Professional Documents
Culture Documents
Download textbook Computational Science Iccs 2018 18Th International Conference Wuxi China June 11 13 2018 Proceedings Part Ii Yong Shi ebook all chapter pdf
Download textbook Computational Science Iccs 2018 18Th International Conference Wuxi China June 11 13 2018 Proceedings Part Ii Yong Shi ebook all chapter pdf
https://textbookfull.com/product/computational-science-
iccs-2018-18th-international-conference-wuxi-china-
june-11-13-2018-proceedings-part-iii-yong-shi/
https://textbookfull.com/product/intelligence-science-ii-third-
ifip-tc-12-international-conference-icis-2018-beijing-china-
november-2-5-2018-proceedings-zhongzhi-shi/
https://textbookfull.com/product/computational-science-and-its-
applications-iccsa-2018-18th-international-conference-melbourne-
vic-australia-july-2-5-2018-proceedings-part-iii-osvaldo-gervasi/
https://textbookfull.com/product/computational-science-and-its-
applications-iccsa-2018-18th-international-conference-melbourne-
vic-australia-july-2-5-2018-proceedings-part-v-osvaldo-gervasi/
Runtime Verification 18th International Conference RV
2018 Limassol Cyprus November 10 13 2018 Proceedings
Christian Colombo
https://textbookfull.com/product/runtime-verification-18th-
international-conference-rv-2018-limassol-cyprus-
november-10-13-2018-proceedings-christian-colombo/
https://textbookfull.com/product/theory-of-cryptography-16th-
international-conference-tcc-2018-panaji-india-
november-11-14-2018-proceedings-part-ii-amos-beimel/
https://textbookfull.com/product/neural-information-
processing-25th-international-conference-iconip-2018-siem-reap-
cambodia-december-13-16-2018-proceedings-part-ii-long-cheng/
https://textbookfull.com/product/computational-science-
iccs-2020-20th-international-conference-amsterdam-the-
netherlands-june-3-5-2020-proceedings-part-i-valeria-v-
krzhizhanovskaya/
https://textbookfull.com/product/computational-science-
iccs-2020-20th-international-conference-amsterdam-the-
netherlands-june-3-5-2020-proceedings-part-iv-valeria-v-
Yong Shi · Haohuan Fu · Yingjie Tian
Valeria V. Krzhizhanovskaya
Michael Harold Lees · Jack Dongarra
Peter M. A. Sloot (Eds.)
LNCS 10861
Computational
Science – ICCS 2018
18th International Conference
Wuxi, China, June 11–13, 2018
Proceedings, Part II
123
Lecture Notes in Computer Science 10861
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, Chennai, 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
Yong Shi Haohuan Fu
•
Computational
Science – ICCS 2018
18th International Conference
Wuxi, China, June 11–13, 2018
Proceedings, Part II
123
Editors
Yong Shi Michael Harold Lees
Chinese Academy of Sciences University of Amsterdam
Beijing Amsterdam
China The Netherlands
Haohuan Fu Jack Dongarra
National Supercomputing Center in Wuxi University of Tennessee
Wuxi Knoxville, TN
China USA
Yingjie Tian Peter M. A. Sloot
Chinese Academy of Sciences University of Amsterdam
Beijing Amsterdam
China The Netherlands
Valeria V. Krzhizhanovskaya
University of Amsterdam
Amsterdam
The Netherlands
This Springer imprint is published by the registered company Springer International Publishing AG
part of Springer Nature
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This year we had 405 submissions (180 submissions to the main track and 225 to the
workshops). In the main track, 51 full papers were accepted (28%). In the workshops,
97 full papers (43%). A high acceptance rate in the workshops is explained by the
nature of these thematic sessions, where many experts in a particular field are per-
sonally invited by workshop organizers to participate in their sessions.
ICCS relies strongly on the vital contributions of our workshop organizers to attract
high-quality papers in many subject areas. We would like to thank all committee
members for the main track and workshops for their contribution toward ensuring a
high standard for the accepted papers. We would also like to thank Springer, Elsevier,
Intellegibilis, Beijing Vastitude Technology Co., Ltd. and Inspur for their support.
Finally, we very much appreciate all the local Organizing Committee members for their
hard work to prepare this conference.
We are proud to note that ICCS is an ERA 2010 A-ranked conference series.
