{"id":"c1e9dce8-0660-4d16-a954-ffb4d2cfa24a","arxiv_id":"2502.00945","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A predictive information decomposition method quantifies emergent dynamics in physiological networks as the balance between synergistic and redundant information, and shows this balance shifts with postural stress.","lead":"This paper introduces a method for detecting when a physiological system behaves as more than the sum of its parts, by measuring how much the past of a network predicts its future. Researchers apply it to heart rate, blood pressure and respiration recordings and find that synergy between signals increases during postural stress.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central empirical claim not secured: the reported net synergy and its sympathetic modulation rest entirely on the MMI redundancy function and q=20 truncation, with no sensitivity analysis provided.","rationale":"We read the paper as making two claims: (i) PrID with VAR is a practical tool; (ii) the MMI-based ΔPID in cardiovascular/respiratory networks is a hallmark of integrated short-term control and rises with sympathetic activation. The mathematical machinery is standard and the simulation study is a useful proof of concept, so (i) is plausible. The load-bearing point is that (ii) is an estimate of a quantity that is not identifiable from data without choosing a PID redundancy measure and a finite q. The reader identified exactly this weakness, and no sensitivity analysis is supplied. An additional formal concern is that Eq. (3) writes I(Xn; X^i) = U_i + R with a single R for N=3, while the coarse-graining described in the 'solution' section defines R as information held by at least two sources; for N>2 these cannot both hold unless R is interpreted as the intersection over all sources, in which case it is not the total redundant information. This does not necessarily invalidate the numerical PID, but it weakens the derivation of Eq. (5) and the 'double counting' interpretation. The proposed q/redundancy-sensitivity test is the cleanest way to decide whether the physiological result is real or an artifact. Consequently, the paper should remain conditional until that test is passed; we do not see grounds to reject, since the framework is coherent and the concern is empirical robustness rather than a demonstrated inconsistency in the main computation.","tokens_in":14904,"tokens_out":13100,"duration_ms":139962,"concrete_test":"Re-run the full physiological analysis (61 subjects; {S,D,H}, {S,D,R}, {S,H,R}; SU and UP) for q = 5, 10, 20, 50 and with at least one alternative redundancy function (e.g., I_broja, I_ccs, or the Gaussian-specific PID of Barrett 2015). Report the sign and per-subject significance of ΔPID and the Wilcoxon SU-UP p-value for each configuration. If the sign of ΔPID or the SU-UP modulation reverses for any plausible alternative, the 'hallmark' claim is not robust; if the pattern is invariant across q and redundancy measures, the conditional reservation is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The physiological conclusion—that the synergy/redundancy balance in cardiovascular-respiratory networks is a hallmark and rises with tilt—is a numerical statement about ΔPID = S − R in Section 'Synergy/redundancy balance'. That number is not a directly measured quantity; it is the output of a PID solved with the minimum-MI redundancy function (Eq. 8) and of restricted VAR models truncated at order q=20 (Eq. 11). Both choices are underdetermined. MMI is one of many admissible redundancy measures; for N=3 the Möbius inversion and first-order coarse-graining propagate this choice into S and R, and different measures (e.g., I_broja, I_ccs, or other Gaussian PIDs) can assign the same joint MI to different synergy/redundancy balances. MMI is known to treat 'equal predictive power' as 'same information', so it can over-weight redundancy; the sign of ΔPID is therefore not a property of the data alone. The q=20 truncation is also consequential: the restricted models in Eq. (11) are theoretically infinite-order, and truncation biases every restricted MI in Eq. (12). Because R is a minimum of these MIs and S is obtained by Möbius inversion, the bias does not cancel; it can push ΔPID in either direction. The paper provides no sensitivity analysis over q or over the redundancy function, so the central 'hallmark' claim is not yet distinguished from an artifact of these two settings.