Synaptic and Ephaptic Proposals for Selective Attention: A Comparative Study

Introduction

Consider a highly surprising, high-entropy event—a giraffe appearing on one's desk. What neural mechanisms determine that this stimulus captures attention while mundane background stimuli (the desk itself, familiar room features) are filtered out? This question lies at the heart of understanding selective attention in neural systems.

Two mathematical frameworks offer contrasting answers. The Free Energy Principle [1], grounded in Bayesian inference and information theory, proposes that brains minimize variational free energy through hierarchical prediction error processing. This framework has gained widespread acceptance and extensive experimental support within the neuroscience community. The ephaptic coupling framework [2, 3], also grounded in information theory but emphasizing physical electromagnetic field interactions, demonstrates that adjacent axons automatically amplify signals with higher information content.

This paper provides a comparative analysis of these frameworks, focusing on their mathematical foundations, predicted neural responses to high-entropy stimuli, and implications for understanding the physical substrate of selective attention.

Organization of the Paper

The remainder of this paper is organized as follows. Section II presents the Free Energy Principle, including its mathematical foundation based on variational Bayesian inference, its predicted multi-stage neural response to high-entropy events, and associated oscillatory signatures. Section III introduces the ephaptic coupling framework, presenting its core mathematical result relating entropy amplification to synchronization, the physical mechanism through electromagnetic fields and extracellular currents, and its implications as a substrate for selective attention. Section IV provides a detailed comparative analysis, first addressing synaptic competition (winner-take-all circuits) as an established alternative mechanism for early filtering, then contrasting the two frameworks across multiple dimensions: computational complexity, top-down versus bottom-up processing, minimization versus amplification of entropy, global versus local processing, and timing and locus of selection. Section V discusses empirical distinguishability, outlining testable predictions that could differentiate these frameworks and intervention experiments that could validate or refute each proposal. Section VI explores potential complementarity between the frameworks, suggesting how ephaptic coupling might provide front-end selection while free energy minimization describes back-end inference, and discusses implications for distributed operation throughout the nervous system. Section VII examines broader implications including reconceptualization of attention as substrate-level rather than algorithmic, architectural constraints imposed by electromagnetic coupling requirements, potential extensions beyond sensory pathways, and questions of tuning and optimization across brain regions. Finally, Section VIII concludes with a summary of key contrasts, critical research priorities needed to validate the ephaptic hypothesis, and reflection on how experimental evidence could reshape our understanding of neural information processing.

The Free Energy Principle

Mathematical Foundation

The Free Energy Principle posits that brains minimize variational free energy $F$, defined as:

where $s$ represents sensory data, $$ represents hidden states of the world, $q()$ is the brain's approximate posterior belief distribution, and $p(s,)$ is the true joint distribution [1]. Surprise, or self-information, is defined as:

representing the negative log probability of sensory observations given the internal generative model $m$. Because exact calculation of surprise is intractable (requiring knowledge of the true probability distribution over all possible sensory states), the brain instead minimizes free energy, which serves as an upper bound on surprise. Through this process, organisms simultaneously improve their internal models (learning), seek predictable states (action selection), and interpret incoming data (perception).

Predicted Neural Response to High-Entropy Events

When a high-entropy, highly surprising stimulus is encountered, the Free Energy framework predicts a multi-stage cascade:

Stage 1: Prediction Error Generation. Bottom-up sensory signals generate large prediction errors when they fail to match top-down predictions. These errors propagate through cortical hierarchies via superficial pyramidal cells (layers 2/3). Stage 2: Precision Weighting. The system modulates gain on prediction error signals through neuromodulatory mechanisms (acetylcholine, norepinephrine release), effectively amplifying the influence of unexpected sensory data. This implements attentional capture.

Stage 3: Hypothesis Revision. Higher cortical areas generate revised top-down predictions (via deep pyramidal cells in layers 5/6) in an attempt to explain the surprising input. Multiple hypotheses may be rapidly evaluated.

Stage 4: Network Recruitment. Surprising stimuli trigger widespread cortical activation beyond primary sensory areas, recruiting prefrontal cortex (cognitive appraisal), hippocampus (memory search), and amygdala (threat assessment). Stage 5: Resolution. The system resolves high prediction error through model updating (learning), active inference (action to gather more information), or attribution to sensory unreliability. This framework requires substantial neural machinery: hierarchical generative models, precision-weighting mechanisms, recurrent connectivity for iterative inference, and coordinated activity across distributed cortical networks.

