimetry of 17 minutes, i.e., between short and long PSS lifetimes. Moreover, the long-to-short PSS ratio could tune the lifetime (Fig. ). Confocal microscopy revealed that the mixed PSS droplets were spherical and could fuse, from which we conclude they were liquid (Fig. ). When we fueled the solution with 75 mM EDC, we observed the droplets' emergence and fusion in the first minutes of the reaction cycle. Toward the end of the reaction cycle, vacuoles inside the droplets appeared (Fig. ), which we previously attributed to a mechanism in which the efflux of deactivated droplet material outcompetes the influx of activated droplet material. Around the same time, we observed the first speckles inside the remaining droplet shell. These speckles were liberated and are independent droplet fragments that survived for tens of seconds until they finally dissolved (Fig. ). Using only short PSS under the same conditions showed no vacuole, speckle formation, or division (Fig. ). These findings support our hypothesis that the long PSS forms solid-like domains inside the liquid coacervate phase that dissolve slightly later than the rest of the droplet. We analyzed the PSS composition of the droplet to validate the mechanism by spinning down the droplet phase and analyzing it using gel permeation chromatography (GPC). Because of the dynamic nature of the active droplets and the long centrifugation times required, we used static droplets as a model instead. We mutated the aspartic acid of the peptide with an asparagine. This model for the activated peptide has an overall charge of +2 and is structurally similar to the activated peptide but does not hydrolyze-we refer to it as the activated peptide model. We used our previously described kinetic model to predict the concentration of the activated peptide at different time points in the reaction cycle. The solution of static droplets was then prepared with a mixture of short and long PSS, the peptide, and the model for the activated peptide. GPC showed that the droplets early in the cycle mainly comprised the short PSS. As the droplets progressed in the reaction cycle and less activated peptides became available, their composition shifted towards long, such that, after 20 minutes, the droplets were made of 80% long PSS (Fig. ). In other words, a mixture that resembles the young droplets seem to favor short PSS whereas older droplet comprises mostly long PSS. |
66afd279c9c6a5c07a0101fd | 5 | We verified the droplet composition also changes in dynamically evolving droplets by confocal microscopy using short (10.8 kDa) Cy3-labeled PSS and long Cy5-labeled PSS (404 kDa). Indeed, the merged confocal micrographs (green short PSS and magenta long PSS) show a transition of white droplets, meaning both PSS lengths are mixed in the coacervate phase in the beginning to nearly completely magenta speckles and at the end of the reaction cycle and magenta offspring droplets (Fig. , intensity line plot Fig. ). In addition, we measured the ratio between the intensities of short and long PSS of coacervate droplets 5 min and 15 min after adding fuel. Five minutes after the fuel addition, the ratio of the fluorescence intensities of the short and long PSS was roughly 1.0. After 15 minutes, the ratio shifted to 1.6, meaning the long PSS was enriched in the coacervate droplet towards the end of the reaction cycle (Fig. ). The confocal microscope findings support the hypothesized mechanism of the division that short PSS is expelled first, while more solid domains comprising the long PSS remain. |
66afd279c9c6a5c07a0101fd | 6 | Taken together, the active droplets consist of a mixture of long and short PSS. During the droplets' short lifetimes, the long PSS forms mature domains as speckles within the droplets. As the fuel depletes, the activated peptide population decreases, preferentially expelling the short PSS. The speckles are expelled from the mother droplet in its final minutes. The offspring speckles move independently in the reaction solution until they finally dissolve. With our mechanistic insights, we tested which parameters affect the droplet division. Given our interest in using these droplets for the de novo synthesis of life, we focussed on tuning the offspring's lifetime and number. To quantify the division, we used continuously imaged microfluidic reactors because, in such reactors, only one large droplet is produced, facilitating the quantification of the selfdivision process. Using the abovementioned conditions, microreactors were produced by combining peptide, PSS, and fuel in a buffered aqueous solution. Immediately after mixing, 5 µL of this aqueous solution was transferred to 50 µL of fluorinated oil and mixed gently. The resulting emulsion comprised millions of polydisperse water droplets ranging from tens to hundreds of microns in diameter-the microreactors. Within these microreactors, we observed the formation of coacervate-based droplets. We imaged the entire reactor in X, Y, and Z with time intervals of roughly 10 seconds until all the droplets had dissolved (Fig. ). Early in the cycle, we observed multiple coacervate droplets that rapidly grew and sank to the bottom of the reactor, where they fused until only one droplet remained. We chose to image with a diameter of roughly 35 µm such that the volume of the final droplet was similar from reactor to reactor at roughly 460 µm 3 (Fig. ). In line with the bulk microscopy data, after roughly 17 minutes, the division process started and produced tens of offspring droplets (Fig. , supporting movie 1). To quantify the division process, we projected all Z-planes of one microreactor to one plane (Fig. ). In the Z-plane projection, we observed the formation of vacuoles, the formation of speckles, and, a few seconds later, the liberation of the speckles as droplet fragments from which we conclude the droplets in the confined space of a microreactor behave similarly to those in bulk. We wrote a Python script that counts the droplets in each frame of the confocal videos (See SI), which revealed the number of droplets rapidly decreased to one because of the fusion. After 17 minutes, the number of droplets increased from one (the "mother") to an average of 12 (the "offspring") before rapidly decreasing when all offspring dissolved. |
66afd279c9c6a5c07a0101fd | 7 | We analyzed the influence of the mother droplet size, the fraction of long PSS, and its length on the division mechanism. First, we varied the size of the mother droplet by analyzing different microreactor sizes. An increased microreactor volume increases the droplet material and, thus, a larger coacervate droplet after fusion (Fig. ). We found that droplets smaller than roughly 145 μm 3 , i.e., those with a 3.3. µm radius showed no division. From there on, the number of offspring fragments increased with the size of the initial coacervate droplet until it leveled off at, on average, 23 offspring (Fig. ). Increasing the coacervate droplet size also increased the variability in the offspring number. We hypothesize that critical droplet size is needed to accumulate sufficient long PSS to produce at least two speckles. Besides, larger droplets lead to more speckles. |
66afd279c9c6a5c07a0101fd | 8 | Next, we analyzed the influence of the fuel concentration. Low fuel amounts lead to short-lived droplets, offering less time for the long PSS to mature and form speckles (Fig. ). Indeed, we found that the coacervates fueled with 25 mM EDC did not show division. Fueling with 50 and 75 mM EDC leads to, on average, eight offspring droplets. Surprisingly, using 100 mM EDC or more decreases the offspring to roughly three (Fig. ). We assume that with the increasing lifetime of the droplets, the chance that speckles combine increases, leading to fewer offspring. The amount of EDC we used did not influence the offspring's lifetime (Fig. ). Next, we kept the total PSS concentration constant at 20 mM as expressed as monomers but varied the fraction of long PSS from 0% to 65%. Increasing the fraction of long PSS leads to the formation of increasingly more viscous droplets, as measured by FRAP experiments (Fig. ). We observed that droplets without long PSS dissolve homogeneously without producing any offspring (Fig. ). In contrast, droplets with as little as 10% PSS displayed a few speckles and consistently produced droplet fragments, albeit only a few. With an increasing fraction of PSS, the number of offspring increased. However, when we tested 40% long PSS or more, the fusion of the droplets was so slow that the initial droplet was rugged and inhomogeneous. At 65% of long PSS, it became evident that the large droplet was merely a conglomerate of the droplets but so viscous that they did not fuse. While the dissolution led to droplet fragments, it was clear that the fragments were simply the remnants of the original, unfused droplets. Given that the original droplets were never homogeneous, we argue that this process is hard to define as division. We also analyzed the offspring's lifetime, which increased with increasing ratio of long PSS from an average of 40 seconds to 70 seconds from 10% to 35%, respectively (Fig. ). Finally, we tested the length of the long PSS on the droplet division behavior and found that a long PSS of 261 kDa at 25% did not produce any droplet fragments. Lengths beyond 261 kDa consistently produced offspring with nearly identical lifetimes (Fig. ). We conclude that the division mechanism is a concerted effort between the short and long PSS-the short is required to produce liquid droplets that enable the long PSS to mature in the next generation of droplets. Too much short PSS hardly produces offspring. Too much long PSS prevents the fusion into the necessary droplet for maturation. |
