Pith. sign in

REVIEW 3 major objections 5 minor 35 references

Decoding the Stability of Transition-Metal Alloys with Theory-infused Deep Learning

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A physics-infused graph neural network predicts transition-metal alloy cohesive energies to within 10 meV/atom and attributes stability differences to distinct physical mechanisms.

desk verdict A genuinely novel theory-infused GNN for cohesive energies, but the segregation mechanism rests on an unsupervised residual term the authors themselves call unreliable. read the letter →

arxiv 2506.03031 v2 pith:O3GA5FCU submitted 2025-06-03 cond-mat.mtrl-sci

Yang Huang , Shih-Han Wang , Shuyi Cao , Luke E. K. Achenie , Hongliang Xin This is my paper
classification cond-mat.mtrl-sci
keywords theory-infusedneuralnetworkcohesiveenergytransition-metalalloyssingle-atomgraphd-bandcouplingagglomerationsegregation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper introduces TinNet, a graph neural network that predicts the cohesive energy of transition-metal alloys with a mean absolute error of 10 meV/atom while decomposing that energy into physically interpretable terms taken from renormalized-atom cohesion theory. Instead of treating the neural network as a black box, the model is built so that each atom's energy is a sum of promotion, renormalization, conduction-band formation, and d-band formation contributions, with the network merely predicting the parameters in that expression. When applied to single-atom alloys, the decomposition attributes agglomeration of dopant atoms into dimers to localized d-orbital coupling, and segregation of a dopant into the subsurface to delocalized, volume-dependent electron effects summarized by a compound descriptor. The model is trained on DFT-computed cohesive energies and physical parameters, and it outperforms a purely data-driven graph network on a nested cross-validation test. The authors present this as an interpretable route to screening alloy stability and guiding catalyst design, and they explicitly caution that the renormalization energy alone is unreliable because it absorbs all systematic and numerical errors.

What carries the argument

The central object is Equation (5), the theory-infused energy expression $$E_{\mathrm{coh}} = \frac{1}{N}\sum_i \left[ \$\Delta$ E_{\mathrm{prom}} + \$\Delta$ E_{\mathrm{ren}} + \frac{3\hbar}{10m}\left(\frac{3\$pi^{2}$ N_s}{\$\beta$ V_{WS}}\right)^{2/3} - \$\alpha$ \frac{W_d}{20} N_d (10-N_d)\right],$$ where $\Delta E_{\mathrm{prom}}$ is the atomic promotion energy, $\Delta E_{\mathrm{ren}}$ is the renormalization energy, the third term is the free-electron conduction-band formation energy, and the last term is the d-band formation energy in the rectangular tight-binding approximation. The graph neural network predicts per-site values of $N_s$, $N_d$, $W_d$, $\alpha$, and $\beta$ rather than the energy itself, so the decomposition is intrinsic to the architecture. The d-band width $W_d$ is the carrier of locality: in tight binding it is proportional to the sum of orbital coupling strengths between a site and its neighbors, which is why the dimer-monomer difference in $W_d$ directly reflects localized d-d bonding. For segregation, the paper combines the two delocalized terms into the descriptor $\Delta N_d\cdot\Delta R$, with $\Delta N_d$ the guest-to-host d-electron count ratio and $\Delta R$ the ratio of relative local space radii between top- and sub-layer configurations.

What would settle it

For a specific single-atom alloy such as Co in Ag(111), compute the top-versus-sub-layer segregation energy with an independent first-principles decomposition that separates renormalization and conduction-band contributions; if the combined delocalized energy does not track the descriptor $\Delta N_d\cdot\Delta R$ across different host metals, the mechanism claiming segregation is dictated by delocalized effects is falsified.

