betaFeedback

WorksPhysics and Astronomy

Fluidic hysterons and memory in flow networks

Abstract

Hysterons provide a minimal description of memory in driven matter: bistable elements with distinct switching thresholds whose interactions generate hysteresis, avalanches, and return point memory or its violation. Experimental realizations have so far been dominated by solid state mechanical systems, where bistability is usually encoded structurally through buckling, snap through, or geometric incompatibility. Here we realize hysteron physics through a hydrodynamic route. A single elastic fiber anchored in a microfluidic channel becomes bistable through nonlinear elastohydrodynamic feedback: viscous loading deforms the fiber, deformation reshapes hydraulic resistance, and flow redistribution modifies the loading. This feedback produces a fluidic hysteron whose onset is organized by a cusp catastrophe in geometric control parameters. A parallel bypass channel acts as a geometric load line that reshapes, and can even eliminate, bistability while simultaneously mediating long ranged hydraulic interactions between fibers. In arrays, varying a single geometric parameter drives a transition from a non interacting Preisach regime with return point memory to an interacting regime with avalanche like switching and return point memory violation. These results establish a passive hydrodynamic route to hysteron networks, in which memory emerges from flow structure feedback and global hydraulic constraints rather than solid state multistability or external control.

Introduction

Living flow networks, from slime molds and fungal mycelia to the microvasculature of the brain, record aspects of their interaction history in their architecture. Flows of nutrients, ions, metabolites, and signaling molecules both transmit information through these networks and remodel them: vessel diameters and mechanical properties adapt in response to shear, pressure gradients, and chemical cues [1, 2, 3, 4, 5, 6]. This interplay between local adaptation and global hydraulic coupling raises a fundamental question: how do distributed memories emerge in flow networks?

Memory, however, is not a privilege of living systems. Under quasistatic cyclic driving, granular packs, crumpled sheets, amorphous solids, and mechanical metamaterials [7, 8, 9, 10, 11] can be trained so that their response depends on loading history rather than instantaneous load. Their deformation is often smooth until interrupted by abrupt, localized rearrangements, such as yield events in disordered solids or snap-through instabilities in architected structures, that reorganize the internal state [12]. Each such rearranging region can be idealized as a bistable switch with distinct up and down thresholds: a hysteron [13, 14]. When many hysterons act independently, their superposition produces hysteresis and return-point memory (RPM), whereby returning the drive to a previous extremum restores the corresponding prior state [13, 15]. When hysterons interact, a flip in one region shifts the thresholds of others, opening new state-transition pathways and enabling cascades of avalanche-like events [16, 17, 18]. This viewpoint has elevated hysteresis from a material response to a design principle: by arranging and coupling hysterons, one can encode counting, memory storage, and reproducible switching sequences as forms of mechanical information processing [19, 20, 21, 22, 23, 24]. Yet nearly all experimental realizations remain solid-state systems, where switching is rooted in snap-through, buckling, or geometric incompatibility.

Fluidic networks offer a distinct setting for such questions. Kirchhoff-like hydraulic constraints couple distant elements, so that a local change in conductance redistributes pressure and flux throughout the network. At the microscale, however, viscous flows obey the linear Stokes equations and are kinematically reversible [25]. History dependence therefore requires additional state variables that do not simply rewind under reversal, for example viscoelastic stresses, fluid–fluid interfaces, activity, or deformable boundaries [26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37]. Existing microfluidic memory and control strategies have exploited such ingredients, often together with external actuation, membrane valves, or pneumatic control [38, 39, 40, 41, 42]. What has remained missing is a passive fluidic counterpart of the hysteron paradigm: an element whose bistability arises intrinsically from quasistatic flow–structure feedback, and whose interactions emerge naturally from global hydraulic constraints.

Here we show that nonlinear elastohydrodynamic feedback between viscous flow and fiber deformation can produce bistability and hysteresis without requiring a structurally bistable element. The minimal realization is a single elastic fiber anchored inside a microfluidic channel: as the imposed flow is ramped quasistatically, the fiber bends and narrows a side gap, redistributing flux between that gap and thin lubrication layers. This closes a feedback loop in which deformation alters the hydraulic resistance, which in turn redirects flow and modifies the hydrodynamic loading. We then introduce a parallel bypass channel as a geometric load-line control that reshapes, and can suppress, bistability while simultaneously tuning long-range hydraulic interactions in fiber arrays. By varying this single geometric parameter, we traverse from a non-interacting Preisach regime to an interacting regime with avalanche-like transitions that violate return-point memory. The same control parameter therefore reshapes the constitutive response of an individual element and the collective interaction topology of the network.

Elastohydrodynamic bistability

We pattern a single elastic fiber of length LL, width bb, and height hh inside a rigid microfluidic channel of width WW and height hch_c, anchored at an upstream tilt angle θ0\theta_0 against the flow direction (Figure 1(c)). As the imposed flow rate QQ is ramped quasistatically, we image the fiber deformation and track the tip angle θ\theta. Short fibers (λ=L/W=0.65\lambda= L/W = 0.65) deform continuously, whereas long fibers (λ=1\lambda= 1) undergo a discontinuous snap above a critical flow rate QcrupQ_{cr}^{up} (Figure 1(a,b) and Supplementary Information Section 1). Varying the fiber length LL and width bb at fixed λ\lambda shifts the deformation curves but preserves the distinction between continuous and discontinuous response, showing that the transition is governed by the geometric ratio λ\lambda rather than by absolute size or stiffness.

