pith. sign in

arxiv: 2311.16981 · v1 · submitted 2023-11-28 · ⚛️ physics.soc-ph

Temporal networks with node-specific memory: unbiased inference of transition probabilities, relaxation times and structural breaks

Giulio Virginio Clemente , Claudio J. Tessone , Diego Garlaschelli This is my paper

Pith reviewed 2026-05-24 05:59 UTC · model grok-4.3

classification ⚛️ physics.soc-ph
keywords temporal networksmaximum entropyIsing modelconfiguration modeltransition probabilitiesrelaxation timesstructural breaksnetwork memory
0
0 comments X

The pith

A mapping of temporal networks to a heterogeneous one-dimensional Ising model disentangles hidden variables that set both structure and link persistence.

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

The paper develops a maximum-entropy framework for time-varying networks that enforces both the observed degree sequence and the observed time-persisting degree sequence at once. An exact mapping to a heterogeneous version of the one-dimensional Ising model supplies closed-form expressions for the parameters that jointly govern static connectivity and temporal memory. Likelihood-based model selection then chooses the coarsest level (global, node-specific, or dyadic) at which memory must be retained. When the resulting model is fitted to an empirical social network, it supplies corrected estimates of dyadic transition probabilities, relaxation times, and the locations of switches between stationary regimes. The central object is the generalized configuration model whose sufficient statistics are the two degree sequences; the Ising mapping renders inference of its parameters analytic rather than numerical.

Core claim

The observed network trajectory is generated by a generalized configuration model conditioned on given degrees and given time-persisting degrees. An exact mapping to a heterogeneous one-dimensional Ising model yields an analytic solution that rigorously separates the hidden variables responsible for static structure from those responsible for link persistence. Model selection via likelihood maximization identifies the most parsimonious structural level (global, node-specific or dyadic) needed to capture memory effects, thereby producing unbiased estimates of dyadic transition probabilities, relaxation times and structural breaks between dynamical regimes.

What carries the argument

Exact mapping of the temporal network trajectory onto a heterogeneous one-dimensional Ising model, which supplies closed-form expressions for the hidden variables that control both degrees and time-persisting degrees.

If this is right

  • Transition probabilities between node pairs become estimable without systematic bias from heterogeneous persistence.
  • Relaxation times can be extracted separately from the effects of degree heterogeneity.
  • Structural breaks are located by testing for changes in the inferred hidden variables across time windows.
  • The framework automatically selects whether memory is best described at the global, node, or dyad level.

Where Pith is reading between the lines

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

  • The same mapping could be used to test whether observed long-range temporal correlations in other systems are artifacts of unaccounted degree persistence.
  • If the model is correct, many existing estimates of memory in social and biological networks would need downward revision once node-specific persistence is controlled.
  • The analytic solution opens the possibility of deriving exact expressions for higher-order temporal statistics such as motif persistence times.

Load-bearing premise

The observed sequence of networks is generated by a maximum-entropy ensemble whose only constraints are the empirical degree sequence and the empirical time-persisting degree sequence.

What would settle it

Generate synthetic trajectories from a process whose constraints include quantities beyond degrees and persisting degrees; the Ising-based estimators should then return systematically biased transition probabilities or select an incorrect structural level.

Figures

Figures reproduced from arXiv: 2311.16981 by Claudio J. Tessone, Diego Garlaschelli, Giulio Virginio Clemente.

Figure 1
Figure 1. Figure 1: FIG. 1. In figure we have the plot of the normalized empirical [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2. Here we show the comparison between the persisting [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: FIG. 3. In figure are depicted the points known as real events, [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: FIG. 4. Here we show the distribution of of average second eigenvalues per node, computed using the model [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5. The figures display boxplots representing the distributions of second eigenvalues associated with links from three [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
read the original abstract

