-
Notifications
You must be signed in to change notification settings - Fork 0
/
Condensed Topic Notes.tex
667 lines (490 loc) · 28.1 KB
/
Condensed Topic Notes.tex
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
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
\documentclass[12pt, titlepage]{article}
\usepackage[letterpaper,margin=2cm]{geometry}
\usepackage{amsmath}
\usepackage{amsthm}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{mathdots}
\usepackage{mathtools}
\usepackage[]{algorithm2e}
\usepackage{Commons}
\usepackage[T1]{fontenc}
\usepackage[sc]{mathpazo}
\usepackage{tikz}
\usepackage{bm}
\usepackage{bbm}
\input{math-boxes}
\title{ORF 526 Condensed Topic Notes}
\date{\today}
\begin{document}
\maketitle
\tableofcontents
\newpage
\section{Elementary Probability Theory}
\subsection{Conditional Probability}
\begin{dfn}[Conditional Probability]{}
\[\prob{A \mid B} = \frac{\prob{A \cap B}}{\prob{B}}\]
\end{dfn}
\begin{theo}[Baye's Theorem]{}
\[\prob{A \mid B} = \frac{\prob{B \mid A} \prob{A}}{\prob{B}}\]
\end{theo}
\subsection{Statistical Independence}
\begin{dfn}[Statistical Independence]{}
Two events $A, B$ are statistically independent if
\[\prob{A \cap B} = \prob{A} \prob{B}\]
\end{dfn}
\subsection{Random Variables}
\begin{dfn}[Elementary Definition of Random Variables]{}
Given a sample space $\Omega$, a random variable is a numeric function on $\Omega$.
\end{dfn}
The distribution of random variables can be defined by a probability distribution function. This can take multiple forms.
\begin{dfn}[Cumulative Density Function]{}
\[F_X(x) = \prob{X \leq x}\]
\end{dfn}
\begin{dfn}[Probability Mass Function]{}
The distribution of discrete random variables can be defined by a function of the form
\[f_X(x) = \prob{X = x} = \prob{\{\omega \in \Omega \mid X(\omega) = x\}}\]
\end{dfn}
\begin{dfn}[Probability Density Function]{}
The distribution of a continuous random variable can be defined by a function $f_X$ where
\[\prob{a \leq X \leq b} = \int_a^b f_X(x)dx\]
Notice that this also defines the cumulative density function as
\[F_X(x) = \int_{-\infty}^x f_X(t)dt\]
\end{dfn}
\subsection{Expected Value}
\begin{dfn}[Expected Value]{}
The expected value of a discrete random variable $X$ is defined as
\[\expect{X} = \sum_{-\infty}^\infty x \prob{X = x}\]
For a continuous random variable, we define it as
\[\expect{X} = \int_{-\infty}^\infty xf_X(x) dx\]
where $f_X$ is the probability density function of $X$.\\
\end{dfn}
\begin{info}[Properties of Expected Value]{}
\begin{itemize}
\item Linearity: For $a, b \in \R$, $\expect{aX + bY} = a \expect{X} + b\expect{Y}$.
\item $\expect{X}$ is finite if and only if $\expect{|X|}$ is finite.
\item $X \geq 0$ A.S., then $\expect{X} \geq 0$.
\item If $X \leq Y$ A.S. and both $\expect{X}$ and $\expect{Y}$ exists (that is, $\min \{\expect{X_+}, \expect{X_-}\} < \infty$ and $\min\{\expect{Y_+}, \expect{Y_-}\} < \infty$), then $\expect{X} \leq \expect{Y}$.
\item If $\expect{|X^b|} < \infty$ and $0 < a \leq b$, then $\expect{|X^a|} < \infty$.
\item If $X, Y$ are independent random variables then $\expect{XY} = \expect{X}\expect{Y}$.
\end{itemize}
\end{info}
\subsection{Variance}
\begin{dfn}[Variance]{}
\[\sigma^2 = \Var(X) = \expect{(X - \expect{X})^2} = \expect{X^2} - \expect{X}^2\]
\end{dfn}
\begin{dfn}[Covariance]{}
\[\Cov(X, Y) = \expect{(X - \expect{X})}\expect{(Y - \expect{Y})}\]
\end{dfn}
\begin{rmk}[Properties of Variance]{}
\begin{itemize}
\item $\Var(X) \geq 0$.
\item $\Var(aX) = a^2\Var(X)$.
\item $\Var(aX + bY) = a^2\Var(X) + b^2\Var(Y) + 2ab\Cov(X, y)$.
\end{itemize}
\end{rmk}
\subsection{Important Distributions}
\subsubsection{Bernoulli Distribution}
\begin{rmk}{}
The Bernoulli distribution is the distribution of the random variable that takes the value $1$ with probability $p$ and the value $0$ with probability $(1-p)$.
\end{rmk}
\begin{info}[Bernoulli Distribution Properties]{}
\begin{itemize}
\item Expected Value: $\expect{X} = p$.
\item Variance: $\Var(X) = p(1-p)$.
\end{itemize}
\end{info}
\subsubsection{Binomial Distribution}
\begin{rmk}{}
The binomial distribution is the probability distribution of the number of successes in a sequence of $n$ experiments, each with probability of success $p$. In other words, the sum of $n$ independent random variables each with a Bernoulli distribution and probability $p$.
\end{rmk}
\begin{info}[Binomial Distribution Properties]{}
\begin{itemize}
\item PMF: $\prob{X = k} = f(k, n, p) = {n \choose k} p^k (1-p)^{n-k}$.
\item CDF: $\prob{X \leq k} = F(k, n, p) = \sum_{i = 0}^k {n \choose i}p^i(1-p)^{n-i}$.
\item Expected Value: $\expect{X} = np$.
\item Variance: $\Var(X) = np(1-p)$.
\end{itemize}
\end{info}
\subsubsection{Geometric Distribution}
\begin{rmk}{}
The geometric distribution describes the number of repeated Bernoulli trials (experiments with a probability of success $p$) needed to achieve one success. For example, how many times must a coin be flipped to get a heads.
\end{rmk}
\begin{info}[Geometric Distribution Properties]{}
\begin{itemize}
\item PMF: $\prob{X = k} = f(k, p) = (1-p)^{k-1}p$.
\item CDF: $\prob{X \leq k} = F(k, p) = 1 - (1 - p)^k$.
\item Expected Value: $\expect{X} = \frac{1}{p}$.
\item Variance: $\Var(X) = \frac{1-p}{p^2}$.
\end{itemize}
\end{info}
\subsubsection{Poisson Distribution}
\begin{rmk}{}
The Poisson distribution describes the probability that a given number of events will occur in a fixed length of time if the events occur at a constant rate and the chance of one occurring is independent of the time since the last event. For example, the number of meteors that strike the earth in one year. $\lambda$ denotes the average number of events in a unit time.
\end{rmk}
\begin{info}[Poisson Distribution Properties]{}
\begin{itemize}
\item PMF: $\prob{X = k} = f(k, \lambda) = e^{-\lambda} \frac{\lambda^k}{k!}$.
\item CDF: $\prob{X \leq k} = F(k, \lambda) = e^{-\lambda} \sum_{i=0}^k \frac{\lambda^i}{i!}$.
\item Expected Value: $\expect{X} = \lambda$.
\item Variance: $\Var(X) = \lambda$.
\end{itemize}
\end{info}
\subsubsection{Exponential Distribution}
\begin{rmk}{}
The exponential distribution describes the time between events in a Poisson point process. That is, a process in which events occur continuously and independently at a constant average rate. $\lambda$ denotes the average number of events in a unit time.
\end{rmk}
\begin{info}[Poisson Distribution Properties]{}
\begin{itemize}
\item PDF: $f(x, \lambda) = \lambda e^{-\lambda x}$.
\item CDF: $\prob{X \leq k} = F(x, \lambda) = \int_{-\infty}^x f(t, \lambda) dt = 1 - e^{-\lambda x}$.
\item Expected Value: $\expect{X} = \frac{1}{\lambda}$.
\item Variance: $\Var(X) = \frac{1}{\lambda^2}$.
\end{itemize}
\end{info}
\subsubsection{Normal Distribution}
\begin{rmk}{}
The normal distribution is a continuous probability distribution with a number of unique properties. It is particularly important due to its relation to the central limit theorem.
\end{rmk}
\begin{info}[Normal Distribution Properties]{}
\begin{itemize}
\item PDF: $f(x, \mu, \sigma^2) = \frac{1}{\sqrt{2\pi \sigma^2}}e^{-\frac{(x - \mu)^2}{2 \sigma^2}}$.
\item CDF: $\prob{X \leq k} = F(x, \lambda) = \int_{-\infty}^x f(t, \lambda) dt = \frac{1}{2} [1 + erf(\frac{x - \mu}{\sigma \sqrt{2}})]$, where $erf$, the error function, is non-elementary.
\item Expected Value: $\expect{X} = \mu$.
\item Variance: $\Var(X) = \sigma^2$.
