=Paper=
{{Paper
|id=Vol-1787/552-556-paper-96
|storemode=property
|title=From parallel to distributed computing as application to simulate magnetic properties-structure relationship for new nanomagnetic materials
|pdfUrl=https://ceur-ws.org/Vol-1787/552-556-paper-96.pdf
|volume=Vol-1787
|authors=Sergiu Mohorianu,Mihail Liviu Craus
}}
==From parallel to distributed computing as application to simulate magnetic properties-structure relationship for new nanomagnetic materials==
https://ceur-ws.org/Vol-1787/552-556-paper-96.pdf
From parallel to distributed computing as application to
simulate magnetic properties-structure relationship for
new nanomagnetic materials
S. Mohorianu1,a, M.-L. Craus1,2,b
1
National Institute of Research and Development for Technical Physics, Romania,
47Mangeron Boulevard, Iasi, Ro-700050, Romania
2
IUCN, Frank Laboratory for Neutron Physics (FLNP)
6, Joliot Curie, Dubna, 141980, Russia
E-mail: a sergium@phys-iasi.ro; b craus@phys-iasi.ro, kraus@nf.jinr.ru
Modern materials science is based in principle on the fundamental experience that the properties of materi-
als are not peremptorily determined by their average chemical composition but they are to a large extent influ-
enced by the distances between atoms and characteristics of their bonds, implicitly by their microstructure. Now,
it is obvious that the outstanding success of magnetic materials for the last two decades may be ascribed to three
relevant accomplishments: -overall improvements in general expertise and techniques in sample synthesis; -a
dramatic refinement and development of new methods and probes for magnetic materials characterization; -the
increasing importance of nano-level studies that led to the ingenious ways of producing nanoparticle samples,
new techniques for element specific studies, goin down to atomic resolution studies and even to single atoms at
surfaces and interfaces. In the last projects completed in recent years we have analyzed and studied magnetic
materials mostly micro and nano scale ferro-, ferrimagnetic and ferroic perovskites, enumerating here
(nano)cobaltites, (nano)manganites and other nanomagnetic materials. Almost all of them, listed up require mas-
sive data processing. At that time, it became obvious to us that it needs another embodiment, namely in the pro-
cessing data activity. In 2010 we introduce parallel computing applications on the simulation of the structure,
magnetic and transport properties to explain the structure-properties relationships for some new nanomagnetic
materials. Knowing quite substantial intersection of the parallel computing and distributed, we think it is of
common sense to introduce our applicative work in magnetism and magnetic materials science modeling proper-
ties, in the context of distributed computing applications. Our latest research specializes in improving techniques
for high-level simulation in the design of nano-materials with controlled magnetic properties. We used a package
built on Linux, called Nmag (with acquiescence) on an open source platform, across a network of parallel com-
puters.
Keywords: computational material science, magnetic properties-structure relationship, parallel and distrib-
uted technologies
The work was supported by: Investigations of Nanosystems and Novel Materials by Neutron Scattering Methods (Project
No. 04-4-1121-2015/2017)
© 2016 M.L. Craus, S. Mohorianu
552
Introduction. Computational materials science - an interdiscipli-nary
field that demands parallel and distributed computing technologies
Mathematical modeling aims to describe through mathematics the different aspects of the real
world, their dynamics and their interaction. Numerical simulation provides accurate and certified solu-
tions to complex mathematical models by means of scientific computing. Modeling and numerical
simulation have become the road-map for mathematics to develop and analyze novel techniques to
solve problems in basic sciences. New science called around 2000 by D. Raabe as computational ma-
terials science [Raabe, 1998], part of a much larger branches, the computational sciences applied, re-
quires: (1) massive data calculations with (2) complex algorithms and (3) sophisticated methods.
There are many different ways to represent knowledge in a computational framework. The builder of
computational models needs an awareness of both: the range and the scope of representational choice.
