The complexity dynamics of grokking
We demonstrate the existence of a complexity phase transition in neural networks by studying the grokking phenomenon, where networks suddenly transition from memorization to generalization long after overfitting their training data. To characterize this phase transition, we introduce a theoretical f...
Gespeichert in:
| Veröffentlicht in: | Physica. D Jg. 482; S. 134859 |
|---|---|
| Hauptverfasser: | , , , , |
| Format: | Journal Article |
| Sprache: | Englisch |
| Veröffentlicht: |
Elsevier B.V
01.11.2025
|
| Schlagworte: | |
| ISSN: | 0167-2789 |
| Online-Zugang: | Volltext |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| Abstract | We demonstrate the existence of a complexity phase transition in neural networks by studying the grokking phenomenon, where networks suddenly transition from memorization to generalization long after overfitting their training data. To characterize this phase transition, we introduce a theoretical framework for measuring complexity based on rate–distortion theory and Kolmogorov complexity, which can be understood as principled lossy compression for networks. We find that properly regularized networks exhibit a sharp phase transition: complexity rises during memorization, then falls as the network discovers a simpler underlying pattern that generalizes. In contrast, unregularized networks remain trapped in a high-complexity memorization phase. We establish an explicit connection between our complexity measure and generalization bounds, providing a theoretical foundation for the link between lossy compression and generalization. Our framework achieves compression ratios 30-40× better than naïve approaches, enabling precise tracking of complexity dynamics. Finally, we introduce a regularization method based on spectral entropy that encourages networks toward low-complexity representations by penalizing their intrinsic dimension. |
|---|---|
| AbstractList | We demonstrate the existence of a complexity phase transition in neural networks by studying the grokking phenomenon, where networks suddenly transition from memorization to generalization long after overfitting their training data. To characterize this phase transition, we introduce a theoretical framework for measuring complexity based on rate–distortion theory and Kolmogorov complexity, which can be understood as principled lossy compression for networks. We find that properly regularized networks exhibit a sharp phase transition: complexity rises during memorization, then falls as the network discovers a simpler underlying pattern that generalizes. In contrast, unregularized networks remain trapped in a high-complexity memorization phase. We establish an explicit connection between our complexity measure and generalization bounds, providing a theoretical foundation for the link between lossy compression and generalization. Our framework achieves compression ratios 30-40× better than naïve approaches, enabling precise tracking of complexity dynamics. Finally, we introduce a regularization method based on spectral entropy that encourages networks toward low-complexity representations by penalizing their intrinsic dimension. |
| ArticleNumber | 134859 |
| Author | Posner, Ingmar Sapora, Silvia Foerster, Jakob Hawes, Nick DeMoss, Branton |
