Empirico-Statistical Analysis of Narrative Material and its Applications to Historical Dating

Volume I

Empirico-Statistical Analysis of Narrative Material and

its Applications

to Historical Dating

Volume I:

The Development of the Statistical Tools

by

A.T. FOMENKO

Department of Geometry and Topology, Faculty of Mathematics,

Moscow University,

Moscow, Russia

KLUWER ACADEMIC PUBLISHERS DORDRECHT / BOSTON / LONDON

A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN 0-7923-2604-0 (Volume I)

ISBN 0-7923-2606-7 (Set of 2 Volumes)

Published by Kluwer Academic Publishers,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Kluwer Academic Publishers incorporates the publishing programmes of

D. Reidel, Martinus Nijhoff, Dr W. Junk and MTP Press.

Sold and distributed in the U.S.A. and Canada by Kluwer Academic Publishers,

101 Philip Drive, Norwell, MA 02061, U.S.A.

In all other countries, sold and distributed by Kluwer Academic Publishers Group,

P.O. Box 322, 3300 AH Dordrecht, The Netherlands.

Translated by O. Efimov

Artwork on the cover by the author

Printed on acid-free paper

All Rights Reserved

© 1994 Kluwer Academic Publishers

No part of the material protected by this copyright notice may be reproduced or utilized in any form or by any means, electronic or mechanical,

including photocopying, recording or by any information storage and

retrieval system, without written permission from the copyright owner.

Printed in the Netherlands

CONTENTS

Foreword Xl

Preface Xlll

Chapter 1. Problems of Ancient and Medieval Chronology 1

§1. The Global Chronological Diagram of Ancient and Medieval

History 1

1.1. The moon's elongation and R. Newton's conjecture 1

1.2. The Dark Ages and the Renaissance epochs 2

1.3. How to substantiate ancient chronology 3

1.4. Statistical dating methods: new possibilities 5

1.5. The duplication effect in ancient history and chronology 7

1.6. The global chronological diagram and the "modern

textbook" of ancient and medieval history 8

1.7. The "modern textbook", a composition of four

identical pieces 10

1.8. Certain corollaries and interpretations 11

1.9. What is to be done with the moon's elongation? 12

References 13

§2. Computation of the Second Derivative of the Moon's Elongation and Statistical Regularities in the Distribution of the Records of

Ancient Eclipses 15

2.1. Parameter D" and R. Newton's paper "Astronomical

evidence concerning non-gravitational forces in the

Earth-Moon system" 15

2.2. Available observations of ancient solar and lunar eclipses 17

2.3. A method of formal astronomical dating 18

2.4. The effect of shifting the dates of eclipses forwards 19

2.5. An example: three eclipses of Thucydides 20

2.6. An example: the eclipse described by Livy 22

2.7. An example: the eclipse described by Livy and Plutarch 23

2.8. An example: the evangelical eclipse described in the New

Testament in connection with the Crucifixion 24

2.9. The oscillation of a new graph of D" about one and the

same value. No nongravitational theories are necessary 25

v

VI

Contents

2.10. Three rigid "astronomical shifts" of ancient eclipses 28

2.11. The complete picture of astronomical shifts 31

2.12. The coincidence of the astronomical shifts with the three

basic chronological shifts in the global chronological

diagram 31

§3. Traditional Chronology of the Flares of Stars and the Dating

of Ancient Horoscopes 32

3.1. Ancient and medieval flares of stars. The star of Bethlehem 32

3.2. Astronomical dating of ancient Egyptian horoscopes 34

3.3. Astronomical dating of the horoscope described in the

Book of Revelation 35

References 37

Chapter 2. New Statistical Methods for Dating 39

§4. Certain Statistical Regularities of Information Density

Distribution in Texts with A Scale 39

4.1. Text with a scale. The general notion 39

4.2. Information characteristics (i.e., informative functions) of

a historical text. Volume function, name function, and

reference function 40

4.3. A theoretical model describing the distribution of

local maxima for the volume function of a historical text.

Primary stock. The information density conservation law 42

4.4. The correlation of local maxima for the volume graphs of

dependent historical chronicles. The surviving-stock graph 43

4.5. Mathematical formalization. The numerical coefficient

d(X, Y), which measures the "distance" between two

historical texts X and Y 44

4.6. Mathematical formulas for computing d(X, Y). Mathematical

corrections of the maxima correlation principle 47

4.7. Verification of the maxima correlation principle against

concrete historical material 49

4.8. A new method for dating historical events. The method

of restoring the graph of the primary and surviving

information stock 52

.

4.9. The discovery of dependent (parallel) historical epochs

traditionally regarded as different 53

4.10. The dynasty of rulers and the durations of their reigns as an

important informative function 54

4.11. Frequency distribution of the rules of kings who lived

from A.D. 1400 to 1800 and from 3000 B.C. to A.D. 1800 55

4.12. The concept of statistically parallel historical texts and epochs 57 4.13. The "written biography" or enquete-code of a historical

character 57

Contents Vll

4.14. A method of comparing the sets of informative functions

for two historical epochs 60

4.15. A computational experiment 61

4.16. The remarkable decomposition of the global chronological

diagram into the sum of four practically indistinguishable

chronicles 62

References 66

§5. A Method of Duplicate Recognition and Some Applications to

the Chronology of Ancient Dynasties 67

5.1. The process of measuring random variables 67

5.2. The distance between two random vectors 67

5.3. Dynasties of rulers. The real dynasty and the

numerical dynasty. Dependent and independent dynasties.

The small-distortion principle 68

5.4. Basic errors leading to controversy among chroniclers as to

the duration of kings' rules 69

5.5. The experimental frequency histogram for the duration of

the rules of kings 70

5.6. Virtual dynasties and a mathematical model for errors made

by the chronicler in measuring the rule duration 70

5.7. The small-distortion principle and a computer experiment 72

5.8. Pairs of dependent historical dynasties previously regarded

as independent 73

5.9. The distribution of dependent dynasties in the "modern

text book" of ancient history 73

5.10. Dependent dynasties in the Bible and parallel with European

history 74

References 75

§6. A New Empirico-Statistical Procedure for Text Ordering and Its

Applications to the Problems of Dating 76

6.1. The chapter generation 76

6.2. The frequency-damping principle 76

6.3. The method of finding the chronologically correct order of

chapters in a historical chronicle 77

6.4. The frequency-duplicating principle and the method of

duplicate recognition 79

6.5. The distribution of old and new duplicates in the Old and

New Testament. A striking example: the Book of Revelation 81

6.6. Duplicates of epochs in the "modern textbook" of ancient

history 83

References 87

V III

Contents

Chapter 3. New Experimental and Statistical Methods for Dating Events of Ancient History, and Their Applications

to the Global Chronology of Ancient and Medieval History 88

§7. Introduction. N .A. Morozov and Modern Results 88

§8. Problems of Historical Chronology 89

8.1. Roman chronology as the "spinal column" of European

chronology 89

8.2. Scaliger, Petavius, Christian chronographers and secular

chronography 90

8.3. Questioning the authenticity of Roman tradition.

Hypercriticism and T. Mommsen 93

8.4. Difficulties in the establishment of Egyptian chronology 94

8.5. Competing chronological versions. De Arcilla, J. Hardouin,

I. Newton and R. Baldauf 96

8.6. Tacitus and Bracciolioni. Cicero and Barzizza 97

8.7. Vitruvius and L. Alberti 98

8.8. "The chaos of medieval datings" (E. Bickerman). Medieval

anachronisms and medieval concepts of time 99

8.9. The chronology of the biblical manuscripts. L. Tischendorf 102

8.10. Vowels in ancient manuscripts 103

8.11. Traditional biblical geography 104

8.12. Problems of geographical localization of ancient events 105

8.13. Modern analysis of biblical geography 107

8.14. Ancient originals and medieval duplicates. Anachronisms

as a common feature in medieval chronicles 110

8.15. Names and nicknames. Handwritten books 112

§9. Astronomical and Mathematical Analysis of the Almagest 113

9.1. Morozov's analysis of the first medieval editions of the

Almagest 113

9.2. On the statistical characteristics of the Almagest. The

structure of the star catalogue 115

9.3. The accuracy of the Almagest's star coordinates 119

9.4. The problem of dating the Almagest from the individual

stars' proper motion 120

9.5. Halley's discovery of the stars' proper motion and the

Almagest 128

§10. Archaeological Dating Methods 131

10.1. Classical excavation methods 131

10.2. Numismatics 132

10.3. The dendrochronological method 133

10.4. The radiocarbon method 133

§11. Astronomical Dating. Ancient Eclipses and Horoscopes 136

§12. New Experimental and Statistical Methods of Dating Ancient

Events 140

Contents

IX

12.1. Introduction 140

12.2. Volume graphs for historical chronicles. The maximum correlation principle. Computational experiments and

typical examples 140

12.3. Method of recognition and dating the dynasties of ancient

rulers. The small-distortion principle 143

12.4. The frequency-damping principle. A method of ordering

texts in time 145

12.5. Applications to Roman and Greek history 148

12.6. The frequency-duplication principle. The duplicate-

discovery method 151

12.7. Statistical analysis of the complete list of all the names

mentioned in the Bible 152

12.8. Statistical analysis of the complete list of all parallel

passages in the Bible 155

12.9. Duplicates in the Bible 157

12.10. The enquete-code or formalized "biography" method 159

12.11. A method for the chronological ordering of ancient maps 160

§ 13. Construction of the Global Chronological Diagram and Certain

Results of Applying the Dating Methods to Ancient History 162

13.1. The "textbook" of ancient and medieval history 162

13.2. Duplicates 162

13.3. Dependent dynasties 163

13.4. The agreement of different methods 165

13.5. Three basic chronological shifts 166

13.6. Biblical history and European history 168

13.7. The beginning of "authentic" history in circa the 10th

century A.O. 173

13.8. The chronological version of Morozov and the author's

conception 174

13.9. The confusion between the two Romes 175

13.10. A universal mechanism which could lead to the chroniclers'

chronological errors 177

13.11. Scaliger, Petavius, and the Council of Trent. Creation of

traditional chronology 178

§14. The "Dark Ages" in Medieval History 180

14.1. Medieval Italy and Rome 180

14.2. Medieval Greece and Athens 187

14.3. The history of religions 193

14.4. Indian history and chronology 195

References 196

Index 203

FOREWORD

Today the methods of applied statistics have penetrated very different fields of knowledge, including the investigation of texts of various origins. These "texts" may be considered as signal sequences of different kinds, long genetic codes, graphic representations (which may be coded and represented by a "text"), as well as actual narrative texts (for example, historical chronicles, originals, documents, etc.). One of the most important problems arising here is to recognize dependent text, i.e., texts which have a measure of "resemblance", arising from some kind of "common origin". For instance, in pattern-recognition problems, it is essential to identify from a large set of "patterns" a pattern that is "closest" to a given one; in studying long signal sequences, it is important to recognize "homogeneous subsequences" and the places of their junction. This includes, in particular, the well-known change-point problem, which is given considerable attention in mathematical statistics and the theory of stochastic processes.

As applied to the study of narrative texts, the problem of recognizing dependent and independent texts (e.g., chronicles) leads to the problem of finding texts having a common source, i.e., the same original (such texts are naturally called dependent), or, on the contrary, having different sources (such texts are naturally called independent). Clearly, such problems are exceedingly complicated, and therefore the appearance of new empirico-statistical recognition methods which, along with the classical approaches, may prove useful in concrete studies (e.g., source determination) is welcome.

The present book by A.T. Fomenko, professor of pure mathematics, is mainly aimed at developing such new methods to be applied to recognizing dependent and independent narrative texts and dating them (with respect to texts with known reliable dates).

The author proposes a new approach to the problem of recognizing dependent and independent narrative (historical) texts based on the several new empiricostatistical models ( regularities) which he has discovered during his extensive statistical emperiments involving various quantitative characteristics of concrete texts, chronicles, originals, and so forth. Verification of these models (statistical hypotheses) on concrete chronicles confirmed the efficacy of the models and made it possible to put forward new methos for dating texts (more precisely, for dating events described in the texts).

The approach proposed in this book is nonstandard and requires attention and diligence on the part of the reader to new and probably unfamiliar logical constructions. At the same time, the basic ideas of the author seem quite natural

Xl

Xli

Foreword

from the viewpoint of modern mathematical statistics and can easily be included in the conceptual system of applied statisticians.

The author's scientific results and ideas are very interesting, and perhaps already today we may speak of the appearance of a new (and rather unexpected) scientific trend in applied statistics whose development is of undisputed importance. This book is a result of a tremendous amount of work done by the author and his colleagues, most of whom specialize in mathematical statistics and its applications.

Since the book is devoted to problems at the interface of several branches of science, the necessity of establishing contact between people of different professional backgrounds becomes obvious. Many concepts and terms customary to specialists in one branch require translation into the language of specialists in another field ofresearch. This should be borne in mind by representatives ofboth the natural sciences and the humanities. Such "difficulties of communication" are typical and must be successfully overcome with any mixed group of scientists working on joint problems. One may hope that many of the readers of this book willjoin to form such an interdisciplinary group in order to successfully continue the studies started here by this well-known mathematician.

Along with developing new empirico-statistical methods for dating events, the book also includes applications to the problem of modern scientific argumentation of the chronology of past events. One should clearly distinguish between the main statistical result obtained by the author (namely, the layered structure of the global "chronological map" and its representation as a "sum" of four layers) and its various interpretations and substantiations. Stating hypotheses and providing substantiations of results are beyond the scope of exact mathematical knowledge, and therefore special care should be taken in formulating conclusions concerning the possible structure of a new "statistical chronology of antiquity" . The author has repeatedly insisted upon the necessity of critical analysis and upon distinguishing between strictly established facts and hypotheses or interpretations concerning these facts.

The concepts proposed by the author are new, sometimes unexpected, and deserve extensive and thorough investigation.

The book is written at a high scientific level, it is a unique phenomenon in the scientific literature in the field of applications of mathematical statistics, and the reader will not remain indifferent to it. The book also enables us to get to know the engaging personality of its author-mathematician and investigator of history ....

I hope that, after perusing the first few pages of the book, the reader will be intrigued to read to the end with unabating interest. He will, at the very least, get to know an interesting set of scientific problems and, perhaps, will even engage in further investigations in this new and promising field of science.

Albert N. Shiryaev President of the International Bernoulli Society for Mathematical Statistics and Probability Theory

PREFACE

This book presents new empirico-statistical methods for the discovery of dependences between texts, on which we base our dating methods. As one of various possible applications, the datings (or dates) of certain events described in ancient and medieval texts are analyzed.

The problem of recognizing dependences (and dependent texts) arises in many branches of applied statistics, linguistics, physics, genetics, and so forth. For example, as applied to source research, the discovery of dependent texts with a common primary source or original (which may not have survived) is of considerable interest. On the other hand, it is useful to have an idea which texts may be called independent or are based on substantially different primary sources and archival data. Meanwhile, the concept of text itself can be treated extremely differently. We can consider a sequence of symbols, signals, codes (of various kinds; for example, genetic codes in DNA chains) as a text, where the general problem in the search for "dependent texts" consists of finding "similar" portions in a given long signal sequence, i.e., textual fragments "duplicating" each other.

Today, there are many methods for finding dependences of this sort. We suggest certain new empirico-statistical procedures which can prove useful both in analyzing narrative texts (such as annals and chronicles) and in studying biological codes to find so-called homologous fragments, and so on.

For the reader's convenience, we divide the contents into several "topics", which may be helpful in getting oriented in the material and in separating reliable statistical evidence from hypotheses. This devision is arbitrary in the sense that the highlights listed in the following are intimately related to each other. Therefore, it would be more correct to speak of the book's "fibres", rather than of its parts. The book's chapters receive different emphasis, and I will briefly describe this accentuation here. I hope that the reader will be able to relate each fragment of the book to some particular ''fibre'', and, in particular, to make out the author's attitude toward each fibre.

The first fibre. The problem of discovering statistically covert dependences and dependent texts is solved, to which purpose a number of new statistical models (or hypotheses) are formulated. They are then checked against sufficiently extensive experimental data consisting of concrete narrative texts like annals or chronicles. It turns out that the suggested models can be confirmed.

Xlll

XIV

Preface

In other words, we managed to discover interesting statistical regularities controlling the chroniclers' process of creating long narrative texts. The discovery of these laws is one of the principal results of our work. And on their basis, the methods for dating the events described are offered, for which the texts under investigation are statistically compared with those whose dating is undisputed. The methods are then verified against sufficiently extensive concrete material. We see that their application to texts describing the events from the 13th to the 20th century supports the efficiency of the method. Namely, the statistical datings obtained are consistent with those that had been known previously and were established by traditional methods. In particular, textual pairs originating from common primary sources, and known a priori as dependent between the 13th and the 20th century, also turn out to be dependent from the point of view of our methods; and pairs of texts known as positively independent prove to be independent from the standpoint of our methods as well.

The discovery of the laws that govern the distribution of information in large historical texts, with the establishment and experimental verification (based on these laws) of new dating methods (there being eight of them at present), is the first basic result of our work. Certainly, the dates we obtained cannot be regarded as absolute and final. Therefore, we will speak in the following only of "statistical datings" , although, for brevity, we will sometimes omit the term "statistical", which is always implied. We thereby regard the obtained empirico-statistical dates only as a formal result of the statistical experiments carried out with narrative texts and do not believe that they are undisputed. Meanwhile, the consistency of these dates with those known earlier and obtained by the classical methods points to the objective character of our results.

The second fibre. This fibre can be called "critical". Here, we analyze the traditional datings of events of the ancient and medieval history of Europe, Egypt, and the Mediterranean. To make it convenient for the reader, we gather here the vast data scattered throughout the scientific literature, known to the specialists of various disciplines (however, often not of general common knowledge), and shall reveal the serious difficulties on the way to justifying the dates of certain ancient events.

We shall inform the reader of the fundamental research of the remarkable Russian scientist and universal scholar N.A. Morozov (1854-1946), Honorary Member of the USSR Academy of Science, who was the first to pose and fully formulate the problem of justifying ancient chronology by means of the methods of natural science, and who collected enormous critical material, putting forward daring hypotheses. We also speak of Isaac Newton's chronological research (questioning the dates of many ancient events), of well-known representatives of the critical school in chronology, and of various others working in the field. We then let major specialists in archaeology, source research, or numismatics speak and often resort to quoting and supplying the opinions of well-known scientists, juxtaposing different points of view so that the reader

Preface

xv

can form his own attitude toward the problems touched upon. The analysis of the dates of ancient events is the basic application of the empirico-statistical methods we worked out. I was therefore forced to analyze possibly all preserved versions of the datings of particular events. As a matter of fact, ancient and medieval texts often differ with respect to the dates of many important events. Attempting to stay as close as possible to the "original" versions (and perhaps to reconstruct them), we usually preferred the versions established in the chronological documents from the 11th to the 16th century. The chronologists of that time were nearer to the ancient events described, which is very important. The versions recorded between the 17th and the 20th century are often the consequence of later, secondary treatment, sometimes blurring the original chronological scheme. The reader should always remember this when looking at the dates given in this book.

Let me clarify this thought. Consider the evolution in time of historical documents and that of attitudes toward the datings of the described events. In the absence of a unique system for denoting dates in antiquity and the Middle Ages, the same events and documents could be dated differently by different chronologists belonging to different epochs. Let an event occur in the year to and be fixed in a document X written in to (or around this time) by a contemporary. X starts "living" when generations succeed each other. Another chronicler living in a later year t could no longer have access to all the necessary information and might "calculate" the date of an event. Denote by D(to, t) the date ascribed to an event in X, and actually occurring in to, by a chronicler who lived in the year t. It is clear that D(to, t) .can be different from to by some positive or negative value. The chronicler's version of the date can turn out to be older (then D(to, t) is less than to) or, on the contrary, younger (then D(to, t) is greater than t). Thus, D(to, t) establishes the point of view elaborated by the chronicler in t with respect to the datings pertaining to X. It is obvious that D(to, t) is dependent either of to or t. We can assume that D(to, to) = to, i.e., the contemporaries mostly date the contemporary events correctly.

Let us construct the graph of the dependence of D(to, t) on t for a fixed to. We then obtain the visual representation of the evolution of the later chroniclers' view of dating an event actually occurring in to. It is convenient to represent it in the form shown in Fig. A. Dating the event to, given by the contemporary chronologists, is denoted by D(to, 1986). In other words, D(to, 1986) indicates the modern version of the dating if the event actually occurred in to. Of course, D(to,1986) can be different from the true dating, for example, be more ancient or younger.

Since the same event in to could be described by the contemporaries in several different documents (this being the typical situation), these individual versions start existing individually as separate texts not related to each other from the viewpoint of subsequent generations. We have represented this fact schematically in Fig. B by doubling or repeating some events and their datings several times. The further evolution of each version is represented by its own

XVI

Preface

D

Figure A. Visual representation of the later chroniclers' view of the evolution of dating an ancient event.

curve, each emanating from the same diagonal but subsequently behaving absolutely independently. Meanwhile, different versions of the description of an event, outwardly totally different, can diverge far from each other from the standpoint of later chronologists. The complete evolution of datings of ancient events is given in Fig. C. Each subsequent epoch finds its own attitude toward the datings of the past events. These versions can vary substantially with time (we give examples of this in the book). Starting with the period from the 16th to the 17th century (see Fig. C), the chronological version of ancient times suggested by I. Scaliger and D. Petavius, is being "stabilized",

t

D

Figure B. Different versions of the description of an event can diverge far from each other from the standpoint of later chronologists.

Preface

XVll

t

D

15-16 th cc, A.D.

----------~---------------------------

Figure C. The complete evolution of dating ancient events.

and the modern point of view coincides, in its basic features, with their chronology. This circumstance shows in the increasing straightening of the "alignment" of the dating trajectories. Today, we have assimilated only this version. However, very little is known about all the previous versions, which often differ sharply from today's. In other words, we are only well aware of the topmost line for the dates D(to, 1986) and know very little of other lines, which obviously make up the bulk of the diagram. Thus, the enormous base of the chronological iceberg is hidden, within which the modern version of ancient chronology has been formed. The basic question formulated in the critical fibre of the book (the second ''fibre'') is related only to the underwater part. It is in this sense that we paid so much attention to the ancient chronological versions of the 10th- to 15th-century scientists.

We now elucidate what is meant by "correct", or "authentic", chronology in terms of the graphs in Figs. A, B, and C. It is the chronology in which the evolution of the date of an ancient event would be represented by approximately vertical lines (see Fig. D). Only in this case can the dates accepted today be regarded as realistic. To verify whether today's ancient chronology satisfies this condition, we should exhibit chronological tables associated with each horizontal line in Fig. C, made up by subsequent chronologists of ancient times. In other words, we should find the originals of those ancient

XVlll

Preface

t / D

!IV 20 th c. A.D.

--~-+~~+-~~-+-r~~+-+-+-+-~~~~----

VV

--t-II-t--t-t~+-+-I--~~--+--+-+-;+/-b'/ - - - - - -- - 15-16 th cc. A.D.

--~-+~;-+-~~-+~~~+-+-~------------~

y

---r~~~~~~~+-~~----------------

/V

----+~~~~~4_~~/---------------------

IV

---~~~~~4~-------------------------

;V

---~~~/_-----------------------------

VV

V

Figure D. Correct or authentic ancient chronology. The evolution of the dates of events is represented by vertical lines.

chronological versions forming the steps of a staircase which the dates were "ascending" . Meanwhile, we have to see that the transition from each version to the previous or subsequent one is represented by vertical lines in Fig. C.

However, an attempt to descend into the past on these "steps", say by jumping over the 20 to 30 years that make up a generation, permits us to move only to the 12th and 13th centuries (with the "staircase" breaking earlier than that). Here we only discover "dating" of pieces that are not united into chronological tables preserved until today, and which fix the viewpoints of the ancient chronologists. Earlier than approximately the 13th century, no sequence of "shorter" predecessors of the chronological table can be found. It is desirable that the "shortening" of the table (respectively, its "extension") occurred approximately by 20 to 30 years, in the hope that the events of this time were described by a contemporary. The important characteristic of the second "fibre" is that the critical material is gathered only here. It thus acquires a new quality and permits us to embrace a larger volume of critical data on the basis of one point of view, accumulated in special works on ancient chronology. We assume that the reader is at least roughly familiar with tradtional ancient chronology (having studied it at school, university,

Preface

XIX

etc.) In general, we do not repeat the traditional version, since we believe it to be known by everyone, but rather focus our attention on the account and criticisms of the competing versions, which are sometimes much different from the traditional one and were developed by many scientists between the 16th and the 20th century.

Within the framework of the second "fibre", we also supply a brief analysis

of the traditional dating methods based on archaeological or radiocarbon data, which is of use if the reader would like to estimate the degree of reliability and accuracy. We shall also pay much attention to the dating of events that are about one, two, or three thousand years old and will demontrate, agairL/ by citing a number of authors, the difficulties that arise. Dating of material more than three thousand years old is beyond the scope of this book.

The third fibre. The author has constructed the so-called global chronological diagram (GCD), which can be regarded as a sufficiently complete and traditional "textbook" for ancient and medieval chronology. All the basic events of ancient history with their traditional dates, lists of the names of principal characters, and so forth, have been plotted on the time axis, and the basic preserved primary sources marked for each epoch. The diagram contains tens of thousands of dates, names, references. Occupying an area of several tens of square metres, it is a convenient collection of statistical data and a guidebook to the building of the traditional version. The graphic representation along the time axis of the principal dates proved useful for the statistical experiments. Since the GCD contains too much material, it was included in this book only in abbreviated form as short tables or graphs and is often replaced by this shorter version. We stress once again that the GCD is based on the traditional dating of ancient events, arising from the Scaliger and Petavius chronology.

