Pith. sign in

REVIEW 4 major objections 4 minor 79 references

Machine-learned dynamics of surface polarons at reduced oxide surfaces

T0 review · 4 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper argues that small polarons—electrons trapped by lattice distortion—hop several orders of magnitude more slowly on the reduced rutile TiO2(110) surface than in bulk rutile, and that this surface bottleneck, not trapping defects al

desk verdict Strong qualitative result on surface polaron suppression, but the room-temperature mobility is an extrapolation from zero observed hops and should be treated as provisional. read the letter →

arxiv 2606.13833 v2 pith:CVWMZFAJ submitted 2026-06-11 cond-mat.mtrl-sci physics.comp-ph

Luca Leoni , Cesare Franchini This is my paper · ORCID
classification cond-mat.mtrl-sciphysics.comp-ph
keywords smallpolaronTiO2(110)surfacemobilitymachinelearninginteratomicpotentialoxygenvacancyhoppingtransportporousO2
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 tries to establish that small polarons—electrons localized at titanium sites and coupled to lattice distortion—move far more slowly on the oxygen-deficient rutile TiO2(110) surface than in bulk rutile, by roughly four orders of magnitude, and that this explains why porous TiO2 conducts electrons much more poorly than single crystals. It reaches this conclusion by using a machine-learned interatomic potential, trained on first-principles molecular dynamics, to simulate several nanoseconds of polaron hopping at 400–700 K, then extrapolating to room temperature. The underlying cause is geometric: surface hopping is confined mostly to flat rows in the subsurface layer, while jumps between layers are rare. A reader should care because the result gives a microscopic, transferable picture of how surface structure, not just defects, controls charge transport in redox-active oxides used in catalysis, photovoltaics, and energy storage.

What carries the argument

The carrying mechanism is a polaron-aware machine-learned interatomic potential: a graph neural network that augments each atom's chemical identity with a one-hot charge-state flag identifying which Ti site currently hosts the polaron, and that predicts both energies/forces and site-resolved spin magnetizations. At each molecular-dynamics step the polaron is reassigned to the Ti atom with the largest predicted magnetization, so lattice motion and charge position evolve together. This enables nanosecond-scale trajectories with explicit hopping events, from which diffusion coefficients and mobilities are obtained via mean-square displacement fits and the Einstein relation.

What would settle it

Run the machine-learned polaron dynamics at 300 K for tens of nanoseconds and count hops: if the observed hopping rate is statistically incompatible with the Arrhenius-extrapolated rate (equivalently, mobility at 300 K differs from about 2×10^-5 cm²/Vs by more than an order of magnitude), the central claim fails. Alternatively, a terahertz or time-resolved conductivity measurement on porous rutile films with controlled particle size that yields a mobility much higher than 10^-4 cm²/Vs and independent of surface fraction would contradict the surface-suppression picture.

Watch

Extended reading notes

Core claim

The central claim is that small-polaron mobility on the reduced rutile TiO2(110) surface is suppressed by several orders of magnitude relative to bulk rutile across the investigated temperature range, with an Arrhenius extrapolation giving approximately 1.9×10^-5 cm² V^-1 s^-1 at 300 K, matching the experimental range for porous TiO2 (7×10^-6 to 5×10^-4). The suppression is attributed to loss of favorable hopping pathways: about 87% of hops occur along [001] Ti rows in the subsurface S-1 layer, about 12% along surface rows, and only about 1% are interlayer hops; inter-row and deeper-layer hops were not observed. The oxygen vacancy acts as an attractive center that biases polaron positions to

Load-bearing premise

The room-temperature mobility and the 'orders of magnitude slower' claim rest on the assumption that polaron hopping follows the same Arrhenius law measured at 400–700 K all the way down to 300 K, where no hop was observed in about 40 ns of simulation; a regime change or increased trapping below 400 K would invalidate the extrapolation.

Editorial extensions

If this is right

  • Porous TiO2's low electron mobility is explained by surface-restricted hopping topology, not solely by trapping defects, so bulk single-crystal mobility values should not be used directly in models of nanostructured or porous electrodes.
  • The machine-learning strategy transfers to other reducible oxide surfaces and defect/adsorbate environments, opening a route to predictive simulations of excess-charge dynamics in catalytic and energy-conversion materials.
  • Because the oxygen vacancy reshapes the polaron free-energy landscape, vacancy concentration and spatial distribution directly control surface transport; engineering vacancy placement could tune conductivity.
  • The computed activation barrier (~386 meV) provides a parameter for device-level transport models of TiO2-based photoelectrodes and sensors.
  • The strong anisotropy of hopping—dominant subsurface row hops, rare interlayer hops—implies that surface transport is effectively quasi-one-dimensional along [001] rows in the second layer.

