Martini2023-Single Atom

pubs.acs.
org/JACS Article
Tracking the Evolution of Single-Atom Catalysts for the CO2

Electrocatalytic Reduction Using Operando X‑ray Absorption
Spectroscopy and Machine Learning
Andrea Martini, Dorottya Hursán, Janis Timoshenko,* Martina Rüscher, Felix Haase, Clara Rettenmaier,
Eduardo Ortega, Ane Etxebarria, and Beatriz Roldan Cuenya*
Cite This: J. Am. Chem. Soc. 2023, 145, 17351−17366 Read Online
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.
ACCESS Metrics & More Article Recommendations *

sı Supporting Information
Downloaded via CHULALONGKORN UNIV on October 15, 2024 at 09:04:43 (UTC).
ABSTRACT: Transition metal-nitrogen-doped carbons (TMNCs) are a

promising class of catalysts for the CO2 electrochemical reduction reaction. In
particular, high CO2-to-CO conversion activities and selectivities were
demonstrated for Ni-based TMNCs. Nonetheless, open questions remain
about the nature, stability, and evolution of the Ni active sites during the
reaction. In this work, we address this issue by combining operando X-ray
absorption spectroscopy with advanced data analysis. In particular, we show that
the combination of unsupervised and supervised machine learning approaches is
able to decipher the X-ray absorption near edge structure (XANES) of the
TMNCs, disentangling the contributions of different metal sites coexisting in the
working TMNC catalyst. Moreover, quantitative structural information about
the local environment of active species, including their interaction with
adsorbates, has been obtained, shedding light on the complex dynamic
mechanism of the CO2 electroreduction.
1. INTRODUCTION the as-prepared samples, as showcased by the striking example