The fourth fibre. The whole set of empirico-statistical methods we developed was applied to the GCD statistical material (see the first "fibre"). All possible pairs of time intervals (epochs) along with the basic texts describing them were considered, and the texts were statistically examined and compared. The "proximity coefficients" or textual "dependence coefficients" were subsequently calculated. If the dependence coefficient for two texts X and Y was the same (in order) as for two a priori, positively dependent texts from the 13th to the 20th century, then X and Y along with the associated time intervals were called "statistically dependent". This was represented in the GCD by denoting the corresponding time segments by the same symbols, for example, by the same letter T. The symbols were chosen arbitrarily. However, if the proximity coefficient was the same (in order) as for two a priori independent texts from the 13th to the 20th century, then X and Y were termed "statistically independent" and hence represented by noncoincident symbols like the letters Hand C. We would like to make it clear that by investigating experimentally reliably dated texts describing the 13th to the 20th century, it was discovered that the proximity coefficients distinguish between

xx

Preface

a priori dependent and independent texts. For example, one of these coefficients, p(X, Y), did not exceed 10-8 for two texts known previously as dependent and was not less than 10-3 for two surely independent texts, which shows the difference of about 4-5 orders. Now, comparing two arbitrary texts X and Y, we can say whether the value of the coefficient is in the zone for dependent or independent texts. It can also be in the zone of "neutral" texts. It goes without saying that the indicated bounds for the values of the coefficient have been found experimentally. Further discovery of dependent and independent texts is then carried out within the framework of the experimental material (which is, though, sufficiently large).

For vast computational experiments, the GCD revealed pairs of statistically dependent texts and the corresponding epochs. The results of applying different methods turned out (and this is very important) to be remarkably consistent; namely, if a pair of texts (and periods) were statistically dependent from the standpoint of one method, then they were also dependent from the point of view of other methods applicable, in principle, to the tests in question. This consistency seems to be important. Our methods discovered no unexpected, formerly unknown duplicates of documents belonging to the period from the 13th to the 20th century. However, for documents preceding the 13th, and especially the 10th century, the same methods led to the quite unexpected discovery of many new statistical duplicates regarded as independent in all respects, and referring to different epochs.

The global chronological diagram showing all statistical duplicates is the second principal empirico-statistical result obtained by the author.

The third basic result is the decomposition of the GCD into the sum of four chronicles practically identical to each other, but shifted by considerable time intervals. To give a rough idea, the third statistical result can be formulated in the following way: The modern "textbook" of traditional ancient and medieval chronology and history is the sum, from the statistical point of view, of the four replicas of one shorter "chronicle".

The principal part of the book concentrates on these three empirico-statistical results. The subsequent "fibres" are mostly of hypothetical and interpretational character. Roughly speaking, they are required so that we may answer the question: What do the obtained ernpirico-statistical results mean?

The fifth fibre. It can be called interpretational. Here, we offer different hypotheses which can explain the regularities discovered and the reasons for the duplicate appearance. We do not regard this material as final. The "shorter" textbook I suggested certainly does not claim completeness and can only be regarded as a possible version. Interpretations of the obtained statistical results can be of different nature and will require much work by many specialists in various fields.

My attitude toward many of the questions discussed is a result of cooperation and numerous discussions. In particular, the statistical results were reported at the Third, Fourth, and Fifth International Vilnius Conferences

Preface

XXI

on Probability Theory and Mathematical Statistics in 1981, 1985, and 1989, respectively; the First World Congress of the International Bernoulli Society for Mathematical Statistics and Probability Theory in 1986 in Tashkent; the seminar "Multidimensional Statistical Analysis and Probabilistic Modelling of Real-Time Processes" by Prof. S.A. Aivazyan (Central EconomicalMathematical Institute, Moscow); the All-Union Seminar on the Stochastic Continuity Model and Stability Problems by Prof. V.V. Kalashnikov (All-Union Systems Research Institute) and Prof. V.M. Zolotaryov (USSR Academy of Science. V.A. Steklov Mathematics Institute); Controllable Processes and Martingales by Prof. A.N. Shiryayev (USSR Academy of Science. V.A. Steklov Mathematics Institute).

I would like to express my indebtedness to the participants in the discussions,

I am also indebted to Acad. Ye.P. Velikhov and Acad. Yu.V. Prohorov for their assistance.

My work received great stimulus from numerous private talks, consultations, and discussions with colleagues, and also from specialists in mechanics and from physicists; in particular, I am much indebted to the Moscow University staff members Prof. V.V. Kozlov, Prof. N.V. Krylov, Prof. M.M. Postnikov, Prof. A.S. Mishchenko, Prof. Ye.M. Nikishin, Prof. V.A. Uspensky, Prof. P.L. Ulyanov (Assoc. Member of the Academy of Science), Prof. Ye.V. Chepurin, Prof. Ye.G. Sklyarenko, Prof. V.I. Piterbarg, Prof. V.V. Moshchalkov, Prof. M.K. Potapov, Prof. N.V. Brandt, Prof. R.N. Kuzmin, Prof. V.V. Surikov, Prof. Yu.P. Gaidukov, Prof. Yu.P. Solovyov, Prof. Ya.V. Tatarinov, Prof. V.V. Alexandrov, Cando Sci. N.N. Kolesnikov, Cando Sci. G.V. Nosovsky, Prof. V.M. Zolotaryov and Prof. A.N. Shiryaryev (USSR Academy of Science. V.A. Steklov Mathematics Institute), Prof. V.V. Kalashnikov and Prof. V. V. Fyodorov (USSR Academy of Science. Systems Research Institute), Prof. S.T. Rachev (Santa Barbara, USA), and D.1. Krystev (Mathematics Institute; Sofia, Bulgaria), Prof. Yu.M. Kabanov (USSR Academy of Science. Central Economical-Mathematical Institute), Prof. A.V. Chernavsky (All-Union Research Information Transmission Problems Institute), Cando Sci. LA. Volodin (Moscow Oil and Gas Institute), Prof. S.V. Matveyev (Chelyabinsk University), and Cando Sci. M.V. Mikhalevich (Kiev University), Prof. Yu.M. Lotman (Tartu University), Prof. V.K. Abalakin (Leningrad), Prof. M.1. Grossman (Moscow), Prof. L.D. Meshalkin (Moscow), Prof. R.L. Dobrishin (Moscow), Prof. I.Z. Schwartz (Moscow), and Cando Sci. S.Yu. Zholkov (Moscow), and Cando Sci. L.E. Morozova (USSR Academy of Science. History Institute).

My special thanks go to my colleagues Prof. V.V. Kalashnikov and Cando Sci. G.V. Nosovsky for their support and collaboration on many problems of mathematical statistics, astronomy, and computer experiments.

I would like to express my gratitude to all of them.

In addition, I would like to thank Cando Sci. G.V. Nosovsky, Cando Sci.

N.S. Kellin, P.A. Puchkov, M. Zamaletdinov, A.A. Makarov, N.G. Chebotaryov, Ye.T. Kuzmenko, V.V. Byasha, T. Turova, L.S. Polyakova, my parents

XXlI

Preface

V.P. Fomenko and Cando Sci. T.G. Fomenko, and my wife Cando Sci. T.N. Fomenko for their help with the primary statistical processing of the historical sources, statistical tables, and frequency graphs.

Much help in computer programming and processing of the statistical material was given by Cando Sci. G.V. Nosovsky, Cando Sci. N.S. Kellin, Cando Sci. N.Ya. Rives, Cando Sci. I.S. Shiganov, P.A. Puchkov, M. Zamaletdinov, and A.V. Kolbasov.

I would like to express my debt to T.G. Zaharova, the Director of the N.A.

Morozov Museum, and V.B. Biryukov for their assistance in studying the material related to Morozov's scientific work.

Further, I would like to thank Kluwer Academic Publishers and, in particular, Prof. M. Hazewinkel and Dr. David J. Lamer for their effort in publishing the book, with special thanks to Dr. Larner, the head of the Science and Technology Division, for his support of this project.

I am also grateful to Mr. V.V. Novoseltsev of the Copyright Agency of the USSR for his help.

And, finally, this book could never have been published without the interested attention and initiative of the distinguished mathematician and editor Prof. M. Hazewinkel, who made it possible to combine and publish an English translation of all my basic works on this problem, which had already been published in the USSR as individual articles.

The book is dedicated to the memory of the outstanding scientist and universal scholar Nikolai Alexandrovich Morozov (1854-1946), Honorary Member of the USSR Academy of Science, author of many profound works in chemistry, physics, mathematics, astronomy, and history. It was he who first posed the problem of scientifically substantiating ancient chronology by using the methods of natural science, and who obtained fundamental results.

In conclusion, I would like to emphasise that, fully aware of the unusualness and unorthodox nature of certain of the obtained results, I nevertheless believe it my scientific duty to present the work to the reader's judgement in hope that it may serve as the next step in working out new statistical methods for the study of narrative sources and in solving the problem of justifying ancient chronological dates.

The book contains only part of the obtained results. I hope to publish others separately. In particular, the following books have recently been published:

A.T. Fomenko, Methods for Statistical Analysis of Narrative Texts and Applications to Chronology, Moscow University Press, 1990 (in Russian).

A.T. Fomenko, V.V. Kalashnikov, and G.V. Nosovsky, Geometrical and Statistical Methods for Dating Ancient Star Catalogues (When Was ptolemy's "Almagest" Compiled in Reality?), (in Russian; English translation in preparation).

CHAPTER 1

PROBLEMS OF ANCIENT AND MEDIEVAL CHRONOLOGY

§1. The Global Chronological Diagram of Ancient and Medieval History!

1.1. The moon's elongation and R. Newton's conjecture

Chronology informs us of how much time has passed since a certain historical fact. Meanwhile, the chronological data of a narrative source describing the fact should be reduced to the modern dating units, i.e., be referred to by B.C. or A.D. This problem proves to be quite complicated, since many a historical inference depends on which date we ascribe to the events discussed in the source.

Modern global chronology embracing the majority of events of the past is the result of the lengthy work of chronologists who lived from the 15th to the 19th century A.D. Thus, all the major events of ancient and medieval history are associated with certain dates in the Julian calendar, which permits us to study historical processes, evolution of scientific and cultural ideas, technological progress, and so forth, within the scope of large time intervals [1],

[2], [4].

However, such research has led to the discovery of certain phenomena which cannot easily be explained for the present. We give an example from natural science, namely, from astronomy. The lunar theory deals, inter alia, with a parameter called the second derivative of the moon's elongation (D"). Depending on time, the values of this parameter should be available for past eras. It can be computed if the ancient eclipse data are known. The problem has been solved by the prominent American astronomer R. Newton [10]. The

1 First published as an article in Kh,imi1lQ. i zh,1Izn', 9(1983), pp. 85-92.

1

2

Problems of Ancient and Medieval Chronology

Chapter 1

o

D-(I)

40 -------------------------

20

·20 ..._ -_._-_--- ....... --~

-1000

o

+1000

+2000

1

Figure 1. R. Newton's graph demonstrating that D"(t) decreases with time. See the astonishing, inexplicable jump at around the first millenium A.D.

graph (Fig. 1) he obtained turned out to be extremely surprising. Newton wrote:

"The most striking feature of Fig. 1 is the rapid decline in D" from about A.D.

100 to about A.D. 1300 This decline means that there was a 'square wave' in the

osculating value of D". Such changes in D", and such values, are unexplainable

by present geophysical theories .... " ([10J, p. 114)

To explain this square wave (one-order jump), Newton was forced to suggest that there should exist some nongravitational interactions in the earth-moon system [11]. These enigmatic forces do not manifest themselves in any other way, which is in itself quite unusual.

Below, we shall see that there is at least one more explanation of the jump in D".

1.2. The Dark Ages and the Renaissance epochs

Let us return to chronology. In the history of Europe and the Mediterranean, there are several Renaissance epochs during which many achievements of ancient scientific thought, lost in the period of the Dark Ages, were discovered. The epoch in the history of Europe when many scientific facts and cultural habits of the past were rediscovered (from the 13th to the 16th century) has been studied most extensively. Such duplication is explicitly traced in astronomy, military engineering, architecture, literature, and many other branches of science and art. For example, the famous Greek fire, which had played such an important role in the sea battles of antiquity, and which had then been forgotten for centuries, was rediscovered only in the Middle Ages.

§1

The Global Chronological Diagram

3

Apart from the classical Renaissance, the Carolingian Renaissance (the time of Charlemagne) is also generally known, when many authors imitated the antique paragons, duplicating the literary themes which had been forgotten earlier. Similar phenomena (termed Restoration) are also known in the history of ancient Egypt. The prominent Orientalist B.A. Thraev noted that the culture of the Saite period had reproduced that of the Old Kingdom: 2,000-year-old texts again went into use, tombs were decorated following the ancient ways, titles that had sunk into oblivion were reintroduced, and so forth.

As we see, duplicates present themselves as a rather frequent phenomenon in history. Naturally, the question arises as to how they are distributed in time: in a random manner or subject to some covert governing law?

1.3. How to substantiate ancient chronology

To calculate the dates of ancient events is not as simple as it may seem at first glance. The final proof of the correctness of certain dates still remains a problem today. It continues to attract the attention of historians and the specialists of physical and chemical dating methods. It is but natural: The further we move from an ancient event in time, the harder it is to date it. The contradictions that often arise in doing so have caused some historians to express doubts regarding the dating of certain events, as suggested by the first chronologists of the 16th to the 18th century, which, by the way, are still accepted with few exceptions at present [12], [13]. A new scientific discipline was born, namely, hypercriticism, which denied not only the correctness of dating a particular event, but also the trustworthiness of certain ancient events. The famous representative of this school, who specialized in the history of ancient Rome, T. Mommsen, noted, in particular, that different versions of dating the foundation of Rome diverged to the extent of 500 years, and that this oscillation influenced the dating of all the documentation counting years since the "foundation of Rome" ([14]; [14*], pp. 513-514).

Chronological problems interested the Egyptologists, too. Thus, H. Brugsch stressed the enormous difference in the determination of the date when Menes had been placed on the throne, writing that the difference between the extreme conclusions was striking, it being equal to 2,079 years. In spite of all the discoveries in this branch of Egyptology, the numerical data were (at the end of the 19th century) still in a very unsatisfactory state ([16]; [16*], pp. 95-97).

Another example: The chronology of certain events in Egyptian history, which was given by Herodotus in his famous Histories, differs by more than a millennium from that accepted today. Herodotus' chronology is much shorter than the modern version; sometimes, he even places near each other (see [17]) rulers who according to the modern version are separated by 18 centuries ([17*], pp. 512, 513, 516).

But especially many discrepancies show up if one compares the dates given in medieval texts with the dating ascribed to them today. The distinguished

4

Problems of Ancient and Medieval Chronology

Chapter 1

modern chronologist E. Bickerman even speaks of "the chaos of medieval datings" ([4], p. 78).

Chronology in its present form was created in a series of fundamental works by the founders of modern chronology as a science, J. Scaliger (1540-1609) and D. Petavius (1583-1652). It became a precise science later; however, the work is not yet completed, and, as Bickerman notes, there is no sufficiently complete investigation of ancient chronology that would satisfy modern requirements ([4]; [4*], p. 90).

It is not surprising that certain sceptical minds have drawn dramatic conclusions from the above-mentioned difficulties. Thus, as early as the 16th century, a professor of Salamanca University, de Arcilla, published two papers in which he stated that the whole of history preceding the 4th century had been falsified (see de Arcilla, Programma Historicae Universalis, Divinae Florae Historicae). The same conclusion was reached by the historian and archaeologist J. Hardouin (1646-1724), who regarded the entire classical literature as the work of 16th-century monks. Isaac Newton devoted many years to historical and chronological studies. Having thoroughly investigated practically the entire historical and theological literature, he wrote Abreges de la Chronologie [19], asserting that the time scale of the chronology of antiquity was unnaturally extended. Newton made up his own tables in accordance with a new version of chronology which related the biblical texts to the history of the Mediterranean. In his book Newton, V.G. Kuznetsov wrote that Newton had collected

"fantastically large volumes of historical material. This was the total of forty years of work, toilsome research and enormous erudition. Newton, in fact, studied all the basic literature in ancient history and all primary sources ... " ([18], pp. 104-105).

"Certainly, being unable to read cuneiform and hieroglyphic texts and having no archaeological data, which were then unavailable, ... Newton was in error to the extent not only of tens or hundreds of years, but even millennia ... " «18], pp. 106-107).

As a matter of fact, many of the most important events of Greek history were chronologically moved forwards by Newton by 300 years, and those of Egyptian history by 1,000 and even 1,800 years.

And now in this century, in his Historie und Kriiik , the German researcher R. Baldauf was proving on the basisff philological arguments that not only ancient but even early medieval history was a later falsification.

An attempt to systematize the considerable critical material and to analyze historical paradoxes and duplicates from the standpoint of natural science was carried out in the work of a scientist with encyclopaedic knowledge, the revolutionary, public figure and honorary academician, N .A. Morozov (1854-1946) [3]. He actually held the opinion of de Arcilla and believed that traditional chronology had been artificially stretched [2]. It should be noted that he apparently came to this idea independently of de Arcilla. Remarkable scientific intuition and strict logical argumentation permitted Morozov to list numerous data in support of such a conjecture. However, his striving to dot all the i's led to poor substantiation of many of his statements; some contained factual

§1

The Global Chronological Diagram

5

errors, and the new chronological version as a whole (including the hypothesis regarding the falsification of ancient history) was rejected, which does not at all lessen his achievements, for the problem is so complicated and many-sided that one mind alone, even if outstanding, is unable to solve it completely.

1.4. Statistical dating methods: new possibilities

To overcome the above difficulties, we should try to consider the subject from a different angle and create a certain independent dating method which is not based on subjective estimation. This done, we can start analyzing the whole of chronology. In my opinion, an approach involving the statistical analysis of various numerical characteristics associated with ancient texts is most suitable for this purpose. The interested reader can learn about concrete methods and some of their applications to the analysis of global chronology from the short bibliography at the end of the section. Here, we shall confine ourselves to a short account of the essentials and give several examples.

We make the immediate reservation that the methods suggested by the author do not pretend to be universal. Moreover, the results obtained by each individual method cannot be regarded as impeccably trustworthy. A sound criterion of their validity is the consistency of the dating obtained by different methods (today, there are seven of them). The general scheme is as follows. First, a statistical hypothesis is formulated for modelling some process (e.g., loss of information with time). Then numerical coefficients are introduced which permit us to quantitatively measure the deviations of experimental curves from those predicted theoretically. Further, the model is checked against a priori true historical material, and if it is confirmed, then the method can be used for the dating of events.

For simplicity, we give an example. Let a period in the history of a region P from a year M to a year N be described in a text X (chronicle or annals) broken into separate chapters X(T), each of which is devoted to the events of a year T. We calculate the volume of all the chapters (number of pages or lines) and represent the obtained data as a volume graph, plotting the years, T, on the horizontal, and the volumes of the chapters on the vertical axis. A similar graph for another text, Y, describing the same events, in general will have different form; most probably, the interests or tendencies of the chronologist will have bearing on it. But how essential are these differences? Is there anything common between the volume graphs? Indeed, there is. But, before stating the details, some words about the information-loss pattern.

The essential characteristic of any graph is its peaks, or extremal points.

In our volume graph, they correspond to the years in which the curve attains local maxima. Such peaks indicate the years described by the chronicle in the time interval under investigation with the finest points of detail. Denote by G(T) the volume of all texts created by contemporary writers and describing a year T. We call it the "primary stock" (Fig. 2a). The precise form of its graph is not known to us, since texts get partly lost in the course of time.

6

Problems of Ancient and Medieval Chronology

Chapter 1

Primary information Surviving stock

stock volume graph volume graph

<, />.

" "

I ......... _-

, , , , ,

, ,

I / I.

a)

M

N

Dependent tests

Textual volume for X

Textual volume for Y

b)

M

N

Independent tests

Textual volume for Y

-:

c)

M

N

Figure 2. Textual volume graphs in a time interval M N: (a) the primary and surviving stocks (curves should exhibit peaks approximately in the same years); (b) the curves for dependent texts are correlated; (c) the curves for independent texts are not correlated.

We now formulate the information-loss model, namely, there will be more texts for those years to which more texts were originally devoted. It goes without saying that to verify the model in this form is difficult, because the graph of the primary stock remains unknown. But one of the corollaries can

§1

The Global Chronological Diagram

7

be verified. Later authors, X and Y, while describing the same period (and not being its contemporaries), will be forced to employ approximately the same set of ancient texts available. Therefore, they will be able to describe best those years from which more texts remain.

Eventually, the model conjecture is formulated as follows: The graphs of the volumes of chapters for two dependent texts X and Y (i.e., describing the same period of history and the same region) must have simultaneous peaks; in other words, the years described in detail in X and Y should coincide or be close (Fig. 2b). On the other hand, if two texts X and Y are independent (either describing essentially different periods of history of the same length, or for different regions), then the graphs of the volume for X and Y attain local maxima at different points (Fig. 2c).

After the mathematical formalization, an experiment was carried out in which the model (maximum correlation principle) was verified for several hundreds of pairs of dependent and independent historical sources. The principle was confirmed, which made it possible to offer a method for dating texts and also for discovering interdependency among them. For example, to date events described in a chronicle, we have to try to choose an a priori dated text such that the volume graphs attain maxima practically simultaneously. If, however, the dating of two comparable texts is unknown, but the peaks in the graphs coincide, then we can assume with a high degree of probability that the texts are dependent, i.e., the events described are close or coincide.

Now, just a few words about some other methods of dating. They are based on the statistical analysis of such parameters as the frequency of mentioning the names of historical characters and that of various astronomical phenomena, the period of the rule of kings in various dynasties, formalized biographical data of historical figures, and so on. All these methods have been verified for undoubtedly true material of the 13th to the 20th century, and their validity has been confirmed [5], [6], [7].

1.5. The duplication effect in ancient history and chronology

The methods briefly described above are applicable not only to the dating of texts. They also permit us to find various literary borrowings, repetitions, literary cliches, citations, and parallels in the texts being compared. For exampie, if in comparing two dynasties of kings a certain dependence is discovered (i.e., if the corresponding graphs of the rule duration are extremely close), then this can be interpreted in different ways. One interpretation consists of our probably having discovered an intentional imitation by the annalist of a certain authoritative source. However, another version is also probable, namely, that we are dealing with duplicates which were never recognized to be identical and are narrating the same events, and which were related to different historical epochs.

Sometimes these methods help discover the proximity of chronicles, i.e., their having originated from the same source. In particular, they make it

8

Problems of Ancient and Medieval Chronology

Chapter 1

possible to indicate the duplication effects discussed at the beginning of the section in connection with the Renaissance epochs. As it turns out, there are essentially more such historical epochs than is usually thought. To avoid a terminological muddle, we will speak in the following simply of duplicates.

It is now time to formulate our problem: to find possibly all duplicates in ancient and medieval history, and if we succeed, to construct on their basis a hypothetical chronology without repetitions and Renaissance periods which are sometimes hard to explain.

1.6. The global chronological diagram and the "modern

textbook" of ancient and medieval history

Before coming to the thorough analysis of historical texts for the purpose of discovering and systematizing duplicates, we have to construct as complete a table of events of the ancient and medieval history of Europe as possible and also that of the Mediterranean region, Egypt, and certainly the Near East, showing their traditional dating. To this end, the author has investigated 15 basic chronological tables and 228 fundamental primary sources (chronicles, annals, records, etc.). Together, these texts contain the description of practically all the basic events of the period from 4000 B.C. to A.D. 1800. All this information was then represented graphically on the plane. Each historical epoch with all of its basic events was shown on the time axis. Meanwhile, each event was represented by a point or horizontal line-segment in accordance with its duration, with the beginning and end of the line-segment being those of the event (e.g., a king's rule). Simultaneous events were represented one above the other, so that any ambiguity or overlapping would be avoided.

Thus, a maximally complete chart was constructed. We will call it the global chronological diagram (GCD); see Fig. 3, upper line. To see which events took place in a particular year according to modern chronology, we have to draw a vertical line through that year on the GCD and collect all the events being intersected.

We have applied the above dating and duplicate-recognition methods to the enormous historical data recorded on the GCD. The entire period of history on the diagram was broken into epochs, for which, roughly speaking, a set of characteristic graphs was calculated. For example, for each epoch (a linesegment on the time axis) in the history of each region, the volume graphs for all the basic primary sources describing this epoch were plotted, and those for different epochs were compared pairwise. As a result of the extensive experiment in which hundreds of texts were investigated, containing altogether tens of thousands of names and hundreds of thousands of lines, the author unexpectedly discovered pairs of epochs which are regarded as independent in traditional history (in every sense), but with extraordinarily close and sometimes even practically indistinguishable graphs.

§1

The Global Chronological Diagram

9

8

-

-

+

§

- +

§I ~

- - + +

8

- •

8

- +

~

- •

8

-

-

•

~ §I

- - . .

