Constitutive Return-Point Memory (RPM)

• Constitutive RPM is a hysteretic memory effect in materials, where the stress–strain curve retraces its path exactly upon reversing strain between fixed bounds.
• It is modeled through Preisach and state-space formulations that incorporate hysterons, rate independence, and no-passing rules to achieve exact minor-loop closure.
• Experimental studies in VACNT foams and knitted fabrics reveal key signatures such as congruent nested loops and wiping-out of sub-extremal excursions.

to=arxiv_search 开号链接 code {"query":"\"constitutive return point memory\" hysteresis Preisach mechanical memory", "max_results": 10} to=arxiv_search tunngatillugu code {"query":"(Dresselhaus et al., 1 Dec 2025) return point memory knitted fabrics", "max_results": 5} to=arxiv_search tunngatillugu code {"query":"(Gupta et al., 16 Sep 2025) VACNT return point memory", "max_results": 5} to=search_arxiv tunngatillugu code {"query":"constitutive return point memory", "max_results": 10} Constitutive return-point memory (RPM) denotes a hysteretic constitutive response in which a material exactly retraces its previous unloading/loading path when driven between two strain limits, provided no new extremum is exceeded. In mechanical stress–strain behavior, if a curve $\sigma(\epsilon)$ is driven monotonically from $\epsilon=0$ up to $\epsilon=\epsilon_1$, then reversed from $\epsilon_1$ down to $\epsilon_2<\epsilon_1$, and then reversed again from $\epsilon_2$ back to $\epsilon_1$, the stress returns precisely along the same path; more generally, for any sequence $$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$, the stress state at $\epsilon_1$ is uniquely recovered. This is directly analogous to return-point memory in ferromagnetic hysteresis, where nested minor loops close exactly and partial excursions are wiped out (Gupta et al., 16 Sep 2025).

1. Definition, signatures, and constitutive meaning

Return-point memory is commonly identified by exact minor-loop closure: whenever the external drive is reversed at some extremum, the system returns exactly to the same internal state when the drive returns to that same value. In the Preisach picture, this means that a system of elementary hysterons subjected to a nested sequence of reversals produces closed minor loops. In constitutive language, RPM is therefore a constraint on admissible hysteretic stress–strain relations, not merely a phenomenological visual feature of a loop (Gupta et al., 16 Sep 2025).

Three experimentally salient signatures recur across constitutive RPM systems. First, minor loops close on themselves with no drift. Second, nested loops are congruent and lie within larger loops. Third, excursions that do not exceed a previous extremum are wiped out: once a larger reversal is imposed, earlier minor loops do not survive as independent memories. These features are explicit in the classical definition of RPM for hysteretic systems and in analyses of the Preisach transition graph, where RPM organizes reachable states into a hierarchy of loops and subloops (Terzi et al., 2020).

Rate independence is central to the constitutive interpretation. A hysteresis operator

$y(t)={\cal W}[H](t)$

is rate independent when its output depends on the values of a piecewise-linear input and on the order of up- and down-segments, but not on their slopes. Under rate independence plus RPM, inputs with the same sequence of local maxima and minima are equivalent, and any excursion confined between previously visited extrema leaves no new memory. This gives a minimal state-space realization in which the state is an equivalence class of past input histories rather than a detailed microscopic trajectory (Langvagen, 2017).

2. Preisach, state-space, and constitutive formulations

The canonical constitutive representation of RPM is the Preisach model. It represents the response as a continuum of elementary hysteretic units, or hysterons, $\epsilon=0$0 with thresholds $\epsilon=0$1,

$\epsilon=0$2

with macroscopic response

$\epsilon=0$3

where $\epsilon=0$4 is the Preisach density. This construction automatically yields nonlocal memory, wiping-out, and congruency (Dresselhaus et al., 1 Dec 2025).