66afd279c9c6a5c07a0101fd | 9 | In our vision of de novo synthesis of life, self-division is critical. Nevertheless, equally important is that the offspring droplets can become the next generation of droplets upon refueling (Fig. ). In this context, a speckle liberated in the form of an offspring droplet should serve as a nucleus for the growth when fuel is available again while suppressing the nucleation of new droplets. We refer to such a droplet as a second-generation droplet. Thus, experimentally, the exact timing of the refueling is critical. Too early in the cycle leads to a regrowth of all mother droplets, and too late results in the complete dissolution of all droplets and a reset of the system (see Fig. ). We tested the refueling experiments using 12.3 mM peptide, 20 mM PSS with 75% short PSS, and 25% long PSS fueled with 50 mM EDC. The first coacervate droplets could be observed 30 seconds after the first fuel addition (Fig. ). The droplets grew and fused for several minutes. After roughly eleven minutes, the vacuolization started, followed by the speckle formation and the liberation into offspring. After 11 minutes, we added a second batch of 50 mM fuel. Because of the flow induced by adding a second batch of fuel, it could be hard to track the offspring droplets. We tracked a few droplets and found that their size increased quickly after adding fuel, e.g., from 1 to 1* in Fig. , and more examples in Fig. . 30 seconds after the second fuel addition, we observed a bigger and smaller population of droplets in the sample. We assume that the bigger ones originated from the offspring droplets and thus are second-generation droplets. The smaller droplets were freshly nucleated, the new first generation. We found the resulting second-generation droplet to be more viscous in the first few minutes than the first-generation. The fusion with other droplets is slower, resulting in non-spherical droplets. Over time, the second-generation droplet takes up more short PSS, resulting in a less viscous droplet that fuses more easily. Towards the end, the vacuole formation, followed by the speckles formation, is observed. Finally, the coacervate droplets' division occurs (Fig. , supporting movie 2). We conclude that we can rescue the offspring by adding a second batch of fuel. In addition, the division mechanism is not affected by the second fuel addition nor by the different morphology of the second-generation droplets due to the initially higher long PSS content. Rescuing the offspring droplets with another batch of fuel is exciting, as it means that the identity of the droplets is not wiped out with every refueling experiment. Ideally, information-containing molecules (like a synthetic genotype) are passed on from the mother to the offspring. As our system does not contain any such genotype, we tested whether such experiments are positive with labeled, positively charged peptide (Cy5-R30) as a model. The R30 is homogenously distributed in the coacervate during the reaction cycle. Towards the end of the cycle, the R30 remains in the offspring droplets until they eventually dissolve (Fig. ). In other words, positively charged molecules like R30 are passed from the mother droplet to the offspring. |
66afd279c9c6a5c07a0101fd | 10 | To synthesize de novo life, we require compartments that can undergo Darwinian evolution. Darwinian evolution is impossible without an analog to a genotype, i.e., information-containing molecules that affect the traits of the compartments and are passed on between generations. The genotype and the compartments must be replicated to ensure that daughter compartments receive these information-containing molecules without dilution over generations. Mutations in the replication process are also needed to increase the genotype space. Finally, selection pressures can select fitter mutants over less fit ones. |
66afd279c9c6a5c07a0101fd | 11 | Our work addresses one step in this vision towards de novo lifethe division of the non-equilibrium compartments into daughter compartments. The division we showed is environmentdependent-when the environment is almost depleted of fuel, the mother droplet releases its offspring. It is questionable how "self" the self-division is in this scenario, but we argue that self-division of life as we know it is also strongly environment dependent-cells only divide when fed with sufficient nutrients. Conceptually, the division we demonstrated is closer to fungal sporulation-a mechanism by which fungi spread offspring during harsh conditions (in our case, starvation) to respawn when conditions are more favorable (in our case, when fuel is abundant). The resulting offspring in our work has no genotype, i.e., information containing molecules passed on from the mother compartment. We envision that such a genotype can come from a selfreplicating molecule that, during the short lifetime of the mother, is replicated at least as often as the number of offspring droplets produced. Given the roughly 12 droplets produced per mother droplet, that would imply one molecule of "genotype" needs to undergo four replication cycles during the roughly 10-minute lifetime of the mother to ensure that, statistically speaking, each daughter could receive one molecule of "genotype". Thus, replication should be fast, on the order of a few minutes per replication cycle. These replicating molecules should affect the behavior of the compartments, a minimal analog of genotype-phenotype mapping. Our work already offers a glimpse of how that could be possiblelong polyanions induce the ability to produce offspring. Put differently, if a replicating molecule produces long polyanions (or is a long polyanion itself), it can change the droplet phenotype by enabling its division. Another mechanism could involve the replicator extending the lifetime of the compartment so that it survives until the next fueling round. |
66afd279c9c6a5c07a0101fd | 12 | We studied the mechanism of the fragmentation of active droplets and tested different parameters to tune the fragmentation. We verified that the long PSS remains inside slightly longer than the coacervate droplets than the short PSS, providing the long PSS enough time to mature inside the coacervate droplet to form denser domains, which are then liberated. Adding a second batch of chemical fuel at the right time can rescue the offspring. Their growth is faster than new nucleated droplets, giving them an advantage. Our findings enable a synthetic, self-sustaining protocell to undergo growth, division, and replication by introducing a set of selfreplicating molecules. The replicator should be designed so that the fragments that contain a replicator survive longer than the ones without. The longer-lived fragments can survive longer until the next batch of fuel is needed. Such coupling between a primitive genotype and phenotype could lead to Darwinian evolution in a synthetic system. |
619e1890802991036ef0e346 | 0 | Kernel and neural network (NN) based machine-learning (ML) methods have in recent years become established as an essential addition to the toolbox of computational chemistry. In particular, ML-based interatomic potentials have had great success in providing energies and forces with quantum mechanical accuracy at a fraction of the cost of first-principles calculations. To achieve size-extensivity and a linear computational scaling with system size, these ML potentials typically rely on a local representation of atomic environments and consequently assume that the energy can be decomposed into local atomic contributions. This simple idea has led to a strong focus of chemical ML research on developing sophisticated representations of local atomic environments and, relatedly, NN architectures that directly embed atoms in their neighborhood. At the same time, it is clear that the assumption of locality does not hold for all systems to the same extent. Indeed, strongly polar or ionic systems display very long-ranged Coulomb interactions. Even for a fairly unpolar (e.g organic) system, the locality of the energy does not necessarily imply that other electronic properties are similarly local. In particular, electronic properties such as molecular orbital energies or dipole moments can break locality assumptions. Consequently, such properties tend to be more challenging to predict with purely data-driven ML methods. Beyond this methodological challenge, dipole moments are an intrinsically interesting target as they govern the asymptotic decay of interactions of neutral molecules and their absorption cross-sections in vibrational spectroscopy. |
619e1890802991036ef0e346 | 1 | A promising route to overcome the limitations of local ML models is to include known physical interactions explicitly. For example, a description of long-range electrostatics can be obtained by learning atom-centered charge distributions (e.g atomic charges, dipoles or partitioned electron densities). A prominent recent example of this is the MuML dipole model of Veit et al., which uses atomic charges and atom-centered dipoles to predict molecular dipole moments. This idea takes advantage of the fact that the charge distributions around atoms can be predicted with reasonable accuracy from local environments, even if their interactions are long-ranged. While this solves some of the issues of local interatomic potentials, there are also significant downsides: Firstly, charge conservation of the overall system is generally not ensured, and secondly non-local charge transfer (e.g through conjugated π-systems) is not captured. These issues can be fundamentally addressed by switching the target of the ML model: Instead of predicting the charge distribution directly, one can predict a charge dependent energy expression. The charge distribution is then obtained by minimizing this energy expression under the constraint that the charge is conserved. This idea is closely related to classical charge equilibration approaches like QEq. In this manner, charge conservation is rigorously ensured, the description of non-local charge transfer is enabled and a simple route to analytical derivatives is provided through a Hellmann-Feynman-like approach. The advantage of this approach, compared to directly predicting the charge density, can perhaps be understood in analogy to the choice of initial guess in density functional theory (DFT) calculations: While it is common practice to construct the initial guess from a superposition of atomic densities, it has been found that the superposition of atomic potentials yields a significantly improved starting point. So far only few examples of ML-based charge equilibration models have been reported, however. Most notably, Goedecker and co-workers applied a NN-based QEq model to ionic crystals. The corresponding models were trained to predict the energies and forces of reference DFT calculations, using the predicted partial charges merely as an intermediate quantity. More recently, Behler, Goedecker and co-workers combined this approach with local NN potentials for the description of organic molecules and MgO surfaces. Here, the charge equilibration models were trained on partial charges from reference DFT calculations. |