The CO2 electrocatalytic reduction reaction (CO2RR), of Cu-based TMNCs, for which the reversible formation of
powered by renewable electricity sources, is a promising metallic Cu clusters under CO2RR conditions was recently
pathway for the abatement of the CO2 emissions from reported.17,18 Thus, only an operando characterization can
industrial sites, where concentrated CO2 is available, and for provide unambiguous answers about the active state of the
the associated production of valuable chemical feedstocks. TMNC catalysts.5,19−21 However, considering the low metal
Nonetheless, suitable active and selective catalysts are still loadings, the lack of long-range ordering in TMNC materials,
needed.1 Transition metal nitrogen-doped carbons (TMNCs) as well as their heterogeneous nature, only a few experimental
have attracted attention as promising electrocatalysts for techniques are up to this task.6,22
CO2RR due to their high activity and selectivity for the CO2 X-ray absorption spectroscopy (XAS) stands out as an
conversion to CO, especially for Ni-based TMNCs.2−4 The element-specific tool that can be applied to a broad range of
structure and the catalytic functionality of these systems differ functional materials under working conditions.5,20,21 Thanks to
significantly from those of bulk or nanostructured materials.5−7 its element-selectivity, XAS can be employed for the
In TMNCs, nitrogen atoms are incorporated in the carbon determination of the oxidation state and local atomic and
matrix forming binding sites for the metal species. These singly electronic structure of the metal sites, shedding light on the
dispersed metal sites are often considered to be the active possible active moieties for the CO2RR. In particular, the
species for the CO2RR reaction,8−11 as well as for other analysis of the extended X-ray absorption fine structure
electrochemical processes such as the oxygen reduction (EXAFS) was shown to be an invaluable tool for the
reaction (ORR).7,12−15 Nonetheless, in addition to singly confirmation of the singly dispersed nature of the TMNC
dispersed metal sites, a large number of different structural
motifs can coexist during the reaction in TMNC catalysts,
hindering the unambiguous identification of the active Received: May 12, 2023
species.16 These can include multiple pyrollic and pyridinic Published: July 31, 2023
nitrogen environments, metallic clusters and carbide, and oxide
and nitride particles.2 We especially highlight here that the
species present under the reaction conditions and participating
in the catalytic processes can differ significantly from those in
© 2023 The Authors. Published by
American Chemical Society https://doi.org/10.1021/jacs.3c04826
17351 J. Am. Chem. Soc. 2023, 145, 17351−17366
Journal of the American Chemical Society pubs.acs.org/JACS Article
catalysts and for the identification of M−Nx moieties (where °C and subjected to carbonization in Ar flow at 1000 °C for 1 h. At
the x denotes the average number of N atoms bound to the this temperature, most of the metallic Zn evaporates from the sample,
metal site M; usually x is equal to 4) as the main building block leaving behind a porous (Zn)-N-doped carbon structure (denoted as
for the TMNC catalysts, at least in the as-prepared state.23−28 N−C). To remove all the crystalline (not single-atomic) Zn species
from the N-C support, we performed an acid washing at room
On the other hand, in the absence of clusters and more temperature using 20 wt% nitric-acid (HNO3, ≥ 65%, Carl Roth)
ordered structural motifs, the information content in the over 24 h, followed by a thorough washing and vacuum filtration of
EXAFS spectra is limited. In this situation, the X-ray the sample with ultrapure water (at least 3 × 600 mL) until the pH of
absorption near edge structure (XANES) part of the XAS the supernatant solution reached a value larger than 5. After drying
spectrum can provide the decisive evidence about the nature of the acid-washed N−C, we impregnated it with a solution of nickel-
the active states. In comparison to EXAFS, XANES is more nitrate (Ni(NO3)2·6H2O, Sigma-Aldrich, 99.999%). 200 mg of N−C
sensitive to the symmetry of the metal center, to the 3D was added to a solution of 20 mL of 6 mM Ni(NO3)2 in isopropanol,
geometry of its environment and distortions.5 Moreover, the suspension was sonicated for 2 h in an ultrasonic bath at ∼30 to
XANES spectra have typically a higher signal-to-noise ratio as 40 °C and then stirred for another 2 h with a magnetic stirrer at room
temperature. We obtained the “precursor” sample denoted here as Ni-
compared to EXAFS data, making them more suitable for time-
TMNC after centrifugation. To get the final catalyst, which we refer
resolved operando investigations, when the data quality in the to as HT-Ni-TMNC, the precursor was subjected to another heat
EXAFS region is often compromised by the low metal loading treatment in Ar flow at 700 °C. Finally, the HT-Ni-TMNC was
and signal attenuation by the electrolyte or the elements washed with ultrapure water, centrifuged, and dried in air at 60 °C.
composing the electrochemical cell. Further adoption of The prepared samples were analyzed ex situ using transmission
XANES analysis for quantitative investigations of working electron microscopy (TEM), X-ray diffraction (XRD), and X-ray
catalysts, however, is hindered by the lack of simple fitting photoelectron spectroscopy (XPS).
approaches. Nonetheless, during the last years, with the For the electrochemical and operando XAS experiments, the
development of reliable XANES ab initio simulation catalysts were deposited onto carbon paper supports (Freudenberg
approaches, machine learning methods, and DFT model- H15C13, Fuel Cell Store). The catalyst ink consisted of 30 mg
catalyst (Ni-TMNC or HT-Ni-TMNC), 75 μL Nafion 117 solution
ing,7,13,21,29−36 the situation has begun to change, making (5%, Sigma-Aldrich), 2.4 mL ultrapure water, and 2.4 mL isopropanol
quantitative XANES-based analysis for TMNC materials finally (C3H7OH, ≥99.8%, Sigma-Aldrich). After sonicating the ink for at
feasible. Despite this initial progress, most of the existing least 60 min, it was spray-coated onto the preheated (90 °C) carbon
studies have focused on understanding the structure of the as- paper until the desired mass loading was reached (1.2 ± 0.1 mg cm−2
prepared TMNC catalysts. Moreover, the few existing for the electrocatalytic measurements, while 1.6 ± 0.1 mg cm−2 was
operando XANES studies were devoted mostly to the ORR used for the operando XAS measurements).
reaction.7,13,37 Little attention has been paid so far to the fact 2.2. Ex Situ Characterization. 2.2.1. Transmission Electron
that the actual working TMNC catalysts could feature Microscopy. TEM images were acquired using a probe-corrected
coexisting metal sites with different nonequivalent environ- JEM-ARM 200F (JEOL, Japan) scanning transmission electron
microscope (STEM) equipped with a cold field emission gun
ments, rendering the commonly used fitting approaches, which (CFEG) operated at 200 kV. The high angle annular dark field
assume a single structural model, inaccurate. (HAADF), annular bright field (ABF), and bright field (BF) detector
Here, operando time-resolved XANES data were used to signals were collected from an electron probe with a 14.2 mrad
unveil the local structure around Ni sites not simply in their as- convergence semiangle and a 90−370, 12−40, 18 mrad collection
prepared or “after reaction” states but under realistic working semiangle, respectively. The beam current was kept at 11 pA and its
conditions during CO2RR. A multistep approach has been resulting electron dose was scaled by the pixel size. Image acquisition
used here. First, we identified the number of different and manipulation were performed with the DigitalMicrograph
coexisting Ni species, their corresponding kinetic profiles, software v2.4 (Gatan, USA).
and XANES spectra using unsupervised machine learning Energy-dispersive X-ray spectroscopy (EDS) spectra and elemental
mapping were acquired using a Talos F200X (ThermoFisher
methodologies, such as the principal component analysis
Scientific, USA) STEM microscope operated at 200 kV and equipped
(PCA) combined with a transformation matrix technique.21,38 with four silicon drift detectors (SDDs). The 72 pA electron beam
In a second step, we deduced the atomistic structures for each with a 10.5 mrad probe convergence semiangle was scanned across
of the identified species through a XANES fitting procedure the region of interest under a continuous frame acquisition mode. The
realized by exploiting a supervised machine learning EDS quantification was performed using the Velox software v1.4.2
approach.21,30 Finally, we validate the predicted structures (ThermoFisher Scientific). To reduce the background signal of the
processing the corresponding EXAFS spectra via Reverse carbon framework, the net elemental maps (baseline intensity counts
Monte Carlo (RMC) simulations,39−41 taking into account the removed) were displayed to highlight the presence of the doping
structural disorder in the local environment around the heavy metals.
identified Ni species. For the postreaction TEM analysis, the CO2RR experiments were
performed using glassy carbon plates as catalyst supports to avoid the
presence of carbon originating from the carbon paper.
2. EXPERIMENTAL SECTION 2.2.2. X-ray Diffraction. Powder X-ray diffractograms were
2.1. Sample Preparation. Ni-based TMNC catalysts were recorded with a Bruker D8 Advance instrument using a Cu anode
synthesized following an impregnation-calcination method as reported (8046.3 eV) between 10 and 90° 2Θ values, with 0.02° step size and 3
in the literature.42 First, we prepared a zeolitic-imidazolate framework s dwell time.
(ZIF-8) precursor in the reaction between zinc-nitrate (Zn(NO3)2· 2.2.3. X-ray Photoelectron Spectroscopy (XPS). XPS spectra were
6H2O, 98%, Acros Organics) and 2-methylimidazole (C4H6N2, 99% acquired with a SPECS Phoibos 150 spectrometer with an Al Kα
Sigma-Aldrich). Specifically, we dissolved 6.78 g of Zn(NO3)2·6 H2O source (300 W, 12.52 kV). Survey spectra were recorded with 100 eV
and 7.87 g of 2-methylimidazole in 800 mL of methanol (CH3OH, pass energy, 0.1 s dwell time, 0.75 eV step size, and 2 scans. The N 1s
≥99.8%, Honeywell), heated the solution to ca. 60 °C, and stirred for and the metal 2p regions were recorded with 30 eV pass energy, 0.3 s
24 h under reflux. Next, we collected the ZIF-8 crystals by dwell time, and 0.15 eV step size using 20 and 60 scans, respectively.
centrifugation and thoroughly washed them two times with methanol Data analysis and fitting were performed using the CasaXPS software.
and once with ethanol. The obtained crystals were dried in air at 60 High-resolution spectra were fitted with 70% Gaussian and 30%
17352 https://doi.org/10.1021/jacs.3c04826
J. Am. Chem. Soc. 2023, 145, 17351−17366
Lorentzian line shapes, and a Shirley background subtraction was XANES part of each XAS spectrum took just a fraction (ca. 1 min) of
applied. The binding energy scale was adjusted by assigning the signal the total acquisition time, making XANES a more reliable probe of
of graphitic carbon to 285 eV. The FWHM for N peaks was fixed rapid processes associated with the changes in the catalyst structure.
between 1.5 and 1.6, and the spectra were fitted based on previously We performed the alignment, the background subtraction, and the
established protocols for similar materials.43−48 normalization of the corresponding XAS profiles using the Athena
2.2.4. Inductively Coupled Plasma Mass Spectrometry (ICP-MS). software.50 For the further processing of the XANES spectra, we
Catalysts were first digested in a microwave digestion system (Anton applied supervised and unsupervised machine learning approaches
Paar, Multiwave GO) at 180 °C for 20 min in an acid mixture provided by the PyFitIt code.51 Finally, we realized the EXAFS data
containing cc. HNO3, cc. H2SO4, cc. HCl in a volume ratio of 2:2:6. fitting employing Reverse Monte Carlo simulations52 using the EvAX
Then, the solutions were filtered and diluted to ca. 50 mL with code,39,41 exploiting the structural models deriving from the XANES
ultrapure water. For the ICP-MS measurement, a 20× dilution of each data analysis as the initial guesses for the EXAFS-based 3D structural
sample in 3% HNO3 was prepared. The measurements were refinement of the catalyst structure.
performed with a ThermoScientific iCAP RQ instrument.
2.3. Electrocatalytic Activity Measurements. The electro- 3. RESULTS
chemical measurements were performed using an Autolab 3.1. Ex Situ Characterization and CO2RR Activity. To
PGSTAT302N potentiostat/galvanostat. The working electrode was
the catalyst-coated carbon paper, usually with a 0.5−1 cm2 geometric
characterize the materials, we first collected the powder X-ray
surface area. The counter electrode was a Pt-mesh and the potentials diffractograms of the precursor (Ni-TMNC) and heat-treated
were measured against a leak-free Ag/AgCl electrode (0.242 V vs (HT-Ni-TMNC) catalyst. The XRD patterns for both samples
standard hydrogen electrode). We report the potentials throughout are typical for an amorphous carbon (see Figure S3). The two
the text versus the reversible hydrogen electrode (RHE). These were broad reflexions at around 25° and 43° are related to the (002)
calculated using the following equation: E (vs RHE) = E (vs Ag/ and (101) planes of graphite. Importantly, no reflections
AgCl) + 0.242 + 0.059 × pH. The solution resistance (Ru) for the IR indicative of the presence of crystalline phases were detected.
correction was determined by electrochemical impedance spectros- To unambiguously prove the absence of Ni/NiO nanoparticles
copy using the high-frequency intercept of the semicircle on the and Zn/ZnO nanoparticles remaining from the N-C support
Nyquist plot with the real axis. The IR-corrected potential (E−IR) was
preparation, we carefully investigated the materials using TEM.
calculated with the following formula E−IR = E − I·Ru, where E is the
applied/measured potential, while I is the applied/measured current. The medium-resolution HAADF-STEM images in Figure 1
CO2RR experiments were performed in a gas-tight two-compart-
ment H-type cell. Its description and schematic depiction are given in
Section S1 and Figure S1. The cathode and anode compartments
were separated by a Selemion anion exchange membrane to avoid
product mixing. CO2 was continuously bubbled through the anolyte
and catholyte with 20 mL min−1 flow rate. The electrolysis was
performed in a 0.1 M KHCO3 solution that was pretreated with an
ion-exchange resin (Chelex 100 Resin sodium form; Bio-Rad) to
remove metal impurities. The gas outlet of the cathode compartment
was directly connected to the injector of the gas chromatograph via a
6-port valve, allowing the online detection of the gaseous products.
Samples were automatically injected every 15 min of the reaction. Gas
products were detected and quantified by an Agilent 7890B gas
chromatograph. The products were separated by different columns
(Molecular sieve 13×, HayeSep Q, and Carboxen-1010 PLOT) and Figure 1. STEM annular dark field (ADF) images of the nickel-
subsequently quantified with a flame ionization detector (FID) as well nitrogen-doped carbon catalysts. The high-resolution HR-STEM
as with a thermal conductivity detector (TCD). images show the disordered carbon structure and the presence of
In the liquid phase, acetate and formate concentrations were individual metal atoms (bright spots) before and after CO2RR.
analyzed by high-performance liquid chromatography (HPLC,
Shimadzu prominence) equipped with a NUCLEOGEL SUGAR show the rhomboid-dodecahedron morphology of the N−C
810 column and refractive index detector (RID). Other liquid carbon, which was inherited from the ZIF-8 precursor. No
products (alcohols and aldehydes) were quantified with a liquid GC nanoparticles were observed in these materials. In the high-
(L-GC, Shimadzu 2010 plus) equipped with a fused silica capillary resolution images, the single metal atoms appear as brighter
column and FID detector. dots, because of the higher Z-contrast of the heavy elements,
2.4. Operando XAS Measurements. We acquired operando Ni compared to that of the carbon/nitrogen atoms. From these
K-edge XAS data at the BESSY II synchrotron (KMC-3 XPP
beamline).49 Operando XAS measurements were performed in our
images, however, we cannot tell whether these are Ni or Zn
home-built single-compartment electrochemical cell.5 The schematic single atoms, as Zn remaining from the ZIF-8 precursor is also
depiction of the cell and its description can be found in Section S1 inherently present in the materials. Nevertheless, energy-
and Figure S2 in the Supporting Information (SI). Measurements dispersive elemental mapping revealed the uniform distribution
were performed in a CO2-saturated 0.1 M KHCO3 electrolyte under a of Ni in our samples, without any visible agglomeration (Figure
static current of −10 mA, which corresponds to a current density of S4). This is a strong indication that Ni is present as single
−15.7 mA/cm2. A Pt mesh was used as a counter electrode, while a atoms in the as-prepared state of the catalysts, which is further
leak-free Ag/AgCl electrode constituted the potential reference. The confirmed by our detailed XAS analysis provided below.
applied current was controlled by a BioLogic potentiostat. We The elemental composition of the catalysts was analyzed by
collected time-resolved spectra with an acquisition rate of ca. 9 min
XPS and ICP-MS (Figure S5 and Tables S1 and S2). The main
per spectrum until no further changes could be observed in the
XANES data. Because of the ultra-dispersed nature of the catalysts, components of the catalysts were C (85−90%), N (6−7%),