One of the main challenges in the study of time-varying networks is the interplay of memory effects with structural heterogeneity. In particular, different nodes and dyads can have very different statistical properties in terms of both link formation and link persistence, leading to a superposition of typical timescales, sub-optimal parametrizations and substantial estimation biases. Here we develop an unbiased maximum-entropy framework to study empirical network trajectories by controlling for the observed structural heterogeneity and local link persistence simultaneously. An exact mapping to a heterogeneous version of the one-dimensional Ising model leads to an analytic solution that rigorously disentangles the hidden variables that jointly determine both static and temporal properties. Additionally, model selection via likelihood maximization identifies the most parsimonious structural level (either global, node-specific or dyadic) accounting for memory effects. As we illustrate on a real-world social network, this method enables an improved estimation of dyadic transition probabilities, relaxation times and structural breaks between dynamical regimes. In the resulting picture, the graph follows a generalized configuration model with given degrees and given time-persisting degrees, undergoing transitions between empirically identifiable stationary regimes.

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

2 major / 2 minor

Summary. The paper develops a maximum-entropy ensemble for temporal network trajectories conditioned on observed node degrees and time-persisting degrees (a generalized configuration model). It claims an exact mapping of this ensemble to a heterogeneous one-dimensional Ising model whose transfer-matrix solution yields closed-form expressions for dyadic transition probabilities and relaxation times, plus a likelihood-based procedure to select among global, node-specific, or dyadic memory models and to detect structural breaks between stationary regimes.

Significance. If the claimed exact mapping holds without approximation, the work supplies an analytic, unbiased route to separate structural heterogeneity from temporal memory, improving estimation of transition probabilities and relaxation times on empirical data. The provision of closed-form expressions and a principled model-selection step for breaks constitutes a clear methodological advance over simulation-based or ad-hoc approaches.

major comments (2)
  1. [§4, Eq. (15)–(18)] §4, Eq. (15)–(18): the mapping writes the partition function as a product of independent per-dyad transfer matrices; because the underlying configuration-model constraints are global (sum of link indicators must exactly equal the prescribed degree sequence), an explicit demonstration is required that the Lagrange multipliers for the degree constraints are recovered exactly by the per-dyad solution rather than only in the sparse/large-N limit.
  2. [§5.1, likelihood (22)] §5.1, likelihood (22): the model-selection procedure for identifying the parsimonious memory level and structural breaks inherits any bias present in the transition probabilities; if the Ising mapping is only approximate under global constraints, the reported likelihood ratios and break locations on the real-world example are correspondingly approximate.
minor comments (2)
  1. [§2] Notation for the time-persisting degree sequence is introduced only in the abstract and final sentence; a clear definition with an equation in §2 would improve readability.
  2. [Figure 3] Figure 3 caption does not state the numerical values of the fitted node-specific memory parameters; adding them would allow direct comparison with the analytic expressions.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their careful reading and constructive comments. Below we address each major comment point by point, defending the exactness of the mapping while agreeing to add an explicit demonstration where requested.

read point-by-point responses

  1. Referee: [§4, Eq. (15)–(18)] §4, Eq. (15)–(18): the mapping writes the partition function as a product of independent per-dyad transfer matrices; because the underlying configuration-model constraints are global (sum of link indicators must exactly equal the prescribed degree sequence), an explicit demonstration is required that the Lagrange multipliers for the degree constraints are recovered exactly by the per-dyad solution rather than only in the sparse/large-N limit.

    Authors: The mapping is exact. The maximum-entropy Hamiltonian is strictly additive over dyads, so the partition function factors exactly into a product of independent per-dyad transfer matrices for any finite N; the Lagrange multipliers for the (global) degree and time-persisting-degree constraints are recovered self-consistently by solving the per-dyad equations that enforce the observed expectations. This holds without invoking the sparse or large-N limit. We will add a short appendix containing the explicit algebraic demonstration that the fixed-point solution of the per-dyad multipliers satisfies the global constraints exactly. revision: yes

  2. Referee: [§5.1, likelihood (22)] §5.1, likelihood (22): the model-selection procedure for identifying the parsimonious memory level and structural breaks inherits any bias present in the transition probabilities; if the Ising mapping is only approximate under global constraints, the reported likelihood ratios and break locations on the real-world example are correspondingly approximate.