Watch

Extended reading notes

Core claim

TinNet's central claim is that the cohesive energy of a transition-metal alloy can be predicted to within about 10 meV/atom while simultaneously yielding a faithful physical decomposition into localized and delocalized electronic effects. The paper reports a test mean absolute error of $10.0 \pm 2.2$ meV/atom against DFT, beating the data-driven baseline, and shows that the learned parameters ($N_s$, $N_d$, $W_d$, $\alpha$, $\beta$) match DFT-computed values closely. For single-atom alloys, the model finds that the dimer-versus-monomer energy difference (agglomeration) is governed by the d-band formation term, which becomes strongly negative when a guest like Re or Os has a d-electron count near five and a d-orbital radius much larger than the host's, creating strong guest-guest coupling. For top-versus-sub-layer placement (segregation), the d-band term is nearly unchanged and the deciding contribution is the sum of renormalization and conduction-band energies, captured by the descriptor $\Delta N_d \cdot \Delta R$ based on the guest-to-host d-electron ratio and the ratio of relative local space radii. The authors explicitly state that the renormalization energy alone is unreliable, so they treat it together with the conduction-band term.

Load-bearing premise

The mechanistic conclusions assume that the energy decomposition learned by the network, especially the renormalization term that is not directly supervised and that the paper says absorbs all systematic and numerical errors, faithfully reflects real electronic interactions rather than just absorbing the model's errors.

Editorial extensions

If this is right

  • TinNet's test error of about 10 meV/atom means cohesive energies across a wide alloy compositional space can be screened without additional DFT calculations.
  • The agglomeration analysis yields a design rule: a guest with d-electron count near five and a d-orbital radius clearly larger than the host's (such as Re or Os in Ag) will tend to form dimers.
  • The segregation analysis yields a complementary rule: a guest with fewer d-electrons than the host and a larger relative local space radius will favor the surface over the subsurface.
  • The facet analysis shows that dense-packed surfaces like Pt(111) owe their stability to both higher coordination and implicitly captured reduced relaxation, which the energy decomposition makes visible.
  • Because only thermodynamic preferences are treated, the predictions indicate tendencies; actual surface configurations also depend on kinetic barriers and adsorbate effects.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same localized-versus-delocalized decomposition could be exported to adsorption energies and diffusion barriers, where d-band coupling already plays a central role, giving a unified interpretable model for alloy surface chemistry.
  • If the renormalization energy were calibrated against a rigorous first-principles energy decomposition, the segregation mechanism could be tested beyond the Cu, Ag, and Au hosts studied here.
  • The d-orbital-radius rule for agglomeration suggests a direct experimental test: scanning tunneling microscopy of dilute Re or Os in Ag should observe dimer formation at low coverage, while guests with smaller d-radii remain dispersed.
  • The paper's own admission that renormalization energy is unreliable implies that the localized-delocalized split is only as trustworthy as the DFT errors feeding it; assessing how those errors are distributed among the learned parameters would sharpen or bound the mechanistic conclusions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The manuscript introduces TinNet, a graph neural network that embeds the renormalized-atom cohesion theory into its architecture. For each atomic site the network predicts physical parameters (Ns, Nd, Wd, alpha, beta, and Eren) and computes the cohesive energy through Eq. (5). The model is trained on DFT data for transition-metal alloys, with direct supervision of Ns, Nd, alpha, beta, and Wd, while Eren is supervised only indirectly through Ecoh. TinNet achieves a test MAE of 10 meV/atom, outperforming a CGCNN baseline, and is applied to single-atom alloys (SAAs) to compute agglomeration and segregation energies. The authors conclude that agglomeration is governed by localized d-orbital coupling, while segregation is dictated by delocalized effects such as wavefunction renormalization.

Significance. The potential value is real: the theory-infused architecture is transparent, the supervised physical parameters (Ns, Nd, alpha, beta, Wd) are meaningful quantities, and the demonstrated extrapolation to SAA systems is useful. The agglomeration analysis, supported by both Shapley-style decomposition and site-wise energy component inspection, is a credible demonstration of the approach. However, the segregation mechanism claim rests on an unsupervised energy term that the authors themselves label unreliable. The paper's own caveats weaken the central mechanistic claim as written, so the significance depends on whether the decomposition can be validated or appropriately reframed.

