서브메뉴
검색
Broadening Our Understanding of Galaxy Mergers with Machine Learning, Simulations, and Observations
Broadening Our Understanding of Galaxy Mergers with Machine Learning, Simulations, and Observations
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105304
- ISBN
- 9798273302532
- DDC
- 523
- 서명/저자
- Broadening Our Understanding of Galaxy Mergers with Machine Learning, Simulations, and Observations
- 발행사항
- [Sl] : University of Colorado at Boulder, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 178 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-07, Section: B.
- 주기사항
- Advisor: Comerford, Julie.
- 학위논문주기
- Thesis (Ph.D.)--University of Colorado at Boulder, 2025.
- 초록/해제
- 요약Galaxy mergers are an integral part of our understanding of how galaxies grow, form new stars, trigger AGN, and evolve morphologically. However, many studies only target major galaxy mergers (mass ratio 1:4) between massive galaxies (stellar mass 10.
- 초록/해제
- 요약 solar masses), despite minor mergers (mass ratio 1:10 1:4) and lower mass galaxies being much more common by number. Additionally, mergers have proved challenging to identify, especially at higher redshifts, where they play a large role in galaxy assembly. In this thesis, I first show that a wide variety of mergers in IllustrisTNG50 experience elevated star formation and black hole accretion rates at 0.2 z 3, and maintain these levels for at least 1 Gyr following a merging event.This enhancement increases as redshift decreases. Next, I discuss using this high spatial resolution sample of simulated galaxies, including low mass and minor mergers, to train a convolutional neural network to identify mergers at z~1 in mock Hubble Space Telescope imaging. I achieve ~73% accuracy, similar to that of local redshifts, for the first time. With explainable AI techniques such as Grad-CAM and UMAP, I investigate the types of galaxies the network can classify well, and which physical properties of galaxies can cause confusion. Then, I show domain adaptation techniques, key for bridging the gap between simulated training data and real observations.Finally, I discuss applying these machine learning methods to create large galaxy catalogs, enabling a more complete understanding of the impact of mergers on cosmic star formation and AGN activity. I present a plan to expand to upcoming large surveys from the Vera C. Rubin Observatory and the Nancy Grace Roman Space Telescope in the future.
- 일반주제명
- Astrophysics
- 일반주제명
- Astronomy
- 키워드
- Galaxy evolution
- 키워드
- Galaxy mergers
- 기타저자
- University of Colorado at Boulder Astrophysical and Planetary Sciences
- 기본자료저록
- Dissertations Abstracts International. 87-07B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360106
■00520260202105304
■006m o d
■007cr#unu||||||||
■020 ▼a9798273302532
■035 ▼a(MiAaPQ)AAI32282962
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a523
■1001 ▼aSchechter, Aimee L.
■24510▼aBroadening Our Understanding of Galaxy Mergers with Machine Learning, Simulations, and Observations
■260 ▼a[Sl]▼bUniversity of Colorado at Boulder▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a178 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-07, Section: B.
■500 ▼aAdvisor: Comerford, Julie.
■5021 ▼aThesis (Ph.D.)--University of Colorado at Boulder, 2025.
■520 ▼aGalaxy mergers are an integral part of our understanding of how galaxies grow, form new stars, trigger AGN, and evolve morphologically. However, many studies only target major galaxy mergers (mass ratio 1:4) between massive galaxies (stellar mass 10.
■520 ▼a solar masses), despite minor mergers (mass ratio 1:10 1:4) and lower mass galaxies being much more common by number. Additionally, mergers have proved challenging to identify, especially at higher redshifts, where they play a large role in galaxy assembly. In this thesis, I first show that a wide variety of mergers in IllustrisTNG50 experience elevated star formation and black hole accretion rates at 0.2 z 3, and maintain these levels for at least 1 Gyr following a merging event.This enhancement increases as redshift decreases. Next, I discuss using this high spatial resolution sample of simulated galaxies, including low mass and minor mergers, to train a convolutional neural network to identify mergers at z~1 in mock Hubble Space Telescope imaging. I achieve ~73% accuracy, similar to that of local redshifts, for the first time. With explainable AI techniques such as Grad-CAM and UMAP, I investigate the types of galaxies the network can classify well, and which physical properties of galaxies can cause confusion. Then, I show domain adaptation techniques, key for bridging the gap between simulated training data and real observations.Finally, I discuss applying these machine learning methods to create large galaxy catalogs, enabling a more complete understanding of the impact of mergers on cosmic star formation and AGN activity. I present a plan to expand to upcoming large surveys from the Vera C. Rubin Observatory and the Nancy Grace Roman Space Telescope in the future.
■590 ▼aSchool code: 0051.
■650 4▼aAstrophysics
■650 4▼aAstronomy
■653 ▼aConvolutional neural networks
■653 ▼aGalaxy evolution
■653 ▼aGalaxy mergers
■653 ▼aGalaxy morphology
■690 ▼a0596
■690 ▼a0606
■71020▼aUniversity of Colorado at Boulder▼bAstrophysical and Planetary Sciences.
■7730 ▼tDissertations Abstracts International▼g87-07B.
■790 ▼a0051
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360106▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