Members
Jiming Wu National Supercomputing Center in Wuxi, China
Lingying Wu National Supercomputing Center in Wuxi, China
Jinzhe Yang National Supercomputing Center in Wuxi, China
Bingwei Chen National Supercomputing Center in Wuxi, China
Yuanchun Zheng University of Chinese Academy of Sciences, China
Minglong Lei University of Chinese Academy of Sciences, China
Jia Wu Macquarie University, Australia
Zhengsong Chen University of Chinese Academy of Sciences, China
Limeng Cui University of Chinese Academy of Sciences, China
Jiabin Liu University of Chinese Academy of Sciences, China
Biao Li University of Chinese Academy of Sciences, China
Yunlong Mi University of Chinese Academy of Sciences, China
Wei Dai University of Chinese Academy of Sciences, China
Program Committee
Kernel Extreme Learning Machine for Learning from Label Proportions . . . . 400
Hao Yuan, Bo Wang, and Lingfeng Niu
A Fast Vertex-Swap Operator for the Prize-Collecting Steiner Tree Problem . . . 553
Yi-Fei Ming, Si-Bo Chen, Yong-Quan Chen, and Zhang-Hua Fu
1 Introduction
Large earthquakes often cause severe damage in cut-and-fill land developed for
housing. It is said that earthquake waves are amplified locally by impedance con-
trast between the cut layer and fill layer, which causes damage. To evaluate this
wave amplification, 3D wave propagation analysis with high spatial resolution
considering nonlinearity of soil properties is required. Finite-element methods
(FEM) are suitable for solving problems with complex geometry, and nonlinear
constitutive relations can be implemented. However, large-scale finite-element
analysis is computational expensive to assure convergence of the numerical solu-
tion.
Efficient use of high performance computers is effective for solving this prob-
lem [2,3]. For example, Ichimura et al. [4] developed a fast and scalable 3D
c Springer International Publishing AG, part of Springer Nature 2018
Y. Shi et al. (Eds.): ICCS 2018, LNCS 10861, pp. 3–16, 2018.
https://doi.org/10.1007/978-3-319-93701-4_1
4 A. Yoshiyuki et al.
nonlinear wave propagation analysis method based on nonlinear FEM, and was
selected as a Gordon Bell Prize Finalist in SC14. Here, computational methods
for speeding up the iterative solver was developed, which enabled large-scale
analysis on distributed-shared memory parallel supercomputers such as the K
computer [5]. In this method, a simple nonlinear model (Ramberg-Osgood model
[6] with the Masing rule [7]) was used for the constitutive equation of soils, and
the program was used for estimating earthquake damage at sites with complex
grounds [8]. However, this simple constitutive equation is insufficient for simu-
lating permanent ground displacement; 3D elasto-plastic constitutive equations
are required to conduct reliable nonlinear wave propagation analysis for soft
grounds. On the other hand, existing elasto-plastic nonlinear wave propagation
analysis programs based on nonlinear FEM for seismic response of soils are not
designed for high performance computers, and thus they cannot be used for large
scale analyses.
In this study, we develop a scalable 3D elasto-plastic nonlinear wave prop-
agation analysis method based on the highly efficient FEM solver described in
[4]. Here, we incorporate a standard 3D elasto-plastic constitutive equation for
soft soils (i.e., super-subloading surface Sekiguchi-Ohta EC model [9–11]) into
this FEM solver. The FEM solver is also extended to conduct self-weight analy-
sis, which is essential for conducting elasto-plastic analysis. This enables large-
scale 3D elasto-plastic nonlinear wave propagation analysis, which is required
for assuring numerical convergence when computing seismic response of soft
grounds.
The rest of the paper is organized as follows. In Sect. 2, we describe the
target equation and the developed nonlinear wave propagation analysis method.
In Sect. 3, we verify the method through a convergence test, apply the method
to an actual site, and measure the computational performance of the method.
Section 4 concludes the paper.
2 Methodology
Previous wave propagation analysis based on nonlinear FEM [4] used the
Ramberg-Osgood model and Masing rule for the constitutive equation of soils.
Instead, we apply an elasto-plastic model (super-subloading surface Sekiguchi-
Ohta EC model) to this FEM solver for analyzing large ground deformation. In
elasto-plastic nonlinear wave propagation analysis, we first find an initial stress
state by conducting initial stress analysis considering gravitational forces, and
then conduct nonlinear wave propagation analysis by inputting seismic waves.
Since the previous FEM implementation was not able to carry out initial stress
analysis and nonlinear wave propagation analysis successively, we extended the
solver. In this section, we first describe the target wave propagation problem with
the super-subloading surface Sekiguchi-Ohta EC model, and then we describe the
developed scalable elasto-plastic nonlinear wave propagation analysis method.