Reading between the lines

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

  • If the Arrhenius extrapolation holds, then at device operating temperatures surface polaron transport must rely on a sparse network of subsurface-row hops, so experimental probes that distinguish subsurface from surface charge carriers could directly test this picture.
  • The results imply a design rule: nanostructuring that increases exposure of (110)-like facets will suppress carrier mobility, so porous electrodes might be optimized by selecting facets that preserve bulk-like hopping rows or by introducing dopants that deepen favorable subsurface pathways.
  • Because the model simulates a single polaron with one vacancy, the strong polaron–vacancy attraction suggests that at realistic carrier densities multi-polaron interactions and vacancy clustering will modify mobilities; extending the approach to two or more excess charges is a natural next test.
  • The paper's claim predicts that measured electron mobility in porous TiO2 should depend systematically on particle size and surface-to-volume ratio; a controlled experimental series varying those parameters could quantitatively validate the surface-confined transport scenario.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The manuscript extends the LEOPOLD machine-learned polaron-dynamics framework from bulk rutile TiO2 to the reduced rutile TiO2(110) surface. The authors construct a polaron-aware NequIP-style potential trained on ~100 ps of DFT+U FPMD with active learning, then perform multiple nanosecond MLMD trajectories at 300–700 K. From mean-square-displacement fits they extract surface polaron mobilities and report a suppression of several orders of magnitude relative to bulk, a dominant subsurface [001] hopping channel, rare interlayer hops, and a temperature-dependent polaron–vacancy attraction. An Arrhenius extrapolation gives a room-temperature mobility of 1.9×10^-5 cm2 V^-1 s^-1, which the authors state is in very good agreement with porous-TiO2 experiments.

Significance. If the quantitative claims hold, this work is significant: it demonstrates a transferable machine-learning strategy for simulating polaron dynamics at oxide surfaces on nanosecond time scales and provides a microscopic rationalization for the much lower electron mobilities observed in porous versus single-crystal rutile TiO2. The qualitative pathway analysis — preferred subsurface [001] hopping and rare interlayer events — is consistent with earlier short FPMD studies and is a useful advance in its own right. The model is not circular in the objectionable sense: it is trained on DFT data, and the porous-TiO2 comparison is an external experimental benchmark. However, the headline room-temperature number and the 'orders of magnitude' claim at 300 K rest on an Arrhenius extrapolation from temperatures at which hopping is actually observed, with no uncertainty quantification. This limits the current support for the central quantitative conclusion and needs to be addressed before the paper can be accepted.

major comments (4)
  1. [Sec. III.A and Fig. 4] The room-temperature value μ = 1.9×10^-5 cm2 V^-1 s^-1 and the claimed 'very good agreement' with porous-TiO2 experiments rest entirely on extrapolating the Arrhenius fit of mobilities computed at 400, 500, 600 and 700 K down to 300 K, because the authors state that 'no hopping event was observed over a total simulation time of ~40 ns' at 300 K. This extrapolation is load-bearing. The paper gives no uncertainty for E_a (386 meV) or μ, no fit residuals, and no check of Arrhenius curvature. I request confidence intervals and a discussion of whether sub-Arrhenius behavior, quantum tunneling, or trap-limited transport could set in below 400 K. If longer or accelerated sampling at 300 K is infeasible, the conclusions should distinguish the directly simulated 400–700 K suppression from the extrapolated room-temperature value, and the 'very good agreement' wording should be softened accordingly
  2. [Sec. III.B] Figure 5(b) shows that at 400 K the polaron–vacancy distance distribution initialized near V_O is strongly peaked at short separations, indicating trap-dominated local exploration rather than homogeneous diffusion. If the 400 K MSD includes this trapping/drift component, the diffusion coefficient entering the Arrhenius fit is not the free-hopping mobility, and the extracted 386 meV barrier mixes trapping and hopping contributions. The claim that the suppression arises from the 'loss of favourable hopping pathways' rather than from the vacancy itself requires a control calculation (e.g., a surface without the vacancy, or trajectories initialized far from V_O with demonstrated linear MSD over the fitted window) or a quantitative estimate of the trapping contribution. As written, the microscopic interpretation is not fully disentangled from the defect potential.
  3. [Sec. III.A, Fig. 4 and Methods] The quantitative mobility estimates lack statistical uncertainties. Each temperature is the average of four independent runs, but Fig. 4 shows no error bars, and the text does not report the number of hopping events per run, the duration of the fitted diffusive regime, or the variance of D across runs. Given the rare-event character of the dynamics (interlayer hops are ~1% of events, and only six such hops occur in a 3 ns trajectory at 700 K), Poisson counting errors are non-negligible. Reporting event counts and error bars is necessary to support the stated orders-of-magnitude comparison and the fitted activation energy.
  4. [Sec. II.A and Sec. III.A] The slab model keeps the two deepest layers fixed, while small-polaron formation requires local lattice distortion. The statement that 'transport to deeper layers was never observed' is therefore partly a consequence of the model constraint: a polaron cannot stabilize in a fixed layer because the surrounding atoms cannot relax. This could bias the surface/bulk comparison and the conclusion that surface confinement controls transport. I ask the authors to test with a thicker relaxed slab, or at least to discuss explicitly how the fixed bottom layers affect the confinement claim and the inferred mobility suppression.
minor comments (4)
  1. [Sec. III.A] Typo: 'hooping trajectories' should be 'hopping trajectories'.
  2. [Conclusions / Sec. IV] The activation barrier of 386 meV is described as within 'approximately 15%' of infrared estimates of 300–330 meV; the actual discrepancy is 17–29%. Please recalculate or reword.
  3. [Supplemental Material, Sec. I] Grammar: 'straightforward its use' should be 'streamline its use'.
  4. [Data Availability] The statement that data and code 'will be made available upon publication' does not allow reviewers to reproduce the results. Consider depositing the LEOPOLD version, configuration files, and representative trajectories in a public repository at the revision stage.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: the surface mobility is an emergent MLMD result, and the room-temperature comparison is an explicitly labeled Arrhenius extrapolation, not a fitted target.

