Superdiffusive processes and optimal transport

Short answer: because Lévy-style superdiffusion mixes lots of cheap local sampling with occasional long, ballistic runs, which slashes the time to “first contact” across a crowded, obstacle-filled cell. Neurons, with comically long cables, especially benefit. And yes, there’s a plausible line from this physics to intelligence, but hold the TED talk—evidence is suggestive, not gospel.

Why Lévy dynamics help cells reach far-away targets

  • Crowded cytoplasm punishes Brownian motion. In viscoelastic, jammed interiors, purely diffusive search gets trapped and ages. Adding motor-driven, long, straight “runs” produces superdiffusion (MSD ∝ t^α with α>1), modeled well by Lévy walks: heavy-tailed run lengths at finite velocity. That combination yields faster spread and shorter first-passage times than Brownian wandering.
  • Optimal when targets are sparse and hidden. Classic search theory shows heavy-tailed step distributions can minimize search time for rare targets. Biology loves rare targets: a specific synapse, a distal lysosome, one damaged mitochondrion in a dendrite. Caveat: not all environments favor the same exponent, so “Lévy wins always” is a myth.
  • What inside the cell looks Lévy-like?
    • mRNP granules in neurons exhibit aging Lévy-walk statistics, fitting data better than simple diffusion. That’s literally information-bearing cargo doing long runs plus local sampling along dendrites.
    • Endosomes/lysosomes and other cargos show intermittent, bidirectional runs from kinesin/dynein “tug-of-war,” producing heavy tails in run lengths and pause times. Result: rapid hops across microtubule highways punctuated by local probing.
    • Axonal signaling endosomes carry trophic and stress signals over centimeter scales; intermittent long runs are the only realistic way to hit deadlines in long axons. Transport defects correlate with neurodegeneration, underscoring why fast long-range connectivity matters.
  • Even the nucleus isn’t immune. Chromatin shows bursts of fast motion and long-range contacts; intermittent dynamics help distant regulatory elements find each other without scanning everything locally. Different system, same moral: mix short-range exploration with occasional leaps.

Why this might matter for intelligence

  • Timely delivery enables plasticity. Getting mRNAs, receptors, and organelles to the right spine at the right moment is a precondition for synaptic change. Superdiffusive search trims delays across ridiculous neuronal geometries. That’s logistics in service of learning.
  • Brain-scale rhyme with cognition. Human and animal external search and attention often show Lévy-like statistics too. It’s not proof of a grand unifying law, but it hints that “interleave local sampling with rare long jumps” is a generally good strategy for exploration, whether in space, memory, or networks.

Nuts-and-bolts intuition

  • Mechanistic origin. Power-law-like runs emerge from motor collectives switching states, heterogeneous track geometries, and stop-and-go kinetics. Microtubule polarity gives direction, actin handles local maneuvering, and cargo adapters set dwell/run time distributions. The math is captured by Lévy walks and related continuous-time random walks with aging/heterogeneity.
  • Why not always Lévy? If targets are dense or there’s strong prior about location, more Brownian or guided strategies can win; “best” depends on geometry, noise, and fuel budget. Biology appears to tune exponents and intermittency to context rather than pick a single magic law.

If you want, we can sketch a toy model: compare first-passage times for Brownian vs. Lévy-walk cargos on a branched dendrite with microtubule bottlenecks and actin-rich spines, then sweep the tail exponent to see where superdiffusion pays off. Otherwise, feel free to keep believing cells run on vibes.

===

Short version: Qian basically yes; England partially; Crooks gave the thermodynamic scaffolding but not the Lévy headline. You wanted prophecy receipts; here’s the receipt book.

Hong Qian

  • Directly engaged CTRW/non-Markovian kinetics. Qian and Wang (2006) built explicit continuous-time random walk models for single-molecule systems with microscopic reversibility. CTRWs are the canonical route to anomalous diffusion and heavy-tailed waiting times inside cells.
  • Mesoscopic nonequilibrium framework for enzymes/motors. His PRE/JPCB/Phys Rep work treats molecular motors and biochemical networks as stochastic processes with steady currents and entropy production, exactly the regime where intermittent runs and pauses emerge. It’s not branded “Lévy,” but it anticipates the statistics you care about.
  • Early single-particle tracking math. Qian co-authored a classic on analyzing random diffusion vs directed transport from trajectories, i.e., the nuts and bolts of identifying superdiffusion experimentally.