619e1890802991036ef0e346 | 2 | These models are directly trained on molecular dipole moments and thus avoid the ambiguity associated with choosing population analysis or projection approaches required in other methods. A closed-form linear algebra expression for training kQEq models is derived and their accuracy is benchmarked on the prediction of molecular dipole moments. Finally, limitations and possible extensions are discussed. |
619e1890802991036ef0e346 | 3 | Different conventional (i.e. non-ML) charge equilibration and electronegativity equalization methods have been proposed in the literature. In the derivation of the charge equilibration approach we largely follow the formalism of Goedecker and coworkers, which is in turn based on the QEq method of Rappé and Goddard. In this context, QEq can be understood as a kind of semi-empirical, orbital-free density functional theory, where the electron density ρ(r) is expanded as: |
619e1890802991036ef0e346 | 4 | where, ρ 0 (r) is a reference density (here the superposition of isolated atom densities) and δρ(r) is a fluctuation term, which describes charge transfer and polarization in the interacting system. We expand δρ into a linear combination of normalized 1s Gaussians centered at the atomic positions r A and of width α A δρ(r) |
619e1890802991036ef0e346 | 5 | where N is the number of atoms and q A are the expansion coefficients. Note that we use the negative of the expansion coefficients q A here, so that these can directly be interpreted as atomic partial charges. With this approximation, the electron density is completely defined via the charges q A and it remains to find their optimal values. |
619e1890802991036ef0e346 | 6 | Here, E 0 is a charge-independent reference energy, which we set to zero throughout. The second term (labeled 'Site-Energy') is the well-known second-order Taylor expansion of the atomic energy with respect to the charge, with the atomic electronegativity χ A and the hardness η A . It provides the energetic contribution incurred by adding or removing electron density from a given atom. The third term (labeled 'Coulomb-Integral') is the classical Coulomb potential of the fluctuation density δρ. This integral can be computed analytically, using the definition of δρ (see Eq. 2): |
619e1890802991036ef0e346 | 7 | which makes it explicit that E[ρ] only depends on the charges q A . We may therefore equivalently use the notation E(q 1 , ..., q N ). Note that this equation has the familiar form of the original QEq formulation, with the slight difference that the hardness parameter in QEq implicitly includes the electrostatic idempotential 2γ AA √ π , whereas here η A only describes the non-classical (e.g. exchange-correlation) contributions to the hardness. |
619e1890802991036ef0e346 | 8 | Given the definitions of ρ and E[ρ], we now search for the density that minimizes the energy functional under the constraint that the total number of electrons is conserved. From the definition of δρ, it can be seen that this is equivalent to the constraint that A q A = Q tot , with the total system charge Q tot . This can be achieved by defining the Lagrangian |
619e1890802991036ef0e346 | 9 | The QEq approach as outlined above only requires three parameters, namely the electronegativity (χ A ), the non-classical contribution to the hardness (η A ) and the atomic size (α A ) for each species in the system. As a flipside of this elegant simplicity, the accuracy and transferability of the QEq method is limited, however. In the kernel Charge Equilibration (kQEq) method proposed herein, we follow the basic idea of Goedecker and coworkers to overcome this limitation. This is achieved by allowing the electronegativity of an atom to change as a function of its chemical environment. Importantly, taking advantage of the fact that both QEq and Kernel Ridge Regression (KRR) are formulated as linear problems, we obtain a closed-form expression for training these models. |
619e1890802991036ef0e346 | 10 | where k is a kernel function, p A is a representation vector that encodes the chemical environment of atom A, w B is a regression coefficient and N train is the number of atoms in the training set. We use the SOAP kernel and representation vector, which are widely used in the construction of interatomic potentials and as descriptors of local environments. We refer to the original literature and a recent review for a detailed account of the corresponding theory and implementation. In general, the kernel function measures the similarity of chemical environments and is defined as: |
619e1890802991036ef0e346 | 11 | However, the choice of a partial charge model is necessarily somewhat arbitrary and does not guarantee an accurate description of electrostatic interactions. We therefore instead use molecular dipole moments µ as a reference, which are unambiguously defined physical observables for finite systems and define the leading order term of molecular interactions in the long-range limit. |
619e1890802991036ef0e346 | 12 | where σ is a regularization hyperparameter and the term w T Kw comes from the use of Tikhonov regularization in a kernel regression model. Note that we use the simplest form of regularization, with a single parameter σ to control for overfitting. In principle different regulatization strenghts could be used for each training point (e.g. proportional to the dipole magnitude). |
619e1890802991036ef0e346 | 13 | The above equations are formulated for a single kQEq problem (i.e. a single molecule or system). In practice we train on multiple systems at once. This can still be achieved with a single linear algebra equation by using blocked matrices for A and R, and by concatenating the dipole vector elements of all training systems into a single vector. |
619e1890802991036ef0e346 | 14 | Up to now, an environment-dependent description of the atomic electronegativity χ A is defined, which can be learned from data. It remains to specify the non-classical contribution to the atomic hardness η A and the atomic radius α A for each element. Herein, we choose these by very simple heuristics: α A is set to be proportional to the original QEq radius of the element in question. These radii are tabulated for all elements up to Lawrencium (Z=103). Empirically, we found that scaling these radii with single global scaling parameter s at = 0.75 yields satisfactory results. Similarly, the non-classical hardness parameter η A is set to zero throughout, as we found that this yields robust models while keeping the empiricism of the method as low as possible. These choices are quite simplistic and further optimization would certainly be possible. As shown below, already these simple defaults provide highly accurate results for the investigated systems though. |
619e1890802991036ef0e346 | 15 | The main hyperparameters to be considered for SOAP are the cutoff radius r cut within which the neighborhood is expanded and the broadness of the Gaussians used to smear out the atomic positions (σ atom ). The choice of these lengthscales governs the range in which the environment of an atom affects its electronegativity and how sensitive it is to changes of the atomic positions. In the following, we keep the ratio between these parameters constant (σ atom = rcut 8 ) so that for larger cutoffs, the atomic positions are smeared out more strongly. The idea of keeping this ratio fixed is based on the fact that the expressiveness of a given atom-centered basis set is limited by the number of basis functions, meaning that it can either provide a high-resolution picture that is short-ranged or a lower-resolution picture that is longer-ranged. Alternatively, one could increase the number of basis functions for larger cutoff radii instead, but this would lead to a significantly increased computational cost. The particular constant of proportionality we use was found to work well empirically. |
619e1890802991036ef0e346 | 16 | To quantify the performance of the kQEq models for predicting molecular dipole moments, we use two complementary metrics. On one hand, we use the mean absolute error (MAE) of predicted absolute dipole moments. This is a common measure of accuracy, which allows direct comparison with previous models. Additionally, we use the root mean squared regularized relative error (RRMSE) as used by Hait and Head-Gordon in ref. [44]. This metric is defined as |
619e1890802991036ef0e346 | 17 | , with an arbitrary threshold of 1 Debye that discriminates between small and large dipole moments. In this way, a seamless transition from absolute error (for small dipoles) to relative error (for large dipoles) is achieved, which is necessary since the pure relative error is otherwise severely distorted towards systems with small dipoles. |
619e1890802991036ef0e346 | 18 | As a first benchmark we trained kQEq models for predicting dipole moments of organic molecules. As reference data, the dipole moments of 7500 random molecules from the QM9 database were calculated at the PBE0/def2-TZVP level, using ORCA (data provided in the SI). This set spans a wide range of small to medium sized molecules containing the elements C, H, N, O and F. From these structures, we randomly selected a validation set (used to optimize the regularisation parameter σ) and a test set of 1000 molecules each. The training sets used below were drawn from the remaining 4000 molecules. Figure depicts the learning curves of different kQEq models, using a range of SOAP cutoffs. For comparison, we also fitted a conventional QEq model to the same data. |
619e1890802991036ef0e346 | 19 | All kQEq models clearly outperform the conventional QEq approach, underscoring the benefit of the additional flexibility obtained by using environment dependent electronegativities. Furthermore, it can be seen that the kQEq models improve continuously when given more data, whereas the MAE of the conventional approach quickly saturates. The model with the smallest SOAP cutoff used here (1.7 Å) shows the best performance for small training sets but stops improving when training on larger sets. Meanwhile, the larger cutoffs we tested (2.6 Å and 3.5 Å) continuously improve and reach an excellent accuracy of 0.15 D (compared to an intrinsic standard deviation of ca. 2 D). |
