pith. sign in

arxiv: 2604.02499 · v1 · submitted 2026-04-02 · ❄️ cond-mat.mtrl-sci

CARBON-2D Topological Descriptor (C2DTD): An Interpretable and Physics-Informed Representation for Two-Dimensional Carbon Networks

Felipe Hawthorne , Marcelo Lopes Pereira Junior , Fabiano Manoel de Andrade , Cristiano Francisco Woellner , Raphael Matozo Tromer This is my paper

Pith reviewed 2026-05-13 20:32 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci
keywords 2D carbon networkstopological descriptormachine learninggraphene defectsring topologystructural representationDFT energy predictionsmall-data regimes
0
0 comments X

The pith

C2DTD combines local geometry, radial signatures and primitive ring counts into a compact vector that predicts energies of 2D carbon networks more accurately than generic features in small-data settings.

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

The paper introduces the CARBON-2D Topological Descriptor as a fixed-length, invariant representation built specifically for two-dimensional carbon structures ranging from pristine graphene to vacancy-rich and amorphous sheets. It assembles three components: local geometric statistics of bonds, a compact radial structural signature, and explicit counts of the smallest rings present in the network. When fed to machine-learning models, this vector delivers lower errors on formation energies and other properties than high-dimensional generic featurizers, even when training data are scarce. Unsupervised embeddings of the descriptor align more closely with the DFT energy landscape, and ablation tests show that the ring-topology part dominates the signal once vacancies trigger reconstruction. The same vector also tracks the continuous shift from hexagon-dominated order to topologically disordered networks as vacancy fractions rise from five to fifteen percent.

Core claim

C2DTD is a fixed-length invariant vector formed by concatenating local bond-length and angle statistics, a binned radial distribution function, and a histogram of primitive ring sizes. Regression models trained on this vector achieve robust accuracy on formation energies of 2D carbon allotropes and defect-engineered graphene sheets, outperforming standard high-dimensional descriptors while remaining directly interpretable. Feature-importance analysis isolates ring topology as the leading energetic driver, particularly after vacancy-induced reconstruction, and the descriptor space maps smoothly onto the DFT energy manifold in unsupervised projections.

What carries the argument

The C2DTD vector, formed by concatenating local geometric statistics, a compact radial structural signature, and explicit primitive ring topology counts into one fixed-length invariant representation.

If this is right

  • Machine-learning models using C2DTD achieve lower prediction errors on formation energies than generic high-dimensional featurizers when training data are limited.
  • Ring-topology features emerge as the dominant contribution to energy variations once vacancies induce local reconstruction.
  • The descriptor registers the continuous loss of hexagonal order as random vacancy concentrations increase from 5 to 15 percent.
  • Unsupervised projections of C2DTD vectors lie closer to the DFT-computed energy landscape than projections obtained from standard structural representations.
  • Ablation and importance rankings derived from C2DTD models identify specific ring motifs that correlate strongly with structural stability.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same three-component construction could be applied to other two-dimensional materials such as hexagonal boron nitride to test whether ring topology remains the leading energetic driver.
  • Because the vector is short and fixed-length, it can be inserted directly into existing high-throughput screening pipelines without increasing computational cost.
  • Feature rankings from C2DTD models could be used to design targeted defect patterns that stabilize desired electronic or mechanical properties.
  • Adding angular or dihedral information to the radial signature might further tighten the correlation with total energy without sacrificing compactness.

Load-bearing premise

That local bond geometry, a radial distribution signature, and counts of the smallest rings together capture the full multi-scale information that determines the energy of any 2D carbon network.

What would settle it

A new test set of amorphous or heavily reconstructed 2D carbon monolayers on which models using only the C2DTD vector produce energy errors that remain large compared with DFT reference values, even after the descriptor has been shown to register the expected topological disorder.

Figures

Figures reproduced from arXiv: 2604.02499 by Cristiano Francisco Woellner, Fabiano Manoel de Andrade, Felipe Hawthorne, Marcelo Lopes Pereira Junior, Raphael Matozo Tromer.

