Abstract
Complex systems are high-dimensional nonlinear dynamical systems with heterogeneous interactions among their constituents. To make interpretable predictions about their large-scale behaviour, it is typically assumed that these dynamics can be reduced to a few equations involving a low-rank matrix describing the network of interactions. Our Article sheds light on this low-rank hypothesis and questions its validity. Using fundamental theorems on singular-value decomposition, we probe the hypothesis for various random graphs, either by making explicit their low-rank formulation or by demonstrating the exponential decrease of their singular values. We verify the hypothesis for real networks by revealing the rapid decrease of their singular values, which has major consequences for their effective ranks. We then evaluate the impact of the low-rank hypothesis for general dynamical systems on networks through an optimal dimension reduction. This allows us to prove that recurrent neural networks can be exactly reduced, and we can connect the rapidly decreasing singular values of real networks to the dimension reduction error of the nonlinear dynamics they support. Finally, we prove that higher-order interactions naturally emerge from the dimension reduction, thus providing insights into the origin of higher-order interactions in complex systems.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 print issues and online access
$209.00 per year
only $17.42 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
Similar content being viewed by others
Data availability
All the details about the real networks data used in the paper, mostly from the network repository Netzschleuder, are given in Supplementary Information section IV. The data to generate Figs. 1, 2 and 4 are available on Zenodo (https://doi.org/10.5281/zenodo.8342130).
Code availability
The Python code used to generate the results of the paper is available on Zenodo (https://doi.org/10.5281/zenodo.8342130). The code for the optimal shrinkage of singular values is a Python implementation of the Matlab codes optimal_singval_threshold (ref. 67) and optimal_singval_shrink (ref. 24), which is partly based on the repository optht by B. Erichson.
References
Bellman, R. Dynamic Programming (Princeton Univ. Press, 1957).
Ganguli, S. & Sompolinsky, H. Compressed sensing, sparsity, and dimensionality in neuronal information processing and data analysis. Annu. Rev. Neurosci. 35, 485 (2012).
Abbott, L. F. et al. The mind of a mouse. Cell 182, 1372 (2020).
Anderson, P. W. More is different. Science 177, 393 (1972).
Strogatz, S. et al. Fifty years of ‘More is different’. Nat. Rev. Phys. 4, 508 (2022).
May, R. M. Simple mathematical models with very complicated dynamics. Nature 261, 459 (1976).
von Neumann, J. The general and logical theory of automata. In John von Neumann Collected Work, Vol. V (ed. Taub, A. H.) 288–328 (Pergamon, 1963).
Wolfram, S. Cellular automata as models of complexity. Nature 311, 419 (1984).
Parisi, G. Statistical physics and biology. Phys. World 6, 42 (1993).
Stein, D. L. & Newman, C. M. Spin Glasses and Complexity (Princeton Univ. Press, 2013).
Funahashi, K. I. & Nakamura, Y. Approximation of dynamical systems by continuous time recurrent neural networks. Neural Netw. 6, 801 (1993).
Fortunato, S. & Newman, M. E. J. 20 years of network community detection. Nat. Phys. 18, 848 (2022).
Bianconi, G. Higher-Order Networks (Cambridge Univ. Press, 2021).
Battiston, F. et al. The physics of higher-order interactions in complex systems. Nat. Phys. 17, 1093 (2021).
Wilf, H. S. The eigenvalues of a graph and its chromatic number. J. Lond. Math. Soc. 1, 330 (1967).
Donath, W. E. & Hoffman, A. J. Lower bounds for the partitioning of graphs. IBM J. Res. Dev. 17, 420 (1973).
Bonacich, P. Factoring and weighting approaches to status scores and clique identification. J. Math. Sociol. 2, 113 (1972).
Restrepo, J. G., Ott, E. & Hunt, B. R. Onset of synchronization in large networks of coupled oscillators. Phys. Rev. E 71, 036151 (2005).
Horn, R. A. & Johnson, C. R. Matrix Analysis (Cambridge Univ. Press, 2013).
Weyl, H. Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen (mit einer Anwendung auf die Theorie der Hohlraumstrahlung). Math. Ann. 71, 441 (1912).
Fan, K. Maximum properties and inequalities for the eigenvalues of completely continuous operators. Proc. Natl Acad. Sci. USA 37, 760 (1951).
Cai, J.-F., Candès, E. J. & Shen, Z. A singular value thresholding algorithm for matrix completion. SIAM J. Optim. 46, 1956 (2010).
Kutz, J. N., Brunton, S. L. & Brunton, B. W. Dynamic Mode Decomposition (SIAM, 2016).
Gavish, M. & Donoho, D. L. Optimal shrinkage of singular values. IEEE Trans. Inf. Theory 63, 2137 (2017).
Kalman, R. E. On the general theory of control systems. In Proc. 1st International IFAC Congress on Automatic and Remote Control (eds. Coales, J. F., Ragazzini, D. J. & Fuller, A. T.) 491–502 (Elsevier, 1960).
Kalman, R. E. Contributions to the theory of time-optimal control. Bol. Soc. Mat. Mex. 5, 102 (1960).
