Using Raman spectroscopy to estimate the dates of carbon-based inks from Ancient Egypt

Abstract

In the first study of its kind, we previously investigated Raman spectra of 17 papyri from the Columbia University Libraries collection, written in carbon-based black ink with well-established dates between 400 BCE and 1000 CE. Using a simple two-peak spectral model for the Raman spectrum of carbonaceous materials, we discovered small but statistically significant systematic variations in the spectral parameters with manuscript date. The purposes of this report are to present: (1) more-powerful statistical examinations of the correlation based on the two-peak spectral model and on a new three-peak model which includes spectral effects for amorphous carbon; (2) new data for additional ancient manuscripts, broadening the investigation; (3) a method for potentially exploiting the observed correlation to predict the dates of undated manuscripts of similar character; and (4) a hypothesis as to the underlying chemical basis of the observed spectral changes. We apply both linear and multiple linear regression analyses to the spectral models. We show that the three-peak model fits the experimental data much better than the two-peak model. Through the use of multiple linear regression, the three-peak model reduces significantly the overall correlation error. We find that our ability to predict dates for manuscripts that are similar to those of the study over the date range 400 BCE to 1000 CE is limited by intrinsic variations in the character of the pigments to about ± 118 years with 68% statistical confidence (or ± 236 years with 95% confidence). We confirm the observed correlation and prediction capabilities through a blind study of six additional ancient manuscripts and two late 19th- or early 20th-century fakes in the Columbia University collection. The data suggest that chemical degradation of amorphous carbon and chemical modification of graphitic crystals may both contribute to the observed spectral differences. This would also account for the magnitude of the variance in spectral character we observe, which possibly reflects differences in pigment manufacture processes as well as the effects of differences in chemical micro-environments after writing. The chemical degradation hypothesis receives additional support from previously unreported examination of two manuscripts from the collection of the Brooklyn Museum. The older of these, dated through conventional papyrological methods to 2323-2150 BCE, deviates from the linear regression models, but in a way consistent with chemical (i.e., non-linear) degradation. For comparison, we report Raman scattering for 8 different modern inks using the same protocol we developed for the ancient inks. Using statistical techniques based on Gaussian Mixture Modeling, we demonstrate differences between the spectral character of the inks of our base study and those prepared from modern reference pigments. This allows us to define a mathematical quantity that can distinguish between spectra of pigments that are spectrally similar to the ancient pigments studied here and those that are not. This new and non-destructive method produces a statistically robust indication of the antiquity and rough period (e.g., Ptolemaic, Roman, etc.) of the inks in question, if not the precise date. We hypothesize that the potential for greater precision is likely limited by the intrinsic variations in the spectral characteristics of the inks from manuscript to manuscript.