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces Predictive Information Decomposition (PrID), a framework to quantify emergent dynamics in multivariate time series by decomposing predictive information into unique, redundant, and synergistic contributions. The authors formulate the decomposition for Gaussian processes using vector autoregressive (VAR) models, apply it to simulated networks and to cardiovascular/respiratory variability series from 61 healthy subjects at rest and during head-up tilt, and report that the synergy/redundancy balance computed via partial information decomposition (PID) increases with sympathetic activation. The manuscript argues that this balance is a hallmark of integrated short-term autonomic control and a potential biomarker.","tokens_in":15186,"tokens_out":2672,"duration_ms":27552,"significance":"If the central claims hold, the paper provides a practical, interpretable tool for quantifying collective dynamics in physiological networks, with a plausible application to autonomic regulation assessment. The theoretical machinery is standard: Eqs. (1)-(5) and the VAR-based mutual information formulas (10)-(12) are internally consistent, the surrogate data procedure is appropriate for the significance tests, and the use of real data from an established protocol strengthens the empirical relevance. The main weakness is that the key numerical results depend on two unvalidated choices—the minimum-MI redundancy function in Eq. (8) and the restricted-model order q=20 in Eq. (11)—and no sensitivity analysis is provided. Because the sign of the reported balance is a statement about the data only after these choices are made, the empirical 'hallmark' claim is not yet fully secured, though it is plausible and testable.","major_comments":[{"comment":"The central empirical finding—net synergy and its increase with tilt—rests entirely on the MMI redundancy function in Eq. (8) and on the first-order coarse-graining adopted from Ref. [12]. The paper does not assess whether the sign of ΔPID in Eq. (4) is robust to alternative redundancy measures (e.g., I_broja, I_ccs, or other Gaussian PIDs). Because MMI is known to treat equal predictive power as shared information, the reported balance could shift systematically under a different admissible redundancy function. A sensitivity analysis over redundancy functions, at least for the three triplets in Fig. 5, is needed to support the 'hallmark' claim.","section":"Partial information decomposition: solution; Synergy/redundancy balance"},{"comment":"The restricted models in Eq. (11) are theoretically infinite-order, and the choice q=20 is stated as 'typically sufficient' without validation. Truncation bias in the restricted MIs computed via Eq. (12) does not cancel when the minimum is taken in Eq. (8) and when Möbius inversion is used to obtain S and R; it can push ΔPID in either direction. The paper should report a sensitivity analysis over q (e.g., q=10, 30, 50) and a check of convergence of the restricted MI values for the analyzed triplets, to rule out the possibility that the results in Fig. 5 are artifacts of truncation.","section":"Practical computation"},{"comment":"The discrepancy between the WMS and PID results in Fig. 5 (WMS largely negative, PID showing net synergy for {S,H,R} and for {S,D,H} during UP) is interpreted as evidence that PID is necessary for assessing emergence. However, this discrepancy could be driven by the specific coarse-graining aggregation rules and the MMI assumption. To make the claim that the PID reveals 'previously unreported modes of interaction' convincing, the authors should validate the PID synergy using a complementary redundancy measure or demonstrate in simulations that the sign of ΔPID is stable under alternative decompositions.","section":"Application to Physiological Networks; Synergy/redundancy balance"}],"minor_comments":[{"comment":"There is a typo in the description of the analyzed variables: 'HP, SAP, SAP and RESP' should read 'HP, SAP, DAP and RESP'.","section":"Application to Physiological Networks, Experimental Protocol"},{"comment":"The paper defines emergence as the prevalence of synergy over redundancy, following Rosas et al. [12], and then interprets the results in those terms. This is legitimate, but the authors should explicitly remind the reader that the 'emergent behavior' conclusion is contingent on this definition, and that other definitions of emergence may lead to different classifications.","section":"General"},{"comment":"The simulated example in Fig. 2 is useful but only demonstrates that the measures behave in a way consistent with the authors' definition of emergence. It would be helpful to state more clearly that this is a consistency check, not an independent validation of the definition.","section":"Simulations, Theoretical Example"}],"recommendation":"major_revision","confidential_remarks":"The reader's report and the stress-test note identify the correct key weakness: the lack of sensitivity analysis for the PID redundancy function and the restricted-model order q. Both concerns are legitimate and directly affect the main empirical conclusion. I agree that the paper should not be accepted in its current form, but the concern is fixable within the manuscript's scope by adding sensitivity analyses. The manuscript is within the journal's scope and makes a useful contribution if robustness is demonstrated. I would encourage the editor to request a revision that includes these analyses."