\end{itemize}
\end{info}
\subsection{Important Inequality Theorems}
\begin{theo}[Markov's Inequality]{}
Suppose $X$ is a non-negative random variable and $a > 0$. Then
\[\prob{X \geq a} \leq \frac{\expect{X}}{a}\]
\end{theo}
\begin{theo}[Chebyshev's Inequality]{}
Let $X$ be a random variable with $\expect{X} = \mu$ finite and non-zero variance $\Var(x) = \sigma^2$. Then for an real $k > 0$,
\[\prob{|X - \mu| \geq k\sigma} \leq \frac{1}{k^2}\]
\end{theo}
\subsection{Important Limit Theorems}
\begin{theo}[Weak Law of Large Numbers]{}
Let $\{X_i\}$ be i.i.d. random variables with mean $\mu$, and let $S_n = \frac{X_1 + X_2 + \dots + X_n}{n}$. Then $S_n$ converges to $\mu$ in probability. That is, for any $\epsilon > 0$,
\[\lim_{n \to \infty} \prob{|S_n - \mu| > \epsilon} = 0\]
That is, for any $\epsilon$, we can guarantee that $S_n$ is within $\epsilon$ of $\mu$ with arbitrary probability given a sufficient $n$.
\end{theo}
\begin{theo}[Strong Law of Large Numbers]{}
Let $\{X_i\}$ be i.i.d. random variables with mean $\mu$, and let $S_n = \frac{X_1 + X_2 + \dots + X_n}{n}$. Then $S_n$ converges to $\mu$ almost surely. That is,
\[\prob{\lim_{n \to \infty} S_n = \mu} = 1\]
\end{theo}
\begin{theo}[Central Limit Theorem]{}
Let $\{X_i\}$ be i.i.d. random variables with mean $\mu$ and finite variance $\sigma^2$, and let $S_n = \frac{X_1 + X_2 + \dots + X_n}{n}$. Then
\[\sqrt{n}(S_n - \mu) \to \mathcal{N}(0, \sigma^2)\]
as $n \to \infty$. That is, as $n \to \infty$, the cumulative distribution function of $\sqrt{n}(S_n - \mu)$ converges pointwise to the CDF of the normal distribution centered at $0$ with variance $\sigma^2$.
\end{theo}
\section{Measure Theory}
\subsection{($\sigma$) Algebras}
\begin{dfn}[($\sigma$) Algebras]
A collection $C$ of subsets of $E$ (the universe) is called an algebra if:
\begin{itemize}
\item $\emptyset \in C$.
\item $A \in C \implies A^c \in C$.
\item $A, B \in C \implies A \cup B \in C$.
\end{itemize}
$C$ is a $\sigma$-algebra if:
\begin{itemize}
\item $\emptyset \in C$.
\item $A \in C \implies A^c \in C$.
\item $A_1, A_2, \dots \in C \implies \bigcup^\infty_{i = 1}A_i \in C$.
\end{itemize}
Noticed the strengthened version of property three.
\end{dfn}
Examples:
\begin{itemize}
\item $\mathcal{E} = \{\emptyset, E\}$ trivial $\sigma$-algebra.
\item $\mathcal{E} = 2^E$, the discrete $\sigma$-algebra.
\end{itemize}
\begin{theo}[Intersections and Unions of $\sigma$-Algebras]{}
\begin{itemize}
\item Any (countable or uncountable) intersection of $\sigma$-algebras is a $\sigma$-algebra.
\item The union of two $\sigma$-algebras is not necessarily a $\sigma$-algebra.\\
\end{itemize}
\end{theo}
\begin{dfn}[Generated $\sigma$-algebra]{}
Let $\mathcal{C}$ be a collection of subsets of $E$. Take all $\sigma$-algebras that contain $\mathcal{C}$. Take their intersection. This $\sigma$-algebra is called the $\sigma$-algebra generated by $\mathcal{C}$, and is denoted by $\sigma(\mathcal{C})$.
\end{dfn}
\begin{dfn}[Borel Algebras and Borel Sets]{}
If $E$ is a topological space, and $\mathcal{C}$ is the collection of all open sets of $E$, then $\sigma (\mathcal{C})$ is called the Borel $\sigma$-algebra. Its elements are called Borel sets. The Borel $\sigma$-algebra is denoted by $\mathcal{B}_E$ or $\mathcal{B}(E)$.
\end{dfn}
\begin{dfn}[Measurable Spaces and Measurable Sets]{}
A pair $(E, \mathcal{E})$ is a measurable space if $\mathcal{E}$ is a $\sigma$-algebra on $E$. The sets in $\mathcal{E}$ are called measurable sets.