Figure 1
Figure 1. Figure 1: Overview of the CARBON-2D Topological Descrip￾tor framework, illustrating (a) the structural diversity of the investigated systems, (b) the descriptor extraction pipeline, (c) the physical mapping of the feature space, and (d) the average ring-size distribution across the dataset. lographic representation and constructs a periodic neighbor graph using a physically motivated cutoff radius consistent with co… view at source ↗
Figure 2
Figure 2. Figure 2: Parity plots of predicted versus DFT total energies for 2D carbon structures evaluated at a test size of 0.3, comprising (a) the predictions derived from the C2DTD framework and (b) the predictions obtained utilizing matminer structural features. The dashed lines indicate perfect agreement between predicted and reference energies [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Interpretability and ablation analysis of the topological descriptor evaluated at a test fraction of 0.3, depicting (a) the top ten features ranked by relative algorithmic gain, (b) the Spearman rank correlation matrix between normalized ring fractions and total energy, and (c-f) the parity plots and predictive metrics for the full descriptor, the isolated ring statistics, the isolated radial distribution … view at source ↗
Figure 4
Figure 4. Figure 4: Unsupervised dimensionality reduction of the descriptor spaces colored by DFT total energy, showing (a) the principal component analysis projection for the C2DTD framework, (b) the corresponding principal component analysis projection for the matminer structural features, and (c) the non-linear t-distributed stochastic neighbor embedding for the C2DTD framework. lar total energy occupying contiguous region… view at source ↗
Figure 5
Figure 5. Figure 5: Interpretability analysis of the topological descriptor evaluated on a defect￾engineered graphene dataset comprising 150 structurally relaxed configurations, illustrating (a) the heatmap of primitive ring fractions across the systematically perturbed dataset, (b-c) the ring distribution for a representative low-defect structure with a 5% vacancy concentration, (d-e) the corresponding distribution for an in… view at source ↗
Figure 6
Figure 6. Figure 6: Prediction of vacancy-induced formation energy for the defect-engineered gra￾phene dataset utilizing the topological descriptor and gradient-boosted trees, presenting (a) the predictive performance and feature-importance ranking for a test fraction of 0.3 and (b) the corresponding performance and feature attribution under a severe data-scarcity regime with a test fraction of 0.8. ring fraction reflects the… view at source ↗
read the original abstract

Two-dimensional (2D) carbon networks, from pristine graphene to defect-rich and amorphous monolayers, exhibit a complex structure-energy landscape governed not only by local bonding but also by medium-range order and network topology. Capturing these multi-scale effects in a compact, interpretable, and data-efficient manner remains a major challenge for machine learning (ML) in low-dimensional materials. In this work, we introduce the CARBON-2D Topological Descriptor (C2DTD), a physically informed structural representation specifically designed for 2D carbon systems. The descriptor integrates local geometric statistics, a compact radial structural signature, and explicit primitive ring topology into a fixed-length, invariant vector that is both computationally efficient and directly interpretable. Benchmarking on diverse datasets of 2D carbon allotropes and defect-engineered graphene sheets demonstrates that C2DTD achieves robust predictive performance in small-data regimes, outperforming generic high-dimensional featurization schemes while preserving physical transparency. Unsupervised manifold analysis reveals a smoother alignment between descriptor space and the DFT energy landscape, and feature-importance and ablation studies confirm that ring topology emerges as a dominant energetic driver, particularly under vacancy-induced reconstruction. Furthermore, controlled simulations with 5-15% random vacancies show that C2DTD naturally captures the progressive transition from hexagon-dominated graphene to topologically disordered networks, enabling both dataset-level and structure-specific interpretation. Owing to its compactness, interpretability, and strong physics-based inductive bias, C2DTD provides a fast and generalizable framework for data-driven modeling, defect analysis, and high-throughput screening of 2D carbon materials.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces the CARBON-2D Topological Descriptor (C2DTD) for two-dimensional carbon networks. It integrates local geometric statistics, a compact radial structural signature, and explicit primitive ring topology into a fixed-length invariant vector. The work claims that this descriptor provides robust predictive performance in small-data regimes for 2D carbon allotropes and defect-engineered graphene sheets, outperforming generic high-dimensional featurization schemes, while maintaining physical transparency and interpretability. Additional claims include smoother manifold alignment with DFT energies and the identification of ring topology as a dominant energetic driver under vacancy reconstruction.

