betaFeedback

WorksPhysics and Astronomy

Correlation-induced suppression of Granger causality and Gaussian transfer entropy

Abstract

In complex dynamical systems, Granger causality and transfer entropy are widely used to infer directional interactions, but they quantify a source's unique predictive contribution after conditioning and can therefore be substantially attenuated by correlation with the retained variables. Here, using our previously developed Linear Operator Causality Analysis (LOCA) framework, which measures causal response from the system propagator, we derive a relationship between predictive and operator-based causality. For a two-dimensional linear system, the fractional suppression of Granger causality equals the squared correlation between source and receiver. We generalize this result to high-dimensional systems using the source covariance remaining after conditioning and show that additional conditioning can only increase the suppression. The same covariance ordering yields a corresponding upper bound for Gaussian transfer entropy. A linearized Burgers advection chain exhibits 36.6% suppression at the convective travel-time peak, and exact nonlinear acoustic Burgers solutions show that the fitted-surrogate penalty is governed by correlation structure rather than nonlinearity and remains robust to sensor noise after operator debiasing. Applied to resting-state fMRI from the Human Connectome Project (80 brain regions, 8 functional networks, 10 subjects), the single-lag Granger causality measure falls 39.4% below the operator response when conditioning only on the receiver and 73.0% when conditioning on the receiver and all other retained variables. The framework provides a quantitative bridge between predictive and operator-based causal measures in high-dimensional dynamical systems.

Significance statement

Directional influence can mean different things: a variable may drive a system’s response yet contribute little unique predictive information once correlated variables are included. This distinction can cause predictive-causality measures to systematically understate operator-based response measures in high-dimensional systems. We show how to quantify this gap for linear dynamical models, and show that additional conditioning can only widen it. In resting-state brain imaging, conditional Granger causality falls on average 73% below the operator response, a gap determined by how strongly brain regions co-fluctuate. The result provides a common framework for Granger causality, Gaussian transfer entropy, and operator-based causality, clarifying when they agree, why they differ, and how those differences should be interpreted across neuroscience, physics, and engineering applications.

Complex dynamical systems often exhibit intricate interactions among variables, where distinguishing causal influence from mere correlation is critical for scientific understanding and design. Granger causality [1] and methods based on information theory [2, 3] are widely used to infer directional interactions from time-series data. However, when variables are correlated, their predictive-causality estimates can differ substantially from intervention- or response-based measures [4, 5]. This distinction has motivated operator-theoretic and interventional alternatives, notably the Dynamical Causal Effects (DCE) framework [6, 7, 8], which grounds causal inference in the system’s propagator rather than statistical association. Here, building on the LOCA framework for linearized fluid flows [9], itself a specific DCE, we derive the closed-form relationship between LOCA and Granger causality, an explicit penalty identity that quantifies the gap between the operator-based causal response and the source’s unique predictive contribution.

Operator theoretic measure of causality

For a continuous-time linear dynamical system x˙(t)=Acx(t)\dot{\mathbf{x}}(t) = \mathbf{A}_{c}\mathbf{x}(t), the causal influence of mode jj on mode ii at time lag τ\tau is given by the matrix exponential element [eAcτ]ij[\mathrm{e}^{\mathbf{A}_{c}\tau}]_{ij} [9]. See Appendix § 12 for details. Here, we develop the discrete-time formulation required for data-driven analysis.

In many practical applications, system data is collected at discrete time intervals, and the dynamics are subject to stochastic forcing. We model such a system with a discrete-time linear stochastic process:

xk+1=Axk+ϵk,(1)\mathbf{x}_{k+1} = \mathbf{A}\mathbf{x}_{k} + \boldsymbol{\epsilon}_{k}, \tag*{(1)}

where xk∈Rn\mathbf{x}_{k} \in\mathbb{R}^{n} is the state vector at time step kk, A∈Rn×n\mathbf{A} \in\mathbb{R}^{n \times n} is the state transition matrix (propagator), and ϵk∈Rn\boldsymbol{\epsilon}_{k} \in\mathbb{R}^{n} is zero-mean process noise with E[ϵkϵjT]=Qδkj\mathbb{E}[\boldsymbol{\epsilon}_{k}\boldsymbol{\epsilon}_{j}^{\mathsf{T}}] = \mathbf{Q}\delta_{kj}.