European history (chronicle E)

E:

T T T T

C.: 040"0!1- c---,I _

1.778-year shift T T T

C3: 0i/""---'H \1[P]11- c.......,I _

1.053-year shift T T T

c, 01 [OJ! I p 111r-- c---,I _

333-year shift

C,: K

~~~------------~--------~

P

Figure 3. The "modem textbook" of European history and its decomposition into the sum of four short isomorphic chronicles.

We illustrate this with an example. The volume graph of the primary sources describing the history of ancient Rome from 753 to 236 B.C. exhibits peaks practically in the same years as a similar graph constructed for medieval Rome from A.D. 300 to 816. To verify this fact, these two time intervals of 500 years in length should be superimposed first (Fig. 4). The same coincidence of the two seemingly independent series of events (antique and medieval) was also discovered by other methods. The GCD happened to include quite a number of duplicates, i.e., pairs of historical epochs which are as close as are undoubtedly dependent texts describing the same historical period. We once again emphasize that the results obtained by different methods are invariably consistent.

10

Problems of Ancient and Medieval Chronology

Chapter 1

:.

· .

· .

· .

· .

· .

· .

· .

· .

· .

.'~ ..

: :

· .

· .

· .

• 0

o •

! :

i : :'~

. '"

.: ....!

! ...•

. .... : ',' .

: .f •

o • 0

. .." .

. .. . :

. :: '.: . .'~..:

..... . ... : ....

· .

· .

· .

· .

· .

. ," ',. "

o

. .

·753 +300

·294 +758

·236 +816

Figure 4. Primary source volume graph, describing ancient (dotted line) and latest (solid line) Roman history: this maximum correlation can hardly be accidental.

1.7. The "modern textbook", a composition of four

identical pieces

Let us once more carefully consider the upper line in Fig. 3. To represent the set of all the discovered epochs of duplicates clearly, they are marked on the GCD by the same geometric symbols and letters (chosen arbitrarily). More precisely, duplicates are designated by the same letters, and the epochs that are considerably different from one another by different ones.

Some of the letters repeat continually (e.g., T repeats eleven times, and C four times). The length of the geometric figures indicates the duration of the corresponding epoch. Say, the black triangles T are associated with periods that are about 20-30 years long, and the rectangles C with periods approximately 300 years long. Certain intervals of time on the GCD are covered by several figures. Thus, the period from ca. A.D. 300 to 550 is represented by four superimposed rectangles n, K, C, P, which means that part of the chart devoted to this period is composed of four pieces designated by different letters. In other words, in the set of events which occurred in the interval from A.D. 300 to 550, those making up the piece Il are first distinguished, then those composing K, and so forth. The events falling into a particular piece are most often associated by what happened in the region. By the way, all the Renaissance epochs noted by the historians are contained in the duplicates on the GCD.

But the main thing is that a rather complicated structure of the GCD is naturally obtained as the result of one quite surprising process. If the four lines (chronicles) Cl, C2, Ca, C4 (also shown in Fig. 3) are distinguished in the chart and are glued together along the vertical line by superimposing, then we shall obtain, as can be expected, the same line on the GCD, consisting of

§1

The Global Chronological Diagram

11

the lettered epochs. But the most surprising fact is that these four chronicles are represented by practically the same series of letters and symbols. The four duplicate pieces differ from one another only by their position on the time axis. Thus, the second chronicle differs from the first one only by a backward shift in time of about 333 years, the third by a shift of already 1,053 years, and the fourth by an approximately 1,778-year-Iong shift. Admitting a certain liberty, we can say that the "modern textbook" of the ancient and medieval history of Europe, the Mediterranean region, Egypt, and the Near East is a composite chronicle obtained by gluing together four practically identical replicas of the abridged chronicle Ct. Three other chronicles are derived from it by redating and renaming the events described, while the whole of C1 is lowered (i.e., shifted back in time) by about 333, 1,053, and 1,778 years, respectively. Thus, the entire GCD can be restored from its part C1.

Another fact to be emphasized is that nearly all the information in the chronicle Cl is concentrated to the right of A.D. 960. The periods P, T, C (to the right of the 10th century A.D.) are very rich in information, whereas K, H, Il (from A.D. 300 to 960) contain very few events.

1.8. Certain corollaries and interpretations

This formal decomposition of the "history textbook" into the sum of four chronicles can be interpreted differently. First, that the periodic behaviour I discovered is possibly accidental. It can be calculated, however, that the probability of such a random event is extremely small. Another possible interpretation is that insufficient written evidence casting light on certain periods of ancient history encumbers the application of statistical methods. Finally, a third possible explanation, which seems to me worth notice, is that the existing global chronology of the period preceding the 13th century A.D. requires quite substantial corrections in certain cases. These will require the redating of certain blocks of events now related to earliest antiquity, for which the chronicles C4, C3, C2 of the modern chronological chart should be distinguished and lifted upwards in accordance with the mentioned shifts. After this formal procedure, the known written history of Europe, the Mediterranean, and so forth, will be abridged, and most of the events now dated as having occurred earlier than the 10th century A.D. will be placed in the interval from the 10th to the 17th century A.D.

This hypothesis can help explain certain long-known paradoxes of traditional chronology, including those mentioned at the beginning of the section. However, I do not at all agree with the assumption of N .A. Morozov and some of his predecessors that the information today available regarding ancient history is, allegedly, a later falsification. The results obtained by new methods of dating show that most of the primary sources which have been prescribed are originals describing real events. Almost all the events mentioned in ancient documents did occur; the question remains only where and when.

12

Problems of Ancient and Medieval Chronology

Chapter 1

Generally speaking, the principal result of the work done is of formal statistical character, and no more. Nonspecialists in history have already attempted to interpret this result in a pseudoscientific manner, with the data of social science being ignored. I am decisively against such conclusions.

1.9. What is to be done with the moon's elongation?

Let us return to the beginning of the chapter, to the moon's elongation and its second derivative. The computation of D" was based on the data of ancient eclipses adopted by traditional chronology. The attempts to explain the surprising square wave in the graph of D" do not touch at all upon the question whether the data of the eclipses were determined correctly. We will assume that an eclipse has been dated correctly if its characteristics exactly described in a historical source coincide with the parameters of the real eclipse offered by chronology.

Morozov suggested a method of "impartial" dating, namely, the comparison of the characteristics of an eclipse given in a primary source with those from astronomical tables. Analysis demonstrates that, while not questioning the chronology of ancient events and a priori regarding it as true, the astronomers often could not find a suitable eclipse in the "desired" century and thus resorted to strained interpretations. For example, in the History of the Peioponnesian War by Thucydides, three eclipses were described, traditionally dated as belonging to the 5th century B.C. However, even in the last century, a discussion around this triad started, being caused by the fact that there were no eclipses with suitable characteristics in the assumed epoch. Still, an exact solution can be found if we extend the interval of the search. One solution is the 12th century A.D., and the second one the 11th century A.D. There are no other solutions.

A similar effect of "shifting the dates forwards" can be extended to those eclipses which are traditionally dated in the interval from A.D. 400 to 900. It is only after A.D. 900 that the traditional dates are satisfactorily consistent with the precise datings given by astronomy, and undoubtedly after A.D. 1300.

But why, in fact, speak of it here? Because such a shift of dates is completely consistent with the GCD being glued together from four identical chronicles. If an earlier and traditional date for an eclipse was assigned to an epoch, say, labelled by C on the GeD, then its precise astronomical date lies much farther to the right on the time axis. It occurs in the period of history denoted on the diagram by the same letter. In particular, the date shift just described is reduced to advancing certain groups of eclipses up by about 333 years, others by 1,053 years, and so on. In such a time advance, the mutual occurrence of dates inside each of these groups is practically unaltered, and the group is advanced as a block.

But what's to be done with D"? Its recalculation on the basis of the reconsidered dates of ancient eclipses showed that the graph (Fig. 5) is qualitatively altered. It cannot now be moved reliably to the left earlier than the 10th

§1

The Global Chronological Diagram

13

40 -------------------------~

as
20 -
as
"0
No data c: Reliable
as data
-
~
Q)
u
0 c:
::::> -20 ·1000

o

+1000

+2000

t

Figure 5. The new graph of D"(t), constructed on the basis of the recalculated dates of ancient eclipses, has no anomalies: there are simply no reliable data to extend it to the left.

century A.D., while in the later period, it almost coincides with the curve already found and is represented by an almost horizontal line. No square wave is found in the second derivative, and no mysterious nongravitational theories should be invented ....

It goes without saying that the work discussed here cannot claim to be the basis for any final conclusion, the more so as the most complicated, multifarious and often subjectively interpreted historical data are analyzed here by strictly mathematical methods. To process the material will certainly require a large variety of methods, purely historical, archaeological, philological, physical and chemical, and, inter alia, mathematical, which as the reader can see, will permit us to look at the problems of chronology from a new angle.

References

[1] Blair, J., Blair's Chronological and Historical Tables, from the Creation to the Present Time, etc. G. Bell & Sons, London, 1882.

[2] Morozov, N.A., Christ. Giz, Moscow-Leningrad, 1926-1932 (in Russian). [3] Nikolai Aleksandrovich Morozov: A Universal Scholar. Nauka, Moscow, 1982 (in Russian).

[4] Bickerman, E., Chronology of the Ancient World. Thames & Hudson, London, 1968.

[4*] Russian translation of [4]. Nauka, Moscow, 1975.

14

Problems of Ancient and Medieval Chronology

Chapter 1

[5] Fomenko, A.T., "Informative functions and related statistical regularities", Abstracts of the Reports of the Third International Vilnius Conference on Probability Theory and Mathematical Statistics. Institute of Mathematics and Cybernetics of AN LSSR 2(1981), pp. 211-212.

[6] Fomenko, A.T., "A method of duplicate recognition and some applications". DAN SSSR 258, 6(1981), pp. 1326-1333 (in Russian).

[7] Fomenko, A.T., "New empirico-statistical methods in ordering texts and applications to dating problems", DAN SSSR 268, 6(1983), pp. 1322- 1327 (in Russian).

[8] Fomenko, A.T., "The jump of the second derivative of the moon's elongation", Celestial Mechanics 25, 1(1981), pp. 33-40.

[9] Newton, I., The Chronology of Ancient Kingdoms amended. To which is prefix'd, a short chronicle from the first memory of things in Europe, to the conquest of Persia by Alexander the Great. J. Tonson, etc., London, 1728.

[10] Newton, R., "Two uses of ancient astronomy", Phil. Trans. Royal Soc., Ser. A., 276(1974), pp. 99-116.

[11] Newton, R., "Astronomical evidence concerning non-gravitational forces in the Earth-Moon system", Astrophys. Space Sci. 16,2(1972), pp. 179- 200.

[12] Scaliger, J., Opus novun. de emendatione temporum. Lutetiae, 1583. [13] Scaliger, J., Thesaurus iemporum: Eusebii ... Chronicorum canonum omnimodae historiae libri duo... Opera ac studio J.J. Scaligeri, etc. Lutetiae, 1606.

[14] Mommsen, T., The History of Rome. Macmillan & Co., London, 1913. [14*] Russian translation of [14]. Giz, Moscow, 1936.

[15] Petavius, D., Opus de doctrina temporum divisum in partes duas, etc.

Lutetiae Parisiorum, 1627.

[16] Brugsch, H., Egypt Under the Pharaohs. John Murray, London, 1891. [16*] Russian translation of [16]. Petersburg, 1880.

[17] Herodotus, The Histories of Herodotus, etc. Everyman's Library, London-New York, 1964.

[17*] Russian translation of [17]. Nauka, Leningrad, 1972.

[18] Kuznetsov, V.G., Newton. Mysl', Moscow, 1982 (in Russian).

[19] Newton, I., Abreges de la Chronologie des Anciens Royaumes. Chez Henri-Albert Gosse, Geneve, 1743.

§2

The Moon's Elongation and Ancient Eclipses

15

§2. Computation of the Second Derivative of the Moon's Elongation and Statistical Regularities in the Distribution of the Records of Ancient Eclipses!

2.1. Parameter D" and R. Newton's paper "Astronomical evidence concerning non-gravitational forces in the Earth-Moon system"

The present section discusses in more detail my results described in [4] (see the list of references at the end of §3). It is known that, for certain problems of computational astronomy, the behaviour of the so-called second derivative of the moon's elongation D"(t) as a function of time t should have been known for large time intervals in the past [13]. Let nM be the acceleration of the moon with respect to ephemeris time, and WE that of the earth. The quantity D" = nM _ 0.033862wE, which is the second derivative of the moon's elongation, is called an acceleration parameter [10], [13]. D" is normally measured in arc seconds per century squared. The dependence of the parameter D"(t) on time has been established in a series of remarkable works by the American astronomer R. Newton [9], [10], [13], who calculated 12 values of the parameter D" on the basis of the investigation of 370 observations of ancient and medieval eclipses, extracted from historical sources ([11], p. 113). In computing the date teel. of the observation of a particular concrete eclipse, the parameter D" can be neglected. Therefore, it can, in turn, be found from the distribution of ancient eclipse dates tecl., which is a priori regarded as known. In R. Newton's papers [9], [10], [11], [13], the computation of D" was based on the dates of ancient eclipses contained in the chronological canons of F. Ginzel and T. Oppolzer [8], [12]. They are generally accepted in the contemporary literature. The results of Newton, related to those of Martin, who studied about 2,000 telescopic observations of the moon from 1627 to 1860, allowed him to construct an experimental curve for D"(t) in the interval from 900 B.C. to A.D. 1900. In the following, we will sometimes designate A.D. by "+", and B.C. by "_". In Fig. 6, the symbol. indicates the values of the parameter D" calculated by means of solar eclipse data, while 0 denotes those of D" which were computed from the lunar eclipse durations fixed in the documents. The sign 6. implies the values of D" calculated on the basis of information regarding the duration of solar eclipses. Finally, \l indicates the values of D" computed from the phases of solar eclipses (see [11]).

Commenting upon the graph of D" obtained, Newton wrote:

"D" has had surprisingly large values and ... it has undergone large and sudden changes within the past 2000 years ... " ([11], pp. 114-115).

1 First published as an article in Operation. Research. and A CS, Vol. 20, Kiev University Press, Kiev, 1982, pp. 98-113 (in Russian).

16

Problems of Ancient and Medieval Chronology

Chapter 1

o

•

D'(t)

Prior graph

40 --------------------

•

20

·20 ..... ---"-_--- ..... _.,.

·1000

o

+1000

+2000 t

Figure 6. Experimental curve for D"(t) in the interval from 900 B.C. to A.D. 1900 (R. Newton).

Newton's paper "Astronomical evidence concerning non-gravitational forces in the Earth-Moon system" [9] was also devoted to the attempts to explain this strange gap (one-order jump) in the parameter D".

Thus, on the basis of Newton's works [9]-[11], we can make the following conclusions.

(1) In the interval A.D. 400-600, the parameter D" starts falling sharply (one-order jump).

(2) Before this interval, until A.D. 300-400, the values of D" do not deviate much from zero.

(3) Starting with about A.D. 1000, the values of D" are close to those of today; in particular, they are practically constant.

(4) In the interval 6000 B.C. to A.D. 1000, the parameter D" undergoes considerable variance, with the oscillation amplitude reaching up to 60" /century2.

Hereafter, the bounds of the time intervals indicated are approximate. Newton wri tes that D" "has even changed its sign near about A. D. 800" ([11], p. 115).

In the following, we shall point out two bounds in the behaviour of the graph D", the first of them being about A.D. 500 (the beginning of the square wave on the graph), and the other one about A.D. 1000 (the end of the square wave).

In the present section, we give the results of a new interpretation and calculations of the graph of D", based on the dates of astronomical observational data made precise, which form the basis for computing the parameter D". The curve of [yl which we obtained has qualitatively different character. In particular, the incomprehensible one-order gap of the graph completely vanishes. As it turns out, the new graph of tr is, in reality, oscillating around a constant numerical value which coincides with the modern one. As a corollary, the

§2

The Moon's Elongation and Ancient Eclipses

17

necessity to invent "nongravitational forces" for the explanation of the "gap" in the graph becomes unnecessary.

2.2. A vailable observations of ancient solar and lunar eclipses

Let X be the set of all available observations of ancient solar and lunar eclipses. Their complete list has been given by F. Ginzel ([8], pp. 167-271). Let A be the set of all eclipses described in the ancient texts X. We have to bear in mind that the same eclipse may be described in several ancient texts. We denote them by Xecl.. Let teel. be the date ascribed to a particular eclipse in accordance with the traditional chronology. These traditional dates have been fixed in the papers of F. Ginzel and T. Oppolzer [8], [12]. They all form a basis for the computation of D"(t). In computing D"(tecd [i.e., at a point tecl. on the time axis t), the theoretical, calculated characteristics of an eclipse, obtained for the date tecl. on the basis of modern lunar theory, are compared with the description of the eclipse, portrayed by the ancient sources Xecl.. The deviation between these two groups of data is exactly what permits us to find the value of the parameter D" at the moment tecI .. This value of D" certainly depends on the choice of the eclipse date, and only those ancient texts are important which contain sufficiently much information about it, e.g., the description of the trajectory, phase, and so on. The analysis of all the ancient texts available (see F. Ginzel [8]) permitted us to distinguish a list of sufficiently complete descriptions of eclipses. We do not have the space to give it here. All our computations in the following are related just to these eclipses.

Newton's attempts to explain the mysterious square wave of the function D"(t) do not touch upon the problem of the precision and correctness of the dates ascribed to the ancient and medieval eclipses by traditional chronology [8], [12]. In other words, the question as to how well the descriptive parameters of an eclipse, fixed in an ancient text, correspond to the calculated eclipse parameters found for the moment tecl. on the basis of lunar theory, was addressed, with tecl. meaning here the date ascribed by traditional chronology [8], [12]. The dating and the description of a given eclipse can be regarded as correct only in the case where the two groups of characteristics, i.e., calculated and fixed in a historical source, coincide. Note that changing the dates of the eclipses will alter the graph of D".

The relation between the problem of calculating the parameter D" and the known investigations of N .A. Morozov [2] was indicated for the first time in the author's paper [4], which, in particular, touched upon the problem of correctly dating ancient eclipses and their descriptions. On the basis of the analysis of considerable factual data, Morozov suggested and partly substantiated his fundamental conjecture that the traditional chronology of the ancient world might be artificially extended in comparison with the real situation. An important role in forming this conjecture was played by the method of astronomical dating.

18

Problems of Ancient and Medieval Chronology

Chapter 1

The descriptions of eclipses from certain ancient texts started to be employed for dating these sources and related events as early as the 16th century. However, the method was applied only for the purpose of obtaining the dates in a somewhat more precise and, usually, quite narrow, prescribed time interval where traditional chronology placed them and the simultaneous eclipse under investigation.

2.3. A method of formal astronomical dating

In paper [2], a method of formal astronomical dating was suggested, consisting of the extraction of the eclipse's descriptive characteristics from a historical text, and then purely mechanically recording all dates of the eclipses with these characteristics from the modern astronomical tables. The recalculation of the ancient eclipse data was performed just by this method in the indicated work, with the dates traditionally being ascribed to the time interval from 700 B.C. to A.D. 400. For the purpose of computing the parameter D", I carried out a new series. of calculations of the ancient and medieval eclipse dates, thus confirming, in particular, the effect of shifting the dates of ancient eclipses forwards from 700 B.C. to A.D. 400 ([2]; [4]; see below).

We now describe the method of formal astronomical dating in more detail.

The papers of Ginzel and Oppolzer [8], [12] supply a list of 89 ancient eclipses and indicate the ancient texts which reported them. The latter are usually (traditionally) dated to have occurred in the interval from 700 B.C. to A.D. 592. A list of the eclipses' descriptive characteristics extracted from an ancient text can be complete to varying degrees. For example, the moment of an eclipse during an entire day can be indicated, but not its phase, and so forth. Besides, the canons of Ginzel and Oppolzer contain the complete and theoretically calculated list of eclipses occurring from 900 B.C. to A.D. 1582, with the basic characteristics including the date and phase of an eclipse, the umbra coordinates, and so on. The problem of dating an eclipse described by an ancient text (and, therefore, that of the accompanying events) is solved as follows. We take the eclipses from the canons [8], [12], all of whose calculated characteristics exactly coincide with those in a historical source. At the same time, it is required that (1) there should be no deviation from the description in the document, and (2) the time interval in which the astronomical solution is sought should not be bounded. This means that we do neglect the a priori outside information of "non astronomical origin".

Analysis shows that requirements (1) and (2) are not fulfilled in the classical works [8] and [12] in the overwhelming majority of cases: The date of an eclipse is usually sought there not in the possible whole historical time interval, but only within nafTOW prescribed limits (normally, one century) in which, according to earlier chronological tradition, an approximate date of the event studied (and, therefore, of the eclipse) was pre-established.

The application of the method of formal astronomical dating to eclipses

§2

The Moon's Elongation and Ancient Eclipses

19

traditionally dating from 700 B.C. to A.D. 400 [2], [4] shows that the written evidence concerning them can be separated into two classes.

(1) Short and vague evidence (without any details). Here, it is often unclear whether the text describes an eclipse at all. In this class, the astronomical dating of written evidence is either senseless or allows for so many possible astronomical solutions that they all fall into practically any of the prescribed historical epochs.

(2) Detailed evidence. In this class, an astronomical solution often turns out to be unambiguous (or there are two or three solutions).

2.4. The effect of shifting the dates of eclipses forwards

One important fact discovered in [2] is that all the eclipses of the second class are not dated by the formal astronomical approach traditionally as, for example, in [8] and [12], by 700 B.C. to A.D. 400, but substantially later. These dates sometimes differ from the traditional ones by several centuries. Meanwhile, these new astronomical solutions (dates) fall into the time interval from A.D. 400 to 1600. The forceful distortion of the dates, made by the earlier chronologists, and fixed in the classical papers [8] and [12], is due to the pressure of chronological tradition. The astronomers had to look for the required astronomical solutions only within a narrow, prescribed time interval. More than that, in most cases, an exact astronomical solution could not be found, and the astronomers had to exhibit an eclipse only partly satisfying the description contained in the ancient document (see the examples below). In making use of the method of formal astronomical dating, this deviation (strained solution) vanishes, which leads to the appearance of new astronomical solutions (dates), although different from the traditional ones.

Continuing the investigations started in [2], the author has also analyzed the medieval eclipses from A.D. 400 to 1600 on the basis of formal astronomical dating methods. It turned out that the effect of shifting the eclipse dates forwards, which had been discovered in the above-mentioned paper only for ancient eclipses, could be extended also to the eclipses traditionally dated as belonging to the interval A.D. 400-900. In particular, many pieces of written evidence (due to the extreme vagueness of their formulations) proved to admit a large spectrum of astronomical solutions distributed over the entire possible period of history. It is only beginning with about A.D. 900, and not A.D. 400, as was suggested originally [2], that the traditional eclipse dates become satisfactorily consistent with the results of the application of the formal astronomical dating method. Finally, this consistency becomes reliable only after about A.D. 1300.

This result agrees with the theory of empirical corrections employed in ([8], pp. 4-6), to revise the formulas for the calculation of eclipse dates. Thus, as 21 "basic eclipses" receiving detailed descriptions in at least 10 ancient texts, we can take those dated to the right of (later than) A.D. 840 (42 reports) and distributed in the interval up to A.D. 1386 [8]. On the other hand, we

20

Problems of Ancient and Medieval Chronology

Chapter 1

recall that it is since A.D. 1300 ([12], p. 114) that the graph of parameter D" has finally been stabilized and aligned. Thus, this moment of the graph's stabilization coincides with the origin of the interval of reliable consistency of the dates of ancient eclipses and with the results of the application of the formal astronomical dating method. An extremely small number of the dates of eclipses in the time interval from A.D. 400 to 900 we discovered and which do not contradict the formal astronomical dating method, is of little importance statistically. At any rate, this is true from the standpoint of the new computation of the graph of D".

Before stating the results of the new computations of the graph of D", we shall dwell at length on the effect of shifting the dates of ancient eclipses in the Middle Ages. Above, we have indicated the difficulties facing the astronomers in dating many of the eclipses traditionally. They were caused by the requirement of traditional chronology to place the dates of eclipses in a narrow, prescribed time interval. The formal astronomical dating method removes these obstacles. Because of considerable factual material, we give here only a short summary, the final results, and typical examples.

2.5. An example: three eclipses of Thucydides

Exa mple 1. Consider the three famous eclipses of Thucydides (the so-called triad; see [8], pp. 176-179, eclipses 6, 8, 9). They are linked into one triad by their having been described in one historical text, namely the History of the Peloponnesian War (Bks. II, 27-28; IV, 51-52; VII, 18-19, 50). The descriptive characteristics of the triad, which are extracted from Thucydides' text unambiguously, are of the following form.

(1) All three eclipses were observed in the Mediterranean region, namely, in a square approximately bounded by the longitudes 15° E. and 30° E. and the latitudes 30° N. and 42° N.

(2) The first eclipse was solar.

(3) The second eclipse was solar.

(4) The third eclipse was lunar.

(5) The time interval between the first and second eclipses was 7 years. (6) The time interval between the second and third eclipses was 11 years. (7) The first eclipse occurred in summer.

(8) The first (solar) eclipse was total (since "the stars were visible"), i.e.,

its phase ~ is 12".

(9) The first eclipse occurred in the afternoon (local time).

(10) The second (solar) eclipse occurred at the beginning of summer. (11) The third (lunar) eclipse occurred at the end of summer.

(12) The second eclipse occurred approximately in March.

Condition 12 is not clearly determined from Thucydides' text and, therefore, is not included in the final list of conditions.

The problem arises to find a triad of eclipses completely satisfying all conditions 1-11. Work [8] gives the traditional astronomical solution, namely, 431, 424, and 413 B.C. However, as has been known long ago, it does not satisfy

§2

The Moon's Elongation and Ancient Eclipses

21