A complementary formulation is state-space based. Under rate independence, RPM, and a reachable demagnetized state, every history is equivalent to a reduced memory sequence parameterized by amplitudes

$\epsilon=0$5

constrained by

$\epsilon=0$6

State evolution consists of appending or incrementing the last coordinate as the field reverses, together with Madelung deletion whenever $\epsilon=0$7, which removes a wiped-out small loop. Read-out functions $\epsilon=0$8 must satisfy branch continuity, Madelung deletion, and tangential continuity at loop-closing surfaces; within this formalism, Rayleigh and Preisach models become particular constitutive realizations (Langvagen, 2017).

From a graph-theoretic standpoint, the Preisach automaton has stable configurations as vertices and two directed maps, $\epsilon=0$9 and $\epsilon=\epsilon_1$0, corresponding to upward and downward threshold crossings. A configuration $\epsilon=\epsilon_1$1 is stable at field $\epsilon=\epsilon_1$2 precisely if

$\epsilon=\epsilon_1$3

RPM then imposes an absorption property on loops: if $\epsilon=\epsilon_1$4 is a loop with $\epsilon=\epsilon_1$5 and $\epsilon=\epsilon_1$6, the intermediate states on one boundary return to the corresponding endpoint on reversal. The resulting loop hierarchy is governed not only by the actual switching fields but, over a large portion of the graph, by the ordering permutation $\epsilon=\epsilon_1$7 that records the sequence in which hysterons flip on the descending sweep (Terzi et al., 2020).

3. Order preservation, no-passing, and exact return

A standard route to exact RPM uses partial order and no-passing. In the zero-temperature Random-Field Blume-Capel model, spin-1 variables $\epsilon=\epsilon_1$8 evolve by energy-lowering Glauber updates under the Hamiltonian

$\epsilon=\epsilon_1$9

With the natural chain $\epsilon_1$0, if two configurations are initially ordered and the applied fields satisfy $\epsilon_1$1 at all times, the order is preserved throughout the dynamics. This no-passing property implies abelian dynamics and, in turn, return-point memory: minor loops close exactly, including in the double-loop hysteresis produced for $\epsilon_1$2 (Aldrin et al., 2021).

The constitutive ingredients distilled from that model are explicit: locally interacting metastable variables, threshold transitions depending only on a monotonic local field, a partial order that makes the update rule order preserving, adiabatic monotonic driving, and deterministic relaxation to a unique metastable state for each field. When these conditions hold, one obtains a single-valued return map $\epsilon_1$3, exact return-point memory $\epsilon_1$4, congruent nested minor loops, and a constitutive law expressible through cavity recursions (Aldrin et al., 2021).

The no-passing route is sufficient but not necessary. In the athermal quasi-static automaton framework, one may replace an assumed intrinsic partial order by dynamic partial orders defined directly from the maps $\epsilon_1$5 and $\epsilon_1$6: $\epsilon_1$7 Their intersection defines a loop partial order, and loop-RPM requires switch-back only across genuine loops $\epsilon_1$8. This weaker condition still forces a rigid intra-loop structure: hierarchical decomposition into subloops, an ordered-tree representation, and a planar state-transition graph inside each loop. It does not, however, strongly constrain inter-loop transitions, so systems with loop-RPM but without no-passing may display long transients or subharmonic response under periodic forcing (Mungan et al., 2018).

A further extension generalizes no-passing and RPM to multiple control fields. For Ising spins $\epsilon_1$9 driven by $\epsilon_2<\epsilon_1$0 control fields $\epsilon_2<\epsilon_1$1 through local fields

$\epsilon_2<\epsilon_1$2

with $\epsilon_2<\epsilon_1$3 and $\epsilon_2<\epsilon_1$4, componentwise monotonic protocols preserve product-order comparability. The corresponding multi-field RPM theorem states that any excursion within a hyperrectangle of previously visited extrema returns exactly to the prior microstate when the maximum corner is re-attained (Croce et al., 22 May 2026).

4. Mechanical realizations in elastically recoverable media

Two recent experimental realizations place constitutive RPM directly in mechanical stress–strain behavior: vertically aligned carbon nanotube (VACNT) foams and knitted fabrics. In both cases the observed hysteresis is rate independent and incompatible with simple viscoelastic fading memory.

