You are on page 1of 20

Data-Driven Pyramidology: From

Pseudoscience to Data Science



Mark Carlotto (mark@carlotto.us)

ABSTRACT

A new method for estimating the ages of Egyptian pyramids using a data-driven approach is
described. Measurements of eight pyramids (Meidum, Bent, Red, Khufu, Khafre, Menkaure,
Sahure, and Neferirkare) reveal correlations that can be exploited and used to estimate the
ages of older pyramids from more recent ones. Preliminary results using a linear regression
model support the possibility that some pyramids could have been built over pre-existing
structures several hundred to several thousand years older than the pyramids themselves.

Introduction
The age of certain pyramids in Egypt, particularly those in Giza, is a long-debated topic within
the archaeological community. The accepted age of Khufu, around 2554 BCE, was challenged
by Bauval (1989, 1993), who calculated an earlier date of 2450 BCE based on the alignment
of two shafts – one to the middle star (Zeta Orionis) in the “belt” of the constellation Orion,
and the other to Thuban, the former pole star. Moving the date of Khufu almost a century
earlier turned out to be consistent with Spence’s revised dating of 2480±5 BCE. Spence
(2000) hypothesized the Egyptians used the simultaneous transit of two circumpolar stars,
Mizar and Kochab, to align the pyramids during a period of time when there was no pole star.

While it is generally assumed the pyramids were aligned using the stars, Isler (1989) argues
that at least some of the pyramids could have been aligned to the sun. In analyzing
correlations between measurements of Old Kingdom pyramids, we propose that regardless
of how they were aligned, other factors not previously considered could have determined
their alignment accuracy. A new method for estimating the age of the Giza pyramids using a
data-driven approach is described that exploits these correlations. Preliminary results
suggest certain pyramids could be several hundred to several thousand years older.

Using Precession to Date the Pyramids


When the earth’s axis is aligned with a star such as Polaris, a line drawn from the star down
to the horizon defines the direction of true north. Circa 3000 BCE, Thuban in the constellation
of Draco served as the pole star (Figure 1). During Egypt’s Pyramid Age, there was no pole
star to use as a geodetic reference. Spence (2000) proposed that the simultaneous transit of
two circumpolar stars, Mizar and Kochab, were used to align the pyramids during this period
of time (Figure 2). In 2467 BCE, the upper-lower culminations of these two stars aligned with
the North Pole. As a result of precession, pyramids constructed in the winter prior to this
date were misaligned to the west, while those constructed after this date were misaligned to
the east. (For pyramids constructed in the summer, the direction of the misalignments is
reversed.)

Table 1 lists the azimuth angle (mean and standard deviation), the length of the base and the
height (both of which we use later on), and the accepted date of construction of the eight
pyramids in Spence’s study: Meidum, Bent, Red, Khufu, Khafre, Menkaure, Sahure, and
Neferirkare. Also included are Djoser and Unas, which are examined in a more limited way.
For convenience, we define the following pyramid variables:

𝜇 and 𝜎 are the mean and standard deviation of the azimuth angle
𝜆 is the length of the base,
t is the year of construction,
𝜀 = |𝜇| is the alignment error.

Plotting the mean azimuth angle versus the year of construction (Figure 3) does not indicate
a strong correlation (𝑅 = 0.55) between the two variables. Only after switching the sign of
the azimuth angles of Khafre and Sahure is a clear linear relationship evident (𝑅 = 0.99),
which supports Spence’s hypothesis. This is a potential problem with the simultaneous
transit method if one does not know the season in which the pyramids were constructed
(Magli 2003). Another, more fundamental problem is how the Egyptians picked these stars
in the first place years before they actually lined up (Nell and Ruggles 2012).

By assuming that misalignment is caused by precession, the dates of all of the pyramids and
thus the accession dates of the associated pharaohs are shifted up in time (Figure 4). Spence
nimbly sidesteps this problem by increasing the uncertainty of more recent dates to avoid
conflict.

