Download the full version of the ebook at

https://ebookmass.com

Introduction to Algorithms for Data Mining

and Machine Learning Xin-She Yang

https://ebookmass.com/product/introduction-to-
algorithms-for-data-mining-and-machine-learning-
xin-she-yang/

Explore and download more ebook at https://ebookmass.com

 Recommended digital products (PDF, EPUB, MOBI) that
you can download immediately if you are interested.

Introduction to algorithms for data mining and machine

learning Yang

https://ebookmass.com/product/introduction-to-algorithms-for-data-
mining-and-machine-learning-yang/

testbankdeal.com

Fundamentals of Machine Learning for Predictive Data

Analytics: Algorithms,

https://ebookmass.com/product/fundamentals-of-machine-learning-for-
predictive-data-analytics-algorithms/

testbankdeal.com

Machine Learning for Signal Processing: Data Science,

Algorithms, and Computational Statistics Max A. Little

https://ebookmass.com/product/machine-learning-for-signal-processing-
data-science-algorithms-and-computational-statistics-max-a-little/

testbankdeal.com

BIG DATA ANALYTICS: Introduction to Hadoop, Spark, and

Machine-Learning Raj Kamal

https://ebookmass.com/product/big-data-analytics-introduction-to-
hadoop-spark-and-machine-learning-raj-kamal/

testbankdeal.com
Machine Learning for Biometrics: Concepts, Algorithms and
Applications (Cognitive Data Science in Sustainable
Computing) Partha Pratim Sarangi
https://ebookmass.com/product/machine-learning-for-biometrics-
concepts-algorithms-and-applications-cognitive-data-science-in-
sustainable-computing-partha-pratim-sarangi/
testbankdeal.com

Machine Learning Algorithms for Signal and Image

Processing Suman Lata Tripathi

https://ebookmass.com/product/machine-learning-algorithms-for-signal-
and-image-processing-suman-lata-tripathi/

testbankdeal.com

Learn Data Mining Through Excel: A Step-by-Step Approach

for Understanding Machine Learning Methods, 2nd Edition
Hong Zhou
https://ebookmass.com/product/learn-data-mining-through-excel-a-step-
by-step-approach-for-understanding-machine-learning-methods-2nd-
edition-hong-zhou/
testbankdeal.com

Absolute Beginner’s Guide to Algorithms: A Practical

Introduction to Data Structures and Algorithms in
JavaScript (for True Epub) Kirupa Chinnathambi
https://ebookmass.com/product/absolute-beginners-guide-to-algorithms-
a-practical-introduction-to-data-structures-and-algorithms-in-
javascript-for-true-epub-kirupa-chinnathambi/
testbankdeal.com

Absolute Beginner's Guide to Algorithms: A Practical

Introduction to Data Structures and Algorithms in
JavaScript Kirupa Chinnathambi
https://ebookmass.com/product/absolute-beginners-guide-to-algorithms-
a-practical-introduction-to-data-structures-and-algorithms-in-
javascript-kirupa-chinnathambi/
testbankdeal.com
 Xin-She Yang

Introduction to
Algorithms for Data Mining
and Machine Learning
Introduction to Algorithms for Data Mining and
Machine Learning
This page intentionally left blank
Introduction to
Algorithms for Data
Mining and Machine
Learning

Xin-She Yang
Middlesex University
School of Science and Technology
London, United Kingdom
Academic Press is an imprint of Elsevier
125 London Wall, London EC2Y 5AS, United Kingdom
525 B Street, Suite 1650, San Diego, CA 92101, United States
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom
Copyright © 2019 Elsevier Inc. All rights reserved.
No part of this publication may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopying, recording, or any information storage and retrieval system, without
permission in writing from the publisher. Details on how to seek permission, further information about the
Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center
and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions.
This book and the individual contributions contained in it are protected under copyright by the Publisher (other
than as may be noted herein).
Notices
Knowledge and best practice in this field are constantly changing. As new research and experience broaden our
understanding, changes in research methods, professional practices, or medical treatment may become necessary.
Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using
any information, methods, compounds, or experiments described herein. In using such information or methods
they should be mindful of their own safety and the safety of others, including parties for whom they have a
professional responsibility.
To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability
for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or
from any use or operation of any methods, products, instructions, or ideas contained in the material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-817216-2

For information on all Academic Press publications

visit our website at https://www.elsevier.com/books-and-journals

Publisher: Candice Janco

Acquisition Editor: J. Scott Bentley
Editorial Project Manager: Michael Lutz
Production Project Manager: Nilesh Kumar Shah
Designer: Miles Hitchen
Typeset by VTeX
Contents

About the author ix

Preface xi
Acknowledgments xiii

1 Introduction to optimization 1
1.1 Algorithms 1
1.1.1 Essence of an algorithm 1
1.1.2 Issues with algorithms 3
1.1.3 Types of algorithms 3
1.2 Optimization 4
1.2.1 A simple example 4
1.2.2 General formulation of optimization 7
1.2.3 Feasible solution 9
1.2.4 Optimality criteria 10
1.3 Unconstrained optimization 10
1.3.1 Univariate functions 11
1.3.2 Multivariate functions 12
1.4 Nonlinear constrained optimization 14
1.4.1 Penalty method 15
1.4.2 Lagrange multipliers 16
1.4.3 Karush–Kuhn–Tucker conditions 17
1.5 Notes on software 18

2 Mathematical foundations 19
2.1 Convexity 20
2.1.1 Linear and affine functions 20
2.1.2 Convex functions 21
2.1.3 Mathematical operations on convex functions 22
2.2 Computational complexity 22
2.2.1 Time and space complexity 24
2.2.2 Complexity of algorithms 25
2.3 Norms and regularization 26
2.3.1 Norms 26
2.3.2 Regularization 28
2.4 Probability distributions 29
2.4.1 Random variables 29
2.4.2 Probability distributions 30
vi Contents

2.4.3 Conditional probability and Bayesian rule 32

2.4.4 Gaussian process 34
2.5 Bayesian network and Markov models 35
2.6 Monte Carlo sampling 36
2.6.1 Markov chain Monte Carlo 37
2.6.2 Metropolis–Hastings algorithm 37
2.6.3 Gibbs sampler 39
2.7 Entropy, cross entropy, and KL divergence 39
2.7.1 Entropy and cross entropy 39
2.7.2 DL divergence 40
2.8 Fuzzy rules 41
2.9 Data mining and machine learning 42
2.9.1 Data mining 42
2.9.2 Machine learning 42
2.10 Notes on software 42

3 Optimization algorithms 45
3.1 Gradient-based methods 45
3.1.1 Newton’s method 45
3.1.2 Newton’s method for multivariate functions 47
3.1.3 Line search 48
3.2 Variants of gradient-based methods 49
3.2.1 Stochastic gradient descent 50
3.2.2 Subgradient method 51
3.2.3 Conjugate gradient method 52
3.3 Optimizers in deep learning 53
3.4 Gradient-free methods 56
3.5 Evolutionary algorithms and swarm intelligence 58
3.5.1 Genetic algorithm 58
3.5.2 Differential evolution 60
3.5.3 Particle swarm optimization 61
3.5.4 Bat algorithm 61
3.5.5 Firefly algorithm 62
3.5.6 Cuckoo search 62
3.5.7 Flower pollination algorithm 63
3.6 Notes on software 64

4 Data fitting and regression 67

4.1 Sample mean and variance 67
4.2 Regression analysis 69
4.2.1 Maximum likelihood 69
4.2.2 Liner regression 70
4.2.3 Linearization 75
4.2.4 Generalized linear regression 77
4.2.5 Goodness of fit 80
Contents vii

4.3 Nonlinear least squares 81

4.3.1 Gauss–Newton algorithm 82
4.3.2 Levenberg–Marquardt algorithm 85
4.3.3 Weighted least squares 85
4.4 Overfitting and information criteria 86
4.5 Regularization and Lasso method 88
4.6 Notes on software 90