System	Core observation	Constitutive mechanism/model
VACNT foams	Minor loops close exactly on themselves; no stress relaxation or creep; quasistatic loops invariant with strain-rate over three decades	Rate-independent nanoscale stick-slip friction; DDSSF spring-slider ensemble (Gupta et al., 16 Sep 2025)
Knitted fabrics	Large hysteresis, return-point memory, wiping-out, and congruent nested minor loops under cyclic uniaxial loading	Extended Preisach model with $\epsilon_2<\epsilon_1$5 and internal entanglement strain $\epsilon_2<\epsilon_1$6 (Dresselhaus et al., 1 Dec 2025)

In VACNT foams, films 1.5–3 mm thick grown by floating-catalyst CVD were punched into $\epsilon_2<\epsilon_1$7 samples, preconditioned by compression to $\epsilon_2<\epsilon_1$8 at rate $\epsilon_2<\epsilon_1$9 for five cycles, and then driven through major and minor loops. The total strain during minor-loop probing was

$\epsilon_2$0

and the dynamic modulus was defined as

$\epsilon_2$1

The constitutive DDSSF model represents each local CNT bundle as a linear spring in parallel with a Coulomb slider: $\epsilon_2$2 Under loading, $\epsilon_2$3; under unloading, $\epsilon_2$4. An assembly of such elements in series yields smooth global hysteresis with RPM because sliders unlock and relock in a well-ordered sequence. The same mechanism allows dynamic softening with increasing $\epsilon_2$5 and static precompression stiffening with increasing $\epsilon_2$6, and in a VACNT–Al periodic array the effective pulse speed $\epsilon_2$7 varied by $\epsilon_2$8–$\epsilon_2$9 over the tested range (Gupta et al., 16 Sep 2025).

In knitted fabrics, acrylic ribbed fabrics tested under uniaxial cyclic loading on an Instron at $\epsilon_1$0 exhibited large hysteresis loops, return-point memory, wiping-out, and congruent nested minor loops. These effects were observed in both wale-wise and course-wise stretching, and in rib and stockinette knits. Stress relaxation at fixed strain followed a slow power law,

$\epsilon_1$1

much slower than the loading/unloading timescale, supporting a rate-independent interpretation. Because the measured loops are asymmetric and evolve with the largest strain in history, the Preisach density and relays were extended to depend on

$\epsilon_1$2

and an internal entanglement strain $\epsilon_1$3 (Dresselhaus et al., 1 Dec 2025).

The knitted-fabric constitutive model decomposes total strain as

$\epsilon_1$4

with adaptive modulus

$\epsilon_1$5

The entanglement strain evolves through the rebound function

$\epsilon_1$6

and Karush–Kuhn–Tucker conditions

$\epsilon_1$7

The constitutive integral becomes

$\epsilon_1$8

Parameters were fit by nonlinear least squares using Matlab fmincon, minimizing

$\epsilon_1$9

The dissipation per cycle,

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$0

was verified to remain positive, and once $$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$1 returned to a previous extremum the simulated stress matched experiment to $$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$2 over many cycles (Dresselhaus et al., 1 Dec 2025).

5. Approximate and emergent constitutive RPM in disordered solids

Not all constitutive memories are exact in the strong Sethna-style sense. In the quenched mesoscopic elasto-plastic (QMEP) model of an amorphous solid, the local stress satisfies

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$3

where the Eshelby kernel is long-ranged and sign-changing,

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$4

Because this non-monotone interaction violates strict no-passing, exact RPM is not guaranteed. Nevertheless, after cyclic training

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$5

the system develops an effective anisotropy or polarization that restores approximate minor-loop closure for $$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$6. Numerically,

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$7

and the mid-cycle read-out distance obeys

$$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$8

up to $$0\to\epsilon_1\to\epsilon_2\to\epsilon_3\ldots\to\epsilon_2\to\epsilon_1$$9. The resulting behavior is described as quasi-RPM, and a minimal Preisach-like model built from polarized local thresholds reproduces the main read-out curves despite neglecting interactions (Kumar et al., 2024).