| Author_xml | – sequence: 1 givenname: Branton orcidid: 0000-0001-6828-6787 surname: DeMoss fullname: DeMoss, Branton email: bdemoss@robots.ox.ac.uk – sequence: 2 givenname: Silvia surname: Sapora fullname: Sapora, Silvia – sequence: 3 givenname: Jakob surname: Foerster fullname: Foerster, Jakob – sequence: 4 givenname: Nick surname: Hawes fullname: Hawes, Nick – sequence: 5 givenname: Ingmar surname: Posner fullname: Posner, Ingmar |
| BookMark | eNp9j71OwzAYRT0UibbwBCyRmBP8l9gZGFAFFKkSS5ktY39unTZxZEeIvD0pYWa6w9W5umeFFl3oAKE7gguCSfXQFP1xTLagmJYFYVyW9QItp0bkVMj6Gq1SajDGRDCxRPf7I2QmtP0Zvv0wZnbsdOtNyoLLDjGcTr473KArp88Jbv9yjT5envebbb57f33bPO1ywzAeci6Z1p-CEScrLcBpKaUzpaQ1rSrgkggrRIlNbYBDaTQAr4UW1LEKCJWWrRGbd00MKUVwqo--1XFUBKuLm2rUr5u6uKnZbaIeZwqma18eokrGQ2fA-ghmUDb4f_kf9xFcJw |
| Cites_doi | 10.1103/PhysRevE.111.014118 10.1007/978-1-4757-3860-5 10.1017/S0960129511000521 10.1145/168304.168306 10.1016/0005-1098(78)90005-5 10.1002/j.1538-7305.1948.tb00917.x 10.1103/PhysRevResearch.2.033312 10.1109/TIT.2010.2048491 |
| ContentType | Journal Article |
| Copyright | 2025 The Authors |
| Copyright_xml | – notice: 2025 The Authors |
| DBID | 6I. AAFTH AAYXX CITATION |
| DOI | 10.1016/j.physd.2025.134859 |
| DatabaseName | ScienceDirect Open Access Titles Elsevier:ScienceDirect:Open Access CrossRef |
| DatabaseTitle | CrossRef |
| DatabaseTitleList | |
| DeliveryMethod | fulltext_linktorsrc |
| Discipline | Engineering Physics |
| ExternalDocumentID | 10_1016_j_physd_2025_134859 S0167278925003367 |
| GroupedDBID | --K --M -~X .~1 0R~ 1B1 1RT 1~. 1~5 29O 4.4 457 4G. 5VS 6I. 7-5 71M 8P~ 9JN AAEDT AAEDW AAFTH AAIKJ AAKOC AALRI AAOAW AAQFI AAQXK AATTM AAXKI AAXUO AAYWO ABAOU ABFNM ABMAC ABNEU ABWVN ABXDB ACDAQ ACFVG ACGFS ACLOT ACNCT ACNNM ACRLP ACRPL ADBBV ADEZE ADGUI ADIYS ADMUD ADNMO ADVLN AEBSH AEIPS AEKER AFFNX AFJKZ AFTJW AGHFR AGQPQ AGUBO AGYEJ AHHHB AIEXJ AIGVJ AIIUN AIKHN AITUG AIVDX ALMA_UNASSIGNED_HOLDINGS AMRAJ ANKPU APXCP ARUGR ASPBG AVWKF AXJTR AZFZN BBWZM BKOJK BLXMC EBS EFJIC EFKBS EFLBG EJD EO8 EO9 EP2 EP3 F5P FDB FEDTE FGOYB FIRID FNPLU FYGXN G-Q GBLVA HMV HVGLF HZ~ H~9 IHE J1W K-O KOM M38 M41 MHUIS MO0 MVM N9A NDZJH O-L O9- OAUVE OGIMB OZT P-8 P-9 P2P PC. Q38 R2- RNS ROL RPZ SDF SDG SDP SES SEW SPC SPCBC SPD SPG SSQ SSW SSZ T5K TN5 TWZ WUQ XJT XPP YNT YYP ~02 ~G- ~HD 9DU AAYXX CITATION |
| ID | FETCH-LOGICAL-c300t-483aab731f86a7efa888fc5829266e4817d7750c9ce4e5caee497a72f36e128d3 |
| ISICitedReferencesCount | 0 |
| ISICitedReferencesURI | http://www.webofscience.com/api/gateway?GWVersion=2&SrcApp=Summon&SrcAuth=ProQuest&DestLinkType=CitingArticles&DestApp=WOS_CPL&KeyUT=001560597300003&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D |
| ISSN | 0167-2789 |
| IngestDate | Sat Nov 29 06:52:14 EST 2025 Sat Oct 11 16:52:49 EDT 2025 |
| IsDoiOpenAccess | true |
| IsOpenAccess | true |
| IsPeerReviewed | true |
| IsScholarly | true |
| Keywords | Algorithmic complexity Grokking Phase transition |
| Language | English |
| License | This is an open access article under the CC BY license. |
| LinkModel | OpenURL |
| MergedId | FETCHMERGED-LOGICAL-c300t-483aab731f86a7efa888fc5829266e4817d7750c9ce4e5caee497a72f36e128d3 |
| ORCID | 0000-0001-6828-6787 |
| OpenAccessLink | https://dx.doi.org/10.1016/j.physd.2025.134859 |
