Visual walkthrough — Real analysis — rigorous epsilon-delta, metric spaces

We only assume you can read a graph: a horizontal axis (the input $x$) and a vertical axis (the output $y$). Everything else is earned below.

---

Step 1 — The vague word we must kill: "approaches"

WHAT. We have a rule that turns a number $x$ into a number $y$. We write it $y=f(x)$; the symbol $f$ is just the name of the rule, and $f(x)$ means "the output the rule gives when you feed it $x$." Our rule for the whole page will be $f(x)=x^2$ (square the input).

We want to make sense of the sentence: "as $x$ gets close to $3$, $f(x)$ gets close to $9$." The words "gets close to" are slippery — how close is close? That vagueness is what we destroy.

WHY. Look at the picture: as the input marble slides toward $3$ from either side, its height slides toward $9$. But "slides toward" is a verb, not something we can check. We need a test a doubter could actually perform.

PICTURE. The curve $y=x^2$, with the input point $3$ marked on the bottom and the target height $9$ marked on the left. The dashed lines show the pairing $3\mapsto 9$.

---

Step 2 — Turn "close" into a measurable distance

WHAT. "$f(x)$ is close to $L$" becomes: the distance between $f(x)$ and $L$ is small. The distance between two numbers $p$ and $q$ is written $|p-q|$ — subtract them, drop the sign. So closeness of output is exactly the single number $|f(x)-L|$.

WHY. A distance is one non-negative number we can compare to a threshold. That is the whole trick: replace a vibe ("close") with an inequality ("$|f(x)-L|$ is less than something").

PICTURE. On the vertical output axis, $L=9$ sits in the middle. We draw a tube of half-height $\epsilon$ around it. The symbol ==$\epsilon$== (Greek "epsilon") is just a chosen positive number — the allowed output error. "Inside the tube" literally means $|f(x)-L|<\epsilon$.

---

Step 3 — The challenge–response game

WHAT. The doubter picks the tube first — any $\epsilon >0$, however tiny. Then it's your job to find a matching input window around $a$ so narrow that every input in it lands inside the tube. The half-width of that input window is called ==$\delta$== (Greek "delta"), another positive number — the allowed input error.

WHY. Order matters. The doubter must be allowed to shrink $\epsilon$ after seeing nothing from you; then you react with a $\delta$. If you had to fix $\delta$ first, a tiny later $\epsilon$ could catch you out. So $\epsilon$ is the challenge, $\delta$ is your response, and $\delta$ is allowed to depend on $\epsilon$.

PICTURE. Two tubes now: the green output tube of half-height $\epsilon$ around $L$ (horizontal band), and a plum input window of half-width $\delta$ around $a$ (vertical band). The claim you must defend: every $x$ in the vertical band produces a point of the curve inside the horizontal band.

---

Step 4 — The full definition, assembled

WHAT. Glue Steps 2 and 3 with the promise "if the input is in the window, then the output is in the tube," and demand it for every challenge.

WHY. The single arrow $⟹$ ("if … then …") is the promise. The "$\forall \epsilon$" ("for all $\epsilon$") is what pins the output arbitrarily close — one lucky $\epsilon$ would prove nothing.

PICTURE. The two bands from Step 3 with the logical flow drawn as an arrow: inside plum window $\Rightarrow$ inside green tube.

Compare with Limits and Continuity: continuity is the same sentence but with $L=f(a)$ and the puncture removed.

---

Step 5 — Now actually WIN: factor out the input distance

WHAT. For our concrete claim, the output error is $|x^2-9|$. We rewrite it so the controllable quantity $|x-3|$ appears explicitly: $|x^2-9|=|(x-3)(x+3)|=|x-3|\cdot |x+3|.$

WHY. Our only knob is $\delta$, and $\delta$ controls $|x-3|$. So we must expose $|x-3|$ as a factor. What's left, $|x+3|$, is the "steepness multiplier" — the amount by which the input error gets magnified into output error.

PICTURE. The curve near $x=3$ with a right-triangle-style rise/run: a small horizontal nudge $|x-3|$ produces a vertical jump $|x^2-9|$, and the ratio of the two is roughly $|x+3|$ — the local steepness.

---

Step 6 — Cage the magnifier (the preliminary restriction)

WHAT. $|x+3|$ is not a constant — it grows as $x$ grows, so we can't just call it a fixed number. We temporarily promise the window is at most $1$ wide: force $|x-3|<1$. That means $2<x<4$, hence $5<x+3<7$, so $|x+3|<7$.

WHY. If we left $|x+3|$ free it could be enormous, and no finite $\delta$ would tame it. By first shrinking the window to width $1$, we bound the magnifier by a concrete number $M=7$. This is the "cap" that turns a moving multiplier into a fixed one.

PICTURE. A plum strip $2<x<4$ under the curve; over that strip the steepness $|x+3|$ stays below the horizontal line at $7$. Outside the strip it would climb past $7$ — that's why we cage it.

---

Step 7 — Choose $\delta$ and slam the door

WHAT. Inside the cage, $|x^2-9|<7|x-3|$. To force this below $\epsilon$, demand $7|x-3|<\epsilon$, i.e. $|x-3|<\epsilon /7$. Two promises must both hold — width $<1$ and width $<\epsilon /7$ — so we take the smaller: $\boxed{\delta =\min \left(1,\frac{\epsilon }{7}\right)}$

WHY. The $\min$ guarantees both restrictions at once: the cage stays valid (so $M=7$ is legal) and the arithmetic $7|x-3|<\epsilon$ goes through. If $\epsilon$ is large, $1$ is the binding cap; if $\epsilon$ is tiny, $\epsilon /7$ is.

PICTURE. The $\epsilon /7$ window (thin) and the $1$ window (wide) drawn together; $\delta$ is the narrower shaded overlap. As $\epsilon \to 0$ the overlap collapses to the $\epsilon /7$ line.

---

Step 8 — Edge & degenerate cases (never leave a scenario unshown)

---

The one-picture summary

WHAT. One frame: the curve $y=x^2$, the green output tube of half-height $\epsilon$ around $9$, the plum input window of half-width $\delta =\min (1,\epsilon /7)$ around $3$, and the guarantee arrow — every $x$ in the window lands in the tube.

Recall Feynman retelling (plain words, no symbols)

A doubter draws a horizontal ribbon of some chosen thinness around the height $9$ and dares me: "make the curve stay inside." I answer by drawing a vertical stripe around the input $3$ so narrow that the piece of curve sitting above it never pokes out of the ribbon. To find how narrow, I notice the curve's height error equals the input error times the local steepness. Near $3$ the steepness is about $6$, and I safely bound it by $7$ once I promise my stripe is under $1$ wide. So if I keep my stripe thinner than the ribbon's thinness divided by $7$ (and under $1$), the curve is trapped. Because I can do this for any thinness the doubter names, the curve truly heads to $9$ — that is what "the limit is $9$" means. And since I used nothing but "distance," the very same story works for vectors, functions, or any space where distance makes sense.

Recall Self-test

Why must $\delta$ be the minimum of the two bounds? ::: Both promises must hold at once — the cage $|x-3|<1$ (so the magnifier $|x+3|<7$ stays valid) and $|x-3|<\epsilon /7$ (so $7|x-3|<\epsilon$). Taking the smaller guarantees both. In the chain $|x^2-9|=|x-3||x+3|$, which factor does $\delta$ control? ::: $|x-3|$ — the input error; $|x+3|$ is the magnifier we must bound. What breaks if the doubter is allowed $\epsilon =0$? ::: It demands a negative distance $|f(x)-L|<0$, impossible; limits are about arbitrarily close, not exactly equal.