all the data of the problem. As a matter of fact, the eclipse of 431 B.C. was not total as required by condition 8. It was only annular with phase 10" for the observation zone and could not be observed as total anywhere on the earth's surface ([8], pp. 176-177). This important circumstance was noted by many authors, e.g., J. Zech, E. H eis, N. Struyck, G. Riccioli, F. Ginzel, and I. Hoffman [8]. A considerable number of astronomical papers were devoted to the recalculation of the phase ~ of the eclipse of 431 B.C., for which various admissible corrections were introduced into the equations of the lunar theory in order to make the phase close to 12". Thus, Dionysius Petavius obtained ~ = 10"25 for the observation zone, Struyck II" ([8], p. 176), Zech 10"38 [14], Hoffman 10"72 ([8], p. 176), and Heis even 7"9(!) ([8], p. 176). In the modern literature, the phase value is assumed to be 10" [8]. We stress once again that, due to its annular form, the first eclipse in 431 B.C. was total nowhere on earth for any latitude and longitude. Accordingly, Ginzel wrote2:

"The insignificance of the eclipse phase was somewhat shocking. ... According to the new calculations, the phase was equal to 10" ...... [8].

Besides, certain other conditions were not fulfilled either. For example, the umbra passed through the observation zone only after 17 hours local time, and even after 18 hours according to Heis [8], which means that condition 9 (the eclipse occurring in the afternoon) is satisfied only approximately.

Certain authors ([8]; see the survey) carried out the calculation of the coordinates of bright planets, thinking that they could have been seen during the annular eclipse, in order to satisfy the important condition 8. However, the obtained results showed clearly that the planets' positions on the celestial sphere during the eclipse of 431 B.C. did not provide for their reliable visibility. If Venus could have been visible, then, for example, Mars was only 3° over the horizon (Heis's computation), while Jupiter and Saturn were below the horizon, and so forth ([8], p. 177). Johnson suggested another astronomical solution for Thucydides' first eclipse, namely, 433 B.C. Although it soon became clear that this solution still did not satisfy the data of the problem posed, it was now for other reasons [8]. Besides, this eclipse had a short phase, namely, 7"8 [8].

The largest variation possible of certain constants in the lunar equations, with the purpose to increase the phase of the eclipse in 431 B.C., was made by Stockwell. However, it yielded only 11"06 for the observation zone, which did not account for the completeness of the eclipse either. The computations were questioned in the literature, too [8].

In this connection, an attempt to revise Thucydides' text itself, and, in particular, condition 8, should be noted also. However, its detailed analysis carried out at the author's request by E.V. Alexeeva (Faculty of Philology, Moscow University) showed that the eclipse characteristics were unambiguously determined from Thucydides. This circumstance had not been questioned earlier, though.

2Translated from the German (tr.)

22

Problems of Ancient and Medieval Chronology

Chapter 1

No other astronomical solutions in 600-200 B.C., which would be more suitable than the traditional solution of 431, 424, 413 D.C., seem to have been found. It is because of this fact that this incorrect "solution" has been retained in spite of the above contradiction repeatedly discussed in the literature.

Meanwhile, the application of the formal astronomical dating method and the extension of the search interval (for astronomical solutions) to 900 D.C.A.D. 1600 yield two and only two exact solutions, the first having been given in paper [2] (Vol. 4, pp. 509, 493-512), while the second one was given by the author of the present work during the repeated analysis of all the eclipses from the indicated interval and the construction of their trajectories on the diagram.

Thus the first solution yields August 2, 1133, March 20, 1140, and August 28, 1151, whereas the second is August 22, 1039, April 9, 1046, and September 15, 1057. Note that the fact of the availability of exact solutions itself is nontrivial. In both exact solutions found, even condition 12 is fulfilled, the one not originally included in the list of basic data. Besides, the first eclipse is total in both solutions (for the observation zone), which is just what was required by condition 8.

2.6. An example: the eclipse described by Livy

Example 2. Consider eclipse 25 (see [8], pp. 189-190) described in the History of Rome by Livy (Bk. XXXVII, 4.4). The characteristics extracted from Livy's text are as follows.

(1) The eclipse was solar.

(2) It occurred 5 days earlier than the ides of July, i.e., on July 10.

(3) The approximate coordinates of its observation zone were 30° < lat. N. < 45° and 10° < long. E. < 25°.

(4) In the observation zone, the moon's trajectory passed below the centre of the sun during the eclipse if the moon and sun were projected on the celestial sphere.

The traditional solution suggested in Ginzel's canon [8] was March 14, 190 B.C. However, since condition 2 was not fulfilled, the astronomers also offered other astronomical solutions, e.g., July 17, 188 B.C. But the conditions of the problem posed were not satisfied in this case either ([8], p. 190). Owing to the absence of other astronomical solutions for the time interval 300-100 B.C., determined beforehand due to the a priori requirements of tradition, and which would satisfy conditions 1-4 better, the traditional one of 190 B.C. was retained in the canon.

Meanwhile, the application of the formal astronomical dating method and extension of the interval in which an exact solution was being sought to the periods from 600 B.C. to A.D. 1600 permits us to reach the following conclusion [2].

(1) As it turns out, there is an eclipse fully satisfying all the conditions of the problem.

§2

The Moon's Elongation and Ancient Eclipses

23

(2) This exact solution is unique for the interval from 600 B.C. to A.D. 1600. (3) It is July 10, A.D. 967.

(4) It is stable with respect to a small perturbation of the initial data, namely a perturbation of the principal condition 2, which means that it remains unique in extending the search interval from July 10 to July 9 and 11, i.e., by one day. This exact solution was found in [2], assuming, naturally, that the Julian names of the months correspond to the Julian calendar.

2.1. An example: the eclipse described by Livy and Plutarch Example 3. Consider the list of descriptive characteristics of eclipse 27 (see [8], p. 190) also described in Livy's History of Rome, Bk. LIV, 36.1. See also Plutarch's Vitae Aemilius Paulus, 17.

(1) The eclipse was lunar.

(2) It occurred on the night of September 4 to September 5.

(3) The observation zone was bounded by lat. 40° and 50° N, and long. 10° and 25° E.

(4) It occurred from 2 to 4 A.M. local time.

(5) Its phase was close to 12", and possibly exceeded 12".

Remark. The phase of a solar eclipse is found by the formula 4> = 12'\, where ,\ is the ratio of the part of the sun's diameter, covered by the moon at the eclipse's maximum, to the whole diameter. However, in the case of a supertotal lunar eclipse, a quantity proportional to its duration is added to the phase of 12" (the moon stays in the shadow of the earth for a long time). Hence, the phase of a lunar eclipse can reach 22"7.

(6) This lunar eclipse occurred after the summer solstice.

The traditional solution given in the canon ([8], p. 190) is June 21, 168 B.C.

This does not satisfy conditions 2 and 6 of our problem. Attempts of many authors to find a better astronomical solution for the interval from 300 to 100 B.C., determined a priori from the requirements of tradition, did not lead to positive results. Omitting the details, we should note that the situation is perfectly similar to the one described in Examples 1 and 2.

Application of the formal astronomical dating method and extension of the search time interval from 600 B.C. to A.D. 1600 permit us to draw the following conclusions [2].

(1) There exist exact astronomical solutions fully satisfying all conditions 1-6.

(2) There are only three exact solutions for the time interval from 600 B.C. to A.D. 1600.

(3) These solutions are (a) the night of September 4 to September 5, A.D. 415, (b) the night of September 4 to September 5, A.D. 955, and (c) the night of September 4 to September 5, A.D 1020 ([2], Vol. 5, pp. 266-272).

(4) For a small perturbation of the initial data, i.e., while considering lunar eclipses occurring not only at night but also at sunset, there arises only one more possible solution, (d) the night of September 4 to Septemer 5, A.D. 434.

24

Problems 0/ Ancient and Medieval Chronology

Chapter 1

Yet another solution is theoretically possible, namely, the lunar eclipse of September 4 to September 5, 106 B.C.; however, it possesses the phase 5"9, which is far too small. If we select those with greatest phases from the above astronomical solutions (see condition 5 of the problem), then two of them are ideally suitable, namely the eclipse of A.D. 955 with the phase of 16"1 and that of A.D. 1020 with the phase of 18"1, the latter being still more adequate than the former.

Upon further perturbation of the initial data of the problem, namely, also considering the night of September 3 to September 4, four other new astronomical solutions present themselves; however, they all relate to the medieval period, occurring in A.D. 453, 936, 1451, and 1416. A perturbation of the initial data in the other direction, namely, considering the night of September 5 to September 6, is impossible, which follows from Livy's text.

2.8. An example: the evangelical eclipse described in the New

Testament in connection with the Crucifixion

Example 4. Let us consider the characteristics of lunar eclipse 36 (see [8], pp. 200-201) described in the New Testament, i.e., the so-called evangelical eclipse (Mt 21:45, Mk 15:33, Lk 23:44-45).

(1) The eclipse was lunar.

(2) It was related to the spring equinox or occurred on the eve of Passover (In 19:14, 19:30-34). For the present, we do not distinguish between Easter and Passover.

(3) It occurred on Friday during Passover (In 19:14, Mt 21:62). (4) It lasted for about three hours (Mk 15:33-34).

(5) It lasted from 0 A.M. to 3 A.M. according to the modern count of hours.

This condition is sometimes questioned in the chronological literature; however, there are valid reasons to believe that the eclipse started approximately at midnight (see [2]).

The traditional solution given in [8] is April 3, A.D. 33 (the date of the crucifixion). However, just like in the above examples, this solution does not satisfy the data of our problem. Namely, although conditions 1, 2, and 3 are fulfilled (the eclipse occurs on the eve of Passover), conditions 4 and 5 are not. In particular, the phase of this eclipse (for the observation zone, i.e., Jerusalem) is so small that it could have been observed only for several minutes as the umbra was already sliding off the rim of the lunar disc.

In spite of the quite controversial descriptive characteristics of this lunar eclipse and the conjecture of some medieval authors and annalists, such as Synkellos, Phlegon, Africanus, and Eusebius, that it had in reality been solar (Lk 23:45), we can, nevertheless, endeavor to apply the formal astronomical dating method. We obtain (see [2]) that the above problem does have at least one solution.

(1) In the time interval from 200 B.C. to A.D. 800, there really exists a lunar eclipse satisfying conditions 1-5.

§2

The Moon's Elongation and Ancient Eclipses

25

(2) In the time interval from 200 B.C. to A.D. 800, this astronomical solution

. .

IS umque.

(3) The solution is March 21, A.D. 368. The (super total) eclipse phase is large and equals 13"3.

It should be noted that the calculations with the purpose of discovering a suitable lunar eclipse were made in the interval only up to A.D. 800 [2]. As a matter of fact, Morozov believed that the related historical events (the crucifixion) could not have occurred later than A.D. 800 ([2], Vol. 1, p. 97). Without imposing this restriction a priori, we extended the calculations along the time axis to embrace the whole interval up to A.D. 1600.

The author has therefore found only one more possible astronomical solution of the problem: the eclipse of April 9, A.D. 1075. There are no other solutions. The eclipse of A.D. 1075 did occur on Friday on the eve of Passover, which was on April 5, 1075. However, its phase was small, <b = 4"8. The moment when half of the period of the eclipse elapsed was at 23 hours and 18 minutes GMT, which means, in particular, that, in the latitude of Rome, for example, it occurred at about midnight. The coordinates of the zenith were lat. 100 N and long. 80 W. Its date (April 3) coincides with the canonical one of the evangelical eclipse (see, e.g., [8], p. 200). The date April 3 is regarded as canonical and traditional. Besides, the lunar eclipse of A.D. 1075 occurred precisely on the eve of Passover, which is consistent with the requirements of the tradition that assumes that the crucifixion occurred on the eve of Passover. Recall that the traditional astronomical solution of April 3, A.D. 33, is also two days prior to Passover, which occurred on April 5, A.D. 33.

2.9. The oscillation of a new graph of D" about one and the same

value. No nongravitational theories are necessary

We now list certain results. If t~l denotes the date of an eclipse obtained by the formal astronomical dating method, and t~~1. is the traditional date given, e.g., in the canons of Ginzel and Oppolzer [8], [12], then we obtain the following result. As it turns out, for all the eclipses of the second class (i.e., those thoroughly described in ancient texts), the following important inequality is valid, namely,

told tnew

eel. > eel.'

Moreover, this shifting of the dates of ancient eclipses of the second class forwards in time (see above) is carried ou t in the following uniform manner. All eclipses of the second class traditionally dated in the interval from 900 B.C. to A.D. 400, turn out to be mechanically shifted further ahead than A.D. 400 into the Middle Ages. It happens that the percentage of eclipses of the first class (inaccurately and vaguely described in the sources) is extremely high from A.D. 400 to 900. The dates of such eclipses are either almost incalculable by astronomical means due to the imprecision of written evidence about them or are shifted forwards again. Starting with A.D. 900, the new dates t~~l are satisfactorily consistent with t~~1 .. Only beginning with A.D. 1300 does this

26

Problems of Ancient and Medieval Chronology

Chapter 1

correlation get quite reliable. The complete picture of shifting the dates of ancient eclipses forwards on the time axis turns out to be rather confusing and complicated because of the nonuniqueness of astronomical solutions. Nevertheless, it obeys a certain law which we shall describe below.

Recall that the behaviour of the graph of D"(t), computed by R. Newton on the basis of the prior eclipse dates t~~t., also makes it possible to naturally distinguish certain time intervals characterized by an essentially different behaviour of the parameter D" on the time axis. Remember that the values of D" oscillate about zero (though very few of them have been computed) earlier than A.D. 400. Then, in the time interval A.D. 400-1000, a considerable chaotic variance of the parameter values is noticeable. Finally, since A.D. 1000 (and still more since A.D. 1300), the values of D" are already close to those known today. Thus, we obtain one important conclusion. It turns out that the time limits discovered for the behaviour of D"(t) almost coincide with the characterization of different shifts of the dates of ancient eclipses forwards on the time axis. This indicates a possible relation between the two important effects, namely, (1) the square wave in the behaviour of D" and (2) the shift of the dates of ancient eclipses forwards due to the application of the formal astronomical dating method.

Let us ascribe to each ancient eclipse from historical sources its new date t~~t calculated by the formal astronomical dating method. The recalculation of the values of parameter D" on the basis of these new dates, which I carried out in [4], is shown in Fig. 7.

o

New graph

40

---------,---.------

I.~ ~

11:: I ,e S I ,~ as I I=> '"0 ,

: • : Reliable :v.: data

, ...

:Vb

: O:O~ nO

, ••........ __ -.J

No data

20

·20 '- ---L-.-__.------j..._.~

·1000

o

+1000

+2000 t

Figure 7. New curve for D"(t) an the interval from 900 B.C. to A.D. 1900 (A. T. Fomenko).

We see that the replacement of the date t~~1. by t~~t' not only shifts the historical events and texts describing the eclipses forwards in time, but also, in most cases, leads to the identification of "former eclipses" (previously

§2

The Moon's Elongation and Ancient Eclipses

27

regarded as ancient) with medieval ones known from other sources. Moreover, the "ancient eclipses" overlap with the medieval ones, many of which were used by R. Newton for the computation of the former graph of D" for the period from A.D. 400 to 1900, thus adding new texts previously treated as antique and traditionally dated as older to the formerly medieval one, with information about medieval eclipses being used to determine the "medieval part" of the graph of D". Therefore, we now have to take into account these new additional data (characteristics of eclipses) which earlier were ascribed to other, presumably ancient eclipses in recalculating the medieval part of the graph of D". A priori, such an extension of the list of descriptive characteristics of certain medieval eclipses could have led to a contradiction with their characteristics known earlier from medieval texts, and, in particular, to a change of the formerly medieval values of DII(t) from A.D. 400 to 1900. However, the detailed investigation of all descriptive characteristics, both old and new, has shown that the formerly medieval values of D" from A.D. 400 to 1900 are almost unaltered.

Based on this result, we can make the following conclusions: The new curve of tr from A.D. 400 to 1900 practically coincides with the former. From 900 B.C. to A.D. 400, the new curve of iJ" is simply undetermined, since no reliable eclipse dates t~~t exist for this time period.

The new graph of iJII is qualitatively different from the former. We see that tr varies along a smooth and nearly constant curve (horizontal in Fig. 7) which oscillates about one and the same value -18" /century2 from A.D. 900 to 1900. The parameter tr undergoes no sharp change. It invariably retains approximately the modern value. No nongravitational theories of the type suggested in [9] are therefore necessary. It is interesting that the variance of the values of ir, quite insignificant from A.D. 900 to 1900, gradually increases in shifting to the left from A.D. 900 to A.D. 400. In our opinion, this fact indicates the vagueness and insufficiency of the observational data contained in ancient historical texts describing this period. Then (see Fig. 7), to the left of A.D. 400, the zone starts where reliable observational data (which may have survived to the present day) are absent, which reflects the natural distribution in time of astronomical observational data supplied by the ancient chroniclers. Apparently, the exactness of observational data and textual descriptions from A.D. 400 to 1000 was extremely low. The precision of observations and descriptions started to improve afterwards as the technology and the instruments improved and became more sophisticated, which is reflected by the gradual decrease in the variance of the values of tr, Finally, in the era of developed astronomy, we see that the curve of tr is all but aligned, and stable from A.D. 900 to 1900.

All the previous results show that the dates of ancient eclipses are shifted forwards in their redating, with the magnitude of the shifts being expressed by the positive quantities t~~l- t~1 .:

28

Problems of Ancient and Medieval Chronology

Chapter 1

2.10. Three rigid "astronomical shifts" of ancient eclipses

We see that, for different eclipse groups previously regarded as ancient, the values of the shifts forwards are much different, which leads to a great confusion in the general picture of redating ancient eclipses. Nevertheless, it so happens that the system of redating and shifting the dates forwards in time, which appears rather chaotic at first glance, allows the derivation of an important regularity, discovered by the author. This law completely agrees with the system of the three rigid chronological shifts in ancient history, discovered by the author in [5] on the basis of quite different statistical investigations (of nonastronomical character), which we shall not define here due to lack of space. However, we formulate only the final result, while referring the reader to the aforementioned work.

Consider the historical time interval from 1600 B.C. to A.D. 1700. Cover it with a system of line-segments denoted arbitrarily by the letters K, H, TI, C, T. Placed on the time axis, they are of different length, and the same letter, or interval, can be repeated several times. Moreover, different intervals designated differently can overlap one another, in which case we will denote the letters of the overlapping intervals by "fractions". In Fig. 8, they are represented for better visualization by different geometrical symbols such as rectangles, triangles, and trapezia. The covering of the obtained time axis will be referred to as the global chronological diagram (GCD). It has the following form:

GCD=T K T H T T K T H T K T T K T H T T PT C
- - - - - -
P C P n n c
-
c P
-
P We will also indicate the end-(boundary- )points of all these intervals, for which we rewrite the chronicle line in the GCD, marking the dates of the beginning and end of each epoch interval. For example, the symbol 1570 T 1550 means that the given copy of the epoch interval T starts in 1570 B.C. and ends in 1550 B.C. If the left-hand number is greater than the right-hand one, then we mean the years B.C.; otherwise, the years A.D. are meant. Thus, we have:

K GCD = 1570 T 1550, 1460 K 1240 T 1226 H 850 T 830, 760 T 753 P 523,

523 T 509 H 82 T 27 27 B.C. K 217 A.D. 217 T 251,

500 C 250 82 B.C. P 217 A.D.

306 K 526 526 T 552 552 H 915

270 T 306 333 n 526 681 n 887 915 T 925 932 T 954

c 650 P 900 ' ,

240 P 580

962 962 P 1254 T 1273 1213 C 1619. 950 C 1300

§2

The Moon's Elongation and Ancient Eclipses

~ § ~ 8 - 8 ~ ~ 8 8 ~ ~ ~ I 8 § § ~
- I";- - - -
~ - - - , + + - - - -
. , , + + + +
Old and New Testament Biblical history shifted
upwards
T European history and chronoligy

E:

T T T T

C.: 040U011""" -C --. 1.778·year shift

T T T

0,: [KJ1/ H \! f!l 11"'- c-I 1.053-year shift

T T T

c, [K] j [OJ i 1 p I 11- C-, 333·year

_- - shift

c,

T T
0' ,

0 T
Co: 0L H ~I p IAI c
Original chronicte
·1700 ·1300 ·900 -500 ·100 +100 +500 +900 +1300 +1700 Figure 8. Statistical duplicates in European history and biblical history (shifted fonoards). Three chronological shifts.

29

30

Problems of AncIent and M edzeval Chronology

Chapter 1

0 fR ~ ~ ~ ~ Ie C"') ~ ~ ~ ,... iii ~ CD tit ~ ll) !i ~ ~ CJ)
t; ~ op- ~ ~ lO
C\I ~ ~
op- op- op- - I 0 I + + + + + - - -
I I I 0 + + +
T K T H T T K T H T K T T K T H T T P T C
- - - - - -
p C p n n c
- -
C p
-
P
K T H T T P T C C4
1.778-year shift K T H T P T C c,
1.053-year shift K T n T p T C c,
333-year shift K H
- T - T P T C C1
p n
p T T C'
H
Different letters represents different periods K - P T C Co
n Fagure 9(1}. Formal decomposition of the "modern textbook" of ancIent Iustory into the sum of four "isomorpluc fibres".

The identical letters indicate the epoch duplicates. For example, the epoch 500 C 250 duplicates 950 C 1300. The GCD resolves into the composition of the three rigid shifts shown in Fig. 9( 1). By adding the horizontal lines vertically (i.e., by superimposing them) and by identifying the same letters placed over each other in the same column, we obviously obtain the complete GeD. Thus, we can write symbolically that the GCD = C1 + C2 + Ca + C4, where the chronicle line C1 equals Co+C'. All four chronicle lines Os, C21 C«, and C4 are almost indistinguishable, Le'I they consist of the same sequences of letters. The chronicle C2 is meanwhile glued to C1 with a backward shift of about 333 years, C3 to C1 + C2 with a backward shift of approximately 1,053 years, and C4 to C1 + C2 + C3 with a backward shift about 1,778 years in length. All three basic rigid shifts are counted from one and the same point (Fig. 7).

We obtain that the three chronological shifts discovered on the basis of statistical methods of non astronomical character (see the author's paper [5])

§2

The Moon's Elongation and Ancient Eclipses

31

manifest themselves also in a rather complicated picture of a forward shift of the revised ancient eclipse dates. It should be noted, as can be seen in the above examples, that for many ancient eclipses, there usually exist several different astronomical solutions, all "enjoying equal rights". Hence, it is apparent that the inverse problem, namely restoration of the three chronological shifts on the GCD based only on astronomical data, cannot be solved reliably today.

2.11. The complete picture of astronomical shifts

We can now outline the complete picture of shifting the redated ancient eclipse dates forwards.

If the traditional date of an eclipse t~1. is represented by a point in the time interval from 1600 B.C. to A.D. 1700, then it will necessarily fall on one of the chronicle lines Gl, G2, G3, and G4, which make up the GCD. G41 C3, and C2 are moved forwards in the inverse shift by 1,778, 1,053, and 333 years, respectively. Meanwhile, they are again identified with the chronicle Cl embracing the events from A.D. 300 to 1619. The following important statement turns out to be valid.

The forward shift of an astronomically redated eclipse by the quantity t~t - t~~1. usually coincides with that of one of the chronicles C2, G3, and C4 into which the date t~~1. fell originally. However, the value of t~t - t~1. sometimes coincides with the difference or sum of certain of the three basic chronological shifts. In other words, the astronomical and chronological shifts are consistent.

Thus, all the dates t~1. of astronomically redated ancient eclipses are broken into certain groups, each of which is shifted forwards by about 1,778, 1,053, and 333 years, or their difference or sum. It is important that the relative position of the dates t~~1. inside each of these groups is then retained qualitatively and that the eclipses fall into the time interval A.D. 400-1800.

This shift of the dates of ancient eclipses can also be described thus: If an eclipse has fallen into some copy of the interval P on the GCD, then, redated astronomically, its new date t~~l is shifted forwards and gets into one of the other replicas of P to the right of the traditional date t:1 .. Further, it turns out that the majority of the dates of eclipses previously regarded as ancient are shifted into the time interval A.D. 800-1700 [5].

2.12. The coincidence of the astronomical shifts with the three basic

chronological shifts in the global chronological diagram

We illustrate our statement regarding the coincidence of the astronomical shifts, arising in redating the eclipses, with the three basic chronological shifts on the GCD by several typical examples.

(1) The triad of eclipses described in Thucydides' History of the Peloponnesian War (see Example 1 above) is closely related to the historical events in ancient Greece, traditionally dated for about the 5th century B.C. [8], [12]. As we have noted, the first eclipse mentioned by Thucydides is placed by

32

Problems of Ancient and Medieval Chronology

Chapter 1

traditional chronology according to the ancient sources at around 431 B.C. [8]. Redated astronomically, its date was shifted forwards, and the eclipse was identified (see the second astronomical solution in Example 1) with the medieval one of A.D. 1039. Therefore, the value of the astronomical forward shifts of the eclipse date is approximately equal to 1039 + 430 = 1469 years, which is close to the difference of two basic chronological shifts on the GCD, namely, 1778 - 333 = 1445 years.

(2) The lunar eclipse of Livy from Example 3 is closely related to the events in ancient Rome, which occurred, according to traditional chronology, in the middle of the 2nd century B.C. [8], [12]. Astronomically redated, it was also shifted forwards, and the eclipse was identified with the medieval one of A.D. 955, the first astronomical solution in Example 3. Thus, the astronomical shift forwards is approximately 1,100 years, which is close to the basic chronological shift on the GCD of 1,053 years. The shift makes the chronicle line C3 coincident with Cl.

(3) The lunar eclipse of Example 4 (see above) is inseparably linked with the events (the crucifixion) occurring, according to the traditional chronology, in about A.D. 33 [8].

Astronomically redated, it was shifted forwards, too, and identified either with that of A.D. 368 or 1075. In the former case, the astronomical shift is 335 years, which all but exactly coincides with the basic chronological 333-year shift of the chronicle line C2 on the GCD; in the latter, the shift forwards is about 1075 - 33 = 1042 years long, which is close to the astronomical shift by 1,053 years.

We should not think that the three astronomical shifts discovered reflect any periodicity in the distribution of authentic eclipse dates belonging to the past. As a matter of fact, in case (1) in Example 1, the astronomically exactly calculated date is only t~~t, A.D. 1039. The first and traditional date of 431 B.C. is astronomically inexact and appears only under the pressure of chronological tradition, which a priori relates the historical events linked with Thucydides' eclipses to the 5th century B.C. However, the above date of 431 B.C. was obtained by the astronomers as the result of a forcibly strained calculation in order to satisfy the requirements of tradition [8].

§3. Traditional Chronology of the Flares of Stars and the Dating of

Ancient Horoscopes

3.1. Ancient and medieval flares of stars. The star of Bethlehem The important fact is that the three discovered chronological shifts on the GCD are very consistent with many other astronomical data not yet linked to the eclipses directly. For example, I have analyzed the traditional chronology of the flares of the so-called novae and supernovae. Let us list the flares of all such stars regarded as reliable in accordance with [3], [6]: 2296 and 2241 B.C., A.D. 185,393,668,902,1006,1054,1184, and 1230, and those of the 16th century (see Kepler's list). The so-called star of Bethlehem described