\end{dfn}
\begin{dfn}[Measurable Rectangles]{}
Let $(E, \mathcal{E})$, $(F, \mathcal{F})$ are two measurable spaces. If $A \subset E$ and $B \subset F$ are measurable sets, then $A \times B$ is called a measurable rectangle.\\
\end{dfn}
\begin{dfn}[Products of Measurable Spaces]{}
The product $(E \times F, \mathcal{E} \otimes \mathcal{F})$ where $\mathcal{E} \otimes \mathcal{F} = \sigma(\{A \times B \mid A \in \mathcal{E}$, $B \in \mathcal{F}\})$, is a measurable space.\\
\end{dfn}
\subsection{Measures}
\begin{dfn}[Measure]{}
$\mu: \mathcal{E} \to \mathbb{R^+}$ is a measure on $(E, \mathcal{E})$ if
\begin{enumerate}
\item $\mu(\emptyset) = 0$
\item If $A_1, A_2, \dots \in \mathcal{E}$ are pairwise disjoint, then $\mu(\bigcup^\infty_{i = 1} A_i) = \sum^\infty_{i = 1}\mu(A_i)$
\end{enumerate}
\end{dfn}
Property (2) is called countable additivity or $\sigma$-additivity.\\
\begin{info}[Properties of Measures]{}
Let $(E, \mathcal{E}, \mu)$ be a measure space with measure $\mu$. Then $\mu$ has the following properties:\\
\begin{enumerate}
\item Finite Additivity. $A \cap B = \emptyset \implies \mu (A \cup B) = \mu(A) + \mu(B)$.
\item Monotonicity. If $A \subseteq B$ then $\mu(A) \leq \mu(B)$. Note that this is clear because $B = A \cup (B \setminus A)$.
\item Sequential Continuity. If $A_n \subset A$ and $A_n$ converges to $A$ as $n \to \infty$ then $\mu(A_n)$ converges to $\mu(A)$ from below.
\item Boole's Inequality / Union Bound. $A_1, A_2, \dots \in \mathcal{E}$, $\mu(\bigcup^\infty_{i = 1} A_i) \leq \sum_{i = 1}^\infty \mu(A_i)$. Note: We can prove this by creating a sequence of disjoint subsets of $\bigcup_{i = 1}^\infty A_i$, then using sequential continuity.
\item If $c > 0$, then $c\mu$ is also a measure, $(c\mu)(A) = c \cdot \mu(A)$.
\item If $\mu_1$, $\mu_2$ are measures then $\mu_1 + \mu_2$ is a measure.
\end{enumerate}
\end{info}
\subsubsection{Examples of Measures}
\begin{enumerate}
\item The Dirac Measure. $x \in E$, $\delta_x(A) =
\begin{cases}
1 & x \in A\\
0 & x \not \in A
\end{cases}$
\item The Counting Measure. $D \in E$, $\mu(A) = $\# of points in $A \cap D$. If $D$ is countable, then $\mu(A) = \sum_{x \in D} \delta_x(A)$.
\item Discrete Measure. $D \subset E$ countable, $m(x)$ is some real value for every $x \in D$. $\mu(A) = \sum_{x \in D} m(x)\delta_x(A)$.
\item The Uniform Measure on $\{1, 2, \dots, n\}$. The discrete measure with $m(x) = \frac{1}{n}$.
\item The Lebesgue Measure. $Leb(A) =$ length of $A$ where $A$ is an interval.
\end{enumerate}
\begin{dfn}[Finite Measure]{}
If $\mu(E) < \infty$ then $\mu$ is called a finite measure.
\end{dfn}
\begin{dfn}[$\sigma$-Finite Measure]{}
We say that a measure $\mu$ is $\sigma$-finite if there exists a measurable countable partition $\{E_n\}$ of $E$ such that $\mu(E_n) < \infty$ for all $n$. Ex: $Leb$ is $\sigma$-finite.
\end{dfn}
For example, the Lebesgue measure is $\sigma$-finite but not finite.
\begin{dfn}[Probability Measure]{}
A probability measure is a measure $\mu$ such that $\mu(E) = 1$.
\end{dfn}
\begin{dfn}[Probability Space]{}
A probability space is a triple $(E, \mathcal{E}, \mu)$ such that $(E, \mathcal{E})$ is a measurable space, and $\mu$ is a probability measure.
\end{dfn}
Probability spaces are often denoted by the triple $(\Omega, \mathcal{F}, \mathbb{P})$.
\subsection{Specification of Measures}
\begin{theo}{}
Let $(E, \mathcal{E})$ be a measurable space. Let $\mu$ and $\nu$ be two measures on $(E, \mathcal{E})$ with $\mu(E) = \nu(E) < \infty$. If $\mu$ and $\nu$ agree on a collection of subsets that is closed under intersections, that generate $\mathcal{E}$, then $\mu = \nu$.
\end{theo}
\begin{theo}[Corollary]{}
Cor: If $\mu$ and $\nu$ are two probability measures on $\mathbb{R}$ with the same cumulative distribution functions, then $\mu = \nu$.
\end{theo}
\begin{theo}[Cumulative Distribution]{}
The cumulative distribution at a point $x$ is $\mu([-\infty, x])$.
\end{theo}
Assume that $\{x\} \in \mathcal{E}$ if $x \in E$. This is true of all standard measurable spaces.\\
\begin{dfn}[Atom]{}
$x$ is an atom of $\mu$ if $\mu(\{x\}) > 0$.
