The story begins in 79 C.E., when the eruption of Mount Vesuvius buried the Roman town of Herculaneum under volcanic ash, entombing a villa library that remains the only intact library known from antiquity. The pyroclastic heat carbonized more than a thousand papyrus scrolls, preserving them in theory but rendering them so brittle that early attempts to physically unroll them, beginning after their discovery in 1752, mostly produced heaps of ashes. Italian authorities eventually halted such efforts. Of the scrolls that were successfully opened, all proved to be texts unknown to scholars, many written by the Epicurean philosopher Philodemus, who lived in Herculaneum a century before the eruption. The surviving scrolls and fragments, roughly 1,800 in number, are now housed in Italy, France and England, and each one represents a book directly from Roman intellectual circles rather than a later medieval copy.

Douglas Seiler, an affiliate of Berkeley SETI and a retired professional who once worked in real estate, banking and invention, conceived the project after hearing about the scrolls while contributing to Panoseti, a new Berkeley telescope designed to search for laser signals from intelligent life in the galaxy. Seiler painstakingly sourced authentic papyrus and reed pens from Egypt and traditional lampblack ink from Japan, then hired high school students to inscribe the material with lines from Star Wars, the Bible and a quote from the 1960s science fiction television series The Outer Limits. He then rolled the papyrus into scrolls and carbonized them in his home laboratory by sealing them in a semi-sealed steel container with very little oxygen and heating it in a high-temperature furnace. The oxygen-poor environment replicated the charring of Vesuvius without reducing the material to dust, producing modern analogues of the ancient artifacts.

The central hypothesis was deceptively simple. Most ancient ink was made from soot, water and a binder such as gum Arabic, meaning that carbon ink sits on carbonized papyrus with almost no contrast in an X-ray image. But if the ink contained lead — a heavy metal that strongly absorbs X-rays — the letters should light up dramatically in a computed tomography scan. With collaborators including retired Berkeley chemists David Kreimer and Elena Kreimer, formerly manager of the College of Chemistry’s Microanalytical Facility, Seiler added calibrated amounts of lead nitrate to lampblack ink to create inks with different lead concentrations. Children of friends then wrote passages on fresh scrolls using traditional reed pens and the lead-spiked inks, and papyrologists Leah Packard-Grams and Jesse Obert of Berkeley’s Archaeological Research Facility scanned the scrolls to confirm the lead concentrations before carbonization.

To image the burned scrolls, Seiler enlisted Jake LaManna, a physicist at the Center for Neutron Research at NIST in Gaithersburg, Maryland, who created a three-dimensional X-ray CT scan of a charred scroll using the center’s laboratory X-ray source. LaManna jury-rigged a stand for the scroll and rotated it in the X-ray beam, recording thousands of slices that a computer assembled into a 3D rendering. Because lead absorbs up to 25 times more X-rays than the charred papyrus, the leaded ink stood out as bright spots against the darker paper, much like an overexposed region of a photographic negative. Ink with a lead concentration as low as 25 micrograms per square centimeter was easily detected. If you look at the images, Seiler said, the letters lit up like a Christmas tree.

A lucky coincidence accelerated the analysis. Michael Cyrus Daugherty, a postdoctoral fellow at NIST, had developed a program to digitally unroll the CT scans of the jelly-roll electrodes inside lithium-ion batteries. LaManna handed him the scroll data, and within a couple of days, after only minor modifications to account for the uneven and changing thickness of the papyrus along its length, Daugherty returned with examples of the scroll successfully unrolled in software. The demonstration showed that the same computational toolkit being applied to industrial imaging can be repurposed for the humanities, and that model scrolls offer a low-risk way to develop and refine unrolling algorithms without ever endangering the priceless originals.

The work builds directly on the ongoing Vesuvius Challenge, launched in 2023 by Brent Seales of the University of Kentucky together with venture capitalists, with a $700,000 Grand Prize offered to the first person to decipher four passages of at least 140 characters each from two Herculaneum scrolls scanned by X-ray CT. Within five months, two independent participants using artificial intelligence deciphered the first word, the Greek term for purple. In 2024, the same team used AI to read 15 full columns — less than a tenth of a scroll owned by the Institut de France — revealing an Epicurean philosophical treatise on perception and pleasure. In 2026, Seales’ team decoded the remains of scroll PHerc. 1667, nearly destroyed by earlier physical unrolling attempts, which appears to be a commentary on Stoic philosophy and may date from the second or third century B.C.E., possibly one of the oldest scrolls in the collection. All of these texts were extracted by virtually unrolling the scans and applying machine learning to detect extremely faint texture and morphological differences associated with carbon ink.