| ParticipantIDs | crossref_primary_10_1016_j_physd_2025_134859 elsevier_sciencedirect_doi_10_1016_j_physd_2025_134859 |
| PublicationCentury | 2000 |
| PublicationDate | November 2025 2025-11-00 |
| PublicationDateYYYYMMDD | 2025-11-01 |
| PublicationDate_xml | – month: 11 year: 2025 text: November 2025 |
| PublicationDecade | 2020 |
| PublicationTitle | Physica. D |
| PublicationYear | 2025 |
| Publisher | Elsevier B.V |
| Publisher_xml | – name: Elsevier B.V |
| References | Power, Burda, Edwards, Babuschkin, Misra (b1) 2022 Vapnik, Chervonenkis (b7) 1971 Ma, Wang, Ma, Wang, Wang, Huang, Dong, Wang, Xue, Wei (b22) 2024 Micikevicius, Narang, Alben, Diamos, Elsen, García, Ginsburg, Houston, Kuchaiev, Venkatesh, Wu (b21) 2017 Nanda, Chan, Lieberum, Smith, Steinhardt (b11) 2023 Shannon (b4) 1948; 27 Vereshchagin, Vitányi (b20) 2004; 56 Li, Farkhoor, Liu, Yosinski (b26) 2018 Rissanen (b3) 1978; 14 Bartlett, Mendelson (b6) 2003; 3 M. Li, P.M.B. Vitányi, An introduction to kolmogorov complexity and its applications, in: Graduate Texts in Computer Science, 1993. Langford, Seeger (b17) 2001 Kolchinsky, Wolpert (b31) 2020; 2 Liu, Kitouni, Nolte, Michaud, Tegmark, Williams (b8) 2022 Liu, Michaud, Tegmark (b9) 2022 Kaplan, McCandlish, Henighan, Brown, Chess, Child, Gray, Radford, Wu, Amodei (b13) 2020 Y. Blau, T. Michaeli, Rethinking lossy compression: The rate–distortion-perception tradeoff, in: International Conference on Machine Learning, 2019. V.N. Vapnik, Principles of risk minimization for learning theory, in: Neural Information Processing Systems, 1991. Baez, Stay (b30) 2012; 22 Goldblum, Finzi, Rowan, Wilson (b16) 2023 Nakkiran, Kaplun, Bansal, Yang, Barak, Sutskever (b12) 2019; 2021 Ebtekar, Hutter (b29) 2025; 111 Aghajanyan, Zettlemoyer, Gupta (b27) 2020 Dettmers, Pagnoni, Holtzman, Zettlemoyer (b25) 2023 Aaronson, Carroll, Ouellette (b2) 2014 Henighan, Kaplan, Katz, Chen, Hesse, Jackson, Jun, Brown, Dhariwal, Gray, Hallacy, Mann, Radford, Ramesh, Ryder, Ziegler, Schulman, Amodei, McCandlish (b14) 2020 Varma, Shah, Kenton, Kram’ar, Kumar (b10) 2023 O. Roy, M. Vetterli, The effective rank: A measure of effective dimensionality, in: 2007 15th European Signal Processing Conference, 2007, pp. 606–610. Lotfi, Finzi, Kuang, Rudner, Goldblum, Wilson (b15) 2024 G.E. Hinton, D. van Camp, Keeping the neural networks simple by minimizing the description length of the weights, in: Annual Conference Computational Learning Theory, 1993. Hu, Shen, Wallis, Allen-Zhu, Li, Wang, Chen (b24) 2021 Ma (10.1016/j.physd.2025.134859_b22) 2024 10.1016/j.physd.2025.134859_b18 10.1016/j.physd.2025.134859_b19 Varma (10.1016/j.physd.2025.134859_b10) 2023 Henighan (10.1016/j.physd.2025.134859_b14) 2020 Kaplan (10.1016/j.physd.2025.134859_b13) 2020 Liu (10.1016/j.physd.2025.134859_b8) 2022 Rissanen (10.1016/j.physd.2025.134859_b3) 1978; 14 Shannon (10.1016/j.physd.2025.134859_b4) 1948; 27 Lotfi (10.1016/j.physd.2025.134859_b15) 2024 Liu (10.1016/j.physd.2025.134859_b9) 2022 Baez (10.1016/j.physd.2025.134859_b30) 2012; 22 Vereshchagin (10.1016/j.physd.2025.134859_b20) 2004; 56 10.1016/j.physd.2025.134859_b23 Nakkiran (10.1016/j.physd.2025.134859_b12) 2019; 2021 Power (10.1016/j.physd.2025.134859_b1) 2022 Aaronson (10.1016/j.physd.2025.134859_b2) 2014 Hu (10.1016/j.physd.2025.134859_b24) 