Image not extracted

Figure 1. Elastohydrodynamic snap-through of an anchored fiber. (a) Short fiber (λ=L/W=0.65\lambda= L/W = 0.65, b=200 μmb = 200\ \mu\mathrm{m}) in a channel of width W=2000 μmW = 2000\ \mu\mathrm{m} deforms continuously under increasing flow, as characterized by the tip angle θ\theta. (b) Long fiber (λ=1\lambda= 1, b=200 μmb = 200\ \mu\mathrm{m}) in the same channel undergoes a discontinuous snap. Solid lines show the prediction of Eq. (1). Insets compare the measured and predicted neutral-axis profiles. (c) Schematic of the device. A fiber of length LL and height hh is anchored at angle θ0\theta_0 in a channel of width WW and height hch_c. The imposed flow rate QQ partitions between the tip gap (flow qgq_g, gap height hgh_g) and the thin lubrication layers (flow qiq_i, layer height hih_i). (d,e) Experimental data collapsed using the elastoviscous number El\mathrm{El}. Short fibers trace a continuous master curve (d), whereas long fibers split into unsnapped (El≲1\mathrm{El} \lesssim1) and snapped (El≳10\mathrm{El} \gtrsim10) branches (e), with the onset of snapping at El≈1\mathrm{El} \approx1. Insets show the corresponding dimensional data. (f) Measured snap-up thresholds Qcr,expupQ_{cr,exp}^{up} versus the scaling estimate Qcr,sclupQ_{cr,scl}^{up} for different anchor angles θ0\theta_0 and fibers with λ≃1\lambda\simeq1.

This distinction arises from the coupling between flow and deformation. Viscous loading bends the fiber and progressively narrows the lateral gap between the fiber and the adjacent sidewall, increasing the hydraulic resistance of the primary gap pathway, qgq_g. At the same time, fluid passes through thin lubrication layers along the sides of the fiber, whose conductance is approximately constant (Figure 1(c)). As the fiber deflects, flow repartitions between these two pathways, changing the pressure drop and traction acting on the fiber. We capture this feedback with a reduced elastohydrodynamic model

that couples lubrication-based hydraulic resistances for the two flow paths to an inextensible Euler–Bernoulli beam description of the fiber (Methods and SI Section 2). Although the fiber itself is not intrinsically bistable, the hydraulic loading depends on its instantaneous shape through the resistance of the narrowing gap, and this nonlinear flow–structure feedback is sufficient to generate bistability. This mechanism differs from conventional fluid-driven snap-through devices, where bistability is already embedded in the elastic structure and the flow merely selects between pre-existing states [30, 43].

Across a wide range of fiber lengths and widths, the measured tip angle θ\theta collapses onto a single master curve when plotted against the elastoviscous number El=μUmL3/D\mathrm{El} = \mu U_m L^3/D, where UmU_m is the mean velocity in the lubrication layers that sets the viscous loading, μ\mu is the fluid viscosity, and D=EID = EI is the flexural rigidity, with I=hb3/12I = hb^3/12 the second moment of area and EE the fiber modulus. Short fibers trace this curve continuously (Figure 1(d)), whereas long fibers appear to skip an intermediate range of El\mathrm{El} (Figure 1(e)): the data cluster near El≲1\mathrm{El} \lesssim1 in the unsnapped state and El≳10\mathrm{El} \gtrsim10 in the snapped state. This apparent gap does not reflect a discontinuity in the imposed flow rate QQ; rather, UmU_m jumps at snapping as the flow repartitions between the lubrication layers and the narrowing gap, and El∝Um\mathrm{El} \propto U_m jumps with it. The collapse therefore identifies El≈1El \approx1 as the onset criterion for snapping in long fibers. Using this criterion, and assuming that the fiber remains close to θ0\theta_0 immediately before snap-up, we infer UmU_m at onset and map it to a predicted critical flow rate QcrupQ_{\mathrm{cr}}^{\mathrm{up}} (SI Section 3.2). The resulting estimate agrees with the measured snap-up thresholds across different fiber dimensions and anchor angles θ0\theta_0 for λ≃1\lambda\simeq1 (Figure 1 (f)).

The distinction between continuous and discontinuous response follows from the constitutive structure of the elastohydrodynamic coupling. The steady fiber angle obeys the reduced relation

tan⁡θ=tan⁡θ0+El6l~p3,(1)\tan\theta= \tan\theta_0 + \frac{El}{6}\widetilde{l}_p^3, \tag*{(1)}

where l~p=lp/L\widetilde{l}_p = l_p/L is the dimensionless projected fiber length (Supplementary Information Section 2). Under flow-controlled driving, solving for the steady response θ(Q)\theta(Q) yields, for long fibers, a discontinuous jump: as QQ increases, the lower stable branch terminates at a critical flow rate QcrupQ_{\mathrm{cr}}^{\mathrm{up}}, forcing the system to snap to the upper branch (Figure 2 (a)). To reveal the full constitutive relation, we invert the problem and compute the flow required to sustain a prescribed configuration, Q(θ)Q(\theta). This uncovers an S-shaped response: two stable branches connected by an intermediate segment with negative slope, dQ/dθ<0dQ/d\theta< 0. This negative-differential branch is unstable under flow-controlled driving and therefore inaccessible during quasistatic ramping; the fiber instead jumps at the two fold points, which define the snap-up and snap-down thresholds, QcrupQ_{\mathrm{cr}}^{\mathrm{up}} and QcrdnQ_{\mathrm{cr}}^{\mathrm{dn}}, and hence the hysteresis window Qcrdn<Q<QcrupQ_{\mathrm{cr}}^{\mathrm{dn}} < Q < Q_{\mathrm{cr}}^{\mathrm{up}}. This multivalued constitutive law, with two stable branches separated by an unstable negative-slope segment, realizes the constitutive structure of a hysteron under flow-controlled driving [17, 44].