full rationale

The paper's central quantitative result—surface polaron mobility suppressed by several orders of magnitude relative to bulk—is obtained by simulating LEOPOLD dynamics and fitting the long-time MSD with the Einstein relation (Sec. III.A). The LEOPOLD model is trained on DFT-FPMD labels (energies, forces, magnetizations), not on the experimental porous-TiO2 mobilities, and the mobility emerges from accumulated hopping statistics. Thus the main 'prediction' is not an input by construction. The room-temperature value µ = 1.9×10^-5 cm2/V s is explicitly stated to be an extrapolation because 'no hopping event was observed over a total simulation time of ~40 ns' at 300 K; the Arrhenius form is assumed, and the experimental range is compared only after the extrapolation, not used to generate it. That is a statistical/extrapolation risk, not a circular reduction. The bulk comparison relies on the authors' prior LEOPOLD paper (Ref. [49]), which is a self-citation; however, the cited bulk LEOPOLD was independently benchmarked against experiment ('yields a room-temperature mobility of 1.6 cm2/Vs, in agreement with recent spectroscopic measurements'), so this is external support rather than a load-bearing self-citation. Surface hopping-pathway assignments are also corroborated by earlier FPMD work. The remaining concerns—four-temperature Arrhenius fit, absence of direct 300 K hops, and temperature-dependent polaron-vacancy trapping—bear on correctness and uncertainty, not on equivalence-by-construction. Overall: minor non-load-bearing self-citation, but no circular derivation chain.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

No new physical entities are postulated; the polaron charge-state vector is a computational construct, not a new particle or force. The main extras the reader is asked to accept are the Hubbard U, the single-polaron slab model, the ML surrogate accuracy, and the Arrhenius extrapolation.

free parameters (2)
  • DFT+U effective Hubbard U (U_eff) = 3.9 eV
    Chosen by hand from previous TiO2 studies; controls polaron localization and hopping barriers in all training data (Methods II.A).
  • Arrhenius activation energy used for extrapolation = 386 meV
    Obtained from Arrhenius fit to LEOPOLD mobilities at 400–700 K; used to extrapolate µ(300 K)=1.9e-5 cm2/Vs; no direct 300 K hops observed (Sec. III.A).
assumptions (5)
  • domain assumption PBE+U with U_eff=3.9 eV captures small-polaron energetics and hopping barriers in rutile TiO2.
    Used throughout DFT training data; polaron localization and barrier heights depend critically on U, which is taken from previous studies and not benchmarked here (Methods II.A).
  • domain assumption The LEOPOLD surrogate trained on ~100 ps of FPMD remains accurate over nanosecond trajectories and at temperatures sampled by MLMD.
    Active learning augments data with hopping configurations, but there is no direct long-timescale DFT validation of hopping rates; polaronic-site force RMSE is 110–139 meV/Å (Supplemental Table I).
  • domain assumption The polaron can be represented as a single localized site determined by argmax magnetization, with the second excess electron from the vacancy canceled by a homogeneous background.
    Eq. (1) and Methods II.A; this enforces the single-polaron manifold and excludes multi-polaron or delocalized states.
  • standard math Einstein relation and Arrhenius thermal-activation law connect MSD, diffusion constant, mobility, and temperature.
    Sec. III.A uses MSD fits and Arrhenius extrapolation; standard transport relations, but the Arrhenius assumption is load-bearing for the 300 K value.
  • domain assumption A five-layer asymmetric slab with fixed bottom layers and Γ-point sampling adequately represents the TiO2(110) surface.
    Methods II.A; finite-size and thickness effects are not quantified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Machine-learned dynamics of surface polarons at reduced oxide surfaces." pith.science (2026). https://pith.science/paper/CVWMZFAJ

