-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathpersistence.html
More file actions
416 lines (346 loc) · 17.8 KB
/
persistence.html
File metadata and controls
416 lines (346 loc) · 17.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
<!DOCTYPE html>
<html lang="en">
<head>
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-5MSKJH3VR6"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'G-5MSKJH3VR6');
</script>
<title>Yonny Sklansky</title>
<meta charset="utf-8" meta name="viewport" content="width=device-width, initial-scale=1">
<link rel="stylesheet" href="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/css/bootstrap.min.css"></link>
<script src="https://ajax.googleapis.com/ajax/libs/jquery/3.3.1/jquery.min.js"></script>
<script src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.7/js/bootstrap.min.js"></script>
<link rel="icon" href="Yonny.jpg"></link>
<script type='text/javascript' src="http://ajax.googleapis.com/ajax/libs/jquery/1.7.1/jquery.min.js"></script>
<script type='text/javascript' src="http://netdna.bootstrapcdn.com/bootstrap/3.0.2/js/bootstrap.min.js"></script>
<style type='text/css'>
ul.nav li.dropdown:hover ul.dropdown-menu {
display: block;
}
</style>
<style>
* {
font-family: 'Helvetica Neue';
font-size: 16px;
}
img {
position: relative;
transform: scale(1);
transition: .5s ease;
background: rgba(255, 255, 255, 0);
border: 3px solid rgba(0, 0, 0, 0);
border-radius: 20px;
z-index: 1;
cursor: pointer
}
.custom-image {
position: relative;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: rgba(255, 255, 255, 1);
transform: scale(1.5);
transition: .5s ease;
z-index: 3;
}
.active {
position: relative;
top: 0;
left: 0;
right: 0;
bottom: 0;
background: rgba(255, 255, 255, 1);
transform: scale(1.6);
transition: .5s ease;
z-index: 3;
border: 3px solid rgba(0, 0, 0, .2);
border-radius: 20px;
}
img:hover {
transition: .5s ease;
border: 3px solid rgba(0, 0, 0, .5);
border-radius: 20px;
z-index: 2;
cursor: pointer
}
a:hover {
cursor: pointer
}
</style>
</meta>
</head>
<body>
<nav class="navbar navbar-default navbar-fixed-top">
<div class="container">
<div class="navbar-header">
<button type="button" class="navbar-toggle" data-toggle="collapse" data-target="#myNavbar">
<span class="icon-bar"></span>
<span class="icon-bar"></span>
<span class="icon-bar"></span>
</button>
</div>
<div class="collapse navbar-collapse" id="myNavbar">
<ul class="nav navbar-nav">
<li><a href="./index.html">Home</a></li>
<li><a href="./About-me.html">About Me</a></li>
<li class="dropdown"><a href="./research.html"> Research <span class="caret"></span></a>
<ul class="dropdown-menu">
<li><a href="./persistence.html">Persistent Homology of the Cosmic Web</a></li>
<li><a href="./voids.html">Void Clustering and Heirarchy</a></li>
<li><a href="./caustics.html">Caustics in Large-Scale Structure</a></li>
</ul>
<li class="dropdown"><a href="./hobbies.html"> Hobbies and Interests <span class="caret"></span></a>
<ul class="dropdown-menu">
<li><a href="./scioly.html">Science Olympiad</a></li>
<li><a href="./drones.html">Drones
</a></li>
</ul>
<li><a href="./CV.pdf" target="_blank">Curriculum Vitae</a></li>
</li>
</ul>
</div>
</div>
</nav>
<div class="container", align = "justify", style="margin-top:50px">
<div class="page-header", align = "left">
<h2>Persistent Homology of the Cosmic Web
<span style="float:right;">
<a href="./research.html"> <h3>Back to research</h3></a>
</span>
</h2>
</div>
<div class="col-sm-7">
<h3 id="background">
Astrophysics Background
</h3>
<p>
In our universe, matter organizes itself on the largest scales into one connected structure known as the <b>cosmic web</b>.
This web is made of smaller sub-structures known as walls, filaments, and nodes, each of which connect to each other to create web-like patterns.
Between these structures lie cosmic voids, which I have studied in more detail <a href="./voids.html">here</a>.
Each of these phenomena are part of a field within cosmology focusing on the creation and evolution of <b>large-scale structure</b> on the megaparsec scale, and studying these can help us understand the origins of the universe and how it might continue to evolve.
<br><br>
An example of such structures can be seen on the figure on the right.
This image is taken from a thin slice in the middle of a dark matter simulation from the IllustrisTNG collection.
The simulation spans a volume of 35 x 35 x 35 Mpc<sup>3</sup>, and the slice is about 0.7 Mpc wide.
It is a snapshot at a redshift of about z=1.
<br><br>
Note that I will be using co-moving Mpc, and to convert to "real" distances, one needs to multiply this by the dimensionless Hubble constant used by IllustrisTNG, h=0.677.
