As a high efficiency and energy saving engine, gas turbine engines are widely used in modern aircraft and power plant.
After burning all the fuel mixture inside the combustion chamber, gas with high temperature and pressure will be exhausted from the combustion chamber to the turbine stage, the method which gas turbine engine use to generate power is using the hot gas to drive the turbine blades in the turbine stage inside the engine, as a consequence turbine blades have to withstand high temperature and stress which is harmful to the turbine structure.
In order to protect the turbine blades inside the engine, effective cooling techniques are applied on the turbine blades, among all those cooling techniques film cooling is considered as the most necessary cooling technique.

During an operation, an application runtime occasionally becomes longer or shorter than previously measured runtime despite running under the same conditions (e.g., with the same load module and input parameters). We call this issue "runtime fluctuation".

The runtime fluctuations disturb efficient operations of the K computer. If runtime exceeds the maximum elapsed time given by a job script, the computational resources will be uselessly consumed due to the incomplete termination of the program. And if runtime is getting short, the job running order should be disordered and rearranged by the job scheduler.

Therefore, Operations and Computer Technologies Division continued to investigate these issues and resolve them. In this talk, I will show some cases of the runtime fluctuations and solutions.

The parallel-in-time integration (PinT) is desired for exploiting effectively the concurrency of the post peta-scale parallel computer which increases the concurrency still more. The progress of PinT study was　slow although the importance of PinT was paid attention to at first time by Nievergelt 50 years ago. However, the study of PinT advances rapidly after the Parareal method was proposed by Lions in 2001. The Parareal method too had many problems. But, direction of the solution to the problems is becoming clear in these past ten years. In addition, the study from the different fields such as a multi-scale, chaos, the optimization etc. becomes active. Therefore the research and development of the PinT is advancing rapidly now. In this presentation, I would like to explain a summary of PinT method from the viewpoints of research and development trend, application field, techniques to be required and prospects of the future development.

In recent years, a variational methods called tensor network methods have been widely applied to obtaining partition functions of classical statistical models or low-energy states of quantum many-body systems. One of the best example is the success of the density matrix renormalization group for many-body systems. In this presentation, I would like to introduce some features of typical tensor network methods and discuss potential bottleneck points in large-scale parallel computations of the tensor network methods.

Scientific data nowadays can contain various physical quantities, attracting researchers and engineers. Although multivariate high resolution simulation has become popular thanks to HPC architectures, even experts struggle to analyze and visualize the complex correlations between different quantities in their data. Our new data analysis web app, fiber surface GUI, eases such analysis for multivariate 3D data. You merely need to specify the region of interest in the familiar scatterplot in your web browser, and the server running in an interactive HPC environment extracts the distribution in the 3D space as 2D surfaces, as known as fiber surface, just as one would extract isosurfaces from 3D scalar data. Hence, importantly, an affordable computer with a weak performance can visualize and even analyze the output - fiber surface is smaller than the original 3D data for an order of magnitude when we compare the data size. In the talk, I introduce fiber surface, its GUI, and HIVE. HIVE is our visualization framework that systematizes your visualization workflow combining supercomputers and visualization-oriented clusters. I demonstrate how you can access the fiber surface GUI and HIVE for your research tasks. I will highly welcome assistance requests for your visualization tasks, as well as offers for new directions.

On the verge of the convergence between high performance computing (HPC) and Big Data processing, it has become increasingly prevalent to deploy large-scale data analytics workloads on high-end supercomputers. Such applications often come in the form of complex work- flows with various different components, assimilating data from scientific simulations as well as from measurements streamed from sensor net- works, such as radars and satellites. For example, as part of the next generation flagship (post-K) supercomputer project of Japan, RIKEN is investigating the feasibility of a highly accurate weather forecasting system that would provide a real-time outlook for severe guerrilla rainstorms. One of the main performance bottlenecks of this application is the lack of efficient communication among workflow components, which currently takes place over the parallel file system. This presentation reports an initial study of a direct communication framework designed for complex workflows that eliminates unnecessary file I/O among components. Specifically, we propose an I/O arbitrator layer that provides direct parallel data transfer among job components that rely on the netCDF interface for performing I/O operations, with only minimal modifications to application code. We present the design and a preliminary evaluation of the framework on the K Computer using RIKEN’s experimental weather forecasting workflow as a case study.

