Atomistic modelling of deformation and failure mechanisms in nanostructured materials

Over thepast twodecades, there has been a whirlwind of worldwide research activities on nanostructured materials, with the long-term promise to tailor-design material properties from the nanoscale and up. In this grand effort, computation is playing an increasingly important role in complementing experiments to unravel fundamental principles that underliemechanical behaviours andproperties (e.g. deformation and failure) of materials with a characteristic length scale below 100 nm. For such materials, atomistic modelling is an especially suitable tool, in combination with other mesoscopic and continuum approaches, for identifying the dominant deformation and failure mechanisms. A comprehensive review on the atomistic modelling and relevant multiscale approaches to material research can be found in the literature [1]. Figure 1 shows a schematic illustration of two typical atomistic modelling strategies, the so-called ab initio molecular dynamics (AIMD) and classical molecular dynamics (CMD) simulations. AIMD provides by far the most accurate description of atomic interactions by considering the electronic structure of materials, and is therefore capable of reliably predicting the energetic, electronic, transport, as well as mechanical properties of nanostructured materials. Unfortunately, at this point of time, AIMD is still an expensive computational tool limited to relatively small systems spanning hundreds of atoms and tens of picoseconds. In comparison, CMD simulations can deal with systems containing tens of millions of atoms subject to nearly instantaneous (up to a few nanoseconds) loading and deformation. In CMD simulations, an empirical potential is typically used to describe interatomic forces that generally include both bonded forces associatedwith the chemical bonding and non-bonded forces such as the van der Waals and electrostatic forces. In principle, the empirical potential should reflect not only the pairwise but also many-body effects through ad hoc functional approximations obtained from fitting against detailed quantum simulations or experimental measurements of material properties. For example, the recently developed reactive force field is an emerging potential that can model reactive bond order during the formation and breakage of chemical bonds. Due to their atomic-scale resolution and increasingly accurate de-


Atomistic modelling of deformation and failure mechanisms in nanostructured materials
Xiaoyan Li 1, * and Huajian Gao 2, * Over the past two decades, there has been a whirlwind of worldwide research activities on nanostructured materials, with the long-term promise to tailor-design material properties from the nanoscale and up.In this grand effort, computation is playing an increasingly important role in complementing experiments to unravel fundamental principles that underlie mechanical behaviours and properties (e.g.deformation and failure) of materials with a characteristic length scale below 100 nm.For such materials, atomistic modelling is an especially suitable tool, in combination with other mesoscopic and continuum approaches, for identifying the dominant deformation and failure mechanisms.A comprehensive review on the atomistic modelling and relevant multiscale approaches to material research can be found in the literature [1]. Figure 1 shows a schematic illustration of two typical atomistic modelling strategies, the so-called ab initio molecular dynamics (AIMD) and classical molecular dynamics (CMD) simulations.AIMD provides by far the most accurate description of atomic interactions by considering the electronic structure of materials, and is therefore capable of reliably predicting the energetic, electronic, transport, as well as mechanical properties of nanostructured materials.Unfortunately, at this point of time, AIMD is still an expensive computational tool limited to relatively small systems spanning hundreds of atoms and tens of picoseconds.In comparison, CMD simulations can deal with systems containing tens of millions of atoms sub-ject to nearly instantaneous (up to a few nanoseconds) loading and deformation.In CMD simulations, an empirical potential is typically used to describe interatomic forces that generally include both bonded forces associated with the chemical bonding and non-bonded forces such as the van der Waals and electrostatic forces.In principle, the empirical potential should reflect not only the pairwise but also many-body effects through ad hoc functional approximations obtained from fitting against detailed quantum simulations or experimental measurements of material properties.For example, the recently developed reactive force field is an emerging potential that can model reactive bond order during the formation and breakage of chemical bonds.Due to their atomic-scale resolution and increasingly accurate de-