major comments (3)
  1. [Fig. 5b discussion (after Eq. 5)] The paper states that 'the renormalization energy absorbs all systematic and numerical errors, making both its sign and magnitude unreliable,' but then uses Delta_Eren, either alone or combined with Delta_Econd, to conclude that segregation is dictated by delocalized effects such as wavefunction renormalization. Because Delta_Eren is not supervised by DFT and is effectively determined as the residual in Eq. (5), errors in Ns, Nd, Wd, alpha, and beta can be absorbed into it. Combining Delta_Eren with Delta_Econd does not remove this degeneracy; it only changes which combination of residuals is examined. The headline segregation mechanism is therefore not independently established. Please provide a direct validation of Delta_Eren (e.g., from a DFT-based decomposition) or demonstrate that the combined delocalized contribution is robust across an ensemble of trained models.
  2. [Fig. 3c] The facet analysis attributes the relative stability of Pt(111) primarily to Delta_Eren, while simultaneously acknowledging that Eren's origin cannot be solely wavefunction compression and likely includes surface-relaxation penalties. Given the stated unreliability of Delta_Eren, this coordination-dependent interpretation is not supported by the model as presented. Either validate Eren for this setting or remove the Eren-based facet interpretation.
  3. [Fig. 5c] The descriptor Delta_Nd * Delta_R is introduced post hoc, and its agreement with segregation-energy trends is only qualitative. No correlation coefficient or uncertainty estimate is reported, and the descriptor is not shown to be uniquely implied by TinNet's decomposition. To support the segregation claim, show a quantitative relationship between the model's predicted delocalized terms and this descriptor, or present a robustness analysis demonstrating that the conclusion does not depend on the residual definition of Eren.
minor comments (5)
  1. [Notation, Eq. (2) and Eq. (5)] The text uses both Delta_Eren and Eren for the same quantity; unify the notation.
  2. [Fig. 5c caption] The descriptor Delta_Nd * Delta_R defines Delta_Nd as a ratio (N_guest_d / N_host_d), not a difference; rename it to avoid confusion.
  3. [Data availability] The GitHub repository link contains a placeholder string of 'X' characters; the actual repository must be provided before publication.
  4. [Typography] There is a typo 'the the segregation energy' in the Fig. 5 caption text; please correct it.
  5. [Fig. 3 caption] The caption 'cohesive energy of 3d, 4d and 5d metals' should say 'transition metals' for clarity.

Circularity Check

1 steps flagged · score 6.0 of 10

Segregation mechanism relies on ΔEren, an unsupervised residual that the paper admits has unreliable sign and magnitude; ΔEren is determined by construction from Ecoh minus the supervised terms.

  1. fitted input called prediction [Eq. 5; model training description; Fig. 5b discussion]
    "Although the model explicitly predicts ΔEren for each atom, we do not have access to its ground truth with DFT; therefore, the model was supervised using Ecoh as the ground truth. ... In our calculations, the renormalization energy absorbs all systematic and numerical errors, making both its sign and magnitude unreliable."

    Equation 5 defines Ecoh as a sum that includes ΔEren plus terms built from Ns, Nd, Wd, α, and β. The training loss supervises Ecoh and the five parameters, but contains no target for ΔEren. Consequently ΔEren is the free component that absorbs the mismatch between the supervised terms and the DFT cohesive energy; its predicted sign and magnitude are a fitting residual rather than a measured or independently constrained physical quantity. The paper itself concedes that ΔEren 'absorbs all systematic and numerical errors, making both its sign and magnitude unreliable.' The abstract then states that segregation is 'dictated by delocalized effects such as wavefunction renormalization,' and Fig. 5b isolates this residual in the Shapley attribution.

full rationale

The cohesive-energy prediction itself is not circular: TinNet is trained against DFT cohesive energies on held-out test sets (MAE 10 meV/atom), and the supervised parameters Ns, Nd, α, β, and Wd have DFT ground-truth labels. The d-band formation energy, which drives the agglomeration conclusion, is therefore an independently supervised physical term rather than a pure fitting residual. The paper's self-citations are not load-bearing: reference 30 supports a standard tight-binding proportionality that is also attributed to reference 31, and references 13 and 34 are contextual rather than premises of the derivation. The substantial circular step is the segregation mechanism. Equation 5 contains ΔEren, but the model is not trained against any ΔEren target; instead it is 'supervised using Ecoh.' Given the other predicted parameters and Ecoh, ΔEren is the residual that makes the formula match the data, and the authors explicitly state that it absorbs all systematic and numerical errors and that its sign and magnitude are unreliable. The Shapley attribution in Fig. 5b then labels this residual as 'wavefunction renormalization' in the abstract's headline claim. The paper's combination of ΔEren with ΔEcond and the geometric ΔNd·ΔR descriptor provides partial independent evidence for a delocalized effect, and the paper is commendably transparent about the unreliability of ΔEren; nevertheless, the specific 'wavefunction renormalization' component in the segregation claim is not independently established and is partly a fitted residual by construction. This warrants a partial circularity score of 6 rather than a higher score, because the total-energy predictions and the agglomeration mechanism retain independent empirical content.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central mechanistic conclusions rest on six learned physical parameters, including two explicit correction factors (alpha, beta) and a renormalization energy that is not directly supervised. The decomposition into localized and delocalized contributions is therefore a property of a fitted model as much as a measurement of the underlying physics.