§3

Flares of Stars and the Dating of Ancient Horoscopes

33

in the New Testament (Mt 2:3), i.e., a flare occurring in about A.D. 1, is also usually added. The study of the astronomical situation in about A.D. 1 with the purpose of discovering the remains of this famous "star" was taken up, for example, by J. Kepler and L. Ideler (see [1], pp.128-129). In the chronological shift backwards by 1,053 years (which corresponds to the superposition of the chronicle C3 on C1 + C2), the time interval A.D. 962-1250 is placed on the time interval 91 B.C.-A.D. 197 (see the chart).

Ancient History from 91 B.C. to A.D. 197 Medieval History in A.D. 962-1250
The complete list of the flares of stars The complete list of the flares
in this epoch, fixed in antique sources, of stars in this epoch, fixed in
IS: medieval sources, is:
the famous flare in A.D. 1; the flare in A.D. 1006;
- the famous flare in A.D. 1054;
the flare in A.D. 185. the flare in A.D. 1230.
(1) The famous "star" of A.D. 1 when (1) The famous flare in A. D. 1054
Christ was born (Mt 2:2, 8, 9-11). of the supernova in Taurus.
Hildebrand was "born" as
reformer of the Church. The dates of these flares are ideally coincident in shifting by 1,053 years (see the GCD).

(2) The flare in A.D. 1 was visible "in the east" (Mt 2:2,9) ("The star which they saw in the east").

(3) The flare of the star in A.D. 185.

(2) The flare of the star in A.D. 1054 was visible "in the eastern sky" [6].

(3) The flare of the star in A.D. 1230

The dates of the flares in A.D. 185 and 1230 can also be made coincident under the same 1,053-year chronological shift with a difference of only 8 years.

(4) The star flare in A.D. 185 lasted 7 months [3], [6].

(4) The star flare in A.D. 1230 lasted 6 months [3], [6].

Thus, the dates (regarded as trustworthy) of all star flares from 900 B.C. to A.D. 390 are obtained from those of the medieval flares of stars from the 10th to the 13th century under the backward 1,053-year chronological shift. This new independent corroboration of the existence of the global 1,053-year chronological shift is interesting, since we have analyzed here the dates given by the written sources of quite irregular astronomical phenomena. Note that the previous traditional dating of these flares was carried out on the basis of a written chronological tradition of "non astronomical origin" .

34

Problems of Ancient and Medieval Chronology

Chapter 1

3.2. Astronomical dating of ancient Egyptian horoscopes

We now turn to an analysis of the astronomical results of [2] of dating the zodiacal positions of the planets, which are described in certain historical sources, namely, the so-called horoscopes. Recall that all planets are placed near the ecliptic, relative to the stars (i.e., on the fixed astral sphere), and their position can be calculated similarly to the method of determining the dates of ancient eclipses. That is, we have to fix the positions of observable planets relative to the zodiacal constellations at some modern moment of time. Then, plotting integral multiples of the (known) sidereal periods of the planets backwards, we can, in principle, calculate horoscopes of the past, i.e., the position of the planets relative to the zodiac at a prescribed moment of time.

Thus, if a horoscope is described in some historical source, then, proceeding analogously with the procedure of calculating ancient eclipse dates, we may attempt to date it. To this end, we have to compare its description in a historical text with the calculated tabular horoscopes and attempt to find a horoscope with the same characteristics.

The seeming simplicity of this idea is made very complicated by the difficulties of the calculations and, which is most important, by various secondary reasons of "non astronomical" character similar to those with which we are already familiar.

In [2], Morozov analyzed the traditional dates of all the basic horoscopes fixed in the surviving ancient sources. Omitting the details, we inform the reader that the result was the same forward shifts of their dates obtained astronomically as occurred previously in the case of ancient eclipses. We give a typical example.

The well-known Egyptologist W. Flinders Petrie in 1901 discovered in Upper Egypt (Athribis) an ancient Egyptian interment dated by the traditional chronology from the 1st century B.C. to the 1st century A.D. The interment was found to contain two graphic images of the planets on the zodiac. The two horoscopes probably indicated the dates of the two tombs. The specialist Knobel [7] attempted to date the horoscopes within the a priori time interval from the 1st century B.C. to the 1st century A.D. However, no exact astronomical solution was found. We make the precise statement that the a priori interval was determined, proceeding from the style and character of the inscriptions in the grave, due to which Knobel was forced to offer only quite approximate values, namely, A.D. 52 and 59. Knobel noted the imprecision, because the position of Venus at that time was different from its representation in the tombs.

Then the Russian astronomer M.A. Vilyev analyzed all the horoscopes from 500 B.C. to A.D. 600; however, he discovered no exact astronomical solution for the Athribis horoscopes. Nevertheless, the extension of the search time interval and the application of the formal astronomical dating method led Morozov to the discovery of an exact astronomical solution, namely, A.D. 1049 and

§3

Flares of Stars and the Dating of Ancient Horoscopes

35

A.D. 1065 ([2), Vol. 6, p. 745). It is important that it is unique in the whole historical interval.

3.3. Astronomical dating of the horoscope described in the Book of

Revelation

Consider another example. An exact astronomical solution for the horoscope described in the Book of Revelation was suggested by Morozov [2]. Though its descriptive characteristics can be extracted from the Book of Revelation only with some controversial interpretation, this circumstance does not encumber the application of the formal astronomical dating method. As it turns out, there are only two exact astronomical solutions in the whole historical time interval, namely, A.D. 395 and A.D. 1249, although the latter was rejected as "too late" ([2], Vol. 1, p. 53). Besides, it is less satisfactory astronomically.

My analysis of the whole collection of horoscope datings given in [2] has shown that their forward chronological shift obtained by the formal astronomical method is also due to the same three basic ones on the GCD (Fig. 8). For example, the forward shift of the dates of the two Athribis horoscopes is approximately 1,000-1,050 years. Recall that traditional chronology dates them to about the beginning of the present millennium. Thus, the astronomical shift is close to the chronological one of 1,053 years (see Fig. 67 in Vol. 2 of this book).

The same situation occurs in our last example, that of the Book of Revelation. The second of the two astronomical solutions for its horoscope, A.D. 1249, yields the forward shift of its creation by about 1,050-1,100 years. Note that the approximate traditional date of writing the Book of Revelation is, according to A. Harnack, Eberhard, J. Martineau, and van Eising, the second half of the 2nd century A.D. In this case, the value of the astronomical shift is thus also close to that of the chronological one by 1,053 years (Fig. 8).

In conclusion, we indicate another interesting astronomical fact [2], which also turned out to agree with the discovered decomposition of the GCD into the sum of three shifted chronicles.

The first Latin edition of Ptolemy's famous Almagest (published in A.D. 1537 in Cologne) contains a catalogue of stars with the indication of their longitudes and latitudes, i.e., coordinates on the celestial sphere. As is clear from the text, the catalogue was made by Claudius Ptolemy himself in the second year of the rule of the Roman emperor Antoninus Pius, traditionally related to A.D. 138-161. It turns out that there exists a reliable method to determine from its star catalogue the date when the Almagest was written [2]. Since it contains ecliptic star coordinates, we can make use of the generally known property of stars to annually increase their longitudes by 50"2 (due to precession). Dividing the difference between the longitudes indicated in the Latin edition and those of the present day by 50"2, we shall obtain the required date. This simple calculation unexpectedly shows that the longitudes of the stars listed in the first Latin edition of the Almagest were observed or

36

Problems of Ancient and Medieval Chronology

Chapter 1

Figure 9(2). Portrait of the "Imperator Caesar Diuus Maximilianus Pius Felix Augustus" (A. Durer).

recalculated by its author in the 16th century A.D. Hence, these astronomical data belong to the time when the book was published.

Morozov's work ([2], Vol. 4), supplies many other arguments in favour of the conjecture that this text was created by the astronomers of the 10th to the 16th century A .0. In our case, the forward shift of the date of writing the Almagest is about 700 or 1,390 years if obtained astronomically. Meanwhile, we compare the date A.D. 1530 (epoch of the first editions of the Almagest) with A.D. 140 (second year of Antoninus Pius' rule). We obtain 1530 - 140 = 1390 years. The value of this shift is also completely consistent with the GCD

§3

Flares of Stars and the Dating of Ancient Horoscopes

37

(Fig. 7), since it practically coincides with the sum of the two basic chronological shifts: 1053 + 333 = 1386 years.

In such a forward shift of dates, the period of Antoninus Pius' rule falls into the epoch when the first editions of the Almagest appeared, namely, A.D. 1528, 1537, 1515(1), 1538, 1542, and 1551. Note, in conclusion, that immediately before this medieval epoch, the emperor Maximilian I Pius (!) Augustus (A.D. 1493-1519) had ruled in the medieval Empire of the Hapsburgs. It is interesting that he was a contemporary of A. Diirer, the creator of the astrographic charts that accompanied Ptolemy's Almagest. The prints were made by Durer in about A.D. 1515. Therefore it cannot be excluded that it was under Maximilian Pius that the astronomical observations fixed in the Almagest were carried out (Fig. 9(2». The statistical analysis of the latitudes in the star catalogue of the Almagest was made in the recent paper [15]. The result is as follows: the latitudes in the star catalogue of the Almagest were observed somewhere in the time-interval A.D. 600-1300. See also [16].

References

[1] Context. Nauka, Moscow, 1978 (in Russian).

[2] Morozov, N .A., Christ. Gosizdat, Moscow-Leningrad, 1926-1932 (in Russian).

[3] Pskovsky, Y.P., Novae and Supernovae. Nauka, Moscow, 1974 (in Russian).

[4] Fomenko, A.T., "On the computation of the second derivative of the moon's elongation", in Controllable Motion Problems: Hierarchal Systems. Perm University Press, Perm, 1980, pp. 161-166 (in Russian).

[5] Fomenko, A.T., "Certain statistical regularities of information density distribution in texts with scale", in Semiotika i Informatika, Vol. 15, VINITI, Moscow, 1980, pp. 99-124 (in Russian).

[6] Shklovsky, I.S., Supernovae. Wiley, New York-London, 1968.

[7] British School of Archaeology in Egypt and Egyptian Research Account.

London, 1908.

[8] Ginzel, F., Spezieller Kanon der Sonnen- und Mondfinsternisse fur Liindergebiete der klassischen Altertumswissenschaften und der Zeitraum von 900 vor Chr. bis 600 nach Chr. Berlin, 1899.

[9] Newton, R., "Astronomical evidence concerning non-gravitational forces in the Earth-Moon system". Astrophys. Space Sci. 16, 2(1972), pp. 179- 200.

[10J Newton, R., Ancient Astronomical Observations and the Accelerations of the Earth and Moon. Johns Hopkins Press, Baltimore, 1970, pp. 32-47. [11] Newton, R., "Two uses of ancient astronomy". Phil. 1rans. Royal Soc., Ser. A, 216(1974), pp. 99-116.

[12] Oppolzer, T., Kanon der Finstemisse ... Mit 160 Tafeln. K.K. Hof- und Staatsdruckerei, Wien, 1887.

38

Problems of Ancient and Medieval Chronology

Chapter 1

[13] The Place of Astronomy in the Ancient World: Joint Symposium of the Royal Society and the British Academy, organized by D.G. Kendall et al. Oxford University Press for the British Academy, Oxford, 1974.

[14} Zech, J., Astronomische Untersuchungen uber die wichtigeren Finsternisse, uielche von den Schriftstellem des classischen Altertums erwahnt werden, etc. S. Hirzel, Leipzig, 1853.

[15] Fomenko, A.T., Kalashnikov, V.V., Nosovsky, G.V. "When was Ptolemy's star catalogue in the Almagest compiled in reality?" Statistical Analysis. Acta Applicandae Mathematicae 17(1989), pp. 203-229.

[16] Kalashnikov, V.V., Nosovsky, G.V., Fomenko, A.T. "The dating of the Almagest based on the variable star configurations". Doklady Akad. Nauk SSSR 301, 4(1989), pp. 829-832 (in Russian).

CHAPTER 2

NEW STATISTICAL METHODS FOR DATING

§4. Certain Statistical Regularities of Information Density Distribution in Texts with A Scale 1

4.1. Text with a scale. The general notion

4.1.1. In the present paper, we list the results of a series of statistical investigations carried out by the author, and which led to the discovery (for narrative texts) of new statistical invariants such as the laws of information density conservation.

By a text X with a scale P, we understand a text (e.g., a narrative one) endowed with a fixed program plan whose items are numbered by a parameter t. Meanwhile, we shall require that X should admit a unique partition associated with this parametrization (program), and such that (1) to each value of the parameter t E P, a certain part X(t) of the text may necessarily correspond; and, vice versa, (2) each phrase (or word) of the text T may necessarily belong to one and only one fragment X (t) for a certain t.

We call such a program plan a scale, or parametrization of the text X.

We will say that the text X is parametrized by t, with X = U X(t), and

X (tI) n X(t2) = 0 if t1 ;f; t2. teP

Such parametrized texts X are well illustrated by those with historical character, e.g., historical monographs, chronicles, annals, college textbooks, diaries, and so forth. In these cases, the part of the parameter t is played by time, i.e., the dates of the events described (according to a system of chronology or any other method of their dating).

Still, the above concept of a text with a scale is considerably more general than in the given example. In particular, the parameter t may range over more complicated domains than the set of natural numbers (dates of events). For

1 First published as an article in Semiotika i InJormatika, Vol. 15, VINITI, Moscow, 1980, pp. 99-124 (in Russian).

39

40

New Statistical Methods for Dating

Chapter 2

example, t can possess a continuous domain in the case where we take as a text with a scale the description of some physical continuous process parametrized by continuous time. Then we can consider the description of the instantaneous state as a "textual fragment" X(t) when the process of interest is going on. The description may be "composed" of some system characterizing the process continuously.

Further, for the applications considered below, the following example is very important. We can assume that a scale P consists of a sequence of disjoint intervals (Ak, Bk) with integral endpoints. In other words, two intervals (Ak, Bk) and (Ac, Be) are disjoint if k :f: c and the union of all of them makes up the initial segment (A, B), with the parameter t ranging over their entire sequence (see the details in Section 4.2.2).

4.1.2. Let X be a certain historical (or, more generally, narrative) text with a scale P, represented as the union of the fragments X(t). For simplicity, let the parameter t range over the positive or negative integers from A to B. For example, we can assume that X describes the events from the year A to the year B, though the quantities t, A, and B can then be measured not only in years, but, say, in months, days, or hours.

4.2. Information characteristics (i.e., informative functions) of a historical text. Volume function, name function, and reference function

4.2.1. Now, we associate each value of the parameter t with a set 11(t), f2(t), ... ,fN(t) of the formal information characteristics of part X(t) of a historical text X, which may describe the events of one year t. As we shall now see, the quantitative information characteristics fi(t} can be quite multifarious. We illustrate this with some basic examples.

Example 1. Let /1(t) = vot:(t), where voIX(t) is the number of lines, pages, signs, or words making up a textual fragment X(t). We then, obviously, get !t(t) > 0, and LA<t<B /1 (t) = 1, with V = vol X denoting the total volume of the text X without illustrations, diagrams, or bibliography. The normalizing condition Lt /1(t) = 1 often happens to be convenient in various instances of such averaging and comparing and is a discrete analogue of the normalizing condition in the case of a continuous parameter, namely

J: h(x)dx = 1.

Example 2. Let /2(t) = "~), where 8(t) is the number of references to a year t in the entire text X, and S that of all the dates (years).

Example 3. Let /3(t) = ml;), where m(t} is the number of the names of historical characters mentioned in a textual fragment X(t), and M that of all the names of all the historical characters from the text X. We shall sometimes also count the number of references to the names of historical figures in the fragment X(t), i.e., we shall take into account their multiplicity, or frequency, of use in the text.

§4

Information Density in Texts

41

Example 4. Let f4(t) = mii~), where mj(t) is the number of references

J

to some concrete name mj in a textual fragment X(t), and Mj is the total

number of references to mj in the entire text X. In general, this name can be ascribed to different historical characters.

We shall give other important examples of the functions fi(t) below. We call them informative, or frequency functions, defined by X.

4.2.2. To determine li(t) means to formally compute an indicated numerical characteristic such as volume, the number of names, and so on, of all textual fragments X(t) of some concrete historical texts X, which may be arranged so that only the partitions of X into the sum of the fragments X(AI:' BI:) are specified uniquely, with X(A1:1 B1:) denoting that part of X in which the events of the interval (Al:I BI:) with integral endpoints are described. Meanwhile, their description inside X(A1:' B1:) may not be specifically related to the years composing (Ak' BI:), due to the convenience of considering an enlarged scale for a parameter which does not range over separate years bu t over separate time intervals, in which case we will denote it by T. It should be noted that, generally, there can exist several different scales for the same text X. Thus, the choice of a particular scale in X may vary in accordance with the problem posed. If a parameter T ranges over a sequence of time intervals (Ak,Bt), then the normalized volume function h(T) is determined

as h (T) = vol ~~~ ~k :"1c_A/c • In other words, we compute the average value of the textual volume function in (At, B1:), where vol(At, Bt) and vol(A, B) are the values of the volume function in (A1:' B1:) and (A, B), respectively. The "averaged" functions for the other examples listed above are defined similarly.

For example, vol( AI: , B1:) is the number of lines, pages, or words making up the fragment X(A1:' B1:) that describes the events in the time interval (AI:, BI:)' Further, vol(A, B) is the total volume of the entire text X represented as the union of its disjoint fragments X(Ak' BIc).

Thus, each informative function I;(t) is given by a certain graph defined on an interval (A, B). All the informative functions under investigation are nonnegative.

4.2.3. Denote by I(t), or I(t, X), some informative function I;(t) for a text X, omitting the subscript i for brevity. Let two (or several) historical texts X and Y be given, which describe the events in an interval (A, B) of the history of one region (e.g., state, town, etc.). It is clear that the informative functions I(t, X) and I(t, Y) constructed for X and Y, respectively, will, in general, be different. This must be so, since their form is influenced by both the individual propensities of the authors and the general attitude toward the events characteristic of the time when the text was written. It is clear that a monograph on art history and one on military history devoted to the same period will accentuate various things differently, which can lead, for instance, to a different distribution of the textual volumes.

A natural question thus arises: How essential are these differences in texts with a statistical approach to the analysis of their informative functions? In

42

New Statistical Methods for Dating

Chapter 2

other words, can we discover certain general, invariant characteristics of the informative functions f(t, X) and f(t, Y), which do not depend (or depend little) on the authors' tendencies, and which are determined mostly by the time interval (A, B) and the region I'?

4.2.4. Let us investigate the behaviour of such important characteristics of the graphs of f(t,X) as the distribution of local maxima ("peaks"). Denote by ri(X) those years t of a time interval (A, B) in which the graph of f(t, X) attains local maxima, i.e., exhibits peaks. Let i vary from 1 to q(X), where q( X) is the total number of local maxima.

In the case where the scale of the text X is enlarged (see the above example), the informative function f is represented by a "steplike" graph. This step function is constant on each separate interval (Ak' Bk)' It is then convenient to take its midpoint as a local maximum attained in (Ak, Bk).

Consider the volume function vol X(t) as a basic example. Its value at a point t equals the volume of a textual fragment X(t), measured in pages, lines, or words. What is the meaning of "the volume function voIX(t) reaches a local maximum at a certain point ri(X) in an interval (A, B)"? It means that the year ri(X) represented as a point in (A, B) is described in the text X in more detail, on a greater number of pages, and with more particulars than the nearby years. What can explain such a difference in the description of different years? One of the possible (and, probably, basic) reasons for this phenomenon can be formulated as follows. The author of X possessed more information for the year ri(X) (e.g., historical texts or more extensive documentation) than for the neighbouring years. It could be for this reason that the events of the year ri(X) of the epoch (A, B) have been described in more detail.

4.3. A theoretical model describing the distribution of local maxima for the volume function of a historical text. Primary stock. The information density conservation law

In Section 4.2.4, we formulated a theoretical model (statistical hypothesis) describing the distribution of local maxima for the textual volume function. We now designate by G(t) the total volume of all the texts written by contemporaries about the events of a year t. We shall measure the volume in lines (pages, words, etc.). It is clear that, in general, we cannot reconstruct the original form of the graph of G(t) today, which is due to a loss of the old texts with time. It is only the part 1«t) of the primary information stock G(t) that has survived. K(t) will be called the surviving information (textual) stock regarding the year t. Denote by Ci the years in which the volume graph for the primary stock G(t) has reached local maxima. It follows that especially many texts were written in these years (for particular reasons which we do not discuss here), i.e., the contemporaries have recorded a particularly large amount of written evidence about them. Let us ask ourselves the following question:

What might be the statistical mechanism of loss and falling into oblivion of

§4

Information Density in Texts

43

textual information which leads to a gradual decrease of the amplitude of the volume graph for the primary stock C(t)?

We formulate the following hypothesis. Though, with time, the amplitude of the volume graph for the surviving textual stock decreases gradually (since ancient texts get lost and destroyed), more remains from those years Ci whose events were described by contemporaries in a considerable number of texts.

In other words, the years Cj in which the volume graph for the primary stock C(t) reached local maxima must be close to those points ki in which that of the surviving stock K(t) attains local maxima. In particular, c; and ki must be close to those years ri(x) in which the volume graph vol X(t) of the text X describing the events in a time interval (A, B) has reached local maxIma.

Since the volume graph for the primary stock C(t) is unknown, it is hard to verify the hypothesis stated in its present form. However, we can verify one of its important corollaries.

4.4. The correlation of local maxima for the volume graphs of

dependent historical chronicles. The surviving-stock graph

4.4.1. Namely, the years of local maxima ri(X) should be close to those denoted by ri(Y) for any two historical texts X and Y describing one historical epoch (A, B) in the history of the same region r.

In other words, the volume graphs vol X(t) and vol Y(t) must attain local maxima, i.e., form peaks, approximately at the same points (years) in the time interval (A, B).

We call this corollary the F -model. To substantiate it intuitively and informally, we assume that two texts X and Y describe the events in a time interval (A, B) in the history of one region. Let this period be considerably removed in time from the authors of X and Y. This means that they are no longer contemporaries of the described ancient events and, therefore, have to employ a collection of historical sources surviving from the historical epoch (A, B), i.e., the surviving information stock. However, we can assume (without making a gross mistake) that this surviving textual set (information stock) is approximately the same for both X and Y. Hence, the volume graphs vol X(t) and vol Y(t) must more or less simultaneously reflect the maxima of the graph of the surviving stock while simultaneously reaching local maxima in (A, B). Thus, a chronicler describes in more detail those years from which more texts have been preserved.

Recall that the years of the local maximum of the volume graph are characterized by especially much surviving information in comparison with the neighbouring ones. Another natural corollary to the basic hypothesis will be given in Section 4.7.2 (the K-model).

4.4.2. We should not think that the amplitudes themselves of the local maxima of the volume function (the absolute values of the textual volumes) of two historical texts describing the same events are close (even with normalizing).

44

New Statistical Methods for Dating

Chapter 2

Simple examples of concrete chronicles show that the values of vol X (ri) and vol Y(ri) may be considerably different. Thus, the absolute values of vol X(rd at the local maximum points rt vary upon altering the text X (i.e., depend on it), which shows that the discovery of the rough "invariants of historical epochs" should not be based on the absolute amplitudes of informative functions. The analysis of "fine amplitude invariants" was carried out by S.T. Rachev and the author.

4.4.3. Let us verify the theoretical F-model against the material of concrete texts (chronicles, etc.), for which we should first formulate the concept of proximity of two sequences of numbers r.(X) and raCY) of local maximum points for two texts X and Y, namely, their volume graphs can, in general, have different numbers of local maxima, and q(X) f; q(Y) (see above). However, as will now be shown, without loss of generality, we can put q(X) = q(Y), for which it suffices to assume that certain ri(Y) coincide with some new, additional local maxima to be added in the case where q(Y) < q(X). In other words, some maxima of the graph of vol yet) are declared to be multiple, i.e., we assume that several maxima coincide.

Equalizing the number of local maxima for two volume graphs can be carried out differently and in accordance with the supposedly multiple maximum points and their multiplicity. We now choose a particular method for equalizing the number of maxima. In the following, we shall perform the minimization of the functions in question with the help of all such equalizing methods.

4.5. Mathematical formalization. The numerical coefficient d(X, Y), which measures the "distance" between two historical texts X and Y

4.5.1. We now describe the mathematical formalization. The points ri(X) break the time interval (A, B) into smaller intervals whose lengths are given by the integers aI, a2,'" ,ap, where al = rl - A, ai = ri - ri-l for 2 < i < q(X), and ap = B - rp-l, their number being p = q(X) + 1. Therefore, we can define a certain integral vector a(X) = (al (X), ... ,ap(X» belonging to the Euclidean space RP of dimension p. Since the sum of all ai is, obviously, equal to B - A, the length of (A, B), we can assume that the endpoint of a(X) lies on a (p - 1 )-dimensional simplex a , which can be given by the equation L:f=l Xi = B - A, where Xi are nonnegative Euclidean coordinates for the space RP. The simplex (T is a closed subset, i.e., with the boundary belonging to it, which follows from the fact that some a. can be zero (the case of multiple maxima).

Now, let us consider two historical texts X and Y. We can construct two integral vectors a(X) and a(Y) with their endpoints on one and the same simplex a (Fig. 10). Meanwhile, we assume that both texts describe events on the same time interval (A, B). Note that we have made use here of the above remark according to which we can assume that the volume graphs for both texts have the same number of local maxima.

§4

Information Density in Texts

45

B-A

o

().~

........ ....

........

....

............

........ ........ ........ ....

o

B-A

Figure 10. Method for comparison of two historical texts X and Y. We can construct and compare two integral vectors.

Consider the difference ~ of the two vectors a(Y) and a(X). The vector ~ belongs to our simplex (Fig. 10 . Let A be the usual Euclidean length of the vector ~. Recall that A = L:i=l(ai(Y) - ai(X))2.