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know this paper because it takes a known information-theoretic machinery (predictive information, PID, MMI redundancy, Gaussian VAR) and produces a genuinely new empirical observation: in three physiological networks, the PID-based synergy/redundancy balance ΔPID = S − R is network-specific and increases from supine to upright tilt. That specific result is not in the cited literature, and it's a plausible step toward a practical biomarker for autonomic control.\n\nWhat the paper does well: the math in Eqs. (1)-(5) and (10)-(12) is standard and internally consistent. The simulation study correctly shows that the WMS decomposition double-counts redundancy when N=3 (Eq. 5), which justifies the PID route. The surrogate data testing is reasonable, and the physiological interpretations—respiration as a common driver, closed-loop baroreflex interactions during tilt—are thoughtful and grounded in prior work.\n\nThe soft spots are real. The central claim that the balance is a 'hallmark' of integrated control is a statement about ΔPID, which is not a directly measured quantity. It is the output of two choices: the MMI redundancy function (Eq. 8) and the truncated restricted-model order q=20 (Eq. 11). MMI is one of several admissible PID redundancy measures; other Gaussian PIDs (I_broja, I_ccs, etc.) can allocate the same joint MI differently between synergy and redundancy. So the sign of ΔPID could change with the PID choice. The q=20 truncation is also consequential: restricted models are theoretically infinite-order, and truncation biases every restricted MI in Eq. (12). Since R is a minimum over these MIs and S comes out of a Möbius inversion, the bias does not cancel; it can push ΔPID in either direction. The paper gives no sensitivity analysis over q or the redundancy function. That is the main weakness.\n\nThere are also two smaller issues: no code or data are shipped, and multiple comparisons across three networks and two conditions are not corrected. Those are minor compared to the sensitivity gap.\n\nDespite these concerns, the paper is not fatally flawed. The method is plausible, the empirical work is real, and the finding might hold up. It deserves a serious referee. I would send it to review and ask the authors to (1) run sensitivity analyses varying q and using at least one alternative Gaussian PID, (2) temper the 'hallmark' language, and (3) make code/data available. If the results are robust to those checks, it becomes a useful contribution.\n\nFor a reading group: maybe; it's a good example of PID applied to physiology, but the lack of sensitivity analysis will make the discussion more about method choice than about the physiological finding.","headline":"A novel, well-executed application of PID-based predictive information to physiological networks, but the headline synergy/redundancy 'hallmark' depends on two under-validated modeling choices and needs sensitivity analysis before it can be believed.","tokens_in":15700,"tokens_out":2965,"would_cite":false,"duration_ms":27692,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Cardiovascular networks become more synergistic when sympathetic tone rises.","keywords":["predictive information","partial information decomposition","causal emergence","synergy and redundancy","vector autoregressive models","cardiovascular networks","autonomic regulation","head-up tilt"],"falsifier":"Recompute $\\Delta_{\\text{PID}}$ for the same supine and tilt data using a different redundancy function, such as the pointwise common change in surprise or the Ibroja measure, and also with restricted-model orders far above $q=20$ (e.g., $q=100$). If the {S,D,H} network no longer turns from net redundancy to net synergy under tilt, or if the tilt modulation becomes non-significant, the claim that sympathetic activation raises causal emergence is an artifact of the MMI choice and truncation.","tokens_in":14711,"feed_emoji":"🫀","tokens_out":5611,"duration_ms":51946,"temperature":0.7,"pith_summary":"This paper introduces a way to detect collective, emergent behavior in networks of physiological signals by decomposing predictive information — how much the present state of the whole network can be predicted from its own past — into unique, redundant, and synergistic contributions. It argues that when synergistic information outweighs redundant information, the network is behaving emergently. Applied to heartbeat, blood-pressure, and respiration series measured at rest and