@misc{pith2026260613833,
  author       = {Pith},
  title        = {Pith review of: Machine-learned dynamics of surface polarons at reduced oxide surfaces},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CVWMZFAJ}},
  note         = {Machine review of arXiv:2606.13833}
}
abstract

Reducible oxides exhibit a rich interplay of electronic, structural, and chemical properties that underpins applications in catalysis, photovoltaics, batteries, and energy storage. This interplay is strongly shaped by excess electrons, often introduced by oxygen vacancies, that localize as small polarons and influence charge transport and surface chemistry. At surfaces, these polarons play a central role in charge localization, mobility, and reactivity, yet their finite-temperature dynamics remain difficult to access from first principles due to the long time scales needed to adequately sample polaron's hopping. To overcome this limitation, we extend machine-learning-assisted polaron dynamics to redox-active oxide surfaces, using oxygen-deficient rutile TiO$_2$(110) as a paradigmatic case. By accessing several nanoseconds of dynamics over a range of temperatures, we show that small-polaron mobility at the reduced rutile TiO$_2$(110) surface is suppressed by several orders of magnitude relative to the corresponding bulk material, providing a microscopic interpretation of the lower electron mobilities observed in porous rutile TiO$_2$ compared with single-crystal samples. This suppressed mobility arises from the loss of favourable hopping pathways: surface polaron motion is largely confined to planar inter-row trajectories within the second topmost layers, with only rare interlayer hopping events. These results establish a transferable machine-learning strategy for investigating polaron dynamics in reducible oxides.

Figures

Figures reproduced from arXiv: 2606.13833 by the authors.

Figure 1
Figure 1. FIG. 1. Representation of the TiO [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Graphical representation of the LEOPOLD loop. The [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. a) Leopold simulated small polaron dynamics on TiO [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Polaron mobility on the oxygen-deficient rutile [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. (a) Structural model of the reduced rutile-TiO [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

79 extracted references · 3 canonical work pages

  1. [1]

    This large difference in probability can be rationalized from the bottom panel of Fig

    jumps. This large difference in probability can be rationalized from the bottom panel of Fig. 3(a): the [111] jump in- volves not only a longer distance between the initial and final Ti sites, but also a reorientation of the polaron or- bital. Both factors reduce the overlap between the initial and final states, thereby strongly suppressing the hop- ping ...

  2. [2]

    These trajec- tories are shown individually at the bottom of Fig

    ), in agreement with previous FPMD studies [9, 45, 46, 60]. These trajec- tories are shown individually at the bottom of Fig. 3(a) over shorter femtosecond time windows. The most fre- quent hopping event by far, accounting for approximately 87% of the total events, involves small-polaron transport along the [001] rows in the subsurface S-1 layer adjacent ...

  3. [3]

    Inter-row hopping between adjacent Ti rows, which are∼3 ˚A apart and connected by oxygen atoms is equally rare

    Small-polaron transport to deeper layers was never ob- served, confirming the clear tendency of small polarons to form and remain confined within the topmost surface layers [9, 61–63]. Inter-row hopping between adjacent Ti rows, which are∼3 ˚A apart and connected by oxygen atoms is equally rare. Each individual hopping event is tracked by monitor- ing the...

  4. [4]

    Bredow and G

    T. Bredow and G. Pacchioni, Electronic structure of an isolated oxygen vacancy at the tio2(110) surface, Chem- ical Physics Letters355, 417 (2002)

    2002

  5. [5]

    -type jump over a 3 ns trajectory indicates that such events are much rarer than the dominanth S

  6. [6]

    Di Valentin, E

    C. Di Valentin, E. Finazzi, G. Pacchioni, A. Selloni, S. Livraghi, A. M. Czoska, M. C. Paganini, and E. Gi- amello, Density functional theory and electron paramag- netic resonance study on the effect of N-F codoping of TiO2, Chemistry of Materials20, 3706 (2008)

    2008

  7. [7]

    Rousseau, V.-A

    R. Rousseau, V.-A. Glezakou, and A. Selloni, Theoretical insights into the surface physics and chemistry of redox- active oxides, Nature Reviews Materials5, 460 (2020)

    2020

  8. [8]

    M. V. Ganduglia-Pirovano, A. Hofmann, and J. Sauer, Oxygen vacancies in transition metal and rare earth ox- ides: Current state of understanding and remaining chal- lenges, Surface Science Reports62, 219 (2007)

    2007

Show all 79 references
  1. [9]

    Franchini, M

    C. Franchini, M. Reticcioli, M. Setvin, and U. Diebold, Polarons in materials, Nature Reviews Materials6, 560–586 (2021)

    2021

  2. [10]