Aligning Structures to the Cardinal Directions


In their study of the Giza pyramids and associated structures, Nell and Ruggles state “the
east-west orientation in relation to sunrise or (in one case) sunset may have been a, or even
the, key factor in many cases” in determining their orientation. The data in Table 2 show the
N-S faces of Khufu, which use the E-W orientation, are better aligned to north than the E-W
faces, which use the N-S orientation. The extra step required to establish an N-S orientation
from E-W will tend to increase error in the N-S direction.

A well-known method of determining the E-W direction uses the “Indian Circle” (Isler 1989).
Recall the operation of a sundial where the gnomon casts a shadow on a flat surface (Figure
5). Instead of determining the time of day from the direction of the sun, we mark the path of
the sun during the course of the day. In the northern hemisphere in summer, the shadow is
cast to the south (Figure 6). On the equinox, the path of the sun is almost exactly east-west.
In winter, the shadow is cast to the north. The intersection of the sun’s path with a circle
drawn around the gnomon intersects at two points. The line between the two points is
almost exactly E-W. A small error is introduced due to the change in the earth’s declination
angle between sunrise and sunset. The error is largest on the equinox and smallest on the
solstices. On the spring equinox (in the northern hemisphere), the error is positive as the
obliquity is increasing, while on the fall equinox, it is negative. According to Dash (2017):
To lay in a baseline for a pyramid, the Egyptians would have had to extend
the line formed by these two points for hundreds [of] meters with little error…
The Egyptians could also have started with a larger gnomon, but they would
also have needed a larger platform on which to trace the shadow, and one
which was precisely leveled. Such a platform would have been available,
however, as the platform around the Great Pyramid is leveled to within a few
centimeters over its entire 920-meter periphery.
As illustrated in Figure 7, the E-W alignment error 𝜀 decreases as the radius of the Indian
Circle increases. The implication is that the alignment error should be negatively correlated
with the size of the structure.

Unlike sites such as Karnak aligned to solstices that can be dated based on changes in
obliquity (Lockyer 1894), the sun offers no direct way of dating sites that are cardinally
aligned. Is it possible to estimate the age of the pyramids using the data in other ways?

Data-Driven Estimation
Returning to the pyramid data in Table 1 we find a number of interesting correlations (Table
3). Base length and mean alignment are negatively correlated, 𝑅 = −0.842, where R is the
normalized correlation coefficient, indicating that alignment errors tend to decrease as the
size of the pyramid increases. This is what one would expect if the Indian Circle or some
variant was used to align the pyramids. There is also a strong positive correlation between
the age of a pyramid and its standard deviation (𝑅 = 0.837). When discussing the quality of
workmanship, Petrie (1885) reports the mean difference, or standard deviation, of a set of
measurements. Using the standard deviation as a measure of the quality of workmanship,
one can say that, on average older pyramids exhibit a higher quality of workmanship. Older
pyramids are also larger, and so age is negatively correlated with size (𝑅 = −0.748). Larger
pyramids also tend to exhibit a higher quality of workmanship (𝑅 = −0.812) and pyramids
with better workmanship tend to be more accurately aligned (𝑅 = 0.789).

That the pyramid data exhibit strong correlations between certain variables suggests the
possibility of using a data-driven approach to estimate the age of one group of pyramids
using data from another group. If we assume the dates of the 5th Dynasty pyramids are more
accurate than those of the previous dynasties, we can perform a regression analysis on these
pyramids and use the regression statistics to develop optimal estimators.

Consider a simple estimator that uses the length of a pyramid’s base 𝜆 to estimate its age

𝑡̂ = 𝑎𝜆 + 𝑏

𝑎
At least two measurement pairs are required to compute the parameters 𝛽 = ; <
𝑏

𝑡! , 𝜆! ; 𝑡" , 𝜆"

using the Normal Equation1

𝛽 = (𝑋 # 𝑋)$! 𝑋 # 𝑌

𝑡 𝜆 1
where 𝑌 = ;𝑡! < and 𝑋 = B ! C. Solving the Normal Equation yields the parameters:
" 𝜆" 1

(𝑡! − 𝑡" )
𝑎=
(𝜆! − 𝜆" )

(𝜆! 𝑡" − 𝜆" 𝑡! )
𝑏=
(𝜆! − 𝜆" )

Using the accepted years of construction and lengths of Sahure’s and Neferirkare’s bases, we
can estimate the year of construction of the other pyramids from the length of their base
according to

𝑡̂ = −1.85𝜆 − 2299

which is plotted in Figure 8. Notice the estimate of Menkaure matches its accepted date
almost exactly. For earlier pyramids up to Meidum, the estimated ages are older than the
accepted ages. According to this estimate, Khufu would be 170 years older.