<br><br>
</p>
</div>
<div class="col-sm-5">
<img src="slice of cosmic web.png" style="width:100%"></img>
<p align = "middle">
Fig. 1: the cosmic web in a simulation
</p>
</div>
<div class="col-sm-7">
<h3 id="persist">
Persistent Homology
</h3>
<p>
In order to study the cosmic web and its "connectedness" mathematically, it is useful to employ the techniques of <b>persistent homology</b>.
This tool has been used in the fields of neuroscience, materials science, and even nuclear collisions.
Here I will give a very brief overview for the "physicist's" persistent homology.
<br><br>
In the context of cosmology, <b>topology</b> studies the shape and connectivity of a structure.
As my colleague Georg Wilding put it, it defines distance not in terms of a specific metric, "but by a notion of neighbourhood or adjacency."
This implies that even physically distant objects in the universe can be considered "connected" through some path or network that links their two positions, effectively making them neighbors.
The types of distinguishable networks these structures form are associated with their <b>homology</b>.
This quantifies the space through the amounts and dimensionalities of <b>holes</b>.
<br><br>
For example, one might expect that a hollow sphere has no holes.
Indeed, in the layman sense, an unpunctured, perfect sphere will not allow you to put a chain through to lock your bike, like a torus might have.
However, this is only referring to the one-dimensional hole, which is the type of hole that is enclosed by tracing a one-dimensional curve on a surface.
There are other types of holes, namely the two-dimensional hole, which the sphere does exhibit.
This hole is enclosed by a two-dimensional surface in a three-dimensional space, and can be called a <b>cavity</b>.
</p>
</div>
<div class="col-sm-5" align="middle">
<p><br><br></p>
<img src="hollow sphere.png" style="width:50%"></img>
<br><br>
<img src="hollow torus.png" style="width:50%"></img>
<p align = "middle">
Fig. 2: a hollow sphere and a hollow torus
</div>
<div class="col-sm-12">
</div>
<div class="col-sm-5">
<br><br>
Computing these holes based on different <b>filtrations</b>, or thresholds of separations, is the essence of persistent homology.
Filtrations can be set on any piece of information one might like, such as a distance or local density.
An example of a distance-based filtration can be observed in the figure to the right.
<br><br>
This is a set of points in a two-dimensional space.
Around each point is a blue circle of equal radii, representing the distance filtration.
Whenever the associated circles of two different points overlap, they are connected by a red line.
Three lines and three points connected together trace a triangle.
Each of these structures is a type of <b>simplex</b>, or the n-dimensional generalization of a triangle.
For instance, the 0-dimensional simplex is a point, the 1-dimensional simplex is a line, then a triangle (2-simplex), a tetrahedron (3-simplex), a pentachoron (4-simplex), and so on.
Connected simplices are said to be part of a <b>simplicial complex</b>.
<br><br>
Simplicial complexes, such as the one in the figure, can represent interesting and quantifiable structures.
Using the distance filtration, one can start with a low distance value (small circles) and increase it gradually, keeping track of when two circles overlap and connect two points together.
When enough points connect to form a loop, it is considered a <b>cycle</b> and its <b>birth time</b> is recorded.
Cycles that are only comprised of triangles are not significant and are therefore not considered to ever be born.
In this example, a birth time is not really a time, but corresponds to the length of the longest line segment in a cycle at the moment a hole is enclosed, as that is the last line segment to have been created.
As the filtration continues to increase, more points connect to each other, until the former cycle is only comprised of triangles.
The moment this happens, the shape is considered to be dead and its <b>death time</b> is recorded.
<br><br>
The birth and death times can be plotted on a <b>persistence diagram</b> such as the one in figure 4.
In the example, there are two cycles, one much larger than the other.
I am able to count the number of cycles by simply counting the number of non-simplex (non-triangle) holes.
We expect the larger cycle to take longer to die, since the circles need to increase to a sufficient radius to overlap with circles on the other side
Thus, we find two points on the diagram corresponding to the two cycles.
<br><br>
</p>
</div>
<div class="col-sm-3" align="middle">
<br><br>
</p>
<img src="./early persistent homology thing.png" style="width:118%"></img>
<br><br><br>
<img src="./persistent homology thing 2.png" style="width:118%"></img>
</div>
<div class="col-sm-3" align="middle">
<br><br>
</p>
<img src="./midway persistent homology thing.png" style="width:118%"></img>
<br><br><br>
<img src="./late persistent homology thing.png" style="width:118%"></img>
</div>
<div class="col-sm-6">
</div>
<div class="col-sm-6" align="middle">
<br>
<img src="./persistence diagram.png" style="width:70%"></img>
<p>
Fig. 4: a persistence diagram corresponding to the simplicial complex
</p>
</div>
<div class="col-sm-12">
<p>
Indeed, they are born around the same time, but one dies much sooner while the other persists.
One can define the <b>persistence</b> of a feature by its distance to the diagonal.
A point of <b>low persistence</b> is close to the diagonal, while a point of <b> high persistence</b> is farther away.
Importantly, points cannot be found under the diagonal, since that corresponds to a feature dying before it is born.