    W.-J. Yin, B. Wen, C. Zhou, A. Selloni, and L.-M. Liu, Excess electrons in reduced rutile and anatase TiO2, Sur- face Science Reports73, 58 (2018)

    2018

  3. [11]

    Di Valentin, G

    C. Di Valentin, G. Pacchioni, A. Selloni, S. Livraghi, and E. Giamello, Characterization of paramagnetic species in N-doped TiO2 powders by epr spectroscopy and dft calculations, The Journal of Physical Chemistry B109, 11414 (2005)

    2005

  4. [12]

    F. Esch, S. Fabris, L. Zhou, T. Montini, C. Africh, P. For- nasiero, G. Comelli, and R. Rosei, Electron localization determines defect formation on ceria substrates, Science 309, 752 (2005)

    2005

  5. [13]

    Di Valentin and A

    C. Di Valentin and A. Selloni, Bulk and surface polarons in photoexcited anatase tio2, The Journal of Physical Chemistry Letters2, 2223 (2011)

    2011

  6. [14]

    Chiesa, M

    M. Chiesa, M. C. Paganini, S. Livraghi, and E. Giamello, Charge trapping in TiO 2 polymorphs as seen by elec- tron paramagnetic resonance spectroscopy, Phys. Chem. Chem. Phys.15, 9435 (2013)

    2013

  7. [15]

    Setvin, C

    M. Setvin, C. Franchini, X. Hao, M. Schmid, A. Jan- otti, M. Kaltak, C. G. Van de Walle, G. Kresse, and U. Diebold, Direct view at excess electrons in tio2 rutile and anatase, Phys. Rev. Lett.113, 086402 (2014)

    2014

  8. [16]

    L. M. Carneiro, S. K. Cushing, C. Liu, Y. Su, P. Yang, A. Alivisatos, and S. R. Leone, Excitation-wavelength- dependent small polaron trapping of photoexcited carri- ers inα-fe2o3, Nature Materials16, 819 (2017)

    2017

  9. [17]

    M. V. Ganduglia-Pirovano, J. L. F. Da Silva, and J. Sauer, Density-functional calculations of the structure of near-surface oxygen vacancies and electron localization on ceo2(111), Phys. Rev. Lett.102, 026101 (2009)

    2009

  10. [18]

    C. S. Ahart, J. Blumberger, and K. M. Rosso, Polaronic structure of excess electrons and holes for a series of bulk iron oxides, Phys. Chem. Chem. Phys.22, 10699 (2020)

    2020

  11. [19]

    Zhang, Z.-K

    D. Zhang, Z.-K. Han, G. E. Murgida, M. V. Ganduglia- Pirovano, and Y. Gao, Oxygen-vacancy dynamics and entanglement with polaron hopping at the reduced ceo2(111) surface, Phys. Rev. Lett.122, 096101 (2019)

    2019

  12. [20]

    Pelatti, E

    S. Pelatti, E. Spurio, D. Catone, P. O’Keeffe, S. Tur- chini, G. Ammirati, F. Paleari, D. Varsano, S. Benedetti, A. di Bona, S. D’Addato, Y. Jiang, P. Zalden, Y. Ue- mura, H. Wang, D. Vinci, X. Huang, F. Lima, M. Bied- nov, D. Khakhulin, C. J. Milne, F. Boscherini, and P. Luch...

    2025

  13. [21]

    Iordanova, M

    N. Iordanova, M. Dupuis, and K. M. Rosso, Charge trans- port in metal oxides: A theoretical study of hematite α-fe2o3, The Journal of Chemical Physics122, 144305 (2005)

    2005

  14. [22]

    Wiktor and A

    J. Wiktor and A. Pasquarello, Electron and hole polarons at the bivo4–water interface, ACS Applied Materials & Interfaces11, 18423 (2019)

    2019

  15. [23]

    Pastor, J.-S

    E. Pastor, J.-S. Park, L. Steier, S. Kim, M. Gr¨ atzel, J. R. Durrant, A. Walsh, and A. A. Bakulin, In situ obser- vation of picosecond polaron self-localisation inα-fe2o3 photoelectrochemical cells, Nature Communications10, 3962 (2019)

    2019

  16. [24]

    H. Liu, H. Cong, G. Yang, C. Gong, J. Ding, Y. y. Fu, J. Cui, K. Song, B. Chen, C. He, N. Zhao, J. Ye, and F. He, Surface hole polaron site tuning governs charge carrier separation in bivo4 photoanodes, Nature Com- munications17, 2562 (2026). 9

    2026

  17. [25]

    Bandaranayake, E

    S. Bandaranayake, E. Hruska, S. Londo, S. Biswas, and L. R. Baker, Small polarons and surface defects in metal oxide photocatalysts studied using xuv reflection– absorption spectroscopy, The Journal of Physical Chem- istry C124, 22853 (2020)

    2020

  18. [26]

    Redondo, M

    J. Redondo, M. Reticcioli, V. Gabriel, D. Wrana, F. Ellinger, M. Riva, G. Franceschi, E. Rheinfrank, I. Sokolovi´ c, Z. Jakub, F. Kraushofer, A. Alexan- der, E. Belas, L. L. Patera, J. Repp, M. Schmid, U. Diebold, G. S. Parkinson, C. Franchini, P. Kocan, and M. Setvin, Real-sp...

