Probability space — sample space Ω, sigma-algebra F, measure P — Kolmogorov axioms: Question bank

4.9.1 · D5 · Maths › Probability Theory & Statistics › Probability space — sample space Ω, sigma-algebra F, measure

Shuru karne se pehle, kuch symbols jinpar questions rely karte hain — inhe ek baar earn karo taaki koi line unhe cold use na kare:

$Ω$ (capital omega) sample space hai: ek bag jisme har woh outcome hai jo ho sakta hai.
$F$ (fancy F) sigma-algebra hai: $Ω$ ke un subsets ki list jinhe hum probability assign karne ke liye allowed hain. Har aisa subset ek event hai.
$P$ probability measure hai: woh rule $P : F \to [0, 1]$ jo Kolmogorov ke teen axioms maanta hai (A1 non-negativity, A2 normalization $P (Ω) = 1$ , A3 disjoint events ke liye countable additivity).
$A^{c}$ ka matlab hai "not $A$ " (complement), $A \cup B$ ka matlab hai " $A$ ya $B$ ", $A \cap B$ ka matlab hai " $A$ aur $B$ ", $\emptyset$ empty set hai (woh event jo kabhi nahi hota).
$A ⊔ B$ (union ke neeche ek bar ke saath) disjoint union hai: exactly $A \cup B$ hi hai lekin saath mein yeh extra promise ki $A \cap B = \emptyset$ , yaani pieces overlap nahi karte. Hum $⊔$ likhte hain $\cup$ ki jagah precisely yeh signal karne ke liye ki "probabilities seedha add karna safe hai".

Spot the error

Error hunt: "

P (\emptyset) = 0

ek axiom hai, isliye hum ise assume kar lete hain."

Wrong — yeh derived hai: disjoint sequence

Ω, \emptyset, \emptyset, \dots

lo, toh A3 deta hai

P (Ω) = P (Ω) + \sum_{i \geq 2} P (\emptyset)

, isliye

\sum_{i \geq 2} P (\emptyset) = 0

; kyunki har term ek hi value

P (\emptyset) \geq 0

(A1) hai, ek repeated non-negative number ka infinite sum sirf tab

0

ho sakta hai jab woh number

0

ho.

Error hunt: "

F

saare outcomes ka set hai."

Wrong — outcomes

ω

Ω

mein rehte hain;

F

Ω

ke subsets ka collection hai (events), ek level upar.

Error hunt: "Kyunki

P

[0, 1]

mein map karta hai, negative values definition se impossible hain, toh A1 redundant hai."

Wrong —

[0, 1]

codomain woh conclusion hai jo hum justify karna chahte hain; A1 ke bina aap non-negativity prove nahi kar sakte, isliye A1 real kaam kar raha hai.

Error hunt: "

A

aur

A^{c}

boundary par overlap karte hain, toh

P (A) + P (A^{c})

over-count karta hai."

Wrong —

A

aur

A^{c}

complement ki definition se disjoint hain (koi cheez ek saath andar aur bahar nahi ho sakti), isliye unki probabilities cleanly add hokar

P (Ω) = 1

deti hain.

Error hunt: "Inclusion–exclusion ke liye hum

A \cup B

A

aur

B

mein split karte hain, jo

P (A) + P (B)

deta hai."

Wrong —

A

aur

B

disjoint nahi hain; sahi disjoint split hai

(A ∖ B) ⊔ (A \cap B) ⊔ (B ∖ A)

, jo

- P (A \cap B)

correction deta hai.

Error hunt: "

{1, 2, 3}

Ω = {1, 2, 3}

par ek sigma-algebra hai."

Wrong — woh

Ω

ka ek subset hai, subsets ka family nahi; yahan sabse chhota sigma-algebra

{\emptyset, Ω}

hai.

Error hunt: "

[0, 1]

par uniform (Lebesgue) measure ke under,

P ([0.5, 0.5 + \frac{1}{n}]) = \frac{1}{n}

kabhi

0

nahi pahunchta, isliye

P ({0.5}) > 0

hai."

Wrong —

P ([a, b]) = b - a

ke saath, hamare paas

{0.5} = ⋂_{n} [0.5, 0.5 + \frac{1}{n}]

hai aur continuity from above deta hai

P ({0.5}) = lim \frac{1}{n} = 0

Why questions

F

complement ke under closed kyun honi chahiye?

Kyunki agar aap pooch sakte ho "kya

A

hua?", toh aap yeh bhi pooch sakte ho "kya

A

nahi hua?"; ek measurable sawaal jiska negation unmeasurable ho, woh logically incomplete hoga.

A3 sirf finite nahi, countable unions ke liye kyun stated hai?