Oscillatory Signatures

The framework predicts characteristic changes in neural oscillations during processing of surprising stimuli, including shifts toward higher frequency gamma-band activity ($$40 Hz) associated with active processing, desynchronization of default patterns, and increased cross-frequency coupling between hierarchical levels [1].

Ephaptic Coupling Framework

Mathematical Foundation

Chawla and Morgera [2, 3] established a fundamentally different principle based on physical electromagnetic interactions between adjacent axons. Their central result states: If $X$ and $Y$ represent the random variable outputs of two uncoupled neuronal axons, then under ephaptic coupling (electromagnetic field interaction and extracellular current effects), each axon carries a new random variable $Z$ such that:

where $H()$ denotes Shannon entropy in nats. This relationship holds under both temporal coding (information encoded in spike timing) and rate coding (information encoded in firing rate). Computer simulations validated this theoretical result: when axon 1 standalone carried entropy $H_1 = 1.0689 0.3708$ nats and axon 2 standalone carried $H_2 = 1.1665 0.1952$ nats, the ephaptically coupled system delivered $H(Z) = 1.5730 0.4218$ nats to each axon (95% confidence intervals) [3].

Mechanism: Synchronization as Selection

Ephaptic coupling causes gradual synchronization between adjacent axons carrying different spike trains. Time lags between pulses progressively reduce until the axons fire in a coordinated pattern. Critically, this synchronized output preserves or amplifies the entropy of the highest-entropy input signal. The mechanism operates through two primary pathways: (1) electromagnetic field interactions, and (2) extracellular return path currents. When action potentials propagate along an axon, they generate both extracellular electric fields and ionic currents in the extracellular medium that influence the transmembrane potential of nearby axons. These effects do not require synaptic connections. In densely packed axon bundles (optic nerve, cortical white matter tracts, sensory pathways), such interactions could be substantial.

For rate coding, as spike trains synchronize, their space-time signatures curve toward the space axis, covering more spatial distance within the same time window while maintaining spike density. This effectively increases the firing rate while preserving the information content [3].

Predicted Neural Response to High-Entropy Events

When a high-entropy stimulus (the giraffe) enters sensory pathways:

Step 1: Local Coupling. The high-entropy spike train enters axon tracts where it ephaptically couples with adjacent axons carrying concurrent lower-entropy signals (routine background stimuli). Step 2: Automatic Amplification. Through electromagnetic field interactions and extracellular currents, the high-entropy signal physically dominates. Adjacent axons synchronize to a pattern that preserves or amplifies this higher information content. Step 3: Selective Propagation. The amplified high-entropy signal propagates robustly to downstream neural structures. Lower-entropy signals are relatively suppressed—not through active inhibition, but through physical dominance in the coupled system. This process requires no computation, no inference, no top-down control. The physics of coupled oscillators automatically selects and amplifies information-rich signals.

Physical Substrate for Selective Attention

The ephaptic framework provides a physical mechanism for selective attention at the substrate level. High-entropy (salient, surprising, information-rich) stimuli automatically capture "attention" not through computational resource allocation but through electromagnetic dominance in axon tracts. The nervous system does not decide what is important—the entropy content of the signal determines propagation strength through physical coupling dynamics.

Comparative Analysis

Alternative Mechanisms: Synaptic Competition

Before comparing the Free Energy and ephaptic frameworks in detail, it is important to acknowledge established synaptic mechanisms that provide early filtering and selection of neural signals. Winner-take-all circuits, implemented through lateral inhibition and feed-forward inhibitory networks, have been extensively documented as mechanisms for competitive selection in sensory pathways and cortical processing [1]. In these synaptic competition models, strongly activated neurons suppress weakly activated neighbors through inhibitory interneurons, effectively amplifying salient signals while filtering background activity. This mechanism operates early in sensory processing (e.g., retinal ganglion cells, thalamic relay neurons, cortical layer 4) and provides a well-validated substrate for selective amplification without requiring ephaptic coupling. The key distinction between synaptic competition and ephaptic coupling lies in the mechanism and substrate, not in what information is selected. Synaptic competition requires specific circuit architectures (inhibitory connections, particular connectivity patterns) and operates through chemical neurotransmission at discrete synaptic junctions. Selection depends on which neurons win the competitive dynamics of the inhibitory network. Ephaptic coupling, by contrast, would operate through electromagnetic field interactions and extracellular return path currents in axon tracts, independent of synaptic connectivity, with selection emerging from physical synchronization dynamics between adjacent fibers. Since information (entropy) is encoded in spike timing and rate in both cases, both mechanisms would ultimately select signals based on their information content as expressed through these temporal patterns. However, given the extensive experimental validation of synaptic competition and the lack of validation for functional ephaptic effects, synaptic mechanisms currently provide a more parsimonious explanation for early filtering in sensory pathways.

