◂ the notebook

phycists.desmondbrown.me

Other projections⟨dev|Desmond⟩⟨ee|Desmond⟩⟨academic|Desmond⟩

No. 08February 2026.
Probability and statistics, MT2204: notes and an exercise

Probability from three axioms, then run forward in time

The course builds probability on three axioms: P(A) ≥ 0, P(Ω) = 1, and additivity over disjoint events. My notes prove six rules from them. Every proof makes the same move: split a set into disjoint pieces, then add.

1P(∅) = 0∅, ∅, ∅, … are disjoint and their union is ∅, so x = x + x + …, and only x = 0 works.
2P(Ac) = 1 − P(A)A and its complement are disjoint and together make Ω.
3A ⊆ B ⇒ P(A) ≤ P(B)B = A ∪ (B ∖ A), two disjoint pieces, and P(B ∖ A) ≥ 0.
4P(A ∪ B) = P(A) + P(B) − P(A ∩ B)A ∪ B is three disjoint pieces, and P(A) + P(B) counts the overlap twice.
5A1 ⊆ A2 ⇒ P(A2 − A1) = P(A2) − P(A1)A₂ = A₁ ∪ (A₂ − A₁), then additivity.
60 ≤ P(A) ≤ 1Nonnegativity below; the third rule with B = Ω above.

Then an exercise put the rules to work. Alice is up to date or behind at the end of each week. Up to date, she stays so with probability 0.8; behind, she catches up with probability 0.6. She starts up to date. The total probability theorem gives each week from the one before, and rule 2 removes P(Bn):

P(Un+1) = 0.8 P(Un) + 0.6 P(Bn)=0.6 + 0.2 P(Un)

Week by week that is 0.8, 0.76, 0.752. My notes stop partway through week two; the figure carries on. The gap to ¾ shrinks fivefold each week, so P(Un) = ¾ + ¼ (0.2)ⁿ, and within a month it hardly matters where she started.

00.51¾0123456780.80.60.760.720.7520.744starts up to datestarts behindP(up to date)week n
00.51¾0123456780.80.60.7520.744starts up to datestarts behindP(up to date)week n
Up to date after n weeks, starting up to date and starting behind. Both settle on ¾, the p that solves p = 0.6 + 0.2p. Computed on this page.

what I don’t understand yet

Bayes’ rule. My notes set two exercises, marbles drawn from one of two boxes and a watch found in a drawer, and solve neither yet. They are the next page.

where this goes

Entry 09 spends probabilities as widths: arithmetic coding gives each symbol a slice of [0, 1) as wide as its probability.