    Authors: Because the mapping derived in §4 is exact (as shown by the demonstration we will add), the closed-form transition probabilities are unbiased and the likelihood (22) is the exact likelihood of the ensemble. Consequently the likelihood-ratio test for memory level and the change-point detection are also exact; the numerical results on the empirical network require no qualification. No revision to §5.1 is needed beyond the clarification supplied for the mapping. revision: no

Circularity Check

0 steps flagged

No circularity: analytic mapping and likelihood-based selection are independent of fitted outputs

full rationale

The derivation begins from a maximum-entropy ensemble conditioned on empirical degrees and time-persisting degrees, then maps this ensemble exactly onto a heterogeneous 1D Ising model whose transfer-matrix solution yields closed-form expressions for the hidden variables. This mapping is a standard re-expression of the Lagrange multipliers and does not redefine any target quantity in terms of itself. Model selection proceeds by direct likelihood maximization over global/node/dyadic memory levels, which is an external statistical criterion rather than a tautology. No self-citation is load-bearing for the central analytic result, and no fitted parameter is relabeled as a prediction. The framework therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the maximum-entropy principle applied to both static degrees and time-persisting degrees; the Ising mapping is treated as an exact equivalence rather than an approximation.

axioms (1)
  • domain assumption The observed network trajectory is the typical realization of a maximum-entropy ensemble constrained by the empirical degree sequence and time-persisting degree sequence.
    Stated in the final sentence of the abstract as the resulting picture of the graph.

pith-pipeline@v0.9.0 · 5731 in / 1235 out tokens · 19384 ms · 2026-05-24T05:59:44.568624+00:00 · methodology

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Maximum entropy temporal networks

    cs.SI 2025-09 unverdicted novelty 6.0

    A maximum-entropy derivation connects non-homogeneous Poisson process intensities for temporal networks to path entropy optimization, producing modular generative models with closed-form log-likelihoods and expectations.

  2. Linking Through Time: Memory-Enhanced Community Discovery in Temporal Networks

    physics.soc-ph 2024-02 unverdicted novelty 6.0

    A memory-enhanced modularity lowers the detectability threshold for communities in Markovian temporal networks.

Reference graph

Works this paper leans on

60 extracted references · 60 canonical work pages · cited by 2 Pith papers

  1. [1]

    V., Zhang, Y., & Spyropoulos, T

    Vasilakos, A. V., Zhang, Y., & Spyropoulos, T. Delay tol- erant networks: Protocols and applications (CRC press, 2016)

    work page 2016

  2. [2]

    Daknet: Re- thinking connectivity in developing nations, Computer 37, 78 (2004)

    Pentland, A., Fletcher, R., & Hasson, A. Daknet: Re- thinking connectivity in developing nations, Computer 37, 78 (2004)

    work page 2004

  3. [3]

    Amaral, L. A. N., Barrat, A., Caldarelli, G., Barab´ asi, A.-L., De Los Rios, P., Erzan, A., Kahng, B., Mantegna, R. N., Mendes, J. F. F., Pastor-Satorras, R., & Others, Virtual round table on ten leading questions for network research, The European Physical Journal B-Condensed Matter 38, 143 (2004)

    work page 2004

  4. [4]

    Dynamical processes on complex networks (Cambridge University Press, 2008)

    Barrat, A., Barth´ elemy, M., & Vespignani, A. Dynamical processes on complex networks (Cambridge University Press, 2008)

    work page 2008

  5. [5]

    Scale-Free Networks: Complex Webs in Nature and Technology, OUP Catalogue (2007)

    Caldarelli, G. Scale-Free Networks: Complex Webs in Nature and Technology, OUP Catalogue (2007). 15

    work page 2007

  6. [6]

    L., Jeong, H., Neda, Z., Ravasz, E., Schu- bert, A., & Vicsek, T

    Barab´ asi, A. L., Jeong, H., Neda, Z., Ravasz, E., Schu- bert, A., & Vicsek, T. Evolution of the social network of scientific collaborations, Physica A, vol. 311, 590 (2002)

    work page 2002

  7. [7]

    The structure of scientific collaboration net- works., Proceedings of the National Academy of Sciences of the United States of America 98, 404 (2001)

    Newman, M. The structure of scientific collaboration net- works., Proceedings of the National Academy of Sciences of the United States of America 98, 404 (2001)

    work page 2001

  8. [8]