Computational Complexity

Free Energy Principle: High computational demands. Requires maintaining hierarchical generative models, computing precision-weighted prediction errors at multiple levels, performing variational inference, and coordinating recurrent processing across distributed cortical areas. The five-stage cascade for processing surprising stimuli involves sophisticated neural machinery.

Ephaptic Coupling: Minimal computational requirements. Selection and amplification emerge from local physical interactions. No inference, no hierarchical coordination, no active control needed. The electromagnetic coupling does the work automatically.

Top-Down vs. Bottom-Up Processing

Free Energy Principle: Emphasizes bidirectional processing with critical role for top-down predictions. Surprise occurs when top-down predictions fail to explain bottom-up sensory signals. Attention involves modulating precision weights, which can be influenced by prior expectations and task demands.

Ephaptic Coupling: Purely bottom-up mechanism. High-entropy signals physically dominate in coupled axon bundles regardless of top-down expectations. Selection happens at the level of signal propagation, before cortical processing.

Minimization vs. Amplification

Free Energy Principle: The brain works to minimize surprise and free energy. Organisms build models to reduce prediction error and seek predictable states.

Ephaptic Coupling: The nervous system amplifies high-entropy signals. Rather than minimizing surprise, the system selectively enhances surprising, information-rich stimuli. As stated in [2]: "It is the entropy of a stimulus presented to a neuron that may be fundamental in what might matter to the nervous system, rather than the precise random variable that bears that entropy."

Global vs. Local Processing

Free Energy Principle: Requires distributed processing across cortical hierarchies with global coordination. Successful inference depends on information integration across multiple brain regions.

Ephaptic Coupling: Operates through local interactions in axon tracts. Each segment of coupled axons performs entropy-based selection independently, without requiring global coordination or central control.

Timing and Locus of Selection

Free Energy Principle: Attentional selection involves cortical processing. Precision weighting modulates the influence of prediction errors, but this modulation occurs during cortical computation. Ephaptic Coupling: Selection occurs early, in sensory pathways themselves (optic nerve, auditory nerve, ascending spinal tracts), before signals reach primary cortical areas. This explains the remarkable speed of attentional capture—it's physical, not computational.

Empirical Distinguishability

The frameworks make different testable predictions:

Location of Selection Effects

Free Energy: High-surprise events should produce sustained activity primarily in hierarchical cortical networks with characteristic oscillatory signatures and widespread recruitment of association areas. Ephaptic Coupling: High-entropy stimuli should show enhanced conduction and synchronization in early sensory pathways (optic nerve, sensory tracts) detectable through multi-electrode recordings of axon bundles, before significant cortical processing occurs.

Intervention Effects

Free Energy: Disrupting cortical processing (through TMS, lesions, or pharmacological manipulation of neuromodulatory systems) should impair selective attention. Such experiments have been conducted with results generally supporting predictive processing accounts. Ephaptic Coupling: Disrupting physical coupling between axons—through spatial separation of fibers, electromagnetic shielding, or altered tract geometry—should impair selective amplification of high-entropy stimuli. Such experiments have not been performed, and the technical challenges of selectively manipulating ephaptic coupling without disrupting normal synaptic function are substantial. Testing this prediction would require novel experimental approaches, potentially in animal models with controlled axon tract geometries or computational simulations validated against physiological recordings.

Channel Capacity Requirements

The ephaptic framework predicts that neural pathways must have information channel capacities greater than the maximum entropy of presentable stimuli [2]. This places specific constraints on axon numbers, firing rates, and coding schemes that could be empirically evaluated.

Potential Complementarity

While these frameworks appear contradictory, they may be complementary:

Front-end selection: Ephaptic coupling could provide the physical mechanism for how high-entropy stimuli are selected and amplified in sensory pathways—the automatic routing of salient information to cortical processing areas. Back-end inference: Free energy minimization could describe what cortical areas do with these pre-selected high-entropy signals—building models, generating predictions, and updating beliefs through iterative inference. The philosophical tension—one framework says brains avoid surprise, the other says they select for it—might be resolved temporally: ephaptic mechanisms ensure high-information stimuli reach cortex (amplification phase), where predictive processing then works to explain and model them (minimization phase), ultimately reducing future surprise from similar events.