the latter parameter was the minimal total time necessary to acquire and O (3.5−7%). The Zn and Ni contents were below 0.5 at
the entire XAS signal, sampling properly the XANES pre-edge and %. Interestingly, the Ni-TMNC (precursor sample) displayed a
white line regions, and measuring the corresponding EXAFS part with lower Ni content and also had a lower Ni/N ratio as compared
a good signal-to-noise ratio. We emphasize that the collection of the to the heat-treated catalyst. We fitted the high-resolution N-1s
J. Am. Chem. Soc. 2023, 145, 17351−17366
spectra (Figure S6a,b and Table S3) with components typically performed the precursor sample. At −0.95 V, the total current
observed for MNC materials.43−48 The main N species were density was −18 mA cm−2 for HT-Ni-TMNC, while only −11
pyridinic N (398.3 eV), followed by N−H (i.e., pyrrolic or mA cm−2 for Ni-TMNC. In addition, the CO partial current
hydrogenated pyridinic N at 400.8 eV) for both catalysts. The density decayed more rapidly over time in the case of Ni-
peak at 399.5 eV is indicative of nitrogen-coordinated metal TMNC (Figure S8), indicating that the heat treatment at 700
sites. We also note that Ni-TMNC contained significantly °C stabilized the Ni single sites. Importantly, no nanoparticle
larger amounts of N−Ox groups, compared to HT-Ni-TMNC or agglomerate formation was observed by TEM analysis after
(16.9 vs 6.8%), indicating that the nitrate groups originating reaction in any of the samples (see Figure 1).
from the Ni(NO)3 decomposed or reduced during the heat 3.2. Qualitative XANES and EXAFS Analysis. For the
treatment at 700 °C. We also recorded the high-resolution Ni operando XAS experiments, the CO2RR was performed under
2p regions of the catalysts. Because of the low metal loading, galvanostatic conditions with −15.7 mA cm−2 applied current
hence weak signal intensity, we could not perform accurate density (Figure S9). Figure 3a shows Ni K-edge XANES
peak deconvolution using the multiple fitting features of Ni spectra collected for the HT-Ni-TMNC catalyst at the
oxides and hydroxides. Instead, we focused on the position of beginning and at the end of the CO2RR process, while the
the main peaks to determine the dominant oxidation state of complete set of XANES spectra collected under CO2RR
Ni present in the samples.3 The 2p3/2 main peak is located at conditions can be found in Figure S10. One can immediately
either 855.2 or 855.5 eV for the HT-Ni-TMNC and the Ni- note that the XANES spectra experience significant changes,
TMNC samples, respectively, which is close to the main peak suggesting strong transformations in the catalyst structure. In
of Ni(OH)2 reported in the literature.53 Thus, the oxidation all cases, the position of the absorption edge matches well that
state of Ni is +2 in these samples. It should be noted that Ni- of reference materials with Ni species in the 2+ state. At the
phthalocyanines (NiPc) also have the main XPS line between same time, the XANES features both, for the initial catalyst and
854.8 and 855.2 eV;54−56 hence, the main peak of HT Ni- its final state, differ clearly from those in the available reference
TMNC may also (partially) originate from a porphyrin-like spectra for NiO and NiPc, suggesting that the local structure
structure. In the spectrum of HT Ni-TMNC, we also observed around the Ni sites is distinct from that in these standard
a weak shoulder at 852 eV (2% of the main peak), originating materials.
from the negligible amounts of Ni0. The heat treatment of the The initial spectrum for the HT-Ni-TMNC catalyst is
Ni-TMNC catalyst resulted in a notable reduction in the characterized by a relatively strong main XANES feature at ca.
satellite intensity (by ∼35%), which may also suggest an 8350 eV (the so-called white line, W.L.), which resembles the
increasing contribution from a NiPc-like structure in the HT W.L. in the reference spectrum for the rocksalt-type NiO.
Ni-TMNC catalyst.3 Nonetheless, for our catalyst, the W.L. feature is broader and
We tested the activity of the catalysts in the CO2RR in an H- less intense than the W.L. of the NiO compound. Under the
type cell configuration in CO2-saturated 0.1 M KHCO3 CO2RR conditions, the W.L. feature further decreases, while
electrolyte between −0.55 and −1.15 V (vs RHE), see Figure the two main pre-edge peaks located at ca. 8333 and 8338 eV
2 for the HT-Ni-TMNC and Figure S7 for the Ni-TMNC. The acquire progressively more intensity. By comparing the spectra
of the HT-Ni-TMNC sample with that of the NiPc reference
and with the available literature data for similar systems,57 we
can assign these pre-edge features to the 1s → 3d−4p
(quadrupole) and the 1s → 4pz (dipole allowed-shakedown
contribution) transitions, respectively. Such pre-edge features
are characteristics of planar Ni complexes. The 1s → 4pz
feature that gets enhanced under CO2RR conditions is
considered to be a fingerprint of the square-planar Ni−N4
configuration with a D4h symmetry.25,58 The intensity of this
feature diminishes with the distortion of the square-planar
environment. At the same time, such geometrical distortions
result in an enhancement of the 1s → 3d−4p peak.26,59−61
Thus, based on a visual examination of the operando spectra,
we can hypothesize that the changes in the Ni K-edge XANES
spectra under CO2RR are associated with a loss of the
Figure 2. Catalytic activity (total current densities) and selectivities octahedral geometry of the Ni(II) sites characteristic for the as-
(faradaic efficiencies) of the HT-Ni-TMNC catalysts in the CO2RR. prepared samples, leading to the formation of distorted square
Catalytic tests were performed under potential control in an H-type planar Ni−N4 units, although the formation of a squared
cell using a CO2-saturated 0.1 M KHCO3 electrolyte. The currents pyramidal or trigonal bi-pyramidal structures cannot be ruled
were normalized by the geometric surface area. Error bars reflect the out, as shown in refs 58, 62. Herein, it is also worth noting that
standard deviation of the measured data for measurements performed
on at least two separate electrodes.
during CO2RR we did not observe any significant shift of the
absorption edge position (see Figure S11a), suggesting that the
Ni oxidation state does not change. This evidence allows us to
potential-dependent catalytic activity is also provided. Both rule out the formation of metallic Ni clusters during the
catalysts produced CO with high selectivity (>90% faradaic reaction, in accordance with the post reaction TEM analysis.
efficiency at potentials below −0.6 V), accompanied by only The qualitative findings from the XANES data are
minor hydrogen evolution as a side reaction, in accordance corroborated with those derived from the examination of the
with previous reports on Ni single atomic catalysts.2−4 In terms EXAFS data. The raw EXAFS data are reported in Figure S11b,
of current density, the heat-treated catalyst, however, out- while Figure 3b shows the magnitude of the Fourier-
J. Am. Chem. Soc. 2023, 145, 17351−17366
Figure 3. Operando Ni K-edge (a) XANES and (b) magnitudes of the Fourier Transformed (FT) EXAFS (phase uncorrected) spectra collected
before and at the end of the CO2RR (after 1 h) for the HT Ni-TMNC sample. The inset in panel (a) shows a magnification of the XANES pre-
edge region. The Fourier transform of the EXAFS spectra was carried out in the k space between 1 and 9 Å−1. XAS measurements were performed
in a CO2-saturated 0.1 M KHCO3 electrolyte under a static current density of −15.7 mA/cm2.
transformed (FT) EXAFS signals. In the as-prepared state, the from the original series of experimental spectra, providing that
FT-EXAFS spectra are dominated by the contribution of the (i) these species exhibit sufficient spectroscopic contrast and
first coordination shell (single peak at ca. 1.5 Å, phase (ii) that their concentration profiles are linearly independent
uncorrected). The intensity and position of this peak resemble (i.e., the ratios of different species are different for at least
those in the reference spectrum for NiO, suggesting similar several collected spectra).
octahedral coordination. The lack of strong peaks at larger R- In our case, we focused on the analysis of a set of XANES
values confirms the single-site nature of the catalyst, as also data (Figure 3a) consisting of nine spectra collected for the
highlighted by the corresponding wavelet transforms of the HT-Ni-TMNC sample during the CO2RR process. This set of
EXAFS spectra reported in Section S4 of the SI. Nonetheless, discretized spectra in the energy region between Emin = 8325
caution is needed when the sample-averaging EXAFS method and Emax = 8430 eV forms a matrix X. The objective of our
is used to draw such conclusions since it cannot rule out the analysis is to decompose this matrix in terms of spectra for
presence of small fractions of disordered clusters.63,64 The pure species and associated concentrations, forming the data
smaller peaks at 2.5 and 3.7 Å in the FT-EXAFS data can be matrices S and C, respectively: X = SC. The number of distinct
tentatively attributed to the interactions between Ni and the chemical species appearing and evolving during the chemical
carbon support. Under CO2RR conditions, the intensity of the reaction can be determined by the principal component
main FT-EXAFS peak decreases, while the intensities of the analysis (PCA). The dataset X can be expressed as X = UΣV.
second and third peak increase, supporting the hypothesis of Here, the columns of matrix U are orthonormal (energy-
significant rearrangements in the local environment of Ni sites dependent) vectors, sorted by their importance to the original
that, nonetheless, still preserve their singly dispersed nature. All dataset (the amount of the XANES variance accounted by the
FT-EXAFS peaks also appear to shift to lower R-values, respective component). We refer to these vectors as principal
indicating a contraction of the interatomic distances between components (PCs). Σ is a diagonal matrix whose elements,
Ni species and their nearest neighbors. called singular values, characterize the importance of each PC
3.3. Spectral Decomposition of the XANES Dataset. in the dataset X. Finally, the rows of matrix V contain the
Quantitative information about the type and relative amount of projections of each spectrum in the dataset X on the
distinct species appearing and evolving during a chemical corresponding PC vector (scaled by the corresponding singular
reaction can often be retrieved by the analysis of the value). We emphasize that the PCs themselves do not
corresponding XANES spectra employing the linear combina- correspond to the spectra of any particular chemical species.
tion fitting. Following this approach, every spectrum of the However, all the spectra in the dataset X, as well as all the
XANES dataset can be written as a linear combination of spectra of the pure compounds can be represented as some
certain XANES standards, which must be selected before- linear combinations of the PCs. The number of significant PCs
hand.21,38 This approach, however, cannot be applied if the (i.e., those with significant corresponding singular values) thus
structural motifs contributing to the experimental data are not defines the dimensionality of the dataset X and corresponds to
known and/or differ significantly from the structures of well- the number of spectroscopically distinct species present during
defined standard compounds.65 This is clearly the case in the the reaction. By analyzing the PCs and their corresponding
studies of TMNC catalysts, where the unique catalytic singular values, we concluded that, in our case, three PCs are
functionality is ensured by the presence of unique structural sufficient to explain all the variations in our original dataset X.
motifs with no analogues among the common bulk standard Indeed, we saw that only the first three PCs have some distinct
materials. In the absence of suitable references, the features that can be linked to particular features in our
decomposition of experimental XANES data in a set of spectra experimental XANES spectra. The fourth and the subsequent
for pure compounds and associated concentration profiles is PCs encode the experimental noise and bear no structural
still possible and can be realized by means of unsupervised information (see Figure S13 of Section S5). This evidence
machine learning methods. Here, as a first step, it is necessary leads us to conclude that the number of independent Ni
to identify the number of spectroscopically distinct species species present in our catalyst is three. A more detailed
present during the reaction. Afterward, the XANES spectra discussion of the determination of a number of species based
corresponding to these pure chemical species can be deduced on PCA is given in Section S5 and Figures S14 and S15.
J. Am. Chem. Soc. 2023, 145, 17351−17366
Figure 4. (a−c) Dark blue polygons: areas of feasible solutions obtained using a spectra-specific constraint requiring that XANES spectra are
nonnegative and have a limited maximal amplitude. The light blue circles show narrower AFS obtained imposing as a further constraint that all the
concentration values should be numbers between 0 and 1. Panel (a) corresponds to the first pure species, while (b, c) to the second and the third
ones, respectively. The three XANES spectra in panel (d) are examples of feasible XANES spectra for the second species, corresponding to three
different solutions depicted in panel (b). The red crosses indicate the single point solution, where it is assumed, in addition, that the first and the
last spectrum in the experimental dataset (i.e., the spectrum corresponding to the as-prepared sample in air and the final spectrum collected under
the CO2RR conditions) correspond to pure species.
Knowing that only three species are present in our sample for relevant structure models (vide infra). It is possible to
during CO2RR, we can write X≈Ũ Σ̃Ṽ , where we approximate identify now the area of feasible solutions (AFS) containing all
the initial dataset X using only the first three columns of U, the those T matrix elements satisfying these constraints.
first three singular values of Σ and the first three rows of V. As Considering that both aforementioned spectra-specific con-
a next step, to retrieve a set of XANES spectra corresponding straints for each species depend only on two T matrix elements
to the pure compounds, we follow the transformation matrix (since each column of the matrix T affects just one of the
(TM) approach as implemented in PyFitIt.66 We note that the spectra in matrix S), the determination of the corresponding
decomposition of matrix X can be further rewritten as X ≈ AFSs is a straightforward linear 2D problem that can be solved
Ũ Σ̃TT−1V∼, where I = TT−1 is a 3 × 3 identity matrix. The by linear programming methods. These AFSs are shown as
role of matrix T is to transform the abstract PCs in the matrix blue polygons in Figure 4. One can see, nonetheless, that while
Ũ into the set of actual XANES spectra corresponding to pure the spectra-specific constraints limit significantly the possible
species: S = Ũ Σ̃T. The remaining part of the decomposition, C values of the T matrix elements, the solution associated with
= T−1Ṽ , will then correspond to the set of concentration the spectral decomposition is far from being unique, requiring
profiles. The problem now is how to determine the nine to impose additional constraints. To this aim, we note that the
unknown elements of the 3×3 matrix T. The number of concentration values in the matrix C should be characterized
unknowns can be decreased to six by introducing the mass by numbers between 0 and 1. Since the matrix C depends on
balance condition (i.e., the condition that the sum of all the inverse of matrix T, the concentration values are functions
concentrations of all species should be equal to 1) together of all 6 unknown elements of matrix T, requiring us to look for
with ensuring the proper normalization of the pure spectra feasible solutions in a 6-dimensional (6D) space. Here we
components. In PyFitIt, this step is realized by normalizing the relied on a brute-force approach, where we constructed 6D
XANES spectra by their variance and fixing the first row of T vectors by randomly sampling points from the three (identical)
Emax 1/2 two-dimensional AFS regions, identified in the previous step,
to a = { Emax
1
Emin
Emin
dE[ 1PCu1(E)]
2
} , where u1 and and checked whether the resulting 6D vector satisfy the
concentration constraints.
σ1PC are the first PC and the corresponding singular value, We also note here that swapping the rows in matrix T would
respectively.67 In our case, we obtained a = − 0.32. We further result in a formally different but physically identical solution
notice that the XANES spectra should be nonnegative, and (therefore, the dark blue polygons in Figure 4a−c have
that the maximal physically reasonable value of each spectrum identical shapes). To remove the ambiguity associated with the
(the W.L. amplitude) is also limited. Here, we assume that it permutation of the spectra of pure species within matrix S, we
should be lower than 1.5, a value slightly larger than the postulate that the species that we labeled as the “first”,
maximum of the W.L. of the initial state XANES spectrum. We “second”, and “third” ones should have, respectively, the
note here that higher W.L. amplitudes are not observed in largest, second largest, and the smallest contribution to the
experimental Ni K-edge data for any of the relevant reference third experimental XANES spectrum in our dataset (i.e., the
materials, nor are they observed in XANES spectra simulations spectrum collected after ca. 18 min under CO2RR). Here, we
J. Am. Chem. Soc. 2023, 145, 17351−17366
have used the third XANES spectrum because we expect all