Since the noise is independent of the initial state x0\mathbf{x}_{0}, the causal influence of state variable jj on variable ii after kk steps is:

∂xi,k∂xj,0=[Ak]ij.(2)\frac{\partial x_{i,k}}{\partial x_{j,0}} = [\mathbf{A}^{k}]_{ij}. \tag*{(2)}

Thus, the matrix power Ak\mathbf{A}^{k} serves as the causality matrix. For the rest of the paper, we normalize this by the signal-to-noise ratio (SNR) to define the operator-theoretic causality measure:

Fj→iop=[A]ij2σxj2σϵi2,(3)F^{\mathrm{op}}_{j \to i} = \frac{[\mathbf{A}]_{ij}^{2}\sigma_{x_j}^{2}}{\sigma_{\epsilon_i}^{2}}, \tag*{(3)}

where σxj2\sigma_{x_j}^{2} is the variance of xjx_j and σϵi2\sigma_{\epsilon_i}^{2} is the variance of the noise component ϵi\epsilon_i. This definition evaluates the propagator at a single lag (k=1k = 1). The notation FopF^{\mathrm{op}} denotes a dimensionless causality magnitude, not the finite-sample regression FF-statistic used to test the statistical significance of Granger causality. FopF^{\mathrm{op}} contains no sample-size or degrees-of-freedom factor. FopF^{\mathrm{op}} asks: when the source fluctuates by a typical amount, how large is the receiver’s next-step response compared with its noise floor? We restrict to off-diagonal terms (i≠ji \ne j) throughout, consistent with the Granger causality convention.

For the general case where we wish to quantify the causal influence of a group of source variables xs\mathbf{x}_{s} on a group of receiver variables xr\mathbf{x}_{r}, ArsA_{rs} denotes the block of AA with rows in rr and columns in ss, so ArsδxsA_{rs}\delta\mathbf{x}_{s} is the one-step receiver response to a source perturbation. We average the squared causal response ∥Arsδxs∥2\lVert A_{rs}\delta\mathbf{x}_{s}\rVert^{2} over perturbations drawn from a specified source distribution δxs∼(0,Σs)\delta\mathbf{x}_{s} \sim(0,\mathbf{\Sigma}_{s}) and normalize by the noise energy (see Appendix § 13 for details):

Fs→rop=tr⁡(ArsΣsArsT)tr⁡(Σεr).(4)F^{op}_{s \to r} = \frac{\operatorname{tr}(A_{rs}\Sigma_s A_{rs}^{T})}{\operatorname{tr}(\Sigma_{\varepsilon r})}. \tag*{(4)}

This is a SNR for the causal pathway ArsA_{rs}: the expected squared causal response under the source’s variability, normalized by the noise floor. For a stationary stochastic system, Σs\Sigma_s is the stationary source covariance. For a transient data ensemble, it is the sample covariance of the pooled source perturbations, as in the acoustic example below.

Granger causality for a two-state system

We now analyze a two-dimensional discrete-time linear system whose causal properties have been studied in [10, 11, 12]:

\begin{pmatrix} x_{k+1}\\ y_{k+1} \end{pmatrix} = \begin{pmatrix} a & c\\ 0 & b \end{pmatrix} \begin{pmatrix} x_k\\ y_k \end{pmatrix} + \begin{pmatrix} \varepsilon_{x,k}\\ \varepsilon_{y,k} \end{pmatrix}. \tag*{(5)}\end{pmatrix}

We assume that the system is stable, with ∣a∣,∣b∣<1|a|, |b| < 1. The only nontrivial causal connection is the effect of yy on xx, governed by c≠0c \ne0. From Eq. (3),

Fy→xop=c2var⁡(y)σε,x2=c21−b2σε,y2σε,x2,(6)F^{op}_{y \to x} = \frac{c^2 \operatorname{var}(y)}{\sigma_{\varepsilon,x}^2} = \frac{c^2}{1-b^2}\frac{\sigma_{\varepsilon,y}^2}{\sigma_{\varepsilon,x}^2}, \tag*{(6)}

where the second equality follows from var⁡(y)=σε,y2/(1−b2)\operatorname{var}(y) = \sigma_{\varepsilon,y}^2/(1-b^2).