PERSPECTIVES
deformation mechanisms in nanocrystalline metals.Full atomistic simulations on three-dimensional (3D) systems with size up to 100 million atoms [2] revealed that, as the mean grain size decreases, there exists a transition from dislocation-to grain boundary (GB)mediated deformation mechanisms below a critical grain size, giving rise to a softening mechanism called the inverse Hall-Petch effect which is opposite to the traditional Hall-Petch law that the flow strength of material be proportional to the inverse square root of grain size.Quasi-3D MD simulations [3] predicted the occurrence of deformation twinning through continuous dislocation slip in nanocrystalline Al, a phenomenon subsequently observed in experiments via transmission electronic microscopy.Further simulations on different fcc metals [4] demonstrated that the deformation mechanisms (partial/full dislocation slip and twinning) strongly depend on the generalized stacking fault energy.These results from MD simulations outlined a comprehensive deformation mechanism map for nanocrystalline metals.
Over the last decade, nanotwinned metals have emerged as a new class of hierarchically structured materials with high density of nanoscale twins embedded in micron-or submicron-sized grains.Due to the presence of twin boundaries (TBs) with high symmetry and low energy, nanotwinned materials exhibit an exceptional combination of ultra-high strength, good tensile ductility, and superior resistance to fracture, fatigue, and wear.MD simulations [5] coupled with nudged elastic band method revealed the atomic-scale reactions responsible for dislocation slip transfer through a TB, and suggested these reactions as the rate-controlling mechanisms contributing to the hardening and ductility of nanotwinned metals.Ultra-large-scale MD simulations [6] on fully 3D systems led to the discovery that, as the twin thickness is reduced, there exists a deformation mechanism transition from dislocation cutting through TBs to detwinning due to the nucleation and motion of partial dislocations along the twin planes.This find-ing corroborates and explains experimentally measured relation between strength and twin thickness, which exhibits a transition from Hall-Petch strengthening to softening at a critical twin size.Combined with experimental results, the simulations further led to a scaling law between the strength of nanotwinned metals and two characteristic dimensions of the material, the mean grain size and twin thickness.MD simulations also played a critical role in revealing two distinct deformation mechanismsshear localization and detwinning-in nanotwinned nanopillars with orthogonal and slanted TBs, and a twin size controlled ductile-brittle transition [7].These simulations were performed side by side with experiments to reveal the dominant atomic-scale processes that are responsible for deformation and fracture in nanotwinned materials and structures.
There have also been successful examples in using MD simulations to explore deformation and failure mechanisms in low-dimensional materials such as carbon nanotubes (CNTs) and graphene.MD simulations of multiwalled CNTs under pure bending [8] produced atomic-level rippling structures in excellent agreement with experimental observations, and revealed a twisting deformation mode of inner tubes induced by the interlayer lattice mismatch associated with curvature.It was recently discovered through MD simulations of bicrystalline graphene under uniaxial tension [9] that the strength of polycrystalline graphene strongly depends on the detailed arrangement of pentagon-heptagon defects (equivalent to dislocations) along GBs.This was later verified by experiments on nanoindentation of polycrystalline graphene via atomic force microscopy.MD simulations [10,11] were also used to study the fracture behaviours of polycrystalline graphene, with predicted crack propagation paths [10] and fracture toughness [11] in good agreement with those observed/measured experimentally.The relationship between mechanical strength of graphene and various defects (dislocations, pre-cracks, and GBs) should play important roles in guiding practical applications of graphene in advanced nanodevices.
In spite of considerable progresses/advances made in using MD simulations to model deformation and failure mechanisms in nanostructured materials, there is still a big gap in quantitative comparisons between MD simulations and experiments.So far MD simulations are only capable of providing qualitative insights in deformation mechanisms, rather than quantitative predictions of experimental measurements, in most studies.The primary reason for this gap is that MD simulations have the inherent limitations in spatial and temporal scales, with loading rates typically several orders of magnitude higher than those accessible in experiments.It is noted that, although many mechanical properties of materials are sensitive to the loading rate, the high loading rates in MD often do not qualitatively alter the physical nature of deformation.To overcome the intrinsic limitations of MD, one potential solution is to establish multiscale modelling by coupling MD with mesoscopic or continuum methods.Another is to rely on the constantly improving capacity and performance of supercomputers (with Peta FLOPS and even faster speed at present) and the developments of more accurate and efficient algorithms.In addition, the development of interatomic potential for diverse types of complex materials (material genome), especially alloys and complex materials, remains an important direction of research in the near future.These developments will inevitably make atomistic modelling more realistic, predictive and useful in materials research, reduce/bridge the gap between MD simulations and experiments, and move us towards the long-term objective of 'material by design' via computation.
The impact of AI: can a robot get into the University of Tokyo?

Noriko H. Arai
The 'Todai Robot Project (Can a robot get into the University of Tokyo?)' was initiated by the National Institute of Informatics (NII), the Japan's only general academic research institution of informatics, in 2011 as an AI grand challenge.The goal of the project is to create an AI system that answers real questions on university entrance examinations.This paper reports the current status of the project including the underlying technologies we have developed thus far, and the results we obtained from evaluations.
There have been various types of AI challenges in the past: the chess and shogi matches against professional players, the quiz show challenge against human champions.Todai Robot Project is unique mainly in the following two senses.First, it targets the integrated intelligent tasks rather than single ones.Examinees must take tests of eight subjects, two social studies, two sciences, Japanese (including Japanese and Chinese classics), English, and two mathematics.The tasks require not only the development of ground-breaking underlying technologies in various AI research areas but also their interdisciplinary research synthesis.Second, it enables us to compare the performance of the software and numerous well-educated students.University entrance examinations in Eastern Asian countries, including Japan, are known to be quite competitive and they cover various skill areas and fields.More than half a million high school graduates take National Center Test for University Admissions (NCTUA), standardized multiple choice style tests, every year in Japan, and less than top 3% students are allowed to take the second written test specially designed to select entrants of the University of Tokyo.
Our research team developed a testbed [1,2] that utilized the resources taken from the history problems asked in NCTUA, and organized international evaluation tasks at NTCIR-9, 10 and 11.The participants pursued mainly two approaches to solving history questions.One is traditional statistical factoid approach.Kanayama and Miyao [3] manually converted original true/false world history questions to a set of factoid-style questions and achieved an accuracy of 65% on the NCTUA data by employing Watson's factoid QA engine as the backend system.Kano [4] developed domain-independent and language-independent keyword-based system, and achieved an accuracy of 51% on the same data.The other approach is to combine a logical inference engine based on their semantic representation with a statistical classifier.Considering that the accuracy rate of Watson at Jeopardy!Challenge was 69%, it was most likely some deeper language analysis

Figure 1 .
Figure 1.A schematic illustration of atomistic modelling strategies.Representative length and time scales for AIMD and CMD simulations are indicated along the axes.Some typical nanostructured materials (such as fullerene, nanoparticle, nanotube, graphene, and nanocrystalline/nanotwinned metals) are showed as illustrative examples for different characteristic length and time scales accessible in atomistic modelling.