619e1890802991036ef0e346 | 20 | The latter is particularly evident when considering the RRMSE, which shows that this model consistently improves the relative error when increasing the training set, whereas some of the other models improve the performance on total dipole moments at the expense of the relative error (i.e. by describing small dipole moments less accurately). |
619e1890802991036ef0e346 | 21 | Overall, these results show that the physical description of long-range contributions in kQEq allows the use of rather small cutoffs for the ML part, effectively focusing on the nearest neighbors. This is beneficial both in terms of transferability of the models and the This indicates that kQEq does a good job of partitioning contributions into long-ranged physical terms and a short-ranged ML model. |
619e1890802991036ef0e346 | 22 | To put this performance into perspective, we compare these models to two recent kernel ML models that are specifically tailored to predicting dipole moments, namely the operator ML approach of Christensen, Faber and Lilienfeld and the aforementioned MuML model of Veit et al.. (see Fig. ). The former uses a modified variant of the Faber-Christensen-Huang-Lilienfeld (FCHL) representation (FCHL ), which can incorporate the response of the ML model to applied electric fields and thus provides a physically rigorous and equivariant route to predicting dipole moments. Meanwhile, the latter uses a decomposition of the total molecular dipole into atomic monopole and dipole contributions, using the equivariant λ-SOAP approach. For reference we also include the learning curve of a naive FCHL model, which simply predicts the total dipole moment as a scalar (taken from ref. [49]). As already discussed in reference [49], the FCHL model is a significant improvement over the scalar approach. It also significantly outperforms conventional QEq for all but the smallest training sets. Meanwhile the MuML and kQEq models display remarkably similar learning curves and represent a further improvement over FCHL . Note that for consistency the QEq and kQEq models in Fig. were trained on the same reference data (calculated at the B3LYP level) as the MuML model from ref. [20]. |
619e1890802991036ef0e346 | 23 | It is also instructive to consider the accuracy of the reference methods themselves. As mentioned above, Hait and Head-Gordon used the RRMSE to benchmark density functional methods against high-level Coupled Cluster references. According to this metric, the best kQEq model reaches an accuracy of 8.1% on the dipole moments of QM9. This can be compared with the reported errors of popular hybrid functionals like PBE0 (5.2%) and B3LYP (7%). However, it should be noted that the benchmark in ref. [44] focuses on very small molecules and includes spin-polarized systems, so that this comparison should not be overinterpreted. Nonetheless, it indicates that kQEq models approach hybrid DFT accuracy at a much reduced cost. |
619e1890802991036ef0e346 | 24 | Having established the high accuracy of the kQEq predicted dipole moments, we next turn to the predicted charges themselves. It is well known that the electron density cannot be unambiguously partitioned into atomic partial charges. Consequently, there is no way to objectively measure the quality of such partitionings. Indeed, there is a fundamental tension between describing the local charge density around each atom versus global properties such as dipole moments or electrostatic potentials, when approximating a continuous charge density with atom-centered spherically symmetric charges. Nonetheless, it is worth considering whether the predicted charges are reasonably intuitive and how they compare to standard population analysis schemes like the one of Mulliken or restricted electrostatic potential fits like the ChelpG scheme. Here, a double SOAP model trained on 2000 training configurations (at the PBE0 level) is used to provide the kQEq charges (see Fig. ). |
619e1890802991036ef0e346 | 25 | Figure depicts the correlation of Mulliken and kQEq charges for all atoms in the test set. While there is a moderate overall correlation between the charges, there are also clear differences. In particular, we see two general trends. First, kQEq charges appear in a broader range (-1 to 2.5) while Mulliken charges lie between -0.5 and 0.5. Second, kQEq charges display a much more pronounced clustering into functional groups. This is particularly evident for oxygen. While the Mulliken analysis assigns similar charges to ether and carbonyl oxygen atoms, kQEq predicts the carbonyl groups to be significantly more polar. Similarly, the polarity of nitrile and fluoride functional groups is much higher in kQEq. |
619e1890802991036ef0e346 | 26 | A similar comparison is shown for ChelpG charges in Fig. . Compared to the Mulliken charges, a somewhat better correlation with kQEq is observed and the charges fall into the same range. In particular, the large differences for polar functional groups (e.g nitrile or carbonyl functional groups) are not observed here. Nonetheless, large deviations can be seen in other cases, particularly for Fluoride functional groups. |
619e1890802991036ef0e346 | 27 | Two illustrative examples of these differences are shown for 1-(2-Methylcyclopropyl)ethanone and 2-Fluoropyrazine in Fig. (both of which are not part of the kQEq training set). At first glance, the partitionings are qualitatively similar, meaning that the signs of most charges match. However, kQEq and ChelpG predict significantly larger absolute charges than Mulliken. In the case of 1-(2-Methylcyclopropyl)-ethanone, this is particularly evident for the carbonyl group, which is much more strongly polarized. Importantly, these differences have strong consequences for the overall description of the molecular electrostatics. Indeed, the dipole moment calculated from the Mulliken charges points in the opposite direction of the actual dipole. In contrast, the ChelpG dipole is well aligned with the reference, but too small in magnitude. Finally, the kQEq charges provide an accurate prediction of both the absolute dipole moment and its direction. The case of 2-Fluoropyrazine is particularly interesting because it illustrates the observed discrepancy between the charges assigned to C-F groups. Here, kQEq predicts the largest absolute charges on the corresponding C and F atoms. Interestingly, the Mulliken and ChelpG charges nonetheless overestimate the molecular dipole moment, while the kQEq prediction provides an excellent fit. This is because the large positive charge on the carbon atom stems not only from charge transfer to Fluorine, but also to the adjacent Carbon and Nitrogen atoms, which partially counter-balance the polarity of the C-F group. |
619e1890802991036ef0e346 | 28 | The generally poor correlation between the charges obtained with different schemes raises some questions about the different roles and interpretations partial charges can have. On one hand, they can reflect a local partitioning of the electron density, as in the case of Mulliken, Hirshfeld or Bader charges. On the other hand, they can reflect the electrostatic potential on the surface of a molecules, as in the case of ChelpG and related schemes. While the latter is arguably less arbitrary (as it is directly tied to a physical observable) it has well-known issues with assigning meaningful charges to atoms that are "buried" in bulky molecules. |
619e1890802991036ef0e346 | 29 | The kQEq model proposed herein does not neatly fit into these categories. Firstly, it is not a charge partitioning scheme but an ML model. Mulliken and ChelpG charges can only be obtained by running a full DFT calculation, whereas the kQEq prediction is much cheaper. Secondly, while kQEq models are trained to reproduce molecular dipole moments, the charges themselves are obtained by minimizing the electrostatic energy expression in Eq. 5. Consequently, the Coulomb interactions between partial charges provide an important physical constraint on how the charge distribution is approximated. |
619e1890802991036ef0e346 | 30 | A well-known drawback of the conventional QEq approach is that it suffers from an delocalization error akin to that observed for local density functionals. This is e.g evident in the fact that QEq models incorrectly dissociate molecules into partially charged atoms, since electronegativity differences between isolated atoms lead to spurious charge transfer. While kQEq could in principle cure this particular pathology (by assigning the same electronegativity to all isolated atoms), the more general delocalization tendencies of QEq will likely be inherited by kQEq to an extent. In this section we explore this issue by testing the performance of kQEq for the organic polymer chains dataset introduced by Veit et al. (see Fig. ). |
619e1890802991036ef0e346 | 31 | This dataset consists of two types of structures. On one hand, glycine polypeptides in the α-helix and β-strand configurations are considered. On the other hand, polyenoic amino acids and n-amino carboxylic acids are included, which consist of a carboxylic acid and an amine group separated by a conjugated double bond or alkane spacer, respectively. Each of these systems shows characteristic changes of the total dipole moment as the polymer length increases. For the polypeptides, each additional amide bond is itself polar, so that the total dipole increases approximately linearly with the system size. However, the precise behaviour depends on the spatial orientation of these bond dipoles and their interactions, so that the α-helix and β-strand configurations show different scalings. The polyeonic amino acid chains also display a linearly increasing dipole moment. In this case this is not due to the addition of polar bonds, however, but due to the polarization of the delocalized πelectrons in the spacer. Finally, the dipole of the n-amino carboxylic acids remains constant upon increasing the chain length, since no polar bonds or delocalized electrons are present. |