A closed-form expression for the corresponding Granger causality measure exists [11] and depends on aa, whereas Fy→xopF^{op}_{y \to x} does not (Appendix § 14.1). To see why, consider how this Granger causality measure is constructed. Granger causality compares an unrestricted model that includes all variables (here simply Eq. (5)) against a restricted model that drops the variable whose causal influence is being tested:

xk+1=a~xk+ε~x,k.(7)x_{k+1} = \tilde{a}x_k + \tilde{\varepsilon}_{x,k}. \tag*{(7)}

The normalized variance-increment Granger causality measure, FGCF^{GC}, is the relative increase in one-step prediction error:

Fy→xGC=σε~,x2−σε,x2σε,x2.(8)F^{GC}_{y \to x} = \frac{\sigma_{\tilde{\varepsilon},x}^2 - \sigma_{\varepsilon,x}^2}{\sigma_{\varepsilon,x}^2}. \tag*{(8)}

This normalized variance increment is distinct from the finite-sample regression FF-statistic used for significance testing, although both are based on the same reduction in prediction error. It is also a monotonic transformation of the Granger–Geweke log variance-ratio measure Fy→x=ln⁡(σ~ε,x2/σε,x2)=ln⁡(1+Fy→xGC)F_{y \to x} = \ln(\tilde{\sigma}_{\varepsilon,x}^2/\sigma_{\varepsilon,x}^2) = \ln(1 + F^{GC}_{y \to x}) [13]. The key point is that a~≠a\tilde{a} \ne a: the omitted variable yy is correlated with xx, so the restricted least-squares estimate absorbs part of the omitted effect into the self-coupling coefficient (Appendix § 14.2):

a~=a+ccov⁡(x,y)var⁡(x).(9)\tilde{a} = a + c\frac{\operatorname{cov}(x,y)}{\operatorname{var}(x)}. \tag*{(9)}

Dropping yy does not remove its influence from the restricted fit. The component of that influence that is predictable from xx is absorbed into the fitted self-coupling coefficient, and only the linearly unpredictable remainder is attributed to yy. The restricted error is therefore not simply εx,k\varepsilon_{x,k} but includes the part of yy’s influence uncorrelated with xx:

ε~x,k=εx,k+c(yk−cov⁡(x,y)var⁡(x)xk).(10)\tilde{\varepsilon}_{x,k} = \varepsilon_{x,k} + c\left(y_k - \frac{\operatorname{cov}(x,y)}{\operatorname{var}(x)}x_k\right). \tag*{(10)}

The term in parentheses is the part of yky_k that cannot be predicted linearly from xkx_k, consistent with the Frisch–Waugh–Lovell theorem [14, 15], which states that the incremental effect of a predictor is determined by the component remaining after projecting out the other retained predictors. Computing the variance of the restricted error above (Appendix § 14.3) and substituting into Eq. (8) yields:

Fy→xGC=c2σϵ,x2(var⁡(y)−cov⁡(x,y)2var⁡(x)).(11)F^{GC}_{y \to x} = \frac{c^2}{\sigma^2_{\epsilon,x}}\left(\operatorname{var}(y) - \frac{\operatorname{cov}(x,y)^2}{\operatorname{var}(x)}\right). \tag*{(11)}

Comparing with Eq. (6) and factoring out var⁡(y)\operatorname{var}(y), we arrive at the central result:

Fy→xGC=Fy→xop[1−ρ2(x,y)],(12)F^{GC}_{y \to x} = F^{op}_{y \to x}\left[1-\rho^2(x,y)\right], \tag*{(12)}

where ρ(x,y)\rho(x,y) is the Pearson correlation between xx and yy. We define the correlation penalty P≡ρ2\mathcal{P} \equiv\rho^2: the fraction of the operator response that the Granger statistic does not attribute to the source. The inverse relation Fy→xop=Fy→xGC/(1−ρ2)F^{op}_{y \to x} = F^{GC}_{y \to x}/(1-\rho^2) gives a direct correction: a Granger estimate plus the pairwise correlation recovers the suppressed LOCA metric, although the correction amplifies uncertainty in the estimated ρ\rho by 1/(1−ρ2)1/(1-\rho^2) and so degrades as ∣ρ∣→1\lvert\rho\rvert\to1. Granger causality vanishes in the limit ∣ρ∣→1\lvert\rho\rvert\to1, where the regression becomes singular and the source is no longer separately identifiable, and coincides with LOCA only at ρ=0\rho= 0; FopF^{op} is thus the upper bound on FGCF^{GC}. Standard Granger causality conditions on the entire past. For fully observed VAR(1) systems, the unrestricted one-lag predictor is already optimal, and restricted-model history can only shrink the restricted error, so FopF^{op} also bounds the full-history statistic (Appendix § 15.6).

