Lab earthquakes show when precursory slip speeds up before rupture

In 62 laboratory earthquakes on 65 cm plastic faults, slow slip sped up clearly before rupture on the smoothest fault only under the lowest side load (1 MPa), and faded as the load rose, a lab-scale clue that does not let anyone predict real earthquakes.

Peer reviewed · Laboratory experiment · Analogue material (PMMA plastic) · Accelerated article preview

Whether a fault shows a detectable build-up of slow slip before it breaks may depend on how smooth it is and how hard it is squeezed, according to laboratory experiments published in Nature Communications on 14 November 2026. Peng Dong, Kaiwen Xia and colleagues at China University of Geosciences (Beijing), the University of Tokyo and Georgia Tech made 62 lab earthquakes on plastic faults. On the smoothest fault, slow slip accelerated before rupture under low stress but not at higher stress. These are centimetre-scale analogue faults, not real ones, and the authors warn that extrapolating to nature requires caution.

Evidence at a glance

Is this a big deal?

A careful step in a long argument, not a breakthrough for public safety. Unlike many earlier experiments, which inferred slip from how far the loading machine moved, this one filmed slip along the whole fault through every cycle, so the team could see where slow slip started and how fast it spread. But one material, three surfaces and three loads make a narrow window. The real test is whether GPS records from subduction zones match the pattern, and the study itself does not check that.

What scientists found

“Precursory slip” means a fault moving slowly and quietly before it breaks in an earthquake. In the lab, the team measured it as moment release, a number that adds up how much the fault slipped and over how much area. If that release rate climbs before rupture, the fault has given a measurable precursor.

They sorted each lab quake into three groups. A pronounced speed-up meant the peak release rate was at least twice the early rate; a moderate one meant between one and two times; and “no acceleration” meant the rate stayed flat. At the lowest side load, pronounced speed-ups dominated on all three surfaces. At higher loads, the smooth fault switched to mostly flat curves, while moderate speed-ups were common on the rougher faults.

In one smooth-fault cycle at 1 MPa, a patch of slow slip appeared near the middle of the fault about 9 seconds before rupture and spread in both directions at an average of 13.6 mm per second. That is roughly 1 km a day, which the authors note is comparable to slow slip seen before subduction-zone earthquakes. The release rate climbed from 1.7 to 6.65 newton-metres per second, almost a fourfold rise. The edge of the slow slip moved toward the spot where the quake later started, and the shear stress there rose by about 0.06 MPa. The authors read this as the slow slip triggering the rupture.

The smooth fault also put about half of its quiet, between-quake slip into the precursory phase at 1 MPa, similar to the rougher faults. At 2 and 3 MPa, that share fell to zero on the smooth fault but not on the rough ones.

How they found it

The team cut a 500 mm square sheet of 20 mm thick PMMA diagonally and pressed the two halves back together. The seam between them became a fault 650 mm long. They sandblasted the surfaces to three roughness levels, measured as root-mean-square height: about 0.1 micrometres (smooth), 1 micrometre (inter-rough) and 5 micrometres (rough). For scale, 0.1 micrometres is one ten-thousandth of a millimetre.

A hydraulic press held a constant side load of 1, 2 or 3 MPa, roughly 10 to 30 times sea-level air pressure, while vertical rams advanced at 1 micrometre per second. Stress built until the fault slipped, then built again, giving a sequence of lab earthquakes. A camera filmed a speckle pattern painted on the plastic at 5 frames per second, and software compared frames to work out how far each side of the fault had moved. Fifteen strain gauges logged stress 1,000 times a second, and seven more logged the rupture itself a million times a second.

Think of dragging a heavy rug across a floor. Before the whole rug moves, a ripple can run along it, and how easily that ripple travels depends on the floor and how hard the rug is pressed down. The analogy breaks down quickly: a fault is locked rock under enormous pressure, often with fluids in it, and the “ripple” is slip of micrometres in the lab and centimetres to metres in nature.

Why it matters to you

Nothing here changes what you should do about earthquakes this year. The U.S. Geological Survey states plainly: “Neither the USGS nor any other scientists have ever predicted a major earthquake.” A prediction, by its definition, must give the date and time, the location and the magnitude. Agencies instead publish probabilities and build early-warning systems, and the practical advice on what to do when the ground shakes is unchanged.

The study matters further down the line. Large subduction-zone earthquakes, where one tectonic plate dives beneath another, cause many of the world’s most damaging quakes and tsunamis. If scientists can learn which faults tend to give a slow, measurable build-up, they will know where extra monitoring is most likely to pay off.

What the evidence does not show

It does not show that real earthquakes can be predicted, and the paper does not claim that. Its own introduction notes that whether such precursory phases are common in nature is “actively debated”, which casts doubt on forecasting a large earthquake from slow slip alone.