Finally, the red points on the left side correspond to the times at which connections are made between "islands" of points, merging from two separate complexes to one.
The point at infinity indicates that the last time more than one complex was present was at a value of just under 1.5, and no new complex islands were found, even at an infinite distance filtration.
<br><br>
There are several important strengths of persistence diagrams when trying to quantify the connectedness of data.
First, it allows for the reduction of a complex dataset to its basic topological features, emphasizing those which the human eye can agree with.
Second, it is robust against noise, and any slight wiggles in the positions of the points in the example would yield a very similar persistence diagram.
Third, it can be used to understand underlying structure of a dataset beyond just its largest features.
<br><br>
In the next section I will describe how I applied persistent homology to the cosmic web, and suggest some future research questions.
<br><br>
</p>
</div>
<div class="col-sm-4">
<h3 id="cosmic">
Cosmic persistence
</h3>
<p>
A natural place to start is to notice the regularity of structure in figure 1.
In order to characterize the connectedness of the cosmic web, a tracer needs to be chosen.
That is, data that accurately and faithfully represents the distribution of matter.
My idea for this project was to make use of information provided by IllustrisTNG itself, which is the mass <b>halos</b> and their corresponding <b>subhalos</b>.
These are simply the galaxy clusters and their constituent galaxies, but including their dark matter contents.
I chose the central subhalo of each halo as its representative.
<br><br>
Figure 5 shows two copies of a thin 31.5 Mpc slice of the 205 Mpc simulation, with only the main subhalos plotted.
The second slice highlights voids detected by persistent homology.
It is able to idenify the holes in this slice by tracing the encompassing loops with a distance-based filtration, as before.
Three of the loops with high persistence are identified by red, green, and magenta.
These are, of course, not the only detected voids, but let's take these as prominent examples.
Visually, it is clear that the loops surround an area which is mostly empty.
It may not be entirely empty because though the slice is thin, there is still some depth in the image and those points may be located in the foreground or background.
</p>
</div>
<div class="col-sm-4" align="middle">
<br>
<img src="31_5 slice.png" style="width:100%"></img>
<br><br>
<img src="31_5 slice loops.png" style="width:100%"></img>
<p>
Fig. 5: A 31.5 Mpc slice of the simulation, with and without the detected voids
</p>
</div>
<div class="col-sm-4" align="middle">
<br><br><br><br><br><br>
<img src="persistence diagram of slice.png" style="width:100%"></img>
<p>
Fig. 6: The associated peristence diagram of the slice, representing 1-dimensional loops
</p>
</div>
<div class="col-sm-12">
<p>
In addition, some lone galaxies may inhabit voids anyways.
Furthermore, the voids appear to be bordering each other, ignoring the "noise" that separates them.
This is promising, and is indicative of the algorithm working as intended, picking out loops and voids, and assigning each one its birth and death times.
<br><br>
<div>
So why did I pick a slice of the simulation and not the whole thing?
First of all, because it's harder to create a visualization of a three-dimensional simulation on a two-dimensional screen.
And second, it is quite computationally expensive.
I ran this on my local laptop hardware before I had any experience with computing clusters, and I anticipate there is a lot of information to learn about this given enough computing power.
</div>
</p>
</div>
<div class="col-sm-7">
<h3 id="next">
Okay, But What's Next?
</h3>
<p>
Now that we have an understanding of the toolset, we can move on to bigger and better things.
There are a lot of possible questions to answer, but the ones I tackled in my research involve <b>cosmic voids</b>.
In the next page, I will move on from 1-dimensional holes to 2-dimensional cavities, and discover the true power of persistent homology.
Click the image to the right to go to voids!
</p>
</div>
<div class="col-sm-5" align="center">
<br><br><br>
<a href="./voids.html">
<img src="./delaunay_volume.png" style="width:66%" class="custom-image"></img>
</a>
</div>
<div class="col-sm-12" align="center">
<br><br>
<hr />
<button id="back-to-top">Back to top</button>
</div>
</div>
<script>
const images = document.querySelectorAll('img');
images.forEach(img => {
img.addEventListener('click', function() {
this.classList.toggle('active');
});
});
</script>
<script>
document.getElementById('back-to-top').addEventListener('click', function(e) {
e.preventDefault();
window.scrollTo({
top: 0,
left: 0,
behavior: 'smooth'
});
});
</script>
</body>
<!-- Bootstrap core JavaScript
================================================== -->
<!-- Placed at the end of the document so the pages load faster -->
<script src="./Sticky Footer Navbar Template for Bootstrap_files/jquery.min.js"></script>
<script>window.jQuery || document.write('<script src="../../assets/js/vendor/jquery.min.js"><\/script>')</script>
<script src="./Sticky Footer Navbar Template for Bootstrap_files/bootstrap.min.js"></script>
<!-- IE10 viewport hack for Surface/desktop Windows 8 bug -->
<script src="./Sticky Footer Navbar Template for Bootstrap_files/ie10-viewport-bug-workaround.js"></script>
</html>