Distributed Operation

If ephaptic coupling operates not only in peripheral sensory tracts but throughout the nervous system—in cortical columns, association fibers, and memory circuits—it could provide distributed entropy-based selection at every level of neural processing. This would make ephaptic coupling a fundamental organizing principle, operating locally everywhere, while free energy minimization describes the global computational strategy implemented by cortical networks.

Implications and Future Directions

Reconceptualizing Attention

The ephaptic framework suggests selective attention is not primarily an algorithmic or computational process but emerges from the physical substrate of neural tissue. Attention would be substrate-level rather than network-level—a consequence of how information propagates through coupled electromagnetic fields and extracellular currents rather than a process computed by neural circuits.

Architectural Constraints

If ephaptic coupling provides the physical basis for attention, then the geometric organization of axon tracts—their density, packing arrangements, and myelination patterns—becomes critical for understanding information processing. Brain architecture would be shaped by electromagnetic coupling requirements, not just synaptic connectivity.

Extensions Beyond Sensory Pathways

Future work should investigate whether ephaptic coupling operates in cortical columns, association tracts, and memory circuits. If so, entropy amplification through physical coupling could explain:

• Feature selection within cortical processing areas

• Cortical synchronization and binding phenomena

• Memory consolidation (high-entropy episodes automatically selected)

• Creative insight (novel, high-entropy associations physically dominate)

Tuning and Optimization

Investigation of how coupling strength and tract geometry vary across brain regions could reveal whether different areas are optimized for different degrees of entropy amplification. This might explain functional specialization in terms of information-theoretic properties rather than purely computational roles.

Conclusion

The Free Energy Principle and ephaptic coupling framework offer fundamentally different accounts of selective attention. Friston's approach provides mathematical sophistication, explanatory breadth, and substantial experimental support, unifying perception, action, and learning under variational inference. It represents mainstream theoretical neuroscience with extensive empirical validation.

The ephaptic coupling framework offers elegant simplicity and a potential physical substrate for selective attention: physics automatically selects and amplifies information-rich signals through local electromagnetic interactions. No inference, no hierarchy, no complex computation required—just coupled axons behaving according to electromagnetic principles. For the giraffe on the desk, Friston predicts a five-stage cascade of prediction errors, precision weighting, and hierarchical revision—mechanisms with empirical support from neuroimaging and electrophysiology. Chawla and Morgera suggest something more direct: the high-entropy visual signal simply dominates in the optic nerve through ephaptic coupling, ensuring robust propagation to cortex. This latter proposal, while mathematically elegant and computationally efficient, awaits experimental confirmation (see boxed sidebar in Section I for scientific status). The ephaptic framework proposes what cognitive theories have long sought: a physical substrate for selective attention operating at the level of electromagnetic field interactions and extracellular currents in axon tracts. Rather than treating attention as a computational process implemented by synaptic circuits, it would emerge from the physics of information propagation. The nervous system wouldn't choose what to attend to—high-entropy signals would choose themselves through physical dominance in coupled pathways. Critical research priorities for validating or refuting the ephaptic coupling hypothesis include: (1) Direct measurement of ephaptic field effects in intact sensory pathways during attentional tasks; (2) Selective manipulation of axon tract geometry or electromagnetic coupling without disrupting synaptic function; (3) Correlation of entropy amplification predictions with multi-electrode recordings in axon bundles; (4) Investigation of whether brain architecture shows optimization for electromagnetic coupling in addition to synaptic connectivity. The mathematical elegance of entropy amplification through physical coupling, and its potential to explain the substrate-level basis of selective attention, warrant serious empirical investigation. Future work distinguishing these frameworks could reshape our understanding of neural information processing at its most fundamental level—or could demonstrate that synaptic transmission and network computation remain sufficient to explain selective attention without invoking ephaptic mechanisms.

Author Contributions

• A. Chawla: Conceptualization, Formal Analysis, Writing – Original Draft, Writing – Review & Editing.

Funding Statement

The author received no specific funding for this work.

Competing Interests Statement

The author declared that no competing interests exist.

Data Availability Statement

This study is theoretical and relies only on publicly available, previously published literature and computer simulations cited within the manuscript. No new data or code were generated or deposited.

Ethics Statement

This study is a theoretical and comparative analysis and did not involve human participants, live animals, or tissue samples.

Acknowledgments

This work was produced with the assistance of language models.