during head-up tilt, the method finds statistically significant net synergy whose balance shifts with sympathetic nervous system activation. The authors conclude that the synergy/redundancy balance is a hallmark of integrated short-term autonomic control and a candidate biomarker.","feed_headline":"Postural stress shifts cardiovascular networks toward synergy","feed_subtitle":"Decomposing predictive information shows net synergy outgrows redundancy, marking integrated autonomic control.","key_machinery":"The load-bearing object is the partial information decomposition of predictive information: $I(X_n; X_{<n}) = \\sum_i U(X_n; X^i_{<n}) + R(X_n; X_{<n}) + S(X_n; X_{<n})$, with the synergy/redundancy balance $\\Delta_{\\text{PID}} = S - R$. The decomposition is solved by building a redundancy lattice over all combinations of source variables and choosing the minimum mutual information (MMI) redundancy function $I_\\cap(X_n^\\alpha) = \\min_j I(X_n; X^{\\alpha_j}_{<n})$, then coarse-graining the atoms into unique, redundant, and synergistic terms. Computationally, all mutual information terms follow from one vector autoregressive fit: the full model gives the predictive information, and restricted models obtained by pruning its covariance structure supply the required MI terms, with restricted-model order set to $q=20$.","core_discovery":"The central claim is that causal emergence in physiological networks can be assessed by predictive information decomposition (PrID): the predictive information $I(X_n; X_{<n})$ is split into unique, redundant, and synergistic components via partial information decomposition, and the balance $\\Delta_{\\text{PID}} = S - R$ serves as a 'strong' measure of emergence. Using vector autoregressive models to compute the required mutual information terms, the paper shows in simulations that net synergy appears when multiple causal interactions and internal dynamics point to the same target, while common-drive or cascade configurations yield net redundancy. In 61 healthy subjects' cardiovascular and respiratory networks, the PID balance is network-specific: redundancy dominates for vascular-respiratory coupling, synergy dominates for cardiovascular-respiratory coupling, and the cardiovascular network shifts from redundant at rest to synergistic during head-up tilt. The paper asserts that this tilt-induced rise in net synergy reflects sympathetic activation and integrated short-term control, a relation not captured by the simpler whole-minus-sum measure.","pith_inferences":["A natural next test is whether the tilt-induced rise in net synergy is blunted in patients with autonomic neuropathy or heart failure; if so, the PID balance could serve as a graded readout of autonomic impairment.","The conclusions depend on the MMI redundancy function; re-running the decomposition with alternative PID measures (e.g., based on pointwise common change in surprise) would show whether the network-specific balances are a property of the data or of the chosen redundancy definition.","Because the computation reduces to a single VAR fit, the framework is cheap enough for real-time monitoring, raising the possibility of tracking emergence during graded stress tests or drug interventions.","The simulation results suggest a general design principle: configurations that concentrate causal influences onto one target generate synergy, so one could deliberately probe for synergy to discover hidden convergence in other multi-channel datasets."],"forward_implications":["Net synergy in a physiological network indicates emergent, integrated control; the method can therefore flag which organ systems act as a collective rather than as independent units.","Because the balance rises under head-up tilt in all three analyzed networks, the measure can serve as a biomarker of sympathetic activation in short-term cardiovascular regulation.","The PID-based balance avoids the multiple counting of redundancy that makes the whole-minus-sum measure always redundancy-dominated for $N \\ge 3$, revealing network-specific synergy and redundancy patterns.","Grouping the cardiovascular variables into different triplets yields distinct balances, meaning the framework can characterize which subsets of physiological signals are engaged in high-order interactions."],"supporting_citations":[{"why":"Introduces the causal-emergence criterion (synergy as emergence) and the coarse-graining rules used to aggregate PID atoms.","marker":"[12]"},{"why":"Defines partial information decomposition and the redundancy lattice that organizes the information atoms.","marker":"[14]"},{"why":"Provides the minimum mutual information redundancy function used to solve the PID for Gaussian systems.","marker":"[21]"},{"why":"Supplies the procedure for deriving restricted VAR models from the full model, used to