This is precisely where the new results could change everything. Lead-free carbon ink produces such a weak signal that most scrolls resist automated reading. Leaded ink, by contrast, produces a signature LaManna estimates is up to 25 times brighter than the surrounding papyrus. Critically, the researchers also demonstrated that an inexpensive handheld X-ray fluorescence scanner can detect leaded ink in a burned scroll, offering a simple screening method to identify which of the 1,800 surviving scrolls would be most amenable to X-ray CT decipherment. No one has systematically searched the Herculaneum collection for leaded ink, although at least one researcher previously found lead in the ink of a fragment left over from the early physical unrolling campaigns. Encouragingly, Packard-Grams used the same handheld technique on the Tebtunis papyri at UC Berkeley — fragments dating from 300 B.C.E. to 300 C.E. excavated in Egypt 126 years ago — and confirmed that some scrolls, generally those written after the first century C.E., contain lead or copper in their ink. If some Herculaneum papyri similarly contain lead, future CT scans can be tuned for sensitivity to that element, mapping the letters far more clearly as the scrolls are virtually unrolled.

The stakes, researchers argue, could hardly be higher. Almost everything from antiquity has been destroyed, and the Herculaneum scrolls are autograph books from the ancient world, frozen in time, rather than copies transmitted through Renaissance scribes. Packard-Grams, a Berkeley archaeology graduate student who consults as a papyrologist and now translates Greek and Egyptian writing in Berkeley’s Center for the Tebtunis Papyri, called the value of the project practically unquantifiable, describing the stakes as the largest in the history of Greek literature. She also hopes that ink chemistry, including any metals added deliberately or accidentally, could one day serve as a signature identifying the author of a specific text. Seiler, who funded the research and is first author of the paper alongside LaManna, Daugherty, David Kreimer, Michael McOsker of University College London and Jens Dopke of the Rutherford Appleton Laboratory, argues that a systematic lead analysis of the remaining papyri should begin now. His informal advisor, Berkeley nuclear engineering professor Karl van Bibber, observed that the gentleman scientist, often imagined as a figure of a bygone era, is alive and well in Berkeley. LaManna, for his part, wants to push further, studying ink composition beyond lead and building robust training datasets to strengthen the algorithms that will one day read the real scrolls. The model scrolls, Seiler noted, prove the concept: if there is lead in those ancient rolls, the letters will be there to see, and someone is going to read them.

Subject of Research: Virtual unrolling and X-ray tomography of carbonized papyrus scrolls from the Herculaneum library using lead-detectable inks and AI-based decipherment.

Article Title: To read 2,000-year-old burned papyrus scrolls, scientists make and burn their own

Article References: To read 2,000-year-old burned papyrus scrolls, scientists make and burn their own. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Herculaneum scrolls, papyrus, Vesuvius, X-ray CT, lead ink, Vesuvius Challenge, papyrology, virtual unrolling, X-ray fluorescence, artificial intelligence, PLOS ONE, ancient texts

The Herculaneum papyri were discovered in the ruins of the town of Herculaneum, near Naples, Italy, and they represent one of the most extraordinary archives of ancient thought ever found. When Vesuvius erupted in 79 CE, the scrolls were covered by an estimated 65 to 70 feet of rock and ash. The extreme heat of the eruption carbonized the papyrus, transforming the delicate rolls into brittle, blackened cylinders that crumble at the slightest touch. While scholars have succeeded in opening and reading some of the scrolls, revealing previously unknown writings by Epicurus and other ancient thinkers, many more have proved simply too fragile to study by physical means. For centuries, the unopened scrolls have sat in collections as tantalizing, inaccessible objects.

In recent years, however, technology has begun to change the calculus. Artificial intelligence and X-ray tomography have allowed researchers to virtually unroll some of the scrolls and recover portions of their text without ever physically touching them. The technique involves scanning a scroll with X-rays in three dimensions and then using software to reconstruct and read the hidden writing layer by layer. Yet the method faces a fundamental physical obstacle: both the ink and the papyrus are made largely of the same material, carbon. Because the writing surface and the writing itself are chemically so similar, X-rays often have great difficulty distinguishing the letters from the background, and entire passages can remain effectively invisible even in the highest-resolution scans.