    Coauthorship networks and patterns of scientific collaboration., Proceedings of the National Academy of Sciences of the United States of America 101 Suppl, 5200 (2004)

    Newman, M. Coauthorship networks and patterns of scientific collaboration., Proceedings of the National Academy of Sciences of the United States of America 101 Suppl, 5200 (2004)

    work page 2004

  9. [9]

    Ghoshal, G., Zlati´ c, V., Caldarelli, G., & Newman, M. E. J. Random hypergraphs and their applications, Physical Review E 79, 66118 (2009)

    work page 2009

  10. [10]

    Computing Short- est, Fastest and Foremost Journeys in Dynamic Net- works, International Journal of Foundations of Computer Science 14, 267 (2003)

    Bui-Xuan, B., Ferreira, A., & Jarry, A. Computing Short- est, Fastest and Foremost Journeys in Dynamic Net- works, International Journal of Foundations of Computer Science 14, 267 (2003)

    work page 2003

  11. [11]

    Network reachability of real-world contact se- quences, Physical Review E 71, 46119 (2005)

    Holme, P. Network reachability of real-world contact se- quences, Physical Review E 71, 46119 (2005)

    work page 2005

  12. [12]

    The structure of information pathways in a social communication net- work, in Proc

    Kossinets, G., Kleinberg, J., & Watts, D. The structure of information pathways in a social communication net- work, in Proc. of the 14th ACM SIGKDD Intl. Conf. on Knowledge Discovery and Data Mining (KDD 2008) (2008) pp. 435–443

    work page 2008

  13. [13]

    Temporal graphs, Physica A: Statistical Mechanics and its Applications 388, 1007 (2009)

    Kostakos, V. Temporal graphs, Physica A: Statistical Mechanics and its Applications 388, 1007 (2009)

    work page 2009

  14. [14]

    Berman, K. A. Vulnerability of scheduled networks and a generalization of Menger’s Theorem, Networks 28, 125 (1996)

    work page 1996

  15. [15]

    Connectivity and inference problems for temporal networks, in Proceedings of the thirty-second annual ACM symposium on Theory of computing (ACM, 2000) p

    Kempe, D., Kleinberg, J., & Kumar, A. Connectivity and inference problems for temporal networks, in Proceedings of the thirty-second annual ACM symposium on Theory of computing (ACM, 2000) p. 513

    work page 2000

  16. [16]

    The diameter of opportunistic mobile networks, Proceed- ings of the 2007 ACM CoNEXT conference on - CoNEXT ’07, 1 (2007)

    Chaintreau, A., Mtibaa, A., Massoulie, L., & Diot, C. The diameter of opportunistic mobile networks, Proceed- ings of the 2007 ACM CoNEXT conference on - CoNEXT ’07, 1 (2007)

    work page 2007

  17. [17]

    Small-world behavior in time-varying graphs, Physical Review E 81, 055101 (2010)

    Tang, J., Scellato, S., Musolesi, M., Mascolo, C., & La- tora, V. Small-world behavior in time-varying graphs, Physical Review E 81, 055101 (2010)

    work page 2010

  18. [18]

    Temporal networks, Physics Reports 519, 97 (2012), temporal Networks

    Holme, P., & Saram¨ aki, J. Temporal networks, Physics Reports 519, 97 (2012), temporal Networks

    work page 2012

  19. [19]

    Complex networks: Patterns of complexity, Nature Physics 6, 480 (2010)

    Pastor-Satorras, R., & Vespignani, A. Complex networks: Patterns of complexity, Nature Physics 6, 480 (2010)

    work page 2010

  20. [20]

    Quantifying social group evolution, Nature 446, 664 (2007)

    Palla, G., Barabasi, A., & Vicsek, T. Quantifying social group evolution, Nature 446, 664 (2007)

    work page 2007

  21. [21]

    Graph evo- lution, ACM Transactions on Knowledge Discovery from Data 1, 2 (2007)

    Leskovec, J., Kleinberg, J., & Faloutsos, C. Graph evo- lution, ACM Transactions on Knowledge Discovery from Data 1, 2 (2007)

    work page 2007

  22. [22]