We now introduce the numerical coefficient d(X, Y) = VO~of~q ,where D = D( X, A) denotes a (p - 1 )-dimensional ball placed in the hyperplane specified by the equation Ef=l Xi = B - A. This ball is centered at a(X) and has the radius A, while D n a denotes its intersection with the simplex. If A is large, then the ball D may contain points located outside the simplex. If, however, the number A is small, and the point a(X) (the center of the ball) does not lie on the boundary of the simplex, then the ball D(X,..\) is small and wholly contained in the simplex, in which case we have the equality Dtva = D. Finally, volS, where S is an arbitrary (p-l)-dimensional subset in the hyperplane Ef=l Xi = B - A, denotes either the usual Euclidean (p - 1)dimensional volume of the subset S (in which case we will say that we are dealing with a continuous model) or the number of integral points in the set S. Recall that a point is said to be integral if its Euclidean coordinates are integers. In the latter case, we will say that we are dealing with a discrete model.

46

New Statistical Methods for Dating

Chapter 2

4.5.2. Assume that a text X is fixed. While varying the text Y, it becomes obvious that the number d(X, Y) which we introduced in Section 4.5.1 can be interpreted (under certain natural assumptions) as the probability of the point aCYl getting randomly into the ball D with fixed radius ..\ and with its center at a(X). This probabilistic interpretation is not at all essential for the following and is given here only as a formalism which is useful for the calculations.

We now introduce on our simplex a (p - i)-dimensional measure" in the following two ways:

Version 1. II(S) = number of il,ltegrai poil,lts il,l the d<?main S consistent with the

r: number of integral pomts 10 the simplex

discrete model (see above).

Version 2 II(S) = Eucli~ean volume of the dC?main S consistent with the continu-

• r: Buclidean volume of the simplex

ous model, where S is a (p - I)-dimensional domain (subset) in the simplex (T.

Further, to interpret the coefficient d(X, Y) probabilistically, we need a hypothesis regarding the distribution of the random vector a(Y) on the simplex. In the simplest case, we assume that a(Y) is uniformly distributed on the simplex, which means that the probability that it gets into some domain S equals the measure ,,(S). Since Jl«T) = 1, 0 < ,,(S) < 1.

Applied to our problem, this conjecture can be confirmed somewhat by the distribution of the random vector a(Y) being consistent with that of the local maxima of the volume graphs vol Yet) for a variable text Y and a variable historical epoch (A, B) with invariable A-B. Since we intend to model the mechanism of information loss and the fall into oblivion, we should take into account the circumstance that if an archive of historical documents is lost during some sort of disaster, or similar circumstances, then it is appropriate to accept the hypothesis that the destruction of any text from this archive is "equally likely" . The assumption can be restated as a conjecture regarding the uniform distribution of the random vector aCYl, where the text Y ranges over all the texts surviving from all possible historical epochs.

To describe this distribution of the local maximum random vector formally is unreal for the moment, because we should calculate all the vectors aCYl for all surviving historical texts Y, which is, obviously, impossible. We could certainly investigate a sufficiently large sample from the set of all existing texts; however, certain complicated problems regarding the representability of such samples and their homogeneity arise here. Besides, such an experimental approach could considerably distort the general form of the distribution, since it is difficult to estimate a priori whether a particular sample from the texts reflects the real mechanism of information loss. At the same time, the assumed uniformity of the distribution of the random vector aCYl well reflects the circumstance that by altering the historical epoch (A, B), we "uniformly mix" all possible reasons for the loss of historical texts.

In certain examples related to other methods of dating (see below), where the total information requiring quantitative processing is sufficiently clear, we have experimentally calculated the desired distribution of the random

§4

Information Density in Texts

47

vector. For example, we have found the empirical frequency histogram for the distribution of the duration of kings' reigns.

We would like to emphasize once again that the probabilistic interpretation of the coefficient d(X, Y) will not be employed below. First, we carry out an experimental calculation of the concrete limits within which d(X, Y) varies for "dependent historical texts" describing the same events. It is the values of d(X, Y) found experimentally that in the following will serve as a standard for comparing other pairs of texts under investigation.

The coefficient d(X, Y) ranges from zero to unity. With the assumed uniform distribution, we can say that the smaller its magnitude, the lower the probability of the random event that the random vector a(Z) distributed on the simplex uniformly happened to be from the vector a(X) at a distance not exceeding the observable one between the points (vectors) a(X) and a(Y).

If ~, the distance between a(X) and a(Y), is sufficiently large, then d(X, Y) may be unity, in which case the above ball D wholly absorbs the simplex a . Therefore, no statement regarding a possible proximity of the texts X and Y can be formulated; these texts are far from each other from the standpoint of their volume functions.

Note that, for any X and Y describing one historical epoch (A, B), we can always find the same scale P, which permits us (in principle, at least) to compare even texts of various nature, namely, textbooks, annals, and so forth.

Section 4.6 is of formal mathematical character and may be omitted on first reading.

4.6. Mathematical formulas for computing d(X, Y). Mathematical

corrections of the maxima correlation principle

4.6.1. In this section, we give certain mathematical formulas for computing and estimating d(X, Y). If a simplex a is specified by the equation L:f=l Xi = s, where all Xi are nonnegative, then its volume in the continuous model, i.e.,

in the usual Euclidean volume, equals ~ Hence,

7r,;l ~p-l(p _ 1)1

d(X Y) < .

, - f(P;l + 1)SP-1JP '

where s = B - A, ~ is the length of the vector a(Y) - a(X), and r is the classical gamma function. For small ~, the inequality turns into an equality.

If B - A, and p and A are all sufficiently small, then we have to make use of the discrete model to calculate d(X, Y) (see above). In order to do this, we have to find the number of integral points in the ball D and the simplex a for the given values of B - A, p, and A. Unfortunately, an exact universal formula cannot be offered in this case; however, we can derive an asymptotic relation, which is omitted due to lack of space.

We can estimate this mathematically, beginning with which numerical values of B - A, P, and A it is possible to resort to the continuous model without

48

New Statistical Methods lor Dating

Chapter 2

committing a gross mistake. As a matter of fact, concrete calculations are more simple within the framework of the continuous than the discrete model. Any concrete numerical estimation is also omitted. Note, in this connection, that an interesting mathematical problem arises here: How can an exact boundary be found, separating the area of application of the discrete model from that of the continuous one (for a prescribed value of admissible error)?

4.6.2. We now indicate the first mathematical correction to the F -model. Since, when equalizing the numbers of the local maxima of the volume graphs of two texts being compared, we are forced to assume certain of the coordinates ai of the vector a(Y) to be equal to zero in the case where q(Y) < q(X), this process is explicitly equivocal. It is clear that we can take any of al to ap as the zero coordinate. We should therefore consider all possible methods of equalizing the numbers of the maxima and then take the least of all the values obtained as the coefficient d(X, Y). To eliminate the asymmetry between the texts X and Y due to the definition of d(X, Y), we have to "symmetrize" the value, i.e., consider d(X,Y)~d(Y,X).

4.6.3. Now we present the second correction of the F-model. We have carried out previous constructions for informative functions of the form I(t, X). However, we should also consider all their possible smoothings in order to establish the invariant, characteristic properties of the graphs better. For example, as the simplest method of smoothing a graph, we can consider the "neighbourwise averaging". For a function I(t), its first smoothing is then constructed as 81/(t) = l(t-l)+/~t)+/(t+l), the second, 82/(t), as the expression 81(81/(t)), and so on. It is evident that this procedure smoothes the graph and eliminates random, small local maxima. Their corresponding coefficient dj(X, Y) is computed after each stage of the smoothing procedure applied to 8j I(t, X) and 8j I(t, Y). As the final value of d(X, Y), we should again take the least of all numbers obtained in such a manner. Having calculated d(X, Y) for the original and nonsmoothed graphs I(t, X) and I(t, Y), we obtain an upper estimate of the final value of the coefficient. Note that if we are comparing two texts X and Y which describe different historical epochs of the same length, then we should make these intervals coincident on the time axis by superimposing.

4.6.4. The third mathematical correction of the F-model is as follows. For greater statistical invariance of the obtained results, we should compare not simply a pair of separate texts X and Y, but two sufficiently large groups of texts XI, ... ,Xn and Y1, ••• ,Yml where nand m are supposed to be sufficiently large. Denote the first text group by {X}, and second by {Y}. Then we have to compare the graphs of the functions I(t, {X}) = /(t, Xl,." ,Xn) and /(t, {Y}) = /(t, Y1, •.• , Ym), where I(t, {X}) is the averaged function defined by the equality I(t, {X}) = k E?=l I(t, Xd. The averaged graphs /(t, {X}) and /(t, {Y}) so formed have been freed of the random local maxima which might have appeared in one or the other text for some particular reason of nonuniversal character. The comparison of the averaged graphs yields a more reliable statistical picture of the evolution in time of written evidence.

§4

Information Density in Texts

49

4.7. Verification of the maxima correlation principle against concrete historical material

4.7.1. Let us now verify the F -model against some concrete historical material. This model will be confirmed if the coefficient d(X, Y) is "small" for the majority of pairs of real historical texts X and Y describing the "same" events in the same time interval (A, B) (we call such texts dependent). On the contrary, for independent texts, i.e., those describing essentially different events (or essentially different time intervals), this coefficient must be "large" . For now, the concept of smallness for the coefficient will not formally be made precise, since we need more experimental data. Meanwhile, the broad statistical experience of natural science permits us to estimate (with the assumption of the uniform distribution of the random vector a(Y» really small probabilities in problems of this kind. Since the explicit construction (from texts) and description of informative graphs for concrete historical texts of large size are sufficiently bulky, we are forced to omit the diagrams here and will give only some typical examples. We first consider the textual volume function vol X(t) (the volume is measured in pages; see also Section 4.2). For example, we have taken the monograph ofV.S. Sergeev, Essays on the History of Ancient Rome [1], as the text X, and the famous History of Rome by Livy [2] as the text Y. The time interval (A, B) described in both books was taken from 758 to 288 B.C., which means that A = -757, B = -287, and the length of the interval B - A = 470 years. The "-" sign denotes the years B.C. The volume graphs of these texts (with respect to years) are shown in Fig. 11. The continuous line has been constructed for Sergeev's text, and the dotted one for Livy's. As it turns out, ;\ = 21, p = 14, and the coefficient d(X, Y) measuring the

Sergeev

............... Uvy

.' . ' , '

· '

· '

· .

· . , '

.. , ,

: ~

, ,

" " , ,

, .

· .

· ,

· .

· . , .

·

· · ·

· .

· .

· "

· , .

-, ' ~

. .

. , : .... :

.

-757

-587 Ancient Roman history textual volume graphs

Figure 11. Volume functions o/two dependent texts: V.S. Sergeev's monograph "Essays on the History of Ancient Rome" and Livy's "History of Rome".

50

New Statistical Methods for Dating

Chapter 2

proximity of the vectors a(X) and a(Y) in the statistical sense equals 2.10-12• The distance between them in the Euclidean sense is 21.

In other words, the probability of random proximity of the texts X and Y (at the distance A = 21) is majorized by 2.10-12. We had obtained it even before minimizing d with respect to smoothing all the graphs (see Section 4.6.3). Therefore, 2 . 10-12 is the upper estimate of the final value of the coefficient. Being very small, it indicates a considerable correlation between the historical texts by Sergeev and Livy from the point of view of the volume graphs. It is but natural, because, describing the "same events", they are a priori dependent. The most important is not the small absolute value of d(X, Y), but the large difference between the values of d(X, Y) for dependent pairs X, Y and independent pairs X, Y.

Similar results are also obtained for the comparison of the different informative functions fi(t,X) and fj(t, Y) for dependent X and Y, which describe the same events in the same time interval (A, B). For example, consider for the above historical texts X by Sergeev and Y by Livy the informative function f2(t, X) = '~), where set) = number of mentions of a year t in X. Compare its graph with the volume graph vol yet) for Y. We have thus considered the time interval from 521 to 294 B.C., i.e., A = -520, B = -293, and the length B - A = 227. It turns out that A = 8, p = 8. The result of the calculations is devol Yet), !2(t, X» = 5 . 10-6.

In the latter example, we have smoothed the graphs once, i.e., compared the graphs of 81 !2(t, X) and 81 vol Yet). Such a small value of the coefficient measuring the probability with which the vectors a(vol Y(t» and a(!2(t, X» may randomly be at the distance A = 8 explicitly expresses the correlation between different informative functions of dependent texts describing the same period in the history of the same region or state.

These results indicate the existence of another special law of information density conservation to be formulated as a K -model below.

Similar results are obtained when investigating other pairs of historical texts. I performed the computational experiment with the help of M. Zamaletdinov (Faculty of Mechanics and Mathematics of Moscow University).

4.7.2. We now formulate another natural model of the evolution of written information. If a historical text X describes events in a time interval (A, B), then there is a correlation between the different informative functions constructed for X. For example, the graphs fi(t,X) and /jet, Y) for different i and j attain local maxima at approximately the same points (years) in (A, B).

The [(-model (statistical hypothesis) is subject to all the mathematical corrections indicated above for the F -model; accordingly, we do not give them here.

Its intuitive substantiation can be extracted from the following observation.

Say, if a year t from a time interval (A, B) is described in a historical text X in more detail than the neighbouring years, then this circumstance must lead not only to a local increase of the number of pages in X, which will have bearing

§4

Information Density in Texts

51

on the function It, but also to an increase of references to the year t (this has bearing on the function f2)' while the number of names of historical characters mentioned at the same time in X increases locally, too (which has bearing on the function 13), and so on. Roughly speaking, the longer a textual fragment X (t), the more names are mentioned there. Certainly, all these statements (models) are valid only "on the average" and for large time intervals.

It is easy to see that the K -model is, actually, a corollary to the same basic statistical hypothesis (information density conservation law) formulated in Section 4.3 and is modelling the mechanism of loss and the fall into oblivion of written evidence. To derive the K-model from it, it suffices to apply the hypothesis to an arbitrary pair of informative functions Ii and /j (and not only to the volume function II = vol). In this sense, both the F- and the K-model can be regarded as corollaries to the basic hypothesis of 4.3.1.

4.7.3. We now turn to the verification of the K -model against concrete historical data. It will be confirmed if the vectors a(li(X» and a(1; (X») are proximal for most sufficiently large concrete texts X, i.e., if the number d(fi(X),!; (X» is sufficiently small. Omitting the diagrams, I shall give just one typical example.

Sergeev's monograph Essays on the History of Ancient Rome was again taken as a text X for which the volume function voIX(t) and the frequency of mentioning particular dates (years) 12(t, X) have been calculated. It turns out that these two graphs are strongly correlated, i.e., their peaks practically coincide. The calculations have shown that A = 25, p = 11, and B - A = 280 years (A = 241 B.C., and B = 521 B.C.). Finally, devol X(t), 12(t, X» = 7 . 10-5. Similar results have been obtained for other chronicles, textbooks, monographs, and so forth, which we investigated. This confirms the K -model. The values of the coefficients obtained in the experiment are close to those obtained by us earlier in verifying the F -model against necessarily dependent pairs of historical texts.

4.7.4. We should not think that the coefficient d(fi(X), h(Y» is generally small for the arbitrary texts X, Y, and the functions fi' 1;. If it were so, then the extreme smallness of the numbers d obtained above (for texts known a priori to be dependent) would not reveal anything.

Recall that, with the assumed uniform distribution of the random vector on a simplex, all points of this simplex are treated equally during the random walk of the vector aCYl (and for a fixed vector a(X». Hence, when the vector aCYl is moving farther from a(X), the ball D(X,..\) enlarges, and the coefficient d(X, Y) approaches unity (the ball occupies a greater and greater part of the simplex).

Concrete computations were then carried out for the independent historical texts X and Y, i.e., texts describing events or periods of history (A, B) and (C, D) known a priori to be different. It turned out that the coefficient d was of the order of unity for a small number of local maxima. For a large number of maxima, this lower bound of the coefficient values for independent texts

52

New Statistical Methods for Dating

Chapter 2

decreases but continues to be several orders greater than its upper bound for a priori dependent texts.

We illustrate this with a typical example. Consider the first and second halves of Sergeev's text as X and Y. They describe different historical epochs, namely, the periods (At B) from 521 to 381 B.C., and (C, D) from 381 to 241 B.C. in the history of ancient Rome. The functions h(t, X) and /2(t, Y) represent the frequencies of mentioning particular dates (years) in the texts. It then turns out that A = 59, B - A = D - C = 140 years, and p = 5. Finally, d = 1/3, which is close to unity in contrast to the above examples of dependent texts.

4.7.S. Our computational experiments confirm both of the theoretical hypotheses, i.e., the F- and K-models. We thereby discover specific laws of information density conservation in historical texts describing sufficiently large time intervals. They are partly manifest in certain quantitative characteristics of textual informative functions such as the distribution of local maxima. Though their intuitive substantiation seems to be rather clear (see above), the number of parameters of the informative functions, which can play the role of "historical epoch invariants" , does not at all include all of them. For example, the absolute amplitudes of the volume graphs, the frequencies of names, and so forth, can be substantially different even for a priori dependent texts. The amplitude is, therefore, not an invariant of the period. The discovery of other "dependent text invariants" is a nontrivial problem and requires more statistical research.

4.8. A new method for dating historical events. The method of restoring the graph of the primary and surviving information stock

4.8.1. We now offer a new method for dating historical events described in ancient texts. The laws of information density conservation permit us to introduce a formal procedure to date the events described in texts with lost or unknown dating.

In fact, let Y be a historical text with undated events. Let it be supplied with dates according to some unknown chronology. For example, let the years t be counted from the unknown absolute date of the foundation of some city. We can assume, nevertheless, that the text Y is parametrized by a time t in the period between the years C and D according to the unknown chronology. How can we restore the absolute dates of the events?

Construct the informative functions h(t, Y) and consider the set of all absolutely dated texts X. We also construct their informative functions /i(t, X) and assume that we can choose X with some /i(t, X), or at once their whole set, close in the sense of smallness of the coefficient d to an informative function /i(t, Y). In other words, d(/i(X), /i(Y» is "small" (i.e., it is close to the values of the coefficient d(/i(Z), /i(V» for surely dependent pairs Z and V). It then means that within the framework of the F - or [(-model substantiated by us, the texts X and Y may be dependent. In particular, the time interval

§4

Information Density in Texts

53

(C, D) is close to (A, B) described in X or simply coincides with (A, B).

Moreover, the smallness of d will indicate not only a possible proximity of the time intervals (A, B) and (e, D), but also that of the events and the historical epochs described (and even the coincidence). Meanwhile, it is important to bear in mind that the descriptions of the same events in X and Y can be outwardly different, with different names or nicknames of the historical characters, different geographical names, and so on. For example, the texts X and Y can be two versions of the description (chronicle) of the history of a region or state, but written in different languages, by different chroniclers, in different countries, or according to different chronologies.

4.8.2. Let Xl, ... ,X Ie be the collection of certain historical texts describing events in the same time interval (A, B). Consider the averaged function of

their volumes, i.e., j«t) = vol(t; Xl, ... , Xk) = t E;=l vol Xj(t).

If the quantity of texts (i.e., the number k) is sufficiently large, then the graph of k(t) can be assumed, due to the computational experiments described above, to be coincident with (or close to) the graph of the original or surviving information stock C(t) or K(t), respectively. Meanwhile, the local maxima of the graph of j«t) indicate those years in the time interval (A, B) for which an especially large amount of written information (texts) has survived.

This circumstance can help in dating available ancient historical texts.

Namely, especially extensive texts (describing individual years) must concentrate close to the local maximum points of the graph of k(t). To avoid a misunderstanding, we stress once again that dating a text here implies dating the events described in it. Besides, the text itself can have been written quite recently, e.g., a textbook on ancient history.

4.9. The discovery of dependent (parallel) historical epochs

traditionally regarded as different

4.9.1. From 1978 to 1979, I made a series of experiments for calculating the coefficients d(/i(X),/i(Y)) for different pairs of historical surveys X and Y which embrace considerable historical periods (A, B) and (C, D) of the same length, i.e., B-A = D-C. This condition is necessary for a formal comparison of the informative functions of different texts.

Unexpectedly, pairs of historical epochs (A, B) and (C, D) traditionally regarded as different (and pairs of corresponding texts X and Y which describe them) were discovered for which the coefficient d turned out to be extremely small, i.e., characteristic of a priori dependent texts (epochs). Let me give one example.

Let Z be the part of F. Gregorovius' work History of the City of Rome in the Middle Ages [3] embracing the events in medieval Rome from A.D. 300 to 754. In other words, A = 300, and B = 754. As a text Y, we take the part of Livy's History of Rome ([2]) describing the antique history of Rome from the year 1 since the foundation of the City (Rome) to the year 459 since the

54

New Statistical Methods for Dating

Chapter 2

foundation of the City. It is traditionally assumed that the year 1 since the foundation of the City coincides with 753 B.C. Therefore, we can take the 460- year-long interval from 753 to 294 B.C. as (e, D). Traditional history certainly regards these two texts, and the events described by them, as independent in all respects. However, the computation shows that d(Z, Y) = 6.10-10 for the volume function, which means that, due to the smallness of the coefficient, the two texts and epochs are "like" dependent texts or epochs.

Recall that Sergeev's text X [11 is necessarily dependent on Livy's text Y [21.

This circumstance was confirmed above by a calculation according to which d(X, Y) = 2.10-12. We should, therefore, expect that from the standpoint of the coefficient d, Gregorovius' and Sergeev's texts, i.e., Z and X, respectively, will turn out to be dependent. Computations fully confirm this assumption, too.

4.9.2. It is\ interesting to investigate the behaviour of d by enlarging the textual time scale. For example, let the parameter t now range not over years, but over separate half-centuries. As a text Z, we take the part of Gregorovius' book [3] which describes medieval Rome from A.D. 300 to 950, i.e., during a 650-year-Iong time interval. As a second text, X, we take the part of Sergeev's book [1] describing the events in ancient Rome from the year 1 to 650 since the foundation of the City, i.e., from 753 to 103 B.C. (assuming that the year 1 since the foundation of the City coincides with 753 B.C.). Ranging over halfcenturies, the parameter t, therefore, assumes 13 values in the indicated time interval. Calculations show that d(Z, X) = 1/50. Here, we have made use of the textual volume graphs.

Similar results are valid for the information function 12 both for the initial (where t ranges over separate years) and the enlarged time scale.

The order of the coefficient d remains practically unaltered also in smoothing the graphs of the functions It and 12, which indicates the stability of the results relative to the smoothing, or averaging operation. We can also see here the advantages of "long time scales", when the length of the time interval described in the texts is of the order of 100-1,000 years. With the functions li(X) and fi(Y) being correlated, the coefficient d then turns out to be especially small for dependent texts. Enlarging the scale certainly makes the picture rougher, which affects the increase of d. In the example given, it increased to 1/50.

4.10. The dynasty of rulers and the durations of their reigns as an

important informative function

We now give an example of another important textual informative function. Consider a continuous (or gap-free) sequence R of some rulers (kings) R1, • •• ,Rn. Let them be indexed by an integral parameter i whose increase is associated with ordering the kings chronologically. As the informative function, we take the duration of a rule, i.e., let 15(i) = the duration of the rule of

§4

Information Density in Texts

55

king Hi who is ith in order. The sequence (dynasty) of the rulers Rb ... ,Rn will be called a dynastic stream for short.

Along with a sequence of numbers representing the durations of the rules, we can define (as above) the vector a( R) of the local maximum points of this graph, which permits us to compare two sequences of rulers by comparing the graphs of the durations of their reigns.

The difference from the previous algorithm lies in the fact that the proximity coefficient d must be calculable in a more complicated way for two dynasties, which is related to the impossibility of regarding the random vector which schematically represents the graph of the rule durations in a dynasty to be distributed uniformly. Because of lack of space, we omit the details of this investigation. It turns out that the distribution of the durations of the reigns of kings (monarchs) is subject to a sufficiently nontrivial law which forms the basis for defining the proximity coefficient for two dynasties.

4.11. Frequency distribution of the rules of kings who lived from A.D. 1400 to 1800 and from 3000 B.C. to A.D. 1800

Figure 12 represents the result of processing the numerical information regarding the durations of reigns contained in J. Blair's chronological tables [4]. The rule durations of historical characters of Europe and of the Mediterranean region are marked off on the horizontal axis S in the time interval from 3000 B.C. to A.D. 1800, where the parameter S ranges not over separate years of a reign, but pairs of years, i.e., 1-2 years, 3-4 years, 5-6 years, ... ,89-90 years. Further, the values of the following function, P(S) = number of historical characters whose rule duration lies in the interval S, are marked off on the vertical axis P. For example, there were 55 personages (included in Blair's tables) ruling from 19 to 20 years. The dotted line in Fig. 12 represents the frequency distribution of the rules of those kings who lived from A.D. 1400 to 1800. The continuous line describes those kings who lived from 3000 B.C. to A.D. 1800 according to traditional chronology. In the investigation of the complete lists of the European and Mediterranean rulers from 3000 B.C. to A.D. 1800, I was assisted by M. Zamaletdinov and P. Puchkov.

The proximity coefficient d for two dynasties should be calculated by taking into account the above histogram of the rule duration frequency. The following experiment also indicates the necessity of resorting to it. All 1200 rulers listed in Blair's chronological tables were aligned in sequence, ordered chronologically inside one dynasty (in one region). Simultaneous reigns were placed in line one after another. We now index the obtained sequence of rulers by i = 1,2, ... ,1200. Let ei be the random variable representing the duration of the ith king's rule. Consider another random variable '1i(k) = ei+k. The sequence '1 is thus obtained from e by shifting as a rigid block through k units (numbers). Let r(k) be the correlation coefficient for e and '1(k). The graph of the variable r( k) for 1 < k < 300 is shown in Fig. 13. The calculations have been performed on a computer by P. Puchkov. We do not have the space here