5 Logistic regression, PCA, LDA, and ICA 91

5.1 Logistic regression 91
5.2 Softmax regression 96
5.3 Principal component analysis 96
5.4 Linear discriminant analysis 101
5.5 Singular value decomposition 104
5.6 Independent component analysis 105
5.7 Notes on software 108

6 Data mining techniques 109

6.1 Introduction 110
6.1.1 Types of data 110
6.1.2 Distance metric 110
6.2 Hierarchy clustering 111
6.3 k-Nearest-neighbor algorithm 112
6.4 k-Means algorithm 113
6.5 Decision trees and random forests 115
6.5.1 Decision tree algorithm 115
6.5.2 ID3 algorithm and C4.5 classifier 116
6.5.3 Random forest 120
6.6 Bayesian classifiers 121
6.6.1 Naive Bayesian classifier 121
6.6.2 Bayesian networks 123
6.7 Data mining for big data 124
6.7.1 Characteristics of big data 124
6.7.2 Statistical nature of big data 125
6.7.3 Mining big data 125
6.8 Notes on software 127

7 Support vector machine and regression 129

7.1 Statistical learning theory 129
7.2 Linear support vector machine 130
7.3 Kernel functions and nonlinear SVM 133
7.4 Support vector regression 135
7.5 Notes on software 137
viii Contents

8 Neural networks and deep learning 139

8.1 Learning 139
8.2 Artificial neural networks 140
8.2.1 Neuron models 140
8.2.2 Activation models 141
8.2.3 Artificial neural networks 143
8.3 Back propagation algorithm 146
8.4 Loss functions in ANN 147
8.5 Optimizers and choice of optimizers 149
8.6 Network architecture 149
8.7 Deep learning 151
8.7.1 Convolutional neural networks 151
8.7.2 Restricted Boltzmann machine 157
8.7.3 Deep neural nets 158
8.7.4 Trends in deep learning 159
8.8 Tuning of hyperparameters 160
8.9 Notes on software 161

Bibliography 163

Index 171
 Visit https://ebookmass.com
now to explore a rich
collection of eBooks and enjoy
exciting offers!
About the author

Xin-She Yang obtained his PhD in Applied Mathematics from the University of Ox-
ford. He then worked at Cambridge University and National Physical Laboratory (UK)
as a Senior Research Scientist. Now he is Reader at Middlesex University London, and
an elected Bye-Fellow at Cambridge University.
He is also the IEEE Computer Intelligence Society (CIS) Chair for the Task Force
on Business Intelligence and Knowledge Management, Director of the International
Consortium for Optimization and Modelling in Science and Industry (iCOMSI), and
an Editor of Springer’s Book Series Springer Tracts in Nature-Inspired Computing
(STNIC).
With more than 20 years of research and teaching experience, he has authored
10 books and edited more than 15 books. He published more than 200 research pa-
pers in international peer-reviewed journals and conference proceedings with more
than 36 800 citations. He has been on the prestigious lists of Clarivate Analytics and
Web of Science highly cited researchers in 2016, 2017, and 2018. He serves on the
Editorial Boards of many international journals including International Journal of
Bio-Inspired Computation, Elsevier’s Journal of Computational Science (JoCS), In-
ternational Journal of Parallel, Emergent and Distributed Systems, and International
Journal of Computer Mathematics. He is also the Editor-in-Chief of the International
Journal of Mathematical Modelling and Numerical Optimisation.
This page intentionally left blank
Preface

Both data mining and machine learning are becoming popular subjects for university
courses and industrial applications. This popularity is partly driven by the Internet and
social media because they generate a huge amount of data every day, and the under-
standing of such big data requires sophisticated data mining techniques. In addition,
many applications such as facial recognition and robotics have extensively used ma-
chine learning algorithms, leading to the increasing popularity of artificial intelligence.
From a more general perspective, both data mining and machine learning are closely
related to optimization. After all, in many applications, we have to minimize costs,
errors, energy consumption, and environment impact and to maximize sustainabil-
ity, productivity, and efficiency. Many problems in data mining and machine learning
are usually formulated as optimization problems so that they can be solved by opti-
mization algorithms. Therefore, optimization techniques are closely related to many
techniques in data mining and machine learning.
Courses on data mining, machine learning, and optimization are often compulsory
for students, studying computer science, management science, engineering design, op-
erations research, data science, finance, and economics. All students have to develop
a certain level of data modeling skills so that they can process and interpret data for
classification, clustering, curve-fitting, and predictions. They should also be familiar
with machine learning techniques that are closely related to data mining so as to carry
out problem solving in many real-world applications. This book provides an introduc-
tion to all the major topics for such courses, covering the essential ideas of all key
algorithms and techniques for data mining, machine learning, and optimization.
Though there are over a dozen good books on such topics, most of these books are
either too specialized with specific readership or too lengthy (often over 500 pages).
This book fills in the gap with a compact and concise approach by focusing on the key
concepts, algorithms, and techniques at an introductory level. The main approach of
this book is informal, theorem-free, and practical. By using an informal approach all
fundamental topics required for data mining and machine learning are covered, and
the readers can gain such basic knowledge of all important algorithms with a focus
on their key ideas, without worrying about any tedious, rigorous mathematical proofs.
In addition, the practical approach provides about 30 worked examples in this book
so that the readers can see how each step of the algorithms and techniques works.
Thus, the readers can build their understanding and confidence gradually and in a
step-by-step manner. Furthermore, with the minimal requirements of basic high school
mathematics and some basic calculus, such an informal and practical style can also
enable the readers to learn the contents by self-study and at their own pace.
This book is suitable for undergraduates and graduates to rapidly develop all the
fundamental knowledge of data mining, machine learning, and optimization. It can
xii Preface

also be used by students and researchers as a reference to review and refresh their
knowledge in data mining, machine learning, optimization, computer science, and data
science.

Xin-She Yang
January 2019 in London
Acknowledgments

I would like to thank all my students and colleagues who have given valuable feedback
and comments on some of the contents and examples of this book. I also would like to
thank my editors, J. Scott Bentley and Michael Lutz, and the staff at Elsevier for their
professionalism. Last but not least, I thank my family for all the help and support.

Xin-She Yang
January 2019
This page intentionally left blank
Introduction to optimization
Contents
1.1 Algorithms
1 1
1.1.1 Essence of an algorithm 1
1.1.2 Issues with algorithms 3
1.1.3 Types of algorithms 3
1.2 Optimization 4
1.2.1 A simple example 4
1.2.2 General formulation of optimization 7
1.2.3 Feasible solution 9
1.2.4 Optimality criteria 10
1.3 Unconstrained optimization 10
1.3.1 Univariate functions 11
1.3.2 Multivariate functions 12
1.4 Nonlinear constrained optimization 14
1.4.1 Penalty method 15
1.4.2 Lagrange multipliers 16
1.4.3 Karush–Kuhn–Tucker conditions 17
1.5 Notes on software 18

This book introduces the most fundamentals and algorithms related to optimization,
data mining, and machine learning. The main requirement is some understanding of
high-school mathematics and basic calculus; however, we will review and introduce
some of the mathematical foundations in the first two chapters.

1.1 Algorithms
An algorithm is an iterative, step-by-step procedure for computation. The detailed
procedure can be a simple description, an equation, or a series of descriptions in
combination with equations. Finding the roots of a polynomial, checking if a natu-
ral number is a prime number, and generating random numbers are all algorithms.