2021 Langford (10.1016/j.physd.2025.134859_b17) 2001 10.1016/j.physd.2025.134859_b28 Goldblum (10.1016/j.physd.2025.134859_b16) 2023 Bartlett (10.1016/j.physd.2025.134859_b6) 2003; 3 Nanda (10.1016/j.physd.2025.134859_b11) 2023 Micikevicius (10.1016/j.physd.2025.134859_b21) 2017 10.1016/j.physd.2025.134859_b5 Vapnik (10.1016/j.physd.2025.134859_b7) 1971 Dettmers (10.1016/j.physd.2025.134859_b25) 2023 Li (10.1016/j.physd.2025.134859_b26) 2018 Ebtekar (10.1016/j.physd.2025.134859_b29) 2025; 111 Aghajanyan (10.1016/j.physd.2025.134859_b27) 2020 Kolchinsky (10.1016/j.physd.2025.134859_b31) 2020; 2 |
| References_xml | – year: 2014 ident: b2 article-title: Quantifying the rise and fall of complexity in closed systems: The coffee automaton – year: 2020 ident: b14 article-title: Scaling laws for autoregressive generative modeling – reference: Y. Blau, T. Michaeli, Rethinking lossy compression: The rate–distortion-perception tradeoff, in: International Conference on Machine Learning, 2019. – reference: G.E. Hinton, D. van Camp, Keeping the neural networks simple by minimizing the description length of the weights, in: Annual Conference Computational Learning Theory, 1993. – year: 2021 ident: b24 article-title: Lora: Low-rank adaptation of large language models – year: 2017 ident: b21 article-title: Mixed precision training – year: 2023 ident: b25 article-title: Qlora: Efficient finetuning of quantized llms – volume: 14 start-page: 465 year: 1978 end-page: 471 ident: b3 article-title: Modeling by shortest data description* publication-title: Autom. – year: 2023 ident: b10 article-title: Explaining grokking through circuit efficiency – volume: 111 year: 2025 ident: b29 article-title: Foundations of algorithmic thermodynamics publication-title: Phys. Rev. E – year: 2020 ident: b13 article-title: Scaling laws for neural language models – year: 2022 ident: b8 article-title: Towards understanding grokking: An effective theory of representation learning – reference: M. Li, P.M.B. Vitányi, An introduction to kolmogorov complexity and its applications, in: Graduate Texts in Computer Science, 1993. – volume: 3 start-page: 463 year: 2003 end-page: 482 ident: b6 article-title: Rademacher and gaussian complexities: Risk bounds and structural results publication-title: J. Mach. Learn. Res. – reference: O. Roy, M. Vetterli, The effective rank: A measure of effective dimensionality, in: 2007 15th European Signal Processing Conference, 2007, pp. 606–610. – year: 2023 ident: b11 article-title: Progress measures for grokking via mechanistic interpretability – volume: 56 start-page: 3438 year: 2004 end-page: 3454 ident: b20 article-title: Rate distortion and denoising of individual data using kolmogorov complexity publication-title: IEEE Trans. Inform. Theory – year: 2018 ident: b26 article-title: Measuring the intrinsic dimension of objective landscapes – volume: 2 year: 2020 ident: b31 article-title: Thermodynamic costs of turing machines publication-title: Phys. Rev. Res. – year: 2023 ident: b16 article-title: The no free lunch theorem, kolmogorov complexity, and the role of inductive biases in machine learning – year: 2022 ident: b1 article-title: Grokking: Generalization beyond overfitting on small algorithmic datasets – volume: 2021 year: 2019 ident: b12 article-title: Deep double descent: where bigger models and more data hurt publication-title: J. Stat. Mech. Theory Exp. – volume: 27 start-page: 623 year: 1948 end-page: 656 ident: b4 article-title: A mathematical theory of communication publication-title: Bell Syst. Tech. J. – year: 1971 ident: b7 article-title: On the uniform convergence of relative frequencies of events to their probabilities – year: 2022 ident: b9 article-title: Omnigrok: Grokking beyond algorithmic data – year: 2024 ident: b15 article-title: Non-vacuous generalization bounds for large language models publication-title: ICML 2024 – year: 2020 ident: b27 article-title: Intrinsic dimensionality explains the effectiveness of language model fine-tuning – year: 2024 ident: b22 article-title: The era of 1-bit llms: All large language models are in 1.58 bits – volume: 22 start-page: 771 year: 2012 end-page: 787 ident: b30 article-title: Algorithmic thermodynamics publication-title: Math. Structures Comput. Sci. – reference: V.N. Vapnik, Principles of risk minimization for learning theory, in: Neural Information Processing Systems, 1991. – year: 2001 ident: b17 article-title: Bounds for averaging classifiers. Carnegie mellon – year: 2018 ident: 10.1016/j.physd.2025.134859_b26 – ident: 10.1016/j.physd.2025.134859_b28 – year: 2022 ident: 10.1016/j.physd.2025.134859_b9 – volume: 111 year: 2025 ident: 10.1016/j.physd.2025.134859_b29 article-title: Foundations of algorithmic thermodynamics publication-title: Phys. Rev. E doi: 10.1103/PhysRevE.111.014118 – year: 2017 ident: 10.1016/j.physd.2025.134859_b21 – year: 2023 ident: 10.1016/j.physd.2025.134859_b25 – year: 2022 ident: 10.1016/j.physd.2025.134859_b8 – year: 2014 ident: 10.1016/j.physd.2025.134859_b2 – year: 2020 ident: 10.1016/j.physd.2025.134859_b13 – ident: 10.1016/j.physd.2025.134859_b5 – volume: 3 start-page: 463 year: 2003 ident: 10.1016/j.physd.2025.134859_b6 article-title: Rademacher and gaussian complexities: Risk bounds and structural results publication-title: J. Mach. Learn. Res. – year: 2023 ident: 10.1016/j.physd.2025.134859_b10 – volume: 2021 year: 2019 ident: 10.1016/j.physd.2025.134859_b12 article-title: Deep double descent: where bigger models and more data hurt publication-title: J. Stat. Mech. Theory Exp. – ident: 10.1016/j.physd.2025.134859_b18 doi: 10.1007/978-1-4757-3860-5 – year: 2020 ident: 10.1016/j.physd.2025.134859_b14 – year: 2024 ident: 10.1016/j.physd.2025.134859_b15 article-title: Non-vacuous generalization bounds for large language models – year: 2022 ident: 10.1016/j.physd.2025.134859_b1 – volume: 22 start-page: 771 year: 2012 ident: 10.1016/j.physd.2025.134859_b30 article-title: Algorithmic thermodynamics publication-title: Math. Structures Comput. Sci. doi: 10.1017/S0960129511000521 – ident: 10.1016/j.physd.2025.134859_b23 doi: 10.1145/168304.168306 – volume: 14 start-page: 465 year: 1978 ident: 10.1016/j.physd.2025.134859_b3 article-title: Modeling by shortest data description* publication-title: Autom. doi: 10.1016/0005-1098(78)90005-5 – year: 2001 ident: 10.1016/j.physd.2025.134859_b17 – year: 2021 ident: 10.1016/j.physd.2025.134859_b24 – ident: 10.1016/j.physd.2025.134859_b19 – volume: 27 start-page: 623 year: 1948 ident: 10.1016/j.physd.2025.134859_b4 article-title: A mathematical theory