Significance. If validated, this descriptor could significantly advance machine learning applications in materials science by offering a compact, interpretable, and physics-informed representation tailored to 2D carbon systems. Its potential for data-efficient modeling, defect analysis, and high-throughput screening in low-data regimes represents a useful contribution, particularly given the emphasis on physical transparency and the ability to capture multi-scale structural effects.

major comments (3)
  1. [Abstract] Strong benchmarking results are asserted, including 'robust predictive performance' and outperformance over generic schemes, yet no numerical metrics, error bars, dataset sizes, or specific validation details are supplied. This makes it impossible to assess the central performance claims from the provided text.
  2. [Descriptor Construction] The primitive ring topology is incorporated into a fixed-length vector, but the manuscript does not specify the ring size cutoff, binning scheme, or handling of larger rings. Given the focus on 5-15% vacancy networks where larger rings become relevant, this omission risks the topology component being incomplete, potentially undermining the claims of capturing the multi-scale energy landscape.
  3. [Results and Discussion] The feature importance and ablation studies are reported to confirm ring topology as the dominant driver, but without quantitative values or detailed controls, the strength of this conclusion cannot be evaluated.
minor comments (2)
  1. [Abstract] The abstract would benefit from a brief mention of the specific datasets used or the target properties being predicted to provide context for the benchmarking claims.
  2. Notation for the descriptor components could be standardized and defined more explicitly in the main text for improved clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us improve the clarity and completeness of the manuscript. We address each major comment point by point below and have made revisions where the feedback identifies gaps in the presented information.

read point-by-point responses

  1. Referee: [Abstract] Strong benchmarking results are asserted, including 'robust predictive performance' and outperformance over generic schemes, yet no numerical metrics, error bars, dataset sizes, or specific validation details are supplied. This makes it impossible to assess the central performance claims from the provided text.

    Authors: We agree that the abstract would benefit from including representative quantitative metrics to allow readers to immediately evaluate the performance claims. The full manuscript reports these details (dataset sizes, cross-validation protocols, error bars, and comparisons) in the Results section. In the revised version we have updated the abstract to incorporate key numerical highlights drawn directly from our benchmarking studies while preserving its concise nature. revision: yes

  2. Referee: [Descriptor Construction] The primitive ring topology is incorporated into a fixed-length vector, but the manuscript does not specify the ring size cutoff, binning scheme, or handling of larger rings. Given the focus on 5-15% vacancy networks where larger rings become relevant, this omission risks the topology component being incomplete, potentially undermining the claims of capturing the multi-scale energy landscape.

    Authors: We acknowledge that the specific implementation parameters for the ring-topology component were not stated explicitly enough. In the revised manuscript we have added a dedicated paragraph in the Descriptor Construction section that specifies the ring-size cutoff, the binning scheme employed, and the explicit handling of rings beyond the cutoff to ensure the descriptor remains complete for the vacancy concentrations studied. revision: yes

  3. Referee: [Results and Discussion] The feature importance and ablation studies are reported to confirm ring topology as the dominant driver, but without quantitative values or detailed controls, the strength of this conclusion cannot be evaluated.

