Data Management and data analysis in Taiwan Guo Chin Liu on behave - - PowerPoint PPT Presentation

Data Management and data analysis in Taiwan

Guo Chin Liu on behave of TGWG

Group Members

‧ Faculty:

• C. Chen(TKU), T. Chiu(NTU, NTNU), S. Haino (AS), C. Lin

(NCHC), F. Lin (NTNU), G. Liu(TKU)

‧ Post Doc.:

• D. Chiou, Y. Inoue, S. Ko

‧ Student:

• B. Chen, Y. Chu, W. Hsu, M. Lin, Y. Pu, J. Peng, C. Yang, U.

Zaman, Y. Zheng http://taipeigravitationalwavegroup.weebly.com/members.html

Supported by NCTS 2-year seed group grant

Plans for the Group

‧ Short term: Data analysis

• C. Lin(NCHC), G. Liu(TKU), students(NTNU)
• focus on CBC

‧ Long term: physics with gravitational wave

• BBH of Brans-Dicke model (F. Lin, D.

Chiou)

• Dark stars(C. Chen)
• Numerical relativity or PN to generate

Waveform

• physics in KAGRA’s sensitive window

10 10

1

10

2

10

3

10

−25

10

−24

10

−23

10

−22

10

−21

Frequency [Hz] Strain [1/√Hz]

initial Virgo initial LIGO Adv LIGO KAGRA (VRSE) Adv Virgo LIGO III Einstein Telescope (ET−D)

Adhikari et al. 2014

Mini Summer School

‧ Date: July 13-15 ‧ Invited speakers: H. Tagoshi, L. Baiotti, S.

Kuroyanagi, T. Li, Y. Itoh

http://taipeigravitationalwavegroup.weebly.com/mini-school.html

Data Management

‧ Mirror data storage in ASGC ‧ possibility to mirror (Proc.) data from AS to

NCHC?

Computing Power

‧ NCHC(National Center for High-performance

Computing)

• 25 K cores (shared)
• 100kNTD/year for 100 cores

‧ Prototype: two servers (12 cores with 400GB MEM;

4cores with 64GB MEM and two GPUs), one workstation (6 cores, 64GB MEM and K80 GPU)

‧ Production: server with 400 cores and 200 GPUs

Pipeline Construction

‧ MCMC: C. Lin ‧ Construct waveform catalogs: G. Liu + NTNU

• GPU, model reduction technique

based on matched filtering, with tools provided by Kagali or LALsuite

GPU(graphics processing unit )

‧ FFTW, LAPACK included in CUDA ‧ experience: gomoku game (Hance and Guozhang) ‧ set up a training schedule in near future

Model Reduction Techniques

‧ decompose the waveform into a set of orthogonal basis

vectors.

‧ reduce the computation costs ‧ applied on

• stellar core-collapse wave form(PCA: Heng. 2009)
• CBC templates (SVD: Cannon et al 2010)

Singular Value Decomposition/ Principal Component Analysis

Model Reduction Techniques

‧ construction of a reduced basis catalog(Field. et

al 2011, Caudill et al. 2014 )

‧ searching the points in parameter space by

greedy sweep algorithm

‧ seek an N dimensional linear space to accurately

represent the space for considered sources

Reduced basis:

Welcome to Join Us