New materials and structures by design is now a modern concept that grows, to be particularly im-
portant for the material sciences and engineering. The outstanding success of magnetic materials for
the last two decades may be ascribed to three relevant accomplishments in the last 15-20 years: i)
overall improvements in general expertise and techniques in sample synthesis; ii) a dramatic refine-
ment and development of new methods and probes for magnetic materials characterization; iii) the
increasing importance of nano-level studies, which has led to the ingenious ways of producing nano-
particle samples and the advent of powerful new techniques for element specific studies, layer-by-
layer and even leading to atomic resolution, suitable for investigating magnetism of nanoclusters, bi-
and trilayer systems and even single atoms at surfaces and/or interfaces. The success at the nano-level
has opened-up new frontiers: - magnetism of nanoclusters; - surface and interface magnetism; - low
dimensional magnetism; - interacting nanostructure magnetism, and others. In terms of theoretical
research the rapid development of theoretical methods based on congruity of computer simulation has
allowed the calculation of increasingly complex details of the formation and interaction of magnetic
moments and has permitted a refined interpretation of experimental data. These continuing work be-
gun in 2005’s by works on some nanomagnetic materials. The new aspects studied in the last year, our
latest research, specialize in improving techniques for high-level simulation in the design of nano-
materials with controlled magnetic properties. New achievements and performance of computing re-
sources have allowed the use of parallel algorithms, distributed calculation appearance in our laborato-
ries. As an interdisciplinary field to simulate the magnetic properties-structure relationship for new
magnetic materials that requires advanced knowledge in several areas was not available to everyone
from the beginning. The starting point of a computer simulation is the development of an idealized
model of a physical system of interest. Theoretical studies to introduce parallel computing in simula-
tion of processes and phenomena encountered in the design of magnetic materials have been made yet
since the 2000’s. Analyzing the actual condition and the capacity of the laboratory, we started to build
a parallel computational system only after years 2010 [Mohorianu, Craus, 2010]. So, in the late years
we conducted an analysis of endowment and human resource capacity and we got to the conclusion
that we can achieve a laboratory infrastructure to support a system of parallel computing for simula-
tion and design of magnetic materials [Mohorianu, 2013]. The theoretical modeling for the magnetic
properties of nanomaterials turns out to be a quite difficult task. That is because it involves the study
of a lot of variables and some apparently non-connected phenomenon that may happen. According
with the complexity of the problem we need to create an appropriate theoretical instrument in order to
explain and to predict how the nature works in our particular problem. It is already well known that
magnetism is a very complex and intriguing phenomenon. Early experiments to elucidate magnetic
phenomena and magnetic materials behavior were based on the measurement of forces and torques
exerted on “samples” placed into magnetic fields produced by current flow through wires. Our mod-
ern understanding of electronic structure is based on the concepts of charge and spin. The key to this
development was the understanding of atomic spectra by means of quantum theory concepts of ex-
change and spin–orbit coupling. Most of the common theory is based on these observations. We try to
553
correlate all these theories and try to exploit some of the newest IT instruments to generate “recipes”
for the experimentalist to produce faster and cheaper magnetic nanomaterials. It seems that today’s
magnetic materials are not only the bulk materials, but atomic engineered wires,, particles or thin film
and multi layer structures that often have one, two, or three dimensions on the nanometric scale as we
show in figure 1. The massive growth of magnetic technologies is due to scientific and technological
developments in four key areas: a) the development of new magnetic materials; b) the progress in the-
oretical developments; c) the developments of new experimental techniques and d) the developments
in simulation techniques.. As things are going now modeling and simulation techniques can be used in
all stages in the development
pment and improvement of new magnetic materials, from the initial formation
of concepts to synthesis and characterization of properties. In this article we describe how a Simulation
and Design Method (SDM) attempt, based on our last results is applied on ssome new type
nanomaterials,
materials, demonstrating the adaptation of parallel and distributed computing simulators. Our new
simulation and design methods allow that by searching on the correlation for the most recent material
properties to enhance the actually known
known magnetic properties. Our IT simulation results are now pr pro-
posed to be checked in labs with the experimental data and we expect a good agreement.