Yan, G. et al. Network control principles predict neuron function in the Caenorhabditis elegans connectome. Nature 550, 519 (2017).
Marčenko, V. A. & Pastur, L. A. Distribution of eigenvalues for some sets of random matrices. Math. USSR-Sbornik 1, 457 (1967).
Féral, D. & Péché, S. The largest eigenvalue of rank one deformation of large Wigner matrices. Commun. Math. Phys. 272, 185 (2007).
Capitaine, M., Donati-Martin, C. & Féral, D. The largest eigenvalues of finite rank deformation of large Wigner matrices: convergence and nonuniversality of the fluctuation. Ann. Probab. 37, 1 (2009).
Benaych-Georges, F. & Nadakuditi, R. R. The eigenvalues and eigenvectors of finite, low rank perturbations of large random matrices. Adv. Math. 227, 494 (2011).
Benaych-Georges, F. & Nadakuditi, R. R. The singular values and vectors of low rank perturbations of large rectangular random matrices. J. Multivar. Anal. 111, 120 (2012).
Pizzo, A., Renfrew, D. & Soshnikov, A. On finite rank deformations of Wigner matrices. Ann. I. H. Poincaré PR 49, 64–94 (2013).
Baik, J., Ben Arous, G. & Péché, S. Phase transition of the largest eigenvalue for nonnull complex sample covariance matrices. Ann. Probab. 33, 1643 (2005).
Valdano, E. & Arenas, A. Exact rank reduction of network models. Phys. Rev. X 9, 031050 (2019).
Beiran, M., Dubreuil, A., Valente, A., Mastrogiuseppe, F. & Ostojic, S. Shaping dynamics with multiple populations in low-rank recurrent networks. Neural Comput. 33, 1572 (2021).
Gao, P. & Ganguli, S. On simplicity and complexity in the brave new world of large-scale neuroscience. Curr. Opin. Neurobiol. 32, 148 (2015).
Beckermann, B. & Townsend, A. On the singular values of matrices with displacement structure. SIAM J. Matrix Anal. Appl. 38, 1227 (2017).
Udell, M. & Townsend, A. Why are big data matrices approximately low rank? SIAM J. Math. Data Sci. 1, 144 (2019).
Gao, J., Barzel, B. & Barabási, A.-L. Universal resilience patterns in complex networks. Nature 530, 307 (2016).
Tu, C., Grilli, J., Schuessler, F. & Suweis, S. Collapse of resilience patterns in generalized Lotka–Volterra dynamics and beyond. Phys. Rev. E 95, 062307 (2017).
Jiang, J. et al. Predicting tipping points in mutualistic networks through dimension reduction. Proc. Natl Acad. Sci. USA 115, E639 (2018).
Laurence, E., Doyon, N., Dubé, L. J. & Desrosiers, P. Spectral dimension reduction of complex dynamical networks. Phys. Rev. X 9, 011042 (2019).
Vegué, M., Thibeault, V., Desrosiers, P. & Allard, A. Dimension reduction of dynamics on modular and heterogeneous directed networks. PNAS Nexus 2, pgad150 (2023).
Kundu, P., Kori, H. & Masuda, N. Accuracy of a one-dimensional reduction of dynamical systems on networks. Phys. Rev. E 105, 024305 (2022).
Thibeault, V., St-Onge, G., Dubé, L. J. & Desrosiers, P. Threefold way to the dimension reduction of dynamics on networks: an application to synchronization. Phys. Rev. Res. 2, 043215 (2020).
Kuehn, C. & Bick, C. A universal route to explosive phenomena. Sci. Adv. 7, 1 (2021).
St-Onge, G., Thibeault, V., Allard, A., Dubé, L. J. & Hébert-Dufresne, L. Social confinement and mesoscopic localization of epidemics on networks. Phys. Rev. Lett. 126, 098301 (2021).
Battiston, F. et al. Networks beyond pairwise interactions: structure and dynamics. Phys. Rep. 874, 1 (2020).
Matheny, M. H. et al. Exotic states in a simple network of nanoelectromechanical oscillators. Science 363, 1057 (2019).
Nijholt, E., Ocampo-Espindola, J. L., Eroglu, D., Kiss, I. Z. & Pereira, T. Emergent hypernetworks in weakly coupled oscillators. Nat. Commun. 13, 4849 (2022).
Gallo, G., Longo, G., Pallottino, S. & Nguyen, S. Directed hypergraphs and applications. Discret. Appl. Math. 42, 177 (1993).
Palla, G., Derényi, I., Farkas, I. & Vicsek, T. Uncovering the overlapping community structure of complex networks in nature and society. Nature 435, 814 (2005).
Yu, S. et al. Higher-order interactions characterized in cortical activity. J. Neurosci. 31, 17514 (2011).
Mayfield, M. M. & Stouffer, D. B. Higher-order interactions capture unexplained complexity in diverse communities. Nat. Ecol. Evol. 1, 1 (2017).
Ferraz de Arruda, G., Tizzani, M. & Moreno, Y. Phase transitions and stability of dynamical processes on hypergraphs. Commun. Phys. 4, 24 (2021).