619e1890802991036ef0e346 | 32 | We again use a "double SOAP" kQEq model trained on 2000 randomly drawn QM9 molecules for comparison. This model has a mixed performance for this test. For polyenonic and n-amino carboxylic acids, the performance is very good. This is both in terms of the qualitative features (linear increase in dipole moments, vs. quick saturation) and the quantitative agreement. Furthermore, kQEq is a strong improvement over conventional QEq here. In contrast, the performance for the polypeptides is less satisfactory. While the linear trend is correctly captured, the magnitude of the dipoles is significantly underestimated, in particular for the β-strand. This behaviour is analogous to the underestimation of dipoles typically observed with local functionals, due to the delocalization error. While this shows that kQEq does not automatically cure all pathologies of QEq, a clear advantage of a ML approach is that it can be improved with more data. Indeed, by including the longest chains of each type explicitly in the training set, a retrained kQEq model can be generated that captures these trends more accurately, see Fig. . Potentially, an improved extrapolation behaviour could be obtained by using a more flexible expression for the site-energy in Eq. 3, effectively tackling the problem at its root. Figure also shows the best MuML model from Ref. [20] for comparison. This model performs quite well overall, in particular for the polypeptide systems, with slightly worse performance for the carboxylic acids. In this context, it is worth discussing the fundamental differences between MuML and kQEq. The former uses a fully local decomposition of the overall dipole moment. This works very well for situations where the charge distribution of a large system is essentially just a sum of smaller fragments (as for the polypeptides). In contrast, long-range charge transfer and polarization effects cannot be described by such a model, which is reflected in the underestimated dipole moments of longer polyenoic amino acid chains. QEq and kQEq are in principle able to describe such non-local effects. The QEq curve for n-amino carboxylic acid reveals that this is not necessarily an advantage however, as large unphysical charge transfer is predicted in this case. In other words, a purely local model will generally lead to more systematic and less dramatic failures than a poor non-local one. Fortunately, kQEq and related methods now provide a framework for sophisticated non-local charge equilibration models. |
619e1890802991036ef0e346 | 33 | More generally, it should be noted that kQEq is not primarily intended as a stand-alone molecular dipole model. Since it is based on an energy functional, it can be used to model long-range electrostatic interactions in combination with local interatomic potentials. Indeed, it remains an open question whether dipole moments alone provide enough information for this purpose. Fortunately, the current approach can easily be expanded to also include higher moments, electrostatic potentials or reference partial charges. This will be explored in future work. |
619e1890802991036ef0e346 | 34 | In this work, we introduced kQEq, a kernel-based approach to charge equilibration in molecules. In contrast to conventional charge equilibration methods like QEq, a data-driven, environment-dependent description of atomic electronegativities is introduced. kQEq models trained on molecular dipole moments display excellent performance, en par with or better than state-of-the-art kernel models, specifically tuned to predicting dipole moments. The formalism presented herein opens the door towards physics-based kernel ML models for predicting atomic charges, to be used in combination with reactive interatomic potentials. Importantly, the presented approach is quite general and can be extended to other fit targets (e.g quadrupole moments and electrostatic potentials) and to more flexible density representations (e.g using atom centered dipoles in addition to partial charges). |
619e1890802991036ef0e346 | 35 | It should also be noted that QEq-based frameworks are likely not equally well suited for different kinds of materials. In principle, one would expect the best performance for metallic systems, where charges are strongly delocalized and mobile. In contrast, polar insulators or interfaces may be less well described due to the intrinsic delocalization error of charge equilibration models. More general site-energy expressions could be developed to overcome these tendencies. Ideally, such developments will ultimately converge with recent developments in ML-based DFT, yielding a new generation of orbital-free density functionals. Work in this direction is ongoing in our groups. |
67cf4fb281d2151a02e8e0b7 | 0 | Polyepoxides are increasingly used to replace heavier materials in structural roles, bonding agents between dissimilar substrates, and matrices for high-performance composite materials, relevant for military and commercial applications. Physical (e.g., roughening) and chemical (e.g., coupling agents) surface treatments dictate, to a great extent, the durability and effectiveness of these interfaces, together with environmental curing conditions, such as temperature, atmospheric composition, mechanical stress, and relative humidity. Therefore, it is crucial to identify any changes at the poly epoxy-inorganic substrate interfaces that can become the source of failure, ideally in real-time. This study aimed to investigate how the quality of a polyepoxy/inorganic substrate interface changes (degrades) during cure and aging under confinement using the unique in situ interface imaging and force measuring capabilities of the Surface Forces Apparatus (SFA) via multiple beam interferometry. |
67cf4fb281d2151a02e8e0b7 | 1 | In this work, we used a stoichiometric mixture of bisphenol-A diglycidyl ether (BADGE) and amineterminated polyoxypropylene glycol (JeffAmine D-230) as a polyepoxy model confined between smooth alumina substrates as an inorganic substrate model, in combination with an SFA under different relative humidity environments. The SFA and alumina fabricated surfaces allowed us to conduct real-time surface imaging and refractive index changes of the confined polyepoxy using fringes of equal chromatic order (FECO) technique. Simultaneously, we monitored curing strains and stresses from prior to gel-point conversion to fully cured over 48 hrs. |
67cf4fb281d2151a02e8e0b7 | 2 | Results support the following: 1) low imposed tensile forces favor complete cure of the polyepoxy at various relative humidity environments, and molecular forces dominate the system during curing; 2) at high imposed tensile forces, regardless of relative humidity, no complete cure was ever observed unless the load (force) experience by the polyepoxy consisted of a compressive load; 3) upon exposure of the polyepoxy to high relative humidity, water diffusion takes place within seconds, increasing the degree of plasticization, preventing cure. Finally, to provide a complete and detailed quantitative picture of the adhesion and failure mechanisms of the polyepoxy-inorganic substrate system studied, we complemented with destructive micro-tensile tests and failure analysis imaging, revealing mainly cohesive (polymer-polymer) rather than adhesive (polymer-substrate) failure modes. Overall, we present a new interface quality measurement method to investigate polyepoxy-inorganic substrate systems expected to be of interest to various technological applications. |
67cf4fb281d2151a02e8e0b7 | 3 | Bisphenol-A diglycidyl ether (BADGE) and poly(propylene glycol) bis(2-amionpropyl ether) (JeffAmine D-230) were purchased from Sigma-Aldrich and used as received. BADGE/JeffAmine was thoroughly mixed stoichiometrically by volume (2.42 mL BADGE /1 mL JeffAmine) for at least 10 minutes with visual confirmation of homogeneity, followed by a degassing step of 5 minutes to remove any air bubbles, enabling reactive groups in each component to fully react with each other, resulting in an optimal cure. The resulting mixture was then deposited on different testing substrates. |
67cf4fb281d2151a02e8e0b7 | 4 | We explored three different surface configurations while designing the aluminum oxide (alumina) surface for use in subsequent experiments. The three surface configurations' fabrication and optical performance are described in S1.1 Surface configurations, nanofabrication, and FECO performance section. Detailed X-ray photoelectron spectroscopy (XPS) for chemical analysis, atomic force microscopy (AFM) for surface roughness, and ellipsometry for the determination of the refractive index of the aluminum and alumina films are provided in the S1.2 Surface chemical analysis and S1.3 Film surface roughness and refractive index analysis sections. Additional characterization consisted of surface stability during curing and aging experiments. The final alumina surface chosen for experiments was stable for at least one week of aging under the experimental conditions tested (humidity 0-100 %RH and immersed, temperature 21-60°C). Furthermore, it was suitable for producing fringes of equal chromatic order (FECO). The surfaces consisted of a 2 nm Ti adhesion-promoting layer, followed by a 42 nm reflective gold layer, followed by a second 2 nm Ti adhesion-promoting layer, and finishing with a 70 -80 nm alumina layer deposited via ion beam deposition (IBD). We call this surface configuration 3. Figures and show the FECO obtained with the common back-silvered mica substrates and the FECO obtained from surface configuration 3. |
67cf4fb281d2151a02e8e0b7 | 5 | A Surface Forces Apparatus (SFA) 2000 (SurForce LLC, Goleta, CA, USA) was used with a normal force measuring configuration. Spring constants were changed for each reported condition using either a weak, intermediate, or stiff spring with spring constants of kweak ≈ 1,000 N/m, kintermediate ≈ 2,500 N/m, or kstiff ≈ 13,000 N/m, respectively. Humidity was monitored with a Rotronic HC2A-IC industrial humidity probe mounted to the SFA chamber. Details on the multiple beam interference (MBI) technique and force-distance measurement technique of the SFA can be found elsewhere. |
67cf4fb281d2151a02e8e0b7 | 6 | We performed quantitative and qualitative wetting analyses of the BADGE/JeffAmine mixtures and single components. Quantitative wetting studies were carried out by measuring the contact angle of BADGE/JeffAmine droplets on various substrates. In contrast, qualitative studies were carried out by depositing large volume droplets (2 mL) and spin coating substrates, which were visually inspected during cure. The final BADGE/JeffAmine droplet on alumina configuration/deposition protocol was based on the wetting results below. |