three species to be significantly contributing to this signal. The
AFSs narrowed down based on these constraints and
conventions are marked with light blue points in Figure 4a−
c. As one can see, there is still some ambiguity in our solution
to the spectral decomposition problem. As an illustration, in
Figure 4d, we show three representative examples of possible
XANES spectra for the second component satisfying the
aforementioned constraints. This ambiguity requires us to
make additional assumptions. Among the possible solutions,
we highlight a particular one, where the first and the last
spectra in the experimental operando XANES dataset X
correspond to pure species, rather than to a mixture of
different chemical components. These solutions correspond to
an intuitively reasonable situation, where the structure of singly
dispersed, homogeneous catalysts is fully transformed under
reaction conditions, but the transformation proceeds through
an intermediate step, where an additional transient species is
Figure 5. (a) XANES spectra for the extracted pure species and (b)
present. Such a scenario is, in part, implied by our PCA results related concentration profiles (filled circles) extracted via TM
(see Figure S15). In Section 3.4, we will confirm this approach from the experimental Ni K-edge XANES data for the
hypothesis by XANES simulation results. HT Ni-TMNC sample. The inset in (a) shows a magnification of the
In situations where the first and last spectra cannot be corresponding XANES pre-edge region. The dashed lines in panel (b)
considered as corresponding to pure species, other additional represent the fits of the obtained concentration profiles with a simple
constraints need to be introduced. Some spectral profiles can model of consecutive first-order reactions model.
be fixed as corresponding to some known reference
compounds. Alternatively, imposing constraints on the
concentration profiles (based, e.g., on some kinetic models) detailed analysis involving XAS data collected with time
can also reduce the ambiguity of the result. Furthermore, in resolution better than 1 min would thus be necessary. Such an
some cases, it is possible to complement the XAS dataset with analysis, however, is outside the scope of this article.
additional spectra collected under different conditions (e.g., 3.4. Machine Learning-Assisted Identification of the
under different applied potentials, pH values, etc.) in order to Structures of the Pure Species. After isolating the XANES
enhance the appearance of some species contributing to the spectra for the pure species via a TM approach, one still needs
heterogeneous structure of the catalyst.38 to deduce the corresponding structures of these three species.
This additional assumption of the purity of the first and last Previous works have demonstrated that ab initio XANES
spectra in our matrix X leads to a single-point solution, simulations are well suited for the task of structure
indicated by the red crosses in Figure 4a−c. The final determination in single site catalysts.7,13 However, the effect
transformation matrix can be written as of possible variations of a large number of relevant local
i a a a y structural parameters (e.g., distances between atoms of
T = jjj 0.78 0.15 0.21zzz. The corresponding spectra and
different types, bonding angles, etc.) on XANES spectra need
k 0.55 1.6 0.17 {
concentrations profiles for these pure species are shown in to be considered in such analysis. Taking into account that the
Figure 5. The obtained concentration profiles appear to be XANES simulations are computationally demanding, the direct
consistent with a simple kinetic three-component model fitting of XANES spectra is thus challenging. Instead, for the
dc1(t ) c1(t ) interpretation of the obtained Ni K-edge XANES profiles, we
(consecutive first-order chemical reactions): dt
= ; employed a supervised machine learning (SML)-based XANES
1 2
dc 2(t ) c1(t ) c 2(t ) dc3(t ) c 2(t ) fitting approach, as implemented in the PyFitIt code. Using
= ; = . Here, τ1→2 and τ2→3 are
dt 1 2 2 3 dt 2 3 this methodology, we first established a non-linear relation
characteristic times (inverse rate constants) for the trans- between the local structural parameters for the Ni species and
formations of the first species into the second and of the the corresponding XANES profiles. For this purpose, we relied
second species into the third, respectively. By solving the on relatively small training sets of ab initio XANES spectra
differential equation system and fitting the obtained concen- (from 200 to 1000 spectra) obtained using the FDMNES
tration profiles, we obtain τ1→2 = 1.0 min and τ2→3 = 1.4 min. code.68,69 We highlight here that the FDMNES code is able to
Overall, the retrieved concentration profiles follow reasonably successfully reproduce XANES spectra of relevant reference
this simple kinetic model, thus confirming also the validity of compounds such as the Ni K-edge XANES spectrum of the
all the physical/chemical assumptions described above. We NiO compound, see Section S16 of the SI text, which gives us
note here that the values of the characteristic times τ1→2 and confidence in the reliability of these simulations for our
τ2→3 are much smaller than the time resolution of our purposes.
experiment (as indicated in Section 2.4, ca. 9 min). The limited The set of structural parameters for which we performed
experimental time resolution results in the ambiguity of the XANES calculations was determined based on an Adaptive
exact time values that should be associated with each datapoint Sampling approach (active learning).70 This sampling scheme
in Figure 5b, This could explain why the concentration profile allows us to ensure higher density of calculated spectra in the
for the intermediate species at t = 8 min deviates from the regions of the structural parameters space, where small
predictions of our kinetic model. To confirm (or disprove) the variations in structure parameters could result in large
validity of such a simple kinetic model for this process, a more variations of the corresponding XANES spectra. Thus, the
J. Am. Chem. Soc. 2023, 145, 17351−17366
total number of required XANES calculations is minimized