    Authors: We agree that the strength of the conclusion would be clearer with explicit quantitative results. In the revised manuscript we have expanded the Results and Discussion section to include the numerical feature-importance scores, the exact performance deltas from the ablation experiments, and the control settings used, allowing direct evaluation of the claim that ring topology is the dominant energetic driver. revision: yes

Circularity Check

0 steps flagged

C2DTD derivation is self-contained with no circular reductions

full rationale

The descriptor is explicitly constructed as a fixed-length vector from three independent structural components—local geometric statistics, radial signature, and primitive ring topology—each defined directly from the atomic coordinates and connectivity of the 2D carbon network. No equations or steps in the provided text show these components being fitted to the target DFT energies or other outputs; instead, the vector is presented as an input representation whose performance is then benchmarked. No self-citations are invoked to justify uniqueness or ansatz choices, and no prediction step reduces by construction to a fitted parameter or renamed input. The derivation chain therefore remains independent of the downstream ML predictions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that local bonding, medium-range order, and network topology together govern the energy landscape of 2D carbon; the descriptor is newly postulated without independent evidence outside this work.

axioms (1)
  • domain assumption The structure-energy landscape of 2D carbon networks is governed by local bonding, medium-range order, and network topology.
    Explicitly stated in the first sentence of the abstract as the governing factors.
invented entities (1)
  • C2DTD descriptor no independent evidence
    purpose: Compact, invariant, interpretable vector representation of 2D carbon networks
    Newly defined in this work by combining three structural components; no external falsifiable evidence supplied in abstract.

pith-pipeline@v0.9.0 · 5619 in / 1321 out tokens · 46022 ms · 2026-05-13T20:32:39.246181+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    Moore, Lucas Lindsay, Zachary H

    Jae Hun Seol, Insun Jo, Arden L. Moore, Lucas Lindsay, Zachary H. Aitken, Michael T. Pettes, Xuesong Li, Zhen Yao, Rui Huang, David Broido, Natalio Mingo, Rodney S. Ruoff, and Li Shi. Two- dimensional phonon transport in supported graphene.Science, 2010. ResearchRabbitId e8fae85f-d2df-4ba9-ad61-e670aa87cbc8. Hawthorneet al.:Preprint submitted to ElsevierP...

    work page 2010

  2. [2]

    science, 306(5696):666–669, 2004

    Kostya S Novoselov, Andre K Geim, Sergei V Morozov, De-eng Jiang, Yanshui Zhang, Sergey V Dubonos, Irina V Grigorieva, and AlexandrAFirsov.Electricfieldeffectinatomicallythincarbonfilms. science, 306(5696):666–669, 2004

    work page 2004

  3. [3]

    Electronicpropertiesofgraphene.physica status solidi (b), 244(11):4106–4111, 2007

    KS Novoselov, SV Morozov, TMG Mohinddin, LA Ponomarenko, Daniel Cunha Elias, Rong Yang, II Barbolina, P Blake, TJ Booth, DJiang,etal. Electronicpropertiesofgraphene.physica status solidi (b), 244(11):4106–4111, 2007

    work page 2007

  4. [4]

    Defect engineering of two-dimensional ma- terialsforadvancedenergyconversionandstorage.Chemical Society Reviews, 52(5):1723–1772, 2023

    Fu Liu and Zhanxi Fan. Defect engineering of two-dimensional ma- terialsforadvancedenergyconversionandstorage.Chemical Society Reviews, 52(5):1723–1772, 2023

    work page 2023

  5. [5]

    Multiscale modeling of graphene and carbon nanostructures: advances in atomistic, coarse-grained, and machine learning approaches.RSC advances, 16(11):10193–10229, 2026

    Iman Najafipour and Santanu Chaudhuri. Multiscale modeling of graphene and carbon nanostructures: advances in atomistic, coarse-grained, and machine learning approaches.RSC advances, 16(11):10193–10229, 2026

    work page 2026

  6. [6]

    Dimitrios Maroudas, Andre R Muniz, and Ashwin Ramasubrama- niam.Structure-propertiesrelationsingraphenederivativesandmeta- materials obtained by atomic-scale modeling.Molecular Simulation, 45(14-15):1173–1202, 2019

    work page 2019

  7. [7]

    Yankun Yin, XuyuanHou, BingzeWu, Jiajun Dong, and Mingguang Yao.Heterogeneousstructurednanomaterialsfromcarbonandrelated materials.Advanced Functional Materials, 34(49):2411472, 2024

    work page 2024

  8. [8]