67cf4fb281d2151a02e8e0b7 | 7 | Measuring the BADGE/JeffAmine wettability of Al surfaces. We measured the contact angle of BADGE/JeffAmine on aluminum and water-oxidized aluminum (2 hrs at 65°C, as described in the SI) to better understand the wetting behavior of the BADGE/JeffAmine polymer using a videobased optical contact angle measuring system. We observed that the cosθ (i.e., the surface energy) value is high (corresponding to a low contact angle), indicating a philic (wetting) solution, as shown in Figure . Interestingly, the wettability of the fully water oxidized aluminum layer was higher (or better), initially suggesting a more favorable substrate than aluminum. It can be due to a contribution of surface roughness and/or porositiy changes of the alumna film formed. Both surface characteristics, surface roughness and porosity, are known to control wetting. Therefore, we conclude that the apparent increased BADGE/JeffAmine wettability of fully water-oxidized Al is due to pores, assuming a Wenzel wetting state. After complete cure (48 hrs), we observed that the initial wetted area decreased, leaving behind a precursor film that suggests autophobic de-wetting. This can be partially explained by the fact that each of the two components that compose the mixture possesses different refractive indices (BADGE = 1.5735; JeffAmine = 1.4466) and, therefore, have distinct substrate adsorption kinetics (determined by the van der Waals interactions). Furthermore, the crosslinking of the polymer mixture decreases the volume and, consequently, the area. |
67cf4fb281d2151a02e8e0b7 | 8 | Wetting with large volumes and spin coating of BADGE/JeffAmine: Spin coating introduces significant stresses that de-stabilizes the film. Attempts to spin-coat different surfaces failed due to the polymer immediately de-wetting the substrate, as exhibited by the film retracting into droplets within seconds to minutes after termination of the spin program. This behavior is, in part, driven by the residual stresses stored in the films due to their viscoelasticity. These residual stresses become a source of destabilization for the BADGE/JeffAmine films, accelerating the de-wetting process, similar to what has been reported for polystyrene spin-coated films. When forcing the droplet to form a film, the non-equilibrium cosθ value becomes 1 and will tend to its equilibrium contact angle (small but finite) unless locked by thermal activation (e.g., hotplates). Despite having identified time, temperature, and spin coating parameters that allowed a 'locked' BADGE/JeffAmine film on the curved substrates, the film thickness remained an uncontrolled parameter. Furthermore, the pre-cure of the films would introduce a possible decrease in BADGE/JeffAmine adhesion when coming into contact with the second Al surface, resulting in lower adhesion measurements/values. |
67cf4fb281d2151a02e8e0b7 | 9 | Self-centering of small droplets on smooth surfaces. A tiny BADGE/JeffAmine droplet (< 0.01 µL) was deposited on alumina to overcome the de-wetting difficulties encountered during spin coating. The droplet was deposited by gently touching the alumina surfaces with a dispensing pipette. The resulting droplets are smaller in diameter than the maximum lateral field of view (≈ 550 µm) that the used SFA setup could capture without lateral scanning. This dimension allows for simultaneous FECO visualization of the BADGE/JeffAmine contact (see Monitoring alumina-BADGE/JeffAmine interface during cure) and a region just outside the BADGE/Jeffamine region (i.e., air or water). When the droplet was deposited on the upper surfaces and then slowly brought into proximity to the second (lower) surfaces, the droplet self-centered (i.e., slid) to the point of closes approach (PCA) of the crossed-cylinder configuration, driven by capillary forces. A shows a 250 µm diameter droplet of BADGE/JeffAmine at the moment of bridging the second surface (tsliding = 0 s). Velocity profiles at different locations across the droplet during the self-centering stage are shown in Figure . The velocity profile clearly shows the velocity gradient that the droplet experiences, driven by the front end of the droplet. Sometimes, the BADGE/JeffAmine droplet splits close to the rear end due to the significant velocity differences, reducing the volume that reaches the PCA. |
67cf4fb281d2151a02e8e0b7 | 10 | The SFA technique allows to measure simultaneously and independently the absolute surface separation, D, between the alumina surfaces and the refractive index of the medium, BADGE/JeffAmine mixture, B/JA. These two parameters (D and B/JA), when properly interpreted, are sufficient to monitor the grade to which the alumina-BADGE/JeffAmine interface changes during cure in different environmental conditions. Interface changes can be asperity formation (roughness), wetting or de-wetting (increase or decrease of the alumina-BADGE/JeffAmine contact area, respectively), condensation of vapors at the three-phase alumina- BADGE/JeffAmine/air contact line, 10 diffusion, or crosslinking of the epoxy matrix. All of the aforementioned changes dictate the effectiveness of the metal oxide-epoxy bond, that is, the alumina-BADGE/JeffAmine bond. The FECO sequence of BADGE/JeffAmine cured at 0 %RH is shown in Figure . Schematics show one possible FECO interpretation. Initially, at tcure = 0 hrs, a BADGE/JeffAmine droplet of 375 µm in diameter bridged surfaces at a separation of 8 µm. During the first hrs of cure (up to tcure = 22 hrs), the surface separation decreased to 25%, a point at which fibrillation FECO signatures appeared. The additional FECO discontinuities, Figure , can show this fibrillation (and possible loss of adhesion area). It is important to note that their presence could be initiated several hours before being detected with FECO (FECO is collected from a cross-sectional area of 450 µm and 10-20 µm in width, typically located around the center of the PCA). It is possible to incorporate lateral scanning capabilities to monitor the detection of any changes across the entire area of interest, as tested and described in S2. Monitoring large areas with FECO. In summary, we show the feasibility of tracking phenomena at the interface in situ and in ipsa hora. |
67cf4fb281d2151a02e8e0b7 | 11 | Monitoring strains and stresses during cure. The red curve in Figure corresponds to BADGE/JeffAmine cured at 50 %RH. The cure strain, Ɛcure, decreased continuously during the 48 hrs and shows a typical degree-of-cure profile as a function of time for epoxy glues; that is an initial fast characteristic time (first hour) with a decrease in strain and increase in strain (molecular forces driving the surfaces towards each other during cure), followed by a slower (several hours) characteristic time with a continuing decrease in strain and increase in strain. The abrupt jump at tcure ≈ 15 hrs was caused by drift and migration of the PCA from the imaging field of view. The last curing experiment example is at 100 %RH, black curves in Figure . The spring constant used to monitor and calculate stress was k ≈ 2,500 N/m. Similar to what was observed for the 50 %RH humidity condition, the curve showed two characteristic time scales: an initially fast increase in strain corresponding to a fast degree of cure, followed by slower characteristic time, where the degree of cure happens much slower. The measured cure stress is within the same order of magnitude, between 250-500 kPa. |
67cf4fb281d2151a02e8e0b7 | 12 | The FECO sequence shown in Figure the BADGE/JeffAmine mixture changes as a function of time immediately after injecting 40 mL of DI water to the chamber. Injection of DI to the chamber saturates the box with water vapor, resulting in a 100 %RH curing environment within minutes. A small droplet of BADGE/JeffAmine was deposited on the top mount and brought into contact with the lower surface. Horizontal dark discontinuities (bands) are due to a refractive index mismatch between the BADGE/JeffAmine drop and air media. After approximately 2 minutes, the FECO lines begin to change shape and become diffuse (Figure and). Water diffuses into the BADGE/JeffAmine droplet almost immediately, dramatically changing the local refractive index of the BADGE/JeffAmine, B/JA. This is reflected and visualized in the distortion of the FECO in Figure D-F. It is reported that water uptake of epoxy resins follows a Fickian behavior, in which the initial stage of water uptake is a smoothly increasing function of the square root of time, t 1/2 , and where the thickness of the droplet determines the cross sectional-area through which water diffuses. Therefore, controlling the absolute surface separation as precise as possible is crucial to quantitatively compare environment effects on the cured BADGE/JeffAmine. Furthermore, undesired water diffusion during initial curing leads to water-induced plasticization, as also seen in other systems. Figure . A sequence of FECO fringes shows how water from the water-saturated environment diffuses into the BADGE/JeffAmine droplet within seconds. Time t = 0 is the time when water was introduced to the dry chamber. The water reservoir is located a few centimeters away from the surfaces. The rough shape of the fringe indicated non-uniform mixture of BADGE/JeffAmine and water, since the refractive index of water (Water ≈ 1.33) is significantly less than BADGE/JeffAmine (BADGE/JeffAmine ≈ 1.57). After 375 s the FECO fringes smoothen out again as the mixture becomes more uniform. |
67cf4fb281d2151a02e8e0b7 | 13 | The mica-BADGE/JeffAmine curing experiments over 48 hrs at varying percentages of relative humidity support the following concluding remarks: 1) compliant springs favor complete cure of the BADGE/JeffAmine even in humid environments. In addition, weaker springs allow for molecular rearrangements (e.g., cross-linking, re-orientation, alignment) to monitor the cure forces acting on the system with higher sensitivity. In other words, molecular forces dominate the system during cure. With higher spring constants, regardless of %RH, no full cure was ever observed unless the load (force) experienced by the BADGE/JeffAmine consisted of a compressive load, such as in the case of BADGE/JeffAmine on mica. |
