Uniqueness of Limits and Basic Examples

Suppose a function approaches some value as its input approaches a point. Can it approach two different values at once? The intuition screams no. But intuition is not proof. The entire logical structure of calculus depends on this "no" being airtight. If limits were not unique, the notation lim⁡x→af(x)\lim_{x\to a} f(x) would be meaningless. This is because it presupposes there is exactly one number worthy of the name. This article proves the Uniqueness of Limits Theorem directly from the Epsilon-Delta Definition of a Limit. It uses nothing but careful quantifier manipulation and the triangle inequality. This is the first real "theorem about the definition itself" that a student meets. It is a template for a huge swath of analysis: assume two candidates satisfy the same defining property, average their competing constraints, and derive a contradiction unless they coincide. Beyond the theorem, we return to the definition. We grind out limits of linear, quadratic, and reciprocal functions by hand. Each time, we choose an explicit δ\delta for a given ε\varepsilon. These computations look tedious. But they are exactly what makes the shortcut rules (sum, product, quotient of limits) trustworthy later. Those rules are theorems, and theorems need proofs. Proofs need to bottom out in the definition somewhere. This article is where that "somewhere" lives for three of the most common function shapes. By the end, you will have a rigorous proof that limits are unique. You will also have three fully worked ε\varepsilon–δ\delta arguments of increasing algebraic complexity. There is a worked example demonstrating exactly where things go wrong without a key hypothesis. Finally, you will have a clear sense of how this foundational work feeds directly into the Algebra of Limits.

Key takeaways
  • A limit, if it exists, is a single, unambiguous real number — this is a theorem, not an assumption.
  • The proof strategy: assume two different limits L1≠L2L_1 \ne L_2, exploit the definition with ε=∣L1−L2∣/2\varepsilon = |L_1-L_2|/2, and derive a contradiction via the triangle inequality.
  • Uniqueness requires that aa be a Limit Point of the domain — points must exist arbitrarily close to aa for the "approaching" to make sense at all.
  • Direct ε\varepsilon–δ\delta verification for f(x)=mx+bf(x)=mx+b, f(x)=x2f(x)=x^2, and f(x)=1/xf(x) = 1/x each requires a different algebraic trick to bound ∣f(x)−L∣|f(x)-L| in terms of ∣x−a∣|x-a|.
  • For non-linear functions (like x2x^2 or 1/x1/x) you must first cap δ\delta (e.g., δ≤1\delta \le 1) to control extra factors before choosing the final δ\delta.
  • These hand-built proofs are the bedrock justifying the later, faster Algebra of Limits rules.
  • Key idea: uniqueness converts "the limit" from a hopeful phrase into a well-defined mathematical object, and direct verification shows the abstract definition actually computes the answers we expect.

Where this lives today

The uniqueness of limits is not merely a formality — it is the load-bearing wall underneath essentially every later definition built on limits. Continuity at a point is defined as lim⁡x→af(x)=f(a)\lim_{x\to a} f(x) = f(a); this statement is meaningless unless "the limit" refers to one specific number to be compared against f(a)f(a). Derivatives are defined as lim⁡h→0f(a+h)−f(a)h\lim_{h\to 0} \frac{f(a+h)-f(a)}{h}; uniqueness guarantees a function cannot have two different slopes at a point. Definite integrals, defined via limits of Riemann sums, similarly need a single well-defined limiting value to deserve the name "the integral." In numerical analysis and scientific computing, algorithms that claim a sequence of approximations converges to "the" answer implicitly rely on uniqueness of limits (and its sequential cousin, uniqueness of limits of sequences) to make sense of convergence claims at all. In topology, uniqueness of limits characterizes Hausdorff spaces: the real line with its usual topology is Hausdorff, and this is exactly the topological property that makes the argument in our proof work (disjoint neighborhoods can always be found around distinct points). In probability theory, uniqueness of limits underlies the well-definedness of expectations and limiting distributions defined via limiting procedures. In cryptography and complexity theory, asymptotic analysis of algorithms (Big-OO notation, limiting behavior of error probabilities) leans on the same guarantee: a growth rate, once shown to exist, is unambiguous. This article's direct verifications, meanwhile, are the scaffolding beneath the Algebra of Limits, which will soon let you compute limits of sums, products, and quotients without repeating this ε\varepsilon–δ\delta labor by hand every time.