    Electron mi- croscopy approaches to unraveling the structure of amorphous mate- rials.Small Methods, 10(3):e01852, 2026

    Sooyeon Hwang, Hyeongjun Koh, and Judith C Yang. Electron mi- croscopy approaches to unraveling the structure of amorphous mate- rials.Small Methods, 10(3):e01852, 2026

    work page 2026

  9. [9]

    Enhanced chemical reactivity of graphene induced by mechanical strain.ACS Nano, 7(11):10335–10343, Nov 2013

    MarkA.Bissett,SatoruKonabe,SusumuOkada,MasaharuTsuji,and Hiroki Ago. Enhanced chemical reactivity of graphene induced by mechanical strain.ACS Nano, 7(11):10335–10343, Nov 2013

    work page 2013

  10. [10]

    Introduction, production, characterization andapplications of defects in graphene.Journal of Materials Science: Materials in Electronics, 32(15):19991–20030, 2021

    WaqasAhmad,ZakaUllah,NazminaImroseSonil,andKarimKhan. Introduction, production, characterization andapplications of defects in graphene.Journal of Materials Science: Materials in Electronics, 32(15):19991–20030, 2021

    work page 2021

  11. [11]

    Nanoscaleoriginandevolutionofkineticallyinducedde- fects in carbon spheres.Carbon, 96:647–660, 2016

    Sanket A Deshmukh, Badri Narayanan, Ganesh Kamath, Vilas G Pol, Jianguo Wen, Dean J Miller, and Subramanian KRS Sankara- narayanan. Nanoscaleoriginandevolutionofkineticallyinducedde- fects in carbon spheres.Carbon, 96:647–660, 2016

    work page 2016

  12. [12]

    Machine learning-driven new material discovery.Nanoscale Advances, 2020

    Jiazhen Cai, Xuan Chu, Kun Xu, Hongbo Li, and Jing Wei. Machine learning-driven new material discovery.Nanoscale Advances, 2020. ResearchRabbitId 19423290-cdf9-4437-bfc3-8e94775c1424

    work page 2020

  13. [13]

    ArtemR.Oganov,ChrisJ.Pickard,QiangZhu,andR.J.Needs.Struc- turepredictiondrivesmaterialsdiscovery.Nature Reviews Materials,

    work page

  14. [14]

    ResearchRabbitId 4352db06-0e4b-4282-839b-07272f62237a

    work page

  15. [15]

    Materials discovery at high pressures.Nature Reviews Materials, 2017

    Lijun Zhang, Yanchao Wang, Jian Lv, and Yanming Ma. Materials discovery at high pressures.Nature Reviews Materials, 2017. Re- searchRabbitId 6a53ba22-6d47-4751-88bc-a3c77ebe0ce0

    work page 2017

  16. [16]

    Interpretable machine learning applications: a promising prospect of ai for materials.Advanced Functional Materials, 35(41):2507734, 2025

    XueJiang,HuadongFu,YangBai,LeiJiang,HongtaoZhang,Weiren Wang, Peiwen Yun, Jingjin He, Dezhen Xue, Turab Lookman, et al. Interpretable machine learning applications: a promising prospect of ai for materials.Advanced Functional Materials, 35(41):2507734, 2025

    work page 2025

  17. [17]

    How to develop machinelearningmodelsforhealthcare.Nature Materials,May2019

    Po-Hsuan Cameron Chen, Yun Liu, and Lily Peng. How to develop machinelearningmodelsforhealthcare.Nature Materials,May2019. ResearchRabbitId 146692582

    work page

  18. [18]

    Charles M. Bishop. Model-based machine learning.Philosophical Transactions of the Royal Society A: Mathematical, Physical and En- gineering Sciences, Feb 2013. ResearchRabbitId 182419347

    work page 2013

  19. [19]

    Raphael M. Tromer. Dynamic collision fingerprints (dcf): Intro- ducing a new descriptor linking lattice interactions to 2d structural data signatures.Journal of Chemical Theory and Computation, 21(16):8106–8118, Aug 2025

    work page 2025

  20. [20]