The aim of this presentation is to understand the recent progress in photochemistry, i.e., molecular processes on the electronically excited states, by using the ab initio electronic structure methods. I focus on the role of potential-energy-surface crossings that induce the nonadiabatic transition between multiple electronic states and on the resultant ultrafast relaxation to the ground state.

The solvers for dense eigenvalue problems are widely used in computational science applications. In the first half of this talk, we will briefly introduce techniques used in the solvers. In the latter half, some topics related to generalized eigensolvers will be presented.

In this talk, we will introduce the key one-sided communication calls in MPI-3, and present the performance analysis of a prototype one-sided communication implementation in FX100.

We are exploring new algorithms to construct biomolecular structures from coherent diffraction patterns of a single particle observed by X-ray free electron laser (XFEL). Coherent diffraction patterns contain useful information on bimolecular conformations. However, the　actual intensity on coherent diffraction pattern from a single particle is very weak. We need to extract useful signal from noisy data. I pay attention to develop and improve an algorithm to detect similarity between pairs of coherent diffraction patterns from an orientation-unknown single particle to extract useful information from noisy data [1,2].
This similarity detection algorithm for noisy diffraction patterns is extremely useful for single particle imaging. For example, this detection is an important step towards classification of a large number of experimental coherent diffraction patterns for reconstructing three dimensional structure of the molecule [1]. For another example, the detection algorithm plays an important role in a hybrid method using computational algorithms to generate hypothetical structural models for searching best agreement with limited experimental data [3].
Improvement of the similarity detection algorithm allows us to reduce the incident beam intensity, thus enabling to alleviate the radiation damage of sample during data acquisition [4,5].
In the presentation, I would like to talk about the possibility of single particle diffraction imaging using XFEL via attainable resolution as a function of the necessary incident beam intensity according to improved similarity detection algorithm.

[1] Tokuhisa, A., Taka, J., Kono, H., Go, N., 2012. Classifying and assembling two-dimensional X-ray laser diffraction patterns of a single particle to reconstruct the three-dimensional diffraction intensity function: Resolution limit due to the quantum noise. Acta Crystallogr. Sect. A Found. Crystallogr. 68, 366–381.
[2] Tokuhisa, A., Arai, J., Joti, Y., Ohno, Y., Kameyama, T., Yamamoto, K., Hatanaka, M., Gerofi, B., Shimada, A., Kurokawa, M., Shoji, F., Okada, K., Sugimoto, T., Yamaga, M., Tanaka, R., Yokokawa, M., Hori, A., Ishikawa, Y., Hatsui, T., Go, N., 2013. High-speed classification of coherent X-ray diffraction patterns on the K computer for high-resolution single biomolecule imaging. J. Synchrotron Radiat. 20, 899–904.
[3] Tokuhisa, A., Jonic, S., Tama, F., & Miyashita, O. 2016. Hybrid Approach for Structural Modeling of Biological Systems from X-ray Free Electron Laser Diffraction Patterns. Journal of structural biology.(in press)
[4] Kai, T., Tokuhisa, A., Moribayashi, K., Fukuda, Y., Kono, H., & Go, N. (2014). Intensity of Diffracted X-rays from Biomolecules with Radiation Damage Caused by Strong X-ray Pulses. Journal of the Physical Society of Japan, 83(9), 094301.
[5] Kai, T., Tokuhisa, A., & Kono, H. 2013. Calculation of Molecular-Structure-Based Damage Caused by Short-Pulse High-Intensity X-ray Lasers. Journal of the Physical Society of Japan, 82(11), 114301.