Ask the tutor: How does this same ε\varepsilon–δ\delta contradiction technique get reused to prove the Algebra of Limits (sum, product, quotient rules)?

DefinitionThe Limit of a Function at a Point

Let f:D→Rf: D \to \mathbb{R} where D⊆RD \subseteq \mathbb{R}, and let aa be a limit point of DD (meaning every open interval around aa contains a point of DD other than aa itself). We say

lim⁡x→af(x)=L\lim_{x \to a} f(x) = L

if for every ε>0\varepsilon > 0 there exists a δ>0\delta > 0 such that, for all x∈Dx \in D,

0<∣x−a∣<δ  ⟹  ∣f(x)−L∣<ε.0 < |x - a| < \delta \implies |f(x) - L| < \varepsilon.

Here ε\varepsilon (a tolerance on the output) is given first, and δ\delta (a tolerance on the input) must be produced in response — δ\delta is generally allowed to depend on ε\varepsilon (and on aa, and on ff).

Intuition

Think of the definition as a game with an adversary. The adversary names a tolerance ε>0\varepsilon > 0, however small and spiteful, demanding that f(x)f(x) land within ε\varepsilon of LL. You must respond with a window δ>0\delta > 0 around aa — excluding aa itself — such that every xx in that window forces f(x)f(x) into the demanded range. You win the definition of the limit if you can always answer, no matter how small ε\varepsilon gets. Uniqueness says: there is only one number LL for which you can always win this game. If a second contender L2≠L1L_2 \ne L_1 also claimed to win, then set ε\varepsilon to be smaller than half the gap between L1L_1 and L2L_2. Both players must trap f(x)f(x) near their own target, but f(x)f(x) is a single number — it cannot be simultaneously close to two far-apart targets. The window where both promises hold is nonempty (by the limit-point hypothesis), so the contradiction is real, not vacuous.

Figure
The gap ∣L1−L2∣|L_1 - L_2| cannot be straddled once ε\varepsilon is small enough
L₁ ε-band around L₁ L₂ ε-band around L₂ If ε < |L₁ − L₂| / 2, bands are disjoint: no single f(x) can lie in both gap = |L₁ − L₂|

The picture makes the algebra almost unnecessary: once ε\varepsilon is small enough, the two ε\varepsilon-bands around L1L_1 and L2L_2 do not overlap. Since f(x)f(x) is forced into both bands simultaneously for xx near aa (by hypothesis), and the bands are disjoint, no such xx exists — but the limit-point hypothesis guarantees such xx must exist. That is the contradiction.

TheoremUniqueness of Limits

Let f:D→Rf: D \to \mathbb{R}, D⊆RD \subseteq \mathbb{R}, and let aa be a limit point of DD. If

lim⁡x→af(x)=L1andlim⁡x→af(x)=L2,\lim_{x\to a} f(x) = L_1 \quad \text{and} \quad \lim_{x \to a} f(x) = L_2,

then L1=L2L_1 = L_2.

Proof

Technique: proof by contradiction, using the triangle inequality to split an assumed error bound. Suppose, for contradiction, that L1≠L2L_1 \ne L_2. Then ∣L1−L2∣>0|L_1 - L_2| > 0, so we may define

ε:=∣L1−L2∣2>0.\varepsilon := \frac{|L_1 - L_2|}{2} > 0.