Figure 1(a) illustrates this for (5) across a range of aa at fixed bb and cc: FopF^{op} is independent of aa while FGCF^{GC} varies through the aa-dependent correlation between xx and yy, and theory agrees closely with values from standard linear regression.

Figure 1. (a) Granger versus operator-based causality for the system of (5) with b=0.99b = 0.99, c=0.4c = 0.4, varying aa, and uncorrelated Gaussian noise, var⁡(ϵx)=var⁡(ϵy)=0.01\operatorname{var}(\epsilon_x) = \operatorname{var}(\epsilon_y) = 0.01. (b) Transfer entropy for the same system. Numerical estimates use 10710^7 timesteps (first 2×1042 \times10^4 discarded); TE uses shuffle-corrected quantile binning (Appendix § 16).

Generalizing to N-dimensional systems

The preceding analysis generalizes to NN-dimensional systems via a three-way partition. Let ss, rr, and oo denote disjoint index sets for source, receiver (target), and other (conditioning) variables, with s∪r∪o={1,…,N}s \cup r \cup o = \{1,\ldots,N\}. Roman entrywise indices i,ji,j, as in Eq. (3) and [9], denote single matrix elements. The set subscripts ss, rr, and oo denote blocks. Define the combined retained predictor zk=[xr,k;xo,k]\mathbf{z}_k = [\mathbf{x}_{r,k};\mathbf{x}_{o,k}] and write the unrestricted model:

