Professional Documents
Culture Documents
Textbook Large Scale Scientific Computing 11Th International Conference LSSC 2017 Sozopol Bulgaria June 5 9 2017 Revised Selected Papers 1St Edition Ivan Lirkov Ebook All Chapter PDF
Textbook Large Scale Scientific Computing 11Th International Conference LSSC 2017 Sozopol Bulgaria June 5 9 2017 Revised Selected Papers 1St Edition Ivan Lirkov Ebook All Chapter PDF
https://textbookfull.com/product/information-security-and-
cryptology-13th-international-conference-inscrypt-2017-xian-
china-november-3-5-2017-revised-selected-papers-1st-edition-
xiaofeng-chen/
https://textbookfull.com/product/cloud-computing-and-service-
science-7th-international-conference-closer-2017-porto-portugal-
april-24-26-2017-revised-selected-papers-donald-ferguson/
https://textbookfull.com/product/studies-on-speech-
production-11th-international-seminar-issp-2017-tianjin-china-
october-16-19-2017-revised-selected-papers-qiang-fang/
https://textbookfull.com/product/computational-aspects-and-
applications-in-large-scale-networks-net-2017-nizhny-novgorod-
russia-june-2017-valery-a-kalyagin/
https://textbookfull.com/product/fundamentals-of-software-
engineering-7th-international-conference-fsen-2017-tehran-iran-
april-26-28-2017-revised-selected-papers-1st-edition-mehdi-
Ivan Lirkov
Svetozar Margenov (Eds.)
LNCS 10665
Large-Scale
Scientific Computing
11th International Conference, LSSC 2017
Sozopol, Bulgaria, June 5–9, 2017
Revised Selected Papers
123
Lecture Notes in Computer Science 10665
Commenced Publication in 1973
Founding and Former Series Editors:
Gerhard Goos, Juris Hartmanis, and Jan van Leeuwen
Editorial Board
David Hutchison
Lancaster University, Lancaster, UK
Takeo Kanade
Carnegie Mellon University, Pittsburgh, PA, USA
Josef Kittler
University of Surrey, Guildford, UK
Jon M. Kleinberg
Cornell University, Ithaca, NY, USA
Friedemann Mattern
ETH Zurich, Zurich, Switzerland
John C. Mitchell
Stanford University, Stanford, CA, USA
Moni Naor
Weizmann Institute of Science, Rehovot, Israel
C. Pandu Rangan
Indian Institute of Technology, Madras, India
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Demetri Terzopoulos
University of California, Los Angeles, CA, USA
Doug Tygar
University of California, Berkeley, CA, USA
Gerhard Weikum
Max Planck Institute for Informatics, Saarbrücken, Germany
More information about this series at http://www.springer.com/series/7407
Ivan Lirkov Svetozar Margenov (Eds.)
•
Large-Scale
Scientific Computing
11th International Conference, LSSC 2017
Sozopol, Bulgaria, June 5–9, 2017
Revised Selected Papers
123
Editors
Ivan Lirkov Svetozar Margenov
Institute of Information and Communication Institute of Information and Communication
Technologies Technologies
Bulgarian Academy of Sciences Bulgarian Academy of Sciences
Sofia Sofia
Bulgaria Bulgaria
Scientific Committee
James Adler Tufts University, USA
Pasqua D’Ambra Istituto per le Applicazioni del Calcolo Mauro Picone,
CNR, Naples, Italy
Emanouil Atanassov Institute of Information and Communication Technologies,
BAS, Bulgaria
Fleurianne Bertrand Universität Duisburg-Essen, Germany
Pavel Bochev Sandia National Laboratories, USA
Ivan Dimov Institute of Information and Communication Technologies,
BAS, Bulgaria
Stefka Dimova Sofia University, Bulgaria
Stefka Fidanova Institute of Information and Communication Technologies,
BAS, Bulgaria
Salvatore Filippone Cranfield University, UK
Krassimir Georgiev Institute of Information and Communication Technologies,
BAS, Bulgaria
Jay Gopalakrishnan Portland State University, USA
Todor Gurov Institute of Information and Communication Technologies,
BAS, Bulgaria
Xiaozhe Hu Tufts University, USA
Oleg Iliev ITWM, Germany
Aneta Karaivanova Institute of Information and Communication Technologies,
BAS, Bulgaria
Mikhail Krastanov Sofia University, Bulgaria
Johannes Kraus University of Duisburg-Essen, Germany
Ulrich Langer Johannes Kepler University Linz, Austria
Raytcho Lazarov Texas A&M University, USA