Image not extracted

Figure 2. Constitutive bistability and cusp scaling of a fluidic hysteron. (a) Full constitutive response of a snapping fiber with λ=1\lambda= 1 and b=200 μmb = 200\ \mu\mathrm{m}. Under flow-controlled driving, the fiber follows the lower stable branch until the snap-up threshold QcrupQ_{\mathrm{cr}}^{\mathrm{up}}, then jumps to the upper branch; on ramp-down it remains snapped until the lower threshold QcrdnQ_{\mathrm{cr}}^{\mathrm{dn}}, producing hysteresis. The jump amplitude Δθ\Delta\theta and hysteresis width ΔQ=Qcrup−Qcrdn\Delta Q = Q_{\mathrm{cr}}^{\mathrm{up}} - Q_{\mathrm{cr}}^{\mathrm{dn}} quantify the bistable response. (b) Jump amplitude Δθ\Delta\theta as a function of distance from threshold. Data for different θ0\theta_0 collapse when plotted against λ−λc\lambda- \lambda_c, following the square-root scaling predicted by the cusp catastrophe analysis. Inset: Δθ\Delta\theta vanishes for λ<λc\lambda< \lambda_c and becomes finite for λ>λc\lambda> \lambda_c. (c) Dimensionless hysteresis width ΔQ/Q~\Delta Q/\widetilde{Q} (Methods) as a function of λ\lambda. The hysteresis width vanishes below threshold and becomes finite in the snapping regime. Inset: the critical flow rates QcrupQ_{\mathrm{cr}}^{\mathrm{up}} and QcrdnQ_{\mathrm{cr}}^{\mathrm{dn}} form the two fold lines that emerge beyond the cusp point. Lines are theoretical model predictions while symbols represent the experimental data.

The jump amplitude Δθ\Delta\theta becomes non-zero only when λ>λc\lambda> \lambda_c, following the scaling Δθ∼(λ−λc)1/2\Delta\theta\sim(\lambda- \lambda_c)^{1/2} (Figure 2 (b)). This square-root onset is the generic scaling of a cusp catastrophe [45, 46], in which two fold lines in the (λ,Q)(\lambda,Q) control plane meet at a cusp point (λc,Qcr)(\lambda_c,Q_{\mathrm{cr}}) (Figure 2 (c), inset). The hysteresis width, ΔQ=Qcrup−Qcrdn\Delta Q = Q_{\mathrm{cr}}^{\mathrm{up}} - Q_{\mathrm{cr}}^{\mathrm{dn}}, likewise becomes finite only beyond λc\lambda_c and grows with increasing λ\lambda (Figure 2 (c)). The critical length ratio λc\lambda_c is set by the anchoring angle θ0\theta_0 and is essentially independent of the flexural rigidity D=EID = EI over the explored range. Whether a fiber snaps is therefore a geometric question, determined by (λ,θ0)(\lambda,\theta_0), whereas the size of the hysteresis window can be tuned by both geometry and stiffness (Supplementary Information Section 4).

Geometric load-line control

To turn the single fluidic hysteron into a tunable circuit element, we add a linear bypass channel of width W2W_2 in parallel with the fiber channel of width W1W_1, and define the control parameter β=W2/W1\beta= W_2/W_1. The limit β=0\beta= 0 recovers the isolated hysteron (Figure 3 (a)). As β\beta increases, an increasing fraction of the imposed flow is diverted through the bypass. Snapping is shifted to larger imposed flow rates, while both the jump amplitude Δθ\Delta\theta and the hysteresis width ΔQ\Delta Q decrease; beyond a critical value of β\beta, snapping is completely suppressed (Figure 3 (a), inset; Supplementary Information Section 5). This behavior follows from the shared pressure drop across the two branches. The total flow satisfies Q=QNL(ΔP,state)+QL(ΔP)Q = Q_{NL}(\Delta P,\mathrm{state}) + Q_L(\Delta P), where QNLQ_{NL} is the flow through the fiber channel and depends on both the pressure drop ΔP\Delta P and the instantaneous fiber state (snapped or unsnapped), while the bypass flow is linear QL=ΔP/Rb(β)Q_L = \Delta P/R_b(\beta). The bypass therefore provides a tunable load line: its linear flow–pressure relation intersects the nonlinear constitutive response of the fiber channel and selects the operating point. Decreasing the bypass resistance flattens the load line, progressively reducing the jump amplitude; beyond a critical β\beta the load line intersects the constitutive curve only once, eliminating bistability entirely (Figure 3b).

Image not extracted

Figure 3. A geometric load line tunes and suppresses the fluidic hysteron. (a) Flow–deformation curves for increasing bypass width ratio β=W2/W1\beta= W_2/W_1, for cases 1–4 with β=0\beta= 0, 0.17, 1.45, and 3.13, respectively. Increasing β\beta shifts snapping to larger imposed flow rates and reduces the jump amplitude Δθ\Delta\theta. Inset: Δθ\Delta\theta decreases monotonically with β\beta and vanishes beyond a critical bypass width. (b) Constitutive response plotted against the flow through the fiber channel, QNLQ_{\mathrm{NL}}. Solid lines show the theoretical prediction, illustrating how the bypass load line allows the system to traverse the negative-differential branch that is inaccessible in the single-channel limit. Symbols are particle-tracking-velocimetry measurements confirming the reduction of flow through the fiber channel with increasing deformation. (c) Passive flow routing for case 2. Snapping switches the preferred flow path from the fiber channel to the bypass, so that the fiber channel acts as a relatively open conduit before snapping and a relatively closed one after.