    2024

  19. [27]

    Wiktor, F

    J. Wiktor, F. Ambrosio, and A. Pasquarello, Role of po- larons in water splitting: The case of bivo4, ACS Energy Letters3, 1693 (2018)

    2018

  20. [28]

    De´ ak, B

    P. De´ ak, B. Aradi, and T. Frauenheim, Oxygen deficiency in tio2: Similarities and differences between the ti self- interstitial and the o vacancy in bulk rutile and anatase, Phys. Rev. B92, 045204 (2015)

    2015

  21. [29]

    Zhang, C

    Y. Zhang, C. Cheng, Z. Zhou, R. Long, and W.-H. Fang, Surface hydroxylation during water splitting pro- motes the photoactivity of bivo4(010) surface by sup- pressing polaron-mediated charge recombination, The Journal of Physical Chemistry Letters14, 9096 (2023), pMID: 37791802

    2023

  22. [30]

    A. R. Elmaslmane, M. B. Watkins, and K. P. McKenna, First-principles modeling of polaron forma- tion in tio2 polymorphs, Journal of Chemical Theory and Computation14, 3740 (2018), pMID: 29874462, https://doi.org/10.1021/acs.jctc.8b00199

    2018 doi

  23. [31]

    Diebold, The surface science of titanium dioxide, Sur- face Science Reports48, 53 (2003)

    U. Diebold, The surface science of titanium dioxide, Sur- face Science Reports48, 53 (2003)

    2003

  24. [32]

    N. A. Deskins and M. Dupuis, Electron transport via polaron hopping in bulk tio2: A density functional theory characterization, Phys. Rev. B75, 195212 (2007)

    2007

  25. [33]

    De´ ak, J

    P. De´ ak, J. Kullgren, and T. Frauenheim, Polarons and oxygen vacancies at the surface of anatase tio2, phys- ica status solidi (RRL) – Rapid Research Letters8, 583 (2014)

    2014

  26. [34]

    Dai and F

    Z. Dai and F. Giustino, Identification of large polarons and exciton polarons in rutile and anatase polymorphs of titanium dioxide, Proceedings of the National Academy of Sciences121, e2414203121 (2024)

    2024

  27. [35]

    P. G. Moses, A. Janotti, C. Franchini, G. Kresse, and C. G. Van de Walle, Donor defects and small polarons on the tio2(110) surface, Journal of Applied Physics119, 181503 (2016)

    2016

  28. [36]

    H. Tang, K. Prasad, R. Sanjin` es, P. E. Schmid, and F. L´ evy, Electrical and optical properties of TiO2 anatase thin films, Journal of Applied Physics75, 2042 (1994)

  29. [37]

    I. Ali, M. Suhail, Z. A. Alothman, and A. Alwarthan, Recent advances in syntheses, properties and applica- tions of tio¡sub¿2¡/sub¿nanostructures, RSC Advances8, 30125–30147 (2018)

    2018

  30. [38]

    Fujishima, X

    A. Fujishima, X. Zhang, and D. A. Tryk, Tio2 photo- catalysis and related surface phenomena, Surface Science Reports63, 515 (2008)

    2008

  31. [39]

    E. J. Marcelis, J. E. ten Elshof, and M. Morales-Masis, Titanium dioxide: A versatile earth-abundant optical material for photovoltaics, Advanced Optical Materials 12, 2401423 (2024)

    2024

  32. [40]

    Dittrich, Porous tio2: Electron transport and applica- tion to dye sensitized injection solar cells, physica status solidi (a)182, 447–455 (2000)

    T. Dittrich, Porous tio2: Electron transport and applica- tion to dye sensitized injection solar cells, physica status solidi (a)182, 447–455 (2000)

    2000

  33. [41]

    S. X. Zhang, D. C. Kundaliya, W. Yu, S. Dhar, S. Y. Young, L. G. Salamanca-Riba, S. B. Ogale, R. D. Vis- pute, and T. Venkatesan, Niobium doped TiO2: Intrinsic transparent metallic anatase versus highly resistive rutile phase, Journal of Applied Physics102, 013701 (2007)

    2007

  34. [42]

    Spreafico and J

    C. Spreafico and J. VandeVondele, The nature of excess electrons in anatase and rutile from hybrid dft and rpa, Phys. Chem. Chem. Phys.16, 26144 (2014)

    2014

  35. [43]