The standard deviation and alignment error are also correlated with age. Since there are two
estimation parameters (independent variables), together with the one dependent variable
(age), at least three measurement triplets are required

𝑡! , 𝜎! , 𝜀! ; 𝑡" , 𝜎" , 𝜀" ; 𝑡% , 𝜎% , 𝜀% ;

Instead of solving it by hand, we use Excel’s Data Analysis package. Since we don’t know the
standard deviation for Unas, we use values for Sahure, Neferirkare, and Meidum to estimate
the ages of the Bent, Red, Khufu, Khafre, and Menkaure pyramids based on their standard
deviations and alignment errors:

𝑡̂ = 16.6𝜎 + 1.85𝜀 − 2654

Like the previous plot the estimated ages (Figure 9) are also lower than the accepted ages of
the five late 3rd /early 4th Dynasty pyramids. We can combine these estimators and use the
base length, standard deviation, alignment error, and accepted date of Sahure, Neferirkare,
Menkaure, and Meidum to estimate the ages of the Bent, Red, Khufu, and Khafre pyramids
from their base length, standard deviation, and alignment error according to:


1 https://mathworld.wolfram.com/NormalEquation.html
𝑡̂ = −1.98𝜎 − 0.817𝜀 − 2.67𝜆 − 2196

Again the estimated ages are older than the accepted ages (Figure 10). A particularly
interesting result uses the base lengths of Sahure, Neferirkare, and Unas to estimate the ages
of the other pyramids (Figure 11). Here the estimated age of Khufu is approximately 800
years older than its accepted age.

Other Measures
The ability to precisely align the parts of a structure is indicative of a high level of technology.
Although the standard deviation is often used to measure the tolerance of an assemblage, it
does not characterize the level of difficulty involved. For example, if we assume that is it
twice as hard to align structures to a tolerance of ±½ inch than it is to a tolerance of ±1 inch,
the ratio 𝜂 = 1/𝜎 is a better measure of the precision and the level of difficulty required to
achieve it.

The ability to orient a structure along a particular direction would also seem to indicate a
high level of technological achievement. For the Indian Circle method (Figure 7), if r is the
measurement uncertainty (a random variable representing the radius of the circles marking
intersections of the solar path with the circle) and d is the distance between the intersections,
the angular uncertainty is given by:

tan(𝜀 ⁄2) ≈ 𝜀 ⁄2 = 2𝑟/𝑑

using the small-angle approximation. The alignment error thus depends on the
measurement uncertainty and the size of the structure, which is proportional to d. Since we
have already considered both of these, the orientation of a structure in a particular direction
does not provide any new information. On the other hand, the alignment of multiple
structures in a particular direction might (Carlotto 2019).

Another measurement not yet considered is the height of the pyramid. But instead of using
the height, consider its volume. The volume of a right rectangular pyramid with base lengths
𝜆 and height h is 𝜐 = 𝜆" ℎ⁄3. Correlations between error, precision, and volume show less
correlation with age but greater correlation among themselves (Table 4).

Using the volume of a pyramid instead of the length of its base leads to estimated ages that
are even more interesting than those in Figure 11. Figure 12 plots the results using data from
Sahure, Neferirkare, and Unas to estimate the age of the other pyramids based on their
volume:

𝑡̂ = −0.002663372𝜐 − 2189

According to this estimate, the late 3rd and 4th Dynasty pyramids could be thousands of years
older.

Using Unas in this and the previous estimator leads to significantly older dates for the early
pyramids. In place of Unas, the following estimator uses data from Sahure, Neferirkare, and
Menkaure to estimate the age of the pyramids based on their volume and precision (Figure
13):

𝑡̂ = 216𝜂 − 0.0016415𝜐 − 2307

Figure 14 summarizes the results of the six previous linear estimators plotting the average
estimated ages of the pyramids along with error bars indicating their range.