56

New Statistical Methods for Dating

Chapter 2

50

.

if\

¥ ·\ •. l···\ •.•.. :\ ...... .".: .... :··.... "

.......................

P(S)

250

200

150

100

o

20 40

60

70

s

Figure 12. Frequency distribution of the rules of kings of Europe and the Mediterranean region from 3000 B.C. to A.D. 1800 and from A.D. 1400 to 1800

r(k)

0.1

·0.1

k

Figure 13. Graph of the variable r(k).

to list the results of other experiments or the related conclusions. That will be carried out in Vol. 2 of this book.

§4

Information Density in Texts

57

4.12. The concept of statistically parallel historical texts and epochs We now formulate the concept of formal, statistical isomorphism (parallel) of historical texts and epochs. Let a time interval (A, B) in the history of some region or state M be described in the historical texts X = {Xl, . .. ,X k }. Let another time interval (C, D) in the history of another and, in general, different region or state H be described in the texts Y = {Y1, ... , Yp}. Consider the set of informative functions fi(X) and fi(Y) of these collections of texts. We will say that the above historical epochs (texts) are formally isomorphic (parallel) if the proximity coefficients d(/i(X), fi(Y)) are "small". More precisely, they must be as small as those of a priori dependent texts (epochs).

The concept of formal isomorphism of epochs described in some historical texts does not at all mean that the epochs or events themselves are identical. Their isomorphism (parallelism) can indicate the convention of ascribing, for some reason, certain historical documents from one epoch to another. Our goal is to solve the global problem, i.e., to describe the collection of all parallel epochs and texts in the whole historical period of written language (see below) for which we extend the stock of informative functions.

4.13. The "written biography" or euquete-code of a historical character

4.13.1. Very heterogeneous historical information has survived concerning the ancient dynasties of rulers. Meanwhile, different sources speaking of the same ruler can be very different in the details concerning the description of his or her activity. Sources can deal with the events during the monarch's rule differently, characterize differently the rulers themselves, and refer to them by different names or nicknames, and so on.

But there exist more or less "invariant" facts whose description is less dependent on a bias or political pressure on the chroniclers. One of these "invariant" parameters, for example, is the duration of a king's rule. Usually, there are no special reasons for which a chronicler would like to considerably distort this value (since it is emotionally neutral). By a "dynasty" of kings, we will understand a continuous (i.e., gap-free) sequence of rulers of one region. We do not assume that the throne should always be passed on hereditarily (from father to son, etc.). With each ruler or prominent statesman playing an important role in a particular period of history, we associate a certain table called an enquete-code, or "formal biography". By a "written biography" of a historical character, we will understand the collection of all preserved evidence about him or her. Normally, this "biography" is a set of individual facts which are rather uncoordinated and traditionally ascribed to the "ruler" by the later historians and chronologists based on the systematization and dating of the available written evidence. Besides, arising from the investigation of the primary sources, this "biography" can have almost nothing to do with the actual biography of the ruler. Strictly speaking, in many cases, we can only guess at

58

New Statistical Methods for Dating

Chapter 2

the ruler's real biography. Therefore, we only deal with "written biographies" in the following. Now, we associate each "written biography" established in the primary sources with the enquete-code, or "formal biography" , by possibly distinguishing all the basic facts from the "written biography". In doing so, we hierarchically order the facts of the "formal biography" in accordance with their decreasing invariance. The facts most often distorted by chroniclers will be nearer the end of the table.

(1) Sex of a personage: (a) male, (b) female.

(2) Length of life of the personage (or at least the year of death).

(3) Duration of the rule. It should be noted here that the end of the rule is nearly always fixed uniquely by the chronicles. It is usually the death of the ruler. The beginning of the rule sometimes (though, rather rarely) admits several versions, among which are the official coronation date, those of conferring the title of "Caesar" or "Augustus", and of the death of a more powerful co-ruler, and so on.

During the statistical investigation, all possible versions of the beginning of a rule are then indicated by all available means, considered to be equally likely, and included in the enquete-code. Note that, in analyzing real chronicles, we found that the number of versions of the date of the beginning of a rule only rarely reached three.

(4) Social status: (a) emperor, king, queen, etc., (b) army commander, (c) politician, public figure, or statesman, (d) scientist, (e) religious leader (pope, bishop, high priest, prophet, etc.).

(5) Cause of death of the personage: (a) natural, (b) on the battlefield, or as the result of a mortal wound, (c) result of conspiracy in peace time, (d) result of conspiracy in war time, (e) due to some special, exotic circumstance.

(6) Natural disasters during the personage's rule: (a) hunger, (b) floods, (c) epidemic diseases, (d) earthquakes, (e) volcanic eruptions. The duration of the event and the year (or years) when it occurred are noted by all available means.

(7) Astronomical phenomena during the rule: (a) did occur, (b) did not occur, (c) solar or lunar eclipses, (d) appearance of comets (often as "swords" in the sky, etc.), (e) "star" flares, (f) horoscopes, i.e., the planetary positions relative to the zodiacal constellations.

(8) Wars during the rule of the personage: (a) did take place, (b) did not take place.

(9) Number of wars (different wars are usually separated in chronicles). (10) Basic time characteristics of the wars 81,82, ... ,8p• Namely, alc year of the king's rule in which the war numbered k , i.e., BIc, took place; hlc = duration of the war BIc; Cl:x = distance in years between the wars BIc and Bx·

(11) Intensity of the war BIc for each k. The intensity of a war can be estimated, for example, by the volume of texts in chronicles devoted to it. Roughly speaking, wars can be divided into two classes, namely, (a) largescale wars, (b) local wars.

(12) Allies, adversaries, and neutral forces in a war BIc (for each k); their

§4

Information Density in Texts

59

number and schematic diagram of their relations (who are allies or enemies, etc.).

(13) Geographical localization ofa war BIc (for each k): (a) near or inside the capital, (b) inside the state, (c) outside the state and where exactly (external war), (d) both external and internal war simultaneously, (e) ci viI war or a war with external enemy.

(14) Final result of the war: (a) victory, (b) defeat, (c) uncertain result. (15) Peace talks: ( a) concluding a peace treaty after the victory of one of the adversaries (who exactly; see paragraph (12», (c) concluding a peace treaty after the defeat of the ruler.

(16) Conquering the capital: (a) did occur, (b) did not occur, (c) specific circumstances of the capital's siege or its fall.

(17) Fate of the peace treaty: (a) it was violated (by whom and under what circumstances), (b) was not violated during the rule.

(18) Detailed description of conquering (or fall of) the capital during the

war.

(19) Diagram of the armies' marches during the war.

(20) Ruler's participation in the war: (a) did occur, (b) did not occur. (21) Conspiracies during the ruler's lifetime: (a) did occur, (b) did not occur. (22) Geographical localization of wars, allies, adversaries.

(23) Name of the capital. A translation of the name is necessary. (24) Name of the state. Translation is necessary.

(25) Geographical localization of the capital (with the terms translated). (26) Geographical localization of the state (with the terms translated). (27) Legislative activities of the ruler (a) reforms and their nature, (b)

issuing of new code of laws, (c) reintroduction of former laws (which exactly). (28) Complete list of all names of the ruler with their translations. As a matter of fact, practically all ancient names have meaningful translations and originally were simply nicknames (such as "mighty", etc.).

(29) Ethnic group of the ruler, members of his or her family, composition

of the family.

(30) Ethnic group of people living in the region or city.

(31) Founding of new cities, capitals, fortresses, harbours, etc.

(32) Religious situation: (a) introduction of a new religion, (b) sectarian struggle (between what sects exactly, names of the leaders and their translations), (c) religious riots and wars, (d) religious meetings, councils, etc.

(33) Dynastic struggle inside the ruler's clan, murders of relatives (if any), usurpers of the throne, adversaries, etc.

(34) Other fragments of the "personage's biography" will not be differentiated in such a detailed manner and will be collected in this item. We will call the information gathered here the "biographical remainder". It is convenient to measure it as a percentage of the whole "biography".

Denote the listed items by EC-l, EC-2, ... , EC-34 [enquete-code, paragraph (1), (2), ... , etc.). The whole enquete-code (i.e., the above table) will be denoted by EC for short. Thus, each "written biography" can be represented

60

New Statistical Methods for Dating

Chapter 2

as a certain formal table of EC, or "formal biography". Certain items in this table may be left blank, which occurs in the case where the corresponding information has been lost in the surviving documents.

4.13.2. If the parameter i ranges over the numbers of consecutive rulers in a dynasty (dynastic stream) R1, ••• , Rn, then the enquete-codes ECi of ~ can be regarded as the set of values of a certain new informative function. Since the set of enquete-codes of the rulers of a dynasty practically completely accumulates all the important information from that epoch, we can assume that the sequence, in fact, describes the epoch (A, B) in the history of the region. It is "covered" by the rulers of the given dynasty. Eventually, we can associate each historic epoch with a set of informative functions v = {Il, 12, ... , EC-dynasties}, where It, 12, ... are the informative functions already familiar to us, and the EC-dynasties are the enquete-code collection for the rulers "covering" (A, B). Since the rulers in one region sometimes reign simultaneously (and are then called co-rulers), different "dynastic jets" should be distinguished from the total dynastic stream, i.e., continuous (gap-free) subsequences of personages among whom there are no co-rulers or very few of them.

Consider now the two epochs (A, B) and (C, D). Associate each with the above set of informative functions, i.e., v = {It, 12, .. , ,EC-dynasties} and v' = {If, I~, .. . ,EC' -dynasties} .

4.14. A method of comparing the sets of informative functions for

two historical epochs

4.14.1. Let us describe a method of comparing the sets of informative functions v and v' for two epochs. The comparison has been carried out in terms of coordinates, i.e., the functions Ii and II of the same kind are compared to each other. For 11,/2,' .. , the coefficients of the above d-type were calculated. The comparison of the dynastic enquete-codes, i.e., the EC-dynasties and EC/-dynasties, is more complicated owing to a finer structure of these informative functions. Without going into details, we only discuss the comparison principle.

We start with the comparison of rule durations. Let T1, ... ,Tn and T{, ... , T~ be two sequences of dynastic rule durations in the epochs (A, B) and (C, D), respectively. We will measure the "distance" between them using the coefficient ,\ introduced in §5; see below. The coefficient ,\ can be called "stream deviation coefficient" (SDC).

4.14.2. Thus, we compared the items of the enquete-codes EC-3 and EC'-3 of two rulers Rand R'. The comparison of the remaining items will be done as follows. Let EC-l, ... , EC-34 be the enquete-code EC for the ruler R, and EC/-I, ... , EC' -34 the enquete-code EC' for R'. We introduce the numerical coefficients E1 , ... ,E34 measuring the proximity (or remoteness) of the items EC-I and EC'-I, ... , EC-34 and EC'-34. We have omitted here the times EC-3 and EC/-3, since they are compared already by means of the SDC. To

§4

Information Density in Texts

61

compare the items EC-j and EC/-j, we introduce the coefficients E-j. Then the following three situations are possible.

(1) The biographical data compared are similar. For example, the items EC-5 and EC/-5 state that both rulers died a natural death, in which case we put E-5 = +1.

(2) The biographical data compared are not explicitly coincident. For example, EC-5 states that a ruler died a natural death, and EC' -5 that he died as the result of a conspiracy. We then put E-5 = -1.

(3) The biographical data compared are neutral in the sense that they are consistent, but not identically coincident. For example, EC-5 states that a ruler "died", whereas EC' -5 states that a ruler "was killed". We then put E-5 = O.

We now introduce the resultant coefficient E = E-l + E-2 + E-4 + ... + E-33. The coefficient E-3 is absent because the rule durations are compared by means of the SDC. Thus, the coefficient E measures the proximity of the enquete-codes of the two rulers R and If. Now, given two sequences of rulers {R;} and {Ri}, we obtain a sequence of coefficients E; comparing them. Finally, we introduce the average coefficient Eav. = ~ L~=l E, measuring the proximity of the complete enquete-codes of the dynasties of {Ri} and {~}. Collecting this information, we might be able to estimate the remoteness or proximity of the enquete-codes of two dynasties and their corresponding epochs by means of the coefficients Eav. and the SDC.

4.15. A computational experiment

From 1978 to 1979, I performed a computational experiment, comparing several hundred pairs of epochs (A, B) and (C, D), i.e., sets of their informative functions v = {It, 12, ... , EC of dynasties of rulers}. The time limits for the experiment were 3000 B.C. and A.D. 1800, with the events localized in Europe, the Mediterranean region, Egypt, and the Near East.

In particular, the experiment showed that, in comparing the enquete-codes of some personages, we often had to put the coefficient E-j equal to zero, which occurred because the information compared there was consistent and, at the same time, unconfirmed. The role of +1 and -1 was therefore heightened. We discovered further that, in the overwhelming majority of concrete enquetecodes, the coefficient E-34 has to be made equal to zero, again because of the consistency of information compared. Recall that item EC-34 of an enquetecode is the personage's "bibliographical remainder" . For a reliable comparison of the items EC-34 and EC/-34 in two biographies, we must be certain that we really possess sufficiently complete "written biographies" of the personages compared. However, to guarantee the completeness of the rulers' enquetecodes (all the more, of whole dynasties) is usually a complicated matter, due to which we have resorted to the following formal method.

For each historical epoch, we have chosen a (possibly) uniform and sufficiently large historical text describing the events. We took either a fundamental monograph of the type of Gregorovius' work or a fundamental primary

62

New Statistical Methods for Dating

Chapter 2

source of Livy's type. Then item EC-34 of the enquete-code was identified with the "bibliographical remainder" for a historical character given in the historical work. For simplicity, the volume of item EC-34 was usually calculated as a percentage of the volume of the entire "biography" of the personage. Meanwhile, as a rule, we did not analyze the components of item EC-34.

We illustrate this with an example. The epoch of regal Rome from 753 to 510 B.C. described by Livy (see [2]) turns out to be sufficiently close to that of the Roman Empire in A.D. 300-552, with the SDC being less than the lower bound for most of the SDC values for independent dynasties. Further, it turns out for these epochs that Eav. = +19, and that the volume of item EC-34 equals 29%. In other words, about 29% of the "biographies" of historical characters in these epochs were "thrown overboard" from the discussed parallels, it being important that this entire "biographical remainder" is zero for the coefficient of EC-34, which means that the information not involved in the parallel is noncontradictory within the framework of the comparison.

It is useful to compare this result, which probably indicates the dependence of the epochs, with the numerical parameters found for an arbitrarily chosen pair of independent epochs. As an example of independent historical epochs, we take the following dynasties.

(1) The dynasty of the Russian grand dukes from Igor (912-944) to Demetrius I (1275-1293).

(2) The dynasty of the Byzantine emperors from Theodosius II (408-450) to Theophilus (829-842) [4]. Calculations show that the SDC is "very large"; further Eav. = -8.7. In this case, the volume of the "remainder" EC-34 is equal to 40%. This pair of dynasties is independent from the standpoint of the coefficients we introduced.

A great difference is obvious between this and the example of the two dependent Roman Empires (see above).

We shall now describe the results of the global computational experiment of comparing different epochs.

4.16. The remarkable decomposition of the global chronological diagram into the sum of four practically indistinguishable chronicles

4.16.1. The application of the above methods to the material of the GCD led to the discovery of isomorphic, parallel historical epochs in the history of Europe and the Mediterranean region (see Fig. 14). The sequence of figures forms a line in the diagram, schematically representing the history of ancient and medieval Europe. Identical geometric figures (denoted by the same letters) schematically represent historical epochs (or some parts of these epochs) which turn out to be formally isomorphic and parallel.

Figure 15( 1) represents the decomposition of the GCD into the sum of four practically indistinguishable chronicles. We can say that the GCD is decomposable into the sum of several shifts of the same chronicle. As it turns out, we can distinguish a shorter part in the GCD, which we call the chronicle

§4

Information Density in Texts

8 8 ~ ..

Biblical history (Old Testament)

+217 +3)8 +552

13

.1254

-1!iOO

-150 -753 -509

Ewopean history

4 5 6

7

10

12

14

Byzantine history

+527 +129 .,204 +1453

Figure 14. Parallel historical epochs in the history of Europe (and the Mediterranean region) and in biblical history.

Cl' We then take another three copies of the same chronicle, each of which is shifted back on the time axis (i.e., from right to left) by 333 years, 1,053 years, and 1,778 years, respectively, after which all the shifted copies of Ct are glued to it and to each other, resulting in a longer GCD chronicle on which, therefore, parallel epochs duplicating each other appear.

4.16.2. We now give a short description of the GCD for Greece, Rome, and Germany. We shall move from left to right in Fig. 14 and list the historical epochs denoted by different geometric symbols. For a detailed description of all the dynastic jets for these cases, see Vol. 2 of this book.

The numbers of the following items correspond to those of the indicated historical epochs (from 1 to 15).

(1) The Trojan kingdom of 1460-1236 B.C. Seven legendary Trojan kings. (2) The Trojan War, ca. 1236-1226 B.C. The fall of Troy, expulsion of the Trojans.

(3) Several dynastic jets of ancient Greek rulers from 1226 to 850 B.C., i.e., from the fall of Troy to the second version of dating the Trojan War according to the ancient authors Hellanic, Damast, and then Aristotle ([5]; [5*], p. 23) immediately before the foundation of Rome.

63

64

New Statistical Methods for Dating

Chapter 2

8 ~ ~ 8 § ~ ~ ~ 8 8 ~ ~ ~ § 8 ~ ~ ~
..... - - ..- .-
~ ..- ..- - . • . . . + + + + + .- ..- ..- ..-
. . . + + + +
f!il T I
Original Co: K IL H ~0" C
...... 333-year shift of C,

c,: K

Sewing C, +C,;

T T T ....____,

1.053-year shift of c, [K]!0![!]!1 C L. .. _ .. _ .... _ ... _._ ..... __ ._ .. _._ .. _ ....

Sewing C, +C, +C,;

1.778-year shift of C,

T T T T p

0!0"0! ,,------,C 1. _ _ _ .

p
ifll wZIII: I
Byzantine history
TT T T T
·1700 ·1300 ·900 ·SOO ·100 +100 +500 +900 +1300 +1700 Figure J5(1). The remarkable decomposition of the GCD into the sum of four short and practically indistinguishable chronicles.

§4

Information Density in Texts

65

(4) The foundation of Rome and the regal period described by Livy, from 760 or 753 B.C. to 522 B.C.

(5) The war with the Tarquins, the kings' "exile" from Rome and the foundation of the ancient Roman republic (522-509 B.C.).

(6) Republican Rome and ancient Greece in 509-82 B.C. The end of classical Greece, start of Hellenism.

(7) Civil wars in Italy during the fall of republican Rome in the 1st century B.C. Beginning of imperial Rome. The Roman Empire (82 B.C.-A.D. 217). (8) Wars in Italy and crisis of the Roman Empire in the middle of the 3rd century A.D. Wars with the Goths, "soldiers' Emperors" (A.D. 217-235-251). (9) Restoration of the Roman Empire under Aurelian, and contemporary

war in Italy (A .0. 270-300).

(10) Roman Empire A.D. 300-535. Western and Eastern Empires.

(11) Gothic War in the middle of the 6th century A.D. in Italy (535-552). (12) Medieval papal Rome from A.D. 553 to the middle of the 10th century. (13) Carolingian Empire, including the empire of Charlemagne 681-887.

Wars of Charlemagne.

(14) Holy Roman Empire of the German Nation in the 10th to 13th century.

The war in the middle of the 13th century; the fall of the dynasty of the Hohenstaufen (1250-1268).

(15) Empire of the Habsburgs (1273-1619 or 1637).

Besides, (10)-(13) also include the medieval dynastic branches of the Eastern Roman and Byzantine Empire.

The parallels marked in Fig. 14 sometimes link all duplicate historical epochs, and sometimes only certain layers within these epochs. Certain epochs in the GCD can branch into several layers parallel to other epochs.

4.16.3. Consider the epoch of ancient Rome from 753 B.C. to 230 B.C. and that of medieval Rome in from 300 to 820. Remember that these epochs are "parallel" in the sense that the coefficient d measuring the proximity of the local maxima of the volume function (for the primary sources describing these periods of history) is very small and equals 6 . 10-11• This parallelism (overlapping) is confirmed by that of the enquete-codes which I found for the rulers' dynasties of these matching epochs. Moreover, I have discovered the parallelism of the events of the epoch of ancient Rome from 753 B.C. to A.D. 300 and that of medieval Rome from A.D. 300 to ca. 1353, which follows. The overlapping of parallel events occurs in shifting their dates by ca. 1,053 years. In other words, this rigid chronological shift can be written as the formula T = X + 300 years, where T are years A.D., and X the years from the foundation of the City (Rome). It is assumed traditionally that the year 1 since the foundation of Rome coincides with 753 B.C. In our forward shift, the "foundation of the city" falls in the year A.D. 300. I discovered this important "uniform forward shift" formula as a result of applying the enquetecode method and the method for the calculation of d. It turns out that Eav. = + 18 in the time interval from the year 1 to 250 since the foundation of Rome (when compared with the duplicate period A.D. 300-550). In the next time

66

New Statistical Methods for Dating

Chapter 2

interval, A.D. 550-820, the small value of the coefficient d calculated above for the duplicate epoch 250-520 since the foundation of Rome is consistent with the existence of a whole series of far-reaching parallelisms linking these two periods, i.e., the antique and medieval ones. An additional analysis of this overlapping (in the interval from A.D. 553 to 820) was then carried out by E.M. Nikishin. From 1978 to 1979, I also investigated the next time interval from the middle of the 9th to the 17th century (which overlaps with the period from 200 B.C. to A.D. 570) with the aid of the enquete-code method. The result is shown in the GCD in Fig. 14. For details, see Vol. 2.

4.16.4. As noted above, the concept of a text with a scale is more general than the examples of historical, narrative texts given above. For example, as a text X, we can take the collection of all the works of one author, as the parameter the numbers of pages (with consecutive pagination), and some quantitative characteristic of the text, for example, the average length of sentences, frequency of conjunctions, and so forth, as the informative function. The question arises whether there exist any conservation laws controlling the behaviour of such informative functions. It turns out that the answer is positive (see the author's paper "Authorial invariants in Russian literary texts of narrative sources" in Methods of Quantitative Analysis of Texts of Narrative Sources, History Institute of AN SSSR, Moscow, 1982, pp. 86-109; in Russian).

We stress that the present chapter is only a brief survey of the theses, the detailed treatment of each of which is rather voluminous and requires an extensive machinery designed for lengthy statistical material.

References

[1] Sergeev, V.S., Essays on the History of Ancient Rome. Sotsekgiz, Moscow, 1938 (in Russian).

[2] Livy, Titus, Works. Cambridge, Harvard University Press; Heinemann, London, 1914.

[3] Gregorovius, F., History of the City of Rome in the Middle Ages. G. Bell & Sons, London, 1900-1909.

[4] Blair, J., Blair's Chronological and Historical Tables, from the Creation to the Present Time, etc. G. Bell & Sons, London, 1882.

[5] Niese, B. Grundrift der riimischen Geschichte nebst Quellenkunde, 2nd ed. Miinchen, 1923.

[5*] Russian transl. (from the 1st ed.), St. Petersburg, 1908.

§5

Chronology of Ancient Dynasties

67

§5. A Method of Duplicate Recognition and Some Applications to the Chronology of Ancient Dynastiea'

5.1. The process of measuring random variables

Let a finite set of points D and a certain many-valued mapping V : D -+ R" which transforms D into a larger, but still finite set of points V(D) be given in the Euclidean space R". For example, V can model a multiple process of measuring a certain random discrete variable e taking values in the set D. The many-valuedness can be caused by the nonuniqueness of the results of measuring due to the existence of random errors. Meanwhile, V(D) can be regarded as the set of values obtained by measuring the given random variable. Note that each true value z of e turns into the set of points (values) V(z) upon its measuring. It represents the original value z of the variable e. In particular, each point of V (z) can be regarded as an approximate value of the true value x,

In studying real processes of measuring random variables, the principal difficulty lies in correctly modelling, by means of the choice of a suitable mapping V, the mechanism of real measuring errors. Now, let the set D of the real values be unknown, and only the set V(D) of the "results of measuring" the desired random variable be known. How can we recognize those points of V(D) which correspond to the same point in the set D? The points (results of measuring) associated with the same real, true value will be called its duplicates, which, in turn, can naturally be called "original".

Let the mapping V be such that the sets V(z) and V(y) are disjoint if the points z and y are different.

5.2. The distance between two random vectors

We introduce a certain natural measure A of the distance between the points in the set V(D). We shall strive to make the points which belong to the same set V(x) [i.e., duplicates) sufficiently close in the sense of the measure A. On the contrary, points from different V(x) and V(y) should be distant in the sense of A.

Let a and b be two points in V(D). Fix a and construct its special neighbourhood Hr. We shall attempt to make the point a the centre of Hv, and let b lie on the boundary of the neighbourhood or close to it.

The simplest of such neighbourhoods is H: = {c E R" : lai - Ci I < lai - bi I, 1 < i < n}. In other words, H~ is simply a parallelepiped with its centre at a and having b as one of its vertices, where a = (at, ... , an), b = (b1, ... ,bn).

For this simplest construction to become suitable in important applications, we have to extend it to model the mechanism of the random errors of interest, which influence the measurements of the true values of the variable e. We

1 First published as an article in DAN SSSR 258, 6(1981), pp. 1326-1333 (in Russian).

68

New Statistical Methods for Dating

Chapter 2

construct such a neighbourhood Hr(a, b) below and thereby introduce the natural measure A, permitting us to estimate the distance between two points a and b. As the basis for defining the measure, we shall take the procedure developed in §4. Namely,

A( b) = vol Hr(a, b)

a, vol V(D) ,

where vol V(D) is the number of points in the whole set V(D), and vol Hr(a, b) that from V(D) in the neighbourhood Hr(a,b).

5.3. Dynasties of rulers. The real dynasty and the numerical dynasty. Dependent and independent dynasties. The small-distortion principle

We shall now describe a concrete problem for whose solution we introduce the measure A. Let a historical text describing a previously unknown dynasty of rulers be discovered with an indication of the duration of their reigns. The question arises whether this historical dynasty is new and not mentioned in the known documents or whether it is one of the rulers' dynasties already known to us but described in a text in unusual terms (with the rulers' names distorted, etc.).