    E. Yagi, R. R. Hasiguti, and M. Aono, Electronic conduc- tion above 4 k of slightly reduced oxygen-deficient rutile tio2−x , Phys. Rev. B54, 7945 (1996)

    1996

  36. [44]

    Hendry, F

    E. Hendry, F. Wang, J. Shan, T. F. Heinz, and M. Bonn, Electron transport in TiO 2 probed by thz time-domain spectroscopy, Physical Review B69, 10.1103/phys- revb.69.081101 (2004)

    2004 doi

  37. [45]

    Hendry, M

    E. Hendry, M. Koeberg, B. O’Regan, and M. Bonn, Lo- cal field effects on electron transport in nanostructured TiO2 revealed by terahertz spectroscopy, Nano Letters6, 755–759 (2006)

    2006

  38. [46]

    Reticcioli, M

    M. Reticcioli, M. Setvin, M. Schmid, U. Diebold, and C. Franchini, Formation and dynamics of small polarons on the rutile tio2(110) surface, Phys. Rev. B98, 045306 (2018)

    2018

  39. [47]

    Janotti, C

    A. Janotti, C. Franchini, J. B. Varley, G. Kresse, and C. G. Van de Walle, Dual behavior of excess electrons in rutile tio2, physica status solidi (RRL) – Rapid Research Letters7, 199 (2013)

    2013

  40. [48]

    A. P. Bart´ ok, M. C. Payne, R. Kondor, and G. Cs´ anyi, Gaussian approximation potentials: The accuracy of quantum mechanics, without the electrons, Phys. Rev. Lett.104, 136403 (2010)

    2010

  41. [49]

    Morita, M

    K. Morita, M. J. Golomb, M. Rivera, and A. Walsh, Models of Polaron Transport in Inorganic and Hybrid Organic–Inorganic Titanium Oxides, Chemistry of Mate- rials 10.1021/acs.chemmater.3c00322 (2023), publisher: American Chemical Society

    2023 doi

  42. [50]

    SHI and X

    Z. SHI and X. GONG, A density functional theory study of polarons on different tio2 surfaces, Journal of Fuel Chemistry and Technology52, 1877 (2024)

    2024

  43. [51]

    P. M. Kowalski, M. F. Camellone, N. N. Nair, B. Meyer, and D. Marx, Charge localization dynamics induced by oxygen vacancies on the tio2 surface, Phys. Rev. Lett. 105, 146405 (2010)

    2010

  44. [52]

    Kresse and D

    G. Kresse and D. Joubert, From ultrasoft pseudopoten- tials to the projector augmented-wave method, Physical Review B59, 1758 (1999)

    1999

  45. [53]

    Behler and M

    J. Behler and M. Parrinello, Generalized neural-network representation of high-dimensional potential-energy sur- faces, Phys. Rev. Lett.98, 146401 (2007)

    2007

  46. [54]

    Dudarev and G

    S. Dudarev and G. Botton, Electron-energy-loss spectra and the structural stability of nickel oxide: An LSDA+U study, Physical Review B57, 1505 (1998)

    1998

  47. [55]

    V. C. Birschitzky, L. Leoni, M. Reticcioli, and C. Fran- chini, Machine learning small polaron dynamics, Phys. Rev. Lett.134, 216301 (2025)

    2025

  48. [56]

    I. G. Austin and N. F. Mott, Polarons in crystalline and non-crystalline materials, Advances in Physics50, 757 (2001)

    2001

  49. [57]

    Kresse and J

    G. Kresse and J. Furthm¨ uller, Efficiency of ab-initio total energy calculations for metals and semiconductors using a plane-wave basis set, Computational Materials Science 6, 15–50 (1996)

    1996

  50. [58]

    S. S. Schoenholz and E. D. Cubuk, Jax m.d. a framework for differentiable physics, inAdvances in Neural Infor- mation Processing Systems, Vol. 33 (Curran Associates, Inc., 2020). 10

    2020

  51. [59]

    J. P. Perdew, K. Burke, and M. Ernzerhof, Generalized gradient approximation made simple, Phys. Rev. Lett. 77, 3865 (1996)

    1996

  52. [60]

    Reticcioli, M

    M. Reticcioli, M. Setvin, X. Hao, P. Flauger, G. Kresse, M. Schmid, U. Diebold, and C. Franchini, Polaron-driven surface reconstructions, Phys. Rev. X7, 031053 (2017)

    2017

  53. [61]

    Nos´ e, A unified formulation of the constant tempera- ture molecular dynamics methods, The Journal of Chem- ical Physics81, 511–519 (1984)

    S. Nos´ e, A unified formulation of the constant tempera- ture molecular dynamics methods, The Journal of Chem- ical Physics81, 511–519 (1984)

    1984

  54. [63]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. John- son, Y. Katariya, C. Leary, D. Maclaurin, G. Nec- ula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, and Q. Zhang, JAX: composable transformations of Python+NumPy programs (2018)