References

  1. [1]CWJ Granger, Investigating Causal Relations by Econometric Models and Cross-spectral Methods. Econometrica 37, 424–438 (1969).
  2. [2]T Schreiber, Measuring Information Transfer. Phys. Rev. Lett. 85, 461–464 (2000).DOI
  3. [3]L Barnett, AB Barrett, AK Seth, Granger causality and transfer entropy are equivalent for Gaussian variables. Physical Review Letters 103, 238701 (2009).DOI
  4. [4]PA Stokes, PL Purdon, A study of problems encountered in Granger causality analysis from a neuroscience perspective. Proceedings of the National Academy of Sciences 114, E7063–E7072 (2017).DOI
  5. [5]L Barnett, AB Barrett, AK Seth, Misunderstandings regarding the application of Granger causality in neuroscience. Proceedings of the National Academy of Sciences 115, E6676–E6677 (2018).DOI
  6. [6]DA Smirnov, Quantification of causal couplings via dynamical effects: A unifying perspective. Physical Review E 90, 062921 (2014).pubmed.ncbi.nlm.nih.gov/25615178
  7. [7]DA Smirnov, II Mokhov, Relating Granger causality to long-term causal effects. Physical Review E 92, 042138 (2015).DOI
  8. [8]DA Smirnov, Generative formalism of causality quantifiers for processes. Physical Review E 105, 034209 (2022).DOI
  9. [9]A Srivastava, V Jiménez-Fernández, LN Cattafesta, STM Dawson, Operator theoretic causal-ity analysis of fluid flows using linearized dynamics. Theoretical and Computational Fluid Dynamics 40, 22 (2026).arxiv.org/abs/2506.08118
  10. [10]L Barnett, AK Seth, Behaviour of Granger causality under filtering: theoretical invariance and practical application. Journal of Neuroscience Methods 201, 404–419 (2011).DOI
  11. [11]L Barnett, AK Seth, The MVGC multivariate Granger causality toolbox: a new approach to Granger-causal inference. Journal of Neuroscience Methods 223, 50–68 (2014).DOI
  12. [12]L Barnett, AK Seth, Granger causality for state-space models. Physical Review E 91, 040101 (2015).DOI
  13. [13]J Geweke, Measurement of linear dependence and feedback between multiple time series. Journal of the American Statistical Association 77, 304–313 (1982).
  14. [14]R Frisch, FV Waugh, Partial time regressions as compared with individual trends. Econometrica 1, 387–401 (1933).DOI
  15. [15]MC Lovell, Seasonal adjustment of economic time series and multiple regression analysis. Journal of the American Statistical Association 58, 993–1010 (1963).
  16. [16]JF Geweke, Measures of conditional linear dependence and feedback between time series. Journal of the American Statistical Association 79, 907–915 (1984).
  17. [17]D McAllester, K Stratos, Formal limitations on the measurement of mutual information in International Conference on Artificial Intelligence and Statistics. (PMLR), pp. 875–884 (2020).
  18. [18]A Kraskov, H Stögbauer, P Grassberger, Estimating mutual information. Physical Review E 69, 066138 (2004).pubmed.ncbi.nlm.nih.gov/15244698
  19. [19]SPG Munoz, M Bounoua, G Franzese, P Michiardi, M Filippone, Tende: Transfer entropy neural diffusion estimation. arXiv preprint arXiv:2510.14096 (2025).
  20. [20]JM Burgers, A mathematical model illustrating the theory of turbulence. Advances in Applied Mechanics 1, 171–199 (1948).DOI
  21. [21]MF Hamilton, DT Blackstock, eds., Nonlinear Acoustics. (Academic Press, San Diego, CA), (1998).
  22. [22]JD Cole, On a quasi-linear parabolic equation occurring in aerodynamics. Quarterly of Applied Mathematics 9, 225–236 (1951).DOI
  23. [23]E Hopf, The partial differential equation u_t + uu_x = μu_xx. Communications on Pure and Applied Mathematics 3, 201–230 (1950).
  24. [24]PJ Schmid, Dynamic mode decomposition of numerical and experimental data. Journal of Fluid Mechanics 656, 5–28 (2010).
  25. [25]JH Tu, CW Rowley, DM Luchtenburg, SL Brunton, JN Kutz, On dynamic mode decomposition: Theory and applications. Journal of Computational Dynamics 1, 391–421 (2014).DOI
  26. [26]STM Dawson, MS Hemati, MO Williams, CW Rowley, Characterizing and correcting for the effect of sensor noise in the dynamic mode decomposition. Experiments in Fluids 57, 1–19 (2016).
  27. [27]AK Seth, AB Barrett, L Barnett, Granger Causality Analysis in Neuroscience and Neuroimaging. J. Neurosci. 35, 3293–3297 (2015).DOI
  28. [28]AK Seth, P Chorley, LC Barnett, Granger causality analysis of fMRI BOLD signals is invariant to hemodynamic convolution but not downsampling. NeuroImage 65, 540–555 (2013).
  29. [29]A Duggento, et al., Multivariate Granger causality unveils directed parietal to prefrontal cortex connectivity during task-free MRI. Sci. Rep. 8, 5571 (2018).DOI
  30. [30]DC Van Essen, et al., The WU-Minn Human Connectome Project: An overview. NeuroImage 80, 62–79 (2013).DOI
  31. [31]SM Smith, et al., Resting-state fMRI in the Human Connectome Project. NeuroImage 80, 144–168 (2013).DOI
  32. [32]FA Roffet, Deep learning and whole-brain networks for biomarker discovery: modeling the dynamics of brain fluctuations in resting-state and cognitive tasks (2024) https://doi.org/10.5281/zenodo.14508470. Deposited 17 December 2024.
  33. [33]RS Desikan, et al., An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 31, 968–980 (2006).pubmed.ncbi.nlm.nih.gov/16530430
  34. [34]BTT Yeo, et al., The organization of the human cerebral cortex estimated by intrinsic functional connectivity. J. Neurophysiol. 106, 1125–1165 (2011).DOI
  35. [35]LN Cattafesta, STM Dawson, A Srivastava, MATLAB code for “correlation-induced suppression of Granger causality and Gaussian transfer entropy” (2026) https://doi.org/10.5281/zenodo.23540187. Deposited 21 October 2026.
  36. [36]MS Hemati, CW Rowley, EA Deem, LN Cattafesta, De-biasing the dynamic mode decomposition for applied koopman spectral analysis of noisy datasets. Theoretical and Computational Fluid Dynamics 31, 349–368 (2017).DOI
  37. [37]H Lütkepohl, New Introduction to Multiple Time Series Analysis. (Springer, Berlin), (2005).
  38. [39]R Marschinski, H Kantz, Analysis of the information flow between financial time series. The European Physical Journal B 30, 275–281 (2002).DOI
  39. [40]GM Ljung, GEP Box, On a measure of lack of fit in time series models. Biometrika 65, 297–303 (1978).
  40. [41]G Krüger, GH Glover, Physiological noise in oxygenation-sensitive magnetic resonance imaging. Magnetic Resonance in Medicine 46, 631–637 (2001).
  41. [42]RT Beyer, Parameter of nonlinearity in fluids. Journal of the Acoustical Society of America 32, 719–721 (1960).
  42. [43]DT Blackstock, Connection between the Fay and Fubini solutions for plane sound waves of finite amplitude. Journal of the Acoustical Society of America 39, 1019–1026 (1966).DOI

Paper details

vlnc:W_6kzumxs7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 2

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents