Visual walkthrough — Uniform continuity — difference from pointwise

4.10.23 · D2 · Maths › Advanced Topics (Elite Level) › Uniform continuity — difference from pointwise

---

Step 1 — Woh picture jo sab shuru karti hai: do dots aur ek box

KYA HAI. Kisi function $f$ ka graph draw karo. Horizontal axis pe do points chuno aur unhe $x$ aur $y$ kaho. "$f(x)$" ka matlab sirf itna hai ki "point $x$ ke upar curve ki height". Toh $x$ aur $y$ ke upar do heights baith jaati hain, $f(x)$ aur $f(y)$.

Ab do numbers matter karte hain:

• horizontal gap $|x-y|$ — dots kitne door hain sideways. Do vertical bars $|\cdot |$ ka matlab hai "distance", yaani "hamesha ek positive amount, direction ignore karo".
• vertical gap $|f(x)-f(y)|$ — curve ki heights kitni door hain upar-neeche.

KYU. Continuity ek trade ka promise hai: "sideways gap chhota rakho aur main promise karta hoon ki upar-neeche ka gap chhota rahega." Iske baare mein reason karne ke liye pehle hame in do gaps ko naam dena hoga aur unhe ek box ki do sides ki tarah dekhna hoga. Woh box hi poori kahaani hai.

PICTURE. Neeche wala red box width $|x-y|$ aur height $|f(x)-f(y)|$ ki hai. Continuity width ko control karke height ko control karti hai.

---

Step 2 — Do dials: $\epsilon$ (goal) aur $\delta$ (effort)

KYA HAI. Do Greek letters aate hain. Yeh sirf do positive numbers ke naam hain:

• $\epsilon$ (epsilon) — height tolerance jo tumhe meet karni hai: "output gap ko $\epsilon$ se chhota karo". Yeh tumhara opponent tumhe deta hai; tum ise choose nahin kar sakte.
• $\delta$ (delta) — width jo tumhe use karni allowed hai: "jab tak input gap $\delta$ se neeche hai". Yeh tum choose kar sakte ho.

Poora game yeh hai: koi $\epsilon$ naam leta hai; tumhe $\delta$ ka jawab dena hoga taaki $\underset{\text{dots within width }\delta }{\underbrace{|x-y|<\delta }}⟹\underset{\text{heights within tolerance }\epsilon }{\underbrace{|f(x)-f(y)|<\epsilon }}.$ Arrow $⟹$ ko "guarantee karta hai" padho. Symbol $<$ ka matlab hai "strictly less than".

YEH DO DIALS KYU AUR EK KYU NAHIN? Kyunki continuity ek cause-and-effect promise hai. Ek dial ($\epsilon$) woh effect set karta hai jo tum chahte ho; doosra ($\delta$) woh cause hai jise tum control kar sakte ho. Tum yeh promise ek single number se state nahin kar sakte — tumhe target aur effort dono chahiye.

PICTURE. Socho ek $\epsilon$-tall horizontal band height $f(x)$ ke around, aur ek $\delta$-wide vertical band $x$ ke around. Rule kehta hai: $\delta$-wide band ke andar ka koi bhi dot $\epsilon$-tall band ke andar land karna chahiye.

---

Step 3 — WOH EK quantifier swap (drawn, sirf stated nahin)

KYA HAI. Ab hum decisive question poochte hain: jab tum $\delta$ choose kar rahe ho, kya tumhe dekhne ki permission hai ki point kahaan hai? Precisely jawab dene ke liye hum dono flavours ka poora rule likhte hain, har piece ke saath — tolerance $\forall \epsilon$, width $\exists \delta$, point sweep, aur closeness implication:

$$\text{Pointwise: }\forall x_0\forall \epsilon >0\exists \delta >0\forall y:|x_0-y|<\delta ⟹|f(x_0)-f(y)|<\epsilon$$ $$\text{Uniform: }\forall \epsilon >0\exists \delta >0\forall x\forall y:|x-y|<\delta ⟹|f(x)-f(y)|<\epsilon$$

Har piece padho: $\forall$ = "har ke liye", $\exists$ = "ek hai", $\epsilon$ = height tolerance (Step 2), $\delta$ = woh width jo tum supply karte ho, aur arrow $⟹$ dono mein same closeness promise laata hai — "agar inputs $\delta$ ke andar hain toh outputs $\epsilon$ ke andar hain." Sirf boxed pieces ka ordering differ karta hai:

• Pointwise: tum pehle ek fixed point $x_0$ pe khade ho, phir ($\epsilon$ diye jaane ke baad) $\delta$ choose karo. Kyunki $\exists \delta$, $x_0$ ke baad baithta hai, tumhari width har point pe re-choose ki ja sakti hai — ise $\delta (\epsilon ,x_0)$ likho, "$\delta$ allowed hai ki dono $\epsilon$ aur location $x_0$ pe depend kare".
• Uniform: $\exists \delta$ points $x,y$ reveal hone se pehle baithta hai, toh ek width saare pairs ko ek saath serve karni chahiye — ise $\delta (\epsilon )$ likho: sirf tolerance pe depend karta hai, kabhi location pe nahin.

YEH VISUALLY KYU MATTER KARTA HAI. Ek aise curve pe jisme steepness change hoti hai, "steep part ke liye ek $\delta$" aur "flat part ke liye ek $\delta$" alag sizes ke hote hain. Pointwise tumhe $\delta$ region by region switch karne deta hai; uniform switching forbid karta hai. Woh single freedom-vs-restriction hi woh cheez hai jise hum ab break hote dekhenge.

PICTURE. Left panel: flat region ko wide $\delta$ chahiye; steep region ko narrow $\delta$ chahiye. Pointwise har ek ko locally pick karta hai (dono theek hain). Uniform ko ONE red $\delta$ pick karni hai — aur narrowest wali jeetti hai.

---

Step 4 — Steepness output gap ko kyun control karti hai: slope factor

KYA HAI. Concrete curve $f(x)=x^2$ lo. Output gap compute karo aur ise factor karo:

$$|f(x)-f(y)|=|x^2-y^2|=\underset{\text{amplifier}}{\underbrace{|x+y|}}\cdot \underset{\text{input gap}}{\underbrace{|x-y|}}.$$

Term by term:

• $|x-y|$ woh input gap hai jo tum control karte ho (Step 1).
• $|x+y|$ ek amplifier hai: yeh input gap ko multiply karta hai output gap produce karne ke liye. Jahan $|x+y|$ bada hota hai, ek tiny input gap ek large output gap mein blow up ho jaata hai.

FACTOR KYU? Kyunki $a^2-b^2=(a+b)(a-b)$ woh ek algebra move hai jo "woh gap jo main control karta hoon" aur "woh multiplier jo main nahin karta" ko separate karta hai. Factoring ke bina, $|x^2-y^2|$ opaque hai; factoring ke baad, danger — growing $|x+y|$ — humein stare kar raha hai.

PICTURE. Output gap (red box ki height) woh input gap hai jo local steepness $|x+y|$ se stretch hua hai. Pair ko rightward slide karo aur same-width box taller aur taller hota jaata hai.

---

Step 5 — Break: $\mathbb{R}$ pe koi single $\delta$ survive nahin karta

KYA HAI. $|x^2-y^2|<\epsilon$ force karne ke liye hume chahiye hoga

$$|x-y|<\frac{\epsilon }{|x+y|}.$$

Jaise $|x+y|\to \infty$ (yaani jaise hum hamesha ke liye right jaate hain), allowed width $\frac{\epsilon }{|x+y|}\to 0$. Ek single fixed $\delta >0$ us target ke neeche nahin reh sakta jo $0$ ki taraf ja raha hai. Toh $\mathbb{R}$ pe uniform continuity fail ho jaati hai.

Failure ko explicit banaate hain — yeh action mein uniform continuity ka negation hai:

YEH CONSTRUCTION KYU? Humne engineer kiye do points jo input mein close hain lekin output mein door, unhe wahan park karke jahan amplifier bada hai. Exactly yahi "not uniformly continuous" ka matlab hai: kuch $\epsilon$ ke liye, koi $\delta$ kaam nahin karta.

PICTURE. Narrow red input gap door right mein baitha hai jahan parabola steep hai; uska box $\epsilon =1$ line se taller hai chahe kitna bhi narrow banao.

---

Step 6 — Rescue: $[0,5]$ pe rehke amplifier cap karo

KYA HAI. $x^2$ ko closed interval $[0,5]$ pe restrict karo. Ab dono $x,y\in [0,5]$ hain, toh amplifier bounded hai:

$$|x+y|\le 5+5=10.$$

Isliye

$$|x^2-y^2|=|x+y||x-y|\le 10|x-y|.$$

$\delta =\frac{\epsilon }{10}$ choose karo. Tab $|x-y|<\delta$ dega $|x^2-y^2|\le 10\cdot \frac{\epsilon }{10}=\epsilon$. Khaas baat yeh hai ki $\delta$ mein koi $x$ nahin — ek width poore interval ko serve karti hai. Yahi uniformity hai.

AB KYU KAAM KARTA HAI. $[0,5]$ pe curve infinitely steep nahin ho sakta; amplifier $10$ pe top out karta hai. Amplifier cap karna cap karta hai kitna ek fixed width stretch ho sakti hai — toh ek fixed width kaafi hai. Yeh miniature mein Heine–Cantor Theorem hai: ek compact (closed aur bounded) interval pe, continuity automatically uniform continuity mein upgrade ho jaati hai. Engine hai Lebesgue number lemma — local $\delta$-balls se cover karo, finite subcover extract karo, minimum $\delta$ lo; finitely many positive numbers ka minimum phir bhi positive hota hai. Unbounded $\mathbb{R}$ pe woh "minimum" infinitely many shrinking numbers ka infimum ban jaata hai, jo $0$ ho sakta hai — aur Step 5 exactly wahi collapse hai.

PICTURE. $[0,5]$ ko local balls mein chop karo har apne $\delta$ ke saath; finitely many interval cover karte hain; sabse chhota red $\delta$ har jagah kaam karta hai.

---

Step 7 — Degenerate warning: bounded lekin NOT closed kaafi nahin hai

KYA HAI. Koi soch sakta hai "$[0,5]$ ne kaam kiya kyunki woh chhota/bounded tha — toh koi bhi bounded interval kaam karega." Yeh sach nahin hai. $f(x)=\frac{1}{x}$ ko open interval $(0,1)$ pe consider karo. Yeh ek set ke roop mein bounded hai, aur $f$ wahan continuous hai, phir bhi $0$ ke paas curve infinity ki taraf shoot karta hai. Amplifier exactly find karte hain, exactly waise jaise humne $x^2$ ko factor kiya. Dono fractions ko common denominator pe rakho:

$$\left|\frac{1}{x}-\frac{1}{y}\right|=\left|\frac{y-x}{xy}\right|=\underset{\text{amplifier}}{\underbrace{\frac{1}{xy}}}\cdot \underset{\text{input gap}}{\underbrace{|x-y|}}.$$

Term by term: $|x-y|$ phir se woh input gap hai jo tum control karte ho; amplifier ab $\frac{1}{xy}$ hai, aur jaise $x,y\to 0^{+}$ yeh blow up karta hai ($xy\to 0$ toh $1/(xy)\to \infty$).

Output gap ko $\epsilon$ se neeche force karne ke liye humhe $|x-y|<\epsilon xy$ chahiye hoga, aur right-hand side origin ke paas $0$ mein collapse ho jaata hai. Concretely, $\epsilon =1$ aur koi bhi candidate $\delta >0$ fix karo; pick karo

$$x=\delta ,y=\frac{\delta }{2}(\text{dono }0\text{ ke paas small }\delta \text{ ke liye}),|x-y|=\frac{\delta }{2}<\delta ,$$

phir bhi

$$\left|\frac{1}{x}-\frac{1}{y}\right|=\left|\frac{1}{\delta }-\frac{2}{\delta }\right|=\frac{1}{\delta },$$

jo $\epsilon =1$ se exceed karta hai jab $\delta <1$ ho. Width ke andar do points, output gap huge — exactly wahi Step 5 mechanism, ab $|x+y|\to \infty$ ki jagah $1/(xy)\to \infty$ se drive ho raha hai.

KYU FAIL HOTA HAI. $(0,1)$ bounded hai lekin not closed: troublesome endpoint $0$ missing hai, toh kuch bhi amplifier $1/(xy)$ ko arbitrarily large hone se nahin rokta jaise hum iske paas jaate hain. Compactness ko closed aur bounded chahiye (Heine–Cantor Theorem ke zariye); "closed" drop karo aur Step 6 ka rescue collapse ho jaata hai.

PICTURE. $0$ ke paas do dots tiny sideways gap ke saath; $1/x$ spike pe unki heights miles apart hain — box arbitrarily thin width ke liye arbitrarily tall hai.

---

Step 8 — Doosra degenerate case: uniform WITHOUT bounded slope ($\sqrt{x}$)

KYA HAI. Doosri direction mein symmetric warning: "uniform continuity ke liye bounded slope chahiye." Aisa nahin hai. $f(x)=\sqrt{x}$ ko $[0,\infty )$ pe lo. Iska slope $f^{\prime }(x)=\frac{1}{2\sqrt{x}}\to \infty$ hai jaise $x\to 0^{+}$ — unbounded — toh yeh Lipschitz nahin hai aur MVT shortcut apply nahin hoti. Phir bhi yeh uniformly continuous hai, aur yahan woh clean derivation hai us bound ki jo ise bachaati hai.

Maano $x\ge y\ge 0$ (zaroorat ho toh relabel karo). Output gap ko conjugate $\frac{\sqrt{x}+\sqrt{y}}{\sqrt{x}+\sqrt{y}}$ se multiply karo — woh standard trick jo square roots clear karti hai:

$$\sqrt{x}-\sqrt{y}=\frac{(\sqrt{x}-\sqrt{y})(\sqrt{x}+\sqrt{y})}{\sqrt{x}+\sqrt{y}}=\frac{x-y}{\sqrt{x}+\sqrt{y}}.$$

Ab $\sqrt{x-y}$ se compare karo. Kyunki $x\ge y\ge 0$ hai, humein milta hai $\sqrt{x}\ge \sqrt{x-y}$ (andar bada, root bada) aur $\sqrt{y}\ge 0$, toh denominator satisfy karta hai $\sqrt{x}+\sqrt{y}\ge \sqrt{x-y}$. Bada denominator fraction ko chhota banata hai:

$$\sqrt{x}-\sqrt{y}=\frac{x-y}{\sqrt{x}+\sqrt{y}}\le \frac{x-y}{\sqrt{x-y}}=\sqrt{x-y}.$$

Symmetric form mein: $|\sqrt{x}-\sqrt{y}|\le \sqrt{|x-y|}$. Output gap input gap ke square root se bounded hai — linear se gentler, phir bhi $0$ ki taraf shrink karta hai jaise input gap shrinks, location mein uniformly. $\delta ={\epsilon }^2$ choose karo:

$$|x-y|<{\epsilon }^2⟹|\sqrt{x}-\sqrt{y}|\le \sqrt{|x-y|}<\sqrt{{\epsilon }^2}=\epsilon .$$

Phir se $\delta ={\epsilon }^2$ mein koi $x$ nahin — uniform.

YEH KYU DIKHAO. Possibilities ke poore space ko fence off karne ke liye: Lipschitz (bounded slope) sufficient hai lekin necessary nahin. Ek curve ek point pe infinitely steep ho sakta hai aur phir bhi uniformly continuous ho sakta hai, provided steepness "integrable enough" ho ki square-root bound hold kare.

PICTURE. $0$ ke paas $\sqrt{x}$ curve vertical hai, phir bhi output gap $\sqrt{\text{input gap}}$ se cap hai — box height tolerance se kabhi nahin nikarti.

---

Ek-picture summary

Upar ki saari baatein ek single comparison mein compress ho jaati hain: kya amplifier poore domain pe bounded hai?

• Bounded → ek $\delta$ kaam karta hai → uniform ($x^2$ on $[0,5]$ ka right box; $\sqrt{x}$ bhi ​ bound ke zariye).
• Unbounded → $\delta$ ko $0$ tak shrink karna padta hai → not uniform ($x^2$ on $\mathbb{R}$; $1/x$ near $0$).

Recall Feynman retelling — plain words mein poora walkthrough

Ek curve draw karo. Uspe do dots rakho. Unke beech ek box appear hota hai: uski width hai woh kitne sideways door hain, uski height hai curve ki values kitni door hain upar-neeche. Continuity ek promise hai: box ko narrow rakho aur main ise short rakhunga. Koi ek height limit $\epsilon$ chillata hai; tum ek width $\delta$ ke saath jawab dete ho jo har box ko us height ke neeche rakhti hai. Woh ek hi sawaal jo do flavours ko alag karta hai woh hai: kya tumhe har region mein alag width choose karni milti hai (pointwise), ya kya tumhe poore page ke liye ek width pick karni hai (uniform)? $x^2$ ke liye, local steepness $|x+y|$ hai — yeh tumhara box taller stretch karta hai jitna door right jaate ho. Poori line pe yeh bina limit ke badhti hai, toh koi bhi single width eventually ek aisa box produce karti hai jo bahut tall hai: continuous lekin uniformly nahin. Line ko $[0,5]$ tak chop karo aur steepness $10$ se exceed nahin kar sakti; ab ek width ($\delta =\epsilon /10$) har box ko tame karti hai — yahi compactness rescue hai (Heine–Cantor). Lekin traps dekho: $(0,1)$ bounded hai phir bhi closed nahin, aur $1/x$ phir bhi missing edge $0$ ke paas explode karta hai (uska amplifier $1/(xy)\to \infty$), toh "chhota interval" kaafi nahin hai; aur $\sqrt{x}$ $0$ pe infinitely steep hai phir bhi uniform rehta hai kyunki uski box height sirf width ke square root ki tarah badhti hai. Ek idea sab pe rule karta hai: uniform continuity ⇔ amplifier poore domain mein bounded hai.

---

Connections

• Continuity (pointwise) — woh weaker promise (Step 3) jise yeh walkthrough strengthen karta hai.
• Heine–Cantor Theorem — Step 6 ka compactness rescue.
• Compactness / Lebesgue number lemma — finite-subcover + minimum-$\delta$ engine.
• Lipschitz continuity — sufficient test ke roop mein bounded amplifier (Steps 6, 8).
• Mean Value Theorem — bounded slope ko Lipschitz bound mein turn karta hai.
• Cauchy sequences — uniform continuity Cauchyness preserve karti hai; pointwise zaroorat nahin.

Concept Map