\end{dfn}
\begin{dfn}[Purely Atomic Measures]{}
$\mu$ is purely atomic if $\exists D \subset E$ such that $\forall x \in D$, $\mu(\{x\}) > 0$ and $\mu(E \setminus D) = 0$.
\end{dfn}
\begin{dfn}[Diffuse Measures]{}
$\mu$ is diffuse if it has no atoms. Ex: $Leb$.
\end{dfn}
\begin{lem}{}
If $\mu$ is a $\sigma$-finite measure on $(E, \mathcal{E})$ then we can write $\mu = \lambda + \nu$ where $\lambda$ is diffuse and $\nu$ is purely atomic.
\end{lem}
\subsection{Completeness and Negligible Sets}
\begin{dfn}[Negligible Set]{}
A measurable set $A$ is negligible if $\mu(A) = 0$. An arbitrary subset of $E$ is negligible if it is contained in a measurable set that is negligible.
\end{dfn}
\begin{dfn}[Complete Measure Space]{}
A measure space is complete if every negligible set is measurable.
\end{dfn}
\begin{lem}[]{}
To make a measure space complete, take $\overline{\mathcal{E}} = \sigma(\mathcal{E} \cup \mathcal{N})$, where $\mathcal{N}$ is the collection of negligible sets. $\forall A \subset \overline{\mathcal{E}}$, $A = B \cup N$ with $B \in \mathcal{E}, N \in \mathcal{N}$. Define $\overline{\mu}(A) = \mu(B)$. This is called the completion of the measure space, $(E, \overline{\mathcal{E}}, \overline{\mu})$. In the case of $(\mathbb{R}, \mathcal{B}_{\mathbb{R}}, Leb)$, the elements of $\overline{\mathcal{B}}_\mathbb{R}$ are called Lebesgue-measurable.\\
\end{lem}
\subsection{Measurable Functions}
\begin{dfn}[Measurable Functions]{}
Let $(E, \mathcal{E})$ and $(F, \mathcal{F})$ be two measure spaces. $f: E \to F$ is measurable relative to $\mathcal{E}$ and $\mathcal{F}$ if $f^{-1}(A) \in \mathcal{E}$ for every $A \in \mathcal{F}$.
\end{dfn}
\begin{theo}[]{}
Let $(E, \mathcal{E})$, $(F, \mathcal{F})$ be measure spaces. $f: E \to F$ is measurable relative to $\mathcal{E}$ and $\mathcal{F}$ if and only if there exists a collection $\mathcal{F}_0$ of subsets of $F$ such that $f^{-1}(B) \in \mathcal{E}$ $\forall B \in \mathcal{F_0}$, and $\mathcal{F}_0$ generates $\mathcal{F}$.
\end{theo}
\begin{prf}[]{}
Left as an exercise
\end{prf}
\begin{theo}[]{}
Let $(E, \mathcal{E})$, $(F, \mathcal{F})$, $(G, \mathcal{G})$ be measure spaces. $f: E \to F$, $g: F \to G$. If $f$ and $g$ are measurable, then $g \circ f$ is measurable.
\end{theo}
\begin{dfn}[]{}
$f: E \to \R$ is $\mathcal{E}$-measurable if it is measurable relative to $\mathcal{E}$ and $\mathcal{B}_\R$.\\
\end{dfn}
\begin{dfn}[Borel Function]{}
If $E$ is a topological space and $\mathcal{E}$ is the Borel $\sigma$-algebra, then we simply say that $f$ is a Borel function.
\end{dfn}
\begin{lem}[]{}
$f: E \to \R$ is $\mathcal{E}$-measurable, if and only if $f^{-1}((-\infty, r]) \in \mathcal{E}$ for all $r \in \R$.\\
\end{lem}
\begin{prf}
From HW1: $\sigma(\{(-\infty, r] \mid r \in \R\}) = \mathcal{B}(\R)$. Then it follows from claim stated last time wrt the inverse of a generating set.\\
\end{prf}
\begin{dfn}[]{}
$f^+:= \max\{f, 0\}$, $f^-:= -\min\{f, 0\}$. Note that $f = f^+ - f^-$.\\
\end{dfn}
\begin{lem}[]{}
$f$ is $\mathcal{E}$-measurable if and only if $f^+$ and $f^-$ are $\mathcal{E}$-measurable.
\end{lem}
\begin{prf}
Left as an exercise
\end{prf}
\begin{dfn}[Indicator Function]{}
An indicator function is of the form
\[\mathbbm{1}_A(x) =
\begin{cases}
1 & x \in A\\
0 & x \not \in A
\end{cases}\]
\end{dfn}
Check: $\mathbbm{1}_A$ is $\mathcal{E}$-measurable if and only if $A \in \mathcal{E}$.\\
\begin{dfn}[Simple Function]{}
A function is simple if $f = \sum_{i = 1}^n a_i \mathbbm{1}_{A_i}$, $a_i \in \R$. Where $A_1, A_2, \dots A_n$ are $\mathcal{E}$-measurable.\\
\end{dfn}
\begin{dfn}[Canonical Form of a Simple Function]
The canonical form of a simple function is $f = \sum_{j = 1}^m b_j \mathbbm{1}_{B_j}$ where $\{B_j\}$ is a partition of $\mathcal{E}$.\\
\end{dfn}
\begin{info}[]{}
Conversely, if a function is $\mathcal{E}$-measurable and takes only finitely many real values, then it is a simple function.
\end{info}
\begin{info}[]{}
If $f$ and $g$ are simple, then so are $f+g$, $f-g$, $fg$, $f/g$, $\max\{f, g\}$, $\min\{f, g\}$.\\
\end{info}
\begin{theo}[]{}
The class of measurable functions is closed under limits.\\
Let $\{f_n\}$ be a sequence of $\mathcal{E}$-measurable functions then $\inf f_n$, $\sup f_n$, $\liminf f_n$, and $\limsup f_n$, defined pointwise, are $\mathcal{E}$-measurable.
\end{theo}
\begin{prf}[]{}
For $\sup f_n = f$, we want to show that $f^{-1}(-\infty, r] \in \mathcal{E}$. Since intersections can be rewritten as unions, and $f(x) \leq r \iff f_n(x) \leq r$ $\forall n$, we have
\[f^{-1}(-\infty, r] = \bigcap^\infty_{n=1} f^{-1}_n(-\infty, r]\]
But we know that $f^{-1}_n(-\infty, r] \in \mathcal{E}$ and since this is a countable intersection, $f^{-1}(-\infty, r] \in \mathcal{E}$.
\end{prf}
Let $f: \overline{\R}_+ \to \overline{\R}_+$ and $d_n(x) = \sum_{k = 1}^{n2^n} \frac{k-1}{2^n} \mathbbm{1}_{[\frac{k-1}{2^n}, \frac{k}{2^n}]} + n \mathbbm{1}_{[n, \infty]}$. We can approximate $f$ by a sequence of simple functions $f_n = d_n \circ f$.\\
\begin{theo}[]{}
A function $f$ is $\mathcal{E}$-measurable if and only if it is the increasing limit of simple functions.\\
\end{theo}
\subsection{Integration of Measurable Functions}
Suppose $(E, \mathcal{E}, \mu)$ is a measure space. Define $f: E \to \R$. We want to find $\int f d_\mu$. That is, the integral of $f$ relative to the measure $\mu$. We denote this $\mu f = \mu(f) = \int f d_\mu = \int \mu(dx)f(x) = \int_E \mu(dx) f(x)$.\\
How we will do this is we will first define integrals over measure spaces for simple functions, then extend this definition by taking limits.\\
\begin{dfn}[Integral of a Measurable Function Relative to a Measure]{}
If $f$ is a simple function, $f = \sum_{i = 1}^n a_i \mathbbm{1}_{A_i}$, where $\{A_i\}$ is a partition of $E$ we define the integral as
\[\int f d\mu = \sum_{i=1}^n a_i \mu(A_i)\]
Now, suppose that $f$ is a measurable positive function, and let $f_n = d_n \circ f$. Then $\int f d\mu = \lim_{n \to \infty} f_n d\mu$.\\
Finally, if $f = f^+ - f^-$, then $\int f d\mu = \int f^+ d\mu - \int f^- d\mu$, provided that at least one of the two integrals on the right are finite. Otherwise, $\int f d\mu$ is undefined.
\end{dfn}
\begin{dfn}[Integrable Relative to a Measure]{}
$f$ is integrable if $\int f d\mu$ exists and is finite.\\
\end{dfn}
\begin{theo}[]{}
$f$ is integrable $\iff$ $\int |f| d\mu < \infty$. Notice $|f| = f^+ + f^-$.\\
\end{theo}
\begin{dfn}[Almost Everywhere]{}
A statement holds almost everywhere (for almost every $x \in E$) if it holds for all $x$ except for $x$ in a negligible set. Denoted $\mu$-a.e. or (a.e.). For probability measures, we say "almost surely".
\end{dfn}
\begin{info}[Properties of Measure Integrals]{}
$a, b \in \R^+$, $f, g \in \mathcal{E}_+$ ($\mathcal{E}$-measurable positive functions).
\begin{enumerate}
\item Positivity: $\mu(f) \geq 0$. $\mu(f) = 0 \implies f = 0$ a.e.
\item Linearity: $\mu(af + bg) = a\mu(f) + b\mu(g)$.
\item Monotonicity: If $f \leq g$ a.e., then $\mu(f) \leq \mu(g)$.
\end{enumerate}
\end{info}
\begin{theo}[Monotone Convergence Theorem]{}
If $f_n \to f$ from below, then $\mu(f_n) \to \mu(f)$ from below.\\
\begin{enumerate}
\item Dirac measure: using the Dirac delta $\delta_{x_0} (f) = f(x_0)$.\\
\item $\mu = \sum_{x \in D} m(x)\delta_x$, $D \subset E$, then $\mu(f) = \sum_{x \in D} m(x) f(x)$. Note that if $E$ is countable, then every measure is of this form ($m(x) = \mu(\{x\})$).
\end{enumerate}
Note that if $E$ is a vector space, we can think of $\mu(f)$ as the inner product $\langle \mu, f \rangle$.\\
\end{theo}
\begin{theo}[]{}
If the Reiman integral of $f$ exists, then $Leb$ integral of $f$ does as well and things are equal. However, the converse is false. Notice that if $E = [0, 1]$, $f = \mathbbm{1}_\Q$, then $Leb(f) = 0$.\\
\end{theo}
Suppose that $A \subset E$, $A \in E$, $f \in \mathcal{E}$. Then $f \mathbbm{1}_A \in \mathcal{E}$ and so $\mu(f \mathbbm{1}_A) = \int f \mathbbm{1}_A d\mu = \int_A f d\mu$.
\begin{theo}[]{}
Let $\{f_n\}$ be a monotone increasing sequence of measurable positive functions. Then $\mu(\lim_{n \to \infty} f_n) = \lim_{n \to \infty} \mu(f_n)$.
\end{theo}
\begin{prf}[]{}
$f := \lim f_n$ is well defined, so $\mu(f)$ is well defined. For all $n$, $f_n \leq f$, so $\mu(f_n) \leq \mu(f)$ and $\lim_n \mu(f_n) \leq \mu(f)$.\\
For the other direction, we want to show that $\lim_{n \to \infty} \mu(f_n) \geq \mu(d_k \circ f)$ $\forall k$.\\
\end{prf}
\begin{lem}[]{}
If $A \in \mathcal{E}$ is negligible, then $\int_A f d\mu = 0$ for all measurable $f$.\\
If $f = g$ a.e., then $\mu(f) = \mu(g)$.\\
If $f \in \mathcal{E}_+$, $\mu(f) = 0$ then $f = 0$ a.e.
\end{lem}
\begin{theo}[Fatou's Lemma]{}
Let $(f_n)_{n \geq 1}$ be a sequence of functions in $\mathcal{E}_+$. Then $\mu(\liminf f_n) \leq \liminf \mu(f_n)$. This follows from MCT (HW).
\end{theo}
\begin{theo}[Dominated Convergence Theorem]
If $f_n$ is a sequence of functions and there exists a function $g$ such that (a) $|f_n| \leq g$ $\forall n \geq $, and (b) $g$ is integrable, then $f:= \lim f_n$ (if it exists) is integrable and $\mu(f) = \lim \mu(f_n)$. This follows from Faton's (HW).\\
Terminology: $g$ dominates $f_n$ for every $n$.
\end{theo}
\begin{theo}[Corollary: Bounded Convergence Theorem]{}
Suppose that $\mu$ is a finite measure, and $|f_n| \leq c < \infty$ ($c$ a constant), and $f:= \lim f_n$ exists. Then $\mu(f) = \lim \mu(f_n)$.
\end{theo}
\begin{theo}[]{}
Let $(E, \mathcal{E})$ be a measurable space and $f$ a measurable function. $L: \mathcal{E}_+ \to \overline{\R}_+$. Then there exists a unique measure $\mu$ on $(E, \mathcal{E})$ such that $L(f) = \mu(f)$ if and only if
\begin{itemize}
\item $f = 0 \implies L(f) = 0$.
\item $L(af + bg) = aL(f) + bL(g)$.
\item If $f_n \to f$ from below, then $L(f_n) \to L(f)$ from below.
\end{itemize}
\end{theo}
\begin{info}[Products of Measure Spaces]{}
\begin{itemize}
\item $(E, \mathcal{E})$, $(F, \mathcal{F})$: $(E \times F, \mathcal{E} \otimes \mathcal{F})$.
\item $(E, \mathcal{E}, \mu)$, $(F, \mathcal{F}, \nu)$: $(E \times F, \mathcal{E} \otimes \mathcal{F}, \mu \times \nu)$.
\item $(\mu \times nu)(A \times B) = \mu(A) \times \nu(B)$.