    Exploration of periodically varying graphs (2009)

    Flocchini, P., Mans, B., & Santoro, N. Exploration of periodically varying graphs (2009)

    work page 2009

  23. [23]

    A Guide to Temporal Net- works (2016)

    Masuda, N., & Lambiotte, R. A Guide to Temporal Net- works (2016)

    work page 2016

  24. [24]

    Grindrod, P., & Higham, D. J. A dynamical systems view of network centrality, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 470, 20130835 (2014), https://royalsocietypublishing. org/doi/pdf/10.1098/rspa.2013.0835

    work page doi:10.1098/rspa.2013.0835 2014

  25. [25]

    Persistence and periodicity in a dynamic proximity network, arXiv 12117343 (2012)

    Clauset, A., & Eagle, N. Persistence and periodicity in a dynamic proximity network, arXiv 12117343 (2012)

    work page 2012

  26. [26]

    Temporal ex- ponential random graph models with btergm: Estimation and bootstrap confidence intervals, Journal of Statistical Software 83 (2018)

    Leifeld, P., Cranmer, S., & Desmarais, B. Temporal ex- ponential random graph models with btergm: Estimation and bootstrap confidence intervals, Journal of Statistical Software 83 (2018)

    work page 2018

  27. [27]

    Logit models and logis- tic regressions for social networks: I

    Wasserman, S., & Pattison, P. Logit models and logis- tic regressions for social networks: I. an introduction to markov graphs and p*, Psychometrika 61, 401 (1996)

    work page 1996

  28. [28]

    W., & Leinhardt, S

    Holland, P. W., & Leinhardt, S. A dynamic model for social networks, The Journal of Mathematical Sociology 5, 5 (1977), https://doi.org/10.1080/0022250X.1977. 9989862

    work page doi:10.1080/0022250x.1977 1977

  29. [29]

    Snijders, T. A. B. The statistical evaluation of so- cial network dynamics, Sociological Methodology 31, 361 (2001), https://onlinelibrary.wiley.com/doi/ pdf/10.1111/0081-1750.00099

    work page doi:10.1111/0081-1750.00099 2001

  30. [30]

    Hanneke, S., Fu, W., & Xing, E. P. Discrete temporal models of social networks, Electronic Journal of Statistics 4, 585 (2010)

    work page 2010

  31. [31]

    E., Macci, C., Monti, A., Pasquale, F., & Silvestri, R

    Clementi, A. E., Macci, C., Monti, A., Pasquale, F., & Silvestri, R. Flooding time in edge-Markovian dy- namic graphs, in Proceedings of the Twenty-Seventh ACM Symposium on Principles of Distributed Comput- ing, PODC ’08 (Association for Computing Machinery, New York, NY, USA, 2008) p. 213–222

    work page 2008

  32. [32]

    Mazzarisi, P., Barucca, P., Lillo, F., & Tantari, D. A dynamic network model with persistent links and node- specific latent variables, with an application to the inter- bank market, European Journal of Operational Research 281, 50 (2020)

    work page 2020

  33. [33]

    Random graph models for dynamic networks, The European Physical Journal B 90 (2016)

    Zhang, X., Moore, C., & Newman, M. Random graph models for dynamic networks, The European Physical Journal B 90 (2016)

    work page 2016

  34. [34]

    Reality mining: Sensing com- plex social systems, Personal and Ubiquitous Computing 10, 255 (2006)

    Eagle, N., & Pentland, A. Reality mining: Sensing com- plex social systems, Personal and Ubiquitous Computing 10, 255 (2006)

    work page 2006

  35. [35]

    A new look at the statistical model identifica- tion, IEEE Transactions on Automatic Control 19, 716 (1974)

    Akaike, H. A new look at the statistical model identifica- tion, IEEE Transactions on Automatic Control 19, 716 (1974)

    work page 1974

  36. [36]

    Maximum-Entropy Net- works: Pattern Detection, Network Reconstruction and Graph Combinatorics (2017)

    Squartini, T., & Garlaschelli, D. Maximum-Entropy Net- works: Pattern Detection, Network Reconstruction and Graph Combinatorics (2017)

    work page 2017

  37. [37]

