Probability space — sample space Ω, sigma-algebra F, measure P — Kolmogorov axioms: Visual walkthrough

Before any symbol, one promise: everything here is just area. Picture the whole sample space $Ω$ (all things that could happen) as a rectangle whose total area is exactly 1 unit — one liter of paint. An event is a region inside that rectangle. Its probability is the fraction of the paint sitting on it. That single mental image carries the entire page.

Step 1 — The stage: a rectangle of area 1

WHAT. Draw $Ω$ as a rectangle. Call its area $1$ . Any event $A$ is a blob inside it.

WHY. The three axioms only talk about numbers $P (A)$ . To see them we need a picture where those numbers are visible quantities. Area is perfect: it is never negative, it adds when regions don't overlap, and a fixed total can be normalized to $1$ . Those are exactly axioms A1, A3, A2 in disguise — so "area" is not a loose analogy, it is a faithful model.

PICTURE. Look at the figure. The whole peach rectangle is $Ω$ ; the magenta blob is $A$ .

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

Step 2 — The adding rule: disjoint pieces add

WHAT. If two events $A$ and $B$ never overlap (share no outcome), the area of " $A$ or $B$ " is just the two areas summed.

WHY. This is axiom A3, countable additivity, drawn in the simplest case. "Disjoint" means the blobs touch nowhere; then covering them with paint uses one blob's worth plus the other's — no paint is counted twice because no region belongs to both.

PICTURE. Two separated blobs; the combined shaded area is literally the sum.

The full axiom A3 allows infinitely many disjoint pieces $A_{1}, A_{2}, \dots$ to add. We only need the picture with two, but keep in the back of your mind: the rule survives even when you slice $Ω$ into a countable infinity of non-overlapping crumbs.

Step 3 — The empty event carries no paint

WHAT. Show $P (\emptyset) = 0$ : the region containing no outcomes has zero area.

WHY. It feels obvious, but nothing is obvious until it is forced by the axioms. The trick: the empty region $\emptyset$ is disjoint from itself and from anything else. So I can stack infinitely many copies of $\emptyset$ next to $Ω$ and the union is still just $Ω$ .

PICTURE. The rectangle $Ω$ , then a stack of infinitely many zero-width slivers (each a copy of $\emptyset$ ) that add nothing to the total.

Step 4 — Splitting the whole: the complement rule

WHAT. Split $Ω$ into a blob $A$ and everything outside it, called $A^{c}$ ("A-complement", the leftover region). Their areas must total $1$ .

WHY. $A$ and $A^{c}$ are disjoint (nothing is inside and outside at once) and together they tile the whole rectangle. So the adding rule from Step 2 applies, and the total is pinned to $1$ by A2. This is the single most-used rule in all of probability: "chance of not A".

PICTURE. The rectangle cut into a magenta $A$ and a violet $A^{c}$ ; the two colors fill it exactly.

Step 5 — Nesting: bigger blob, more paint (monotonicity)

WHAT. If $A$ sits entirely inside $B$ (written $A \subseteq B$ ), then $A$ cannot have more area than $B$ .

WHY. Draw $B$ as an outer blob and $A$ as an inner blob fully contained in it. The ring between them, $B ∖ A$ (" $B$ minus $A$ ", the part of $B$ outside $A$ ), is a genuine region with its own non-negative area. Since $B$ is $A$ plus that ring, and the ring can't subtract area, $B$ is at least as big as $A$ .

PICTURE. Nested blobs: inner $A$ (magenta), outer $B$ , orange ring $B ∖ A$ between them.

Step 6 — Overlap: why you must subtract (inclusion–exclusion)

WHAT. When $A$ and $B$ do overlap, adding their areas double-counts the overlap $A \cap B$ . Fix it by subtracting that overlap once.

WHY. This is the case Step 2 forbade. Pour paint on $A$ , then on $B$ : the lens-shaped region where they cross ( $A \cap B$ , " $A$ and $B$ ") gets painted twice. The paint on the union should be counted once, so we remove one copy of the double-counted lens.

PICTURE. Two overlapping blobs; the central lens is shaded darker to show the double count that must be subtracted.

Step 7 — Degenerate case: a point of zero area that still happens

WHAT. On a continuous stage like $Ω = [0, 1]$ , a single point has area $0$ , yet it can still occur. "Probability $0$ " is not "impossible".

WHY. Shrink a tiny interval around a point $0.5$ down to nothing. Each interval $[0.5, 0.5 + \frac{1}{n}]$ has area (length) $\frac{1}{n}$ , and $\frac{1}{n} \to 0$ as $n$ grows. The point is the limit of these shrinking intervals, so its area is the limit $0$ . But you can land exactly on $0.5$ when you throw a dart at the line — the outcome is real, its area is nil.

PICTURE. A number line $[0, 1]$ with nested intervals collapsing onto the point $0.5$ , their lengths $\frac{1}{1}, \frac{1}{2}, \frac{1}{3}, \dots$ marching to $0$ .

The one-picture summary

Every rule on this page is one statement about area inside a rectangle of total area 1:

add disjoint pieces (Step 2),
the leftover completes the whole (Step 4, complement),
nested means no bigger (Step 5, monotonic),
overlap is double-counted, so subtract it (Step 6),
and shrinking to nothing gives area $0$ (Steps 3 & 7).

Recall Feynman retelling of the whole walkthrough

Picture a rectangular table top holding exactly one liter of spilled paint — that's $Ω$ with total probability $1$ . Any shape you draw on the table is an event, and its probability is just how much paint sits inside it. Rule one: you can't have negative paint. Rule two: the whole table holds all one liter. Rule three: two shapes that never touch — their paint just adds. From only that: an empty shape holds no paint; the paint outside a shape is one liter minus the paint inside; a shape inside a bigger shape can't hold more paint than the bigger one; and if two shapes overlap you painted the overlap twice, so subtract it once. Finally, on a smooth table a single dot has no area at all, so no paint — but your finger can still land on it. That's the whole of basic probability, and it's all just area.

Recall Rapid self-test

Why is $P (\emptyset) = 0$ forced? ::: Stack infinitely many empties beside $Ω$ ; the union is still $Ω$ , so the pile of $P (\emptyset)$ terms sums to $0$ , and each non-negative term must be $0$ . Where does $P (A) \leq 1$ come from? ::: Complement rule $P (A) + P (A^{c}) = 1$ with $P (A^{c}) \geq 0$ . When does $P (A \cup B) = P (A) + P (B)$ hold exactly? ::: Only when $A \cap B = \emptyset$ ; otherwise subtract $P (A \cap B)$ . Does $P ({0.5}) = 0$ mean $0.5$ can't occur on $[0, 1]$ ? ::: No — probability $0$ is "almost never", the point is still a possible outcome.

Connections: parent 4.9.01 Probability space — sample space Ω, sigma-algebra F, measure P — Kolmogorov axioms (Hinglish) · builds toward Conditional probability & independence · Random variables · Inclusion–exclusion principle · Borel sigma-algebra · Measure theory & Lebesgue integration · Expectation and Lebesgue integral.