Discussion
Although the data used in this study are quite limited in scope, the results support the
possibility that late 3rd and 4th Dynasty pyramids could be significantly older than their
currently accepted age.

Vigato (2021) proposes that what later became the Giza pyramids were pre-existing
structures modified and expanded by Khufu, Khafre, and Menkaure. In particular, he
suggests the original structure that later became the pyramid of Khufu might have resembled
Mastaba 3038, thought to have been built toward the end of the First Dynasty, circa 2900
BCE. The possibility that the pyramids could be at least a few centuries older is supported
both by radiocarbon and optically stimulated luminescence (OSL) dating. Radiocarbon dates
of wood found within Khufu and Khafre are several hundred years older than the accepted
ages of the pyramids.2 Although neither Khufu nor Khafre has been OSL dated, Liritzis and
Vafiadou (2014) found Menkaure’s granite casing could be as much as a millennium older.
Samples from other structures on the Giza necropolis also indicate earlier dates.

In his survey of the Giza pyramids, Petrie (1885) states:
During the course of building there was evidently a great change in the style
of the work; a change, however, belonging more to the builders than to the
masons... In the King’s Chamber the masonry is very fine, both in its accuracy
of fitting and in the squareness and equal height of all the blocks; but the
builders were altogether wrong in their levels, and tilted the whole chamber
over to one corner… It would be difficult to suppose any architect allowing
such errors of building, after so closely restricting the variations of masons’
work, it strongly suggests that the granite had been prepared for the chamber
long before it was built, and that the supervision was less strict as the work
went on.
The change in workmanship noted by Petrie is at the level of the 35th course of masonry. It
is conceivable that the courses below could have been built first as part of an earlier structure
with those above added by Khufu. In this way, the original pyramid could be older and still
satisfy the two star/shaft alignments discovered by Bauval.

2 “Dating the Pyramids”, Archaeology, Vol. 52, No.5, September/October 1999.
Any theory attempting to estimate the ages of the Giza pyramids must also explain their size
and precision. An order of magnitude increase in the volume of the 4th Dynasty pyramids
over those in earlier and later dynasties is an anomaly that has yet to be explained by
mainstream Egyptology. Similarly, there has been no attempt to explain the increase in
precision in the 3rd and 4th Dynasties, followed by a marked decrease in the 5th Dynasty.
Precision is a measure of workmanship. Why did it improve only to decline in only a century?

The product of volume times precision (𝛼 = 𝜐𝜂 = 𝜐/𝜎) provides a single number, a
technology index, that can be used to characterize the level of technology involved in
building a pyramid. Figure 15 plots the technology index versus accepted date. How does
one explain what appears to be a century-long technological anomaly in an otherwise early
Bronze Age culture? If we assume certain pyramids are older, changing the order, the
anomaly becomes a decline. Figure 16 plots the technology index versus estimate date. The
four trendlines in the figure are based on different dating. The mean trend is the average of
the six age estimates. The min/max trends are one standard deviation above and below the
mean. The trend marked “same” plots the ages computed from the last estimator that uses
volume and precision. All of these trendlines are highly correlated with the estimated ages.
That the technology index versus estimate date decreases suggests pyramids thought to be
late 4th/early 5th Dynasty could have been built during an earlier period of technological
decline.

Although a great deal of uncertainty remains over their exact ages, it is possible the Bent,
Red, Khafre, and Khufu pyramids were either built much earlier or were built over pre-
existing structures. That these pyramids may pre-date the dynastic period suggests the
possibility they were not built by the Egyptians but by their predecessors.


References

W. M. Flinders Petrie (1885) The Pyramids and Temples of Gizeh, Second Edition, Published
by Histories & Mysteries of Man Ltd., London, 1990.

J. Norman Lockyer (1894) The Dawn of Astronomy, Cassell and Company, Ltd.

Martin Isler (1989) “An Ancient Method of Finding and Extending Direction,” Journal of the
American Research Center in Egypt.

Robert G. Bauval (1989) “A Master-Plan for the Three Pyramids of Giza Based on the
Configuration of the Three Stars of the Belt of Orion,” Discussions in Egyptology, ISSN 0266-
3083.