Functionally, the bypass converts the anchored fiber into a passive flow router. Before snapping, most of the flow passes through the fiber channel; after snapping, the large increase in its hydraulic resistance diverts the flow into the bypass (Figure 3c). The device therefore switches between relatively open and relatively closed flow states without requiring intrinsic elastic multistability, external actuation, or membrane-based valving [30, 36].

Interacting fluidic hysterons and avalanches

Equipped with a tunable fluidic hysteron, we now move from single-element bistability to the collective switching of coupled hysterons [14]. In the no-bypass limit (β=0\beta= 0), the two fibers experience the same imposed flow but do not interact, and therefore behave as independent hysterons (Figure 4a,Figure 4b). Labeling the unsnapped state by 0 and the snapped state by 1, two fibers with nested hysteresis loops (fiber 1 nested within fiber 2) traverse the sequence 00 →\to 01 →\to 11 →\to 10 →\to 00 under a full ramp-up/ramp-down cycle, as shown by the transition graph in Figure 4b [47]. The θ1\theta_1–θ2\theta_2 phase space reflects this independence: the system visits four well-separated clusters corresponding to the four binary states, and transitions occur along the coordinate axes, indicating that only one fiber switches at a time.

Image not extracted

Figure 4. Transition pathways in a two-hysteron fluidic network. (a) Schematic of two fibers coupled through a parallel bypass channel with width ratio β=W2/W1\beta= W_2/W_1. The individual constitutive curves are nested, with fiber 1 switching inside the hysteresis loop of fiber 2. (b) Non-interacting limit (β=0\beta= 0). The transition graph follows the classical Preisach sequence 00→01→11→10→0000 \to01 \to11 \to10 \to00, and the θ1\theta_1–θ2\theta_2 phase space contains four discrete clusters connected only by axis-aligned transitions. (c) Weakly interacting regime. A single avalanche appears during ramp-up, producing one diagonal jump in the phase space (n=1n = 1). (d) Strongly interacting regime. Avalanches occur on both ramp-up and ramp-down, producing two diagonal jumps and reversing the unsnapping order relative to the snapping order (n=2n = 2). Lines represent the theoretical model and symbols represent experimental data.

Introducing a bypass channel couples the fibers through global hydraulic constraints. The two fibers remain spatially separated, but the state of either fiber changes the effective resistance of the nonlinear branch and thereby the flow seen by the other. The interaction is long-ranged and effectively all-to-all within the fiber branch, with strength controlled by the geometric parameter β=W2/W1\beta= W_2/W_1.

For small β\beta, the coupling already produces qualitatively new behavior. When one fiber snaps, 00 →\to 01, the resistance of the fiber branch increases, diverting flow into the bypass and reducing the flow through the remaining unsnapped fiber. The second fiber therefore snaps only at a larger imposed flow. When it eventually does so, the additional increase in branch resistance reroutes enough flow to unsnap the first fiber, producing an avalanche-like transition 01 →\to 10; as the flow rate increases further, the first fiber snaps up again arriving at 11 state (Figure 4c). In the θ1\theta_1–θ2\theta_2 phase space, this appears as a diagonal jump rather than a sequence of axis-aligned transitions. We denote by nn the number of such avalanches, with n=0n = 0 for the non-interacting limit and n=1n = 1 for this weakly coupled regime.

At larger β\beta, corresponding to stronger coupling, a second avalanche appears on ramp-down. The unsnapping order is now reversed relative to the snapping order, 11 →\to 10 →\to 01 →\to 00, and the phase space shows two diagonal jumps (Figure 4d). In hysteron language, this is an “antiferromagnetic-like” interaction: the snapped state of one fiber favors the unsnapped state of the other, producing avalanche-mediated pathways that are forbidden in the non-interacting Preisach limit [16, 23, 48].

Collective memory and return-point-memory violation

To probe collective memory, we study arrays of N=10N = 10 fibers in series under standard write–read protocols (Figure 5) (a,b)), in which the drive is reversed at a sequence of nested turning points and then returned to earlier extrema to test whether the system recovers its prior state (Figure 5 (d,e) inset). In the non-interacting limit (β=0\beta= 0), the array exhibits return-point memory (RPM): minor loops remain enclosed within the major loop, and returning to a previously visited extremum restores the corresponding earlier state [14]. This is observed both in the mean tip angle ⟨θ⟩\langle\theta\rangle and in the average binary state ⟨s⟩=1N∑isi\langle s\rangle= \frac{1}{N}\sum_i s_i, where si=0,1s_i = 0,1 denotes the unsnapped or snapped state of the iith fiber (Figure 5 (d) and Supplementary Information Section 6).

Image not extracted

Figure 5. Collective memory in arrays of fluidic hysterons. (a) Schematic of an array of N=10N = 10 fibers connected in series with a parallel bypass channel. (b) Measured snap-up and snap-down thresholds, QcrupQ_{\mathrm{cr}}^{\mathrm{up}} and QcrdnQ_{\mathrm{cr}}^{\mathrm{dn}}, for each of the ten fibers, showing the spread of individual switching flow rates. (c) Number of avalanches nn as a function of the bypass width ratio β\beta. Stronger interactions produce progressively more avalanches per cycle. (d) Mean fiber angle ⟨θ⟩\langle\theta\rangle under nested write–read protocols in the non-interacting limit (β=0\beta= 0). Minor loops remain enclosed within the major loop, demonstrating return-point memory (RPM). Inset: quasistatic flow-driving protocol. (e) Mean fiber angle ⟨θ⟩\langle\theta\rangle for an interacting network (β≈0.3\beta\approx0.3). Minor loops extend beyond the major loop and the system fails to return to previously visited states, demonstrating violation of RPM. Inset: quasistatic flow-driving protocol.