Ivan Lirkov Institute of Information and Communication Technologies,
BAS, Bulgaria
Gabriel Luque University of Málaga, Spain
Svetozar Margenov Institute of Information and Communication Technologies,
BAS, Bulgaria
Joseph Pasciak Texas A&M University, College Station, USA
Mauro Perego Sandia National Laboratories, USA
Daniela di Serafino Università degli Studi della Campania Luigi Vanvitelli,
Italy
Gerhard Starke University of Duisburg-Essen, Germany
Vladimir Veliov TU Vienna, Austria
VIII Organization
Invited Papers
On the Winter Wave Climate of the Western Black Sea: The Changes
During the Last 115 Years . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
Vasko Galabov and Hristo Chervenkov
Contributed Papers
1 Introduction
2 Energy Laws
where u represents the fluid velocity and ν is the fluid viscosity, accounting for
the dissipation in the system. Applying the so-called least-action principle results
in the integral equation,
∂u
+ ∇p, y = ∇ · ν∇u, y , ∀y ∈ V,
∂t
where we assume an incompressible fluid, ∇ · u = 0, and an appropriate Hilbert
space, V. Here, we use ·, · to denote the L2 (Ω) inner product. In strong form,
we obtain the time-dependent Stokes’ equations (assuming appropriate boundary
conditions):
∂u
+ ∇p − ∇ · ν∇u = 0, (3)
∂t
∇ · u = 0. (4)
Note that the energy law can also be derived directly from the PDE itself. First,
we consider the weak form of (3)–(4), multiplying (3) by u and (4) by p and
integrate over Ω. After integration by parts we obtain the following relations:
∂u
0= + ∇p − ∇ · ν∇u, u + ∇ · u, p
∂t
∂u
= , u + ∇p, u − ∇ · ν∇u, u + ∇ · u, p
∂t
1 ∂
= u, u + ∇p, u + ν∇u, ∇u − u, ∇p .
2 ∂t
Here, we have assumed that the boundary conditions are such that the boundary
terms, resulting from the integration by parts, vanish. Hence, we have
1 ∂
u, u = − ν∇u, ∇u .
2 ∂t
This approach can also be applied to other PDEs, such as the heat equation,
to show similar energy dissipation relations. Let ν be the thermal diffusivity of
the body Ω, and u its temperature. Then the PDE describing the temperature
distribution in Ω is as follows,
∂u
− ∇ · ν∇u = 0, on Ω, u = 0, on ∂Ω. (5)
∂t
As before, we multiply (5) by u and integrate over Ω to obtain that
∂u ∂u
0= − ∇ · ν∇u, u = , u − ∇ · ν∇u, u
∂t ∂t
1 ∂
= u, u + ν∇u, ∇u
2 ∂t
Hence,
1 ∂
u, u = − ν∇u, ∇u ,
2 ∂t
which is the scalar version of (2).
6 J. H. Adler et al.
For the remainder of the paper, we analyze (2), specifically how closely the
FOSLS method can approximate the energy law discretely. We will consider
both the scalar (heat equation) and the vector version (Stokes’ equation) in the
numerical results, as the form of the energy law is identical. First, we discuss
how moving to a finite-dimensional space affects the energy law.
3 Heat Equation
First, we consider the heat equation, assuming a constant diffusion coefficient
ν = 1 for simplicity, homogeneous Dirichlet boundary conditions, and a given
initial condition:
∂u(x, t)
= Δu(x, t) ∀x ∈ Ω, ∀t > 0 (6)
∂t
u(x, t) = 0 ∀x ∈ ∂Ω, ∀t ≥ 0 (7)
u(x, 0) = u0 (x) ∀x ∈ Ω̄. (8)
To discretize the problem in time, we consider a symplectic, or energy-conserving,
time-stepping scheme such as Crank-Nicolson. Given a time step size, τ , and time
tn = τ n, we approximate un = u(x, tn ) with the following semi-discrete version
of (6),
un+1 − un Δun+1 + Δun
=
τ 2
To simplify the calculations later, we introduce an intermediate approxima-
tion, un+ 12 , and re-write the semi-discrete problem as
un+ 12 − un
τ = Δun+ 12
2
un+ 12 (x) = 0 ∀x ∈ ∂Ω, n = 0, 1, 2, . . . (9)
un+1 = 2un+ 12 − un
Remark 1. To obtain the semi-discrete energy law for (9), we perform a similar
procedure as done in Sect. 2, where we multiply the first equation in (9) by
un+ 12 and integrate over the domain. After some simple calculations, we obtain
the corresponding energy law, using L2 −norm notation:
||un+1 ||2 − ||un ||2
= −||∇un+ 12 ||2 (10)
2τ
To use the FOSLS method, we now put the operator into a first-order system.