    Jaynes, E. T. Information theory and statistical mechan- ics, Phys. Rev. 106, 620 (1957)

    work page 1957

  38. [38]

    The statistical physics of real-world networks, Nature Reviews Physics 1 (2019)

    Cimini, G., Squartini, T., Saracco, F., Garlaschelli, D., Gabrielli, A., & Caldarelli, G. The statistical physics of real-world networks, Nature Reviews Physics 1 (2019)

    work page 2019

  39. [39]

    Park, J., & Newman, M. E. J. Statistical mechanics of networks, Phys. Rev. E 70, 066117 (2004)

    work page 2004

  40. [40]

    Fitness-dependent topo- logical properties of the world trade web, Physical review 16 letters 93, 188701 (2004)

    Garlaschelli, D., & Loffredo, M. Fitness-dependent topo- logical properties of the world trade web, Physical review 16 letters 93, 188701 (2004)

    work page 2004

  41. [41]

    Maximum likelihood: Extracting unbiased information from complex networks, Physical review

    Garlaschelli, D., & Loffredo, M. Maximum likelihood: Extracting unbiased information from complex networks, Physical review. E, Statistical, nonlinear, and soft matter physics 78, 015101 (2008)

    work page 2008

  42. [42]

    Dynamic social network analysis using latent space models, in Advances in Neural Infor- mation Processing Systems, Vol

    Sarkar, P., & Moore, A. Dynamic social network analysis using latent space models, in Advances in Neural Infor- mation Processing Systems, Vol. 18, edited by Y. Weiss, B. Sch¨ olkopf, and J. Platt (MIT Press, 2006)

    work page 2006

  43. [43]

    De- tection and localization of change points in temporal net- works with the aid of stochastic block models, Journal of Statistical Mechanics: Theory and Experiment 2016 (2016)

    De Ridder, S., Vandermarliere, B., & Ryckebusch, J. De- tection and localization of change points in temporal net- works with the aid of stochastic block models, Journal of Statistical Mechanics: Theory and Experiment 2016 (2016)

    work page 2016

  44. [44]

    Detecting change points in the large-scale structure of evolving networks, Proceedings of the AAAI Conference on Artificial Intelligence 29 (2014)

    Peel, L., & Clauset, A. Detecting change points in the large-scale structure of evolving networks, Proceedings of the AAAI Conference on Artificial Intelligence 29 (2014)

    work page 2014

  45. [45]

    Change point estimation in a dynamic stochastic block model (2018)

    Bhattacharjee, M., Banerjee, M., & Michailidis, G. Change point estimation in a dynamic stochastic block model (2018)

    work page 2018

  46. [46]

    Change-point detection in dynamic networks via graphon estimation (2019)

    Zhao, Z., Chen, L., & Lin, L. Change-point detection in dynamic networks via graphon estimation (2019)

    work page 2019

  47. [47]

    A cluster analysis method for grouping means in the analysis of variance, Biometrics 30 (1974)

    Scott, A., & Knott, M. A cluster analysis method for grouping means in the analysis of variance, Biometrics 30 (1974)

    work page 1974

  48. [48]

    Exactly Solved Models in Statistical Mechan- ics, Dover books on physics (Dover Publications, 2007)

    Baxter, R. Exactly Solved Models in Statistical Mechan- ics, Dover books on physics (Dover Publications, 2007). Appendix A: Supplementary

    work page 2007

  49. [49]

    Finding ⃗ αfor the three memory-less models Here we show how to derive the functional form of pij for the three specification of the memory-less model, and how estimate the value of ⃗ αsuch that the constraints are reproduced. The most general Hamiltonian between the three is: H1,nm(G) = 1 T X t X j<i θijaij(t) = X j<i θijaij (A1) We can note that in this...

    work page

  50. [50]

    To begin, we explicitly rewrite the Hamiltonian (31) in the most general case, considering constraints at the level of links ( H1,m)

    Mapping to the one-dimensional Ising model Here we provide a proof of the main results presented in the text. To begin, we explicitly rewrite the Hamiltonian (31) in the most general case, considering constraints at the level of links ( H1,m). H1(G) ≡ H 2,nm + H1,m = NX i=1 αi¯ki) + X j<i βijaij(t)aij(t + 1) (A6) = 1 T TX t=1   NX i=1 αiki(t) + X i<j βi...

