MnEdgeNet Accurate Decomposition of Mixed Oxidation States for Mn XAS and EELS L23 Edges without Reference and Calibration Huolin L. Xin Mike Hu

2025-05-02 0 0 3.26MB 20 页 10玖币

侵权投诉

MnEdgeNet—Accurate Decomposition of Mixed Oxidation States for

Mn XAS and EELS L2,3 Edges without Reference and Calibration

Huolin L. Xin*, Mike Hu

Department of Physics and Astronomy, University of California, Irvine, CA 92697, United

States

*Correspondence should be addressed to H.L.X. (huolin.xin@uci.edu)

Abstract: Accurate decomposition of the mixed Mn oxidation states is highly important

for characterizing the electronic structures, charge transfer and redox centers for electronic,

electrocatalytic and energy storage materials that contain Mn. Electron energy loss

spectroscopy (EELS) and soft X-ray absorption spectroscopy (XAS) measurements of the

Mn L2,3 edges are widely used for this purpose. To date, although the measurement of the

Mn L2,3 edges are straightforward given the sample is prepared properly, an accurate

decomposition of the mix valence states of Mn remains non-trivial. For both EELS and

XAS, 2+, 3+, 4+ references spectra need to be taken on the same instrument/beamline and

preferably in the same experimental session because the instrumental resolution and the

energy axis offset could vary from one session to another. To circumvent this hurdle, in

this study, we adopted a deep learning approach and developed a calibration-free and

reference-free method to decompose the oxidation state of Mn L2,3 edges for both EELS

and XAS. To synthesize physics-informed and ground-truth labeled training datasets, we

created a forward model that takes into account plural scattering, instrumentation

broadening, noise and energy axis offset. With that, we created a 1.2 million-spectrum

database with a three-element oxidation state composition label. The library includes a

sufficient variety of data including both EELS and XAS spectra. By training on this large

database, our convolutional neural network achieves 85% accuracy on the validation

dataset. We tested the model and found it is robust against noise (down to PSNR of 10)

and plural scattering (up to t/λ = 1). We further validated the model against spectral data

that were not used in training. In particular, the model shows high accuracy and high

sensitivity for the decomposition of Mn3O4, MnO, Mn2O3 and MnO2. In particular, the

accurate decomposition of Mn3O4 experimental data shows the model is quantitatively

correct and can be deployed for real experimental data. Our model will not only be a

valuable tool to researchers and materials scientists but also can assist experienced electron

microscopists and synchrotron scientists in automated analysis of Mn L edge data.

Introduction

X-ray absorption spectroscopy (XAS)1 and electron energy loss spectroscopy (EELS)2,3

are two techniques that can probe the unoccupied electronic states providing bonding

information of materials. In particular, the L2,3 edges are widely used to determine the

oxidation state of transition metals1,4,5. The transition metal L2,3 edges probe the

unoccupied d orbitals and therefore the edge onset and the edges’ fine structures and shapes

are sensitive to the oxidation state of the d-block metal ions, in particular the 3d transition

metals, such as V, Ti, Mn, Fe, and Ni5-8. For example, using the near edge fine structures

in the Mn L2,3 edges, the oxidation states of Mn ions in a material can be determined by

decomposing the spectrum into a linear combination of Mn2+, Mn3+, and Mn4+ reference

spectra9,10. This decomposition, in principle, is simple but in reality, it is non-trivial because

the energy axis is not always calibrated, and the instrument/beamline does not always have

the instrumental broadening. Without proper calibration, an energy offset is present

between the experimental spectrum and the references which prevents accurate oxidation

states decomposition. In order to avoid the problem, standard reference samples such as

MnO, Mn2O3, MnO2 need to be measured in the same experimental session to avoid any

energy offsets as well as change in instrumental broadening9,11. Still with this procedure,

there are other factors that could prevent the proper energy axis calibration for example

temperature fluctuations would result in an energy shift in the monochromator. Basically,

if the XAS measurements are separated multiple hours in time, the spectra taken could have

a slight energy offset. In EELS, the energy offset could change more rapidly and is more

unpredictable than XAS. Typically, the energy offset is very sensitive to the DC stray field.

For example, the passing of a truck or the movement of a nearby elevator, all could change

the energy offset if the TEM is not fully shielded. This problem is now mitigated with the

dualEELS instruments but there are still many single EELS instruments under active

service. Moreover, all historical data were acquired without the dualEELS correction. In

addition, nonlinearity of parallel EELS spectrometer is present in EELS in a nontrivial way

because the nonlinearity is not only present in the dispersion device, i.e. the magnetic prism.

There is another complex nonlinearity present in the magnification lenses, a series of

quadrupole. Therefore, it is extremely difficult to calibrate the energy onset of EELS edges

unless strict protocols are followed as described by Tan et al11.

Another complication is that EELS’ near-edge fine structures change with sample thickness

due to plural scattering. As the sample gets thicker, signals close to the edge onset would

be multiple scattered to higher energy losses. This would result in a shape change of the

spectrum11. For example, for latter 3d transition metals’s L2,3 edges, as the sample gets

thicker, the L2/L3 ratio increases—this problem has rendered the reference-free L2,3 ratio

method inaccurate for EELS11. In addition, for XAS, the background and the near edge

structures could be different between the TEY and TFY modes. That also renders the L2,3

ratio method unreliable. Moreover, for early 3d transition metals, there are no established

reference-free methods because of the L2,3 anomaly.