free parameters (6)
  • alpha (d-band correction factor) = learned per site, not reported
    Multiplies the d-band formation term in Eq. 5 and is trained to match DFT-derived ground truth values; it absorbs errors in the tight-binding approximation, so Shapley contributions partly reflect this fit.
  • beta (conduction-band volume correction) = learned per site, not reported
    Divides the Wigner-Seitz volume in the free-electron term (Eq. 5) and is trained; it adjusts the conduction term to reproduce DFT cohesive energies.
  • Eren (renormalization energy) = learned per atom, not directly supervised
    Predicted by the network but no DFT ground truth; it is only constrained through the total cohesive energy and is acknowledged to absorb systematic and numerical errors.
  • Ns (conduction electron count) = learned per site, supervised; values not reported
    Predicted by the network and supervised with DFT-derived values; it enters the free-electron conduction term.
  • Nd (d-electron count) = learned per site, supervised; values not reported
    Predicted by the network and supervised with DFT-derived values; it determines the d-band occupancy in the cohesion formula.
  • Wd (d-band width) = learned per site, supervised; values not reported
    Predicted by the network and supervised with DFT-derived values; it sets the magnitude of the d-band formation energy in Eq. 4.
assumptions (6)
  • domain assumption Renormalized-atom cohesion theory decomposes cohesive energy into promotion, renormalization, conduction band formation, and d-band formation terms.
    Used in Eq. 1-4 and throughout; this is the physical framework the model embeds.
  • domain assumption The d-band formation energy can be approximated by the rectangular-band tight-binding expression -Wd/20 times Nd times (10-Nd).
    Used in Eq. 4; relies on tight-binding theory and a rectangular density of states.
  • domain assumption The conduction band formation energy follows the free-electron model with the given scaling in Eq. 3.
    Used in Eq. 3; assumes free-electron-like s-electrons and Wigner-Seitz volume.
  • domain assumption DFT calculations provide reliable ground truth for Ns, Nd, Wd, alpha, and beta.
    The model is supervised with these values; if the definitions of these parameters from projected DOS are inconsistent, the decomposition is unphysical.
  • domain assumption The graph neural network can learn site-specific physical parameters that transfer to unseen single-atom alloy configurations.
    The extrapolation to SAAs depends on the learned representation generalizing from the training alloy set.
  • domain assumption Experimental promotion energies from the NIST database are appropriate for use in Eq. 1 in the condensed phase.
    The promotion energy term uses atomic spectroscopic values and is independent of the environment, as assumed by the cohesion theory.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Decoding the Stability of Transition-Metal Alloys with Theory-infused Deep Learning." pith.science (2026). https://pith.science/paper/O3GA5FCU

@misc{pith2026250603031,
  author       = {Pith},
  title        = {Pith review of: Decoding the Stability of Transition-Metal Alloys with Theory-infused Deep Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O3GA5FCU}},
  note         = {Machine review of arXiv:2506.03031}
}
read the original abstract

We introduce an interpretable deep learning framework that predicts the cohesive energy of transition-metal alloys (TMAs) by embedding cohesion theory within graph neural networks (GNNs). Beyond accurate prediction of cohesive energy, a key indicator of thermodynamic stability, the model offers mechanistic insights by disentangling energy contributions into physically meaningful components. These data-driven interpretations reveal periodic trends and stability principles governing transition metals. We apply the model to single-atom alloys (SAAs) to assess their thermodynamic resilience against two destabilizing processes: agglomeration (adatom clustering) and segregation (migration into the subsurface). Our analysis shows that these phenomena are governed by distinct physical factors-agglomeration is primarily influenced by localized d-orbital coupling, while segregation is dictated by delocalized effects such as wavefunction renormalization. This model thus serves as an explainable AI tool for understanding and guiding the design of stable TMAs, with implications for catalysis and materials discovery.