67cf4fb281d2151a02e8e0b7 | 14 | The amount of water present in the environment is consistent qualitatively with what has been reported for epoxy glues, leading to changes in the degree of BADGE/JeffAmine plasticization. That is, water aids in maintaining a viscous behavior even after 48 hrs of cure time, conditions that, if favorable, are enough to fully cure the BADGE/JeffAmine volumes used during this test. Using force springs with controlled variable spring stiffness is needed to monitor high sensitivity cure forces (strains) and will need to switch to high spring constants to allow separation during the tensile test after reaching full cure (high stiffness spring). After each cure conditions, we performed tensile test on the 48 hrs cured BADGE/JeffAmine droplets on alumina substrate. Given the variability of the droplet volumes across experiments and droplet diameter, we consequently normalized the measured forces over the initial droplet area and defined tensile strain, similar to what is described for cure stress and cure strain. |
67cf4fb281d2151a02e8e0b7 | 15 | Figure shows four tensile stress-strain curves, all at 50 %RH, with different imposed external mechanical forces. It is very interesting to note that for BADGE/JeffAmine cured under high external mechanical forces (red curve), the system retained a very viscous behavior, as seen by the extensive strains before failure. The curve was obtained with full motor range until failure. This was not the case for BADGE/JeffAmine cured under low external mechanical forces (inset pink and black), where the system shows linear elastic properties. Separation forces for BADGE/JeffAmine cured at 100 %RH showed a "super-plastic" or very viscous behavior and it was unclear to determine the stress at failure. |
67cf4fb281d2151a02e8e0b7 | 16 | We imaged the BADGE/JeffAmine on the top and bottom substrates for each condition investigated after the 48 hrs of cure and followed by a tensile test (motor and manual, or motor only) with a laser confocal microscope. Imaging allowed us to gain insight into the modes of failure (adhesive, cohesive, or mixed), and gave us a second metric of the droplet diameter (the shape of the FECO fringes being the first metric). It allowed us to determine if the failures were of a brittle (sharp edges) or plastic (e.g., fibrillation) nature, which indicate the degree of cure. Mechanical imposed external forces seem to drastically influence how long the BADGE/JeffAmine droplets take to fully cure. Surfaces mounted on weak springs appeared fully cured and showed primarily brittle fractures with sharp edges. These conditions reached full cure, and we interpret the external mechanical forces to be weak enough to allow the molecular forces to drive optimal crosslinking (molecular alignment, rotation, etc.), as shown in Figures and. Also, when the BADGE/JeffAmine was cured under compressive loads (one test only, Figure ), BADGE/JeffAmine reached a complete cure. The fracture surfaces showed brittle or clean fracture over large areas, accompanied by fibrillation on edges. BADGE/JeffAmine droplets cured on surfaces mounted on stiff springs did not reach full cure regardless of the %RH, and in some instances, failing (full surface separation) before reaching the 48 hrs of cure. These surfaces present significantly more fibrillation, the signature of plastic deformation, and almost no clean surface areas. |
67cf4fb281d2151a02e8e0b7 | 17 | Additionally, environmental conditions (%RH) seem to have a significant effect on the degree of plasticization. Increasing %RH allows the BADGE/JeffAmine to retain viscous behavior under the investigated times and rates, while lower %RH seems to help gain a more elastic behavior. In summary, in all unaged cases, we observe mainly cohesive failure. In conditions such as the ones shown in Figure (stiff spring to maintain constant separation distance and smooth alumina surfaces), it could be argued that the failure mode is mixed (cohesive and adhesive). However, it is very unlikely that no precursor film has formed during the early stages of cure. |
67cf4fb281d2151a02e8e0b7 | 18 | We monitored in ipsa hora and in situ the curing of a model two-component polyepoxy, BADGE/JeffAmine, onto model oxide surfaces, alumina, using the SFA's combined force measuring and interface visualization capabilities. We found that (1) low tensile forces promote complete curing across different humidity levels, dominated by molecular interactions; (2) high tensile forces prevent full curing unless compressive forces are applied; (3) high humidity rapidly induces water diffusion, increasing plasticization and preventing cure. Additional micro-tensile tests and failure analysis imaging revealed primarily cohesive (polymer-polymer) rather than adhesive (polymer-substrate) failure modes. This study introduces a novel method for assessing interface quality in polyepoxy-inorganic substrate systems, which has implications for various technological applications requiring durable bonding and adhesion performance. |
62b682efe84dd1d29902903d | 0 | As our reliance on eco-friendly and renewable energy grows, there is an urgent need to explore various sustainable sources. In this race, solar energy is considered an enormous energy source that can offer a permanent solution to the world's energy crisis. However, a few hours of sunshine in a day, uneven power density distribution on the earth due to geographical area, and low power conversion efficiencies of photovoltaic devices threaten its widespread acceptability. Of several other renewable energy resources, splitting water to generate hydrogen as a fuel and oxygen as a byproduct is the most viable technique. However, in several electrochemical reactions, the energy conversion rate and efficiency are often limited by oxygen evolution reaction (OER) expressed as equation (1) in aqueous media. This reaction proceeds via multiple state reactions involving 4 electron oxidation, thus becoming kinetically sluggish. Therefore, to kinetically accelerate this reaction, the overpotentials (η) need to be lowered. Thus, several OER electrocatalysts exhibiting lower η have been explored so far. 4𝑂𝑂𝑂𝑂 -→ 2𝑂𝑂 2 𝑂𝑂 + 4𝑒𝑒 -+ 𝑂𝑂 2 ; 𝐸𝐸 = 1.23 𝑉𝑉 𝑣𝑣𝑒𝑒𝑣𝑣𝑣𝑣𝑣𝑣𝑣𝑣 𝑅𝑅𝑂𝑂𝐸𝐸 |
62b682efe84dd1d29902903d | 1 | Over the past few decades, numerous 3d transition metal (TM)-based layered oxide cathodes have been explored to act as OER catalysts. It has also been reported that the properties of a Co-based, efficient OER electrocatalyst, particularly Co3O4 , can be effectively tuned up by lithium insertion/(de)insertion as in LiCoO2. Apart from a single 3d TM containing OER electrocatalysts, a dual 3d TM Fe-substituted LiNiO2 has also been reported to be an efficient electrocatalyst. The higher activity of this doubly doped layered oxide is attributed to the higher oxidation states of TMs and their stabilization in layered structures. So far, several layered oxides have only been looked upon as a suitable host for stabilizing the highly oxidized TM species necessary for effective OER catalysis. Though layered oxides are better suited to stabilize oxidized TM species, it has also been predicted that the electronic state regulation is another avenue to enhance the activity of perovskite electrocatalysts. In this realm, doping is an effective strategy to tailor the oxidation states of TM ions in layered oxides, to effectively promote hole hopping between TM n to TM (n+1) ions. 12. However, the binding energy (BE) of surface oxygen is another vital aspect in OER catalysis, which has rarely been considered in previous reports. Thus, efforts must be substantiated to understand the role of BE of TMs with surface oxygen in the presence of doped cations/anions. Herein we report, for the very first time, a uniquely combined 3d/4d TMs-based cation disordered rocksalt (DRX) cathode material with Na-substitution in the lattice framework by the facile solid-state route. This uniquely combined 3d/4d TMs DRX cathode is believed to entirely change the stacking sequence to that of just 3d TMs cathode, while doping of monovalent cation substitution assists in promoting the oxidation state of TMs. The X-ray diffraction and other characterization techniques reveal that Na-substitution significantly assists in OER. The Na-substituted disordered cathode achieves a small Tafel slope of 67.5 mV dec -1 , η of 270 mV at a current density of 10 mA cm -2 , better than that of the IrO2 benchmark catalyst, and long-term stability over 133 hours. The machine learning aid to search for the minimum energy structure along with density functional theory (DFT) calculations reveals that the oxygen BE is ~6.51 eV in pristine. In comparison, it is much lower (~5.45 eV) in Nasubstituted cathode. The higher activity in Na-substituted DRX cathode is attributed to the facile desorption of O2 from the surface (Scheme 1). Scheme1. Schematic illustration of facile O2 desorption from the surface. |
62b682efe84dd1d29902903d | 2 | A solid-state synthesis route was used to obtain DRX Li1.22Ru0.61Ni0.16O2 and Li1.22Ru0.61Na0.05Ni0.10O2 hereafter denoted as 1 and 2, respectively. The bulk composition was determined using inductively coupled plasma optical emission spectroscopy (ICP-OES), closely following the theoretical composition (Table and Experimental Section). The crystal structure determined using high-power powder X-ray diffraction (HP-PXRD) reveals that the most characteristic XRD peaks (Figure ) in both cathodes are assigned to the C2/c space group of Li2RuO3. The merged peaks ~63° indicate disordered structures of the synthesized cathodes. Compared to 1, the XRD patterns of 2 show the asymmetric peak broadening effect (marked by the green circle in Figure ), a signature of lattice defects/strains/dislocation possibly arising from slightly larger radii of Na + doped in Li slab. Doping-induced lattice defects are expected to enhance electrochemical performances. To elucidate the electronic structures, the core-level XPS data were obtained. The XPS peak fitting data (Figure and Supplementary Discussion S1) reveal that Na doping brought significant changes in electronic structure. Before moving to the discussion of XPS peak fitting, we disclose the BE of Na. The BE value of ~1071.38 eV (Figure ) corresponds to Na + states of Na, indicating that Na has been electro-positively doped in the crystal framework. The Ni 2p (Figure and S1a) possesses two spin-orbit doublets at 855.28 (Ni 2p3/2) and 872.48 eV (Ni 2p1/2) accompanied by two satellites at 861.28 and 879.68 eV. The BE ~ 855.28 eV corresponds to Ni 3+ , whereas deconvolution of Ni 2p1/2 reveals two sets of energy band: 872.48 eV and 873.88 eV for OH -/OOH. This XPS revelation of Ni hints toward the presence of more Ni 3+ species favoring the electrophilicity of adsorbed O, thereby catalyzing the formation of -OOH species from OH -(these steps are considered as rate-determining steps in alkaline media). The electronic structure of Ru in 2 is also different from 1 (Figure ). The Ru 3d5/2 can only be fitted with a doublet (separated by~1 eV), indicating the presence of both metallic (280.68 eV) and oxidized ruthenium on surface (Figure ). This oxidized ruthenium can be inferred as oxides with low BE (281.98 eV) and with high BE (283.78 eV). The BEs at 285.38 and 286.68 eV correspond to RuOx/Ru. Most importantly, XPS results of the O 1s core-level (Figure ), fitted with three components apportioned as oxidative oxygen (O2 2-/O -: BE ~529.58 eV), carbonate/hydroxides/oxyhydroxides species (CO3 2-/OH -/OOH: BE ~531.48 eV), and absorbed water (H2O: BE ~534.88 eV) suggest a relatively more significant number of O2 2-/O-species in 2. This indicates a surplus amount of surface oxygen vacancies compared to 1 (Figure ) with hydroxides (BE ~531.7 eV) and oxygen species (BE ~533.4 eV). Firstly, Na doping has significantly modified the electronic structure, which agrees with previous reports on layered cathodes. Secondly, the excess of O2 2-/O-species on the surface of 2 is beneficial for catalyzing OER reactions in alkaline media. 