Visual walkthrough — Conditional probability

This page builds the formula $P(A\mid B)=\frac{P(A\cap B)}{P(B)}$ from absolute zero. No formula is used before we can see it. By the end you will understand every symbol as an area you can point at. This is the visual companion to Conditional probability — read it slowly, one picture at a time.

---

Step 1 — The whole world is a box of area 1

WHAT. We start with the sample space: the collection of everything that could possibly happen. Draw it as a rectangle. We stretch the rectangle so its total area equals $1$.

WHY area 1? Because probability is just "what fraction of all possibilities?" — and a fraction of the whole is easiest to see as area covered $÷$ total area. If the total area is exactly $1$, then "area of a region" is "its probability" directly, with no division. That single choice makes every later picture honest.

PICTURE. The outer black rectangle below is the world $\Omega$ (read "omega", the Greek letter we use for everything). Its area is written inside: $1$.

---

Step 2 — An event is a blob inside the box

WHAT. An event is just "some of the outcomes" — a coloured blob inside the box. Call one blob $B$. The area of $B$ is its probability, written $P(B)$.

WHY a blob and not a number yet? Because we want to see how two events overlap. A bare number like "$0.4$" hides the geometry. A blob shows exactly which outcomes belong to $B$ and which don't.

PICTURE. The shaded region is $B$. Everything outside it is $\neg B$ ("not B"). Since the whole box has area $1$, the area of $B$ plus the area of $\neg B$ must equal $1$.

$\underset{\text{area of the blob}}{\underbrace{P(B)}}+\underset{\text{area outside}}{\underbrace{P(\neg B)}}=\underset{\text{whole box}}{\underbrace{1}}$

Every symbol here is an area you can point to.

---

Step 3 — Two events overlap: the intersection

WHAT. Now draw a second event $A$ as another blob. The two blobs generally overlap. That overlap — the outcomes that are in $A$ and in $B$ at the same time — is written $A\cap B$ (read "A intersect B"). Its area is $P(A\cap B)$.

WHY the intersection matters. Conditional probability is about the question "given B, how much A?". The only part of $A$ that can still happen once we know $B$ is true is the part of $A$ that lives inside $B$ — which is exactly $A\cap B$. So this overlap is the star of the whole derivation.

PICTURE. The red lens in the middle is $A\cap B$ — where the two blobs share ground.

$\underset{\text{outcomes in BOTH }A\text{ and }B}{\underbrace{A\cap B}}⟶\underset{\text{area of the red lens}}{\underbrace{P(A\cap B)}}$

---

Step 4 — Learning $B$ throws the rest of the world away

WHAT. Someone tells you: "$B$ definitely happened." Every outcome outside $B$ is now impossible. We erase the whole box except the $B$ blob.

WHY erase? Because information removes possibilities. Before, our universe was the full box. Now our universe is only $B$. Anything not in $B$ can no longer occur, so it should not count toward any fraction anymore.

PICTURE. The greyed-out region is now dead. The only surviving world is the blob $B$ (outlined in red). Inside it, the red lens $A\cap B$ is still alive — that is the only piece of $A$ that survived.

This is the whole idea of conditioning: shrink the universe to $B$.

---

Step 5 — Re-measure inside the new, smaller world

WHAT. Inside the surviving world $B$, we ask: what fraction of it is also $A$? That fraction is the conditional probability $P(A\mid B)$ (read "probability of A given B").

WHY divide? Here is the crux. After Step 4 our surviving world is $B$, whose area is $P(B)$ — and that area is generally not $1$. But probabilities must be fractions of the whole current world. So we must rescale $B$ back up to total area $1$ before we can read off a probability. Rescaling area $P(B)$ up to $1$ means dividing every area inside by $P(B)$.

The surviving-$A$ region has area $P(A\cap B)$. Divide it by the surviving-world area $P(B)$:

$P(A\mid B)=\frac{\stackrel{\text{area of red lens (survivor)}}{\overbrace{P(A\cap B)}}}{\underset{\text{area of new whole world}}{\underbrace{P(B)}}}$

PICTURE. Left: the shrunken world $B$ with the red lens inside it. Right: we blow $B$ up to fill a fresh unit box — the red lens grows by the same factor, and its new area is exactly $P(A\mid B)$.

---

Step 6 — The degenerate case: what if $P(B)=0$?