Introducing bypass-mediated interactions (β>0\beta> 0) destroys RPM. Avalanches allow the system to access transition pathways that are forbidden in the non-interacting limit (Figure 5 (c)), and minor loops extend beyond the major loop (Figure 5 (e)): the system visits states not encountered on the initial ramp and fails to recover previously visited intermediate states upon reversal. This is consistent with the general expectation that interacting hysterons admit a profusion of transition graphs, many of which violate RPM; frustrated, antiferromagnetic-like interactions of the kind produced by our bypass coupling provide a particularly direct route to such violations [16, 18, 48]. The bypass ratio β\beta therefore provides a single geometric control parameter that continuously tunes the network from Preisach to non-Preisach collective behavior.

Discussion

Together, these results establish a fluidic realization of the hysteron framework developed in solid-state mechanical systems [11, 14]. Two recent studies have explored related territory in network analogues and fluidic networks: Altman et al. [49] realized collective memory states and avalanches in an electronic analogue of bistable flow networks while Martinez-Calvo et al. [50] developed fluidic memristors, where hysteresis emerges collectively from volume storage and nonlinear resistance. By contrast, in our system a single fiber already acts as a hysteron. Its bistability is geometric in origin: the snapping transition has the structure of a cusp catastrophe, controlled by the length ratio λ\lambda and anchor angle θ0\theta_0, while stiffness sets the flow scale and tunes the width of the hysteresis window. Cusp catastrophes are generic organizing structures for the onset of bistability in nonlinear systems, from mean-field first-order transitions to optical bistability [45, 51]; here, this structure emerges from the coupling between viscous flow and elastic deformation rather than from an intrinsic elastic energy landscape.

The fluidic setting also changes the nature of interactions between hysterons. In solid-state systems, coupling is typically mediated by short-ranged elastic stresses or contact forces. Here, a local change in resistance redistributes pressure and flux throughout the network through Kirchhoff-like hydraulic constraints. In the presence of a bypass, this leads to an all-to-all coupling within the fiber branch, with interaction strength controlled by the single geometric parameter β=W2/W1\beta= W_2/W_1. The resulting antiferromagnetic-like interaction, in which snapping of one fiber disfavors snapping of the others, generates avalanches and violates return-point memory, consistent with the predicted proliferation of transition pathways in frustrated hysteron networks [16, 17, 23, 48]. Our fluidic setting therefore provides a tunable, optically resolved realization of interacting-hysteron physics, in which interaction strength is set by geometry rather than by local contact mechanics or material disorder.

The ability to tune both the local constitutive response and the interaction strength using geometry suggests a broader design principle for memory-bearing flow networks. Embedding fluidic hysterons in networks beyond a simple linear series, such as loops, trees, or disordered graphs, would introduce topology-dependent interactions and provide experimental access to questions about how network architecture shapes transition pathways, frustration, and memory capacity [18, 52, 53]. In this sense, fluidic hysteron networks provide a complementary experimental setting for the statistical physics of interacting hysterons, with interactions mediated by conservation laws rather than elastic contact or stress redistribution.

Finally, this system provides a minimal physical analogue of living flow networks, which couple local mechanical adaptation, such as compliance and shear-dependent remodeling, to global hydraulic constraints [54]. Real biological networks are far richer, involving active remodeling, growth, and biochemical signalling. Yet our results show that passive ingredients alone suffice to generate history-dependent collective response, identifying flow–structure feedback as a sufficient mechanism for memory in fluidic networks. A central open question is therefore which aspects of memory and adaptation in living and soft adaptive flow networks genuinely require active control, and which are already latent in the passive mechanics of coupled deformable elements under global constraints.

**Note added.—**After submission of this manuscript, Stinissen et al. reported two intrinsically bistable inflatable membranes coupled by an equal-volume-change constraint, with preset volume selecting the pair’s state-diagram topology and whether avalanches occur [55]. By contrast, the hysterons here are elastic fibers that are not intrinsically bistable: bistability arises from through-flow elastohydrodynamic feedback, and interactions from bypass-mediated hydraulic redistribution in a flow network.

Methods

Device fabrication

Experiments were performed in polydimethylsiloxane (PDMS) microfluidic devices fabricated by standard soft lithography. Channel moulds were patterned on silicon wafers using a laser writer. A PDMS prepolymer and curing agent mixture (10:1 by weight) was poured onto the mould and cured for 4 h at 65°C. The cured PDMS slab was then bonded to a glass slide spin-coated with a thin layer of uncured PDMS mixture. The channels used in this study had widths W=800W = 800, 1200, 1600, and 2000 μm\mu\mathrm{m} and a height hc=80 μmh_{\mathrm{c}} = 80\ \mu\mathrm{m} (Figure 1(c)).

Deformable hydrogel fibers were fabricated in situ using stop-flow lithography [56]. One wall of each channel contained PDMS microposts that served as anchors for the fibers. A photosensitive solution of PEGDA, Milli-Q water, and photoinitiator was injected into the channel using a syringe pump. The flow was then stopped and the solution exposed to masked ultraviolet light, photopolymerizing the hydrogel structures. This procedure yielded fibers with Young’s modulus E=0.30±0.04E = 0.30 \pm0.04 MPa, height h=70h = 70 μm, width b∈[100,400]b \in[100,400] μm, and length ratio λ=L/W∈[0.65,1.00]\lambda= L/W \in[0.65,1.00]. Fibers were anchored at upstream tilt angles θ0∈{−50∘,−40∘,−30∘,−20∘}\theta_0 \in\{-50^\circ,-40^\circ,-30^\circ,-20^\circ\}. Some devices additionally contained a parallel bypass channel of width W2W_2 alongside the fiber channel of width W1W_1, characterized by the width ratio β=W2/W1\beta= W_2/W_1 (inset of Figure 3 (a)).