Since we have reduced the problem to a reaction-diffusion type problem, we
introduce a new vector V = ∇u, and use the H 1 -elliptic equivalent system [8,9]:
⎛ ⎞ ⎛2 ⎞
−∇ · Vn+ 12 + τ2 un+ 12
un+ 12 τ un
Lτ = ⎝ Vn+ 12 − ∇un+ 12 ⎠ = ⎝ 0 ⎠ . (11)
Vn+ 12
∇ × Vn+ 1 0
2
Energy Laws for FOSLS 7
Note that Dirichlet boundary conditions on the continuous solution, u, gives rise
to tangential boundary conditions on V, V × n = 0, where n is the normal
vector to the boundary.
Next, we consider a finite-dimensional subspace of a product H 1 space, V h ,
and perform the FOSLS minimization of (11) over V h :
⎛ 2 h ⎞
u τ un
h h
un+ 1 , Vn+ 1 = arg min Lτ − ⎝ 0
⎠
V ,
2 2
(u,V )∈V h 0
uhn+1 = 2uhn+ 1 − uhn .
2
For each n, the above minimization results in the following weak set of equations:
h ⎛ 2 h⎞
un+ 1 τ un
Lτ h
2
− ⎝ 0 ⎠ , Lτ φh = 0 ∀φh ∈ V h , (12)
Vn+ 1
2 0
where the inner products and norms are all in L2 (scalar or vector, depending
on context), unless otherwise noted.
Note, that with the introduction of V, the discrete form of the FOSLS energy
law can now be written,
The goal of the remainder of this Section is to show how well this energy law is
satisfied. To do so, we make use of the following assumption.
Assumption 1. Assume that the initial condition is smooth enough and the
projection onto the finite-element space has the following property,
u0 − uh0 1 ≤ Chp u0 p+1 ,
H H
un+1 Hp ≤ C un Hp
2 2
− Δu + u = un , in Ω, u = 0 on ∂Ω, (15)
τ τ
where the exact semi-discrete solution un , at the previous time step, is used in
the right-hand side. By the triangle inequality,
h
un+ 1 − un+ 12 ≤ uhn+ 1 − ũhn+ 1 + ũhn+ 1 − un+ 12 . (16)
2 2 2 2
By Lemma 3, we have
h h h
un+ 1 − ũhn+ 1 ≤ uhn+ 1 − ũhn+ 1 + Vn+ 1 − Ṽ
n+ 1
2 2 2 2 H1 2 2 H1 (17)
≤ C2 (τ ) uhn − un .
Energy Laws for FOSLS 9
The functions ũhn+ 1 and un+ 12 are, respectively, FOSLS and exact solutions
2
of the same boundary value problem (15). Hence, from Lemma 2, we have
C(τ )
h C(τ ) p
ũn+ 1 − un+ 12 ≤ hp u n H p−1 ≤ h un Hp (18)
2 τ τ
Combining (16), (17) and (18), we obtain (14).
Finally, we have the following result on the approximation of the exact energy
law (13).
h
un
Theorem 1. Let be the solution to the FOSLS system, (12), at time
Vnh
h h
step n (with un+ 1 and Vn+ 1 defined as before). There exists C(τ ) > 0 such that
2 2
⎛ ⎞
h uh 1
un+1 2 − uhn 2 n+ 2
h 2 2 h ⎜ h ⎟ h
+ Vn+ 1
≤ C(τ ) u n − u n min V
⎝ n+ 1 ⎠ − Lτ φ .
2τ 2 τ φ h ∈V h 2
0
Proof. To simplify the notation, define the energy law we wish to bound as
uhn+1 2
− uhn 2
Enh := h
+ Vn+ 1
2
.
2τ 2
Note that
2 2 h uh − uh
1 uhn+1 − uhn un+1 − uhn h n+ 12 n
h
= , un+ 1 = τ , un+ 1 ,
2 τ τ 2
2
2
and
h 2 h h h h h
Vn+ 1 = −∇ · Vn+ 1,u
n+ 1 + Vn+ 1 − ∇u
n+ 1,V
n+ 1 ,
2 2 2 2 2 2
Vn+ 1 − Vn+ 1
2
τ
2 H1
2 2
≤ C(τ ) Mnh uhn − un + Mnh uhn − un .