Figures

Figures reproduced from arXiv: 2506.03031 by the authors.

Figure 1
Figure 1. Cohesion processes in transition metals as described by [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) The TinNet architecture for accurate and interpretable [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. TinNet-predicted: (a) total cohesive energy of [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: (a) Dopant diffusion between terrace and step sites on [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: (a) Model structures representing SAAs with a guest metal [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

35 extracted references · 34 canonical work pages

  1. [1]

    J. W. Schwank, (1985)

    work page 1985

  2. [2]

    Kitchin, J

    J. Kitchin, J. K. Nørskov, M. Barteau, and J. Chen, The Journal of chemical physics120, 10240 (2004)

    work page 2004

  3. [3]

    Zhang, M

    J. Zhang, M. B. Vukmirovic, Y . Xu, M. Mavrikakis, and R. R. Adzic, Angewandte Chemie117, 2170 (2005)

    work page 2005

  4. [4]

    Antolini, J

    E. Antolini, J. R. Salgado, and E. R. Gonzalez, Journal of Power Sources160, 957 (2006)

    work page 2006

  5. [5]

    C. W. Bezerra, L. Zhang, H. Liu, K. Lee, A. L. Marques, E. P. Marques, H. Wang, and J. Zhang, Journal of Power Sources 173, 891 (2007)

    work page 2007

  6. [6]

    J. C. Meier, C. Galeano, I. Katsounaros, J. Witte, H. J. Bon- gard, A. A. Topalov, C. Baldizzone, S. Mezzavilla, F. Schüth, and K. J. Mayrhofer, Beilstein journal of nanotechnology5, 44 (2014)

    work page 2014

  7. [7]

    Zhang, Y

    J. Zhang, Y . Yuan, L. Gao, G. Zeng, M. Li, and H. Huang, Advanced Materials33, 2006494 (2021)

    work page 2021

  8. [8]

    Q. Gao, X. Han, Y . Liu, and H. Zhu, ACS catalysis14, 6045 (2024)

    work page 2024

Show all 35 references
  1. [9]

    Zhang, A

    T. Zhang, A. G. Walsh, J. Yu, and P. Zhang, Chemical Society Reviews50, 569 (2021)

    2021

  2. [10]

    Giannakakis, M

    G. Giannakakis, M. Flytzani-Stephanopoulos, and E. C. H. Sykes, Accounts of chemical research52, 237 (2018)

    2018

  3. [11]

    J. Liu, F. R. Lucci, M. Yang, S. Lee, M. D. Marcinkowski, A. J. Therrien, C. T. Williams, E. C. H. Sykes, and M. Flytzani- Stephanopoulos, Journal of the American Chemical Society 138, 6396 (2016)

    2016

  4. [12]

    Zhang, G

    X. Zhang, G. Cui, H. Feng, L. Chen, H. Wang, B. Wang, X. Zhang, L. Zheng, S. Hong, and M. Wei, Nature Commu- nications10, 5812 (2019)

    2019

  5. [13]

    Huang, S.-H

    Y . Huang, S.-H. Wang, L. E. Achenie, K. Choudhary, and H. Xin, The Journal of Chemical Physics161(2024)

    2024

  6. [14]

    F. R. Lucci, M. T. Darby, M. F. Mattera, C. J. Ivimey, A. J. Therrien, A. Michaelides, M. Stamatakis, and E. C. H. Sykes, The journal of physical chemistry letters7, 480 (2016)

    2016

  7. [15]

    Réocreux, P

    R. Réocreux, P. L. Kress, R. T. Hannagan, V . Çınar, M. Sta- matakis, and E. C. H. Sykes, The Journal of Physical Chemistry Letters11, 8751 (2020)

    2020

  8. [16]