1.1.1 Essence of an algorithm

In essence, an algorithm can be written as an iterative equation or a set of iterative
equations. For example, to find a square root of a > 0, we can use the following
iterative equation:
1 a
xk+1 = xk + , (1.1)
2 xk
where k is the iteration counter (k = 0, 1, 2, . . . ) starting with a random guess x0 = 1.
Introduction to Algorithms for Data Mining and Machine Learning. https://doi.org/10.1016/B978-0-12-817216-2.00008-9
Copyright © 2019 Elsevier Inc. All rights reserved.
2 Introduction to Algorithms for Data Mining and Machine Learning

Example 1
As an example, if x0 = 1 and a = 4, then we have

1 4
x1 = (1 + ) = 2.5. (1.2)
2 1

Similarly, we have

1 4 1 4
x2 = (2.5 + ) = 2.05, x3 = (2.05 + ) ≈ 2.0061, (1.3)
2 2.5 2 2.05
x4 ≈ 2.00000927, (1.4)
√
which is very close to the true value of 4 = 2. The accuracy of this iterative formula or algorithm
is high because it achieves the accuracy of five decimal places after four iterations.

The convergence is very quick if we start from different initial values such as
x0 = 10 and even x0 = 100. However, for an obvious reason, we cannot start with
x0 = 0 due to division by
√zero.
Find the root of x = a is equivalent to solving the equation

f (x) = x 2 − a = 0, (1.5)

which is again equivalent to finding the roots of a polynomial f (x). We know that
Newton’s root-finding algorithm can be written as

f (xk )
xk+1 = xk − , (1.6)
f  (xk )

where f  (x) is the first derivative or gradient of f (x). In this case, we have
f  (x) = 2x. Thus, Newton’s formula becomes

(xk2 − a)
xk+1 = xk − , (1.7)
2xk

which can be written as

xk a 1 a
xk+1 = (xk − )+ = xk + ). (1.8)
2 2xk 2 xk

This is exactly what we have in Eq. (1.1).

Newton’s method has rigorous mathematical foundations, which has a guaranteed
convergence under certain conditions. However, in general, Eq. (1.6) is more general,
and the gradient information f  (x) is needed. In addition, for the formula to be valid,
we must have f  (x) = 0.
Introduction to optimization 3

1.1.2 Issues with algorithms

The advantage of the algorithm given in Eq. (1.1) is that√it converges very quickly.
However, careful readers may have asked: we know that 4 = ±2, how can we find
the other root −2 in addition to +2?
Even if we use different initial value x0 = 10 or x0 = 0.5, we can only reach x∗ = 2,
not −2.
What happens if we start with x0 < 0? From x0 = −1, we have
1 4 1 4
x1 = (−1 + ) = −2.5, x 2 = (−2.5 + ) = −2.05, (1.9)
2 −1 2 −2.5
x3 ≈ −2.0061, x4 ≈ −2.00000927, (1.10)
which is approaching −2 very quickly. If we start from x0 = −10 or x0 = −0.5, then
we can always get x∗ = −2, not +2.
This highlights a key issue here: the final solution seems to depend on the initial
starting point for this algorithm, which is true for many algorithms.
Now the relevant question is: how do we know where to start to get a particular
solution? The general short answer is “we do not know”. Thus, some knowledge of
the problem under consideration or an educated guess may be useful to find the final
solution.
In fact, most algorithms may depend on the initial configuration, and such algo-
rithms are often carrying out search moves locally. Thus, this type of algorithm is
often referred to as local search. A good algorithm should be able to “forget” its initial
configuration though such algorithms may not exist at all for most types of problems.
What we need in general is the global search, which attempts to find final solutions
that are less sensitive to the initial starting point(s).
Another important issue in our discussions is that the gradient information f  (x) is
necessary for some algorithms such as Newton’s method given in Eq. (1.6). This poses
certain requirements on the smoothness of the function f (x). For example, we know
that |x| is not differentiable at x = 0. Thus, we cannot directly use Newton’s method
to find the roots of f (x) = |x|x 2 − a = 0 for a > 0. Some modifications are needed.
There are other issues related to algorithms such as the setting of parameters, the
slow rate of convergence, condition numbers, and iteration structures. All these make
algorithm designs and usage somehow challenging, and we will discuss these issues
in more detail later in this book.

1.1.3 Types of algorithms

An algorithm can only do a specific computation task (at most a class of computational
tasks), and no algorithms can do all the tasks. Thus, algorithms can be classified due
to their purposes. An algorithm to find roots of a polynomial belongs to root-finding
algorithms, whereas an algorithm for ranking a set of numbers belongs to sorting
algorithms. There are many classes of algorithms for different purposes. Even for the
same purpose such as sorting, there are many different algorithms such as the merge
sort, bubble sort, quicksort, and others.
 Visit https://ebookmass.com
now to explore a rich
collection of eBooks and enjoy
exciting offers!
4 Introduction to Algorithms for Data Mining and Machine Learning

We can also categorize algorithms in terms of their characteristics. The root-finding

algorithms we just introduced are deterministic algorithms because the final solutions
are exactly the same if we start from the same initial guess. We obtain the same set of
solutions every time we run the algorithm. On the other hand, we may introduce some
randomization into the algorithm, for example, using purely random initial points.
Every time we run the algorithm, we use a new random initial guess. In this case, the
algorithm can have some nondeterministic nature, and such algorithms are referred
to as stochastic.√Sometimes, using randomness may be advantageous. For example, in
the example of 4 = ±2 using Eq. (1.1), random initial values (both positive and neg-
ative) can allow the algorithm to find both roots. In fact, a major trend in the modern
metaheuristics is using some randomization to suit different purposes.
For algorithms to be introduced in this book, we are mainly concerned with al-
gorithms for data mining, optimization, and machine learning. We use a relatively
unified approach to link algorithms in data mining and machine learning to algorithms
for optimization.

1.2 Optimization

Optimization is everywhere, from engineering design to business planning. After all,

time and resources are limited, and optimal use of such valuable resources is crucial.
In addition, designs of products have to maximize the performance, sustainability, and
energy efficiency and to minimize the costs. Therefore, optimization is important for
many applications.

1.2.1 A simple example

Let us start with a very simple example to design a container with volume capacity
V0 = 10 m3 . As the main cost is related to the cost of materials, the main aim is to
minimize the total surface area S.
The first thing we have to decide is the shape of the container (cylinder, cubic,
sphere or ellipsoid, or more complex geometry). For simplicity, let us start with a
cylindrical shape with radius r and height h (see Fig. 1.1).
The total surface area of a cylinder is

S = 2(πr 2 ) + 2πrh, (1.11)

and the volume is

V = πr 2 h. (1.12)

There are only two design variables r and h and one objective function S to be min-
imized. Obviously, if there is no capacity constraint, then we can choose not to build
the container, and then the cost of materials is zero for r = 0 and h = 0. However,
Introduction to optimization 5

Figure 1.1 Design of a cylindric container.

the constraint requirement means that we have to build a container with fixed volume
V0 = πr 2 h = 10 m3 . Therefore, this optimization problem can be written as

minimize S = 2πr 2 + 2πrh, (1.13)

subject to the equality constraint

πr 2 h = V0 = 10. (1.14)

To solve this problem, we can first try to use the equality constraint to reduce the
number of design variables by solving h. So we have
V0
h= . (1.15)
πr 2
Substituting it into (1.13), we get

S = 2πr 2 + 2πrh
V0 2V0
= 2πr 2 + 2πr 2 = 2πr 2 + . (1.16)
πr r
This is a univariate function. From basic calculus we know that the minimum or max-
imum can occur at the stationary point, where the first derivative is zero, that is,
dS 2V0
= 4πr − 2 = 0, (1.17)
dr r
which gives

V0 3 V0
r3 = , or r = . (1.18)
2π 2π
Thus, the height is

h V0 /(πr 2 ) V0
= = 3 = 2. (1.19)
r r πr
Random documents with unrelated
content Scribd suggests to you:
describit, dicens: Cum de Pictavis bellum sit et Andegavinis, Inque
die Martis fuit et Sancti Benedicti, Circa forte Caput Wultonnæ
contigit esse, Annus millenus tunc sexagesimus unus.”
That entry comprises all the direct information on the subject. The
Angevin monastic chronicles and Fulk Rechin do not mention it at all.
Neither do the Gesta Cons. in the right place; but they mix it up with
the war between Geoffrey Martel and William the Fat in 1033. By the
light of the Chron. S. Maxent., it seems possible to disentangle the
two stories. It even seems possible to make sense of a passage in
the Gesta which never can be sense as it stands, by understanding it
as referring to Geoffrey the Bearded instead of his uncle: “Willelmus
Pictavensium comes consulatum Sanctonicum suum esse volebat et
vi preoccupatum tenebat, quia patrui sui fuerat. Martellus eumdem
consulatum reclamabat quia avi sui fuerat, cujus heredes absque
liberis mortui erant; et ideo ad heredes sororis avi sui debere reverti
affirmabat” (Gesta Cons., Marchegay, Comtes, p. 126). This is the
story by which the Gesta-writer professes to explain the cause of the
war of Geoffrey Martel and William the Fat, of which he then gives
an elaborate account, ending with William’s capture and the
consequent surrender of Saintes to Geoffrey. But the story is utterly
senseless; the claims of William and Martel as therein stated are
alike devoid of all show of reason. In the account of the war itself,
too, there are strong traces of confusion; Saintes is assumed to have
passed back into the duke’s hands, of which there is no sign
elsewhere; and to crown all, the scene of the battle in which William
is taken is laid, not as by the Chron. S. Maxent. (a. 1032,
Marchegay, Eglises, p. 392) and Fulk Rechin (Comtes, p. 378), at S.
Jouin-de-Marne or Montcontour, but at Chef-Boutonne. The question
then arises: Can this wild tale in the Gesta, which is quite impossible
as an explanation of Martel’s war with William V., be interpreted so
as to explain his successor’s war with William VII.?
“Willelmus [VII., alias Guy-Geoffrey] Pictavensium comes
consulatum Sanctonicum suum esse volebat et vi præoccupatum
tenebat [having presumably seized it on Martel’s death], quia patrui
sui [for patrui read fratris—William the Fat—or patris, William the
Great] fuerat. Martellus [Barbatus] eumdem consulatum reclamabat,
quia avi sui [Fulconis Nerræ] fuerat, cujus hæredes [i.e. G. Martellus]
absque liberis mortui essent; et ideo ad hæredes sororis avi sui
[read avunculi sui—Martel’s sister, the Bearded one’s mother]
debere reverti affirmabat.”
Read in this way, the story is quite reasonable and intelligible, and
the rest of the Gesta’s account might stand almost intact, except the
capture of the duke, which of course is dragged in from the earlier
war. The confusion between the Williams of Aquitaine is easily
accounted for, and so is that between the Geoffreys of Anjou,
especially as all the Geoffreys after Martel occasionally took to
themselves his cognomen.

Note D.
THE DESCENDANTS OF HERBERT WAKE-DOG.

Not the least puzzling matter connected with the Cenomannian

wars is the genealogy of the sovereign house of Maine. The
succession of the counts themselves—Hugh I. (or David), Herbert I.
(Wake-dog), Hugh II., Herbert II.—is plain enough, as also that each
was the son of his predecessor. But the filiation of the women of the
family—Margaret, Gersendis, Paula and Biota—is far from being
equally clear.
1. As to Margaret, there is no real doubt. Orderic does once
(Duchesne, Hist. Norm. Scriptt., p. 683) call her a daughter of
Herbert [II.]; but his own statements in two other places (ib. pp. 487
and 532), as well as Will. Poitiers (ib. p. 190), shew that this is a
mere slip. Margaret was clearly a daughter of Hugh II. and sister of
Herbert II.
2. As to Biota. Orderic (as above, p. 487) calls her “Hugonis
Cenomannensium comitis filiam”; in Will. Poitiers (ib. p. 189) she is
“soror Hugonis”; and Mr. Freeman (Norm. Conq., 3d ed., vol. iii. p.
200, and note T, p. 676) adopts the latter version. Biota, then, was a
daughter of Herbert Wake-dog and sister of Hugh II. But were
Gersendis and Paula her sisters or her nieces?
3. The fullest and most distinct statement of the Cenomannian
pedigree is that of Orderic in Duchesne, Hist. Norm. Scriptt., p. 532:
“Hugo filius Herberti . . . Bertam . . . in conjugium accepit; quæ
filium nomine Herbertum et tres filias ei peperit. Una earum data est
Azsoni Marchiso Liguriæ. Alia nomine Margarita Rodberto filio
Guillelmi Ducis Neustriæ desponsata est . . . Tertia vero Joanni
domino castri quod Flecchia dicitur nupsit.”
With regard to this last marriage, it is to be observed that in the
speech which Orderic puts into the mouth of Elias of La Flèche,
addressing Hugh of Este (ib. p. 684), he says nothing about his
mother at all, but makes him trace his descent from Herbert Wake-
dog through his grandmother, whom he calls Herbert’s daughter:
“Filia Herberti comitis Lancelino de Balgenceio nupsit, eique . . .
Joannem meum genitorem peperit.” The name of John’s wife, Paula,
comes from another passage of Orderic (ib. p. 768); but he there
says nothing about her parentage, merely calling her son Elias
“Hugonis Cenomannorum consulis consobrinus.” The houses of Le
Mans and La Flèche cannot have intermarried twice in two
succeeding generations; one of Orderic’s statements must be wrong;
but which, I cannot decide.
The last point is the parentage of Gersendis, the wife of Azzo of
Este; and as the whole tone of Elias’s speech (as above) implies that
he and her son were related to the counts of Le Mans in the same
degree, the solution of this question might almost be held to decide
the previous one also. This seems to be Mr. Freeman’s opinion, and
he regards Orderic’s statement quoted above as conclusive that
Gersendis and Paula were both daughters of Hugh II., and sisters
therefore of Margaret and Herbert II., in spite of the biographer of the
bishops of Le Mans (Mabillon, Vet. Anal., p. 308), who expressly
says that Gersendis was a daughter of Herbert Wake-dog, and the
continuator of Will. Jumièges, who says:—“Cenomannenses . . .
consilium ineunt cum Heliâ filio Joannis de Flecâ . . . ut filiam
cujusdam comitis Langobardiæ, neptem videlicet Hereberti quondam
Cenomannensis comitis ex primogenitâ filiâ, in matrimonium ducat.”
Will. Jumièges, l. viii. c. 5 (Duchesne, Hist. Norm. Scriptt., p. 294).
This re-appears in R. Diceto (Stubbs, vol. i. pp. 183, 184;
Marchegay, Comtes, p. 334) in the following form:—“Helias, filius
Johannis de Flecâ, Sibillam, filiam cujusdam comitis Longobardiæ,
neptem scilicet Hereberti quondam Cenomannorum comitis, duxit
uxorem, et cum eâ comitatum Cenomanniæ suscepit.” But this is
certainly wrong; for the first wife of Elias was Matilda of Château-du-
Loir, and the second was Agnes of Perche.
What Elias could have had to gain by the marriage thus proposed
for him it is impossible to guess, as he himself certainly was quite as
nearly related to the counts of Maine as this oddly-described bride
could have been. Mr. Freeman (Norm. Conq., 3d ed., vol. iii., note T,
p. 676), takes the description as favouring Orderic’s theory, and
remarks: “The words could only have been written by one who
looked on Gersendis as a sister of Herbert.” “Neptem Hereberti,”
then, he interprets, “niece of Herbert [II].” But is it not a much simpler
interpretation of the whole phrase—“neptem Hereberti ex
primogenitâ filiâ”—to read it “granddaughter of Herbert [I.] through
his eldest daughter”? In that case, we should have another witness
on the side of the bishops’ biographer.
There is another curious bit of evidence which at first glance
seems also to tell in his favour. I do not think that it really proves
anything about the matter; but it is worth examining for other
reasons. M. d’Arbois de Jubainville (Comtes de Champagne, vol. i.
p. 392, note 5), declares it proved on documentary evidence that
Stephen-Henry of Blois, the father of our King Stephen, was the son
of Theobald III. by his first marriage with Gersendis of Maine. About
the marriage itself there is no doubt, nor about the divorce which
followed it; and the latter had taken place in 1049 at latest, for
Theobald was excommunicated for that very cause by the Council of
Reims. Most historians seem however to have supposed that
Gersendis was then a mere child, and that the mother of Stephen, as
well as of Theobald’s other children, was his second wife, Adela of
Valois. M. de Jubainville, in support of his opinion, refers especially
to two charters. One is in Gallia Christiana, vol. viii., instr. col. 548. It
has no date, and says nothing about Stephen’s mother or his
stepmother; I therefore cannot see its bearing on the question. The
other is in Bernier, Histoire de Blois, preuves, pp. xiii–xiv. In it
Stephen-Henry, in the year 1089, grants certain lands to Pontlevoy
“pro animæ meæ et uxoris et Theobaldi patris mei et matris meæ
Gandree . . . remedio”; and has the grant confirmed “nomine . . .
Alæ uxoris meæ, Alæ uxoris Thebaudi comitis,” etc. This certainly
seems to shew that Adela was not his mother, though it does not
necessarily follow that “Gandree” represents Gersendis. If it does,
Stephen-Henry must have been born in 1049 at latest, and therefore
Gersendis cannot possibly have been a daughter of Hugh II., who
was not married till 1040 at the very earliest.
The greatest puzzle in the whole matter, however, is this: If
Stephen-Henry was really the eldest son of Gersendis of Maine, how
does it happen that neither in 1073, nor in 1089, nor in any of the
Cenomannian revolutions and wars, do we hear a single word about
his claims upon the county? M. d’Arbois de Jubainville’s suggestion
in fact opens a question much more important and much more
obscure than that of the age and parentage of Gersendis. He
certainly seems to have proved that Adela of Valois was not
Stephen’s mother; but has he proved that Gersendis was? The only
bit of evidence, direct or indirect, which it seems possible to bring to
bear upon this matter is a passage in the Historia Pontificalis (Pertz,
Mon. Germ. Hist., vol. xx. p. 531) where it is said that the cause of
our King Stephen was upheld by some of the Roman cardinals who
claimed kindred with him “eo quod avia ejus Lumbarda fuerit.” Now,
as the second husband of Gersendis was a Lombard, this may come
from some confused idea about her. But it also suggests another
possible solution of the whole question about Stephen-Henry’s
mother. Theobald and Gersendis were divorced in 1049 at latest; the
first record in which Adela appears as Theobald’s wife is dated 1061
(Jubainville, Comtes de Champagne, vol. i. p. 393, note 3). May not
the mysterious “Gandrea” of the charter of 1089 have been an Italian
lady who was married to Theobald, became the mother of his heir,
and died, between those two dates?
 Note E.
THE SIEGE OF LA FLÈCHE AND TREATY OF BLANCHELANDE.

There are two questionable points connected with these matters:

1. the date; 2. the geography.
1. The only original writer who gives a detailed account of both
siege and treaty is Orderic, who carries his story straight on from the
quelling of the revolt of Maine in 1073 to the siege of La Flèche, as if
it had all happened in the same year, before William returned to
England with his troops. On the other hand, none of the Angevin
writers mention La Flèche under date 1073; but the Chronn. S. Albin.
and S. Flor. Salm. (Marchegay, Eglises, pp. 26, 189) have “Exercitus
de Fissâ,” the former in 1077, the latter in 1078; and in the Art de
vérifier les Dates these entries are interpreted as referring to the
siege which was followed by the treaty of Blanchelande. M. Voisin
(Les Cénomans, p. 414) dates the whole affair 1085; he gives no
reason and seems to be quite unsupported. The choice lies therefore
between Orderic’s date and that of the Angevin chronicles. Mr.
Freeman (Norm. Conq., vol. iv. pp. 560–563) follows Orderic, and I
have done the same.
2. As to the geography. Orderic (Duchesne, Hist. Norm. Scriptt., p.
533) says that to meet William the Angevin and Breton host, leaving
La Flèche, “Ligerim fluvium audacter pertransierunt.” Now this must
be wrong, as the Loire is a long way south of La Flèche. It is clear
that for Ligerim, “Loire,” we must read Liderim, “Loir,” as Mr.
Freeman says (Norm. Conq., vol. iv. p. 562, note 2). Even crossing
the Loir seems rather a strange proceeding; for La Flèche being on
the right or north bank of that river, they must have crossed it to the
southward—i.e. away from Normandy. How came it that William,
marching against them out of Normandy, had gone so far down to
the south of them?
There is however a further question as to the actual place of the
treaty, which Mr. Freeman (as above, p. 562) places at Bruère in the
Passais. If such was the case, Orderic’s story of the crossing of the
river becomes quite hopeless, as Bruère is a long way north-west of
La Flèche. But there is another version. J. Pesche in his Dictionnaire
historique de la Sarthe, vol. i. p. 168, under “Blanchelande ou
Blanche-bruyère,” says: “Vaste espace de terrain infertile, où croît
abondamment le lichen des rennes, dont la blancheur lui aura fait
donner son nom; situé entre La Flèche et Le Lude, côtoyé par la
route qui conduit de l’une à l’autre de ces deux villes.” It is this which
Pesche and, following him, M. Voisin (Les Cénomans, p. 414, note
1) mark as the scene of the treaty. So does M. Prévost in a note to
Orderic, vol. ii. p. 258, and he adds that a farm there still in 1840
bore the name of Blanchelande. If this theory is correct, Orderic’s
geography is quite right and clear; the besiegers of La Flèche, on the
north side of the Loir, crossing over to its southern bank, would
march straight upon the “white moor.” William must then have
crossed higher up and made a circuit to the south-east of them. The
only question remaining would be, what was his reason for this
movement? To which there was doubtless a good military answer.
With regard to the second siege of La Flèche by Fulk Rechin, in
1081, there is a very strange story in the Chron. Rain. Andeg.
(Marchegay, Eglises, p. 13). We are there told that Fulk not only took
and burned the castle (as the Chron. S. Albin., ib. p. 26, also states
under the same year) in revenge, for John’s rebellion against him,
but also punished King William for his previous relief of the castle, by
so worsting him in battle that he retreated after giving hostages for
peace, among whom were his brother the count of Mortain and his
own son! Mr. Freeman says nothing of this very apocryphal-looking
story. Is it anything more than an Angevin travesty of Robert’s
homage to Fulk at Blanchelande?

Note F.
THE MARRIAGE OF GEOFFREY AND MATILDA.
 The date of this marriage is commonly given as 1127. A
comparison of evidence seems however to lead to the conclusion
that its true date is 1128.
1. The Angevin chronicles never mention the marriage at all. The
Gesta Cons., Will. Jumièges and several other writers mention it
without any kind of date. The English Chronicle, Sim. Durh., Will.
Malm. and Hen. Hunt. give no distinct date, but imply that the
proposal was immediately followed by the wedding. They speak as if
Robert and Brian had taken Matilda over sea and married her to
Geoffrey without more ado.
2. Orderic mentions the marriage in two places. In the first
(Duchesne, Hist. Norm. Scriptt., p. 763) he gives no clue to the date;
in the second (ib. p. 889) he dates it 1129.
3. The Chron. Fiscannense (Rer. Gall. Scriptt., vol. xii. p. 778)
dates it 1127.
4. A charter of agreement between the bishop of Séez and the
convent of Marmoutier (printed in Gilles Bry’s Hist. de Perche, p.
106) has “signum Henrici Regis quando dedit filiam suam Gaufredo
comiti Andegavensi juniori.” It is dated “anno ab Inc. Dom. 1127,
Indictione VI.”
5. The last witness is John of Marmoutier, the author of the
Historia Gaufredi Ducis. From him we might have expected a distinct
and authentic statement; but he does not mention the year at all. He
says that Geoffrey was knighted on Whit-Sunday and married on its
octave, and that he was then fifteen years of age (Hist. Gaufr. Ducis,
Marchegay, Comtes, pp. 236, 233). Afterwards, in speaking of the
birth of Henry Fitz-Empress, he says that it took place in the fourth
year of his parents’ marriage (ib. pp. 277, 278). Henry was born on
Mid-Lent Sunday, March 5, 1133; if therefore the writer reckoned
backwards from the Whitsuntide of that year, his words ought to
mean that the marriage was in 1129. But as he goes on to state that
Matilda’s third son was born in the sixth year of her marriage, and
that Henry I. died “anno eodem, ab Incarnatione videlicet Domini
1137,” it is impossible to say what he did mean. Whether he is
collecting the traditions of the ancient counts or writing the life of his
own contemporary sovereign, John’s chronology is pursued by the
same fate; whenever he mentions a date by the year, he is almost
certain to make it wrong. But that he should have done the like in his
reckoning of days, or even of his hero’s age, by no means follows.
To consider the latter point first: Geoffrey the Handsome was born on
August 24, 1113 (Chron. S. Albin. ad ann., Marchegay, Eglises, p.
32). Therefore, if John meant that he was past fifteen at his
marriage, it must have been in 1129. But if he only meant “in his
fifteenth year,” it would be 1128. In that year the octave of Pentecost
fell on June 17; Geoffrey then lacked but two months to the
completion of his fifteenth year; and considering Matilda’s age, it is
no wonder that the panegyrist tried to make her husband out as old
as possible. It is in fact plain that such was his intention, for though
he places Geoffrey’s death in the right year, 1151, he gives his age
as forty-one instead of thirty-eight (Hist. Gaufr. Ducis, Marchegay,
Comtes, p. 292).
The most important matter, however, is John’s statement that the
wedding took place on the octave of Pentecost. The date in this case
is not one casually slipped in by the writer in passing; it comes in a
detailed account of the festivities at Rouen on the occasion of
Geoffrey’s knighting, which is expressly said to have occurred at
Pentecost, and to have been followed by his marriage on the octave.
Now this leaves us on the horns of a dilemma fatal alike to the date
in the Chron. Fiscann., 1127, and to that of Orderic, 1129. For, on the
one hand, Will. Malm. (Hist. Nov., l. i. c. 3, Hardy, p. 692) says that
Matilda did not go to Normandy till after Whitsuntide [1127]; and Hen.
Hunt., l. vii. c. 37 (Arnold, p. 247), adds that the king followed her in
August (Sim. Durh., ed. Arnold, vol. ii. pp. 281, 282, really witnesses
to the same effect; for his chronology of the whole story is a year in
advance). Consequently, as Mrs. Everett Green remarks, “the union
could not have taken place before the spring of the following year,
1128” (Princesses of England, vol. i. pp. 107, 108). On the other
hand, it is plain that Fulk was present at his son’s wedding; but
before Whitsuntide 1129 Fulk was himself married to the princess of
Jerusalem (Will. Tyr., l. xiii. c. 24).
 From all this it results: 1. If Geoffrey and Matilda were married in
1127, it cannot have been earlier than September, i.e. at least three
months after Whitsuntide. 2. If they were married in 1129, it must
have been quite at the beginning of the year, and Orderic must, on
this occasion at least, have made his year begin in English fashion,
at Christmas. 3. If they were married at Whitsuntide, it can only have
been in 1128.
We have in short to choose one out of three authorities: the
Chronicle of Fécamp, Orderic and John of Marmoutier—for the Séez
charter, as Mrs. Everett Green remarks (Princesses, vol. i. p. 108),
proves nothing more than that the betrothal had taken place in 1127.
Of these three, the first is certainly of least account. Orderic, on the
other hand, is on most other subjects a far better authority than
John. But his chronology is very little better than John’s, at any rate
towards the close of his work; his whole account of Henry’s later
years is sketchy and confused; while John is Geoffrey Plantagenet’s
own special biographer, writing within sixty years of the event, from
materials furnished by personal followers of his hero. I cannot but
regard him as our primary authority on this subject, and believe on
his testimony that the real wedding-day of Geoffrey and Matilda was
the octave of Pentecost, June 17, 1128.
 CHAPTER V.
GEOFFREY PLANTAGENET AND STEPHEN OF

BLOIS.

1128–1139.

All the mental and bodily gifts wherewith nature had endowed the
most favoured members of the Angevin house seemed to have been
showered upon the eldest son of Fulk V. and Aremburg of Maine.
The surname by which he is most generally known, and which an
inveterate usage has attached to his descendants as well as to
himself, is in its origin and meaning curiously unlike most historical
surnames; it seems to have been derived simply from his boyish
habit of adorning his cap with a sprig of “planta-genista,” the broom
which in early summer makes the open country of Anjou and Maine
a blaze of living gold. With a fair and ruddy countenance, lit up by the
lightning-glance of a pair of brilliant eyes; a tall, slender, sinewy
frame, made for grace no less than for strength and activity:—[628] in
the unanimous opinion of his contemporaries, he was emphatically
“Geoffrey the Handsome.” To this prepossessing appearance were
added the charms of a gracious manner and a ready, pleasant
speech;[629] and beneath this winning exterior there lay a
considerable share of the quick wits of his race, sharpened and
developed by such a careful education as was given to very few
princes of the time. The intellectual soil was worthy of the pains
bestowed upon it, and brought forth a harvest of, perhaps, somewhat
too precocious scholarship and sagacity. Geoffrey’s fondness for the
study of the past seems to have been an inheritance from Fulk
Rechin; the historian-count might have been proud of a grandson
who carried in his memory all the battles fought, all the great deeds
done, not only by his own people but also in foreign lands.[630] Even
Fulk the Good might have approved a descendant who when still a
mere boy could shine in serious conversation with such a “lettered
king” as Henry I.;[631] and Fulk the Black might not have been
ashamed of one who in early youth felt the “demon-blood” within him
too hot to rest content in luxury and idleness, avoided the corrupting
influences of mere revelry, gave himself up to the active exercises of
military life,[632] and, while so devoted to letters that he would not
even go to war without a learned teacher by his side,[633] turned his
book-learning to account in ways at which ruder warriors and more
unworldly scholars were evidently somewhat astonished.[634] Like
his ancestor the Black Count, Geoffrey was one of those men about
whom their intimate associates have a fund of anecdotes to tell. The
“History” of his life put together from their information, a few years
after his death, is chiefly made up of these stories; and through the
mass of trite moralizing and pedantic verbiage in which the compiler
has imbedded them there still peeps out unmistakeably the peculiar
temper of his hero. Geoffrey’s readiness to forgive those who threw
themselves upon his mercy is a favourite theme of his biographer’s
praise; but the instances given of this clemency indicate more of the
vanity and display of chivalry in its narrower sense than of real
tenderness of heart or generosity of soul. Such is the story of a
discontented knight whose ill-will against his sovereign took the
grotesque form of a wish that he had the neck of “that red-head
Geoffrey” fast between the two hot iron plates used for making a
wafer-cake called oublie. It chanced that the man whose making of
oublies—then, as now, a separate trade—had suggested the wish of
this knight at St.-Aignan shortly afterwards made some for the eating
and in the presence of Count Geoffrey himself, to whom he related
what he had heard. The knight and his comrades were presently
caught harrying the count’s lands; and the biographer is lost in
admiration at Geoffrey’s generosity in forgiving not only their
depredations, but the more heinous crime of having, in a fit of ill-
temper after dinner, expressed a desire to make a wafer of him.[635]
On another occasion we find the count’s wrath averted by the
charms of music and verse, enhanced no doubt by the further charm
of a little flattery. Four Poitevin knights who had been taken captive
in one of the skirmishes so common on the Aquitanian border won
their release by the truly southern expedient of singing in Geoffrey’s
hearing a rime which they had composed in his praise.[636] A touch
of truer poetry comes out in another story. Geoffrey, with a great train
of attendants and noble guests, was once keeping Christmas at Le
Mans. From his private chapel, where he had been attending the
nocturnal services of the vigil, he set out at daybreak at the head of a
procession to celebrate in the cathedral church the holy mysteries of
the festival. At the cathedral door he met a poorly-dressed young
clerk, whom he flippantly saluted: “Any news, sir clerkling?”—“Ay, my
lord, the best of good news!”—“What?” cried Geoffrey, all his
curiosity aroused—“tell me quick!”—“‘Unto us a Child is born, unto
us a Son is given!’” Abashed, Geoffrey asked the youth his name,
bade him join the other clergy in the choir, and as soon as mass was
over went straight to the bishop: “For the love of Him Who was born
this day, give me a prebend in your church.” It was no sooner
granted than taking his new acquaintance by the hand, he begged
leave to make him his substitute, and added the further gift of a stall
in his own chapel, as a token of gratitude to the poor clerk whose
answer to his thoughtless question had brought home to him,
perhaps more deeply than he had ever felt them before, the glad
tidings of Christmas morning.[637] From another of these anecdotes
Geoffrey seems, as far as we can make out, to have been the
original hero of an adventure which has since, in slightly varying
forms, been attributed to several other princes, from Charles the
Great down to James the Fifth of Scotland, and which indeed may
easily have happened more than once. Led away by his ardour in
pursuit of the chase—next to literature, his favourite recreation—the
count one day outstripped all his followers, and lost his way alone in
the forest of Loches. At last he fell in with a charcoal-burner, who
undertook to conduct him back to the castle. Geoffrey mounted his
guide behind him; and as they rode along, the peasant, ignorant of
his companion’s rank, and taking him for a simple knight, let himself
be drawn into conversation on sundry matters, including a free
criticism on the government of the reigning count, and the
oppressions suffered by the people at the hands of his household
officers. When they reached the gates of Loches, the burst of joy
which greeted the wanderer’s return revealed to the poor man that
he had been talking to the count himself. Overwhelmed with dismay,
he tried to slip off the horse’s back; but Geoffrey held him fast, gave
him the place of honour at the evening banquet, sent him home next
day with a grant of freedom and a liberal gift of money, and profited
by the information acquired from him to institute a thorough reform in
the administration of his own household.[638]

[628] Hist. Gaufr. Ducis (Marchegay, Comtes), p. 233.

[629] Ib. pp. 232, 233.

[630] Hist. Gaufr. Ducis (Marchegay, Comtes), p. 232.

[631] Ib. p. 235.

[632] Ib. p. 233.

[633] Ib. p. 276.

[634] See the story of the siege of Montreuil-Bellay, Hist. Gaufr.

Ducis (Marchegay, Comtes), p. 286.

[635] Hist. Gaufr. Ducis (Marchegay, Comtes), pp. 257–260.

[636] Ib. pp. 253–256.

[637] Ib. pp. 274–276.

[638] Hist. Gaufr. Ducis (Marchegay, Comtes), pp. 240–250.

Such stories as these, while they help us to form some picture of

the manner of man that Geoffrey was, set him before us in the
romantic light in which he appears to the best advantage. When one
turns from them to a survey of his life as a whole, one is struck with a
sense of something wanting in him. The deficiency was in truth a
very serious one; it was a lack of steady principle and of genuine
feeling. The imaginative and impulsive vein which ran through all the
more refined characters of his race lay in him very near the surface,
but it did not go very deep. His imagination was sensitive, but his
heart was cold; his impulses sprang from the play of a quick fancy,
not from the passion of an ardent soul. One more story may furnish a
slight, but significant, illustration of his temper. For some wrong done
to the see of Tours Geoffrey was once threatened by the archbishop
with excommunication. Either the earlier or the later Fulk of
Jerusalem would have almost certainly begun by a reckless defiance
of the threat, and the later one, at least, would almost as surely have
ended by hearty penance. Geoffrey began and ended with a jest:
“Your threats are vain, most reverend father; you know that the
archbishop of Tours has no jurisdiction over the patrimony of S.
Martin, and that I am one of his canons!”[639] In all the sterling
qualities of a ruler and a man, the hasty, restless, downright Fulk V.
was as superior to his clever charming son as Fulk the Black was
superior to Geoffrey Martel. But it is only fair to bear in mind that
Geoffrey Plantagenet’s life was to a great extent spoilt by his
marriage. The yoke which bound together a lad of fifteen and a
woman of twenty-five—especially such a woman as the Empress
Matilda—could not fail to press heavily on both parties; but the one
most seriously injured by it was probably the young husband. Even
in a political point of view, to him personally his marriage was more
of a hindrance than an advantage; it cut him off from all chance of
striking out an independent career. The man himself was in fact
sacrificed to his posterity. Chained down while his character was yet
undeveloped to the irksome position of a mere appendage to King
Henry’s heiress;—plunged suddenly, and for life, into a sphere of
interests and duties alien from his own natural temper and
inclinations:—weak, selfish, unprincipled as Geoffrey too plainly
shewed himself to be, still it was well not only for him but for others
that he had enough of the dogged Angevin thoroughness to carry
him safely and successfully, if not always gloriously, through his
somewhat dreary task till he could make it over to the freer, as well
as stronger, hands of his son.

[639] Hist. Gaufr. Ducis (Marchegay, Comtes), p. 252.

 The hope which inspired both the king of England and the count of
Anjou when they planned their children’s marriage can only have
been the hope of a grandson in whom the blood of both would be
united, who would gather into his own person all conflicting claims,
and in whom all feuds would have an end. On this depended all King
Henry’s schemes for the future; on this were concentrated all his
desires, on this were founded all his plans and arrangements during
the last seven years of his reign. In the internal history of England
those years are an almost complete blank; they are in fact simply
seven more years of the administration of Bishop Roger of Salisbury,
for Henry himself spent almost the whole of them upon the continent.
His work was finished, and all that remained to do was to maintain
the order of things which he had established so as to hand it on in
full working to his successor. He must, however, have begun to
doubt the success of his schemes when Geoffrey and Matilda
separated little more than twelve months after their marriage. At first,
everything had seemed to be turning in favour of Henry’s
arrangements. Six weeks after the wedding, the death of William the
Clito, wounded in a skirmish with a rival claimant of the county of
Flanders,[640] removed the only competitor whom the king could
deem likely to stand in the way of his plans for the descent of the
crown. In the spring Fulk’s departure for Holy Land left the young
couple sole masters at Angers. All things looked tranquil and secure
when Henry returned to England in July 1129. He had, however,
been there only a few days when he learned, to his great indignation,
that his daughter had been sent away with scorn by her husband,
and had betaken herself with a few attendants to Rouen.[641] There
she remained for nearly two years, while Geoffrey was busy with a
general revolt among his barons. East and west and south and north
had all risen at once; the list of rebels includes the chief landowners
in all parts of the Angevin dominions, from the old eastern outpost
Amboise to Laval on the Breton border, and from Sablé on the
confines of Anjou and Maine to Montreuil-Bellay, Thouars and
Mirebeau in the Aquitanian territory of Loudun, and the yet more
remote fief of Parthenay in Poitou.[642] It seems as if the disaffected
barons, worsted in their struggle with Fulk, had only been waiting till
he was out of the country, and now, when Geoffrey by his quarrel
with his wife had deprived himself of all chance of help from his
father-in-law, they closed in upon the boy-count with one consent,
thinking to get him into their power and wring from him any
concessions they pleased. They unintentionally did him an immense
service, for by thus suddenly throwing him upon his own resources
they made a man of him at once. No one knew better than Geoffrey
Plantagenet that he was not the first count of Anjou who had been
left to shift for himself in difficult circumstances at the age of fifteen;
and he faced the danger with a promptitude and energy not
unworthy of Fulk Nerra’s representative. One after another he
besieged the rebel leaders in their strongholds; one after another
was forced, tricked or frightened into submission. Once, while
besieging Theobald of Blazon in the great fortress of Mirebeau,
Geoffrey was blockaded in his turn by the count of Poitou, whom the
traitors had called to their aid; even from this peril, however, his
quick wit and youthful energy extricated him in triumph; and the
revolt was finally crushed by a severe punishment inflicted on its
most powerful leader, Lisiard of Sablé. Geoffrey ravaged the whole
of Lisiard’s estates, razed his castle of Briolet, seized that of Suze
and kept it in his own hands for the rest of its owner’s life; while to
guard against further dangers from the same quarter, by the advice
of his faithful barons he reared, for the express purpose of defence
against incursions from Sablé, a fortress to which he gave the name
of Châteauneuf, on the left bank of the Sarthe, just below the bridge
made famous by the death of Count Robert the Brave.[643]

[640] Ord. Vit. (Duchesne, Hist. Norm. Scriptt.), pp. 886, 887.

[641] Sim. Durh. Gesta Reg. a. 1129.

[642] Hist. Gaufr. Ducis (Marchegay, Comtes), p. 263.

[643] For the barons’ revolt, see Hist. Gaufr. Ducis

(Marchegay, Comtes), pp. 263–268. The strange and not very
clear story of the double siege of Mirebeau is in pp. 265, 266.
“Exercitus de Mirebello” is recorded in Chronn. S. Albin. and S.
Flor. Salm. a. 1130 (Marchegay, Eglises, pp. 33, 191). The
Chron. S. Albin. also records the building of Châteauneuf, a.
 1131; the Hist. Gaufr. Ducis, p. 270, connects it with the revolt of
a lord of Sablé, but apparently with the later revolt of Lisiard’s
son Robert—which, however, the date in the chronicle shows to
be a mistake.

King Henry had joined his daughter in Normandy in the summer of

1130; in July of the next year they returned to England together.
They were soon followed by a message from Geoffrey, who was now
becoming awake to his rights and duties as husband of King Henry’s
heiress, and having made himself thoroughly master in his own
dominions felt it time to demand the return of his wife. A great
council held at Northampton on September 8 decided that his
request should be granted;[644] and the assembled prelates and
barons repeated their homage to Matilda as her father’s destined
successor.[645] She then went back to her husband, by whom she
was, if not warmly welcomed, at least received with all due courtesy
and honour.[646] Fortunately for the ill-matched couple, they were
both of that cold-blooded temperament to which intense personal
affection is not a necessary of life. Henceforth they were content to
work together as partners in political enterprise, and to find in
community of worldly interests a sufficient bond of union. On Mid-
Lent Sunday—March 5, 1133—the bond was made indissoluble by
the birth of their son and heir. Most fittingly, the child to whom so
many diverse nationalities looked as to their future sovereign[647]
was born not in the actual home of either of his parents, but in that
city of Le Mans which lay midway between Normandy and Anjou,
which had so long been the ground of their strife, and had at last
been made the scene of their union.[648] He was baptized in the
cathedral church by the bishop of the diocese on Easter Eve,
receiving the name of his grandfather Henry, and was then, by his
mother’s special desire, solemnly placed under the protection of the
local patron saint on the same altar where his father had been
dedicated in like manner thirteen years before.[649]

[644] Hen. Hunt., l. vii. c. 41 (Arnold, p. 252).

 [645] Will. Malm. Hist. Nov., l. i. c. 6 (Hardy, p. 698).

[646] Hen. Hunt. as above.

[647] “Quem multi populi dominum expectant.” Ord. Vit.

(Duchesne, Hist. Norm. Scriptt.), p. 763.

[648] Acta Pontif. Cenoman., c. 36 (Mabillon, Vet. Anal., p.

322). Cf. Chronn. S. Albin. and S. Serg. a. 1133 (Marchegay,
Eglises, pp. 33, 144, 145), Chron. S. Flor. Salm. a. 1133 (ib. p.
191, giving a wrong day), Hist. Gaufr. Ducis (Marchegay,
Comtes), pp. 277, 278, also wrongly dated.

[649] Acta Pontif. Cenoman. as above.

To King Henry the birth of his grandson was the crowning of all his
hopes. The greatest difficulty which had hitherto stood in the way of
his scheme for the descent of the crown—the objection which was
sure to be made against Matilda on account of her sex—would lose
more than half its force now that she could be regarded as regent for
her infant son; and Henry at once summoned another great council
at which he again made the archbishops, bishops, earls and barons
of his realm swear fealty to the Empress “and also to her little son
whom he appointed to be king after him.”[650] All things seemed as
safe as human foresight could make them when in the beginning of
August he crossed over to Normandy.[651] Signs and wonders in
earth and sky, related afterwards as tokens of coming evil,
accompanied his voyage;[652] but nearly two years passed away
before the portents were fulfilled. In the spring Matilda joined her
father at Rouen, and there, shortly before Whitsuntide, her second
son was born.[653] The old king’s pleasure in his two little
grandchildren was great enough to keep him lingering on in
Normandy with them and their mother, leaving England to the care of
Bishop Roger, till the middle of the following year,[654] when there
came tidings of disturbance on the Welsh border which made him
feel it was time he should return.[655] His daughter however set
herself against his departure. Her policy is not very clear; but it
seems impossible to acquit her of playing a double game and
secretly instigating her husband to attack her father while the latter
was living with her in unsuspecting intimacy and confidence.
Geoffrey now suddenly put forth a claim to certain castles in
Normandy which he asserted had been promised to him at his
marriage.[656] Henry denied the claim; the Angevin temper burst
forth at once; Geoffrey attacked and burned the castle of Beaumont,
whose lord was like himself a son-in-law of Henry, and altogether
behaved with such insulting violence that the king in his wrath was
on the point of taking Matilda, who was with him at Rouen all the
while, back with him to England. But he now found it impossible to
leave Normandy. The land was full of treason; many barons who
only disguised their real feelings from awe of the stern old king had
been gained over in secret to the Angevin cause; among those
whose fidelity was most suspected were Roger of Toëny and William
Talvas the lord of Alençon, who had been restored to the forfeited
estates of his family at the intercession of Geoffrey’s father in 1119.
Roger’s castle of Conches was garrisoned by the king; William
Talvas was summoned to Rouen more than once, but the conscious
traitor dared not shew his face; at last Henry again seized his
estates, and then, in September, Talvas fled across the border to be
received with open arms by the count of Anjou.[657] The countess
pleaded warmly with her father for the traitor’s pardon, but in vain.
When she found him inexorable, she suddenly threw off the mask
and shewed on which side her real sympathies lay by parting from
the king in anger and going home to her husband at Angers.[658]
Father and daughter never met again. In the last week of November
Henry fell sick while hunting in the Forest of Lions; feeling his end
near, he sent for his old friend Archbishop Hugh of Rouen to receive
his confession and give him the last sacraments. His son Earl Robert
of Gloucester hurried to the spot at the first tidings of his illness; his
daughter made no sign of a wish for reconciliation; yet when the earl
and the primate asked for his final instructions concerning the
succession to the crown, he remained true to his cherished purpose
and once more bequeathed all his dominions on both sides of the
sea to Matilda and her heirs for ever.[659] He died on the night of
December 1, 1135.[660]