Since lim⁡x→af(x)=L1\lim_{x\to a} f(x) = L_1, applying the Epsilon-Delta Definition of a Limit with this ε\varepsilon, there exists δ1>0\delta_1 > 0 such that for all x∈Dx \in D,

0<∣x−a∣<δ1  ⟹  ∣f(x)−L1∣<ε.(1)0 < |x-a| < \delta_1 \implies |f(x) - L_1| < \varepsilon. \tag{1}

Since lim⁡x→af(x)=L2\lim_{x \to a} f(x) = L_2, applying the definition again with the same ε\varepsilon, there exists δ2>0\delta_2 > 0 such that for all x∈Dx \in D,

0<∣x−a∣<δ2  ⟹  ∣f(x)−L2∣<ε.(2)0 < |x-a| < \delta_2 \implies |f(x) - L_2| < \varepsilon. \tag{2}

Let δ:=min⁡(δ1,δ2)>0\delta := \min(\delta_1, \delta_2) > 0 (the minimum of two positive numbers is positive). Because aa is a limit point of DD, the interval (a−δ,a+δ)(a - \delta, a+\delta) contains a point x0∈Dx_0 \in D with x0≠ax_0 \ne a; that is, there exists x0∈Dx_0 \in D with

0<∣x0−a∣<δ.0 < |x_0 - a| < \delta.

(This step is exactly where the limit-point hypothesis is used: without it, no such x0x_0 need exist, and the argument below would never engage.) Since δ≤δ1\delta \le \delta_1, we have 0<∣x0−a∣<δ10 < |x_0 - a| < \delta_1, so (1) gives

∣f(x0)−L1∣<ε.|f(x_0) - L_1| < \varepsilon.

Since δ≤δ2\delta \le \delta_2, we have 0<∣x0−a∣<δ20 < |x_0 - a| < \delta_2, so (2) gives

∣f(x0)−L2∣<ε.|f(x_0) - L_2| < \varepsilon.

Now apply the triangle inequality:

∣L1−L2∣=∣L1−f(x0)+f(x0)−L2∣≤∣f(x0)−L1∣+∣f(x0)−L2∣<ε+ε=2ε.|L_1 - L_2| = |L_1 - f(x_0) + f(x_0) - L_2| \le |f(x_0) - L_1| + |f(x_0) - L_2| < \varepsilon + \varepsilon = 2\varepsilon.

But 2ε=∣L1−L2∣2\varepsilon = |L_1 - L_2| by our definition of ε\varepsilon. So we have derived

∣L1−L2∣<∣L1−L2∣,|L_1 - L_2| < |L_1 - L_2|,

which is false — a strict inequality of a number with itself. This is a logical contradiction. Therefore the assumption L1≠L2L_1 \ne L_2 is false, and we conclude L1=L2L_1 = L_2. ■\blacksquare

∎

Worked examples

We now use the definition itself — not the theorem just proved, and not any algebra-of-limits shortcuts, since those are proved using results like this one — to verify three concrete limits. Each proof has the same architecture: start from the desired conclusion ∣f(x)−L∣<ε|f(x) - L| < \varepsilon, manipulate algebraically to isolate ∣x−a∣|x - a|, and then declare a δ\delta that forces the inequality. We write these as scratch work (informal, working backward) followed by a clean forward proof.

Example 1 — a linear function. Verify directly from the definition that

lim⁡x→3(4x−5)=7.\lim_{x \to 3} (4x - 5) = 7.

Scratch work. We want to bound ∣(4x−5)−7∣=∣4x−12∣=4∣x−3∣|(4x-5) - 7| = |4x - 12| = 4|x-3|. If we want 4∣x−3∣<ε4|x-3| < \varepsilon, it suffices to have ∣x−3∣<ε/4|x - 3| < \varepsilon/4. So δ=ε/4\delta = \varepsilon/4 should work — no capping needed, since the coefficient 44 is a constant multiplier with no dependence on xx. Proof. Let ε>0\varepsilon > 0 be given. Choose δ=ε/4\delta = \varepsilon/4. Note δ>0\delta > 0 since ε>0\varepsilon > 0. Suppose x∈Rx \in \mathbb{R} satisfies 0<∣x−3∣<δ0 < |x - 3| < \delta. Then

∣(4x−5)−7∣=∣4x−12∣=4∣x−3∣<4δ=4⋅ε4=ε.|(4x-5) - 7| = |4x - 12| = 4|x - 3| < 4\delta = 4 \cdot \frac{\varepsilon}{4} = \varepsilon.