Consider n consecutive authentic rulers (kings). Let the true rule durations of these kings be PI I P2, ... ,Pn, respectively. We call this sequence a real dynasty. Note that the same real dynasty of rulers is often described in the primary sources from different standpoints by different chroniclers. But there exist more or less "invariant facts" concerning these rules, and their description depends little on the tastes of the author of a primary source (chronicler). Such facts include, for example, the duration of a king's rule, since there are usually no special reasons for which the chronicler should considerably and intentionally distort it. Nevertheless, chroniclers often encounter serious difficulties in calculating the regal rule duration, which leads to giving different values to the duration of the rule of the same historical character in different historical documents.

Thus, each author (chronicler), while describing a real dynasty P = (PI,P2, ... ,Pn), calculates the duration ai of a king's rule and obtains a certain sequence of numbers a == (al' a2,' .. ,an), This sequence of numbers represented as an integral vector a in the space Rn will be called a numerical dynasty. Another chronicler, while describing the same real dynasty of kings, will possibly obtain another vector b from Rn, i.e., another numerical dynasty. Thus, one and the same real dynasty can be represented as different numerical dynasties in different documents.

As the set D described in 5.1, we take a sufficiently large set of real dynasties of length n, i.e., D = {p = (PI, ... , Pn)}. We formulate the following theoretical model (statistical hypothesis).

§5

Chronology of Ancient Dynasties

69

The small-distortion principle. If two numerical dynasties are sufficiently close (in the sense of the measure >..), then they indeed represent the same real dynasty of kings, i.e., they are merely two different versions of its description.

Such numerical dynasties will be called dependent. On the contrary, if two numerical dynasties represent two real dynasties of kings, known a priori as different, then the numerical dynasties are much different from one another (in the sense of the measure >..). Such numerical dynasties will be called independent.

Later in Section 5.4, before verifying this model experimentally, we shall give an exact description of the measure >... Meanwhile, we identify the set of all numerical dynasties describing real historical dynasties from the set D in the space H" (see above) with the set V(D).

5.4. Basic errors leading to controversy among chroniclers as to the duration of kings' rules

We now point out concrete errors most often leading to the controversy among chroniclers as to the duration of the rules of kings.

(a) Permutation of the names of (or confusion between) two neighbouring rulers.

(b) Replacing two neighbouring rulers by one, the duration of whose rule was assumed to be equal to the sum of the rule durations of both.

( c) Computational error by a chronicler. The longer the duration of a king's rule, the greater the error that arises in its computation.

It turns out that these three basic types of concrete errors made by chroniclers can be sufficiently simply described by means of a suitable mapping, V : D --+ R", Let P be a certain real dynasty in the set D. We call the dynasty (vector) c a virtual variation (virtual vector) of the dynasty p and write c = v(p) if the following conditions are fulfilled, namely, each coordinate c, of c coincides with one of the three coordinates of the original vector p, i.e., Pi-I, Pi, Pi+l, or with Pi + Pi+l·

It is clear that each of such virtual vectors, or virtual dynasties, can be

regarded as a numerical dynasty and be obtained from a real dynasty p, because of chroniclers' errors of type (a) and (b).

Eventually, we take as V(D) the union of all virtual vectors (virtual dynasties) c = v(p), where P ranges over all the real dynasties of D. It remains to model an error of type (c).

Let a piecewise smooth, nonnegative function aCt) be given on the positive half-axis t > O. In our case, the role of aCt) will be played by the probability density of a random variable 1] to be specified below. Put h(t) = f(a(t», where f(8) is a certain monotonically decreasing function of a parameter 8, given on the half-axis 8 > 0, and such that lim j'{s] = +00 when 8 --+ +0. For example, as f(s), we can take the function ;. If 1] is a discrete random variable with probability density aCt), then the quantity h(t) becomes greater as 1] assumes the value t with lesser probability. In our problem, we take as 1] the duration of a king's rule in a dynasty. Let t range over all positive integers

70

New Statistical Methods for Dating

Chapter 2

(i.e., possible values of the rule duration). If t is a certain fixed rule duration, or the value of '1, then a(t) will mean the number of historical characters ruling for t years (see Fig. 12). We call h(t) the error amplitude in measuring a rule t years long. The graph aCt) in Fig. 12 shows that short rules are most frequent, and, conversely, long rules are rare.

5.5. The experimental frequency histogram for the duration of the

rules of kings

§4 showed the experimental frequency histogram I obtained for the rule durations of authentic kings (Fig. 12). If t is the value (rule duration) taken by the random variable '1 with large probability, then the amplitude of chroniclers' errors h(t) decreases. In other words, the values of short rule durations of frequently mentioned kings have been calculated by the chroniclers better than long ones, which are rarely encountered. We now indicate the error function h( t) of chroniclers, which we calculated for the probability density of the random variable, namely the "rule duration" (see §4).

Break the interval from zero to 100 on the integral axis t into smaller segments of the form (10k, 10k + 9), where k = 0,1, ... ,9. Then the amplitude h(t) of the chroniclers' error has the following form, namely,

I 2

h(t) = 3

5([1'0] - 1)

when 0 <t < 20 when 20 < t < 30 when 30 < t < 100.

5.6. Virtual dynasties and a mathematical model for errors made

by the chronicler in measuring the rule duration

Consider a rectangular parallelepiped IT(a,b) in the space fl!1 and denote it simply by IT. Its orthogonal projections 7r; = ai ± (Ia; - bil + h(ai» onto the coordinate axes in the space Rn will be given by intervals with the following endpoints, namely,

if 0 < ai < 20

if 20 < a; < 30 if 30 < ai < 100,

where [y] denotes the integral part of the number y. Thus, if 0 < ai < 20, then the rule duration at, and also bi, of two kings numbered i in the dynasties compared, is considered by us only approximately, to the accuracy of ±1 year. In other words, this is an error of the chronicler, made in measuring the rule duration. If 20 < ai < 30, then the chronicler's error is already equal to ±1.5 years, and so forth. We now fix two dynasties a and b.

It remains to model the fact that the assignment of a point (dynasty) c from the set of virtual dynasties V(D) to the parallelepiped IT can be considered only approximately, with some allowance. Hence, we have to make the

§5

Chronology of Ancient Dynasties

71

boundary of TI less distinct. Let r be a certain fixed number. Consider a real dynasty P from the set D. Assume that at least r coordinates Pi of this vector p, i.e., r values for the rule duration, have fallen onto the projections 1ri of TI. We assume, in addition, that a certain virtual dynasty c = v(p) of p has fallen entirely into TI. We say that such vectors (dynasties) p from Dare r-close to the parallelepiped TI determined by the two fixed dynasties a and b.

Eventually, we define the neighbourhood H; (a, b) of a by considering the union of TI and all the virtual variations of dynasties (vectors) p from the set D, which happened to be r-close to TI (see Fig. 15(2)).

(=15) k

Parallelepiped n(M. N)

rr::-------------------."

1 <, ,," ,

1 ....... ,,"" I

1 ........... ",,~, --- .........

1 ......."." 1

I .... ,," I

I ,,>c... 1

1 ,," M ......... t

I,,'" ....1

I."." '... I

I ,,"" .......... I

~------------------~ H

A(M1)

n(M.H)

o

1

Figure 15(2). Parallelepiped determined by the two fixed dynasties M and N.

As a proximity measure for two dynasties a and b, we take the ratio of the number of dynasties (vectors) of the set V(D), which are in the neighbourhood Hr(a,b), to the total number of dynasties (vectors) in the set V(D). In counting the number of dynasties in Hr(a, b), we do not count the virtual variations of a and b, which are different from them.

The constructed number A has an important probabilistic interpretation.

Indeed, construct the function 4> of the probability density for the random vector from the set V(D) of all virtual dynasties, with the vector ranging over the dynasties. We divide the space R" into standard cubes of sufficiently small size, so that no point of V(D) has fallen onto any of their boundaries. If x is an interior point of some cube, then we take the following value as ¢( x), namely,

¢(x) = number of points of the set V(D) in this cube

total number of points in the set V(D) .

72

New Statistical Methods for Dating

Chapter 2

However, if z is on the boundary of the cube, then we put 4>(z) = O.

It is clear that the above measure ..\( a, 6) is the integral of 4> with respect to the set H; (a, 6), provided that H r( a, 6) consists of the cubes of our partition. Since we model here approximate calculations of the chroniclers, we can assume that this condition is fulfilled (if the cubes are sufficiently small).

Finally, the number ..\( a, 6) can now be regarded as the probability of the fact that the random vector distributed in the space Rn with the density function tP has fallen into the neighbourhood H; with its centre at a point (dynasty) a and the "radius" 16 - e] + h(a).

5.7. The small-distortion principle and a computer experiment

To verify the theoretical model of Section 5.3 (the small distortion principle), the chronological tables of J. Blair [1] and F. Ginzel [2] were used; they contain practically all the basic chronological data that survived for real historical dynasties. I have made a complete list of all dynasties of length n = 15 from the history of Europe, the Mediterranean region, the Near East, and Egypt from 4000 B.C. to A.D. 1800. The data have then been supplemented by information from 14 other chronological tables. The obtained list D turned out to represent certain real kings by several different numerical dynasties (due to the difference in the chroniclers' opinion). We now indicate the basic historical dynasties included in D.

Bishops and popes in Rome, Saracens, high priests in Judaea, Greeks in Bactria, exarchs in Ravenna, all dynasties of Pharaohs and other Egyptian rulers, dynasties of the Byzantine Empire, Roman Empire, Spain, Russia, France, Italy, the Ottoman Empire, Scotland, Lacedaemon (Sparta), Germany, Sweden, Denmark, Israel, Babylonia, Syria, Sicyon, Judaea, Portugal, Parthia, the Bosphorus, Macedonia, Poland, and England.

The total number of dynasties making up the virtual set V(D) in the space R15 turned out to be approximately equal to 15 . 1011. If, for some pair of virtual dynasties a and b, the number ..\( a, b) is sufficiently small, then the observable proximity of the dynasties a and 6 is a rare event; the rarer it is, the less is the coefficient ..\( a, b). As r in the numerical experiment, we have taken 11 equal to 1 + in by the "two-thirds" rule.

The author then performed an extensive computational experiment to determine ,\(a, b) for different pairs of dynasties a and b. The result fully confirmed the model of Section 5.3. Namely, the coefficient >.(a, b) turned out to oscillate for surely dependent numerical dynasties in the interval from 10-12 to 10-8. If, on the contrary, the numerical dynasties a and b are surely independent, then the coefficient >.(a, b) was not less than 10-3. The great (5th-order) difference between surely dependent and surely independent dynasties is manifest. The remaining ones make up a small percentage of the total number of dynasties.

Obviously, the above result permits us to solve the problem of distinguishing dependent numerical dynasties.

§5

Chronology of Ancient Dynasties

73

5.S. Pairs of dependent historical dynasties previously regarded as

independent

Our experiment has discovered several special pairs of historical dynasties a and b which previously had been regarded as independent in all senses; however, the value of the coefficient ..\( a, b) for them is the same as for pairs of surely dependent dynasties. There are only several dozen such special pairs among the 106 of dynastic pairs studied.

We shall illustrate this with some examples.

(1) The Roman Empire from 82 B.C. to A.D. 217 and the Roman Empire from 270 to 526, where ..\ = 1.3 . 10-12.

(2) The Holy Roman Empire of the German Nation from 962 to 1254 and the Habsburgs Empire from 1273 to 1619, where ..\ = 1.2.10-12.

(3) The Roman Empire from 270 to 553 (see example 1) and the Holy Roman Empire of the German Nation (see example 2) from 962 to 1254, where ..\ = 2.3 . 10-10•

(4) The Carolingians, the empire of Charlemagne from 681 to 887 and the Eastern Roman Empire from 333 to 527, where ..\ = 8.25 . 10-9•

5.9. The distribution of dependent dynasties in the "modern

textbook" of ancient history

All the above results have been analyzed in the following manner. I have constructed the aCD for all the historical dynasties described in Section 5.7 (see also §4 and [3]), for which the rule duration periods for all the rulers of the indicated dynasties from list D and the dates of all basic events occurring in the time interval from 4000 B.C. to A.D. 1800 were marked off on the horizontal time axis (as horizontal intervals of different length).

The GCD was then subjected to the procedure of discovering duplicates, or dependent epochs. All the discovered historical dynastic pairs a and b, and the corresponding historical epochs, for which the coefficient ..\(a, b) turned out to be anomalously small, of the order from 10-12 to 10-8, were marked on the GCD. We will call such dynasties (and also the epochs) duplicates.

Recall that the theoretical model of Section 5.3 has been confirmed by the results of the experiment performed, from which it follows that the anomalous small value of ..\( a, b) most probably indicates the dependence of the historical dynasties and their corresponding epochs.

We now describe the portion E of the GCD in the time interval from 1600 B.C. to A.D. 1800. We represent the result as a schematic line E made up of consecutive letters indicating on the time axis different dynasties and the corresponding epochs. Duplicate epochs will be represented by identical letters. Because of the enormousness of the data, we give here only a rough sketch of the aCD (for details, see Vol. 2 of this book). The letters in the numerator and denominator of a fraction represent simultaneous epochs. Thus, the epoch E has the form:

74

New Statistical Methods for Dating

Chapter 2

E=T KTH TT K T H T K T T K TH T T PT C
- - - - - -
P C P n n c
c P
-
P
1600 B.C. 753 B.C. 82 B.C. A.D. 250 A.D. 962 A.D. 1619 It is obvious that E contains a repetition, duplicating each other's epochs.

Besides, it decomposes into the sum of four almost identical copies of shorter chronicle lines (see Fig. 9). We can schematically write that E = C1 + C2 + C3 + C4• Line C1 is obtained by gluing the chronicles Co and C' together.

Thus, all four chronicle lines Cl, C2, C3, and C4 are practically identical, being only different in their position on the time axis.

5.10. Dependent dynasties in the Bible and parallel with European history

. There are also other pieces (chronicles) in the GCD containing duplicates.

Consider an example: the chronicle line B embracing the events from 4000 to 586 B.C., described in the Old Testament. We borrowed their chronology from Blair's traditional tables (Tables 1-7 in [1]; more precisely, see the data from Columns 1 and 2 of Tables 1, 2, and 3; Columns 1, 2, and 3 of Table 4; Columns 1 and 2 of Table 5; Column 1 of Table 6; and finally, Column 3 of Table 7). The historical events making up the chronicle line B have been described in Genesis, Exodus, Leviticus, Numbers, Deuteronomy, the Book of Joshua, the Book of Judges, the Book of Ruth, the First Book of Samuel, the Second Book of Samuel, the First Book of the Kings, the Second Book of the Kings, the First Book of Chronicles, the Second Book of Chronicles, the Book of Ezra, the Book of Nehemiah, and the Book of Esther. Traditionally, these events are believed to have occurred in the Near East, i.e., in a region different from that determined by the events composing the chronicle line E (Europe, the Mediterranean region; see above). The application of our method for duplicate recognition and those methods described in §4 lead to the discovery of duplicates in B, which are distributed as follows (for details, see Vol. 2).

B=T K T H T K T K T H T T P T ~, n

p

where the epoch Ca is part of C. It is not accidental that we have employed the same symbols in describing the biblical chronicle B as for the European chronicle E. We see that B coincides with a certain part of E, i.e., overlaps it (there is parallelism of events). Namely, the following equality is valid:

References

75

E = T ]( T H T (T K T HT K T

H

II KTPTTPTC)

p

C P

T II C p

or

E = T K T H T (chronicle B = Old Testament) PCPTn

C

p

The length of B equals ca. 2,300 years. Thus, the complete "textbook of modern history", i.e., the GCD, contains not only shortened redated chronicles of the forms E and B, but also parallel, or isomorphic, i.e., nearly coincident chronicles of considerable length. Meanwhile, they are traditionally treated today as chronicles describing different historical epochs.

The following general result is valid. The whole GCD, and not only the above chronicles E and B, can be completely restored from its lesser part Co describing the events placed to the right of A.D. 300, it being important that most of the events in Co are placed, in reality, even to the right of A.D. 960. In particular, the chronicle B can be practically completely restored from its lesser part which describes the events from A.D. 960 to 1400.

References

[1] Blair, J., Blair's Chronological and Historical Tables from the Creation to the Present Time, etc. G. Bell & Sons, London, 1882.

[2] Ginzel, F., Handbuch der mathematischen und technischen Chronologie, etc. Leipzig, 1906-1914.

[3] Fomenko, A.T., "On the computation of the second derivative of the moon's elongation", in Controllable Motion Problems: Hierarchical Systems. Perm University Press, Perm, 1980, pp. 161-166 (in Russian).

76

New Statistical Methods for Dating

Chapter 2

§6. ANew Empirico-Statistical Procedure for Text Ordering and Its Applications to the Problems of Dafing''

6.1. The chapter generation

This section presents one of the new methods for dating ancient events worked out on the basis of statistical principles which I initially formulated and verified in [1] and which I presented at the Third International Vilnius Conference on Probability Theory and Mathematical Statistics.

The goal of the method discussed in this section is to find a chronologically correct order of separate fragments of historical texts and to discover among them various duplicates, or repetitions, i.e., parts describing the same events.

We call the fragment of a historical text describing the events of (approximately) one generation a chapter generation (or simply chapter). Let a historical text X embrace the events in a sufficiently large time interval (A, B), i.e., from a year A to a year B. Assume that this text is broken (or can be broken) into separate chapters X (T), where T denotes the number of a generation (historical characters) described in a fragment of the text X (T). Meanwhile, we assume that numbering of the chapters X (T) is determined by their order in the text X. The obvious question arises: Have these chapters been ordered chronologically correctly by the author? If, however, the correct (chronological) numeration of the chapters has been lost (is unknown or doubtful), how can it be restored? In other words, how can the events described in the chapters X(T) be ordered chronologically correctly in time?

6.2. The frequency-damping principle

Let a time interval (A, B) described in a text be sufficiently large, i.e., tens of hundreds of years long. Then, as I discovered while quantitatively processing the information contained in a large set of concrete historical texts, the following important circumstance should be taken into account. It turns out that in the overwhelming majority of cases, different historical characters bear different full names in the text. This can be explained easily, though. As a matter of fact, a chronicler is interested in distinguishing between different historical characters in order to avoid any ambiguity. The simplest method to achieve this is to give different full names to different characters. The fact can be justified by checking experimentally.

We now formulate the theoretical frequency-damping principle.

In the chronologically correct ordering of chapter generations of a text X, the author changes historical characters while proceeding from the description of the events of one generation to those of the subsequent one. Namely, when describing those generations prior to a fixed one numbered To for a given ordering of chapters, the chronicler mentions no characters of To. With

1 First published as an article in DAN SSSR, 268, 6(1983), pp. 1322-1327 (in Russian).

§6

Text Ordering and Problems of Dating

77

a chronologically correct ordering of chapters, this can be explained by the simple fact that these personages have not yet been born. Then, when describing To, the chronicler speaks of the historical characters of this generation most often in chapter X(To). This is quite understandable, for the historical events described by the author are related to the personages born at that time. Finally, proceeding with the description of subsequent generations, the chronicler mentions the characters preceding the generation To less and less, which is also natural, because the author describes the new historical events of subsequent centuries whose personages certainly overshadow the deceased characters and the memories of them.

Since, due to the above remark, we can assume that the "identity" name = historical character is valid (see above), we shall now investigate the totality of all full names of personages mentioned in a text under investigation. As a rule, the term "full" will be omitted.

Consider the set of the names of personages first appearing, for a given ordering of chapters, in a chapter To of a text X. Denoting the number of mentions of all the names in the chapter X(To) by K(To, To), we count each name with its multiplicity and calculate the frequency of its being mentioned.

We then see how many times these names have been mentioned in a chapter X(T) and obtain a certain number K(To, T). We stress once again that if a certain name is encountered several times, then all these references are taken into account.

Thus, for each number To, we obtain a certain numerical graph K(To, T), where the argument T is variable. We can now reformulate the frequency-damping principle as follows.

In numbering the chapters chronologically correctly (i.e., chapters describing the same events), with duplicates being absent among them, each graph of ]«To, T) must have the following (theoretical) form. The function K(To, T) vanishes to the right of the point To while reaching its absolute maximum at the point To itself and decreasing monotonically to the right of To (Fig. 16).

The experimental check has completely confirmed (on the average) this frequency-damping principle for several dozen historical texts with a prescribed chronologically correct ordering of chapters (see [5]).

6.3. The method of finding the chronologically correct order of

chapters in a historical chronicle

We now describe the method of finding the chronologically correct order of chapters in a historical text X (or in a whole set of texts). Number all the chapters of the text X in a certain order, e.g., in which they occur in the text itself. We then determine the graph of K (To, T) described above for each separate chapter X (To). The number of these graphs will equal that of the chapters in the text X. All these values K(To, T) (for the variables To and T) are naturally organized into a square matrix K {T} of order n x n, where n is the total number of chapters in the text.

In the ideal (theoretical) case, the matrix K {T} has the form shown in

78

New Statistical Methods for Dating

Chapter 2

K{T}

a)

b)

Kaveraged (t)

1 n

Figure 16. The frequency-damping principle (for the case of chronologically correct ordering of chapters in the chronicle).

Fig. 16a. Namely, all the absolute maxima (lines and columns of the matrix) are concentrated on the principal diagonal. Then, the farther off the principal diagonal, the smaller (monotonically) the values K(To, T). The computational experiment has shown for real historical texts that, with a chronologically correct ordering of the chapters in a text X, the numbers I«To, T) decrease, on the average, monotonically not only with respect to the rows of the matrix I( {T} but also to its columns (see Fig. 16b).

In other words, the frequency of names (personages) of prior origin, i.e., from the earlier chapters X(T) mentioned in the fragment X(To), gradually decreases as the generation T creating them moves farther away from the generation To under investigation. It thus turned out that an increase in the age of a historical character (name) almost always causes a decrease in the frequency of references to this personage (name) in the subsequent chapters X (To). To estimate the rate and character of the frequency-damping graph for the references to a name, we can make use of the following averaged graph,

§6

Text Ordering and Problems of Dating

79

namely,

K (t) = Ei-To=t K (To, i)

avo t'

n-

where t = 0,1,2, ... , n - 1.

It is clear that it is obtained by averaging the square matrix K {T} with respect to all diagonals parallel to the principal.

Certainly, the experimental graphs K (To, T) may turn out not to be coincident with the theoretical graph for a concrete text.

It is obvious that, upon varying the original numbering of chapters X(T), the matrix K {T} and its entries also vary. As a matter of fact, there occurs a rather complicated redistribution of the names first appearing in a certain chapter X(To). Let us change the order of chapters of the text X by means of various permutations, which we denote by (1. We also designate the new chapter numeration corresponding to a permutation (1 performed by (1T. While calculating the new matrix K {(1Tl for each of these chapter permutations, we will seek (1, i.e., an order (1 of the text chapters, such that all or almost all frequency graphs of references to the names K(To, T) will have the almost theoretical form shown in Fig. 16. In particular, we will try to make the graph Kav.(t) maximally close to the ideal, monotonically damping graph in Fig. 16.

The order of the textual chapters, for which the deviation of the experimental matrix from the theoretical (damping) is the least, should be taken as chronologically correct and required.

This method of chapter ordering permits us to date ancient events. In fact, let a certain historical text Y be given for which it is only known that it describes some events from a historical epoch (A, B). Assume that we already have another dated text X describing the same epoch more or less completely. Let X be separated into the chapter generations X(T). How can we learn which generation exactly has been described in the text Y in question? We shall make use of the text X. Add Y to the collection of chapters X (T) of X, for which it suffices to assume that Y is a new chapter of X, and ascribe a certain number To to it, i.e., insert the chapter Y in place of To in the text X. Then, employing the above method, we find the optimal, i.e., chronologically correct order of all the chapters of the text X with the chapter Y added. Meanwhile, we shall therefore also find a chronologically correct place for the new chapter Y. The relative position which the text Y will occupy among other chapters of X should evidently be taken as the one desired. We thereby date the ancient events described in Y relative to the chapters of the text X.

This dating method has been checked against historical texts with an a priori known dating of the events described. The efficiency of the method has been fully confirmed (see [5]).

6.4. The frequency-duplicating principle and the method of duplicate recognition

We now account for a new method of duplicate recognition in a text and describe the frequency-duplicating principle. Let a historical epoch (A, B) be

80

New Statistical Methods for Dating

Chapter 2

described in some text X which is divided into individual chapters X(T). Assuming that they have been numbered generally in a chronologically correct manner, we suppose that there are two duplicates among the chapters, i.e., two chapters or fragments of the text, describing the events of one and the same generation. In other words, these chapters repeat each other, but are placed by the chronicler in different locations in the text X. Consider the simplest situation where the same chapter is repeated in X twice, numbered To as a chapter X(To) and Co as a chapter X(Co). We take To < Co.

1

n

K(Co1 T)

1

n

Figure 17. A new method of duplicate recognition (frequency-duplication principle).

It is evident that the frequency graphs /«To, T) and /«Co, T) have the form represented in Fig. 17. The first graph K (To, T) clearly does not satisfy the frequency-damping principle (two maxima). Hence, we have to permute the chapters of the text somehow to achieve better agreement with the theoretical damping graph in Fig. 16. Furthermore, we can see that the second graph vanishes, i.e., K(Co, T) = 0, which is explained by the fact that there are no new names appearing in the chapter X (Co) for the first time (they all have already appeared in the earlier chapter X(To». It then becomes evident that the best coincidence of the experimental frequency graph with the theoretical one in Fig. 17 is achieved when we juxtapose these two duplicates (i.e., chapters X (To) and X (Co» or simply identify them.

Thus, if the chapters of the text, which in general are numbered chronologically correctly, contain two whose frequency graphs have the form approximately represented in Fig. 17, then they are probably duplicates and should be identified. This is exactly what we call the frequency-duplicating principle. A similar reasoning is also valid for the case of several duplicates in a text.