compute all MI terms.","marker":"[22]"},{"why":"Describes the VAR-based implementation of information decomposition in cardiovascular networks.","marker":"[16]"},{"why":"Supplies the experimental protocol and dataset of heartbeat, pressure, and respiration series in supine and tilt.","marker":"[18]"},{"why":"Documents the redundancy-dominant pattern previously observed in cardiovascular interactions, the baseline the PID results are contrasted with.","marker":"[19]"},{"why":"Gives earlier synergy and redundancy measures via predictability and transfer entropy, the approach the new method extends.","marker":"[30]"}],"fun_headline_variants":["Tilt test shifts cardiovascular networks to synergy","Net synergy marks integrated autonomic control in networks","Predictive info decomposition reveals synergy in physiological networks","Synergy beats redundancy as emergence marker in networks","VAR decomposition quantifies emergent network dynamics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reported synergy and redundancy balances rest on a specific definition of redundant information (the minimum of the individual mutual informations) and on approximating each restricted model's history with 20 lags; if either choice is wrong for these signals, the sign and modulation of net synergy could change.","fun_headline_variants_meta":{"raw":{"variants":["Tilt test shifts cardiovascular networks to synergy","Net synergy marks integrated autonomic control in networks","Predictive info decomposition reveals synergy in physiological networks","Synergy beats redundancy as emergence marker in networks","VAR decomposition quantifies emergent network dynamics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00062,"raw_usage":{"total_tokens":2896,"prompt_tokens":987,"completion_tokens":1909,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":603,"completion_tokens_details":{"reasoning_tokens":1852}},"tokens_in":603,"tokens_out":1909,"duration_ms":13433,"temperature":1.0,"reasoning_tokens":1852,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-09T17:09:22.346260+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute $\\Delta_{\\text{PID}}$ for the same supine and tilt data using a different redundancy function, such as the pointwise common change in surprise or the Ibroja measure, and also with restricted-model orders far above $q=20$ (e.g., $q=100$). If the {S,D,H} network no longer turns from net redundancy to net synergy under tilt, or if the tilt modulation becomes non-significant, the claim that sympathetic activation raises causal emergence is an artifact of the MMI choice and truncation.","supporting_citations":[{"cited_title":"Mea- suring integrated information: Comparison of candidate measures in theory and simulation,","cited_arxiv_id":null,"evidence_quote":"Introduces the causal-emergence criterion (synergy as emergence) and the coarse-graining rules used to aggregate PID atoms."},{"cited_title":"Syn- ergy, redundancy, and multivariate information mea- sures: an experimentalist’s perspective,","cited_arxiv_id":null,"evidence_quote":"Defines partial information decomposition and the redundancy lattice that organizes the information atoms."},{"cited_title":"Local measures of information storage in complex distributed computation,","cited_arxiv_id":null,"evidence_quote":"Provides the minimum mutual information redundancy function used to solve the PID for Gaussian systems."},{"cited_title":"Exploration of synergistic and redundant information sharing in static and dynamical gaussian sys- tems,","cited_arxiv_id":null,"evidence_quote":"Supplies the procedure for deriving restricted VAR models from the full model, used to compute all MI terms."},{"cited_title":"Predictability, complexity, and learning,","cited_arxiv_id":null,"evidence_quote":"Describes the VAR-based implementation of information decomposition in cardiovascular networks."},{"cited_title":"Assessing high-order links in cardiovascular and respiratory net- works via static and dynamic information measures,","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental protocol and dataset of heartbeat, pressure, and respiration series in supine and tilt."},{"cited_title":"Basic cardiovas- cular variability signals: mutual directed interactions ex- plored in the information domain,","cited_arxiv_id":null,"evidence_quote":"Documents the redundancy-dominant pattern previously observed in cardiovascular interactions, the baseline the PID results are contrasted with."},{"cited_title":"Mechanisms of causal interaction between short-term rr interval and systolic ar- terial pressure oscillations during orthostatic challenge,","cited_arxiv_id":null,"evidence_quote":"Gives earlier synergy and redundancy measures via predictability and transfer entropy, the approach the new method extends."}],"review_version":1}