Recovering ionized bubble size distribution from 21cm power spectrum with ANN

2024/5/23, SWIFAR cosmology workshop

1. Recovering ionized bubble distribution from
21cm power spectrum with ANN
SWIFAR cosmology workshop(2024/5/23)
1
Hayato Shimabukuro(島袋隼士)
Yunnan university,SWIFAR

2. The history of the universe
©NAOJ
Dark Ages~~~No luminous object exists().z>30
Epoch of Reionization(EoR)~~~UV photons by luminous objects ionize
neutral hydrogen in the IGM ( ).z∼6−15
Cosmic Dawn~~~First stars and galaxies form ( ).z∼20−30
2

3. The history of the universe
©NAOJ
Dark Ages~~~No luminous object exists().z>30
Epoch of Reionization(EoR)~~~UV photons by luminous objects ionize
neutral hydrogen in the IGM ( ).z∼6−15
Cosmic Dawn~~~First stars and galaxies form ( ).z∼20−30
2

4. The history of the universe
©NAOJ
Dark Ages~~~No luminous object exists().z>30
Epoch of Reionization(EoR)~~~UV photons by luminous objects ionize
neutral hydrogen in the IGM ( ).z∼6−15
Cosmic Dawn~~~First stars and galaxies form ( ).z∼20−30
These epochs have not been
observed yet
2

5. 21cm line emission
Proton Electron
21cm line emission(1.4GHz)
(Neutral) hydrogen atom is a good tracer for IGM through the dark ages to EoR.
singletTriplet
21cm radiation: neutral hydrogen atom emits 21cm line emission due to hyperfine structure.
Transition
!Tb=
TS!T!
1+z
(1!exp(⌧⌫))
⇠27xH(1 +!m)
✓
H
dvr/dr+H
◆✓
1!
T!
TS
◆✓
1+z
10
0.15
⌦mh
2
◆
1/2✓
⌦bh
2
0.023
◆
[mK]
Brightness temperature
Red : cosmology Blue : astrophysics
3

6. 21cm line emission
We can map the distribution of HI in the IGM with 21cm line.
Liu & Shaw (2020)
To describe 21cm signal statistically…
Redshift
4

7. 21cm line emission
We can map the distribution of HI in the IGM with 21cm line.
Liu & Shaw (2020)
21cm global signal: Sky-averaged 21cm line
signal
To describe 21cm signal statistically…
Redshift
4

8. We can map the distribution of HI in the IGM with 21cm line.
Redshift
To describe 21cm signal statistically…
Liu & Shaw (2020)
21cm line emission
5

9. We can map the distribution of HI in the IGM with 21cm line.
21cm line power spectrum
h!Tb(k)!Tb(k
0
)i=(2⇡)
3
!(k+k
0
)P21
Redshift
To describe 21cm signal statistically…
Liu & Shaw (2020)
21cm line emission
5

10. HERA collaboration 2022b
Astrophysics from 21cm upper limits
•They obtained some constraints on astrophysical
parameters from 21cm power spectrum upper limits.
•HERA firstly constrains X-ray parameters at high
redshift.
•High redshift galaxies suggest higher X-ray luminosity
and lower metallicity than local galaxies.
Astrophysics with 21cm line signal is
around the corner!?
Shimabukuro+ (2023)
6

11. HERA collaboration 2022b
Astrophysics from 21cm upper limits
•They obtained some constraints on astrophysical
parameters from 21cm power spectrum upper limits.
•HERA firstly constrains X-ray parameters at high
redshift.
•High redshift galaxies suggest higher X-ray luminosity
and lower metallicity than local galaxies.
Astrophysics with 21cm line signal is
around the corner!?
6

