CPLUOS Astrophysics team meeting

Friday 11 Mar 2022, 14:00 → 15:10 Asia/Seoul

물리학과

https://uos-ac-kr.zoom.us/j/8264902652

14:00 → 15:00 COSMOLOGY

14:00 Hyeonmo's Report - Head-on Collsion of FDM/CDM Halos(Writing Paper)

Speaker: Hyeonmo Koo (University of Seoul)

Head-on Collsion of FDM/CDM Halos

• Important things
  • Snapshots of velocity
    • v_i=40.0, T=0.0, T=0.14, T=0.28, T=0.42, T=0.56, T=0.7
    • v_i=80.0, T=0.0, T=0.07, T=0.14, T=0.21, T=0.28, T=0.35
  • FDM, CDM Collision Result
    • Initial Velocity - Velocity Decrease
    • FDM is higher than CDM

14:10 Young's report. FoF and MST algorithm

Speaker: Young Ju

FoF and MST algorithm

• Present number of cluster with changing linking-length. The results of DBSCAN does not appear in the graph.
• The DBSCAN and Hierarchical return the same results.
• Why is it happened? Change the free-parameter of DBSCAN
• DBSCAN has two free-parameter
  • 1. Eps = liking-length
  • 2. Min_sample = how many neighbours in the cluster
• Change the "min_sample" from 10 to 3. Because I defined minimum member of cluster as 3.
• Check the hierarchical clustering and DBSCAN again
• Hierarchical clustering : there are two criteria
  • 1. Single : calculate euclidean distance
  • 2. Ward : calculate variance of distance
  • Single : distance_threshold = 1.0
  • Ward : distance_threshold = 1.0
  • DBSCAN : eps = 1.0, min_sample = 3
• When min_sample is very small, the result of DBSCAN return the same result of Hierarchical
• What is alternative? Using the "Ward" method
• "Single" : distance_threshold = 4.0
• "Ward" : distance_threshold = 4.0
• "Ward" : distance_threshold = 50.0
• When I use the "Ward", should use very high distance_threshold. This does not compare with other algorithms.
• Hierarchical clustering
  • 1. Every point in the data starts as individual cluster.
  • 2. Nearest two cluster is merged, if their distance in under distance_threshold.
  • 3. "Single" use the euclidean distance, "Ward" use the sum of squared, variance.
• DBSCAN
  • 1. Check the number of point in certain eps range.
  • 2. If the number of point is larger than min_sample, then it is "Core point".
  • 3. The cluster is composed of the core point and border point.
• Why does FoF find number of cluster less than MGS?
• FoF finds the many groups to identify the largest cluster or structure not accounting for substructure.
• The aim is different from MGS. The difference comes from this properties.

14:20 Hannah's Report

Speaker: Hannah Jhee (University of Seoul)

Filament finders

• Tidal tensor/velocity hessian → Cosmic web(cluster, wall, filament, void...)
• DisPerSE : density based(Morse theory)
• T-REX : graph-based
• Plan
  • Create mock scatter data on a 3d line with gaussian displacement from the line
  • Test DisPerSE on it

14:30 CHOA(Cosmology of High-Order Statistics)

Speaker: Sumi Kim (University of Seoul)

CHOA(Cosmology of High-Order Statistics)

• 40 OpenMP threads

14:40 CLML

Cosmology with Large scale structure using Machine Learning

Speaker: Se Yeon Hwang

CLML_03.11.pdf

Cosmology with Large scale structure using Machine Learning(CLML)

• Consider analysing the large structure within astronomical data with Machine Learning techniques(CNN)
• Then, use properties(number density, mass density, velocity) in various combinations to extract cosmological information.
• Have x, y, z position, mass, vx, vy, vz
• Data
  • Shape = (32, 32, 32, 1)
  • Shape = (32, 32, 32, 2)
  • Shpae = (32, 32, 32, 3)
  • Shpae = (32, 32, 32, 4)

14:50 Dr.Sabiu's Report - N-PCF

Speaker: Dr Cristiano Sabiu (University of Seoul)

15:00 → 15:10 Discussion