Since ε>0\varepsilon > 0 was arbitrary and we exhibited a valid δ\delta in each case, this proves lim⁡x→3(4x−5)=7\lim_{x\to 3}(4x-5) = 7. ■\blacksquare Note the asymmetry deliberately built in: the point a=3a=3 is not 00, the slope 44 is not 11, and the limit value 77 is not aa or 00 — this confirms the argument is not secretly relying on some cancellation special to "nice" numbers.

Example 2 — a quadratic function. Verify directly from the definition that

lim⁡x→2(x2+3x)=10.\lim_{x \to 2} (x^2 + 3x) = 10.

Scratch work. We want to bound

∣(x2+3x)−10∣=∣x2+3x−10∣=∣(x−2)(x+5)∣=∣x−2∣ ∣x+5∣.|(x^2 + 3x) - 10| = |x^2 + 3x - 10| = |(x-2)(x+5)| = |x-2|\,|x+5|.

Unlike the linear case, the second factor ∣x+5∣|x+5| is not constant — it depends on xx, so we cannot simply divide ε\varepsilon by it. The fix: since we only care about xx near 22, we may first restrict attention to xx within distance 11 of 22, i.e. impose δ≤1\delta \le 1 as a cap. If ∣x−2∣<1|x - 2| < 1, then 1<x<31 < x < 3, so 6<x+5<86 < x + 5 < 8, hence ∣x+5∣<8|x+5| < 8. With this cap in place,

∣x−2∣ ∣x+5∣<8∣x−2∣,|x-2|\,|x+5| < 8|x-2|,

and we want 8∣x−2∣<ε8|x-2| < \varepsilon, i.e. ∣x−2∣<ε/8|x-2| < \varepsilon/8. So we should take δ=min⁡(1,ε/8)\delta = \min(1, \varepsilon/8): small enough to satisfy both the cap and the final bound. Proof. Let ε>0\varepsilon > 0 be given. Choose

δ=min⁡(1,ε8).\delta = \min\left(1, \frac{\varepsilon}{8}\right).

This is positive since both 11 and ε/8\varepsilon/8 are positive. Suppose x∈Rx \in \mathbb{R} satisfies 0<∣x−2∣<δ0 < |x-2| < \delta. Since δ≤1\delta \le 1, we have ∣x−2∣<1|x - 2| < 1, i.e. 1<x<31 < x < 3. Adding 55 throughout, 6<x+5<86 < x + 5 < 8, so in particular ∣x+5∣<8|x + 5| < 8 (as x+5>6>0x+5 > 6 > 0, the absolute value is just x+5x+5 itself, and it's bounded above by 8). Since δ≤ε/8\delta \le \varepsilon/8, we have ∣x−2∣<ε/8|x-2| < \varepsilon/8. Combining these two bounds:

∣(x2+3x)−10∣=∣x−2∣ ∣x+5∣<ε8⋅8=ε.|(x^2+3x) - 10| = |x-2|\,|x+5| < \frac{\varepsilon}{8} \cdot 8 = \varepsilon.

Since ε>0\varepsilon > 0 was arbitrary and a valid δ\delta was produced in each case, this proves lim⁡x→2(x2+3x)=10\lim_{x \to 2}(x^2 + 3x) = 10. ■\blacksquare

Example 3 — a reciprocal function. Verify directly from the definition that

lim⁡x→51x=15.\lim_{x \to 5} \frac{1}{x} = \frac{1}{5}.

Scratch work. We must bound

∣1x−15∣=∣5−x5x∣=∣x−5∣5∣x∣.\left|\frac{1}{x} - \frac{1}{5}\right| = \left|\frac{5 - x}{5x}\right| = \frac{|x-5|}{5|x|}.

Here the denominator 5∣x∣5|x| is a problem for two reasons: it depends on xx, and (worse) if xx could be near 00, this denominator could be tiny, blowing the fraction up uncontrollably. We must cap δ\delta to keep xx away from 00. Since we're centering at a=5a = 5, capping δ≤1\delta \le 1 forces 4<x<64 < x < 6, safely bounding xx away from zero: ∣x∣>4|x| > 4. With that cap, 15∣x∣<15⋅4=120\dfrac{1}{5|x|} < \dfrac{1}{5\cdot 4} = \dfrac{1}{20}, so

∣x−5∣5∣x∣<∣x−5∣20.\frac{|x-5|}{5|x|} < \frac{|x-5|}{20}.

We want ∣x−5∣20<ε\dfrac{|x-5|}{20} < \varepsilon, i.e. ∣x−5∣<20ε|x - 5| < 20\varepsilon. So take δ=min⁡(1,20ε)\delta = \min(1, 20\varepsilon). Proof. Let ε>0\varepsilon > 0 be given. Choose

δ=min⁡(1,20ε).\delta = \min(1, 20\varepsilon).

This is positive. Suppose xx is in the domain (so x≠0x \ne 0) and 0<∣x−5∣<δ0 < |x - 5| < \delta. Since δ≤1\delta \le 1: ∣x−5∣<1|x-5|<1, so 4<x<64 < x < 6. In particular x>4>0x > 4 > 0, so ∣x∣=x>4|x| = x > 4, giving

5∣x∣>20  ⟹  15∣x∣<120.5|x| > 20 \implies \frac{1}{5|x|} < \frac{1}{20}.

Since δ≤20ε\delta \le 20\varepsilon: ∣x−5∣<20ε|x - 5| < 20\varepsilon. Now combine:

∣1x−15∣=∣x−5∣5∣x∣<∣x−5∣⋅120<20ε⋅120=ε.\left|\frac{1}{x} - \frac{1}{5}\right| = \frac{|x-5|}{5|x|} < |x - 5| \cdot \frac{1}{20} < 20\varepsilon \cdot \frac{1}{20} = \varepsilon.

(In the middle step we used 15∣x∣<120\frac{1}{5|x|} < \frac{1}{20} together with ∣x−5∣>0|x-5|>0 to multiply the inequality ∣x−5∣5∣x∣<∣x−5∣20\frac{|x-5|}{5|x|} < \frac{|x-5|}{20}, then substituted the bound on ∣x−5∣|x-5|.) Since ε>0\varepsilon > 0 was arbitrary, this proves lim⁡x→51x=15\lim_{x \to 5} \frac{1}{x} = \frac{1}{5}. ■\blacksquare

Figure
Choosing δ = min(1, 20ε) traps x away from the singularity at 0
0 (excluded) 4 5 = a 6 cap δ ≤ 1 keeps x in (4,6), away from 0
Counterexamplewhy aa must be a limit point of the domain

The Uniqueness Theorem requires aa to be a limit point of DD — points of DD other than aa must exist arbitrarily close to aa. Drop this hypothesis, and uniqueness collapses. Let D={0}∪{1}D = \{0\} \cup \{1\}, a two-point set, and define f:D→Rf: D \to \mathbb{R} arbitrarily, say f(0)=3f(0) = 3, f(1)=99f(1) = 99. Consider a=0a = 0. Since 00 is an isolated point of DD (there is no point of DD other than 00 within, say, distance δ=1/2\delta = 1/2 of 00), the statement

0<∣x−0∣<δ  ⟹  ∣f(x)−L∣<ε0 < |x - 0| < \delta \implies |f(x) - L| < \varepsilon

is vacuously true for any LL whatsoever, provided δ≤1/2\delta \le 1/2: there is simply no x∈Dx \in D satisfying 0<∣x∣<δ0 < |x| < \delta, so the implication has a false hypothesis and is automatically true, regardless of what LL is or what ε\varepsilon is. Concretely: fix any ε>0\varepsilon > 0 and take δ=1/2\delta = 1/2. For L=3L = 3: is it true that for all x∈Dx \in D, 0<∣x∣<1/2⇒∣f(x)−3∣<ε0 < |x| < 1/2 \Rightarrow |f(x) - 3| < \varepsilon? The only element of DD within (−12,12)(-\tfrac12,\tfrac12) is x=0x=0 itself, which is excluded by the strict inequality 0<∣x∣0 < |x|. So there is no xx to check, and the implication holds. By the identical argument, it is also true for L=5,000,000L = 5{,}000{,}000, or literally any real number. So by the raw definition, lim⁡x→0f(x)=L\lim_{x \to 0} f(x) = L holds for every real number LL simultaneously when aa is isolated — "the limit" is not a single number at all. This is precisely why the definition of limit is only stated for limit points aa: the hypothesis is not decoration, it is what prevents this vacuous, useless degeneracy. Our proof of uniqueness used it in exactly one place — producing the point x0x_0 with 0<∣x0−a∣<δ0 < |x_0 - a| < \delta — and this counterexample shows that without it, that step (and the whole theorem) fails.

Scope, edge cases & common traps

SituationWhat happensWhy
aa is a limit point of DD (standard case)Limit, if it exists, is uniqueProof above applies directly
aa is an isolated point of DD"Limit" is vacuously satisfied by every LLNo xx near aa to test the implication
ff undefined at aa but aa is a limit point of D∖{a}D \setminus \{a\}Uniqueness still holdsDefinition never uses f(a)f(a); only 0<∣x−a∣0 < |x-a| matters
Choosing δ\delta for non-linear ff without a capBound on ∣f(x)−L∣|f(x)-L| may depend on xx in an unbounded wayExtra factor (e.g. ∣x+5∣|x+5|, 1/∣x∣1/|x|) must be controlled first
Capping δ\delta too generously (e.g. forgetting the cap entirely near a pole)The "constant" bound used isn't actually validMust verify the capped region excludes singularities

The mathematician behind it

Origins · 1 minKarl Weierstrass and how this was discovered
KW
Karl Weierstrass
1815–1897 · German analysis, Berlin school

Weierstrass is the figure most responsible for turning "limit" from an intuitive, motion-based idea — Newton and Leibniz spoke of quantities "flowing" or becoming "infinitely small" — into the static, purely arithmetic ε\varepsilon–δ\delta statement used throughout this article. Working largely without a university post for much of his early career (he taught secondary school in Prussia for over a decade), Weierstrass later became one of the most influential lecturers in Berlin, insisting on total rigor at a time when calculus textbooks were still full of appeals to infinitesimals and geometric intuition. His formulation is the one every modern analysis course, and this article, states as the definition.

History
- 1821: Augustin-Louis Cauchy publishes Cours d'Analyse, giving an early rigorous-in-spirit but still verbal definition of limit.
- 1817–1850s: Bernard Bolzano anticipates several rigorous ideas in relative obscurity, unpublished and largely unread in his lifetime.
- 1850s–1860s: Karl Weierstrass, lecturing in Berlin, formalizes the ε\varepsilon–δ\delta definition in the form used today.
- 1861: Weierstrass's Berlin lectures cement ε\varepsilon–δ\delta arguments as the standard tool for proving uniqueness and continuity results rigorously.
- Late 1800s–present: The ε\varepsilon–δ\delta framework becomes the universal foundation for real analysis, topology, and beyond.
Why can't we just define the limit as "the value the function approaches" without worrying about uniqueness?

That informal phrasing hides an assumption: that there is a single value being approached. Before proving uniqueness, nothing guarantees this. The ε\varepsilon–δ\delta definition is a precise existential/universal statement ("for every ε\varepsilon there's a δ\delta such that...") attached to a specific candidate LL; it's entirely possible, a priori, that two different candidates L1L_1 and L2L_2 both satisfy this statement for the same function and point. The Uniqueness Theorem is what rules this out, retroactively justifying the informal phrase "the value it approaches" as referring to something well-defined. Without the theorem, writing "lim⁡x→af(x)=L\lim_{x\to a} f(x) = L" and later using "the limit" as a definite noun phrase would be logically unjustified.

Does uniqueness of limits require the function to be continuous?

No — continuity is a separate, stronger condition, and it isn't assumed anywhere in the proof. The theorem only requires that aa be a limit point of the function's domain; it says nothing about the function's behavior at aa, or even whether ff is defined at aa at all. In fact, the classic case where limits are computed and shown unique — like lim⁡x→0sin⁡xx\lim_{x\to 0} \frac{\sin x}{x} — involves a function not even defined at the limit point. Uniqueness is purely a statement about the definition's logical structure: whatever number satisfies the ε\varepsilon–δ\delta condition is forced to be one specific number, regardless of how wild or discontinuous ff is elsewhere.