Qi, L. & Luo, Z. Tensor Analysis (SIAM, 2017).
Watanabe, S. & Strogatz, S. H. Constants of motion for superconducting Josephson arrays. Phys. D 74, 197 (1994).
Brunton, S. L., Budišić, M., Kaiser, E. & Kutz, J. N. Modern Koopman theory for dynamical systems. SIAM Rev. 64, 229 (2022).
Valente, A., Pillow, J. W. & Ostojic, S. Extracting computational mechanisms from neural data using low-rank RNNs. In Advances in Neural Information Processing Systems (eds. Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K. & Oh, A.) 24072–24086 (Curran Associates, Inc., 2022).
Holland, J. H. Hidden Order: How Adaptation Builds Complexity (Addison-Wesley, 1995).
Montanari, A. N., Duan, C., Aguirre, L. A. & Motter, A. E. Functional observability and target state estimation in large-scale networks. Proc. Natl Acad. Sci. USA 119, e2113750119 (2022).
Sanhedrai, H. et al. Reviving a failed network through microscopic interventions. Nat. Phys. 18, 338 (2022).
Desrosiers, P. & Roy-Pomerleau, X. One for all. Nat. Phys. 18, 238 (2022).
Martin, C. H. & Mahoney, M. W. Implicit self-regularization in deep neural networks: evidence from random matrix theory and implications for learning. J. Mach. Learn. Res. 22, 1 (2021).
Gower, J. Properties of Euclidean and non-Euclidean distance matrices. Linear Algebra Appl. 67, 81 (1985).
Gavish, M. & Donoho, D. L. The optimal hard threshold for singular values is \(4/\sqrt{3}\). IEEE Trans. Inf. Theory 60, 5040 (2014).
Donoho, D., Gavish, M. & Johnstone, I. Optimal shrinkage of eigenvalues in the spiked covariance model. Ann. Statis. 46, 1742 (2018).
Scheffer, L. K. et al. A connectome and analysis of the adult Drosophila central brain. eLife 9, 1 (2020).
Malinowski, E. R. Theory of error in factor analysis. Anal. Chem. 49, 606 (1977).
Sánchez, E. & Kowalski, B. R. Generalized rank annihilation factor analysis. Anal. Chem. 58, 496 (1986).
Abdi, H. & Williams, L. J. Principal component analysis. WIREs Comput. Stat. 2, 433 (2010).
Almagro, P., Boguñá, M. & Ángeles Serrano, M. Detecting the ultra low dimensionality of real networks. Nat. Commun. 13, 6096 (2022).
Lynn, C. W. & Bassett, D. S. Compressibility of complex networks. Proc. Natl Acad. Sci. USA 118, e2023473118 (2021).
Perry, P. O. Cross-Validation for Unsupervised Learning. PhD thesis, Stanford Univ. (2009).
Städter, P., Schälte, Y., Schmiester, L., Hasenauer, J. & Stapor, P. L. Benchmarking of numerical integration methods for ODE models of biological systems. Sci. Rep. 11, 2696 (2021).
Sompolinsky, H., Crisanti, A. & Sommers, H.-J. Chaos in random neural networks. Phys. Rev. Lett. 61, 259 (1988).
Acknowledgements
We are grateful to G. Eilerstein for sharing the code to extract the weight matrices from the repository NWS, G. Jékely for sharing the neuronal and desmosomal connectomes of Platynereis dumerilii, C. Murphy for useful discussions on artificial neural networks, G. St-Onge for his comments on the preprint and X. Roy-Pomerleau for helping us to explore the microbial dynamics numerically. We thank É. Boran for his fundamental contribution to linear algebra. This work was supported by the Fonds de recherche du Québec—Nature et technologies (V.T. and P.D.), the Natural Sciences and Engineering Research Council of Canada (V.T., A.A. and P.D.) and the Sentinelle Nord programme of Université Laval, funded by the Canada First Research Excellence Fund (V.T., A.A. and P.D.).
Author information
Authors and Affiliations
Contributions
All authors contributed to the formulation of the study, the interpretation of the results and the editing of the paper. V.T. and P.D. obtained the mathematical results and conceived the conceptual basis of the project. V.T. led the writing of the manuscript, wrote the supplementary information with P.D., designed the figures, wrote the code and performed the numerical experiments to generate the results. V.T., A.A. and P.D. contributed to the code and analysed the data to generate Fig. 1.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Physics thanks Jianxi Gao and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Supplementary information
Supplementary Information
Supplementary mathematical details to support the results of the paper, Fig. 1–12, discussion and details of the real network data.
Supplementary Table 1
The table lists all the real networks that were used in the paper. All the names have a clickable link to the source of the data, along with their number of vertices N, their rank and their effective ranks.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Thibeault, V., Allard, A. & Desrosiers, P. The low-rank hypothesis of complex systems. Nat. Phys. 20, 294–302 (2024). https://doi.org/10.1038/s41567-023-02303-0
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s41567-023-02303-0
This article is cited by
-
Intrinsic simplicity of complex systems
Nature Physics (2024)