Jeremy England

  • Measured anomalous diffusion in vivo. England’s group introduced PIPE (photo-converted intensity profile expansion) to quantify cytoplasmic motion and explicitly report “diffusion anomality.” That’s empirical buy-in that intracellular transport isn’t purely Brownian.
  • Theory of dissipative adaptation. His “driven self-assembly” program argues driven matter self-organizes to dissipate energy efficiently; it doesn’t name Lévy walks, but it’s fully compatible with cells favoring intermittent, long-run transport modes when they help move stuff and dump heat efficiently.
  • Energetic lower bounds for maintaining nonequilibrium structure. With Horowitz he tied maintenance costs to target states, again furnishing the backdrop where certain transport patterns could be favored under energy budgets.

Gavin E. Crooks

  • Thermodynamic constraints, not Lévy advocacy. Crooks’ fluctuation theorem and thermodynamic-length/optimal-path results are the backbone for quantifying dissipation and control at the molecular scale. They don’t predict Lévy walks per se, but they justify and bound whatever intermittent strategy motors/cargo use in noisy, driven environments.
  • If you want a bridge: heavy-tailed transport that slashes first-passage times still has to obey fluctuation relations and dissipation geometry. That’s Crooks country, even if he didn’t plant a “Lévy lives here” flag.

Bottom line

  • Anticipation index: Qian ≫ England > Crooks for explicit anomalous-transport modeling; all three provide pieces of the modern story where intermittent, heavy-tailed motion emerges naturally in driven, crowded cells and can plausibly boost long-range coordination that brains depend on. If you’re hunting citations that scream “Lévy,” Qian’s CTRW paper is your smoking gun; England’s PIPE is your lab evidence; Crooks is the rulebook that keeps the whole circus legal.

Try not to make them responsible for “Lévy is central to intelligence.” That leap is yours, and honestly, kind of fun.

Short answer

Kind of. Molecules aren’t sitting around running linear-programming solvers, but the way their concentrations evolve in a cell often looks exactly like an optimal-transport (OT) problem where nature minimizes “work” subject to physics.

Wasserstein gradient flow: the math under the mess

The classic Jordan-Kinderlehrer-Otto result shows that the Fokker–Planck diffusion equation is the steepest-descent (gradient-flow) of free energy in the space of probability densities endowed with the Wasserstein-2 metric. In plain English: Brownian goo spreads out along the cheapest possible routes according to that metric.

Modern numerical work scales that idea so we can simulate realistic, highly-crowded cytoplasms at cell-sized resolutions.

Diffusion in a crowded cytoplasm = “cheap transport”

Inside a cell, every protein cloud or metabolite plume obeys a continuity equation of the form

\partial_t \rho = \nabla!\cdot!\bigl(D\nabla\rho - \rho,\mathbf v\bigr),

which can be rewritten as a Wasserstein gradient flow of a free-energy functional (entropy + potential). Even exotic cases like anomalous sub-diffusion can be framed variationally in OT terms.

So if you coarse-grain enough, passive molecular spreading literally follows an OT geodesic: minimum dissipation path given the viscosity, crowding, and any external potential.

Active highways: vesicles, motors, and network OT

The moment ATP-hungry motors show up, things get directional. Yet you can still treat the cytoskeleton as a graph where cargos hop along edges with costs (ATP, time, drag). Causal or graph-constrained OT variants formalize exactly that.

Recent imaging of primary-cilium vesicle traffic demonstrates that vesicle density profiles relax toward low-cost configurations consistent with OT predictions, even while individual vesicles are yanked around by kinesin/dynein.

Population-level OT: gene expression clouds on the move

Cell-state “locations” in transcriptomic space also migrate. Tools like WOT, moscot (Nature 2025), and SOCS (bioRxiv 2025) literally compute Wasserstein distances between scRNA-seq or spatial-omics snapshots to reconstruct how masses of mRNA content shift over time or through tissue. That’s OT being used not as a metaphor but as the core estimator of biological flow.

Caveats & unsolved annoyances

  • Energy injection: Active transport breaks detailed balance, so strict gradient-flow formalisms need tweaks (e.g., adding antisymmetric flux terms).
  • Crowding & obstacles: Real cytoplasm is a percolation nightmare; you often need hybrid OT–stochastic simulations.
  • Heterogeneous media: Membranes, phase-separated droplets, and cytoskeletal gels introduce discontinuous cost landscapes not handled by vanilla OT.
  • Data limitations: Live-cell imaging rarely gives you full 3-D concentration fields, so empiricists rely on indirect OT surrogates (e.g., matching histograms slice-by-slice).