τ τ
Combining the two terms completes the proof.
To provide a better bound for the FOSLS energy law (13), we introduce a mea-
sure for the truncation error defined as
⎛ ⎞
uh 1 (v)
n+
1 ⎜ h 2 ⎟ h
δn = max min V
⎝ n+ 1 ⎠(v) − Lφ , (19)
v∈H p+1 (Ω) v H p+1 φ h ∈V h 2
0
Energy Laws for FOSLS 11
Proof. Using the definitions of uhn+ 1 and un+ 12 , the triangle inequality, and
2
Lemma 4,
2
≤ C1 (τ )hp un Hp + (C2 (τ ) + 1) uhn − un .
τ
An induction argument then gives
n
2 j−1 n
uhn − un ≤ C1 (τ )h p
(C2 (τ ) + 1) un−j Hp + (C2 (τ ) + 1) uh0 − u0 .
τ j=1
With Assumption 1,
n
2 j−1 n
uhn − un ≤ C1 (τ )hp (C2 (τ ) + 1) un−j Hp + (C2 (τ ) + 1) u0 H p+1 .
τ j=1
Then, with the definition of δ and the result from Theorem 1, the proof is com-
plete.
We note that the bound in Corollary 1 is a rather pessimistic one. At a
fixed time, t, we expect the quality of both the fully discrete and semi-discrete
approximations to the true solution to improve as τ → 0 and more time-steps
are used to reach time t; thus, uhn − un should decrease as τ → 0 for n = t/τ .
Furthermore, for the unforced heat equation, we expect both uhn and un to
decrease in magnitude with n, but this is not accounted for in the bound in
Corollary 1. The bound above worsens with smaller τ and bigger n, showing
the limitations of bounding uhn − un by terms depending only on u0 and the
finite-element space.
Remark 2. As shown in the numerical experiments, Sect. 5, the constant δ
defined in (19) is of order hp for a smooth
solution. This indicates that the
energy law (13) holds with order O h2p . While the theoretical justification
of such statement may be plausible, it is nontrivial as the discrete quantities
involved in the definition of δ do not possess enough regularity (they are just
finite-element functions, only in H 1 ).
12 J. H. Adler et al.
4 Stokes’ Equations
Next, we return to the time-dependent Stokes’ equations, (3)–(4). For simplicity,
we again assume ν = 1, and rewrite the equations using Dirichlet boundary
conditions for the normal components of the velocity field, and zero-mean average
for the pressure field,
∂u(x, t)
− Δu(x, t) + ∇p(x, t) = 0 ∀x ∈ Ω, ∀t > 0 (21)
∂t
∇ · u(x, t) = 0 ∀x ∈ Ω, ∀t > 0 (22)
n · u(x, t) = 0 ∀x ∈ ∂Ω, ∀t ≥ 0 (23)
u(x, 0) = g(x) ∀x ∈ Ω̄, (24)
p(x, t) dV = 0 ∀t ≥ 0. (25)
Ω
To use the FOSLS method, we put the operator into a first-order system in a
similar fashion to the heat equation. Least-squares formulations are well-studied
for Stokes’ system and we consider a simple, velocity-gradient-pressure formu-
lation, where a new gradient tensor, V = ∇u, is used to obtain an H 1 -elliptic
equivalent system [4,6,14]:
⎛ ⎞ ⎛ ⎞
−∇ · Vn+ 12 + ∇pn+ 12 + τ2 un+ 12 2
un
⎛ ⎞ ⎜ ⎟ ⎜ τ
un+ 12 ⎜ ∇ · un+ 12 ⎟ ⎜ 0 ⎟ ⎟
⎜ ⎟ ⎜
Lτ ⎝Vn+ 12 ⎠ = ⎜ Vn+ 12 − ∇un+ 12 ⎟=⎜ 0 ⎟ ⎟. (27)
⎜ ⎟ ⎝
pn+ 12 ⎝ ∇ × Vn+ 12 ⎠ 0 ⎠
∇trVn+ 1 0
2
Note that the weak system is similar to (12) and the energy law is identical
to (13) in vector form. Thus, all the above theory still holds subject to enough
regularity of the solution to the time-dependent Stokes’ equations [21,22] and a
suitable generalization of the definition of δ.