Robert G. Bauval (1993) “Cheops’s Pyramid: A New Dating Using the Latest Astronomical
Data,” ibid.

Kate Spence (2000) “Ancient Egyptian chronology and the astronomical orientation of
pyramids,” Nature, Vol. 408.

Giulio Magli (2003) “On the astronomical orientation of the IV dynasty Egyptian pyramids
and the dating of the second Giza pyramid.” See https://arxiv.org/abs/physics/0307100

Erin Nell and Clive Ruggles (2012) “The Orientations of the Giza pyramids and associated
structures,” The Journal of Astronomy in Culture, Vol. 25.

Ioannis Liritzis and Asimina Vafiadou (2014) “Surface Luminescence dating of some
Egyptian Monuments,” Journal of Cultural Heritage.

Glen Dash (2017) “Occam’s Egyptian razor: the equinox and the alignment of the pyramids,”
The Journal of Ancient Egyptian Architecture, Vol. 2.

Mark Carlotto (2019) “Geographical Alignments of Megalithic Sites in Northern Europe – An
Unexpected Finding.” See http://dx.doi.org/10.2139/ssrn.3486984

Marco M. Vigato (2021) “15 Reasons why Khufu did NOT build the Great Pyramid,” See
https://www.academia.edu/45014007/15_Reasons_why_Khufu_did_NOT_build_the_Great_
Pyramid





Table 1 Pyramid data. Mean azimuth, standard deviation, and error are in arc minutes. Length and
height are in meters. (Multiple sources including Spence and Wikipedia)

Dynasty Pyramid Year Az (Mean) Az (Stdev) Length Height Error


3 Djoser -2640 180 121 63 180
3 Meidum -2600 -20.6 1 144 94 20.6
4 Bent -2583 -17.3 0.2 189 101 17.3
4 Red -2572 -8.7 0.2 220 104 8.7
4 Khufu -2554 -3.4 0.2 230 147 3.4
4 Khafre -2522 -6 0.2 215 136 6
4 Menkaure -2489 12.4 1 105 66 12.4
5 Sahure -2446 -23 10 79 47 23
5 Neferirkare -2433 30 10 72 52 30
5 Unas -2317 17.1 58 43 17.1


Table 2 Comparison of average azimuth angles computed from Petrie’s table showing the north and
south sides of Khufu are more precisely aligned than the east and west sides.

Faces Casing Sockets Core Average


N-S -3.5 -1.04 -5 -3.17
E-W -3 925 -6.45 -5.5 -5.31


Table 3 Correlations between mean azimuth, standard deviation, length, alignment error, and year for
all pyramids except Djoser and Unas. Strong correlations are highlighted in bold type.

Year Az (Mean) Az (Stdev) Length


Az (Mean) 0.556
Az (Stdev) 0.837 0.285
Length -0.748 -0.346 -0.812
Error 0.485 0.202 0.789 -0.842


Table 4 Precision and volume are more highly correlated with each other and less correlated with year
than standard deviation and length.

Year Az (Mean) Precision Volume


Az (Mean) 0.556
Precision -0.638 -0.274
Volume -0.567 -0.224 0.909
Error 0.485 0.202 -0.789 -0.858





Figure 1 Effect of precession on the location of the north celestial pole. Currently, the pole points toward
Polaris. Circa 3000 BCE, it pointed toward Thuban, a faint star in the constellation Draco (top). For
more than 4000 years there was no pole star, including the period of Egypt’s Pyramid Age (bottom).
Reference star chart courtesy Sky Publishing Corporation.

Figure 2 Spence’s simultaneous transit method estimates the location of the north pole from two
circumpolar stars (top). In 2467 BCE, when two stars, Mizar and Kochab, were aligned vertically the
extension of the line through the two stars down to the horizon pointed precisely north (right). Horizon
view courtesy Your Sky.3



3 https://www.fourmilab.ch/yoursky/
40

30 Neferirkare

20

Menkaure
10

0
Khufu
Khafre
-10 Red

Bent
-20 Meidum
Sahure
-30
-2650 -2600 -2550 -2500 -2450 -2400


40

30 Neferirkare

Sahure
20

Menkaure
10
Khafre
0
Khufu
-10 Red

Bent
-20 Meidum

-30
-2650 -2600 -2550 -2500 -2450 -2400


Figure 3 Mean azimuth angle vs. assumed date of construction (top). Plot after angles for Khafre and
Sahure have been reversed (bottom). Angles are in arc minutes.
-2300 Accepted Spence
-2350

-2400

-2450

-2500

-2550

-2600

-2650

-2700
Meidum

Khufu

Khafre

Neferirkare
Menkaure

Sahure
Bent

Red

Figure 4 Spence’s method shifts the date of all pyramids up in time.