of communication publication-title: Bell Syst. Tech. J. doi: 10.1002/j.1538-7305.1948.tb00917.x – year: 2023 ident: 10.1016/j.physd.2025.134859_b11 – volume: 2 year: 2020 ident: 10.1016/j.physd.2025.134859_b31 article-title: Thermodynamic costs of turing machines publication-title: Phys. Rev. Res. doi: 10.1103/PhysRevResearch.2.033312 – volume: 56 start-page: 3438 year: 2004 ident: 10.1016/j.physd.2025.134859_b20 article-title: Rate distortion and denoising of individual data using kolmogorov complexity publication-title: IEEE Trans. Inform. Theory doi: 10.1109/TIT.2010.2048491 – year: 2024 ident: 10.1016/j.physd.2025.134859_b22 – year: 1971 ident: 10.1016/j.physd.2025.134859_b7 – year: 2020 ident: 10.1016/j.physd.2025.134859_b27 – year: 2023 ident: 10.1016/j.physd.2025.134859_b16 |
| SSID | ssj0001737 |
| Score | 2.472396 |
| Snippet | We demonstrate the existence of a complexity phase transition in neural networks by studying the grokking phenomenon, where networks suddenly transition from... |
| SourceID | crossref elsevier |
| SourceType | Index Database Publisher |
| StartPage | 134859 |
| SubjectTerms | Algorithmic complexity Grokking Phase transition |
| Title | The complexity dynamics of grokking |
| URI | https://dx.doi.org/10.1016/j.physd.2025.134859 |
| Volume | 482 |
| WOSCitedRecordID | wos001560597300003&url=https%3A%2F%2Fcvtisr.summon.serialssolutions.com%2F%23%21%2Fsearch%3Fho%3Df%26include.ft.matches%3Dt%26l%3Dnull%26q%3D |
| hasFullText | 1 |
| inHoldings | 1 |
| isFullTextHit | |
| isPrint | |
| journalDatabaseRights | – providerCode: PRVESC databaseName: Elsevier SD Freedom Collection Journals 2021 issn: 0167-2789 databaseCode: AIEXJ dateStart: 19950101 customDbUrl: isFulltext: true dateEnd: 99991231 titleUrlDefault: https://www.sciencedirect.com omitProxy: false ssIdentifier: ssj0001737 providerName: Elsevier |
| link | http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwtV1RT9swED5tZZPgATG2CRhDkbbHpWoSJ7YfYYAGggppbOpb5Dg2ajslqC3Qn885ttOiTtOYtBcrimQnuS8-fz6dvwP4zHTGzU4ixNVah4RlRSgUJzjxOElKIiVpsip_XtB-nw0G_MrV0Zs25QRoVbH5nN_-V6jxHoJtjs4-A-52ULyB1wg6tgg7tn8NfJMoruaGYZe25HyTsXEzqcdjv1Y5RnplgeouUn-P1aUrpX40aUoMt1EYYVSsmnjp8Nf9sHXop7XhkBb7czGui4Vbe7BeqO-19118IU7dQbs26LVy8MXGIdG_mkO0y46U2DJCK07ZxgdGXROrMeKscdqNEsKcEvhTtevvZmQzMFKzXpJk9CWsxTTlrANrh2cng_N2mY2oFUT1b-IlpZrkvZVH_Z52LFGJ6y3YdHuA4NBi9wZeqGobNpaUIbfhtYVl-hY-IZ7BAs_A4xnUOvB4voMfpyfXX7-FrrJFKJNeb2YiuEIUNIk0ywRVWjDGtExZzJEvKcIiWlKkcpJLRVQqhVKEU0FjnWQKCUWZvIdOVVdqB4JCZcjY4wJnmSakLJGA0aLUUVSyNGU9uQtf_Ifnt1bAJPeZfaO8sVNu7JRbO-1C5o2TOw5muVWOaP6p496_dvwA64vfbh86s8md-giv5P1sOJ0cONQfAeuQU30 |
| linkProvider | Elsevier |
| openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=The+complexity+dynamics+of+grokking&rft.jtitle=Physica.+D&rft.au=DeMoss%2C+Branton&rft.au=Sapora%2C+Silvia&rft.au=Foerster%2C+Jakob&rft.au=Hawes%2C+Nick&rft.date=2025-11-01&rft.pub=Elsevier+B.V&rft.issn=0167-2789&rft.volume=482&rft_id=info:doi/10.1016%2Fj.physd.2025.134859&rft.externalDocID=S0167278925003367 |
| thumbnail_l | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=0167-2789&client=summon |
| thumbnail_m | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=0167-2789&client=summon |
| thumbnail_s | http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=0167-2789&client=summon |