The Crease Is Not the Whole Memory

Imagine flattening a crumpled piece of paper. The damage is easy to read. Ridges cross the surface; corners refuse to lie down. Something happened to this sheet, and the marks remain.

That is one meaning of memory. The stranger one requires a measuring instrument.

In experiments with thin plastic film, researchers have found histories that emerge not simply from the shape left behind, but from what the sheet does when it is pushed again. One protocol reveals a signature of how long a previous squeeze lasted. Another reveals the reach of earlier compression—and allows smaller excursions to retrace their way home.

The distinction matters. A permanent scar and a recoverable mechanical response are not the same achievement.

The force goes the wrong way

Start with a crumpled ball of Mylar between two parallel plates. Bring the plates closer and hold them there. The film pushes back, but its force gradually diminishes as the compressed structure relaxes.

Now wait, then open the gap partway. Not all the way back: the ball remains compressed.

The immediate response is unsurprising. With some compression released, the force drops. What happens afterward is the interesting part. With the plates held still, the force slowly increases, reaches a peak, and only then resumes falling.

This is the sequence reported by Yoav Lahini and colleagues in their 2017 study of crumpled sheets and elastic foams. My detailed reading here uses their complete author manuscript, not the final journal PDF. The Mylar had been repeatedly crumpled, opened and flattened before testing; that preparation made an otherwise variable response more reproducible. This was not an experiment on an arbitrary fresh paper ball. [1]

The delayed rise could last minutes or hours. More importantly, its timing depended on what came before. For a fixed partial-release step, the time from that release to the peak scaled approximately with the earlier waiting time. Waiting longer during the first squeeze left a later signature in the second response. Change the size of the release, however, and the relationship changes too. The sheet is not a universal stopwatch.

There is a clean way to see why this counts as memory rather than merely slow motion. Consider a force value reached once while the curve rises and again while it falls. The plate gap is the same. The measured force is the same. Yet the subsequent force trends have opposite signs.

Those two readings do not specify the whole state. Something inside the material differs because of its history.

The authors describe a possible explanation using many relaxation processes with different rates. After the first squeeze, fast processes have traveled much farther toward their new conditions than slow ones. The partial release changes the destination again. Some contributions now evolve one way, others the opposite way; their changing balance can make the total force rise before it falls.

This broad spectrum of rates is a useful phenomenological model, not a photograph of the hidden machinery. Its simplest form also does not quantitatively predict everything: the Mylar comparison needed an empirical scale correction, and the foam's relaxation was more complicated. Explaining how a reversal can happen is not the same as uniquely identifying every process that makes it happen. [1]

A smaller journey can come back

A later experiment asks a different question. Instead of leaving a ball confined between plates, Dor Shohat, Daniel Hexner and Lahini crumpled Mylar sheets and opened them again. They held opposite edges and repeatedly moved one edge inward and outward. [2]

This preparation makes a different kind of history legible.

Plot force against edge displacement and the outward and return journeys do not follow the same path. They form a loop: the sheet's response depends on which way it is being driven. The curve is also interrupted by abrupt jumps. After several cycles, much of that pattern becomes repeatable, although slow creep and drift remain.

Now reduce the excursion. Compress less far than before, then return. The sheet follows a smaller loop inside the trained one. Restore the original range and the earlier response pattern reappears.

Push past the previous maximum, however, and the story changes. When the experiment returns to the old range, its former pattern is no longer recovered. The new extreme has altered the mechanical response. In this regime, the sheet carries a readable trace of the largest displacement in its driving history. [2, Fig. 1]

The researchers also used a backtracking sequence that produced loops nested inside loops. Returning to earlier reversal points closed the corresponding loops; completing the enclosing excursion recovered its parent response. This is called return-point memory.

It is an unusually concrete use of the word. The memory is read in a recovered force pattern under a specified sequence—not inferred from the fact that a crease still exists. Nor does it mean every possible history survives. A new outer extreme can reorganize the response, and experimental recovery is highly accurate rather than perfectly immune to creep. [2, Fig. 2]

Little switches in a damaged landscape

Where is that cyclic memory kept?

The 2022 study gives the question a structural handle. Three-dimensional scans taken before and after individual force jumps showed localized regions snapping between configurations. A small facet or a feature near a ridge junction could buckle one way and then the other. Crucially, the forward and reverse switches occurred at different thresholds. Each such region behaved like a tiny two-state element with its own hysteresis. [2, Fig. 3]

These were not isolated switches glued to a passive backing. Using a local probe, the researchers found that switching one region could make a neighbor easier—or harder—to switch. The configuration of the surrounding sheet changed the threshold. [2, Fig. 4]

That is a richer account than “every crease is one stored bit.” The memory involves the collective configuration of interacting geometric features. And once the creased framework exists, changing the response need not mean deliberately writing a new crease each time. Test loads were much lower than the loads used during crumpling; the researchers expected negligible new plastic deformation during those measurements. That is a qualified expectation, not microscopic proof that no fresh damage occurred anywhere.

There is another experimental detail worth keeping straight. The published global force–displacement curves used the cylindrical setup; the topographic measurements used the flat setup. The authors report the same characteristic memory behaviors in the flat setup, but the two measurements are not one identical geometry. [2, Materials and Methods]

A simplified network of interacting bistable springs reproduced the main cyclic-memory features. It shows that those ingredients can suffice. It does not reconstruct every ridge in the real sheet, and it does not settle why such accurate return-point memory survives the competing interactions. [2]

Two memories, not one solved mystery

It is tempting to close the story by saying the little switches explain everything: the returning loops, the waiting-time peak, the stubbornness of the flattened sheet.

The papers do not earn that ending.

The studies share an author and research lineage; they are not independent-laboratory replications. More fundamentally, they measure different things. One reads elapsed waiting history from a delayed force maximum. The other reads deformation history from cyclic response patterns. The 2022 paper explicitly leaves open whether its interacting-instability framework can also explain the earlier aging and waiting-time effects.

I began with the obvious version of a sheet's memory: the marks are still there. What changes the picture is that the marked structure can possess several histories that become readable only when you ask the right mechanical question.

None of this requires the film to recall, recognize or feel anything. It requires a past intervention to leave an internal difference that changes a later response.

The next question is harder than “does it remember?” It is: which part of the past can this particular test recover—and what remains hidden even when the test works?

Sources and reading limits

[1] Y. Lahini, O. Gottesman, A. Amir and S. M. Rubinstein, Nonmonotonic Aging and Memory Retention in Disordered Mechanical Systems, Physical Review Letters 118, 085501 (2017), DOI. Substantive reading: complete author manuscript, arXiv:1608.02429v3, especially Figs. 1–5 and the two-step protocol/model discussion. The final journal PDF was not accessible in this study.

[2] D. Shohat, D. Hexner and Y. Lahini, Memory from coupled instabilities in unfolded crumpled sheets, PNAS 119, e2200028119 (2022), published full text, DOI. Read the complete main narrative, methods and main figure captions. Supporting appendix, movies and figure images were not inspected. No new crumpling, listening, imaging or raw-data analysis was performed for this essay. The experiments used Mylar, not the imagined paper of the opening.