The turning point described in the new study comes from an unexpected discovery about the chemistry of the ancient ink itself. Some of the letters in fragments of the Herculaneum scrolls have been found to contain lead. Lead is a heavy metal that stands out vividly in X-ray imaging, and since X-rays can far more easily distinguish lead from papyrus than carbon from carbon, the presence of even small amounts of the metal could make text legible. The authors of the new paper therefore suggest a triage strategy: scan the surviving scrolls for lead first, and then attempt virtual unrolling on those scrolls that contain it, prioritizing the ones most likely to yield readable text.

Testing such an idea on the genuine Herculaneum scrolls themselves was not possible, since the irreplaceable artifacts cannot be subjected to experimental risks. Instead, the team did something remarkable: they recreated carbonized scrolls in the laboratory. The researchers wrote on new papyrus using ink prepared with various concentrations of lead, then heated the finished scrolls in a high temperature furnace until they carbonized. The result was a set of model scrolls that mimicked the tight packing of layers and the waviness of the ancient charred artifacts, replicating the very conditions that make unrolling algorithms struggle with the real thing.

The experiments delivered encouraging results on two fronts. First, X-ray fluorescence, a technique that maps the elemental composition of a sample, was able to detect lead in the recreated scrolls at each of the lead concentration levels the team tested. This means that even relatively modest amounts of lead in the ink should be detectable in a survey scan, validating the idea that lead detection could serve as a practical screening tool for the museum and library collections of unopened Herculaneum scrolls. A scroll that lights up with a lead signature would be flagged as a promising candidate for more intensive imaging.

Second, the team demonstrated that the text itself could be recovered. Using X-ray tomography combined with a custom software program, the researchers were able to re-read some of the words they had written on the carbonized model scrolls before charring them. In effect, the full pipeline from carbonization to detection to virtual unrolling to machine-assisted reading was exercised end to end. The reconstructed letters emerged from scans of a scroll that, to the naked eye, looked like an unremarkable blackened rod, precisely the form that has defeated conservators since the eighteenth century.

Beyond the immediate promise of reading more of the surviving scrolls, the laboratory-made artifacts serve a second, equally important purpose. Because the researchers know exactly what is written inside their recreated carbonized scrolls, they can use them as ground-truth training data for the algorithms that attempt to decipher real scrolls. Virtual unrolling depends on software that must infer the geometry of tightly packed, warped, and fused layers of papyrus, and that task becomes dramatically harder when the internal structure is unknown, as it is with the ancient originals. The model scrolls offer a rare test case in which the correct answer is already known, allowing developers to measure how well an algorithm performs and to refine it systematically. Success on the recreated scrolls, the team argues, could translate into improved success rates when the same algorithms are turned loose on the ancient texts.

The implications reach well beyond a single library. The Herculaneum papyri are the only intact library known to have survived from antiquity, and scholars believe many of the unopened scrolls may contain works by Epicurean philosophers, and possibly texts otherwise lost to history entirely. Every previously unknown book recovered from the collection has reshaped understanding of ancient philosophy, literature, and science. A reliable way to identify which scrolls are most likely to yield readable text could transform the pace of discovery, focusing expensive and time-consuming imaging campaigns on the artifacts with the highest probability of success. Douglas Seiler captured the spirit of the achievement in a characteristically understated remark: “It’s amazing what you can get electrons to do.”

The work, conducted with support from the Department of Commerce Radiation Physics Division and the Center for Neutron Research at NIST, demonstrates how physics, chemistry, and machine learning can converge on one of archaeology’s oldest unsolved problems. The recreated scrolls are published openly in PLOS One, and the freely available article provides full technical detail for other laboratories to build upon. For the curators who guard the carbonized library and for the scholars who have spent lifetimes waiting to read it, the message of the new study is straightforward: the combination of lead-sensitive X-ray fluorescence screening, tomographic imaging, and increasingly capable unrolling algorithms offers a realistic path into the interiors of scrolls that have kept their secrets since the afternoon the volcano buried them. The long-lost words of Herculaneum may not remain lost for much longer.

Subject of Research: Experimental recreation of carbonized Herculaneum papyrus scrolls to test lead-based X-ray imaging and virtual unrolling

Article Title: We may soon be able to read long-lost ancient scrolls damaged by eruption of Mt. Vesuvius

Article References: We may soon be able to read long-lost ancient scrolls damaged by eruption of Mt. Vesuvius. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Herculaneum papyri, Mt. Vesuvius, carbonized scrolls, X-ray tomography, X-ray fluorescence, lead ink, virtual unrolling, machine learning, papyrus, PLOS One, ancient texts, Pompeii eruption