• Tidak ada hasil yang ditemukan

for Clean Energy Conversion

N/A
N/A
Protected

Academic year: 2023

Membagikan "for Clean Energy Conversion "

Copied!
113
0
0

Teks penuh

Introduction

Statement of the Problem

The key barrier to these sustainable energy conversions is the development of efficient, selective and stable catalysts. For example, Pt alloy nanomaterials7-12, especially PtNi alloys13-14, have been shown to have improved performance. Similarly, the improved catalytic activity and selectivity was observed in Copper nanomaterials for the reduction of CO to C2+ products16 and in Gold nanomaterials for reduction of CO2 to CO17.

Organization of the Thesis

Predicted structures of the active sites responsible for the enhanced reduction of carbon dioxide by gold nanoparticles. Distributions and partitions of the data are listed in Table B-1 and the training trajectory is shown in Figure B-4. Distribution and distribution of the data is listed in Table B-2 and the training trajectory is shown in Figure B-4(b).

Atomistic Explanation of the Dramatically Improved Oxygen

Introduction

Fortunelli and Goddard (FG) used the ReaxFF reactive force field27 to predict the structure of the delocalized Debe NP and found that starting with 70% Ni led to a porous Pt with significant internal area exposed to the surface. Nanowires (NW) led to 2 nm Pt Jagged NW (J-PtNW) with 50 times faster ORR than current state-of-the-art Pt/C, but they found that all Ni was extracted. Starting with J-PtNW synthesized computationally using the ReaxFF reactive force field, we randomly selected 500 of the more than 10,000 surface sites and performed Quantum Mechanics (QM) calculations on clusters within 8 Å of the surface site.

Results and Discussion

We show in Figure A-7 of the Supporting Information three examples illustrating the correlation between d-OO and 𝐺D at room temperature. The centers of the sampled bridges are shown in VP in Supporting Information Figure A-9. Thus, we evaluated the performance of the entire nanowire by plotting the statistics of our sample on the full NW.

Figure 2-1. Bridge Nanocluster Model. We first cut two spheres of size R around Pt-1 (cyan  atom) and Pt-2 (pink tom), where Pt-1 is specifically for water adsorption and Pt-2 is for  oxygen adsorption
Figure 2-1. Bridge Nanocluster Model. We first cut two spheres of size R around Pt-1 (cyan atom) and Pt-2 (pink tom), where Pt-1 is specifically for water adsorption and Pt-2 is for oxygen adsorption

Conclusion

Nature of the active sites for CO reduction on copper nanoparticles; Suggestions for optimizing performance. Termination of this structure to expose the (100) and (111) surfaces results in sites that are concave or convex with respect to the (100) planes. b) The four types of sites for adsorbed *OCCOH on the surface of the minimal periodic structure. The neural network used in this study is of the type by Behler and Parrinello6.

Identifying Active Sites for CO2 Reduction on Gold Nanoparticles

Introduction

Moreover, it would be quite impractical to use QM with cluster models of 5 to 10 thousand surface areas. To solve this puzzle, we propose here a machine learning strategy that uses QM calculations with clustering models for hundreds of 5 to 10 thousand surface sites of ReaxFF grown NPs, which we iteratively train to achieved 0.05 eV accuracy. Our previous experimental studies of Au3Fe alloy core NPs led to spectacular results, a 100-fold increase in mass activity compared to AuNPs, a 500 meV improvement in overpotential for >80% faradaic efficiency (FE) of CO2 to CO27 compared to Au foil.

Results and Discussion

Among these 100 sites, 96 sites cannot absorb even the important reaction intermediate (HOCO) and the other 4 sites show very high ∆𝐸,+*+. We constrained the ratio of sites from different coordination groups to be equal within each of the three datasets, but the three groups are completely independent. The root mean square error (RMSE) of the validation set reached a minimum at the training epoch of 19000 with the training set RMSE at 0.0563 eV and the validation set RMSE at 0.0591 eV.