Flow control and imaging

The volumetric flow rate QQ was imposed using a Harvard Apparatus syringe pump. Quasistatic ramps in the range Q∈[10−3,100]Q \in[10^{-3},10^0] μL s−1^{-1} were chosen to ensure negligible inertia (Re⁡≪1\operatorname{Re} \ll1). Fiber deformation was imaged in top view at 1 Hz using brightfield microscopy. The tip angle θ\theta was extracted from binary masks of the fiber centerline using custom image-analysis code based on intensity thresholding. Uncertainty was estimated from repeated segmentation and calibration.

Reduced elastohydrodynamic model

The fiber was modeled as an inextensible Euler–Bernoulli beam with linear constitutive law M=EIκM = EI\kappa, clamped at the anchor and free at the tip. The surrounding flow was decomposed into parallel pathways: a variable-height tip-gap pathway with hydraulic resistance Rg(θ)R_g(\theta) and constant-thickness lubrication-layer pathways with resistance RiR_i. Each pathway was described using lubrication approximations appropriate for slender gaps (Supplementary Information Section 2). Flow partitioning between these pathways determined the pressure and shear tractions acting on the fiber, which were projected onto an effective distributed load in the beam equation. Steady states were obtained by solving the coupled nonlinear elastohydrodynamic system for the tip angle θ\theta, and, when needed, the full fiber shape. Stability was assessed by identifying segments of the constitutive curve with dQ/dθ<0dQ/d\theta< 0.

Elastoviscous number and geometric parameters

The length ratio is λ=L/W\lambda= L/W. A geometric prefactor Λ=3b(hc+h)hi2\Lambda= \frac{3b(h_c+h)}{h_i^2} collects the remaining geometric parameters. The elastoviscous number is defined using the mean velocity in the lubrication layers, UmU_m, as El⁡=ΛμUmL3D\operatorname{El} = \frac{\Lambda\mu U_m L^3}{D}, where D=EID = EI is the flexural rigidity. For long fibers, UmU_m was inferred from the hydraulic partition model using the measured flow rate QQ and device geometry. A corresponding characteristic flow scale, Q~:=6qiAμL2=3DhiAμL2\widetilde{Q} := \frac{6q_i}{A\mu L^2} = \frac{3Dh_i}{A\mu L^2}, combines the elastic and viscous contributions into a single control scale used to nondimensionalize the hysteresis width in Figure 2 (c) (Supplementary Information Section 2).

Transition graphs, state labels, and RPM protocols

Each fiber was assigned a binary state si∈{0,1}s_i \in\{0,1\} according to whether θi\theta_i lay on the lower (unsnapped) or upper (snapped) branch. The switching thresholds QcrupQ^{up}_{cr} and QcrdnQ^{dn}_{cr} were determined from ramp-up and ramp-down experiments. Transition graphs were constructed by listing the stable states and directed transitions under monotonic increases or decreases of QQ. The return-point-memory protocols in Figure 5 (d,Figure 5e) employed nested turning points Q1<Q2<…Q_1 < Q_2 < \ldots, with controlled partial cycles; the readout followed the recovered state at repeated extrema. The collective state of an array was characterized by the mean angle ⟨θ⟩:=1N∑iθi\langle\theta\rangle:= \frac{1}{N}\sum_i\theta_i and the mean binary state ⟨s⟩:=1N∑isi\langle s\rangle:= \frac{1}{N}\sum_i s_i, where θi\theta_i and sis_i are the tip angle and binary state of the iith fiber, respectively.

Data availability

All data and analysis codes are available upon publication.