    2018

  55. [64]

    Solbrand, H

    A. Solbrand, H. Lindstr¨ om, H. Rensmo, A. Hagfeldt, S.- E. Lindquist, and S. S¨ odergren, Electron transport in the nanostructured tio2-electrolyte system studied with time-resolved photocurrents, The Journal of Physical Chemistry B101, 2514–2518 (1997)

    1997

  56. [65]

    See Supplemental Material at [URL will be inserted by publisher] for details on the training specifics, which in- cludes Ref. [69,70]

  57. [66]

    Kopidakis, E

    N. Kopidakis, E. A. Schiff, N.-G. Park, J. van de Lage- maat, and A. J. Frank, Ambipolar diffusion of photocarri- ers in electrolyte-filled, nanoporous tio¡sub¿2¡/sub¿, The Journal of Physical Chemistry B104, 3930–3936 (2000)

    2000

  58. [67]

    A. C. Papageorgiou, N. S. Beglitis, C. L. Pang, G. Teobaldi, G. Cabailh, Q. Chen, A. J. Fisher, W. A. Hofer, and G. Thornton, Electron traps and their effect on the surface chemistry of tio¡sub¿2¡/sub¿(110), Proceedings of the Na- tional Academy of Sciences107, 2391 (2010), ht...

    2010 doi

  59. [68]

    Reticcioli, I

    M. Reticcioli, I. Sokolovi´ c, M. Schmid, U. Diebold, M. Setvin, and C. Franchini, Interplay between adsor- bates and polarons: Co on rutile tio2(110), Phys. Rev. Lett.122, 016805 (2019)

    2019

  60. [69]

    Reticcioli, U

    M. Reticcioli, U. Diebold, and C. Franchini, Modeling po- larons in density functional theory: lessons learned from tio2, Journal of Physics: Condensed Matter34, 204006 (2022)

    2022

  61. [70]

    Z. Zeng, X. Liang, Z. Fan, Y. Chen, M. Simoncelli, and B. Cheng, Thermal transport of amorphous hafnia across the glass transition, ACS Materials Letters7, 2695–2701 (2025). Supplemental material to ”Machine-learned dynamics of surface polarons at reduced oxide surfaces” Luca ...

    2025

  62. [71]

    H. Idriss, Infrared spectroscopic measurements of ex- cited polaronic states of tio2 (110) rutile and tiow (101) anatase single crystals, Journal of the American Chemi- cal Society147, 41067–41078 (2025)

    2025

  63. [73]

    V. C. Birschitzky, F. Ellinger, U. Diebold, M. Reticcioli, and C. Franchini, Machine learning for exploring small polaron configurational space, npj Computational Mate- rials8, 125 (2022)

    2022

  64. [74]

    V. C. Birschitzky, I. Sokolovi´ c, M. Prezzi, K. Palot´ as, M. Setv ´ ın, U. Diebold, M. Reticcioli, and C. Franchini, Machine learning-based prediction of polaron-vacancy patterns on the tio2(110) surface, npj Computational Materials10, 89 (2024)

    2024

  65. [75]

    Wadowski, A

    A. Wadowski, A. D. S. Parmar, F. Ka´ skosz, J. Byggm¨ astar, J. S. Wr´ obel, M. J. Alava, and S. Bon- fanti, Efficient training of machine learning potentials for metallic glasses: Cuzral validation (2025)

    2025

  66. [77]

    V. C. Birschitzky, L. Leoni, M. Reticcioli, and C. Franchini, Machine learning small polaron dynamics, Phys. Rev. Lett. 134, 216301 (2025)

    2025

  67. [78]

    Batzner, A

    S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials, Nature Communications13, 10.1038/s41467-022-29939-5 (2022)

    2022 doi

  68. [79]

    Wadowski, A

    A. Wadowski, A. D. S. Parmar, F. Ka´ skosz, J. Byggm¨ astar, J. S. Wr´ obel, M. J. Alava, and S. Bonfanti, Efficient training of machine learning potentials for metallic glasses: Cuzral validation (2025)

    2025

  69. [80]

    Z. Zeng, X. Liang, Z. Fan, Y. Chen, M. Simoncelli, and B. Cheng, Thermal transport of amorphous hafnia across the glass transition, ACS Materials Letters7, 2695–2701 (2025)

    2025

  70. [110]

    The occurrence of only sixh S,S−1

    direction. The occurrence of only sixh S,S−1

  71. [111]

    polaron encoding

    ). II. METHODS A. First-principles calculations All first-principles calculations were performed with the GPU implementation of V ASP 6.5.1 [51, 52], in- cluding spin polarization and using the Perdew–Burke– Ernzerhof exchange-correlation functional [53]. Elec- tronic correlat...

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.