In general, the predicted performance of the other surface locations on AuNPs surrounds that of the low-index surfaces, as the nanoparticle surfaces are irregular and disordered. A more detailed description of the model and structure equilibrium is included in B4 of Supporting Information. AR is then the area to the right of the dotted line, as the total area of ​​each plot is normalized to 1.

As defined, the AR (active ratio) is the area to the right of the dashed line for each plot. Here we avoid the exact modeling of the spill process, which requires complex and expensive calculations. To further verify the neural network prediction results, we randomly selected 5 sites from each of the 7 groups and performed DFT cluster calculations (without solution).

These a-values ​​are then mapped back onto the particle to visualize the catalytic activity of the entire surface.

Figure 3-1. Overall Structure of Neural Network.  We use two-body terms(C_2) and three- three-body terms(C_3) to represent the local geometric feature of the selected surface atom (gray  atom) extracted from 8 Å nanocluster model
Figure 3-1. Overall Structure of Neural Network. We use two-body terms(C_2) and three- three-body terms(C_3) to represent the local geometric feature of the selected surface atom (gray atom) extracted from 8 Å nanocluster model

Conclusion

Mechanism and kinetics of the electrocatalytic reaction Responsible for the high cost of hydrogen fuel cells. Selectivity of CO 2 reduction on copper electrodes: The role of the kinetics of elementary steps. This approach allows us to map the machine-learned CO adsorption energies on the surface of the copper nanoparticles to construct the active site visualization (ASV).

Based on the configurations of the adsorbed *OCCOH on the copper clusters, there are 4 ways to place the intermediate on this surface, as shown in Figure 4-4(b). Predicted faradaic efficiency (FE) of the concave site on the minimal periodic structure compared to experimental data. Finally, we used machine learning to match the structure–activity relationship between the local structures of the copper nanoparticle and the theoretical CO adsorption energies.

And the complete set of ReaxFF parameters is available in the supporting information of the referenced work2. Distribution of the energy between the DFT values ​​and the neural network predicted values ​​for the training set, validation set and the test set. RMSEs of the training set, validation set, and test set as functions of the neural network sizes.

Based on the configurations of the adsorbed *OCCOH on the copper clusters, there are 4 ways to place the intermediate on the surface, as shown in the following figure, Figure C-4.

Introduction

Rapid progress is being made in the development of new catalysts that are highly active and selective to electrochemically reduce CO or CO2 to specific chemical fuels and feedstocks1-2. At the same time, using sequestered CO2 as the feedstock will reduce the amount of excess atmospheric CO2 by completing the carbon cycle with carbon sequestration via artificial photosynthesis or other forms of renewable energy sources4-5. After decades of development, copper remains the only catalyst capable of reducing CO or CO2 by more than two electrons to generate valuable products in non-trivial quantities.

Recently, oxide-derived copper nanoparticles (NP) have been shown to significantly improve both the activity and selectivity of CO and CO2 reduction towards C2+ products7. Based on early temperature programmed desorption (TPD) experiments, the improved performance of the oxide-derived metal NP was hypothesized to result from strong CO adsorption sites8. In this work, we focus on elucidating which local Cu structures lead to the optimum properties for CO reduction to C2+ products.

Results and Discussion

The predicted structure leads to XRD spectra and TEM images consistent with those of the experimental NP structures. Section C2 of the Supporting Information shows that the root mean squared error (RMSE) of the CO binding energy (ΔECO) on the training set is 0.111 eV, while for the validation set RMSE = 0.117 eV and for the test set RMSE = 0.123 eV. These results are consistent with the TPD experiments, which show a broad peak centered at 275K only for the copper NP, indicating that ~7–15% of the surface leads to stronger CO adsorptions than low-index copper surfaces.