Development of Scalable Three-Dimensional Elasto-Plastic Nonlinear Wave 5
with ⎧ n
⎪
⎪q = qn−1 + Kn δun ,
⎪ n
⎨u = un−1 + δun ,
(2)
⎪
⎪vn = −vn−1 + dt
2
δun ,
⎪
⎩ n
a = −an−1 − dt
4 n−1
v + 4 n
dt2 δu .
Cne = α∗ Me + β ∗ Kne ,
current stress state is always on it. We can take into account plastic deforma-
tion in the normal yield surface and reproduce smooth change from elastic state
to plastic state by introducing the subloading surface. On the other hand, the
superloading surface is a yield surface defined outside of the normal yield surface.
It is similar in shape to the normal yield surface and the subloading surface. Rel-
ative contraction of the superloading surface (i.e., the expansion of the normal
yield surface) describes the decay of the structure as plastic deformation pro-
ceeds. At the end, the superloading surface and the normal yield surface become
identical. Similarity ratios of the subloading surface to the superloading surface,
of the normal yield surface to the superloading surface are denoted by R, R∗ ,
respectively (0 < R ≤ 1, 0 < R∗ ≤ 1). 1/R is overconsolidation ratio and R is
the index of degree of structure. As plastic deformation proceeds, the subloading
surface expands and the superloading surface relatively contracts. The expan-
sion speed Ṙ and contraction speed R˙∗ are calculated as in Fig. 1. D, ˙p are the
coefficient of dilatancy, the plastic volumetric strain speed and m, a, b, c are the
degradation parameters of overconsolidated state and structures state, respec-
tively. Using this R and R∗ , a yield function of the subloading surface is described
as f (σ , v p ) in Fig. 1. Here, M, nE , σ , σ0 are the critical state parameter, the
fitting parameter, the effective stress tensor, the effective initial stress tensor
and η ∗ , p , q are the stress parameter proposed by Sekiguchi and Ohta, the effec-
tive mean stress, the deviatoric stress. The following stress-strain relationship is
obtained by solving the simultaneous equations in Fig. 1.
⎛ ⎞
⎜ e Ce : ∂f ⊗ ∂f
: C
e
˙
σ = ⎝C − ∂σ ∂σ
⎟
⎠ : ˙ ,
∂f e : ∂f − ∂f ∂f m (ln R) ∂f ∂f ∂f ∂f
: C ∂v p ∂p
+ ∂σ − a (R∗ )b (1 − R∗ )c ∂R ∗ ∂σ
∂σ ∂σ D ∂R
ep
= C : ˙ , (3)
where,
2
e
Cijkl = K − G δij δkl + G (δik δjl + δil δjk ) ,
3
Λ 3 (1 − 2ν )
K= p , G = K,
M D (1 − Λ) 2 (1 + ν )
Ce (Cijkl
e
), Cep are the elasticity tensor, the elasto-plasticity tensor and
K, G, Λ, ν are the bulk modulus, the shear modulus, the irreversibility ratio,
the effective Poisson’s ratio, respectively.
of Eq. (1) by using adaptive conjugate gradient (CG) method with multi-grid
preconditioning, mixed precision arithmetics, and fast matrix-vector multiplica-
tion based on the Element-by-Element method [12,13]. Instead of storing a fixed
preconditioning matrix, the preconditioning equation is solved roughly using an
another CG solver. In Algorithm 1, outer loop means the iterative calculation of
the CG method solving Ax = b, and the inner loop means the computation of
preconditioning equation (solving z = A−1 r by CG method). Since the precondi-
tioning equation needs only be solved roughly, single-precision arithmetic is used
in the preconditioner, while double precision arithmetic is used in the outer loop.
Furthermore, the multi-grid method is used in the preconditioner to improve
convergence in the inner loop itself. Here, a two-step grid with second-order
tetrahedral mesh (FEMmodel) and first-order tetrahedral mesh (FEMmodelc )
is used. Specifically, an initial solution of z = A−1 r is estimated by computing
zc = Ac −1 rc , which reduces the number of iterations in solving z = A−1 r. In
order to reduce memory footprint, memory transfer sizes, and improve load bal-
ance, a matrix-free method is used to compute matrix-vector products instead
of storing the global matrix on memory. This algorithm is implemented using
MPI/OpenMP for computation on distributed-shared memory computers.
We enable initial stress analysis and nonlinear wave propagation analysis
successively by changing the right hand side of Eq. (1). The calculation algorithm
for each time step of the elasto-plastic nonlinear wave propagation analysis is
shown in Algorithm 2. Here, the same algorithm is used for both the initial
stress analysis and the wave propagation analysis. In the following, we describe
initial stress analysis and nonlinear wave propagation analysis after initial stress
analysis.
