Reading Without Unrolling: Virtual Unwrapping and Lead-Ink Imaging of Herculaneum Papyri

The PLoS ONE work takes a different-element approach to the weak carbon-versus-carbon contrast: the researchers sourced papyrus and reed pens from Egypt and traditional lampblack ink from Japan, inscribed papyri with inks containing various concentrations of lead, rolled them up and placed them in a high-temperature furnace to replicate the roasting the ancient scrolls underwent at Herculaneum, then scanned the still-rolled burnt scrolls as thousands of slices in the Neutron and X-ray Tomography system at the US National Institute of Standards and Technology to build a 3D rendering, and virtually unrolled them with a custom version of a program originally designed to unroll scans of thin-film coils inside lithium-ion batteries. Earlier work often struggled to spot letters because standard CT techniques assess carbon-based materials and both ink and papyrus contain carbon; this study shifts detection toward lead in the ink and turns readability into a repeatable experimental question using recreated samples. The evidence comes from researcher-made carbonized scroll surrogates and a purpose-built scan-reconstruct-unwrap pipeline, i.e. a feasibility demonstration; the news story also quotes Thomas Coward, a classicist at the University of Nottingham who was not involved in the PLoS ONE work, on the value of overcoming ink-detection problems.

The arXiv paper reports complete virtual unwrapping and complete reading of the unopened roll PHerc. 1667: using high-resolution phase-contrast micro-CT on the BM18 beamline at the European Synchrotron Radiation Facility together with improved computational unrolling and machine learning, it reconstructs 22 columns or column-equivalents over approximately 860 cm2 of preserved writing surface, and describes PHerc. 1667 as the first Herculaneum papyrus to be fully digitally unrolled and read for extended scholarly study without physical opening. The paper states that previous work on sealed Herculaneum scrolls recovered only localized readings or limited surface regions, whereas this work extends the pipeline to the preserved writing surface of an intact roll under explicit coverage and papyrological-review criteria. The transcription was produced and interpreted at column level by eight qualified papyrologists, with acceptance requiring that the sampled surface be geometrically tight to the papyrus layer in the CT volume, that letterforms remain consistent across repeated renderings or inference passes, and that the transcription be independently endorsed or resolved by consensus; the complete unwrapping amounted to 31 wraps and 1231 cm2 of papyrus surface, using semi-automated segmentation tooling with roughly 25 hours of manual annotation per wrap.

In the same paper, PHerc. 139 yields title evidence read as Philodemus, On Gods, Book 8; the authors note that until now only the first book was known, preserved in PHerc. 26, so this discovery shows for the first time that On Gods comprised at least eight books. The book number is written on the same line as the title, unlike the more commonly attested two-line practice in Herculaneum titles, and a similar layout is documented in the title of the unopened roll PHerc. 172. This extends the payoff of virtual unwrapping from running prose to library metadata, namely author, work title and book number, so a sealed book can be identified even when continuous prose reading remains incomplete. The evidence is the author, work-title and book-number sequence recovered in a title-bearing region of the reconstructed surface; the paper notes the title was written by the same scribe who copied the text, and that the presence of Philodemus, by far the most extensively attested Epicurean philosopher in the Herculaneum library, allows assignment to the 1st century BC to 1st century AD with a terminus ante quem of 79 AD.

The paper treats PHerc. Paris 4 as an independent imaging-validation case: in the optimized high-resolution BM18 scan, thanks to phase retrieval, ink-bearing strokes appear as bright, surface-associated deposits with apparent cross-sectional thicknesses of approximately 10-20 micrometres, can be segmented in three dimensions and projected onto the flattened surface, where they coincide with the Vesuvius Challenge Grand Prize Banner region. This supplies a physical check on surface-conditioned ink recovery, showing that under at least one optimized scan regime and ink preservation state the model's signal corresponds to directly visible surface-associated deposits rather than only to rendering-based inference. The evidence is ink directly visible and three-dimensionally segmentable in the same optimized tomographic volume, spatially coinciding with the first multi-line reading from a still-rolled Herculaneum papyrus.

This evidence set speaks to two audiences: humanities scholars interested in ancient texts and library history, and engineering readers interested in non-invasive imaging and machine-learning pipelines. The paper states explicitly that the workflow does not imply that all sealed Herculaneum rolls are automatically readable, since performance depends on scroll general preservation state, scan quality, layer separation, deformation, local ink contrast and surface preservation. PHerc. 1667 itself is only a portion of a scroll, about 8 cm in height out of the usual 19-24 cm of a full roll and 2 cm in diameter out of 4-6 cm for a full roll, and it had undergone repeated unsuccessful physical opening attempts; its title may have been located in the now-lost upper portion, so authorship and title cannot be established with certainty. The PLoS ONE lead-ink route was validated on recreated samples, and its transfer to real holdings remains to be examined.

Worth watching: how the lead-ink detection route performs on real Herculaneum scrolls and how signals differ across ink recipes and degradation states; how the geometric bottleneck (mergers, holes and sheet switches in highly compressed regions) and the radiometric bottleneck (weak and variable ink segmentation across ink recipes and local degradation states) ease as scan conditions improve; whether the untranscribed material in PHerc. 1667, restricted to the surviving portions of the first three text columns amounting to 33 cm2, can be securely read with better imaging; and whether title and author metadata can be recovered systematically across more sealed rolls. In addition, the loaded arXiv text is the full paper, but the external news page is truncated at a subscription prompt, so some details mentioned in the story cannot be further checked from that page.