The three vertical dashed lines correspond to the CO adsorption energies of single crystal surfaces of and (111)15. b) Active site visualization (ASV) of the predicted CO adsorption energies on the nanoparticles. However, we examined the sites with the lowest DEOCCOH and found that all involve square sites similar to those on the (100) surface. Compared to the random sites on the surface of the copper nanoparticles, we found that the square sites were actually more favorable, as shown by the shift in the distribution of DEOCCOH in Figure 4-3(a). a) Distributions of DEOCCOH on the surface of copper nanoparticles.

With the new distribution of only the square locations, we extracted the common features of the most selective locations by further examining the square locations with the lowest DEOCCOH. Of the ABC stack of the FCC copper, the smallest grain size must contain at least 6 layers, corresponding to ABCACB stack, where the A layers are double boundaries. The full *OCCOH structure for (c) is shown in the top view, as shown in the right column.

An experimental copper structure with abundant twin boundaries25 is also extrapolated based on boundary density.

Figure  4-1.  Schematics  of  the  machine  learning  model.  For  each  surface  site  (red),  we  extract  a  copper  cluster  including  all  the  atoms  within  8 Å   from  a  copper  NP
Figure 4-1. Schematics of the machine learning model. For each surface site (red), we extract a copper cluster including all the atoms within 8 Å from a copper NP

Conclusion

The periodic table is then made by replicating 220 individual unit cells along the axis of NW. We randomly selected a nanocluster carved from J-PtNW and calculated the adsorption energy of the reactants (H2O and O) and products (OH and OH) of O"45 hydration. Here, for triangle structures, we took the average bond length of the three Pt atoms, that form the triangle.

And for Concave-Up Rhombuses, we took the average bond length of the four vertices of the rhombus. Atomistic origin of the remarkable oxygen reduction activity of Pt3Ni7 fuel cell catalysts. Identification of selective sites for the electrochemical reduction of CO products to C 2+ on copper nanoparticles by combining reactive force fields, density functional theory and machine learning.

12 Structures from the SurfaceDefect group. The central atom is pink, while the atoms on the same layer are white. As we have already shown 1, 7, the adsorption energy of CO can be adequately described on the nanoparticle surface by considering all atoms within 8A of the surface site. Since we only consider the chemical behavior of the selected surface site, only the molecular descriptions of this site are used as input data for the neural network.

The set of piecewise cosine functions is described by 4 quantities: the inner cutoff 𝑅|®®¬°, the outer cutoff 𝑅±²€¬°, the number of two-body functions 𝑀8, and the number of three-body functions 𝑀2 . Once the structure of the neural network is constructed, the weights and biases of the neural network model are initialized using the Xavier initializer8. Then, using the 8:1:1 split of the data set for training, validation, and testing, we obtain the following RMSE as a function of training iterations, as shown in Figure C-2(a).

Figure  A-1.  Surface  Extraction  using  Surface  Vector  Based  Methodology.  The  angle  threshold  is  optimized  to  30°  here
Figure A-1. Surface Extraction using Surface Vector Based Methodology. The angle threshold is optimized to 30° here

Gambar

Figure 2-2. Generating All Bridge Pairs for Data Sampling.  Starting from the J-PtNW with  6926 Pt atoms, we identified 3881 surface atoms using surface vector methodology
Figure  2-3.  d-OO  Distribution  among  500  Sampled  Bridge  Nanoclusters.  Our  sample  showed a broad range of d-OO, from 2.50 Å to 6.01 Å
Figure 2-6. Identification Results of Barrier-Less Sites (d-OO <= 2.68 Å). Among the 72  barrier-less sites, 30.8% of them are in the Concave-up Rhombus group and 43.6% of them  are in the Triangle group
Figure  2-7.  Statistics  of  Triangle  Group  among  143  out  of  500  Sampled  Bridge  Nanoclusters
+7

Referensi

Dokumen terkait

The aim of this study is to find the answer of the problem, “How is the mastery of eleventh year students of SMUN I Wiradesa in using conjunctions but, and, or, before, and after ?―

Journal of Perioperative Nursing Volume 35 Number 1 Autumn 2022 acorn.org.au e-25 • providing a voice for perioperative nursing • ensuring the health of the perioperative nursing