Kyunki continuous distributions, limits, aur integrals ke liye infinitely many disjoint pieces add karne padti hain; finite additivity akela shrinking sets ke limit ka

P

express nahi kar sakta. Dekho Measure theory & Lebesgue integration.

Har subset ko measure karne ki bajaye

F

ki zaroorat kyun hai?

Uncountable

Ω

par, har subset ko measurable insist karna countable additivity se contradict karta hai (Vitali's construction), isliye hum ek well-behaved family tak restrict karte hain.

A2 mein

P (Ω) = 1

specifically kyun fix hai, koi aur number kyun nahi?

Yeh ek normalization choice hai: "kuch na kuch hota hai" certain hai, aur certainty ko

1

kehna probabilities ko poore ka fraction ki tarah behave karaata hai. Koi bhi total rescale ho sakti hai, lekin

1

woh convention hai jo ratios ko chances ki tarah padhne deta hai.

Do alag events dono ki probability

0

kyun ho sakti hai jabki ek

\emptyset

ho aur doosri nahi?

Probability size measure karti hai, possibility nahi;

\emptyset

mein koi outcome nahi (truly impossible) jabki

{0.5}

mein ek outcome hai zero length ka (possible but negligible).

Independence Kolmogorov ke axioms mein kyun nahi aata?

Independence

P

ke zariye define ki gayi events ke beech ek extra relationship hai, na ki har

P

ko satisfy karna zaroori rule; yeh Conditional probability & independence mein ek layer upar rehta hai.

Edge cases

Edge case: Kisi bhi

Ω

par sabse chhota possible sigma-algebra kya hai?

Trivial wala

{\emptyset, Ω}

— yeh teeno closure rules satisfy karta hai aur sirf yeh poochne deta hai "kya kuch hua?".

Edge case:

Ω = {a}

(ek single outcome) par

P

kya hona zaroori hai?

P ({a}) = P (Ω) = 1

by A2 aur

P (\emptyset) = 0

; yeh space forced hai, koi freedom nahi.

Edge case: Kya ek event

A

aisa ho sakta hai jisme

P (A) = 1

ho lekin

A \neq = Ω

Haan —

[0, 1]

mein set

[0, 1] ∖ {0.5}

ki probability

1

hai lekin woh poora space nahi hai; aise

A

ko "almost sure" kehte hain.

Edge case: Agar

A_{1} \supseteq A_{2} \supseteq \dots

\emptyset

ki taraf shrink karein, toh

P (A_{n})

ka kya hoga?

Continuity from above (A3 ka consequence) se,

P (A_{n}) \to P (⋂_{n} A_{n}) = P (\emptyset) = 0

Edge case: Agar

A_{1} \subseteq A_{2} \subseteq \dots

upar ki taraf grow karein, toh

P (⋃_{n} A_{n})

kya hai?

Continuity from below (twin property, yeh bhi A3 se) se,

P (⋃_{n} A_{n}) = lim_{n} P (A_{n})

— badhte limit ki probability, probabilities ka limit hai; jaise

P ([0, 1 - \frac{1}{n}]) = 1 - \frac{1}{n} \to 1 = P ([0, 1))

Edge case: Kya

P (A \cup B) \leq P (A) + P (B)

kabhi violate hota hai?

Kabhi nahi — inclusion–exclusion deta hai

P (A \cup B) = P (A) + P (B) - P (A \cap B)

aur

P (A \cap B) \geq 0

, isliye union bound hamesha hold karta hai; equality iff

A, B

disjoint hain (probability-

0

overlap tak). Yeh Inclusion–exclusion principle ke zariye generalize hota hai.

Edge case: Kya union bound countably many events ke liye survive karta hai?

Haan —

P (⋃_{i = 1}^{\infty} A_{i}) \leq \sum_{i = 1}^{\infty} P (A_{i})

kisi bhi countable family ke liye (overlapping ho ya nahi); union ko disjoint shells

A_{i} ∖ (A_{1} \cup \dots \cup A_{i - 1})

mein split karo, A3 apply karo, aur har shell ki probability monotonicity se

\leq P (A_{i})

hai. Yeh "countable subadditivity" hai (Boole's inequality).

Recall Ek-line summary lock karne ke liye

Axioms measure $P$ ko constrain karte hain; closure rules event list $F$ ko constrain karti hain; probability- $0$ size ke baare mein hai, possibility ke baare mein nahi; aur "hamesha additive" sabse common trap hai.

Question bank — Probability space — sample space Ω, sigma-algebra F, measure P — Kolmogorov axioms

True or false — justify

Spot the error

Why questions

Edge cases