WHAT. Suppose $B$ has zero area — it never happens. Then Step 5 asks us to divide by $0$, which is meaningless.

WHY it breaks. You cannot "shrink the universe to $B$ and then measure fractions of it" if $B$ has no outcomes to shrink into. There is no surviving world to re-measure. That is exactly why the definition carries the fine print $P(B)>0$.

PICTURE. A box whose "$B$" is a single line — zero area. Nothing survives conditioning; the fraction is undefined.

---

Step 7 — Two extreme overlaps: disjoint vs. contained

WHAT. We check the two limiting shapes the overlap can take, so no case surprises you.

Case A — no overlap (disjoint). If $A$ and $B$ never happen together, the lens vanishes: $P(A\cap B)=0$, so $P(A\mid B)=\frac{0}{P(B)}=0.$ Knowing $B$ happened makes $A$ impossible.

Case B — $A$ swallows $B$ (containment). If $B$ lies entirely inside $A$, then everywhere $B$ happens, $A$ happens too: $A\cap B=B$, so $P(A\mid B)=\frac{P(B)}{P(B)}=1.$ Knowing $B$ happened makes $A$ certain.

WHY show both ends? Any real conditional probability lands between these extremes, $0\le P(A\mid B)\le 1$. Seeing the endpoints proves the formula never produces nonsense.

PICTURE. Left: disjoint blobs → answer $0$. Right: $B$ inside $A$ → answer $1$.

---

Step 8 — Multiply the picture the other way = Bayes

WHAT. The red lens has one area, but we can build it from two directions:

• Start in world $B$, then take the $A$-fraction: $P(A\cap B)=P(A\mid B)P(B)$.
• Start in world $A$, then take the $B$-fraction: $P(A\cap B)=P(B\mid A)P(A)$.

Both give the same lens area, so we may set them equal and solve:

$P(A\mid B)P(B)=P(B\mid A)P(A)⟹\boxed{P(A\mid B)=\frac{P(B\mid A)P(A)}{P(B)}}$

WHY this is powerful. It lets us flip the conditioning — computing $P(A\mid B)$ from the reverse $P(B\mid A)$. That flip is the engine behind Bayes-Theorem and every classifier built on it, like Naive-Bayes-Classifier.

PICTURE. The same lens measured two ways: "$B$ then $A$" (top path) and "$A$ then $B$" (bottom path) meet at the identical shaded region.

The denominator $P(B)$ itself is usually expanded by summing over pathways — that is the Law-of-Total-Probability, and stacking these multiplications gives the Chain-Rule-of-Probability.

---

The one-picture summary

Everything above collapses into a single storyboard: whole box → two blobs → learn $B$ → erase the rest → rescale → read the fraction.

Recall Feynman retelling (say it out loud)

Picture the world as a box with area exactly one, so area is probability. Draw two smudges, $A$ and $B$; where they overlap is "both happen", the red lens. Now a friend says "B for sure happened" — so cross out everything outside $B$; that dead region can't count anymore. Your world is now just the $B$ smudge, but it's smaller than one, so blow it back up to full size — divide every leftover area by $P(B)$. The lens grows with it, and its new size is the answer: $P(A\mid B)=P(A\cap B)/P(B)$. If $B$ had no area at all, there's nothing to blow up — undefined. If $A$ and $B$ never touch, the lens is empty and the answer is $0$; if $B$ sits wholly inside $A$, the lens is $B$ and the answer is $1$. Finally, that same lens can be sliced starting from $A$ instead of $B$ — set the two slices equal and you've just derived Bayes' theorem.

Recall

What does dividing by $P(B)$ physically do? ::: It renormalizes — rescales the shrunken world $B$ back to total area $1$ so fractions are honest. When is $P(A\mid B)$ undefined? ::: When $P(B)=0$; there is no surviving world to measure. Disjoint $A,B$ gives $P(A\mid B)=$? ::: $0$ — knowing $B$ makes $A$ impossible. $B$ fully inside $A$ gives $P(A\mid B)=$? ::: $1$ — knowing $B$ makes $A$ certain. Setting the lens area equal from both directions yields what? ::: Bayes' theorem, $P(A\mid B)=P(B\mid A)P(A)/P(B)$.

Read the numbers-and-worked-examples companion back on Conditional probability, or the same story in Hinglish.