    work page

  51. [51]

    (A14), by adapting it from the known solution of the one-dimensional Ising model [48]

    Partition function We now show the solution of the dyadic model defined by eq. (A14), by adapting it from the known solution of the one-dimensional Ising model [48]. The periodic- ity condition (30) ensures that all sites (time steps) are statistically equivalent, i.e. ⟨σij(1)⟩ = ⟨σij(2)⟩ = · · · = ⟨σij(T )⟩ (A19) The system is therefore translationally (...

    work page

  52. [52]

    19 We do this because we will use some of the passages in both the calculations to compute the autocorrelation function

    Expected values We start calculating the expected value ofσij(t) in two different ways. 19 We do this because we will use some of the passages in both the calculations to compute the autocorrelation function. In the first approach we use the free energy per spin, defined by the following equation: Fij = − 1 T lnZ(Bij, Jij) (A32) And differentiating Fij wi...

    work page

  53. [53]

    Maximum likelihood estimation The above expressions allow us to calculate all the rel- evant expected properties of the time series generated by the model, once the parameters B and J are set to the values B∗ and J∗ maximizing the likelihood P(G∗|B, J) of the observed graph trajectory G∗, where P(G|B, J) is given by eq. (A31). Equivalently, in terms of th...

    work page

  54. [54]

    We note that the case of no memory can be translated by imposing βi = 0 ∀i, that is Jij = 0 ∀i, j

    From Memory to Memory-Less Here we show how, starting form the full model and turning off the memory of the system we obtain the same expression of pij in the memory-less case. We note that the case of no memory can be translated by imposing βi = 0 ∀i, that is Jij = 0 ∀i, j. This implies that: λ+ ij = 2coshBij; (A70) λ− ij = 0 (A71) q sin(Bij)2 + e−4Jij =...

    work page

  55. [55]

    In this section, we describe how this matrix is defined

    Deriving The Stochastic Matrices The model is built such that each link is associated with a stochastic matrix. In this section, we describe how this matrix is defined. We start using the quantities defined in the model con- necting them in order to introduce the coupled probabil- ities of events for each link:

    work page

  56. [56]

    The probability that two nodes have a connection at time t and t + τ (with τ = 1): P [aij(t + τ) = 1 & aij(t) = 1] = ⟨aij(t)aij(t + 1)⟩ = qij (A76)

    work page

  57. [57]

    The probability that two nodes have a connection at time t and not t + τ: P [aij(t + τ) = 0 & aij(t) = 1] = ⟨aij(t)(1 − aij(t + 1))⟩ = pij − qij (A77)

    work page

  58. [58]

    The probability that two nodes don’t have a con- nection at time t having a connection a time t + τ: P [aij(t + τ) = 1 & aij(t) = 0] = ⟨(1 − aij(t))aij(t + 1)⟩ = pij − qij (A78)

    work page

  59. [59]

    This implies that each link tra- jectory is independent evolving: it is not influenced by the behavior of the others

    The probability that two nodes don’t have a con- nection at time t and t + τ: P [aij(t + τ) = 0 & aij(t) = 0] = ⟨(1 − aij(t))(1 − aij(t + 1))⟩ = 1 − 2pij + qij(A79) Defined these elements, we can write the transition matrix Pij, equal in the form for each couple of nodes: Pij = " qij pij pij −qij pij pij −qij 1−pij 1−2pij+qij 1−pij # (A80) Once we find th...

    work page

  60. [60]

    In [30] is proposed a class of models to deal with time-varying graphs, making a Markovian assumption on the trajectory of the network: P(Gt, Gt−1, Gt−2,

    Temporal Exponential Random Graphs as Maximum Entropy probability Here we show how the Temporal Exponential Random Graphs models (TERGMs), introduced by Hanneke in 2010 [30], can be seen as a maximum entropy probabil- ity distribution. In [30] is proposed a class of models to deal with time-varying graphs, making a Markovian assumption on the trajectory o...

    work page 2010