5 Numerical Experiments
For the numerical results presented here, we use a C++ implementation of the
FOSLS algorithm, using the modular finite-element library MFEM [20] for man-
aging the discretization, mesh, and timestepping. The linear systems are solved
by direct method using the UMFPACK package [10].
103
p=1
p=2
100 p=3
h2 convergence
Energy Law Error
h4 convergence
10−3
h6 convergence
10−6
10−9
10−12
0 2 4 6
# Refinement Levels,
Fig. 1. Energy law error, (13), vs. number of mesh refinements, (h = 21 ), for the
FOSLS discretization of the heat equation, (12), using various orders of the finite-
element space (p = 1 - linear; p = 2 - quadratic; and p = 3 - cubic). One time step is
performed with τ = 0.005.
14 J. H. Adler et al.
Figure 1 displays the convergence of the energy law to zero as the mesh is
refined for a fixed time step. The convergence is O h2p , where p is the order of
the finite-element space being considered, confirming Theorem 1. It also suggests
that the constant δ is O (hp ), as is remarked above.
100
10−1 p=1
p=2
p=3
10−4
10−4
Energy Law Error
10−12 10−10
10−13
10−16
0 20 40 60 0 0.2 0.4 0.6 0.8 1
# Time Steps, n Time Step Size, τ
(a) (b)
Fig. 2. Energy law error, (13), vs. (a) number of time steps, n (with fixed τ = 0.005),
and (b) time step size, τ , for the FOSLS discretization of the heat equation, (12), using
various orders of the finite-element space (p = 1 - linear; p = 2 - quadratic; and p =
1
3 - cubic). Mesh spacing is h = 32 .
Figure 2 indicates how the timestepping affects the convergence of the energy
law. As discussed above, taking more time steps decreases the error in the energy
law, showing that we can improve the results on the bound, un − uhn . On the
other hand, if only one time step is taken, the convergence slightly worsens
for small τ , which is consistent with the constants found in Theorem 1 and
Corollary 1.
ENGLAND: A. D. 1899.
Dealings with anti-missionary demonstrations in China.
{211}
ENGLAND: A. D. 1899-1900.
Renewed investigation of the Old-Age Pension question.
(4) Has not received poor relief, other than medical relief,
unless under circumstances of a wholly exceptional character,
during twenty years prior to the application for a pension;
(6) Has not an income from any source of more than 10s. a
week; and
Total deductions
1,360,700
In round figures.
£10,300,000
ENGLAND: A. D. 1899-1901.
The Newfoundland French Shore question.
ENGLAND: A. D. 1900.
Industrial combinations.
ENGLAND: A. D. 1900.
Naval strength.
{212}
For the fourth time, Lord Salisbury was called to the lead in
government, and formed his Ministry anew, making considerable
changes. He relieved himself of the conduct of Foreign Affairs
(which was transferred to the Marquis of Lansdowne), and took,
with the office of Prime Minister, that of Lord Privy Seal. Mr.
Brodrick, who had been an Under Secretary, succeeded Lord
Lansdowne as Secretary of State for War. Mr. Balfour continued
to be First Lord of the Treasury, and Leader of the House; Mr.
Chamberlain remained in the Colonial Office. Mr. Goschen
retired.
{213}
Among all the tributes to the Queen that were called out by
her death none seem so significant and so fully drawn from
knowledge of what she was in her regal character, as the words
that were spoken by Lord Salisbury in the House of Lords, at
the meeting of Parliament on the Friday following her death.
"My lords." he said, "the late Queen had so many titles to our
admiration that it would occupy an enormous time to glance at
them even perfunctorily; but that on which I think your
lordships should most reflect, and which will chiefly attach
to her character in history, is that, being a constitutional
monarch with restricted powers, she reigned by sheer force of
character, by the lovableness of her disposition, over the
hearts of her subjects, and exercised an influence in moulding
their character and destiny which she could not have done more
if she had bad the most despotic power. She has been a great
instance of government by example, by esteem, by love; and it
will never be forgotten how much she has done for the
elevation of her people, not by the exercise of any
prerogative, not by the giving of any commands, but by the
simple recognition and contemplation of the brilliant
qualities which she has exhibited in her exalted position. My
lords, it may be, perhaps, proper that those who, like noble
lords opposite and myself, have had the opportunity of seeing
the close workings of her character in the discharge of her
duties as Sovereign, should take this opportunity of
testifying to the great admiration she inspired and the great
force which her distinguishing characteristics exercised over
all who came near her.
{214}
The position of a Constitutional Sovereign is not an easy one.
Duties have to be reconciled which sometimes seem far apart.