In this study, we use self-weight analysis as initial stress analysis. Gravity is
considered by calculating the external force vector in Eq. (1) as
fn = fn + ρgNdV, (4)
8 A. Yoshiyuki et al.
where ρ, g, and N are density, gravitational acceleration and the shape func-
tion, respectively. We apply the Dirichlet boundary condition by fixing vertical
displacement at bottom nodes of the model.
During nonlinear wave propagation analysis, waves are inputted from the
bottom of the model. Thus, instead of using Dirichlet boundary conditions at
Development of Scalable Three-Dimensional Elasto-Plastic Nonlinear Wave 9
the bottom of the model, we balance gravitational forces by adding reaction force
to the bottom of the model obtained at the last step of initial stress analysis
(step t0 ). Here, the reaction force
− f t0 + qt0 −1 , (5)
is added to the bottom nodes of the model in Eq. (1). Here, f n is calculated as
in Eq. (4).
3 Numerical Experiments
3.1 Verification of Proposed Method
As we cannot obtain analytical solutions for elasto-plastic nonlinear wave propa-
gation analysis, we cannot verify the developed program by comparing numerical
solutions with analytical solutions. However, we can compare 1D numerical anal-
ysis results with the same elasto-plastic constitutive models with 3D numerical
analysis results on a horizontally stratified soil structure to verify the consistency
between the 1D and 3D analyses as well as the numerical convergence with fine
discretization of the analyses. As we use the results of the 1D analysis (stress and
velocity) with the same elasto-plastic models as the boundary condition at base
and side faces of the 3D model for 3D analyses, we can check the consistency
between the 3D and 1D analyses and their numerical convergence by checking
the uniformity of 3D analysis results in the x − y plane.
10 A. Yoshiyuki et al.
quake in 1995 (Fig. 3, Kobe wave). We pull back this wave to the bedrock and
input it to the bottom of the 3D model. Since the major components of the
response is influenced by waves below 2.5 Hz, we conduct analysis targeting fre-
quency range between 0.1 and 2.5 Hz. We first conduct self-weight analysis with
dt = 0.001 s × 700,000 time steps, and then conduct nonlinear wave propagation
analysis with dt = 0.001 s × 40,000 time steps using the Kobe wave. Instead of
loading the full gravitational force at the initial step, we increased the gravi-
tational force by 0.000002 times every time step until 500,000 time steps for
both the 1D and 3D analyses. For the 3D analysis, we used the Oakleaf-FX
system at the University of Tokyo consisting of 4,800 computing nodes each
with single 16 core SPARC64 IXfx CPUs (Fujitsu’s PRIMEHPC FX10 mas-
sively parallel supercomputer with a peak performance of 1.13 PFLOPS). For
the model with minimum element size of 1 m, the degrees-of-freedom was 85,839,
and the 3D analysis took 20,619 s using 576 CPU cores (72 MPI processes × 8
OpenMP threads). For the model with minimum element size of 2 m, the degrees-
of-freedom was 14,427, and the 3D analysis took 12,278 s by using 64 CPU cores
(8 MPI processes × 8 OpenMP threads).
Results of the 1D and 3D analyses are shown in Figs. 4 and 5. From Fig. 4, we
can see that the time history of displacement on ground surface for each analysis
are almost identical. Figure 5 shows the displacement distribution at surface
of the 3D analysis. We can see that the difference of displacement values at
each point is converged within about 0.75%. Although not shown, the maximum
difference was about 2% for the case with element size of 2 m. We can see that
the 3D analysis results converge to the 1D analysis results by using sufficiently
small elements (in this case, 1 m elements).
Another random document with
no related content on Scribd:
¹⁰And Cush begat Nimrod: he began to be a
mighty one in the earth.
10. And Cush begat Nimrod] From the parallel passages in
Genesis (x. 10, 11) it is apparent that Nimrod is the name of an
individual, the traditional founder of the Babylonian-Assyrian Empire.
As Cush is here called the father of Nimrod and in verse 8 is the son
of Ham, Hebrew tradition would appear to have regarded Hamites as
the founders of the Babylonian power. Possibly the Redactor of
Genesis who combined these verses which belong to the tradition of
J with verses 5‒9 which are from “P” may have thought so. But in the
independent “J” narrative it is very probable that Cush, father of
Nimrod, represents the third or Kassite dynasty (Κοσσαῖοι) which
held sway in Babylon from about 1750‒1200 b.c. Even so, the
identification of Nimrod himself remains a puzzle, and it is not yet
possible to say whether he is a legendary or an historical character,
or partly both.
began to be a mighty one in the earth] i.e. was the first grand
monarch (for the idiom, compare Genesis ix. 20). In Genesis x. 9, he
is further and quaintly described as “a mighty hunter before the
Lord.”