    Montoya, Jiming Chen, Kyle Bystrom, Maxwell Dylla, Kyle Chard, Mark Asta, Kristin A

    Logan Ward, Alexander Dunn, Alireza Faghaninia, Nils Zimmer- mann, Saurabh Bajaj, Qi Wang, Joseph H. Montoya, Jiming Chen, Kyle Bystrom, Maxwell Dylla, Kyle Chard, Mark Asta, Kristin A. Persson, G. Jeffrey Snyder, Ian Foster, and Anubhav Jain. Matminer: Anopensourcetoolkitformaterialsdatamining.Computational ma- terials science, 2018. ResearchRabbitId 68a...

    work page 2018

  21. [21]

    Virtual prediction of material properties.Materials Today: Proceedings, 2022

    Arpan Kumar. Virtual prediction of material properties.Materials Today: Proceedings, 2022. ResearchRabbitId dc61d392-999f-42f7- a7dc-8a57df9cba6f

    work page 2022

  22. [22]

    Manifolds of quasi- constantsoapandacsffingerprintsandtheresultingfailuretomachine learnfour-bodyinteractions.Journal of Chemical Physics,2021

    Behnam Parsaeifard and Stefan Goedecker. Manifolds of quasi- constantsoapandacsffingerprintsandtheresultingfailuretomachine learnfour-bodyinteractions.Journal of Chemical Physics,2021. Re- searchRabbitId dff5586a-8b02-4a99-a612-4342cd981643

    work page 2021

  23. [23]

    Atomicenergyingrainboundaries studied by machine learning.PHYSICAL REVIEW MATERIALS,

    SongXin-yuanandChuangDeng. Atomicenergyingrainboundaries studied by machine learning.PHYSICAL REVIEW MATERIALS,

    work page

  24. [24]

    ResearchRabbitId de2b1dc6-1f16-4f6d-9786-1805a68ccff0

    work page

  25. [25]

    Systematic identification of atom-centered symmetry functions for the development of neural network poten- tials.The journal of physical chemistry

    Mohammed Wasay Mudassir, Sriram Goverapet Srinivasan, Mahesh Mynam, and Beena Rai. Systematic identification of atom-centered symmetry functions for the development of neural network poten- tials.The journal of physical chemistry. A, 2022. ResearchRabbitId f9109f19-c2eb-4ed8-85f1-50c178428674. Hawthorneet al.:Preprint submitted to ElsevierPage 19 of 20 CA...

    work page 2022

  26. [26]

    Schütt, Pan Kessel, Michael Gastegger, Kim A

    Kristof T. Schütt, Pan Kessel, Michael Gastegger, Kim A. Nicoli, Alexandre Tkatchenko, and K. Müller. Schnetpack: A deep learn- ing toolbox for atomistic systems.Journal of Chemical Theory and Computation, 2019. ResearchRabbitId 13ecd332-06c7-4b0e-9915- cb1e8ca6da3a

    work page 2019

  27. [27]

    Interpretability in the medical field: A systematic mapping and review study.Applied Soft Computing,2021

    Hajar Hakkoum, Ibtissam Abnane, and Ali Idri. Interpretability in the medical field: A systematic mapping and review study.Applied Soft Computing,2021. ResearchRabbitId89b89cd6-efc4-4a01-a753- 33484cc46938

    work page 2021

  28. [28]

    Explainable ai: A review of machine learning interpretability meth- ods.Entropy, 2020

    PantelisLinardatos,VasilisPapastefanopoulos,andSotirisKotsiantis. Explainable ai: A review of machine learning interpretability meth- ods.Entropy, 2020. ResearchRabbitId bccdecf9-21c9-402f-94a6- 9cd61af2343b

    work page 2020

  29. [29]

