An extended linear algebra library for electronic structure calculation and its optimization for many-core processors
| Project/Area Number |
26286087
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (B)
|
| Allocation Type | Partial Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Research Field |
Computational science
|
|---|
| Research Institution | The University of Electro-Communications |
|---|
Principal Investigator |
Yamamoto Yusaku 電気通信大学, 大学院情報理工学研究科, 教授 (20362288)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
横川 三津夫 神戸大学, 先端融合研究環, 教授 (70358307)
星 健夫 鳥取大学, 工学研究科, 准教授 (80272384)
|
|---|
| Co-Investigator(Renkei-kenkyūsha) |
IMAMURA TOSHIYUKI 理化学研究所, 計算科学研究センター, チームリーダー (60361838)
INADOMI YUICHI 九州大学, 大学院システム情報科学研究院, 研究員 (70437747)
|
|---|
| Research Collaborator |
Vajtersic Marian ザルツブルグ大学, コンピュータサイエンス学科, 教授
|
|---|
| Project Period (FY) |
2014-04-01 – 2018-03-31
|
| Project Status |
Completed (Fiscal Year 2017)
|
| Budget Amount *help |
¥11,440,000 (Direct Cost: ¥8,800,000、Indirect Cost: ¥2,640,000)
Fiscal Year 2016: ¥4,290,000 (Direct Cost: ¥3,300,000、Indirect Cost: ¥990,000)
Fiscal Year 2015: ¥3,250,000 (Direct Cost: ¥2,500,000、Indirect Cost: ¥750,000)
Fiscal Year 2014: ¥3,900,000 (Direct Cost: ¥3,000,000、Indirect Cost: ¥900,000)
|
|---|
| Keywords | 固有値計算 / 連立1次方程式 / 直交化 / 並列計算 / メニーコア / 電子状態計算 / 計算科学 / 確率的アルゴリズム / 並列化 / 非線形固有値問題 / 理論流体力学 / 連立1次方程式解法 / 有機高分子系 / 前処理付き反復法 / 線形計算 / 固有値 / 行列関数 / 通信回避型アルゴリズム / 電子波ダイナミクス / 固有値問題 |
|---|
| Outline of Final Research Achievements |
In this project, we aimed at developing solvers for numerical linear algebra functions used in electronic structure calculations. The main achievements of this project are as follows. (1) We performed an error analysis of the CholeskyQR2 method, which is a promising communication-avoiding algorithm for the QR decomposition, and proved its numerical stability. (2) We developed a parallel linear equation solver based on the one-way dissection for quantum wave dynamics. (3) We developed a generalized eigenvalue solver EigenKernel, which is a hybrid solver that combines three eigenvalue solvers for dense matrices. These results will be useful for accelerating electronic structure calculations on many-core processors.
|
Report
(5 results)
Research Products
(42 results)