References

  1. [1]Atsushi Tero, Seiji Takagi, Tetsu Saigusa, Kentaro Ito, Dan P. Bebber, Mark D. Fricker, Kenji Yumiki, Ryo Kobayashi, and Toshiyuki Nakagaki. Rules for biologically inspired adaptive network design. Science, 327(5964):439–442, 2010.DOI
  2. [2]Axel R. Pries, Michael Höpfner, Ferdinand le Noble, Mark W. Dewhirst, and Timothy W. Secomb. The shunt problem: control of functional shunting in normal and tumour vasculature. Nature Reviews Cancer, 10(8):587–593, 2010.DOI
  3. [3]Henrik Ronellenfitsch and Eleni Katifori. Global optimization, local adaptation, and the role of growth in distribution networks. Phys. Rev. Lett., 117:138301, Sep 2016.DOI
  4. [4]Marcus Roper and Agnes Seminara. Mycofluidics: The fluid mechanics of fungal adaptation. Annual Review of Fluid Mechanics, 51(Volume 51, 2019):511–538, 2019.DOI
  5. [5]Mirna Kramar and Karen Alim. Encoding memory in tube diameter hierarchy of living flow network. Proceedings of the National Academy of Sciences, 118(10):e2007815118, 2021.DOI
  6. [6]Loreto Oyarte Galvez, Corentin Bisot, Philippe Bourrianne, Rachael Cargill, Malin Klein, Marije van Son, Jaap van Krugten, Victor Caldas, Thomas Clerc, Kai-Kai Lin, Félix Ka- hane, Simon van Staalduine, Justin D. Stewart, Victoria Terry, Bianca Turcu, Sander van Otterdijk, Antoine Babu, Marko Kamp, Marco Seynen, Bas Steenbeek, Jan Zomerdijk, Evelina Tutucci, Merlin Sheldrake, Christophe Godin, Vasilis Kokkoris, Howard A. Stone, E. Toby Kiers, and Thomas S. Shimizu. A travelling-wave strategy for plant–fungal trade. Nature, 639(8053):172–180, 2025.
  7. [7]Nathan C. Keim and Paulo E. Arratia. Mechanical and microscopic properties of the reversible plastic regime in a 2d jammed material. Phys. Rev. Lett., 112:028302, Jan 2014.pubmed.ncbi.nlm.nih.gov/24484046
  8. [8]K. L. Galloway, E. G. Teich, X. G. Ma, Ch. Kamm, I. R. Graham, N. C. Keim, C. Reina, D. J. Jermolmack, A. G. Yodh, and P. E. Arratia. Relationships between structure, memory and flow in sheared disordered materials. Nature Physics, 18(5):565–570, 2022.DOI
  9. [9]Dor Shohat, Daniel Hexner, and Yoav Lahini. Memory from coupled instabilities in unfolded crumpled sheets. Proceedings of the National Academy of Sciences, 119(28):e2200028119, 2022.DOI
  10. [10]Chaviva Sirote-Katz, Dor Shohat, Carl Merrigan, Yoav Lahini, Cristiano Nisoli, and Yair Shokef. Emergent disorder and mechanical memory in periodic metamaterials. Nature Communications, 15(1):4008, 2024.
  11. [11]Joseph D. Paulsen and Nathan C. Keim. Mechanical memories in solids, from disorder to design. Annual Review of Condensed Matter Physics, 16(Volume 16, 2025):61–81, 2025.DOI
  12. [12]Muhittin Mungan, Eric Clément, Damien Vandembroucq, and Srikanth Sastry. Self-organization, memory, and learning: From driven disordered systems to living matter. Annual Review of Condensed Matter Physics, 17(Volume 17, 2026):207–232, 2026.DOI
  13. [13]Ferenc Preisach. Über die magnetische nachwirkung. Zeitschrift für physik, 94(5):277–302, 1935.
  14. [14]Nathan C Keim, Joseph D Paulsen, Zorana Zarevac, Srikanth Sastry, and Sidney R Nagel. Memory formation in matter. Reviews of Modern Physics, 91(3):035002, 2019.DOI
  15. [15]James P. Sethna, Karin Dahmen, Sivan Kartha, James A. Krumhansl, Bruce W. Roberts, and Joel D. Shore. Hysteresis and hierarchies: Dynamics of disorder-driven first-order phase transformations. Phys. Rev. Lett., 70:3347–3350, May 1993.pubmed.ncbi.nlm.nih.gov/10053845
  16. [16]Martin van Hecke. Profusion of transition pathways for interacting hysterons. Phys. Rev. E, 104:054608, Nov 2021.DOI
  17. [17]Gentian Muhaxheri and Christian D Santangelo. Bifurcations of inflating balloons and interacting hysterons. Physical Review E, 110(2):024209, 2024.DOI
  18. [18]Dor Shohat and Martin van Hecke. Geometric control and memory in networks of hysteretic elements. Physical Review Letters, 134(18):188201, 2025.DOI
  19. [19]Nidhi Pashine, Daniel Hexner, Andrea J. Liu, and Sidney R. Nagel. Directed aging, memory, and nature’s greed. Science Advances, 5(12):eaax4215, 2019.
  20. [20]Tian Chen, Mark Pauly, and Pedro M. Reis. A reprogrammable mechanical metamaterial with stable memory. Nature, 589(7842):386–390, 2021.DOI
  21. [21]Hadrien Bense and Martin van Hecke. Complex pathways and memory in compressed corrugated sheets. Proceedings of the National Academy of Sciences, 118(50):e211436118, 2021.DOI
  22. [22]Lennard J. Kwakernaak and Martin van Hecke. Counting and sequential information processing in mechanical metamaterials. Phys. Rev. Lett., 130:268204, Jun 2023.
  23. [23]Jingran Liu, Margot Teunisse, George Korovin, Ivo R Vermaire, Lishuai Jin, Hadrien Bense, and Martin van Hecke. Controlled pathways and sequential information processing in serially coupled mechanical hysterons. Proceedings of the National Academy of Sciences, 121(22):e2308414121, 2024.DOI
  24. [24]Joseph D. Paulsen. Mechanical hysterons with tunable interactions of general sign. Nature Communications, 17:2799, 2026.DOI
  25. [25]EM Purcell. Life at low reynolds number. American Journal of Physics, 45(1), 1977.
  26. [26]Alex Groisman, Markus Enzelberger, and Stephen R. Quake. Microfluidic memory and control devices. Science, 300(5621):955–958, 2003.pubmed.ncbi.nlm.nih.gov/12738857
  27. [27]Manu Prakash and Neil Gershenfeld. Microfluidic bubble logic. Science, 315(5813):832–835, 2007.
  28. [28]Michael J. Fuerstman, Piotr Garstecki, and George M. Whitesides. Coding/decoding and reversibility of droplet trains in microfluidic networks. Science, 315(5813):828–832, 2007.pubmed.ncbi.nlm.nih.gov/17204610