    Carvalho, Eduardo M

    Diogo V. Carvalho, Eduardo M. Pereira, and Jaime S. Cardoso. Machine learning interpretability: A survey on methods and met- rics.Electronics, 2019. ResearchRabbitId 8d888c37-a5c9-4ada- 847d-0d3e3fae094d

    work page 2019

  30. [30]

    Global mapping of structures and properties ofcrystalmaterials.Journal of Chemical Information and Modeling, 63(12):3814–3826, 2023

    Qinyang Li, Rongzhi Dong, Nihang Fu, Sadman Sadeed Omee, Lai Wei, and Jianjun Hu. Global mapping of structures and properties ofcrystalmaterials.Journal of Chemical Information and Modeling, 63(12):3814–3826, 2023

    work page 2023

  31. [31]

    Manifolds of quasi- constant soap and acsf fingerprints and the resulting failure to ma- chinelearnfour-bodyinteractions.The Journal of Chemical Physics, 156(3), 2022

    Behnam Parsaeifard and Stefan Goedecker. Manifolds of quasi- constant soap and acsf fingerprints and the resulting failure to ma- chinelearnfour-bodyinteractions.The Journal of Chemical Physics, 156(3), 2022

    work page 2022

  32. [32]

    Non-hexagonal-ring defects and structures induced by healing and strain in graphene and functionalized graphene.Nan- otechnology, 24(3):035708, 2013

    Joice da Silva-Araújo, AJM Nascimento, Hélio Chacham, and RW Nunes. Non-hexagonal-ring defects and structures induced by healing and strain in graphene and functionalized graphene.Nan- otechnology, 24(3):035708, 2013

    work page 2013

  33. [33]

    Explor- ing the configurational space of amorphous graphene with machine- learned atomic energies.Chemical Science, 13(46):13720–13731, 2022

    ZakariyaEl-Machachi,MarkWilson,andVolkerLDeringer. Explor- ing the configurational space of amorphous graphene with machine- learned atomic energies.Chemical Science, 13(46):13720–13731, 2022

    work page 2022

  34. [34]

    Xgboost: extreme gradient boosting.R package version 0.4-2, 1(4):1–4, 2015

    Tianqi Chen, Tong He, Michael Benesty, Vadim Khotilovich, Yuan Tang, Hyunsu Cho, Kailong Chen, Rory Mitchell, Ignacio Cano, Tianyi Zhou, et al. Xgboost: extreme gradient boosting.R package version 0.4-2, 1(4):1–4, 2015

    work page 2015

  35. [35]

    High-throughput screening of two-dimensional planar sp2 carbon space associated with a la- beled quotient graph.The Journal of Physical Chemistry Letters, 12(47):11511–11519, 2021

    Xizhi Shi, Shifang Li, Jin Li, Tao Ouyang, Chunxiao Zhang, Chao Tang, Chaoyu He, and Jianxin Zhong. High-throughput screening of two-dimensional planar sp2 carbon space associated with a la- beled quotient graph.The Journal of Physical Chemistry Letters, 12(47):11511–11519, 2021

    work page 2021

  36. [36]

    Elstner, D

    M. Elstner, D. Porezag, G. Jungnickel, J. Elsner, M. Haugk, Th. Frauenheim,S.Suhai,andG.Seifert. Self-consistent-chargedensity- functionaltight-bindingmethodforsimulationsofcomplexmaterials properties.Physical Review B, 58(11):7260–7268, September 1998

    work page 1998

  37. [37]

    Hourahine, B

    B. Hourahine, B. Aradi, V. Blum, F. Bonafé, A. Buccheri, C. Ca- macho, C. Cevallos, M. Y. Deshaye, T. Dumitrică, A. Dominguez, S.Ehlert,M.Elstner,T.vanderHeide,J.Hermann,S.Irle,J.J.Kranz, C.Köhler,T.Kowalczyk,T.Kubař,I.S.Lee,V.Lutsker,R.J.Maurer, S. K. Min, I. Mitchell, C. Negre, T. A. Niehaus, A. M. N. Niklasson, A. J. Page, A. Pecchia, G. Penazzi, M. P....

    work page 2020