12. SKA-Mid SKA-Low
7

13. SKA-Mid SKA-Low
7

14. SKA-Mid SKA-Low
7

15. IFT
21cm Image
BSD
Incomplete IFT due to limited number of antenna in interferometer.
visibility
•However, we do not observe 21cm image directly by radio interferometer!
•We first observe visibility and perform Inverse Fourier Transformation (IFT) to obtain
21cm image. Then, compute BSD.
Bubble size distribution(BSD)
Giri+ (2017)
•Bubble size distribution(BSD) tells us the information on EoR sources, ionizing efficiency,
radiative feedback, etc. We can measure BSD from 21cm image map.
8

16. IFT
21cm Image
BSD
Incomplete IFT due to limited number of antenna in interferometer.
visibility
•However, we do not observe 21cm image directly by radio interferometer!
•We first observe visibility and perform Inverse Fourier Transformation (IFT) to obtain
21cm image. Then, compute BSD.
Bubble size distribution(BSD)
Giri+ (2017)
•Bubble size distribution(BSD) tells us the information on EoR sources, ionizing efficiency,
radiative feedback, etc. We can measure BSD from 21cm image map.
8

17. 21cm power spectrum BSD
visibility
We can directly compute 21cm power spectrum from visibility without Inverse Fourier
Transformation.
Avoid information loss by incomplete IFT.
Bubble size distribution(BSD)
Giri+ (2017)
9

18. 21cm power spectrum BSD
visibility
We can directly compute 21cm power spectrum from visibility without Inverse Fourier
Transformation.
So, how do we try it??
Avoid information loss by incomplete IFT.
Bubble size distribution(BSD)
Giri+ (2017)
9

19. •Training network with training
dataset, ANN can approximate any
function which associates input and
output values.
y=f(x)
• Applying trained network to unknown
data(test data) for prediction.
yANN=f(xtest)
• ANN consists of input layer, hidden
layer and output layer. Each layer has
neurons.
nonlinear regression Problem
Artificial Neural Network (ANN)
https://www.snexplores.org/article/explainer-what-is-a-neuron10

20. •Training network with training
dataset, ANN can approximate any
function which associates input and
output values.
y=f(x)
• Applying trained network to unknown
data(test data) for prediction.
yANN=f(xtest)
• ANN consists of input layer, hidden
layer and output layer. Each layer has
neurons.
nonlinear regression Problem
Artificial Neural Network (ANN)
10

21. Artificial Neural Network (ANN)
Example
(To protect the right of publicity,
I don’t use real women. Instead, I
generated this woman by AI)
11

22. Artificial Neural Network (ANN)
Example
(To protect the right of publicity,
I don’t use real women. Instead, I
generated this woman by AI)
11

23. Application of ML to 21cm studies
Hey, ChatGPT, tell me how we can apply ML to 21cm studies.
12

24. Application of ML to 21cm studies
Hey, ChatGPT, tell me how we can apply ML to 21cm studies.
12

25. Application of ML to 21cm studies
Hey, ChatGPT, tell me how we can apply ML to 21cm studies.
12

26. Application of ML to 21cm studies
Hey, ChatGPT, tell me how we can apply ML to 21cm studies.
12

27. Application of ML to 21cm studies
Hey, ChatGPT, tell me how we can apply ML to 21cm studies.
Seems correct?
12

28. •Emulator + MCMC→parameter estimate
•Parameter estimate
•Others
(e.g.) Mertens+ (2017), Li+ (2019), Chardin+
(2019), Yoshiura+ (2020), Shimabukuro+
(2022), Bianco+ (2021,2023)
(e.g.) Kern+ (2017), Schmit+ (2018),
Aviad+ (2020), Bevins+ (2021),
Yoshiura+ (2023)
21cm study + ML
10
20
30
40
50
60
10 20 30 40 50 60
R
mfp,ANN
[Mpc]
R
mfp,true
[Mpc]
10
20
30
40
50
60
10 20 30 40 50 60
ζ
ANN
ζ
true
1
10
100
1 10 100
T
vir,ANN
[K/10
3
]
T
vir,true
[K/10
3
]
(e.g.) Shimabukuro&Semelin (2017),
Gilet+ (2018), Nicolas+ (2019), Doussot+
(2019), Choudhury+(2020,2021,2022),
Zhao+ (2022a,b), Vivekanand+ (2023)
13

29. Recovering statistics from another statistics
•We can associate statistics with other statistics by the ANN. Can we recover statistics from
another statistics by the ANN?
Statistics A Statistics B
14

30. Recovering statistics from another statistics
•We can associate statistics with other statistics by the ANN. Can we recover statistics from
another statistics by the ANN?
Statistics A Statistics B
21cm power spectrum
14

31. Recovering statistics from another statistics
•We can associate statistics with other statistics by the ANN. Can we recover statistics from
another statistics by the ANN?
Statistics A Statistics B
21cm power spectrum
Bubble size distribution
(BSD)
14

32. Recovering statistics from another statistics
•We can associate statistics with other statistics by the ANN. Can we recover statistics from
another statistics by the ANN?
Statistics A Statistics B
21cm power spectrum
Bubble size distribution
(BSD)
Can we recover BSD from 21cm PS with ANN?
14

33. Recovered BSD
Black: Distribution obtained by
21cm 3D image directly.
Red: Distribution obtained by
ANN.
15

34. Recovering BSD from 21cm PS
HS, Mao, Tan (2022)16

35. Including thermal noise
21cm PS with thermal noises
(SKA level)
Errors are estimated by 10
realizations thermal noises
17

36. More discussions? Please read our paper or invite me to the colloquium!
18

37. More discussions? Please read our paper or invite me to the colloquium!
What is the scale dependence of the 21cm PS to recover BSD?
18

38. More discussions? Please read our paper or invite me to the colloquium!
What is the scale dependence of the 21cm PS to recover BSD?
How accurate we can recover for other models?
18

39. More discussions? Please read our paper or invite me to the colloquium!
What is the scale dependence of the 21cm PS to recover BSD?
How accurate we can recover for other models?
Comparing the direct method (MCMC with 21cm PS
and generating BSD from the parameters)
18

40. Reconstruction of HI distribution from LAE
Publications of the Astronomical Society of Japan,(2014),Vol.00,No.0 7
Fig. 4.Sky distribution of the LAEs atz=5.7.Theredsquares,magentadiamonds,andblackcirclesrepresent positions of narrowband bright (<23.5mag),
medium-bright (23.5−24.0mag), and faint (24.0−25.0mag) LAEs, respectively. The large red open square indicatestheLAEswithspatiallyextendedLyα
emission (Shibuya et al. 2017b). The gray shades denote either the areas with no HSC data or the masked regions with a bad data quality. The scale on the
map is marked in angles (degrees) and the projected distances(comovingmegaparsecs).
Fig. 5.Same as Figure 4, but for the LAEsz=6.6.ThelargeredopensquaresindicatetheLAEswithspatially extended Lyαemission including Himiko
(Ouchi et al. 2009a) and CR7 (Sobral et al. 2015). See Shibuyaet al. (2017b) for more details.
Input :
Lyman-alpha emitter galaxies
Output :
HI distributionYoshiura, HS+ (2021)
cGAN
Shintaro Yoshiura (NAOJ)

41. 21cm cosmology + machine learning
has big potential!
20

42. 21cm cosmology + machine learning
has big potential!
Collaboration is welcome!!
20

43. Backup
21

44. Recovering statistics from another statistics
•We can associate statistics with other statistics by the ANN. Can we recover statistics from
another statistics by the ANN?
Statistics A Statistics B
21cm power spectrum
21cm global signalPreliminary
22

45. Recovering statistics from another statistics
21cm power spectrum
Bubble size distribution
(BSD)
(Note) It does not mean 21cm PS contains beyond Gaussian information. Training enables us
to connect a data space to another data space.23

46. Other models
•The distribution of relative errors for other models.
•The peak of relative errors comes to ~0 .
24

47. Comparison with MCMC
25