(see Figure 7a).71
We used the structural parameters sets and the respective
calculated XANES spectra as the training datasets for an SML
algorithm. Here, the objective is to construct a mathematical
model μ̂ (E; p) that is a function of the energy E and a set of
the corresponding structural parameters p. μ̂ (E; p) must be
able to interpolate between the points in the training datasets
and thus to provide XANES spectra for those structural
parameters p for which the explicit XANES calculations have
not been carried out. For this purpose, we relied on a radial
basis function algorithm (RBF). Here, the SML model is
constructed as a linear combination of a set of basis functions:
N
(E ; p) = i = 1 wi(E)K ( p pi ) + PE(p), where N is the
number of calculated spectra in the training set, K(r) is a linear
radial basis function, and PE(p) is a second-order polynomial
depending on p with energy-dependent coefficients. We
obtained then the unknown factors wi and the polynomial
coefficients by requesting that the model should describe the
XANES spectra in the training data set as closely as possible.
To this end, wi and the polynomial coefficients were obtained
through the least squares and ridge regression methods,
respectively.30,51,71 We tested the accuracy of the trained SML
routine using 10-fold cross-validation. Herein, the training set
was divided randomly into 10 parts, nine of which were used
for the algorithm training, while the last one for the validation.
In accordance to the PyFitIt convention, the SML accuracy is
2
i i i
then defined as 1 2 , where the summation is
i i
M
carried out over all the points in the validation data set. μi is
the spectrum calculated by FMDNES for the ith combination
of parameters pi, μ̂ i is the approximated spectrum yielded by
SML, while M is the average XANES spectrum evaluated over Figure 6. Set of possible deformations applied to the structure models
all the spectra composing the training set. In all cases discussed used for the construction of the SML training data set and XANES
below, we achieved an accuracy value higher than 0.98, fitting.
indicating that our SML yields spectra closely match those
obtained using the exact FDMNES simulations. The obtained pure components, respectively. Here we keep the guessed C−
SML routine can then be used to efficiently generate XANES O distance close to 1.1 Å. We used these structural models
spectra for different sets of structure parameters, and thus with varied interatomic distances and angles to construct the
allows us the direct fitting of XANES spectra for the pure training data set and, consequently, to establish the
species identified by the TM. corresponding interpolating functions μ̂ (E; p). The parameters
Based on the qualitative analysis of the XANES data and on that were allowed to vary and the ranges of their possible
the metal single sites models proposed in the prior literature values are shown in Figure 6 and listed in Table 1.
for TMNC catalysts for the CO2RR2,6,72,73 and ORR,7,13,74 we We have also considered other possible distortions of the
focus on several distinct families of possible candidate structure models, such as the shift of the Ni atom out of the
structures for different states of our HT-Ni-TMNC sample. pyridine ring plane, as suggested by refs 2, 6, 22, 72−74, 76,
First, as discussed above, our qualitative analysis of the XANES 77, see Figure S18 and Table S5. However, we found that such
data suggested that in the as-prepared state, Ni appears to be distortions do not agree with the observed W.L. and post-edge
octahedrally coordinated, with a local structure resembling that XANES features in our spectra (see Section S8 of the SI). In
of nickel oxide. Therefore, we considered a model where a Ni addition, we tried also to quantify and classify the importance
atom is surrounded by four pyridine ligands sited at 1.95 Å,75 of the different chosen structural parameters on the XANES
with two additional axial oxygen atoms added at a distance of spectra introducing a normalized standard deviation estima-
2.0 Å from the Ni absorber (see model 1 in Figure 6). We tor,30 as explained in Section S9 and Table S8. From the
expect this model to be representative of the structure of our obtained results one can see that the largest variations of the
as-prepared sample. Furthermore, our qualitative analysis XANES spectra are associated with the stretching of the
suggested that under CO 2RR conditions, the catalyst pyridine ring and with the changes in the Ni−O and Ni−CO
experiences significant changes in the local structure, but the bond lengths. Smaller, but not negligible, changes in the
singly dispersed cationic nature of Ni species is preserved. We XANES spectra are also caused by the in-plane displacement of
have found that the replacement of one (model 2 in Figure 6) the Ni atoms and by the variations in the C−O distance. Less
or both (model 3 in Figure 6) axial O atoms with a CO ligand pronounced effects on the XANES spectra are caused by the
in the afore discussed model leads to XANES spectra change of the Ni − C − O angle, while the rotation of the CO
resembling the determined spectra for the second and third group around the C−Ni−C axis, analogous to that suggested
J. Am. Chem. Soc. 2023, 145, 17351−17366
Table 1. List of Structural Parameters for the Models Shown derived by the TM method from the experimental data.
in Figure 6 Employed in the Fit of the XANES Spectra for Technical details concerning the choice of the FDMNES
Pure Components simulations parameters and parameters of spectra convolution,
as well as more details about the XANES fitting procedure are
range of
parameter description variation given in Sections 6 and 7 of the SI text. The final best-fit results
MODEL 1 (COMPONENT 1) are shown in Figure 8.
p1 contraction/expansion of the pyridine ring [−0.2: +0.2] Å
p2 shift of the Ni atom and the two O atoms [0: +0.3] Å
toward the edge of the pyridine ring
p3 contraction/expansion of the two Ni−O [−0.2: +0.2] Å
bonds
MODEL 2 (COMPONENT 2)
p2 shift of the Ni atom, the O atom, and the CO [0: +0.3] Å
group toward the edge of the pyridine ring
p3 contraction/expansion of the axial Ni−C [−0.2: +0.2] Å
bond
p4 contraction/expansion of the Ni−O bond [−0.2: +0.2] Å
p5 contraction/expansion of the C−O bond [−0.2: +0.2] Å
p6 Ni−C−O bond angle [135: 180] °
MODEL 3 (COMPONENT 3)
p2 shift of the Ni atom, and of the CO groups [0: +0.3] Å
toward the edge of the pyridine ring
p3 contraction/expansion of the axial Ni−C [−0.1: +0.2] Å
bonds
p4 contraction/expansion of the C−O bonds [−0.2: +0.2] Å
p5 Ni−C−O bond angle [135: 180]°
Figure 8. Comparison of the XANES components for pure species, as