21 Scanning electron microscopy (SEM) images reveal a distinguishable feature in their morphology (Figure and Supplementary Discussion S2). The particles in 2 show cornered-shaped structures with smaller particles decorating larger particles, thus creating rougher surfaces. These rough surfaces expose multiple active sites facilitating oxygen evolution reactions. A better sintering ability with enhanced particle connectivity is an additional feature brought by Na doping. The lattice fringe distance of 0.475 nm revealed under high-resolution transmission electron microscopy (HRTEM) image corresponds to the disordered structure (Figure ). As expected Na + (marked by the pink arrow) distributes uniformly inside the lattice (Figure ). Few other exciting features that promote OER kinetics can be explored under HRTEM. For instance, defective zones (marked by the yellow arrow) can also be seen in Figure of Na-DRX, whereas no such beneficial features can be seen in 1 (Figure ). These defective zones are termed as stacking fault defects, primarily arising from the differences in ionic radii of Na + and Li + . These defective sites can modulate the electronic structure and tune surface properties of 2, thus optimizing the adsorption energies of OER steps. The autocorrelated image reveals the Na-doping and defects zone (Figure ). |
62b682efe84dd1d29902903d | 3 | The energy-dispersive X-ray spectroscopy (EDS) mapping images show a uniform distribution of elements with a slight increase in elemental O on the surface. To further corroborate the increased oxygen vacancies through Na-doping, the electrochemical oxygen intercalation in 1 and 2 was probed through cyclic voltammetry (CV) performed on a three-electrode system in oxygen saturated 1M KOH electrolyte environment. As illustrated in Figure , redox peaks appear as oxygen ions are inserted into and extracted from the accessible lattice vacancy sites. It is important to note that 2 shows positive-shifted redox peaks (marked by arrows) compared to 1, which is in harmony with the increased BE reflected by XPS. Furthermore, the linear sweep voltammetry (LSV) curve of 2 exhibits a low η of 270/390 mV (on NF) to yield current densities of 10/250 mA cm -2 (Figure and). This catalytic activity is much better than 1 and several other cathode materials deployed for water catalysis (Table ). As the reaction kinetics directly relates to the electrochemically active surface area, we also evaluated double-layer capacitance (Cdl) measurements to compare the electrochemically active surface areas of 1 and 2 (Figure ). The double-layer capacitance 2 (Cdl =1.63 mF cm -2 ) is higher than 1 (Cdl =1.25 mF cm -2 ), which suggests that the former possesses a relatively higher density of active sites for catalytic reactions, thus further boosting the OER activities. |
62b682efe84dd1d29902903d | 4 | To get better insights into OER kinetics, the Tafel slopes of the cathodes was obtained from steady-state polarization curve. Figure manifests that Tafel slope of 2 (67.5 mV dec -1 ) is lower than 1 (72.2 mV dec -1 ) and benchmark IrO2 (86 mV dec -1 ). The lower Tafel slope value of 2 indicates that the Na + substitution into DRX matrix plays a vital role in promoting the electrochemical OER kinetics, thus improving the intrinsic catalytic activity. |
62b682efe84dd1d29902903d | 5 | To evaluate the impact of Na + doping on electrical conductivity in 2, we studied its impedance spectra (Figure ). The semicircle is related to the charge-transfer resistance (Rct) of the cathode materials. The electrochemical impedance spectroscopy (EIS) spectra show that the charge-transfer resistance of 2 (Rct =16.01 Ω) is lower than 1 (Rct = 20.89 Ω) and much lower than the commercial IrO2 (72.8 Ω), indicating a better electron transfer capability across the electrode/electrolyte interface, thereby indicating a superior OER kinetics and an improved overall electrochemical performance (Figure ). |
62b682efe84dd1d29902903d | 6 | The long-term stability of cathode materials is another crucial factor that deters the wide applicability of DRX materials for water splitting reactions. We, therefore, investigated the stability in 1M KOH. The chronopotentiometry results suggest that 2 has superb stability of more than 5.5 days and even beyond without an appreciable increment in overpotential (Figure ). In contrast, 1 shows a tendency to increase in potential even after a few hours. This superb stability of 2 is attributed to the following factors. The Na doping in the lattice increased the oxidation states of Ru and Ni, thereby modifying the electronic structures. As reported earlier, higher oxidation states of TMs in layered structure are beneficial for OER. Additionally, these higher oxidation states can be easily stabilized in DRX layered structures, which is essential for sustained OER reactions. Furthermore, it has also been widely accepted that high-valence transition metal cations facilitate the adsorption of OH-with the catalysts to form adsorbed -OOH species to promote OER reaction steps readily. The above results reveal the enhanced electrochemical performance of 2 compared to not only 1 but also to many other cathode materials (Figure ). It is generally assumed that the oxygen BE at the surface largely influences the OER activity. For this, a series of DFT calculations were conducted to investigate the role of Nadoping on surface oxygen atoms as shown in Figure . The favorable locations of Ni and Na in 2 are found with Genetics Algorithm (GA). DFT calculations established that while Ni atoms prefer to occupy the Ru layer (Figure ), Na atoms prefer to locate in Li layer (Figure ). Surprisingly, the average oxygen Bes in both cases are almost same (∼7 eV), but while it is almost uniform in 1, but 2 shows higher fluctuations. In particular, the lowest BE for oxygen is 5.45 eV in 2 and 6.51 eV in 1. The Na-substitution promotes the OER activity by reducing the BE of some oxygen atoms at the surface and accentuating the facile diffusion of O2 gas, which is consistent with experimental results. The total density of states (DOS) reveals that the electrons' occupancy near the Fermi level between 1 and 2 is different. In addition, the projected DOS of oxygen 2p orbital in 2 (Figure ) is substantially different from that in 1 (Figure ) and shows a shift in O p-band center toward the Fermi level , which leads to better TM-O overlap and creates reduced charge transfer resistance well supported with Figure . In summary, we have enrooted the idea of engaging cation disordered cathode materials for OER application synthesized by a facile solid-state route. Na-doping in 2 solved the issue of instability prevailing in disordered cathodes but also contradicted the former idea that layered structures alone can stabilize TM oxidized species for improved OER performances. Nonetheless, 2 performs much better than 1 and many other layered cathode materials, including dedicated OER electrocatalysts. 2 requires a low η of 270/390 mV to display current densities of 10/250 mA cm -2 and demonstrate superb stability for more than 5.5 days, which is exceptionally better than many of the OER electrocatalysts reported to date. The significant enhancement in the OER performance is attributed to several intriguing factors. First, the disordered structure provides better interaction between Li and TMs, often unavailable in layered cathodes where Li and TMs occupy a fixed position. Secondly, the induction of Na + promoted the oxidation state of TMs (Ni and Ru), favoring OER kinetics. Thirdly, DFT calculations predicted that Na + accentuates the facile evolution of O2 from the surface layer, thereby alleviating the sluggish reaction kinetics often observed in the OER. We believe that our finding of engaging DRX cathode materials for the OER application will pave a novel design strategy for future materials of this family for high-performance electrocatalysts for energy conversion reactions. |
62b682efe84dd1d29902903d | 7 | We opted for a conventional solid-state reaction method with a slight modification to synthesize cathode active materials calcined at 950 °C. Cation disordered rocksalt (DRX) cathode materials precisely Li1.22Ru0.61Ni0.16O2 (1) and Li1.22Ru0.61Na0.05Ni0.10O2 (2), were synthesized. The powders of LiNO3, NaNO3, Ni(CH3COO)2.4H2O, and RuO2 were stoichiometrically calculated and thoroughly homogenized using mortar and pestle for around 10 minutes. Subsequently, the powders were mixed with acetone, followed by ball milling for the next four hours. To avoid loss during high-temperature synthesis, an excess of 5 wt % of LiNO3 and NaNO3 over was taken. The homogenized mixture was then dried overnight in a vacuum oven at 100 °C. The powders were again mortar-pestled and finally, powers were transferred in an alumina crucible for sintering at 950 °C for 12 h at a ramp rate of 5 °C/min in a tube furnace in the ambient atmosphere. Subsequently, the furnace was cooled to room temperature and samples were grounded manually to obtain a fine powder. |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.