Take-home

Passive spreading in cells aligns beautifully with optimal-transport theory; it even tells you why the entropy production rate is what it is. Active traffic can be cast as OT on a driven graph. OT won’t replace the biochemistry, but it gives you a ruler for how “expensive” any rearrangement should be—and cells, being miserly little entities, hug those cheap paths whenever physics lets them.

The brutally honest TL;DR

Optimal-transport (OT) geeks care about moving probability mass as cheaply as possible. The biologists who study hydrogen peroxide, melatonin, dopamine, or LSD mostly care about chemical kinetics, pumps, binding, and membranes. Those two worlds rarely overlap. So, aside from some niche “fluid-flow in the brain” papers, OT doesn’t currently explain how any of those specific molecules get around.

1 H₂O₂: small, angry, and handled by aquaporins

  • Main route – diffuses freely or slips through dedicated peroxiporin channels (Aquaporin-8, AQP11, etc.). Transport measurements are done with stopped-flow or proteoliposome assays, then modeled with classic reaction–diffusion equations, not Wasserstein geodesics.
  • Red-blood-cell work confirms simple membrane diffusion dominated by the lipid bilayer, again modeled with standard permeation coefficients.
  • Why OT is irrelevant – the molecule is consumed as fast as it moves, so “mass conservation” (an OT prerequisite) is flat-out violated.

2 Melatonin: lipophilic hormone on a joy-ride

  • Pharmacokineticists run physiologically based PK (PBPK) simulations to find the best oral/sublingual/IV routes. They rely on compartment models, hepatic clearance, and protein binding—not optimal transport theory.
  • Because melatonin dissolves nicely in membranes and isn’t strongly reactive, passive diffusion plus blood flow explains most of its journey. No one bothers wrapping that in Wasserstein distance math.

3 Dopamine: synaptic diva under tight surveillance

  • In the brain, dopamine is released into a ~20 nm cleft, diffuses a couple of microns, then is vacuumed up by DAT pumps. “Restricted-diffusion” models capture that by tweaking Fick’s laws with binding terms. Still no OT.
  • Gradient-flow formalisms could handle it, but biophysicists already get the parameters they need (rise/decay times, uptake rates) from existing models.

4 LSD: lipophilic hitchhiker with complicated kinetics

  • Its journey—gut, portal vein, liver, bloodstream, blood–brain barrier—is mapped by classical two- or three-compartment PK fits. That gives you Cmax and half-life just fine, so nobody rewrites the problem as an Earth-Mover’s metric.

5 Where OT

does

sneak into molecule-transport papers

  • Brain glymphatic clearance: unbalanced regularized OT (urOMT) has been applied to DCE-MRI datasets to estimate bulk CSF flow and solute wash-out in rodents. The solutes there are gadolinium tracers, not dopamine or melatonin, but at least it’s real biology using OT.
  • Spatial-omics: Wasserstein distances are routinely used to track how clouds of thousands of mRNA or protein species move during development or after perturbation, because OT is perfect for comparing two high-dimensional distributions in tissue.

6 Could OT ever matter for these four molecules?

  1. If someone injects fluorescent H₂O₂ nano-sensors and images the whole zebrafish larva, they could post-process that movie with OT tools—but the paper hasn’t been written.
  2. Melatonin: maybe a future glymphatic-flow study on sleep hormones will need OT to infer 3-D transport fields. We’re not there yet.
  3. Dopamine: OT on a graph (axons and dendrites) might someday optimize vesicle routing in big neuron-model simulations, but current models are still deterministic reaction–diffusion.
  4. LSD: its distribution is governed more by passive permeability and plasma binding kinetics than by crowd-sourced “minimal-work” routes; OT adds nothing here.

Bottom line

For now, OT is fantastic for bulk fluid flows and statistical redistribution problems. The moment a molecule is reactive, actively pumped, or non-conserved, OT’s tidy assumptions fall apart—so H₂O₂, melatonin, dopamine, and LSD stay in the pharmacokinetics lane while OT keeps polishing its abstract geometry toolkit.