Fig. 1. Nano-materials
materials and nano-structures.
nano structures. Dominant dimension classification
Applied problem. Experimental and Theoretical App
proach
Currently nanomagnetism research involves investigating the basic magneetic, magneto-optical,
galvano-magnetic, magneto-transsport phenomena associated with reduced dimenssionality. The idea of
extracting valuable information from data is not new. It is new distributed com mputer processing and
data storage technologies, which h allow gigabytes, even terabytes of data to remaain on-line, available
for processing by client/serveer applications. It’s new as well, creating algorithms to study
micromagnetism, parallelizing th hese algorithms, the use of some Artificial Neuural Network applica-
tions and the development of ad dvanced algorithms for knowledge discovery. The parallelization of
some of these algorithms is onee of our last attempts. The challenges of simulaating the relationship
structure-magnetic properties off nanomagnetic materials appeared lately. Simuulation and design in
case of magnetic material struccture was started for some perovskites like maaterials. We studied:
cobaltites oxide with perovskite structure, multiferroics materials of type ReTO3 and ReT2O5 (Re-rare
earth, T- transition metal), mangganites type La1-xRexMnO3. As a result a variety of magnetic different
behavior appears in response to external conditions: temperature, chemical dopinng, magnetic or elec-
tric fields, pressure. The NMAG [NMag package] MPI package was implementedd in the laboratory in
order to simulate of new magnetiic materials in bulk presentation, micro or nano magnetic: perovskites
generic CaTiO3, enumerating h here cobaltites and manganites as nanoferrim magnetics, nanomulti-
ferroics and nanoferromagnetic materials. We construct a database (DB) with manny of these materials.
The basic idea is to obtain a fine correlation between the material structure and/orr composition and the
magnetic material properties ussing either ANN or some micromagnetics soft ass IT instrument. The
analyst has a near-infinite num mber of approaches that can be taken in the course of a simula-
tion/numerical experiment. The approach that is used will depend on 1) the kind of material to be ana-
lyzed, 2) the form of the materiial, 3) the problem type that is required to be solved, 4) the experi-
mental or instrumental techniquue that can be employed, and 5) the known limiitations of the instru-
mental / simulation method. Naanometer-sized magnetic ‘object’ are theoretically placed at the limit
between classical and quantum magnetism so, detecting their magnetic propertiies is technologically
very challenging and from the co omputational point of view a very difficult task. Thus, we experimen-
554
tally implemented a Parallel Computational System (PCS) working with som me academic software
packages that were adapted to our current modelling problems. Accumulation based on logistics and
scientific development in recent years allowed the creation and development of some modern compu-
tational strategies in our laborato
ory.
Method. Paradigms
Our latest research specializzes in improving techniques for high-level simulaation programs in the
design of nano-materials with coontrolled magnetic properties. We have tried adaapting across multiple
platforms, several programs: OO OMMF (finite differences), Magpar (finite elemeents) and other codes
(Mumax, MicroMagnum). We prresented here as an example what we use in som me work constructing
the parallel computing network. The package built on Linux (Python language), called NMag (finite
elements) used. This is on an opeen source platform, working across a network of parallel computers or
Processor Elements (PE). Nmag‘s numerical core as part of the nsim (multi-physsics library) has been
designed to carry out numericall computation on several CPUs simultaneously.. The protocol is the
wide spread Message Passing Interface (MPI). The best ones in this appliication are probably
MPICH1. We give data from thee package builder on a test for a fairly small systeem: 4114 mesh nodes,
1522 surface nodes, BEM size 18MB. NMag uses the hybrid finite element methhod/boundary element
method to compute the demagneetization field. The HLib library is available for academic use, and we
ort parallel execution. It is thus stored on the masster node, and cannot
have installed it, does not suppo
be distributed over several nodees. Simulations using the Hlib library can use MPI yet, which means
parallelized execution. The meshhes used in micromagnetic simulations were obtaiined with NetGen and
usually represent idealized geommetries [NetGen], for example a nano-wire mightt be modeled using a
completely smooth cuboid mesh..
Results
The present work gives so ome few milestones on how to design new nannomagnetic materials
(wires), using simulation methodds strictly correlated with the experimental validaation. We start from a
simple and suitable geometry off the experimental data. So, we considered for the first round of the
simulation process a nanomagneetic wire whose magnetic properties have been sppecified. We enter the
specific length values with a 50 nm radius and 1000 nm in length. This calibratioon was set to work on
the simulation tool. Netgen prooduces a mesh network shown in figure 2, with (P)=703 points;
(E)=1940 elements; (SE)=1256 SurfacesElements in this case. Same geometry was studied for differ-
ent wire symmetries, once takingg Ox axis of symmetry then taking Oz. Differencces occur for different
symmetries. Thus, taking Oz as axis we have: P=695; E=1902; SE=1252. In an another step we built
a new configuration of two, six and twelve were built to simulate the interactioons between wires in-
cluding short range magnetic intteractions, and exchange coupling, see figure 3.
Fig. 2 Netgen mesh for a nanowire Fig. 3. Magnetic interaction
teraction nanowire prepared
555
So, we were expected some differences in the simulation results. Part of this work seek to esta
estab-
lish the necessary parameters for the theoretical model, capable to explain and later to obtain in expe
exper-
imental laboratory help to give new magnetic nanomagnetic wires with predicted properties: Curie and
transition temperature variation with concentration and implicit the structure dependence, or magnetic
hysteresis.. We give here as example what NMag computes es the hysteresis loop shown in figure 4, for a
wires, for this magnetic ‘objects’ with K1 (magnetic uniaxial anisotropy) being (a) 0,
series of nano-wires,
(b) 10 and (c) 100 (J/m^3). Nmag uses the hybrid finite element method/boundary element method
(hybrid FEM/BEM)) to compute the demagnetisation field. For our parallel network we obtain for 4
PE’s a relatively close score of 2.03 SS (Simulation – Speedup) that agree the Amdahl’s lowlow. The gen-
uine progress on the computational methods, on the experimental IT-based
IT ed technique, dedicated to the
magnetic materials simulation, now allow us a better understanding of our experimental data.
a) b) c)
Fig. 4. The computed hysteresis loop (magnetic nanowires with K1 a) zero; b) 10; c) 100 (J/m^3))
Conclusions
Low dimensional magneticc systems, such as thin films, wires, multilayers,, and surfaces exhibit
many scientifically interesting and technologically useful properties. Modeling has now a very im-
portant position in the developmeent and improvement of new materials for appliccations. Modeling and
simulation techniques affect all stages in the development and improvement of new materials, from
the initial formation of concepts to synthesis and characterization of material propperties. We have been
applied successfully in the identiification and classification of some nanomagneticc characteristics from
a large amount of data. These methods prove themselves to be better candidatess for the discovery of
new aspects between structure and properties of magnetic nanomaterials. Ouur extensive research
practice in the properties study of magnetic materials, closely depending on new theories and theoreti-
cal models describing their prop perties led us to an intensive use of the numericcal experiment in our
research. Our methods based on n numerical experiment and theoretical modelinng involves checking
with the experimental data and d validation stage. Simulations in magnetic matterial usually require
large computing power in orderr to produce realistic results in an acceptable exxecution time. Today
parallel computation becomes a standard way to achieve this, for new complex and realistic simula-
tion programs. Regular algorithm ms can be realized as parallel programs in a straaightforward way but
when especially more complex algorithms are involved some more effort is needded to exploit the spe-
cific algorithms and the parallel hardware platform to get fast and efficient runs.
References
Raabe D. Computational Material Science, The simulation of materials microstructures and properties
// Wiley-VCH, 1998.
Mohorianu S., Craus M.-L. Perovskites-like
Perovskites like magnetic materials properties prediction by innovative
computational simulation IT-based
IT techniques // BPI, sec. Mat.Mec.Fiz. 2010. Vol. II. Iasi,
Romania.
Mohorianu S. New Landmark In The Study Of Geometry And Parallel Computation Applied To Na-
no-Wires Magnetic Properties // Proc. CNFA 2013. The 5-th Nat. Conf. of Appl . Physics, 23-24
May 2013. – Iaşi, Romania.
NMag package [Electronic resource] : http://nmag.soton.ac.uk .
Netgen [Electronic resource]: http://www.hpfem.jku.at/netgen/.
556