for the Cu-CO sites in ref 78, in turn, did not affect the extracted from the experimental data, with the best-fit results. Insets
calculated XANES spectra significantly and we thus concluded show the final structure models obtained in the XANES fitting.
that our approach is insensitive to this parameter (see Section
S9 of the SI text) and we thus neglected it in the fit.
For the discussed sets of structure models, we carry out
XANES simulations (Figure 7b−d), train the interpolating Tables 2 and S10 report the final refined structural
functions, and employ them to fit the XANES components parameters for all structure models, in particular, average
Figure 7. (a) Points in a structural parameter space obtained using the adaptive sampling employed to establish the μ̂ (E; p) interpolating functions
for model 1 depicted in Figure 6. (b) Calculated spectra for the structure parameters corresponding to the points indicated in (a). (c, d)
Representative calculated spectra for models 2 and 3 are shown in Figure 6.
J. Am. Chem. Soc. 2023, 145, 17351−17366
Table 2. Interatomic Distances and Ni−CO Ligand Angles, effects. Thus, we achieve the best possible agreement between
Calculated for the Final Structure Models Obtained through the experimental Ni K-edge EXAFS data and the theoretical
the XANES Fitting Shown in Figure 8a EXAFS spectra calculated for the given structure model. Note
that the sensitivity of EXAFS to disorder effects is much higher
bond/angle XANES best-fit value
than that of XANES. While these effects can be largely
MODEL 1 (COMPONENT 1). Misfit (Fmin): 0.02% neglected in the XANES modeling, they need to be taken into
Ni−N (two pyridinic N atoms closer to Ni) 1.75 (2) Å account in the interpretation of the EXAFS data. The maximal
Ni−N (two more distant pyridinic N atoms) 2.03 (2) Å allowed atom displacements from their starting positions in our
Ni−O 2.00 (2) Å RMC simulations are 0.4 Å; thus, the overall 3D structure of
MODEL 2 (COMPONENT 2). Misfit (Fmin): 0.02%
the material and coordination numbers do not change in the
Ni−C (C of the CO group) 1.91 (3) Å
RMC-EXAFS fit. The RMC approach allows us to fit EXAFS
Ni−N (two pyridinic N atoms closer to Ni) 1.69 (2) Å
data and account explicitly for the contributions of distant
Ni−N (two more distant pyridinic N atoms) 1.97 (2) Å
coordination shells, multiple scattering effects, as well as for the
Ni−O 2.07 (4) Å
non-Gaussian shapes of bond-length distributions. Thus, we
C−O (C and O of the CO group) 1.04 (6) Å
can fully benefit from all the information, encoded in EXAFS
Ni − C − Ô bond angle 149 (5)°
data, and can reliably fit the EXAFS data for strongly
MODEL 3 (COMPONENT 3). Misfit (Fmin): 0.02%
disordered materials such as TMNCs. All of these factors
Ni−C (C of the CO group) 1.78 (3) Å
make RMC well suited for the interpretation of EXAFS data in
Ni−N (two pyridinic N atoms closer to Ni) 1.82 (3) Å
TMNC catalysts, providing that the initial structure model is
Ni−N (two more distant pyridinic N atoms) 2.17 (3) Å
C−O (C and O of the CO group) 1.27 (4) Å
available.
Ni − C − Ô bond angle 170 (5)°
To fit the EXAFS spectrum for the as-prepared HT-Ni-
a TMNC, we use as an initial structure the model obtained from
The uncertainties of the last digit are shown in parentheses. They are XANES analysis (Figure 8a). To fit the EXAFS spectrum for
derived from the ones shown in Table S10. For the details of the
misfit quantity (Fmin) calculations, see Section S7 of the SI text. the final state of the HT-Ni-TMNC catalyst under CO2RR
conditions, we start with the structure model as shown in
Figure 8c. Considering that a single structure model, as
distances between Ni and its nearest neighbors plus the Ni− depicted in Figure 8, contains only one absorbing Ni atom,
CO ligand bond angle. while the experimental EXAFS spectra are averaged over a
To demonstrate that the obtained solution is well within the large number of Ni species, to properly describe the bond
region where our constructed interpolating functions are length distributions the structure models that are optimized in
accurate, in Figure S25, we compared the XANES spectra, our RMC-EXAFS simulations consist of 64 replicas of the
obtained by using the constructed interpolation functions, with models shown in Figure 8, placed at sufficiently large distances
the spectra directly calculated by FDMNES for the final from each other. A similar approach was previously used by us
structure models. to fit the EXAFS spectra of small oxide39 and metal79
3.5. Validation of the Structural Models via Reverse nanoparticles. RMC fits of the EXAFS spectra are performed
Monte Carlo EXAFS Analysis. To check whether the in k and R-space simultaneously, using wavelet transform to
structure models derived based on XANES data fitting are compare experimental and theoretically simulated EXAFS
consistent also with the available EXAFS data, we perform spectra. The refs 39, 41 provide more details on RMC-EXAFS
reverse Monte Carlo (RMC) simulations.52 In the RMC- simulations.
EXAFS approach, we start with a 3D structure model obtained Figures 9 and S27 show the RMC fits for the as-prepared
using the XANES fitting procedure and slightly move the HT-Ni-TMNC catalyst and HT-Ni-TMNC catalyst in its final
atoms in the model around their initial positions in a random state under CO2RR conditions. RMC simulations yield
process in order to include the thermal and static disorder structure models that are in an excellent agreement with the
Figure 9. (a) Results of the RMC-EXAFS simulations. Comparison of Fourier-transformed experimental Ni K-edge EXAFS spectra for as-prepared
HT-Ni-TMNC and for HT-Ni-TMNC catalysts in the final state under CO2RR conditions with the corresponding theoretical EXAFS data,
calculated for the final structure models obtained in RMC simulations. (b) RDFs calculated from the atomic coordinates in the final RMC models.
Partial RDFs for Ni−N, Ni−O, and Ni−C atoms are shown. FT-EXAFS spectra and RDFs are shifted vertically for clarity. Fourier Transforms in
(a) are carried out in the k-range between 3 and 10 Å−1. RMC fits are carried out in k- and R-spaces simultaneously using wavelet transform, in the
k-range between 3 and 10 Å−1 and the R-range between 1 and 6 Å, including multiple scattering contributions with up to 5 backscattering atoms.
J. Am. Chem. Soc. 2023, 145, 17351−17366
Figure 10. Main reaction steps suggested on the basis of the PCA and the XANES and EXAFS fitting results.
experimental EXAFS data, both in wavelet space (Figure S27) significantly the relative intensities of the pre-edge features.
and in R-space (Figure 9a). Thus, this confirms that the The correct sampling of these deviations in the XANES
structure models obtained from XANES fitting do not modeling would require ab initio molecular dynamics or
contradict the available experimental EXAFS data. The Metropolis Monte Carlo approaches,81 which are outside the
structures obtained from the RMC-EXAFS fits are best scope of this article. Here, we just note that, statistically, there
analyzed in terms of radial distribution of atoms of different could be some configurations characterized by one Ni−O or
types around the central Ni atom. Figure 9b shows such partial Ni−CO ligand located closer to the Ni site, i.e., with a more
radial distribution functions (RDFs) both, for the as-prepared pyramidal geometry characterized by a more intense 1s → 4pz
HT-Ni-TMNC catalyst and the same catalyst in its final state pre-edge feature.
under CO2RR conditions. The broad, split shapes of the RDF Finally, combining the results emerging from the PCA, TM
peaks indicate the strongly disordered structure of the HT-Ni- method, and the XANES and EXAFS analyses, it is possible to
TMNC material. In particular, we note that all RDF peaks of propose the mechanism of the Ni speciation during the
Ni−N contribution in the as-prepared sample are split, CO2RR reaction as depicted in Figure 10.
suggesting that the Ni atom is located at an off-center position The initial state, before the reaction starts, is represented by
with respect to the square defined by 4 nearest N atoms, in a Ni site in a nearly octahedral coordination, where Ni is
agreement with the conclusions from XANES fitting. coordinated with two axial O atoms at ca. 2 Å (belonging, e.g.,
Furthermore, in our RMC-EXAFS results for the as-prepared to two O atoms from the adsorbed water molecules) and
sample, the average Ni−O bond length is similar to that of the shifted from the pyridines ring center by 0.2 Å. During
longest Ni−N bond, and both are ca. 2.0 Å long, which is in CO2RR, one O atom appears to be replaced by a CO ligand,
excellent agreement with our results from XANES data fitting deriving from the activation of the CO2 molecule on the Ni
(Table 2). For the sample under CO2RR, the width of the Ni−
center. We note that spectra very similar to our Ni K-edge
N RDF peaks is similar to that in the as-prepared sample, but
spectrum for the intermediate state were reported in the
the individual subpeaks cannot be resolved, suggesting a
literature for the [Fe(tren(py)3)]2+ ground state82,83 (Fe K-
further increase in the structural disorder under reaction
conditions. The shortest Ni−C bond, associated with the edge XANES) and for Co-based TMNC complexes (Co K-
interactions between Ni and CO adsorbates, results in a very edge XANES).84 These Fe K-edge and Co K-edge XANES
broad RDF peak between 1.4 and 2.2 Å, with the maximum at spectra in the prior works were attributed to octahedrally
ca. 1.6−1.8 Å, in excellent agreement with the Ni−CO bond coordinated species, resembling the structure model suggested
length obtained in the XANES fit (ca. 1.78 Å). Note, however, by our XANES fit. For this intermediate state, we have found
that due to a very large disorder factor, the contribution of this that the position of Ni with respect to the pyridine ring is
bond to the experimental EXAFS spectrum is very small, and similar to that in the as-prepared state, while the ring itself is
thus, the sensitivity of EXAFS to the presence of this bond is slightly more contracted (p1 = − 0.09 Å). The length of the
expected to be low. Ni−O bond is increased to 2.07 Å, while the Ni−CO bond
length is found to be ca. 1.91 Å, with a Ni − C − Ô angle of
4. DISCUSSION 149° and with a C−O distance of 1.04 Å.
The intermediate state is consequently converted into the
Focusing on Figures 8 and 9, we conclude that both the
simulated XANES and EXAFS spectra for the final structure final state through the substitution of the remaining O atom
models agree well with the available experimental data. This with an additional CO ligand. The fit of the final state with a
gives us confidence that the atomistic structure models, geometry possessing a Ni site with a lower coordination
constructed based on the TM method and XANES fitting, number (i.e., with none or just one CO group) did not result
are representative of the structures of the as-prepared catalysts, in a good agreement with the experimental data (Section S8).
intermediate state, and the final state achieved under CO2RR The XANES fitting results for the final catalyst state suggest
conditions. In the case of the XANES analysis, a particularly an expansion of the pyridine ring of +0.07 Å, a larger shift of
good agreement between the experimental and simulated the Ni site outside the ring center and the presence of two CO
spectra is observed in the post-edge region. A reasonable groups sited at a distance of 1.78 Å from the Ni site. In this
agreement between the XANES simulations and the extracted configuration, the first coordination shell for Ni effectively
spectra for the pure species is achieved also in the pre-edge consists of four atoms: two N and two O. The remaining two
region, despite the fact that the approximations employed in N atoms are found to be located at larger distances (ca. 2.17
the FDMNES code are known to be less accurate in this Å). The XANES analysis indicates also that the Ni−C−O
region.30,80 angle is ca. 170°. On the other hand, looking at the map of
Moreover, as evident from our RMC-EXAFS results, the correlations between different variables (Figure S26), it is
structures around each Ni site are flexible, resulting in large evident that all the Ni−C−O angle values in the range between
structural disorder. By attempting to reproduce the exper- the 165° and 175° could also provide a good agreement with
imental XANES data with a single structural model, the the experimental data. In particular, a nearly collinear Ni−C−
structural disorder effect is neglected. This could affect O configuration with just slightly expanded average C−O
J. Am. Chem. Soc. 2023, 145, 17351−17366
distance could also fit well the XANES spectrum provided by species. Our results confirm that the single Ni sites are the
the TM method. active species for the CO2RR, but also reveal their dynamic,
The shift of the Ni site inside the pyridines ring deserves a heterogeneous nature and adaptation to the reaction
special mention. As shown in Section S9, we found this conditions. In particular, our data suggest the direct influence
parameter to be highly relevant for obtaining a good fit with of the interactions between the Ni site and CO adsorbates on
the experimental data. The off-center displacement of Ni is, in the XANES and EXAFS spectra, allowing us to get insight into
fact, necessary to reproduce properly the W.L. and post edge the reaction mechanism, and highlighting the importance of
regions for all three spectral components identified by the TM- operando spectroscopic investigations. We believe that the
XANES approach. RMC-EXAFS approach reinforces this approach developed here will be helpful for understanding the
finding yielding a very broad distribution of Ni−N bonds for active states of other TMNC catalysts as well, including single
the working catalyst, suggesting, again, a strong off-center atom catalysts based on different transition metals and could
displacement of Ni. The symmetry breaking in this case is also be applied for the understanding of other reactions, such
likely explained by the presence of defects in the carbon as the oxygen reduction reaction, the photocatalytic CO2
support. Indeed, both experimental observations85 and conversions,93,94 and many others where single metal sites
theoretical simulations86 suggest that Ni−N sites are not are considered to be attractive catalytic motifs.
distributed uniformly over the carbon support, but are
localized in the vicinity of the edges of graphene layers (e.g.,
next to the pores in the support). This breaks the bi-directional
symmetry of the system, allowing the distortions of the Ni−N4