GNOMON

(x,y)


Figure 5 Path of the shadow cast by a solar gnomon.


South-North (y)

West-East (x)

South-North (y)

West-East (x)

South-North (y)

West-East (x)


Figure 6 Paths of the sun on the summer solstice (top), equinox (middle), and winter solstice (bottom)
in the northern hemisphere. Up is north in the graphs.
!

A B

Figure 7 Comparison of angular errors 𝜀 caused by measurement uncertainties in using the Indian circle
method. The small, filled circles represent the uncertainty in marking the location of intersections
between a circle drawn around a gnomon and the path of the sun (dotted line). The smaller Indian circle
(A) has a larger error 𝜀 than the larger circle (B).



-2200 Year Est. Year

-2300

-2400

-2500

-2600

-2700

-2800

-2900

Khafre
Meidum

Bent

Red

Khufu

Neferirkare
Menkaure

Sahure


Figure 8 Estimating the ages of pyramids from their base length using data from Sahure and
Neferirkare.



-2200 Year Est. Year

-2300

-2400

-2500

-2600

-2700

-2800

-2900
Khafre
Meidum

Bent

Red

Khufu

Neferirkare
Menkaure

Sahure



Figure 9 Estimating the ages of pyramids from their alignment error and standard deviation using data
from Sahure, Neferirkare, and Meidum.


-2200 Year Est. Year

-2300

-2400

-2500

-2600

-2700

-2800

-2900

Khafre
Meidum

Bent

Red

Khufu

Neferirkare
Menkaure

Sahure

Figure 10 Estimating the ages of the pyramids from their base length, alignment error, and standard
deviation using data from Sahure, Neferirkare, Menkaure, and Meidum.


-2200 Year Est. Year

-2400

-2600

-2800

-3000

-3200

-3400

-3600
Meidum

Khufu

Unas
Neferirkare
Khafre

Menkaure

Sahure
Djoser

Bent

Red



Figure 11 Estimating the ages of pyramids from their base length using data from Sahure, Neferirkare,
and Unas.


0 Year Est. Year
-1000
-2000
-3000
-4000
-5000
-6000
-7000
-8000
-9000
-10000

Khafre
Meidum

Bent

Red

Khufu

Unas
Neferirkare
Menkaure

Sahure
Djoser



Figure 12 Estimating the ages of pyramids from their volume using data from Sahure, Neferirkare, and
Unas.

-2000 Year Est. Year


-2500

-3000

-3500

-4000

-4500

-5000
-5500

-6000
Khafre
Meidum

Bent

Red

Khufu

Neferirkare
Menkaure

Sahure



Figure 13 Estimating the ages of pyramids from their volume and precision using data from Sahure,
Neferirkare, and Menkaure.


Estimat ed Accepted

-2000
-2500
-3000
-3500
-4000
-4500
-5000
-5500
-6000
Khafre

Meidum
Bent
Khufu

Red

Neferirkare
Menkaure

Sahure


Figure 14 Estimated ages of estimators developed and tested.




14000000
Khufu
12000000

Khafre
10000000

8000000 Red

6000000 Bent

4000000

2000000
Menkaure Neferirkare
Meidum Sahure
0
-2650 -2600 -2550 -2500 -2450 -2400

Figure 15 Product of precision and volume vs. accepted year shows a sharp rise and fall over about a
century.
Mean Min Max Same

14000000

12000000

10000000

8000000

6000000

4000000

2000000

0
-10000 -9000 -8000 -7000 -6000 -5000 -4000 -3000 -2000

Figure 16 Product of volume and precision vs. estimated year. Four trend lines represent different rates
of technological decline based on estimated age ranges.

You might also like