  29. [29]José Alvarado, Jean Comtet, Emmanuel De Langre, and AE Hosoi. Nonlinear flow response of soft hair beds. Nature Physics, 13(10):1014–1019, 2017.DOI
  30. [30]Michael Gomez, Derek E Moulton, and Dominic Vella. Passive control of viscous flow via elastic snap-through. Physical review letters, 119(14):144502, 2017.DOI
  31. [31]Anneline H. Christensen and Kaare H. Jensen. Viscous flow in a slit between two elastic plates. Phys. Rev. Fluids, 5:044101, Apr 2020.DOI
  32. [32]Jean-Fran Çois Louf, Jan Knoblauch, and Kaare H. Jensen. Bending and stretching of soft pores enable passive control of fluid flows. Phys. Rev. Lett., 125:098101, Aug 2020.
  33. [33]Keunhwan Park, Aude Tixier, Magnus Paludan, Emil Østergaard, Maciej Zwieniecki, and Kaare H Jensen. Fluid-structure interactions enable passive flow control in real and biomimetic plants. Physical Review Fluids, 6(12):123102, 2021.
  34. [34]Ofek Peretz, Anand K Mishra, Robert F Shepherd, and Amir D Gat. Underactuated fluidic control of a continuous multistable membrane. Proceedings of the National Academy of Sciences, 117(10):5217–5221, 2020.
  35. [35]Olek N Christov. Soft hydraulics: from newtonian to complex fluid flows through compliant conduits. Journal of Physics: Condensed Matter, 34(6):063001, nov 2021.
  36. [36]Hemanshul Garg, Pier Giuseppe Ledda, Jon Skov Pedersen, and Matteo Pezzulla. Passive viscous flow selection via fluid-induced buckling. Physical Review Letters, 133(8):084001, 2024.
  37. [37]Camille Jorge, Amélie Chardac, Alexis Poncet, and Denis Bartolo. Active hydraulics laws from frustration principles. Nature Physics, 20(2):303–309, 2024.arxiv.org/abs/2305.06078
  38. [38]Minsoung Rhee and Mark A. Burns. Microfluidic pneumatic logic circuits and digital pneumatic microprocessors for integrated microfluidic systems. Lab Chip, 9:131–13143, 2009.DOI
  39. [39]James A. Weaver, Jessica Melin, Don Stark, Stephen R. Quake, and Mark A. Horowitz. Static control logic for microfluidic devices using pressure-gain valves. Nature Physics, 6(3):218–223, 2010.DOI
  40. [40]Bobak Mosadegh, Chuan-Hsien Kuo, Yi-Chung Tung, Yusuke Torisawa, Tommaso Bersano-Begey, Hossein Tavana, and Shuichi Takayama. Integrated elastomeric components for autonomous regulation of sequential and oscillatory flow switching in microfluidic devices. Nature Physics, 6(6):433–437, 2010.pubmed.ncbi.nlm.nih.gov/20526435
  41. [41]Philip N. Duncan, Transon V. Nguyen, and Elliot E. Hui. Pneumatic oscillator circuits for timing and control of integrated microfluidics. Proceedings of the National Academy of Sciences, 110(45):18104–18109, 2013.DOI
  42. [42]Siavash Ahrar, Manasi Raje, Irene C Lee, and Elliot E Hui. Pneumatic computers for embedded control of microfluidics. Science Advances, 9(22):eadg0201, 2023.
  43. [43]Pier Giuseppe Ledda, Hemanshul Garg, Vitus Østergaard-Clausen, Lucas Krumenacker Rudzki, Ahmad Madary, and Matteo Pezzulla. Fluid-induced snap-through instability of spherical shells. Phys. Rev. Lett., 135:234002, Dec 2025.DOI
  44. [44]Gentian Muhaxheri, Victoria Antonetti, and Christian D Santangelo. A catastrophic approach to designing interacting hysterons. New Journal of Physics, 27(12):125004, 2025.DOI
  45. [45]Tim Poston and Ian Stewart. Catastrophe theory and its applications. Courier Corporation, 1978.
  46. [46]Steven H Strogatz. Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering. Chapman and Hall/CRC, 1994.
  47. [47]M. Mert Terzi and Muhittin Mungan. State transition graph of the preisach model and the role of return-point memory. Phys. Rev. E, 102:012122, Jul 2020.DOI
  48. [48]Chloe W Lindeman, Travis R Jalowiec, and Nathan C Keim. Generalizing multiple memories from a single drive: The hysteron latch. Science Advances, 11(5):eadr5933, 2025.DOI
  49. [49]Lauren E. Altman, Nadia Aguilar, Douglas J. Durian, Miguel Ruiz-Garcia, and Eleni Kafriori. Collective behavior and memory states in flow networks with tunable bistability. Nature Communications, 2026.
  50. [50]Alejandro Martínez-Calvo, Matthew D Bibiano, Anneline H Christensen, Eleni Katifori, Kaare H Jensen, and Miguel Ruiz-García. The fluidic memristor as a collective phenomenon in elastohydrodynamic networks. Nature Communications, 15(1):3121, 2024.
  51. [51]Luigi A. Lugiato. Theory of optical bistability. In E. Wolf, editor, Progress in Optics, volume 21, pages 69–216. North-Holland, Amsterdam, 1984.
  52. [52]Komal Bhattacharyya, David Zwicker, and Karen Alim. Memory formation in adaptive networks. Phys. Rev. Lett., 129:028101, Jul 2022.DOI
  53. [53]Komal Bhattacharyya, David Zwicker, and Karen Alim. Memory capacity of adaptive flow networks. Phys. Rev. E, 107:034407, Mar 2023.DOI
  54. [54]Mathieu Le Verge-Serandour and Karen Alim. Physarum polycephalum: Smart network adaptation. Annual Review of Condensed Matter Physics, 15(Volume 15, 2024):263–289, 2024.
  55. [55]Katrien Stininssen, Franco Nicolas Piñan Bausaldo, and Benjamin Gorissen. Interacting parallel fluidic hysterons. Advanced Science, page e76158, 2026/07/16 2026.
  56. [56]Dhananjay Dendukuri, Shelley S Gu, Daniel C Pregibon, T Alan Hatton, and Patrick S Doyle. Stop-flow lithography in a microfluidic device. Lab on a Chip, 7(7):818–828, 2007.pubmed.ncbi.nlm.nih.gov/17593999

Paper details

vlnc:W_ww9a2n7n
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

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