Zidon his firstborn] From the time of David downwards Tyre takes
precedence of Zidon in any mention of the Phoenician cities in the
Old Testament, but Zidon was the older of the two cities, as is here
implied and as the Roman historian Justin (xviii. 3) asserts. So we
find the Phoenicians in the earlier books of the Old Testament called
Zidonians, not Tyrians (e.g. Judges iii. 3; 1 Kings v. 6). Homer also
refers not to Tyre but to Zidon.
Heth] i.e. the Hittites, a northern non-Semitic race, who from
about 1800‒700 b.c. were a great power, extending over part of Asia
Minor and northern Syria from the Orontes to the Euphrates. The
references to them in the Old Testament make it probable that Hittite
settlements were to be found in various parts of Palestine. This fact
and their dominant influence, circa 1300 b.c., throughout Canaan
and Phoenicia probably accounts for their inclusion as a “son” of
Canaan.
the Amorite] compare Numbers xiii. 29, xxi. 21; Judges i. 35. The
name (probably a racial one) was frequently used of the pre-
Israelitish inhabitants of Canaan (“Canaanites” being the
geographical description). In a more restricted sense it was used to
denote the people of Sihon, east of the Jordan.
Elam] is the name of a land and nation north of the Persian Gulf
and east of Babylonia, and is often referred to in the Old Testament.
Though settled by Semites at a very early date, it was subsequently,
circa 2280 b.c., possessed by a non-Semitic race, who even
extended their power over Babylonia itself. The inclusion of Elam
among the Semites is doubtless due to its proximity to Asshur, and,
though not strictly correct, is very natural.
Lud] the name suggests the Lydians, but how this non-Semitic
people situated on the west coast of Asia Minor comes to be
included with Asshur and Aram as a son of Shem is a mystery.
Possibly therefore a Semitic region, called Lubdu, between Tigris
and Euphrates is meant.
20. Joktan begat Almodad] All the names of the sons of Joktan
here given, so far as they have been identified, represent peoples
situated in south Arabia or on the west coast of the Red Sea lying
over against south Arabia. The only familiar name is that of the
unidentified Ophir, which possibly but not certainly may be the “El
Dorado” to which Solomon sent his fleet for gold.
24‒27.
The Descent of Abraham from Shem.
Hadad] The name begins with the Hebrew letter Ḥēth and
therefore differs from the Hadad of verse 46 and of verse 50 and of 2
Chronicles xvi. 2 in which the first letter is Hē, a softer guttural than
Ḥeth.
36. Teman] Amos i. 11, 12; Habakkuk iii. 13. The word means
South, and is applied in the first passage to Edom itself, in the
second to the wilderness of Edom, both being south of Canaan.
before ... Israel] i.e. before Saul; or possibly “before David,” if the
phrase means before the reign of the first Israelitish king over Edom.
For the use made of this statement in the discussion of the date of
the Hexateuch, see Chapman, Introduction to the Pentateuch, p. 40,
in this series.
Bela the son of Beor] possibly the same as the familiar Balaam
son of Beor, the consonants of the names differing in Hebrew only by
the final m. See, however, Gray, Numbers (International Critical
Commentary), pp. 315, 324.
Chapter II.
1, 2 (compare Genesis xxxv. 22b‒26).
The Sons of Israel.
3‒17.
Descendants of Judah to the Sons of Jesse.
Zimri] LXX. (B) Ζαμβρεί (β being merely euphonic) here and also
Joshua vii. 1 where Hebrew has “Zabdi.” LXX. is probably right in
identifying the two. Either form might arise from the other by easy
textual corruption.
Ethan ... Dara] Read, Darda with Vulgate, Targum, Peshitṭa The
same four names in the same order occur 1 Kings iv. 31 as the
names of wise men whom Solomon surpassed in wisdom. They are
there called sons of “Mahol” who may have been either a nearer or
remoter ancestor than Zerah. Ethan however is there called the
Ezrahite (= probably “son of Zerah”). [Psalms lxxxviii., lxxxix. bear
respectively the names “Heman the Ezrahite,” “Ethan the Ezrahite,”
but these (it seems) were Levites (compare xv. 17, 19, where see
note).]
16. sons of Zeruiah] Joab and his brothers are always thus
named after their mother; perhaps their father died while they were
young, or we may have a relic here of the ancient method of tracing
kinship through the mother.