■ ASSOCIATED CONTENT
* Supporting Information
sı
structural motif. The presence of nonsquare-like motifs and The Supporting Information is available free of charge at
their importance in electrocatalytic activity were reported also https://pubs.acs.org/doi/10.1021/jacs.3c04826.
for Fe−N−C catalysts, where the distorted geometry can be
tracked by Mössbauer spectroscopy.87 We also note that such Details of cells used for electrocatalytic activity and
distortions in the M−N4 structural motif may appear in EXAFS operando XAS measurements, ex situ characterization of
spectra as a decrease in the effective M−N coordination the catalysts and its electrocatalytic activity measure-
number. And, indeed, several recent EXAFS-based works on ments, supplementary XANES and EXAFS data, results
M−N−C catalysts suggest that the presence of metal sites with of wavelet transform analysis of EXAFS spectra;
apparently reduced M−N coordination numbers could be description of the principal component analysis,
decisive for the CO2RR activity.24,59,88−92 FDMNES simulations, XANES normalization and
Overall, the scheme shown in Figure 10 with the gradual fitting; XANES fitting results with additional structure
replacement of O or OH axial ligands by CO adsorbates under models; analysis of the importance of individual fitting
working conditions matches well the one recently reported for parameters; validation of the accuracy of the employed
molecular Co single site catalysts for photocatalytic CO2 machine learning method; details of the RMC-EXAFS
reduction,84 suggesting that the mechanism proposed here simulations; XANES and EXAFS data for the Ni-TMNC
could be common in single site catalysts. catalyst and XANES fitting results for this catalyst
One should note that according to our findings XANES (PDF)
spectra are extremely sensitive to the details of the Ni local
environment. Indeed, even slight changes in the bond distances
and bond angles result in significant variations in the XANES
features. As a result, the final XANES spectra could change
■ AUTHOR INFORMATION
Corresponding Authors
depending on the sample preparation and from one sample to Janis Timoshenko − Department of Interface Science, Fritz-
another, even if the structure around the active Ni site remains Haber Institute of the Max Planck Society, 14195 Berlin,
qualitatively similar. To illustrate that, in Sections S14 and 15 Germany; Email: janis@fhi-berlin.mpg.de
of the SI we show the results stemming from an analogous Beatriz Roldan Cuenya − Department of Interface Science,
XANES analysis carried out for a Ni-TMNC sample that, Fritz-Haber Institute of the Max Planck Society, 14195
unlike the HT-Ni-TMNC sample, was not exposed to the Berlin, Germany; orcid.org/0000-0002-8025-307X;
high-temperature treatment. We note that the spectrum for the Email: roldan@fhi-berlin.mpg.de
final state of the Ni-TMNC sample differs remarkably from the
spectrum for the HT-Ni-TMNC. Nonetheless, the final Authors
structure, featuring the Ni−N4 site coordinated with two CO Andrea Martini − Department of Interface Science, Fritz-
ligands, is similar in both cases. This explains the similarity in Haber Institute of the Max Planck Society, 14195 Berlin,
catalytic properties observed for both of these catalysts. Germany; orcid.org/0000-0001-8820-2157
Dorottya Hursán − Department of Interface Science, Fritz-
5. CONCLUSIONS Haber Institute of the Max Planck Society, 14195 Berlin,
In this work, we provided the first quantitative analysis of the Germany; Present Address: Department of Physical
structural changes experienced by Ni-based TMNC catalysts Chemistry and Materials Science, University of Szeged,
under CO2RR conditions. Only through the combination of Aradi Square 1, Szeged 6720, Hungary
time-resolved operando XAS measurements, unsupervised and Martina Rüscher − Department of Interface Science, Fritz-
supervised machine learning approaches and simulation-based Haber Institute of the Max Planck Society, 14195 Berlin,
XANES and EXAFS data fitting, we were able to identify the Germany
structures of the initial catalysts and those of their intermediate Felix Haase − Department of Interface Science, Fritz-Haber
and final states under reaction conditions, as well as to Institute of the Max Planck Society, 14195 Berlin, Germany;
reconstruct the concentration profiles of the different Ni orcid.org/0000-0003-1646-4312
J. Am. Chem. Soc. 2023, 145, 17351−17366
Clara Rettenmaier − Department of Interface Science, Fritz- Atom Co-N-5 Catalytic Site: A Robust Electrocatalyst for CO2
Haber Institute of the Max Planck Society, 14195 Berlin, Reduction with Nearly 100% CO Selectivity and Remarkable Stability.
Germany J. Am. Chem. Soc. 2018, 140, 4218−4221.
Eduardo Ortega − Department of Interface Science, Fritz- (12) Mehmood, A.; Gong, M. J.; Jaouen, F.; Roy, A.; Zitolo, A.;
Khan, A.; Sougrati, M. T.; Primbs, M.; Bonastres, A. M.; Fongalland,
Haber Institute of the Max Planck Society, 14195 Berlin,
D.; et al. High loading of single atomic iron sites in Fe-NC oxygen
Germany; orcid.org/0000-0002-0643-5190 reduction catalysts for proton exchange membrane fuel cells. Nat.
Ane Etxebarria − Department of Interface Science, Fritz-Haber Catal. 2022, 5, 311−323.
Institute of the Max Planck Society, 14195 Berlin, Germany (13) Zitolo, A.; Ranjbar-Sahraie, N.; Mineva, T.; Li, J. K.; Jia, Q. Y.;
Complete contact information is available at: Stamatin, S.; Harrington, G. F.; Lyth, S. M.; Krtil, P.; Mukerjee, S.;
https://pubs.acs.org/10.1021/jacs.3c04826 et al. Identification of catalytic sites in cobalt-nitrogen-carbon
materials for the oxygen reduction reaction. Nat. Commun. 2017, 8,
Funding 957.
(14) Saveleva, V. A.; Kumar, K.; Theis, P.; Salas, N. S.; Kramm, U. I.;
Open access funded by Max Planck Society. Jaouen, F.; Maillard, F.; Glatzel, P. Fe-N-C Electrocatalyst and Its
Notes Electrode: Are We Talking about the Same Material? ACS Appl.
The authors declare no competing financial interest. Energy Mater. 2023, 6, 611−616.
■ ACKNOWLEDGMENTS
XAS measurements were carried out at the KMC-3 XPP
(15) Saveleva, V. A.; Ebner, K.; Ni, L. M.; Smolentsev, G.; Klose, D.;
Zitolo, A.; Marelli, E.; Li, J. K.; Medarde, M.; Safonova, O. V.; et al.
Potential-Induced Spin Changes in Fe/N/C Electrocatalysts Assessed
by In Situ X-ray Emission Spectroscopy. Angew. Chem., Int. Ed. 2021,
instrument at the BESSY II electron storage ring operated by 60, 11707−11712.
the Helmholtz-Zentrum Berlin für Materialien und Energie. (16) Artyushkova, K.; Serov, A.; Rojas-Carbonell, S.; Atanassov, P.
D.H. thanks the funding provided by the Alexander von Chemistry of Multitudinous Active Sites for Oxygen Reduction
Humboldt Foundation. F.H. and C.R. acknowledge support by Reaction in Transition Metal-Nitrogen-Carbon Electrocatalysts. J.
the IMPRS for Elementary Processes in Physical Chemistry. Phys. Chem. C 2015, 119, 25917−25928.
■ REFERENCES
(1) Handoko, A. D.; Wei, F. X.; Jenndy; Yeo, B. S.; Seh, Z. W.
(17) Weng, Z.; Wu, Y. S.; Wang, M. Y.; Jiang, J. B.; Yang, K.; Huo, S.
J.; Wang, X. F.; Ma, Q.; Brudvig, G. W.; Batista, V. S.; et al. Active
sites of copper-complex catalytic materials for electrochemical carbon
dioxide reduction. Nat. Commun. 2018, 9, 415.
Understanding heterogeneous electrocatalytic carbon dioxide reduc-
tion through operando techniques. Nat. Catal. 2018, 1, 922−934. (18) Karapinar, D.; Huan, N. T.; Sahraie, N. R.; Li, J. K.; Wakerley,
(2) Liang, S. Y.; Huang, L.; Gao, Y. S.; Wang, Q.; Liu, B. D.; Touati, N.; Zanna, S.; Taverna, D.; Tizei, L. H. G.; Zitolo, A.;
Electrochemical Reduction of CO2 to CO over Transition Metal/N- et al. Electroreduction of CO2 on Single-Site Copper-Nitrogen-Doped
Doped Carbon Catalysts: The Active Sites and Reaction Mechanism. Carbon Material: Selective Formation of Ethanol and Reversible
Adv. Sci. 2021, 8, No. 2102886. Restructuration of the Metal Sites. Angew. Chem., Int. Ed. 2019, 58,
(3) Ju, W.; Bagger, A.; Hao, G. P.; Varela, A. S.; Sinev, I.; Bon, V.; 15098−15103.
Roldan Cuenya, B.; Kaskel, S.; Rossmeisl, J.; Strasser, P. Under- (19) Wang, Y.; You, L. M.; Zhou, K. Origin of the N-coordinated
standing activity and selectivity of metal-nitrogen-doped carbon single-atom Ni sites in heterogeneous electrocatalysts for CO2
catalysts for electrochemical reduction of CO2. Nat. Commun. 2017, reduction reaction. Chem. Sci. 2021, 12, 14065−14073.
8, 944. (20) Bordiga, S.; Groppo, E.; Agostini, G.; van Bokhoven, J. A.;
(4) Moller, T.; Ju, W.; Bagger, A.; Wang, X. L.; Luo, F.; Thanh, T. Lamberti, C. Reactivity of Surface Species in Heterogeneous Catalysts
N.; Varela, A. S.; Rossmeisl, J.; Strasser, P. Efficient CO2 to CO Probed by In Situ X-ray Absorption Techniques. Chem. Rev. 2013,
electrolysis on solid Ni−N−C catalysts at industrial current densities. 113, 1736−1850.
Energy Environ. Sci. 2019, 12, 640−647. (21) Timoshenko, J.; Frenkel, A. I. ″Inverting″ X-ray Absorption
(5) Timoshenko, J.; Roldan Cuenya, B. In Situ/Operando Spectra of Catalysts by Machine Learning in Search for Activity
Electrocatalyst Characterization by X-ray Absorption Spectroscopy. Descriptors. ACS Catal. 2019, 9, 10192−10211.
Chem. Rev. 2021, 121, 882−961. (22) Varela, A. S.; Ju, W.; Bagger, A.; Franco, P.; Rossmeisl, J.;
(6) Varela, A. S.; Ju, W.; Strasser, P. Molecular Nitrogen-Carbon Strasser, P. Electrochemical Reduction of CO2 on Metal-Nitrogen-
Catalysts, Solid Metal Organic Framework Catalysts, and Solid Doped Carbon Catalysts. ACS Catal. 2019, 9, 7270−7284.
Metal/Nitrogen-Doped Carbon (MNC) Catalysts for the Electro- (23) Bi, W. T.; Li, X. G.; You, R.; Chen, M. L.; Yuan, R. L.; Huang,
chemical CO2 Reduction. Adv. Energy Mater. 2018, 8, No. 1703614. W. X.; Wu, X. J.; Chu, W. S.; Wu, C. Z.; Xie, Y. Surface
(7) Zitolo, A.; Goellner, V.; Armel, V.; Sougrati, M. T.; Mineva, T.; Immobilization of Transition Metal Ions on Nitrogen-Doped
Stievano, L.; Fonda, E.; Jaouen, F. Identification of catalytic sites for Graphene Realizing High-Efficient and Selective CO2 Reduction.
oxygen reduction in iron- and nitrogen-doped graphene materials. Adv. Mater. 2018, 30, No. 1706617.
Nat. Mater. 2015, 14, 937−942. (24) Fan, Q.; Hou, P. F.; Choi, C. H.; Wu, T. S.; Hong, S.; Li, F.;
(8) Abbas, S. A.; Song, J. T.; Tan, Y. C.; Nam, K. M.; Oh, J.; Jung, K. Soo, Y. L.; Kang, P.; Jung, Y. S.; Sun, Z. Y. Activation of Ni Particles
D. Synthesis of a Nickel Single-Atom Catalyst Based on Ni−N4−xCx into Single Ni−N Atoms for Efficient Electrochemical Reduction of
Active Sites for Highly Efficient CO2 Reduction Utilizing a Gas CO2. Adv. Energy Mater. 2020, 10, No. 1903068.
Diffusion Electrode. ACS Appl. Energy Mater. 2020, 3, 8739−8745. (25) Hou, Y.; Huang, Y.-B.; Liang, Y.-L.; Chai, G.-L.; Yi, J.-D.;
(9) Jiang, K.; Siahrostami, S.; Zheng, T. T.; Hu, Y. F.; Hwang, S.; Zhang, T.; Zang, K.-T.; Luo, J.; Xu, R.; Lin, H.; et al. Unraveling the
Stavitski, E.; Peng, Y. D.; Dynes, J.; Gangisetty, M.; Su, D.; et al. Reactivity and Selectivity of Atomically Isolated Metal–Ni-
Isolated Ni single atoms in graphene nanosheets for high-performance trogen Sites Anchored on Porphyrinic Triazine Frameworks for
CO2 reduction. Energy Environ. Sci. 2018, 11, 893−903. Electroreduction of CO2. CCS Chem. 2019, 1 (4), 384−395.
(10) Millet, M. M.; Algara-Siller, G.; Wrabetz, S.; Mazheika, A.; DOI: 10.31635/ccschem.019.20190011.
Girgsdies, F.; Teschner, D.; Seitz, F.; Tarasov, A.; Leychenko, S. V.; (26) Koshy, D. M.; Chen, S. C.; Lee, D. U.; Stevens, M. B.;
Schlogl, R.; et al. Ni Single Atom Catalysts for CO2 Activation. J. Am. Abdellah, A. M.; Dull, S. M.; Chen, G.; Nordlund, D.; Gallo, A.;
Chem. Soc. 2019, 141, 2451−2461. Hahn, C.; et al. Understanding the Origin of Highly Selective CO2
(11) Pan, Y.; Lin, R.; Chen, Y. J.; Liu, S. J.; Zhu, W.; Cao, X.; Chen, Electroreduction to CO on Ni, N-doped Carbon Catalysts. Angew.
W. X.; Wu, K. L.; Cheong, W. C.; Wang, Y.; et al. Design of Single- Chem., Int. Ed. 2020, 59, 4043−4050.
J. Am. Chem. Soc. 2023, 145, 17351−17366
(27) Li, X. G.; Bi, W. T.; Chen, M. L.; Sun, Y. X.; Ju, H. X.; Yan, W. XPS binding energy shift for nitrogen-containing graphene-like
S.; Zhu, J. F.; Wu, X. J.; Chu, W. S.; Wu, C. Z.; et al. Exclusive Ni−N- structures. Chem. Commun. 2013, 49, 2539−2541.
4 Sites Realize Near-Unity CO Selectivity for Electrochemical CO2 (45) Matanovic, I.; Artyushkova, K.; Strand, M. B.; Dzara, M. J.;
Reduction. J. Am. Chem. Soc. 2017, 139, 14889−14892. Pylypenko, S.; Atanassov, P. Core Level Shifts of Hydrogenated
(28) Zheng, Y. L.; Han, J. Y.; Takele, L.; Xie, F.; Zhang, Y.; Sun, J. Pyridinic and Pyrrolic Nitrogen in the Nitrogen-Containing
Q.; Han, B.; Chen, J.; Gao, Y.; Tang, Z. Y. Poly-phenylenediamine- Graphene-Based Electrocatalysts: In-Plane vs Edge Defects. J. Phys.
derived atomically dispersed Ni sites for the electroreduction of CO2 Chem. C 2016, 120, 29225−29232.
to CO. Inorg. Chem. Front. 2019, 6, 1729−1734. (46) Ortega, K. F.; Arrigo, R.; Frank, B.; Schlogl, R.; Trunschke, A.
(29) Benfatto, M.; Della Longa, S.; Pace, E.; Chillemi, G.; Padrin, C.; Acid-Base Properties of N-Doped Carbon Nanotubes: A Combined
Natoli, C. R.; Sanna, N. MXAN: A new program for ab-initio Temperature-Programmed Desorption, X-ray Photoelectron Spec-
structural quantitative analysis of XANES experiments. Comput. Phys. troscopy, and 2-Propanol Reaction Investigation. Chem. Mater. 2016,
Commun. 2021, 265, No. 107992. 28, 6826−6839.
(30) Guda, A. A.; Guda, S. A.; Lomachenko, K. A.; Soldatov, M. A.; (47) Ott, S.; Du, F.; Luna, M. L.; Dao, T. A.; Selve, S.; Roldan
Pankin, I. A.; Soldatov, A. V.; Braglia, L.; Bugaev, A. L.; Martini, A.; Cuenya, B.; Orfanidi, A.; Strasser, P. Property-reactivity relations of
Signorile, M.; et al. Quantitative structural determination of active N-doped PEM fuel cell cathode catalyst supports. Appl. Catal., B
sites from in situ and operando XANES spectra: From standard ab 2022, 306, No. 121118.
initio simulations to chemometric and machine learning approaches. (48) Spath, F.; Zhao, W.; Gleichweit, C.; Gotterbarm, K.; Bauer, U.;
Catal. Today 2019, 336, 3−21. Hofert, O.; Steinruck, H. P.; Papp, C. Hydrogenation and
(31) Hayakawa, K.; Hatada, K.; D’Angelo, P.; Della Longa, S.; dehydrogenation of nitrogen-doped graphene investigated by X-ray
Natoli, C. R.; Benfatto, M. Full quantitative multiple-scattering photoelectron spectroscopy. Surf. Sci. 2015, 634, 89−94.
analysis of X-ray absorption spectra: Application to potassium (49) Zizak, I. G. P. The KMC-3 XPP beamline at BESSY II. J. Large-
hexacyanoferrat(II) and -(III) complexes. J. Am. Chem. Soc. 2004, Scale Res. Facil. JLSRF 2017, 2017, 3, 332, DOI: 10.1021/
126, 15618−15623. jacs.6b00332.
(32) Sarangi, R.; Benfatto, M.; Hayakawa, K.; Bubacco, L.; Solomon, (50) Ravel, B.; Newville. ATHENA, ARTEMIS, HEPHAESTUS:
E. I.; Hodgson, K. O.; Hedman, B. MXAN analysis of the XANES data analysis for X-ray absorption spectroscopy using IFEFFIT. J.
energy region of a mononuclear copper complex: Applications to Synchrotron Radiat. 2005, 12, 537−541.
bioinorganic systems. Inorg. Chem. 2005, 44, 9652−9659. (51) Martini, A.; Guda, S. A.; Guda, A. A.; Smolentsev, G.; Algasov,
(33) Tavani, F.; Martini, A.; Capocasa, G.; Di Stefano, S.; A.; Usoltsev, O.; Soldatov, M. A.; Bugaev, A.; Rusalev, Y.; Lamberti,
Lanzalunga, O.; D’Angelo, P. Direct Mechanistic Evidence for a C.; et al. PyFitit: The software for quantitative analysis of XANES
Nonheme Complex Reaction through a Multivariate XAS Analysis. spectra using machine-learning algorithms. Comput. Phys. Commun.
Inorg. Chem. 2020, 59, 9979−9989. 2020, 250, No. 107064.
(34) Timoshenko, J.; Anspoks, A.; Cintins, A.; Kuzmin, A.; Purans, (52) Gurman, S. J.; McGreevy, R. L. Reverse monte-carlo simulation
J.; Frenkel, A. I. Neural Network Approach for Characterizing for the analysis of exafs data. J. Phys.: Condens. Matter 1990, 2, 9463−
Structural Transformations by X-Ray Absorption Fine Structure 9473.
Spectroscopy. Phys. Rev. Lett. 2018, 120, No. 225502. (53) Biesinger, M. C.; Lau, L. W. M.; Gerson, A. R.; Smart, R. S. C.
(35) Timoshenko, J.; Lu, D. Y.; Lin, Y. W.; Frenkel, A. I. Supervised The role of the Auger parameter in XPS studies of nickel metal,
Machine-Learning-Based Determination of Three-Dimensional Struc- halides and oxides. Phys. Chem. Chem. Phys. 2012, 14, 2434−2442.
ture of Metallic Nanoparticles. J. Phys. Chem. Lett. 2017, 8, 5091− (54) Muralidharan, S.; Hayes, R. G. Satellites in the X-ray
5098. photoelectron-spectra of metalloporphyrins. J. Chem. Phys. 1979, 71,
(36) Timoshenko, J.; Wrasman, C. J.; Luneau, M.; Shirman, T.; 2970−2974.
Cargnello, M.; Bare, S. R.; Aizenberg, J.; Friend, C. M.; Frenkel, A. I. (55) Ottaviano, L.; DiNardo, S.; Lozzi, L.; Passacantando, M.;
Probing Atomic Distributions in Mono- and Bimetallic Nanoparticles Picozzi, P.; Santucci, S. Thin and ultra-thin films of nickel
by Supervised Machine Learning. Nano Lett. 2019, 19, 520−529. phthalocyanine grown on highly oriented pyrolitic graphite: An
(37) Jia, Q. Y.; Liu, E. S.; Jiao, L.; Pann, S.; Mukerjee, S. X-Ray XPS, UHV-AFM and air tapping-mode AFM study. Surf. Sci. 1997,
Absorption Spectroscopy Characterizations on PGM-Free Electro- 373, 318−332.
catalysts: Justification, Advantages, and Limitations. Adv. Mater. 2019, (56) Wei, S. T.; Zou, H. Y.; Rong, W. F.; Zhang, F. X.; Ji, Y. F.;
31, No. 1805157. Duan, L. L. Conjugated nickel phthalocyanine polymer selectively
(38) Martini, A.; Borfecchia, E. Spectral Decomposition of X-ray catalyzes CO2-to-CO conversion in a wide operating potential
Absorption Spectroscopy Datasets: Methods and Applications. window. Appl. Catal., B 2021, 284, No. 119739.
Crystals 2020, 10, 664. (57) Baker, M. L.; Mara, M. W.; Yan, J. J.; Hodgson, K. O.; Hedman,
(39) Timoshenko, J.; Anspoks, A.; Kalinko, A.; Kuzmin, A. Local B.; Solomon, E. I. K- and L-edge X-ray absorption spectroscopy
structure of nanosized tungstates revealed by evolutionary algorithm. (XAS) and resonant inelastic X-ray scattering (RIXS) determination
Phys. Status Solidi A 2015, 212, 265−273. of differential orbital covalency (DOC) of transition metal sites.
(40) Timoshenko, J.; Kuzmin, A.; Purans, J. Reverse Monte Carlo Coord. Chem. Rev. 2017, 345, 182−208.
modeling of thermal disorder in crystalline materials from EXAFS (58) Colpas, G. J.; Maroney, M. J.; Bagyinka, C.; Kumar, M.; Willis,
spectra. Comput. Phys. Commun. 2012, 183, 1237−1245. W. S.; Suib, S. L.; Baidya, N.; Mascharak, P. K. X-ray spectroscopic
(41) Timoshenko, J.; Kuzmin, A.; Purans, J. EXAFS study of studies of nickel-complexes, with application to the structure of nickel
hydrogen intercalation into ReO3 using the evolutionary algorithm. J. sites in hydrogenases. Inorg. Chem. 1991, 30, 920−928.
Phys.: Condens. Matter 2014, 26, No. 055401. (59) Mou, K. W.; Chen, Z. P.; Zhang, X. X.; Jiao, M. Y.; Zhang, X.
(42) Li, J. Z.; Zhang, H. G.; Samarakoon, W.; Shan, W. T.; Cullen, P.; Ge, X.; Zhang, W.; Liu, L. C. Highly Efficient Electroreduction of
D. A.; Karakalos, S.; Chen, M. J.; Gu, D. M.; More, K. L.; Wang, G. F.; CO2 on Nickel Single-Atom Catalysts: Atom Trapping and Nitrogen
et al. Thermally Driven Structure and Performance Evolution of Anchoring. Small 2019, 15, No. 1903668.
Atomically Dispersed FeN4 Sites for Oxygen Reduction. Angew. (60) Ren, W. H.; Tan, X.; Yang, W. F.; Jia, C.; Xu, S. M.; Wang, K.
Chem., Int. Ed. 2019, 58, 18971−18980. X.; Smith, S. C.; Zhao, C. Isolated Diatomic Ni-Fe Metal-Nitrogen
(43) Artyushkova, K. Misconceptions in interpretation of nitrogen Sites for Synergistic Electroreduction of CO2. Angew. Chem., Int. Ed.
chemistry from x-ray photoelectron spectra. J. Vac. Sci. Technol., A 2019, 58, 6972−6976.
2020, 38, No. 031002. (61) Li, C.; Ju, W.; Vijay, S.; Timoshenko, J.; Mou, K.; Cullen, D. A.;
(44) Artyushkova, K.; Kiefer, B.; Halevi, B.; Knop-Gericke, A.; Yang, J.; Wang, X.; Pachfule, P.; Brückner, S.; et al. Covalent Organic
Schlogl, R.; Atanassov, P. Density functional theory calculations of Framework (COF) Derived Ni−N-C Catalysts for Electrochemical
J. Am. Chem. Soc. 2023, 145, 17351−17366
CO2 Reduction: Unraveling Fundamental Kinetic and Structural (77) Zhou, M.; Jiang, Y.; Wang, G.; Wu, W. J.; Chen, W. X.; Yu, P.;
Parameters of the Active Sites. Angew. Chem., Int. Ed. 2022, 61, Lin, Y. Q.; Mao, J. J.; Mao, L. Q. Single-atom Ni−N-4 provides a
No. e202114707. robust cellular NO sensor. Nat. Commun. 2020, 11, 3188.
(62) Soldatov, A. V.; Smolentsev, G.; Kravtsova, A.; Yalovega, G.; (78) Tavani, F.; Fracchia, M.; Tofoni, A.; Braglia, L.; Jouve, A.;
Feiters, M. C.; Metselaar, G. A.; Joly, Y. X-ray absorption near-edge Morandi, S.; Manzoli, M.; Torelli, P.; Ghigna, P.; D’Angelo. Structural
spectroscopic study of nickel catalysts. Radiat. Phys. Chem. 2006, 75, and mechanistic insights into low-temperature CO oxidation over a
1866−1868. prototypical high entropy oxide by Cu L-edge operando soft X-ray
(63) Resasco, J.; DeRita, L.; Dai, S.; Chada, J. P.; Xu, M. J.; Yan, X. absorption spectroscopy. Phys. Chem. Chem. Phys. 2021, 23, 26575−
X.; Finzel, J.; Hanukovich, S.; Hoffman, A. S.; Graham, G. W.; Bare, S. 26584.
R.; Pan, X.; Christopher, P. Uniformity Is Key in Defining Structure- (79) Timoshenko, J.; Frenkel, A. I. Frenkel, A. I. Probing structural
Function Relationships for Atomically Dispersed Metal Catalysts: The relaxation in nanosized catalysts by combining EXAFS and reverse
Case of Pt/CeO2. J. Am. Chem. Soc. 2020, 142, 169−184. Monte Carlo methods. Catal. Today 2017, 280, 274−282.
(64) Longo, A.; Giannici, F.; Casaletto, M. P.; Rovezzi, M.; Sahle, C. (80) Joly, Y. FDMNES manual. http://fdmnes.neel.cnrs.fr/ 2021,
J.; Glatzel, P.; Joly, Y.; Martorana, A. Dynamic Role of Gold d- FDMNES manual, Manual.
Orbitals during CO Oxidation under Aerobic Conditions. ACS Catal. (81) Vila, F.; Rehr, J. J.; Kas, J.; Nuzzo, R. G.; Frenkel, A. I. Dynamic
structure in supported Pt nanoclusters: Real-time density functional
2022, 12, 3615−3627.
theory and X-ray spectroscopy simulations. Phys. Rev. B2008, 78(),
(65) Timoshenko, J.; Shivhare, A.; Scott, R. W. J.; Lu, D. Y.; Frenkel,
121404(R), DOI: 10.1103/PhysRevB.78.121404.
A. I. Solving local structure around dopants in metal nanoparticles
(82) Huse, N.; Cho, H.; Hong, K.; Jamula, L.; de Groot, F. M. F.;
with ab initio modeling of X-ray absorption near edge structure. Phys. Kim, T. K.; McCusker, J. K.; Schoenlein, R. W. Femtosecond Soft X-
Chem. Chem. Phys. 2016, 18, 19621−19630. ray Spectroscopy of Solvated Transition-Metal Complexes: Decipher-
(66) Smolentsev, G.; Guilera, G.; Tromp, M.; Pascarelli, S.; ing the Interplay of Electronic and Structural Dynamics. J. Phys. Chem.
Soldatov, A. V. Local structure of reaction intermediates probed by Lett. 2011, 2, 880−884.
time-resolved x-ray absorption near edge structure spectroscopy. J. (83) Khalil, M.; Marcus, M. A.; Smeigh, A. L.; McCusker, J. K.;
Chem. Phys. 2009, 130, 174508. Chong, H. H. W.; Schoenlein, R. W. Picosecond X-ray absorption
(67) Martini, A.; Guda, A. A.; Guda, S. A.; Dulina, A.; Tavani, F.; spectroscopy of a photoinduced iron(II) spin crossover reaction in
D’Angelo, P.; Borfecchia, E.; Soldatov, A. V. Estimating a Set of Pure solution. J. Phys. Chem. A 2006, 110, 38−44.
XANES Spectra from Multicomponent Chemical Mixtures Using a (84) Xiang, S. T.; Huang, P. P.; Li, J. Y.; Liu, Y.; Marcella, N.; Routh,
Transformation Matrix-Based Approach. In Synchrotron Radiation P. K.; Li, G. H.; Frenkel, A. I. Solving the structure of “single-atom”
Science and Applications; Di Cicco, A., Giuli, G., Trapananti, A., Eds.; catalysts using machine learning − assisted XANES analysis. Phys.
Springer International Publishing: Cham, 2021; pp 65−84. Chem. Chem. Phys. 2022, 24, 5116−5124.
(68) Guda, S. A.; Guda, A. A.; Soldatov, M. A.; Lomachenko, K. A.; (85) Cheng, Y.; Zhao, S. Y.; Li, H. B.; He, S.; Veder, J. P.;
Bugaev, A. L.; Lamberti, C.; Gawelda, W.; Bressler, C.; Smolentsev, Johannessen, B.; Xiao, J. P.; Lu, S. F.; Pan, J.; Chisholm, M. F.; Yang,
G.; Soldatov, A. V.; et al. Optimized Finite Difference Method for the S. Z.; Liu, C.; Chen, J. G.; Jiang, S. P. Unsaturated edge-anchored Ni
Full-Potential XANES Simulations: Application to Molecular single atoms on porous microwave exfoliated graphene oxide for
Adsorption Geometries in MOFs and Metal-Ligand Intersystem electrochemical CO2. Appl. Catal., B 2019, 243, 294−303.
Crossing Transients. J. Chem. Theory Comput. 2015, 11, 4512−4521. (86) Maulana, A. L.; Saputro, A. G.; Prasetyo, Y.; Mahyuddin, M. H.;
(69) Joly. X-ray absorption near-edge structure calculations beyond Iqbal, M.; Yudistira, H. T.; Wenten, I. G.; Dipojono, H. K. Two-
the muffin-tin approximation. Phys. Rev. B 2001, 63, No. 125120. Electron Electrochemical Reduction of CO2 on B-Doped Ni−N−C
(70) Tereshchenko, A.; Pashkov, D.; Guda, A.; Guda, S.; Rusalev, Y.; Catalysts: A First-Principles Study. J. Phys. Chem. C 2021, 125,
Soldatov, A. Adsorption Sites on Pd Nanoparticles Unraveled by 19247−19258.
Machine-Learning Potential with Adaptive Sampling. Molecules 2022, (87) Kramm, U. I.; Herranz, J.; Larouche, N.; Arruda, T. M.;
27, 357. Lefevre, M.; Jaouen, F.; Bogdanoff, P.; Fiechter, S.; Abs-Wurmbach,
(71) Guda, A. A.; Guda, S. A.; Martini, A.; Kravtsova, A. N.; Algasov, I.; Mukerjee, S.; et al. Structure of the catalytic sites in Fe/N/C-
A.; Bugaev, A.; Kubrin, S. P.; Guda, L. V.; Š ot, P.; van Bokhoven, J. A.; catalysts for O2-reduction in PEM fuel cells. Phys. Chem. Chem. Phys.
et al. Understanding X-ray absorption spectra by means of descriptors 2012, 14, 11673−11688.
and machine learning algorithms. npj Comput. Mater. 2021, 7, 203. (88) Wang, X. Q.; Chen, Z.; Zhao, X. Y.; Yao, T.; Chen, W. X.; You,
(72) Hossain, M. D.; Huang, Y. F.; Yu, T. H.; Goddard, W. A.; Luo, R.; Zhao, C. M.; Wu, G.; Wang, J.; Huang, W. X.; Yang, J.; Hong, X.;
Z. T. Reaction mechanism and kinetics for CO2 reduction on nickel Wei, S.; Wu, Y.; Li, Y. Regulation of Coordination Number over
single atom catalysts from quantum mechanics. Nat. Commun. 2020, Single Co Sites: Triggering the Efficient Electroreduction of CO2.
Angew. Chem., Int Ed. 2018, 57, 1944−1948.
11, 2256.
(89) Zheng, W. Z.; Yang, J.; Chen, H. Q.; Hou, Y.; Wang, Q.; Gu,
(73) Lin, S. C.; Chang, C. C.; Chiu, S. Y.; Pai, H. T.; Liao, T. Y.;
M.; He, F.; Xia, Y.; Xia, Z.; Li, Z. J.; et al. Atomically Defined
Hsu, C. S.; Chiang, W. H.; Tsai, M. K.; Chen, H. M. Operando time-
Undercoordinated Active Sites for Highly Efficient CO2 Electro-
resolved X-ray absorption spectroscopy reveals the chemical nature reduction. Adv. Funct. Mater. 2020, 30, No. 1907658.
enabling highly selective CO2 reduction. Nat. Commun. 2020, 11, (90) Yan, C. C.; Li, H. B.; Ye, Y. F.; Wu, H. H.; Cai, F.; Si, R.; Xiao,
3525. J. P.; Miao, S.; Xie, S. H.; Yang, F.; Li, Y.; Wang, G.; Bao, X.
(74) Li, Y. G.; Wu, Z. S.; Lu, P. F.; Wang, X.; Liu, W.; Liu, Z. B.; Ma, Coordinatively unsaturated nickel-nitrogen sites towards selective and
J. Y.; Ren, W. C.; Jiang, Z.; Bao, X. H. High-Valence Nickel Single- high-rate CO2 electroreduction. Energy Environ. Sci. 2018, 11, 1204−
Atom Catalysts Coordinated to Oxygen Sites for Extraordinarily 1210.
Activating Oxygen Evolution Reaction. Adv. Sci. 2020, 7, (91) Rong, X.; Wang, H. J.; Lu, X. L.; Si, R.; Lu, T. B. Controlled
No. 1903089. Synthesis of a Vacancy-Defect Single-Atom Catalyst for Boosting CO2
(75) Database, C. O. http://www.crystallography.net/cod/2100746. Electroreduction. Angew. Chem., Int. Ed. 2020, 59, 1961−1965.
html. (accessed January 1, 2022). (92) Zhao, C. M.; Dai, X. Y.; Yao, T.; Chen, W. X.; Wang, X. Q.;
(76) Yang, H. B.; Hung, S. F.; Liu, S.; Yuan, K. D.; Miao, S.; Zhang, Wang, J.; Yang, J.; Wei, S. Q.; Wu, Y. E.; Li, Y. D. Ionic Exchange of
L. P.; Huang, X.; Wang, H. Y.; Cai, W. Z.; Chen, R.; Gao, J.; Yang, X.; Metal Organic Frameworks to Access Single Nickel Sites for Efficient
Chen, W.; Huang, Y.; Chen, H. M.; Li, C. M.; Zhang, T.; Liu, B. Electroreduction of CO2. J. Am. Chem. Soc. 2017, 139, 8078−8081.
Atomically dispersed Ni(i) as the active site for electrochemical CO2 (93) Hu, Y. G.; Zhan, F.; Wang, Q.; Sun, Y. J.; Yu, C.; Zhao, X.;
reduction. Nat. Energy 2018, 3, 140−147. Wang, H.; Long, R.; Zhang, G. Z.; Gao, C.; Zhang, W.; Jiang, J.; Tao,
J. Am. Chem. Soc. 2023, 145, 17351−17366
Y.; Xiong, Y. Tracking Mechanistic Pathway of Photocatalytic CO2

Reaction at Ni Sites Using Operando, Time-Resolved Spectroscopy. J.
Am. Chem. Soc. 2020, 142, 5618−5626.
(94) Wang, Y. O.; Chen, E. Q.; Tang, J. W. Insight on Reaction
Pathways of Photocatalytic CO2 Conversion. ACS Catal. 2022, 12,
7300−7316.
J. Am. Chem. Soc. 2023, 145, 17351−17366

Martini2023-Single Atom

Uploaded by

Copyright:

Available Formats

Martini2023-Single Atom

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Martini2023-Single Atom

Uploaded by

Copyright:

Available Formats

pubs.acs.

Tracking the Evolution of Single-Atom Catalysts for the CO2

ACCESS Metrics & More Article Recommendations *

ABSTRACT: Transition metal-nitrogen-doped carbons (TMNCs) are a

1. INTRODUCTION the as-prepared samples, as showcased by the striking example

have used the third XANES spectrum because we expect all

total number of required XANES calculations is minimized

Figure 8. Comparison of the XANES components for pure species, as

Y.; Xiong, Y. Tracking Mechanistic Pathway of Photocatalytic CO2

You might also like