    Cheng, Y

    X. Cheng, Y . Wang, Y . Lu, L. Zheng, S. Sun, H. Li, G. Chen, and J. Zhang, Applied Catalysis B: Environmental306, 121112 (2022)

    2022

  9. [17]

    Q. Gao, B. Yao, H. S. Pillai, W. Zang, X. Han, Y . Liu, S.-W. Yu, Z. Yan, B. Min, S. Zhang,et al., Nature Synthesis2, 624 (2023)

    2023

  10. [18]

    M. T. Darby, E. C. H. Sykes, A. Michaelides, and M. Sta- matakis, Topics in catalysis61, 428 (2018)

    2018

  11. [19]

    R. T. Hannagan, G. Giannakakis, M. Flytzani-Stephanopoulos, and E. C. H. Sykes, Chemical Reviews120, 12044 (2020)

    2020

  12. [20]

    A. J. McCue and J. A. Anderson, Journal of Catalysis329, 538 (2015)

    2015

  13. [21]

    Ouyang, K

    M. Ouyang, K. G. Papanikolaou, A. Boubnov, A. S. Hoff- man, G. Giannakakis, S. R. Bare, M. Stamatakis, M. Flytzani- Stephanopoulos, and E. C. H. Sykes, Nature Communications 12, 1549 (2021)

    2021

  14. [22]

    Xie and J

    T. Xie and J. C. Grossman, Physical review letters120, 145301 (2018)

    2018

  15. [23]

    G. L. Hart, T. Mueller, C. Toher, and S. Curtarolo, Nature Re- views Materials6, 730 (2021)

    2021

  16. [24]

    Ubaru, A

    S. Ubaru, A. Mi˛ edlar, Y . Saad, and J. R. Chelikowsky, Physical Review B95, 214102 (2017)

    2017

  17. [25]

    Hodges, R

    L. Hodges, R. Watson, and H. Ehrenreich, Physical Review B 5, 3953 (1972)

    1972

  18. [26]

    Gelatt Jr, H

    C. Gelatt Jr, H. Ehrenreich, and R. Watson, Physical Review B 15, 1613 (1977)

    1977

  19. [27]

    Ralchenko, Memorie della Societa Astronomica Italiana Supplementi8, 96 (2005)

    Y . Ralchenko, Memorie della Societa Astronomica Italiana Supplementi8, 96 (2005)

    2005

  20. [28]

    Hammer and J

    B. Hammer and J. K. Nørskov, inChemisorption and Reac- 7 tivity on Supported Clusters and Thin Films: Towards an Un- derstanding of Microscopic Processes in Catalysis(Springer,

  21. [29]

    Brooks and B

    M. Brooks and B. Johansson, Journal of Physics F: Metal Physics13, L197 (1983)

    1983

  22. [30]

    Huang, S.-H

    Y . Huang, S.-H. Wang, M. Kamanuru, L. E. Achenie, J. R. Kitchin, and H. Xin, Physical Review B110, L121404 (2024)

    2024

  23. [31]

    J. R. Kitchin, J. K. Nørskov, M. A. Barteau, and J. Chen, Phys- ical review letters93, 156801 (2004)

    2004

  24. [32]

    Clementi and D.-L

    E. Clementi and D.-L. Raimondi, The Journal of Chemical Physics38, 2686 (1963)

    1963

  25. [33]

    S. L. Bueno, A. Leonardi, N. Kar, K. Chatterjee, X. Zhan, C. Chen, Z. Wang, M. Engel, V . Fung, and S. E. Skrabalak, ACS nano16, 18873 (2022)

    2022

  26. [34]

    Huang, S.-H

    Y . Huang, S.-H. Wang, X. Wang, N. Omidvar, L. E. Achenie, S. E. Skrabalak, and H. Xin, The Journal of Physical Chemistry C128, 11183 (2024)

    2024

  27. [35]

    P. L. Kress, S. Zhang, Y . Wang, V . Çınar, C. M. Friend, E. C. H. Sykes, and M. M. Montemore, Journal of the American Chem- ical Society145, 8401 (2023). All source data along with Jupyter notebooks for data pre- processing, model development, andpost hocanalysis are availabl...

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.