For both EELS and XAS spectroscopy, one interesting observation is that human operators

with sufficient training can identify spectral features and assign oxidation states to

transition metal L2,3 edges with high confidence. This points to the direction that deep

learning could be successful in solving the L2,3 oxidation state decomposition problem.

Pate et al in 2021 discussed using deep learning to denoise high frame rate spectra.12

Chatzidakis and Botton in 2019 introduced the idea of translation-invariance for classifying

EELS edges.13 They built a convolutional neural network (CNN) for oxidation state

classification and showed that with a translation-invariant training, moving the energy axis

does not change the Mn 2+, 3+, 4 oxidation state classification. This is a very important

step in demonstrating that spectral features are like spatial features in images—they can be

classified by a CNN network regardless of their absolute energy positions in the spectrum.

However, there are still problems remained to be resolved: 1) how to quantitatively

decompose mixed oxidation states; 2) is it possible to build one model that work for both

XAS and EELS spectroscopy that have drastically different energy resolution; 3) is it

possible to build a model that is not affected by plural scattering, i.e. the thickness effect

in EELS.

To address the three challenges defined above, in this study, we present a reference-free,

calibration-free deep learning approach to determine the accurate oxidation states

decomposition of 3d transition metal based on the L2,3 near edge fine structures. To

demonstrate the validity of the method, in this study, we use Mn as an example because

Mn is technologically important in catalysis, energy storage and electronic materials.

Determining the composition of the mixed oxidation states is extremely important for

understanding the charge transfer phenomenon happening at the device interfaces. The

method we present in this study is not a simple classification of Mn2+, 3+ and 4+ edges

but an accurate and quantitative decomposition of the mixed Mn oxidation states. Instead

of having a classification/binary type label, we created a three-element ground truth vector

that quantitatively describe the composition of Mn2+, 3+ and 4+ in each Mn spectrum, i.e.

[%Mn2+, %Mn3+, %Mn4+].

To achieve this goal, we built a 1.2 million-spectrum ground truth labeled library with 50%

XAS data and 50% EELS data. In building the mixed oxidation state library, we paid

special attention to normalizing the Mn L2,3 edges correctly, and including experimental-

like uncertainties such as both Gaussian and Lorentzian type instrumental broadening,

energy offset and detector noise. To include the plural scattering effect in the training

library, we developed a forward model to correctly introduce the thickness effect to the

L2,3 edges. Using this physics-informed large training library, we show that the deep

convolutional regression model we trained is robust against plural scattering and noise. The

overall accuracy of model in determining the mixed valence state reaches 85% on the

validation data set. We also validated the data on “unknown unknowns”, i.e. Mn3O4

spectra that have never been used for training—the accurate decomposition of Mn3O4

experimental data shows the model is quantitatively correct and can be deployed for real

experimental data.

Methods

In this method section, we will describe 1) how to build a ground-truth oxidation state

labeled Mn edge library, 2) how to construct the neural network, and 3) how to train it.

For building the library, the technical challenges lie in 1) how to obtain a wide variety of

XAS and EELS Mn2+, Mn3+, Mn4+ reference spectra; 2) how to normalize or ratio the

2+, 3+ and 4+ spectra correctly; 3) how to include the EELS’s plural scattering effect

(thickness effect) into the training sets; 4) how to include the various experimental

uncertainties including instrumental broadening, energy offset, detector noise, etc. In the

following subsections, we will address each aforementioned challenge.

Collection of Mn reference spectra

To have sufficient varieties of data that can capture the features of the EELS and XAS Mn

2+, 3+ and 4+ edges, in this study, we digitized a large number of experimental EELS and

XAS Mn spectra that were documented in the literature. In Figure 1, we presented all

spectra that were used for making the training library. (The Mn 2.67+ spectrum was not

included in the training library.) In Table 1, we listed the compounds for which we digitized

the spectra and their original references.

All data are standardized to range from 630.5 eV to 669.4 eV with 0.1 eV increments (338

data points). For missing data, the left side of the spectra is padded with zero and the right

side is padded with the end value of the spectra. Fig xx shows examples of the standardized

data.

Figure 1. The presentation of the EELS and XAS Mn L2,3 edges included in making the

training library. The Mn 2.67+ presented is not included in the training library.

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

MnEdgeNet—AccurateDecompositionofMixedOxidationStatesforMnXASandEELSL2,3EdgeswithoutReferenceandCalibrationHuolinL.Xin*,MikeHuDepartmentofPhysicsandAstronomy,UniversityofCalifornia,Irvine,CA92697,UnitedStates*CorrespondenceshouldbeaddressedtoH.L.X.(huolin.xin@uci.edu)Abstract:Accuratedecompositionof...

展开>> 收起<<

MnEdgeNet Accurate Decomposition of Mixed Oxidation States for Mn XAS and EELS L23 Edges without Reference and Calibration Huolin L. Xin Mike Hu.pdf

共20页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

MnEdgeNet Accurate Decomposition of Mixed Oxidation States for Mn XAS and EELS L23 Edges without Reference and Calibration Huolin L. Xin Mike Hu

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: