Buy Ebook Mathematical Problems in Meteorological Modelling 1st Edition András Bátkai Cheap Price
Buy Ebook Mathematical Problems in Meteorological Modelling 1st Edition András Bátkai Cheap Price
Buy Ebook Mathematical Problems in Meteorological Modelling 1st Edition András Bátkai Cheap Price
com
Mathematical Problems in
Meteorological Modelling 1st
https://textbookfull.com/product/mathematical
-problems-in-meteorological-modelling-1st-
edition-andras-batkai/
textbookfull
More products digital (pdf, epub, mobi) instant
download maybe you interests ...
https://textbookfull.com/product/mathematical-modelling-in-real-
life-problems-case-studies-from-ecmi-modelling-weeks-ewald-
lindner/
https://textbookfull.com/product/mathematical-modelling-of-waves-
in-multi-scale-structured-media-1st-edition-alexander-b-movchan/
https://textbookfull.com/product/mathematical-modelling-in-solid-
mechanics-1st-edition-francesco-dellisola/
https://textbookfull.com/product/mathematical-modelling-1st-
edition-seppo-pohjolainen-eds/
Mathematical Methods and Modelling in Applied Sciences
1st Edition Mehmet Zeki Sarikaya (Editor)
https://textbookfull.com/product/mathematical-methods-and-
modelling-in-applied-sciences-1st-edition-mehmet-zeki-sarikaya-
editor/
https://textbookfull.com/product/modelling-in-mathematical-
programming-methodology-and-techniques-jose-manuel-garcia-
sanchez/
https://textbookfull.com/product/language-and-culture-in-
mathematical-cognition-volume-4-mathematical-cognition-and-
learning-1st-edition-daniel-b-berch/
https://textbookfull.com/product/exercises-and-problems-in-
mathematical-methods-of-physics-giampaolo-cicogna/
https://textbookfull.com/product/methods-of-mathematical-
modelling-fractional-differential-equations-1st-edition-harendra-
singh/
Mathematics in Industry 24
The European Consortium for Mathematics in Industry
András Bátkai
Petra Csomós
István Faragó
András Horányi
Gabriella Szépszó Editors
Mathematical
Problems in
Meteorological
Modelling
MATHEMATICS IN INDUSTRY 24
Editors
Hans Georg Bock
Frank de Hoog
Avner Friedman
Arvind Gupta
André Nachbin
Tohru Ozawa
William R. Pulleyblank
Torgeir Rusten
Fadil Santosa
Jin Keun Seo
Anna-Karin Tornberg
SUBSERIES
Managing Editor
Michael Günther
Editors
Luis L. Bonilla
Otmar Scherzer
Wil Schilders
More information about this series at http://www.springer.com/series/4650
András Bátkai • Petra Csomós • István Faragó •
András Horányi • Gabriella Szépszó
Editors
Mathematical Problems
in Meteorological Modelling
123
Editors
András Bátkai Petra Csomós
Institute of Mathematics Institute of Mathematics
ERotvRos Lorand University ERotvRos Loránd University
Budapest, Hungary Budapest, Hungary
MTA-ELTE Numerical Analysis MTA-ELTE Numerical Analysis
and Large Networks Research Group and Large Networks Research Group
Budapest, Hungary Budapest, Hungary
Gabriella Szépszó
Hungarian Meteorological Service
Budapest, Hungary
When I first began studying pure mathematics, I could never have foreseen that I
would spend the better part of my professional life in the area of meteorology. At
that time, designing a proof for some conjecture seemed much more appealing than
dealing with complex, high-dimensional weather forecasting models. Nevertheless,
the presence of rich dynamics for already low-dimensional dynamical systems
eventually piqued my curiosity for what the atmosphere, which could be considered
the ultimate chaotic dynamical system, might hold within it.
In the early 1990s, people started to give some thought on how to deal with
chaotic behaviour in the context of day-to-day weather forecasting. The opportunity
to contribute to an operational approach that would probe the predictability of the
atmosphere has been an exciting one. The Ensemble Prediction Systems, which at
the time were being developed at a few places worldwide, were all designed to
inform the forecasters about the occurrence of different possible weather regimes. It
was already a mathematical challenge to optimally select the model runs comprising
the ensemble. Given the computer capacity, it was only possible to conduct a
relatively small number of model runs, compared to the model phase dimension.
During the last two decades, predictability research has produced an important
shift from a deterministic to a probabilistic view in the field of weather and
climate forecasting. A closer connection between ensemble forecasting and data
assimilation naturally follows from this transition: the flow-dependent information
that ensembles provide regarding uncertainties for certain areas can be employed to
improve the use of observations in data assimilation, which in turn leads to better
performing ensembles. This hybrid data assimilation approach is a promising new
development that presents ample opportunities for mathematicians to excel, e.g. by
exploring the role of model error.
Without a doubt, the increase in computer performance has been a decisive factor
for the progress in meteorological modelling. As such, it is essential to be prepared
for the next generations of computer architectures, which will most certainly require
a high level of scalability for numerical models. Close interaction between computer
hardware and software experts, numerical scientists and modellers would therefore
seem to be a prerequisite to progress once exascale computers become the standard.
vii
viii Foreword
This book provides an attractive and varied introduction to topics where math-
ematicians and modellers from the meteorological community can cooperate with
and help one another. I sincerely hope it will stimulate mathematicians to engage
in solving problems which operational weather centres face now and in the near
future. I would also like to warmly acknowledge the editors and contributors for
their valued initiative and excellent work.
ix
x Preface
We hope this volume can help reach this goal and reach many mathematicians.
We have divided the book into three parts, moving from mathematical and numerical
problems through air quality modelling to advanced applications in data assimilation
and probabilistic forecasting. We are confident that many mathematicians working
in numerical analysis, partial differential equations and stochastic analysis will find
a source of motivation and inspiration for their future research work.
Lastly, we of course wish to warmly thank the reviewers for their valued
contributions.
xi
xii Contents
Index . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 263
Part I
Numerical Methods for Meteorological
Problems
In Chap. 4, the authors derive and show the convergence of a parallel numerical
method for general nonlinear parabolic problems which usually arise in air pollution
transport problems. Due to the chemical part, they face with nonlinear elliptic
problems.
As seen from this short summary, Chaps. 1 and 2 treat numerical methods
being the right choice when modelling the large-scale dynamics of the atmosphere,
Chap. 3 deals with the modelling of turbulence occurring on much smaller scales,
and Chap. 4 gives a generic treatment of problems arising in, for instance, air quality
modelling. All of these fields will be covered in Parts II and III of this issue.
Chapter 1
On a Conservative Finite-Difference Method
for 1D Shallow Water Flows Based
on Regularized Equations
Abstract We deal with the 1d shallow water system of equations and exploit its
special parabolic regularization satisfying the energy balance law. We construct a
three-point symmetric in space discretization such that the discrete energy balance
law holds and check that it is well-balanced. The results of numerical experiments
for the associated explicit finite-difference scheme are also given for several known
tests to confirm its reliability and some advantages. The practical error behavior is
also analyzed.
1.1 Introduction
The 1d and 2d shallow water equations are widely exploited in hydrodynamics. The
vast literature exists on their numerical solving, in particular, see [14] and [1, 3, 12,
15, 17–19], etc.
Several years ago a special regularization of the equations together with explicit
two-level in time and symmetric in space finite-difference schemes based upon it
were suggested in [5, 8]. The approach is further applied for various problems
in [4, 9–11]. Previously the similar approach was developed for the more com-
plicated Euler and Navier–Stokes equations and its practical efficiency in various
A. Zlotnik ()
Department of Higher Mathematics at Faculty of Economic Sciences, National Research
University Higher School of Economics, Myasnitskaya 20, 101000 Moscow, Russia
e-mail: [email protected]
V. Gavrilin
Department of Mathematical Modelling, National Research University Moscow Power
Engineering Institute, Krasnokazarmennaya 14, 111250 Moscow, Russia
e-mail: [email protected]
compressible gas flow computations was demonstrated [6, 7, 16]; it is called a quasi-
gas dynamic (QGD) approach. Moreover, the regularized shallow water system of
equations is nothing more than a special case of the main version of the barotropic
QGD system, see [20, 22].
In this chapter, following [21–23], we change the space discretization in the
above mentioned schemes in order to guarantee the discrete energy balance law
not available originally. We present the proof of the law and check the important
property that the new discretization is well-balanced. The results of numerical
experiments for the associated explicit in time finite-difference scheme are also
given. We treat known tests such as a dam break (a disintegration of discontinuity)
and subcritical, transcritical and supercritical flows over a hump to confirm the
reliability and accuracy compared with the original scheme. In addition, the
practical error behavior is analyzed as mesh refines. Here we confine ourselves by
the 1d case; the 2d case should be considered in a forthcoming paper.
1.2.1. The 1d shallow water system of equations consists of the mass and momentum
balance equations
@t h C @x .hu/ D 0; (1.1)
where 0 < x < X, t > 0 and @t and @x denote the time and space derivatives. The
main unknown functions are the water depth h.x; t/ > 0 (measured from the bottom
mark b.x/) and the velocity u.x; t/. The pressure, the Navier–Stokes viscous stress
and the density of external force are given by
Here the density of the mass flux j, the viscous stress ˘ and the regularized density
h are given by
j D h.u w/; w D wO C u@x .hu/; wO D .hu@x u C @x p hF/; (1.7)
h h
˘ D @x u C huw
O C gh@x .hu/; h D h @x .hu/; (1.8)
where D .h; u/ > 0 is the relaxation multiplier. If one puts D 0, the system
becomes the original system (1.1)–(1.3).
Note that the following simplified formula holds
˚
wO D u@x u C g@x .h C b/ f : (1.9)
The regularized system (in equivalent form) was derived from the original one
in [5, 8] in 1d and 2d cases. On the other hand, the latter system is nothing more
than the special case of the main version of the barotropic QGD system of equations
from [20] (its earlier version can be found in [24]) for the gas density D h and the
pressure law p./ D 0:5g2 with the adiabatic exponent D 2. In 1d this is due to
formulas (1.9) and .u@x p C p@x u/ D gh@x .hu/.
For the barotropic QGD system of equations, the Petrovskii parabolicity, the
energy balance law and the linearized stability of the equilibrium solutions were
proved in [20, 22]; its very short derivation can be also found there. Remind that,
first, the original system (1.1) and (1.2) is hyperbolic for D 0 or of the composite
hyperbolic-parabolic type for > 0 and, second, the Petrovskii parabolicity implies
a unique solvability (local in time) for the Cauchy problem in Hölder spaces, see
[24] for details. These results demonstrate the dissipative nature of the barotropic
system. They straightforwardly imply the corresponding results for the regularized
shallow water system of equations, in particular, the following important one.
Theorem 1.1 For the regularized shallow water equations (1.5)–(1.8), the follow-
ing pointwise energy balance law holds
˚ ˚
0:5@t g.h C b/2 C hu2 C @x h.u w/ g.h C b/ C 0:5u2 ˘ u
˚ 2 ˚ 2
C.@x u/2 C g @x .hu/ C h u@x u C g@x .h C b/
˚
D h fu C h u@x u C g@x .h C b/ f : (1.10)
In the law on the left, the second term is a divergence one whereas the third
(Navier–Stokes) term and fourth and fifth relaxation terms are non-negative; the
property remains valid for > 0.
6 A. Zlotnik and V. Gavrilin
Notice that, for the equilibrium solutions h D hS .x/ > 0 and u D 0, both
systems (1.1)–(1.3) and (1.5)–(1.8) for f D 0 are reduced to the equation
The operators Œ, ./˙ and ı map functions given on !N h to ones defined on !h
whereas Œ and ı map functions given on !h to those defined on !h D fxi giD1
N1
.
We need two difference product rules
equations as follows
@t h C ı Œh.Œu w/ D 0; (1.17)
@t .hu/ C ı Œh.Œu w/Œu C Œ p D ı ˘ C Œh F (1.18)
on !h for t > 0. Here the following discretizations for the density of the mass flux
j, the viscous stress ˘ and the regularized density h are exploited
j D Œh.Œu w/; w D wO C Œuı.hu/; wO D ŒhŒuıu C ıp ŒhF ;
Œh Œh
(1.19)
˘ D ıu C ŒhŒuwO C gŒhı.hu/; h D Œh ı.hu/; F D gıb C f : (1.20)
The main unknown functions h and u together with the functions p and b are defined
on the main mesh !N h whereas the functions j, w, w,
O ˘ , h , , and f are defined
on the auxiliary mesh !h . Note that ıp D gŒhıh and therefore similarly to (1.9) the
following formula holds
O D fŒuıu C gı.h C b/ f g:
w (1.21)
We assume that h > 0 and derive the counterpart of the energy balance
law (1.10).
Theorem 1.2 For the semi-discrete method (1.17)–(1.20), the following pointwise
energy balance law holds
˚ ˚
0:5@t g.h C b/2 C hu2 C ı j .g.h C b/ C 0:5u uC / ˘ Œu C B
˚ 2
C .ıu/2 C g ı.hu/ C ŒhfŒuıu C gı.h C b/g2
D Œh f u C ŒhfŒuıu C gı.h C b/gf ; (1.22)
the mass balance equation (1.17) and twice formula (1.14) to obtain:
@t h u2 D ı j u2 D ı . jŒu2 / C jı.u2 / ;
ı . jŒu/ u D ı . jŒu2 / jŒuıu :
We add equality (1.23) and the last one. Formulas (1.14) and (1.16), respectively,
imply
˚
Œıp C gh ıb u D ıp C gŒhıb ı.hu/ gıb Œu C ı B ;
ı ˘ u D ı .˘ Œu/ C Œ˘ ıu ;
where
CfŒhŒu.wO C f / C gŒhı.hu/gıu
1 On a Conservative Finite-Difference Method 9
.h; u/
˚
D Œhgı.h C b/ C ŒhŒuıu .wO C f / C gıh Œuı.hu/ C gŒhı.hu/ ıu
˚ 2 ˚
D Œh Œuıu C gı.h C b/ C ı.hu/ gıh Œu C gŒhıu :
Finally, using the formula ıh Œu C Œhıu D ı.hu/, see (1.13), we rewrite .h; u/ as
the sum of two squared terms and thus pass from (1.25) to the desired result (1.22).
t
u
For f D 0 and suitable boundary conditions at x D 0; X, for example, for the
periodic ones, clearly law (1.22) implies non-increasing of the total energy in time
X
N1
˚
0:5@t g.h C b/2 C hu2 i O i 6 0;
iD0
where O 0 D .h1 C hN /=2, that is the discrete counterpart of the energy inequal-
ity (1.11).
In the discrete energy balance law (1.22), the term u uC is like the geometric
mean for u2 . The summand ı B is the divergence mesh imbalance [additional with
respect to (1.10)] with B D O.2max / for the functions of the continuous argument
provided that h, @x h, @x u and @x b are bounded.
For the equilibrium solution h D hS .x/ > 0 and u D 0, method (1.17)–(1.20) for
f D 0 is reduced to the equations [with the help of formula (1.15)]
˚
ı ıp.hS / C gŒhS ıb D 0; ıp.hS / C gŒhS ıb D 0 on !h :
Notice that ıp.hS / C gŒhS ıb D gŒhS ı.hS C b/. By virtue of the first equation
gŒhS ı.hS C b/ D C0 const on !h , and next C0 D 0 due to the second equation.
Consequently
in [5, 8, 11] (generalized for the non-uniform !N h and written in our notation), where
Q and fQ are defined on !N h . Moreover, inserting formula (1.21) into the second
formula (1.19) leads to
˚
wD 2ŒhŒuıu C Œu2 ıh C gı.h C b/ f
Œh
Œ
Q ˚
wD Q gı.h C b/ fQ
ı.hu2/ C Œ
Œh
was applied. Since ı.hu2 / D 2ŒhŒuıu C Œu2 ıh but Œu2 6 Œu2 [see (1.13)
and (1.24)], these two formulas for w also differ.
On the other hand, inserting formula (1.21) into the first formula (1.20) gives
˚
˘ D ıu C ŒhŒu ı g.h C b/ C 0:5u2 f C gŒhı.hu/
on !h for 0 6 m 6 M 1, where hO and hu b are the values for the upper time level.
After computing them, we set uO WD hu=h assuming that hO > 0. Here we also exploit
b O
the same expressions (1.19) and (1.20).
1 On a Conservative Finite-Difference Method 11
For solving the inviscid shallow water system of equations, the viscous and
relaxation terms are considered as artificial regularizers with and in the form
4 p
D Œ p or D 0; D ˛ ; c D gŒh
3 c
on !h like in [7, 16], where c is the counterpart for the velocity of sound and 0 <
˛ < 1. To satisfy a stability condition of the Courant-Friedrichs-Lévy type, the time
step at the current time level is chosen as
t D ˇ min
16i6N jŒui1=2 j C ci1=2
with 0 < ˇ < 1, and thus t varies from one time level to another. The scheme
parameters ˛ and ˇ are adjusted experimentally in each problem.
Notice that, at least for the uniform space mesh, the approximation order of the
semi-discrete method (1.17)–(1.20) [with respect to the regularized system (1.5)–
(1.8)] is O.2 / with D X=N for smooth solutions but it is reduced to O./ for
the finite-difference scheme owing to the above choices of and t.
1.2.5. In our computations below, we use a uniform mesh in space and take f D 0.
We investigate the convergence of functions v D h; u and hu for our finite difference
scheme at the final time tfin as the space mesh refines; this is a standard matter,
for example see [13, 15, 17], and especially important in the case of non-smooth
solutions where the theoretical information is poor. In those rare cases where a
sought function vex is known, we compute the average absolute value of the error
on the mesh (the scaled mesh L1 -norm)
1 Xˇˇ .N/ ˇ
N
.N/
Eex Œv WD vex i vi ˇ; (1.28)
N C 1 iD0
.N/
where vex i and vi are the exact and approximate quantities at xi D iX=N.
Otherwise, we apply two formulas
where v .N/ is the pseudo-exact solution computed for the finer mesh with the step
D X=N. Here N is a multiple of 2N, and due to our numerical experience it is
desirable that 4N 6 N. The second formula is associated to the Runge rule for the
practical error estimation and is cheaper in implementation (v .N/ is not required).
We compare both of them with (1.28) if possible.
12 A. Zlotnik and V. Gavrilin
ˇ ˇ
Below we plot the dependence of ˇln E.N/ Œvˇ (the subscripts are omitted) on ln N.
Moreover, we approximate them by linear functions using the least square method
and thus get the corresponding approximate error orders pex Œv, pŒv and pR Œv.
1.3.1. We first compute a flow in a channel of the length X D 1500 m. In its central
part, a rectangular ledge is symmetrically located (having the height 8 m and the
length 375 m), and the other bottom is flat. At the center of the channel, a dam is
situated. Initially the water level H0 .x/ equals 20 m to the left and 15 m to the right
of the dam, and the flow is at rest, i.e. the initial velocity is zero: u0 .x/ D 0. The
dam breaks at the moment t D 0. The open boundary conditions are imposed for h
and u at the both boundaries x D 0; X, i.e., for example, h0 D h1 and u0 D u1 for
the left boundary.
The computational results are shown in Fig. 1.1 for t D 15 s and t D 60 s, for
N D 400; the scheme parameters are ˛ D 0:35 and ˇ D 0:5. They are in a good
agreement with [15] for N D 200 and in [11] for N D 400 (for other schemes).
H u
25 3
2.5
20
10 1.25
0 0
100 200 300 400 100 200 300 400
H u
25
3.5
20
2.5
10 1.5
0.5
0
100 200 300 400 100 200 300 400
Fig. 1.1 Dam break in the channel with the rectangular ledge: H and u at t D 15 s (above) and
t D 60 s (below)
1 On a Conservative Finite-Difference Method 13
4.5 E(h)
E(u)
| ln E |
3.5
2.5
In Fig. 1.2, the errors in h and u for t D 15 s are presented for N D 200, 400, 800
and 1600 as well as N D 6400. We note that the error polylines are not completely
linear. The error orders pŒh 0:946 and pŒu 0:908 are rather close to each other
but both less than 1.
1.3.2. Next we consider a channel of the length X D 25 m with a flat bottom
except for the small hump of the parabolic shape in its middle part:
(
0:2 0:05.x 10/2 ; 8 m x 12 m
b.x/ D
0; otherwise:
Initially the water level is constant: H0 .x/ CH , and the flow is at rest. The left
boundary conditions are hujxD0 D Chu for the discharge together with the open
boundary condition for h, and the right boundary conditions are HjxDX D CH (in
general, up to a certain time moment) together with the open boundary condition
for u.
This problem may seem simple but only at first glance. There are three types of
flows in it: subcritical, transcritical and supercritical depending on values of the flow
parameters. In what follows, examples of all three types of flows are considered.
Results of computations are presented at tfin D 200 s (at this moment the flows
become stationary for the chosen values of the parameters). Note that, for these
flows, the exact discharge at the final time is known: hu Chu . As a rule, namely
the computation of hu causes difficulties. The results are accurate enough for N
listed below; they are comparable with those obtained in [3] (for N D 100), [12]
(for N D 300) and [1, 18, 19] (for N D 200) by other schemes but much better for
hu than in [11] (for the same N).
(a) Subcritical flow. This is the simplest type of flows. Here CH D 2 m and Chu D
4:42 m2 /s are taken. At tfin , the water level is almost flat with only a small cavity
above the hump.
We select ˛ D 0:9, ˇ D 0:2 and D 0. Figure 1.3 shows the water level H
and the discharge hu for N D 400. In the vicinity of the hump edges, “hubbles”
14 A. Zlotnik and V. Gavrilin
H hu
2.5 4.422
4.42
1
0 4.418
100 200 300 400 100 200 300 400
Fig. 1.3 Subcritical flow over the hump: H and hu at tfin D 200 s
14
E(h)
E(u)
11
E(hu)
E (h)
| ln E |
R
E R (u)
E R (hu) 8
E ex (hu)
of hu values are observed (damping as N increases) but the absolute error in the
discharge Eabs WD max j.hu/i Chu j 7:375e-5 is small.
0iN
In Fig. 1.4, errors in h, u and hu are shown for N D 100, 200, 400 and 800 as
well as N D 3200. The same values are taken for two other flow types below.
The polylines of errors in h and u are more close to linear ones than for the
previous problem (though this is not the case for jln Eex Œhuj). Now all the orders
pŒh 1:193, pR Œh 1:139, pŒu 1:183 and pR Œu 1:121 are all close to each
other and slightly greater than 1. They are essentially less than pex Œhu 1:709.
Comparing the results by formulas (1.28) and (1.29) for hu, we see that the error
orders pR Œhu 2:023 and pŒhu 2:017 are close to each other but both
overestimate the value pex Œhu.
(b) Transcritical flow. Here CH D 0.66 m and Chu D 1:53 m2 /s are chosen. For this
and the next types of flows, the right boundary condition H.X; t/ D CH is posed
only for t 40 s whereas it is replaced by the open boundary condition for h
for t > 40 s. In this case, the behavior of the stationary water level H is more
complicated exhibiting much more sharp change (its values on the left and right
Another random document with
no related content on Scribd:
noteworthy—there are no altar rails, but their place is taken by three
massive oaken benches, covered at all times with the “houseling
linen,” fair white cloths—the use of which goes back to very early
times. The benches which now stand across the presbytery floor,
close to the topmost of the flight of steps leading up to it, are the
remains of ten such benches, which were made in Puritan times for
communicants to sit on as they received the sacred elements. When
the custom of kneeling was revived, these benches were placed on
the steps; and on “Sacrament Sundays,” the clerk, after morning
prayer, went to the lectern and bade those who were prepared to
receive the Holy Communion to draw near, whereupon intending
communicants left the nave and knelt at the benches, or in the choir
stalls, until the officiating clergy brought them the sacramental bread
and wine. In 1852, when sundry changes were being made in the
arrangements of the church, all these benches except three were
removed—the three which were in use as altar-rails.
The beautiful triplet of windows, over the altar, end the long vista
seen from the west. Beneath the presbytery floor is a vaulted crypt.
This is not, as many crypts are, dark and gloomy, but well lit by
triangular windows, which from the outside are seen to be just above
the level of the churchyard. This crypt does not retain its original
altar, but its place is marked by a piscina on the southern side; two
arches open out into the choir aisles, through which those kneeling
in the aisles might look down on the priest officiating at the altar of
the crypt.
There are several monuments worthy of notice, but by far the most
interesting is the Beaufort altar-tomb on the south side of the
presbytery. This was erected, to the memory of her father and
mother, by the Lady Margaret. foundress of Christ’s and St. John’s
Colleges, at Cambridge, well known for her many benefactions, and
from the fact that she was the mother of Henry VII. As the history of
this family—the Beauforts—is interesting, and its details little
touched on by the writers of English history, it may not be out of
place, especially as they were connected with Wimborne, to give an
outline of it here. Everyone knows that John of Gaunt was the third
son of Edward III. and the father of Henry IV., but many know only in
a general way that the House of Tudor traced through him their claim
to the English crown. John of Gaunt married Blanche, great-
granddaughter of Edmund Crouchback, second son of Henry III.,
who was created Earl of Lancaster in 1257. Her father, on account of
his valour in the French wars of Edward III., had been made a Duke
—a new title as far as England was concerned, for the only English
Duke that had been previously created was the Duke of Cornwall,
better known as the Black Prince. Through Blanche, his wife, John of
Gaunt succeeded to the estates of the Duke of Lancaster, among
them to the Castle of Beaufort, in Anjou. He was himself in 1362
created Duke of Lancaster. Among Blanche’s maids of honour was
one Kate, daughter of Sir Payne Roet, and widow of Sir Owen
Swynford. When Blanche died, John married Constance of Castile,
but took unto himself Kate Swynford as his mistress; by her he
became the father of four children, all born at Beaufort Castle. As
they were illegitimate, they took the name of their birthplace as a
surname. The eldest of these was John de Beaufort, and the
second, Henry, the celebrated Cardinal Beaufort. When Constance
died, these four children were legitimatised by a Bull of Urban VI.,
then by Richard II., then by Act of Parliament. The Duke then
married Kate. On January 13th, 1396, John de Beaufort was created
Earl of Somerset. He died in 1410, leaving four sons and two
daughters; the eldest, Henry, did not long survive his father, and his
title and estates passed to his next brother, John. He greatly
distinguished himself in the French wars, and was made Duke of
Somerset, Earl of Kendale, Lieutenant of the Duchy of Aquitaine and
Captain-General of the whole realm of France and Normandy. In
1436 the Duke of Bedford, the Regent, died, and as the King, Henry
VI., was still a minor, another regent had to be appointed. The Duke
of Lancaster thought he should have obtained this important post,
but it was conferred on the Duke of York, and Lancaster therefore
retired from active service, and in 1440 married Margaret, widow of
Oliver St. John, and daughter of John, Lord Beauchamp, of Bletsoe
Manor, Bedfordshire. This John, Duke of Somerset, and Margaret,
his wife, are they whose figures lie side by side in alabaster on their
altar-tomb at Wimborne. Their right hands are clasped together;
angels guard their heads; his feet rest on a dog, hers on an
antelope; he is clad in complete armour, the face and right hand
alone bare; the left hand holds the right-hand gauntlet, which he has
taken off before taking the lady’s hand. On the apex of the arch,
above the tomb, hangs the helm which he, during his life, used to
wear in tournaments. Their only child was born in 1441—Margaret,
of whom mention has been made. Her father died in 1444, aged
thirty-nine years, and the Duchy of Somerset became extinct in the
Beaufort family. His death took place at Kingston Lacy, an estate
close to Wimborne, belonging to the Beaufort family. His widow and
daughter went to live on the Bedfordshire property. In about four
years time, the widow married her third husband, Lord Welles. Young
Margaret, when only nine years of age, was sought by the Duke of
Suffolk as a wife for his son, John de la Pole, and by King Henry VI.
as wife for his half-brother, Edmund Tudor, Earl of Richmond, son of
Catherine, the Queen of Henry V., by her second husband, Sir Owen
Tudor. Margaret was a clever girl, well educated, knowing even then
Latin and French; but sorely distraught was she to know which of the
two suitors to choose: so she consulted an old gentlewoman, who
advised her to commit the matter to St. Nicholas. She took the
advice, prayed to the saint, and fell asleep, and about four o’clock
next morning, whether sleeping or waking she could not tell, saw one
standing in her room, habited in a bishop’s robes, who bade her
accept Edmund Tudor as a husband. She told her mother, and she
was betrothed to the Earl of Richmond, and they were married in
1455, when she was fourteen years of age and he twenty-four. They
lived at Pembroke Castle, which belonged to Jasper Tudor, Earl of
Pembroke. In 1456 her only son, Henry, afterwards Henry VII., was
born, and shortly after this her husband died. He was buried at
Caermarthen Abbey, and when the monastery was suppressed, his
body was removed to the Cathedral Church at St. David’s. His
mother, anxious to keep quite aloof from party strife (for the War of
the Roses had already broken out), lived on at Pembroke, educating
her son. In 1459 she married her second husband, Sir Humphrey
Stafford; widowhood, for one of exalted rank, not being a desirable
condition in those times of war and turmoil. It has been seen that her
mother was thrice married, and Margaret followed her example, for
when Sir Humphrey died in 1481, she, at the end of a year, being
then about forty years of age, married Thomas, Lord Stanley. After
fifteen years she separated from him with his consent, in order to
devote herself to a religious life, and retired to the convent at
Woking, in Surrey. It must not be supposed that she had a peaceful
or happy life. Her thoughts were centred on her only son, and many
were the years of separation from this son that his mother had to
endure. The story of his wanderings, his dangers, his detention in
Brittany, are too long to be told here—suffice it to say that Richard III.
became so odious to the chief nobles that at last it was arranged that
Henry Tudor, Earl of Richmond, Margaret’s son, should be
recognised by the Lancastrian party as their leader, and should claim
the throne; and that in order to gain the adherence of those
members of the Yorkist party who were opposed to Richard, a
marriage should take place between Henry and Elizabeth, the
daughter of Edward IV. Both the mothers agreed to the union: but
the first attempt at invasion by Richmond was a failure, and Richard
seemed free from all danger; and with a view to win over his Yorkist
opponents, he made up his mind to marry Elizabeth himself,
although she was his own brother’s child. This step led to a fresh
invasion—this time a successful one—and the death of Richard on
Bosworth field virtually placed the crown on Henry’s head (1485). So
at last the Lady Margaret’s troubles were at an end, and she saw her
son crowned and wedded, and the red and white roses twined
together. It remains only to speak of her benefactions. Those at
Cambridge are well known; and the writer of this chapter, who once
held at Christ’s College one of the scholarships she founded,
cherishes the memory of the royal and pious lady with all due
gratitude. But it is of her benefactions at Wimborne that mention
must now be made. About 1498 she built the beautiful monument to
her father and mother in the Minster; founded a chantry, where, for
her own soul, and for those of her son, her parents, and ancestors,
Mass was to be duly said. She founded and endowed the Grammar
School, though, as its endowments were greatly added to by Queen
Elizabeth, its name was afterwards changed from the Lady
Margaret’s to that of Queen Elizabeth’s Free Grammar School.
The Lady Margaret just outlived her son; both died in the same
year, 1509, but she rather later than he.
Of the other monuments space forbids mention of any, save that of
Anthony Etricke, if it can be called a monument, for it is really the
sarcophagus that contains the body of this eccentric magistrate, who
lived at Holt, and was recorder of Poole. He desired, for some
reason, to be buried neither above the ground nor under it, neither
within the church nor outside of it; and in order to carry out this
strange wish, he got permission to cut a niche in the south wall,
partly below the level of the surface of the churchyard, and in it fixed
a slate sarcophagus. In this he ordered his coffin to be deposited
when he died; and he made all the preparations he could
beforehand, even to the painting of the date of his death on the side
of the slate coffin. He had a presentiment that he should die in 1691,
but he lived till 1703, so that the real date of his death had to be
placed on the coffin. The other date was not obliterated, but the new
one was painted on the other, and the two may be seen there to-day.
The arms of his family are painted on the lid, and, as he left twenty
shillings per annum to keep coffin and niche in good repair, the
sarcophagus is bright and smart as paint, gilding, and varnish can
make it.
The Chained Library, Wimborne Minster.