\end{itemize}
\end{info}
\begin{theo}[Fubini]{}
Suppose that $f: E \times F \to \overline{\R}$ such that $\int \int_{E \times F} |f| d(\mu \times \nu) < \infty$. Then $\int \int_{E \times F} f d(\mu \times \nu) = \int_F (\int_E f(x, y) \mu(dx)) \nu(dy) = \int_E (\int_F f(x, y) \nu(dy)) \mu(DX)$.\\
\end{theo}
\begin{theo}[Tonelli]{}
If $f \geq 0$ then the same conclusions hold.
\end{theo}
\subsection{Absolute Continuity of Measures}
\begin{dfn}[Absolute Continuity of Measures]{}
$(E, \mathcal{E})$ with measures $\mu$ and $\nu$. We say that $\mu$ is absolutely continuous with regard to $\nu$ if $\forall A \in \mathcal{E}$, $\nu(A) = 0 \implies \mu(A) = 0$. Denote this by $\mu << \nu$.
\end{dfn}
Example: If a measure on $\R$ has a density (e.g., $\mu(dx) = \frac{1}{\sqrt{2\pi}} e^{-\frac{x^2}{2}} dx$, the standard Gaussian measure), then it is absolutely continuous with regard to the Lebasgue measure.\\
Example: Discrete distributions with the same support.\\
\begin{theo}[]{}
Suppose that $\mu$ is $\sigma$-finite, and that $\nu << \mu$. Then there exists a positive $\mathcal{E}$-measurable function $p$ such that $\int_E \nu(dx) f(x) = \int_E \mu(dx) p(x)f(x)$, $\forall f \in \mathcal{E}_+$.\\
Moreover, $p$ is unique up to equivalence (if this holds for $p'$, then $p = p'$ a.e.).\\
\end{theo}
\begin{dfn}[Radon-Nikodym Derivative]{}
This function $p$ is called the Radon-Nikodym derivative of $\nu$ with regard to $\mu$. We write this as $p(x) = \frac{\nu(dx)}{\mu(dx)}(x)$, or $p = \frac{d\nu}{d\mu}$.\\
\end{dfn}
If we care about $\nu$, but it is difficult to use. If $\nu << \mu$, then we can perform calculations using the nicer $\mu$.\\
\begin{dfn}[Singular Measures]{}
$\mu$ is singular with regard to $\mu$ if there exists some set $D \in \mathcal{E}$ such that $\mu(D) = 0$ and $\nu(E \setminus D) = 0$.
\end{dfn}
\subsection{Products of Measure Spaces}
A product of measure spaces
\[\bigotimes_{i = 1}^n (E_i, \mathcal{E}_i, \mu_i)\]
can be seen as $n$ mutually independent random variables (i.e. $n$ coin tosses).\\
How do we define a countably infinite product of measure spaces ($\bigotimes_{i = 1}^\infty (E_i, \mathcal{E}_i, \mu_i)$)?\\
Let $\mathcal{R}$ be the collection of all finite dimension measurable rectangles. That is, all sets of the form $\{ x \mid x_1 \in B_1, \dots, x_n \in B_n, x_{n + 1} \in \R, x_{n + 2} \in \R, \dots \}$ where $n \in \N$ and $B_i \in \mathcal{B}(\R)$. Then $\mathcal{B}_C = \sigma(\mathcal{R})$. We define the measure as
\[\mu(\{x \mid x_1 \in B_1, \dots, x_n \in B_n, \dots\}) = \mu_1(B_1)\mu_2(B_2)\dots\mu_n(B_n)\]
\begin{theo}[Kolmogarov's Extension Theorem]{}
Suppose $\{mu_n\}_{n \geq 1}$ is a sequence of probability measures, where $\mu_n$ is a probability measure on $(\R^n, \mathcal{B}_{\R^n})$ that is consistent. That is
\[\mu_{n + 1}(\{x_1 \in B_1, x_2 \in B_2, \dots, x_n \in B_n, x_{n+1} \in \R^n\}) = \mu_n(\{x_1 \in B_1, x_2 \in B_2, \dots, x_n \in B_n \})\]
For all $n \in \N$ and all $B_1, B_2, \dots, B_n \in \mathcal{B}(\R)$. Then there exists a unique probability measure $\mathbb{P}$ on $(\R^\N, \mathcal{B}_C)$ such that $\mathbb{P}(\{w \mid w_1 \in B_1, \dots, w_n \in B_n\}) = \mu_n(B_1 \times B_2 \times \dots \times B_n)$.
\end{theo}
\section{Probability Spaces}
%Treating it as an integral
\subsection{Expected Value}
\subsection{Almost Sure and Almost Everywhere}
\subsection{Inequalities and Bounds}
\subsection{Borel-Cantelli Lemmas}
\subsection{Law of Large Numbers, Central Limit Theorem}
\subsection{Weak Convergence}
\section{Markov Chains}
\end{document}