An emergent version arises in cyclic inverse design of elastic moduli in disordered sphere packings. There the trained variables are species diameters $\epsilon_1$0, updated by gradient descent,

$\epsilon_1$1

Repeated training between $\epsilon_1$2 and $\epsilon_1$3 drives the system to a marginally absorbing manifold $\epsilon_1$4 satisfying

$\epsilon_1$5

Within the training interval, return-point measures in parameter space and particle-position space remain nearly zero; outside it they become $\epsilon_1$6 or finite. Specifically,

$\epsilon_1$7

is $\epsilon_1$8 inside $\epsilon_1$9 and $y(t)={\cal W}[H](t)$0 outside, with the same qualitative behavior for

$y(t)={\cal W}[H](t)$1

The proposed mechanism, Gradient Discontinuity Learning, attributes this memory to sign-changing perpendicular gradient jumps across Type-2 discontinuity surfaces in parameter space (Zu et al., 1 Sep 2025).

These cases separate three regimes of constitutive memory. Exact RPM follows from order-preserving hysteretic structure; quasi-RPM emerges from training-induced polarization even when strict no-passing fails; and parameter-space return-point memory can emerge from cyclic optimization rather than from the physical constitutive response itself. This suggests that constitutive RPM is best understood as a family of history-dependent return maps rather than a single microscopic mechanism.

6. Scope, misconceptions, and broader significance

A common misconception is that any closed mechanical loop reflects viscoelastic memory. The VACNT and knitted-fabric studies argue otherwise. VACNT foams showed no stress relaxation, no creep, and rate invariance over three decades, leading to a rate-independent frictional interpretation rather than viscoelastic fading memory (Gupta et al., 16 Sep 2025). Knitted fabrics exhibited stress relaxation $y(t)={\cal W}[H](t)$2, much slower than the loading/unloading timescale, and their hysteresis deviated from the two standard models of hysteresis that usually apply to solid-state materials, viscoelasticity and plasticity (Dresselhaus et al., 1 Dec 2025).

Another misconception is that RPM is synonymous with no-passing. Classical sufficient conditions do use partial order, no-passing, and adiabaticity, but loop-RPM can be formulated directly from the dynamics-induced orders $y(t)={\cal W}[H](t)$3 and $y(t)={\cal W}[H](t)$4 without requiring a globally valid intrinsic order. In that broader setting, RPM rigidly constrains intra-loop structure while permitting substantial freedom in inter-loop transitions, including long transients and subharmonic response (Mungan et al., 2018). Conversely, exact microstate return under componentwise monotone multi-field driving shows that RPM can persist in vector-valued drive spaces, with monotonic protocols forming a commutative abelian sector and non-monotonic protocols generally becoming non-commutative (Croce et al., 22 May 2026).

Constitutive RPM is also not restricted to magnetic analogies. The experimental systems already span nanostructured foams and loop-based textiles, while the knitted-fabric interpretation proposes yarn contacts in entangled regions as bistable elements and suggests that crochet, torus-knot tesselations, or even woven fabrics with intermittent contacts may exhibit similar memory. In that picture, tuning loop topology or knit tension changes memory through the density $y(t)={\cal W}[H](t)$5 of effective hysterons (Dresselhaus et al., 1 Dec 2025). In VACNT arrays, the same constitutive memory enables tunable wave speed and is connected to amplitude-dependent shock limiters, elastodynamic lensing, and passive matched filtering for mechanical analog computing (Gupta et al., 16 Sep 2025).

At a more abstract level, the significance of constitutive RPM lies in its status as a robust constraint on reachable states. In Preisach-type descriptions, the set of reachable states on the main loop, $y(t)={\cal W}[H](t)$6, is equal to the number of increasing subsequences contained in the permutation $y(t)={\cal W}[H](t)$7. In state-space realizations, the memory can be compressed to a reduced sequence of reversal amplitudes $y(t)={\cal W}[H](t)$8. In multi-field systems, bounded excursions inside a hyperrectangle recover the same exact microstate when the corner is revisited. These formulations all encode the same constitutive principle: once the relevant extrema are specified, intermediate histories inside those bounds do not generate independent memory traces (Terzi et al., 2020).