It does not show that the same pattern holds in rock. The authors scale their 1 to 10 MPa range of fault stress to roughly 20 to 200 MPa in rock, because rock is about 20 to 30 times stiffer, and they call this a rough estimate that ignores other factors. All three surfaces sit at the smooth end of what lab faults usually use, so “rough” here is rough only compared with the other two.

It also cannot separate cause neatly. A rougher surface changes how stress and friction vary along the fault, and the authors say they “cannot fully decouple” roughness from that frictional patchiness. “No acceleration” means none the camera could detect: a final burst lasting milliseconds could still occur, and the reported slip-front speeds carry an uncertainty of about ±50%.

How this fits what we already knew

Field geophysicists have caught slow slip before several big quakes. Continuous GPS in Guerrero, Mexico, recorded a slow slip event that began in February 2014, two months before the magnitude 7.3 Papanoa earthquake of 18 April, starting next to the future quake’s origin, as Radiguet and colleagues reported in Nature Geoscience in 2016. A 2012 Science paper led by Aitaro Kato, a co-author of the new study, described slow slip propagating in the run-up to the 2011 magnitude 9.0 Tohoku-Oki earthquake in Japan, although a magnitude 7.3 foreshock complicates that case, as one referee of the new paper pointed out.

Many of these quakes struck on subduction megathrusts, the giant faults where plates meet, which earlier studies describe as relatively smooth. Several lab studies had instead found smooth faults giving less warning; a 2024 Nature Communications paper by Thomas Goebel and colleagues on Westerly granite, for example, linked low roughness to shorter precursory periods. The new experiments suggest both can be true: a smooth fault can hide its build-up under high stress and show it under low stress. The authors link that to subduction zones, where pore fluids at high pressure are thought to lower the effective squeeze on the fault. That link is their proposal, not something they tested. For how plates and the crust fit together, see our guide to the layers of Earth.

What the peer reviewers said

SciQuest has not yet obtained independent comment. The paper’s open peer review file gives outside views. Giuseppe Petrillo, one of three referees, called the dataset valuable and wrote that “the detectability of precursory slip remains one of the central unresolved issues in earthquake physics”, while asking for more caution when extending the results to megathrusts. Barnaby Fryer warned that “the smoothest sample is therefore only smooth relative to the other two”. An anonymous referee pointed out that most earlier experiments, including some on smooth faults, already show a little accelerating slip just before failure. The authors softened several claims in response.

What happens next

The obvious test is in the field: whether slow slip before real quakes is clearer where faults are smoother and fluid pressures higher. The authors argue that combining slow-slip records with foreshocks, fluid measurements and other signals may give more reliable precursory information than any one signal. Their data are deposited on Figshare, so other groups can re-analyse the 62 cycles, and the final typeset version of the paper is still to come.

In one sentence

On plastic lab faults, slow slip built up clearly before rupture on smooth surfaces only under light loading, which helps explain conflicting lab and field results but brings no one closer to predicting a real earthquake’s date, place and size.

Read next

• How to survive an earthquake: the safest place to be when it hits
• From crust to core: a detailed look at the layers of Earth
• What are moonquakes, and how do they differ from earthquakes?

Sources and disclosure

1. Dong, P., Xia, K., Kato, A., Obara, K., Peng, Z. & Peng, J. Geophysical conditions favoring precursory slip before major earthquakes. Nature Communications (2026), published 14 November 2026. SciQuest read the full accepted-manuscript text, Methods, figure legends and the Supplementary Information.
2. Transparent peer review file for the paper (CC BY 4.0; referees Giuseppe Petrillo, Barnaby Fryer and one anonymous referee).
3. U.S. Geological Survey. Can you predict earthquakes? (FAQ, accessed 15 November 2026).
4. Radiguet, M. et al. Triggering of the 2014 Mw7.3 Papanoa earthquake by a slow slip event in Guerrero, Mexico. Nature Geoscience 9, 829–833 (2016) (abstract).
5. Kato, A. et al. Propagation of slow slip leading up to the 2011 Mw 9.0 Tohoku-Oki earthquake. Science 335, 705–708 (2012) (bibliographic record).
6. Goebel, T. H. W. et al. A laboratory perspective on accelerating preparatory processes before earthquakes and implications for foreshock detectability. Nature Communications 15 (2024).

The authors declare no competing interests. The work was funded by China’s National Science and Technology Major Project (Deep Earth Probe and Mineral Resources Exploration), the National Natural Science Foundation of China and the Fundamental Research Funds of China University of Geosciences, Beijing.

Disclosure: This article was drafted with AI assistance and checked against the sources above. The graphic shows only categories and values stated in the paper; the fourfold rise and the air-pressure comparison are SciQuest’s arithmetic from the paper’s numbers. Spotted an error? See our corrections policy.