In the quadratic example, why did we cap δ\delta at exactly 11, and would another number work?

The choice of 11 is arbitrary — any fixed positive number would work, it just changes the resulting bound on the variable factor. If we had capped at δ≤1/2\delta \le 1/2 instead, we'd get 1.5<x<2.51.5 < x < 2.5, giving 6.5<x+5<7.56.5 < x+5 < 7.5, so ∣x+5∣<7.5|x+5| < 7.5, leading to δ=min⁡(1/2,ε/7.5)\delta = \min(1/2, \varepsilon/7.5). Both choices produce valid proofs of the same limit; they simply produce different (equally correct) explicit δ\delta's. The convention of capping at 11 is just a common, clean default that keeps arithmetic simple — there is nothing canonical about it, and the existence of a valid δ\delta for each ε\varepsilon is all the definition requires, not a unique or "best" δ\delta.

Is the δ\delta produced in these proofs the largest one that works?

Almost never, and it doesn't need to be. The definition of a limit only requires that some δ>0\delta > 0 work for each ε\varepsilon; it says nothing about optimality. In the linear example lim⁡x→3(4x−5)=7\lim_{x\to3}(4x-5)=7, our δ=ε/4\delta = \varepsilon/4 actually is the largest possible (the bound 4∣x−3∣<ε4|x-3|<\varepsilon is tight), but in the quadratic and reciprocal examples, the capping step (using ∣x+5∣<8|x+5|<8 instead of the true supremum of ∣x+5∣|x+5| on the relevant interval) deliberately sacrifices tightness for algebraic simplicity. Finding the "best possible" δ\delta is rarely the goal in analysis; producing any valid, explicitly justified δ\delta for arbitrary ε\varepsilon is sufficient and is the standard practice.

How does this uniqueness theorem relate to uniqueness of limits of sequences?

It's the same idea in a different but closely related setting. A sequence (an)(a_n) can be viewed as a function f:N→Rf: \mathbb{N} \to \mathbb{R}, and lim⁡n→∞an=L\lim_{n\to\infty} a_n = L has its own ε\varepsilon–NN definition (find NN such that n>N⇒∣an−L∣<εn > N \Rightarrow |a_n - L| < \varepsilon). The uniqueness proof carries over almost verbatim: assume two limits L1≠L2L_1 \ne L_2, set ε=∣L1−L2∣/2\varepsilon = |L_1-L_2|/2, get N1,N2N_1, N_2 from each definition, take N=max⁡(N1,N2)N = \max(N_1,N_2), then any n>Nn > N (and such nn always exists, since N\mathbb{N} is unbounded) yields the same contradiction via the triangle inequality. The "limit point" role is played automatically here since N\mathbb{N} is infinite and unbounded above.

Why do the linear, quadratic, and reciprocal cases require increasingly complicated proofs?

Because the algebraic distance between f(x)f(x) and its claimed limit LL involves increasingly complicated expressions in xx. For linear ff, ∣f(x)−L∣|f(x)-L| is exactly a constant times ∣x−a∣|x-a| — no extra work needed. For quadratic ff, factoring produces ∣x−a∣|x-a| times a linear leftover factor, which varies with xx and must be bounded using a capped neighborhood. For the reciprocal function, the leftover factor is not just variable but can blow up to infinity as x→0x \to 0, so the cap must specifically prevent xx from approaching the domain's singularity. Each additional layer of algebraic complexity in ff translates into one more layer of "control the extra factor first" in the proof.

Once we have the Algebra of Limits, do we still need these direct ε\varepsilon–δ\delta verifications?

Not for computing everyday limits — that is precisely the point of building the Algebra of Limits next: it lets you conclude lim⁡x→2(x2+3x)=10\lim_{x\to2}(x^2+3x) = 10 by combining lim⁡x→2x=2\lim_{x\to2} x = 2 (a nearly immediate ε\varepsilon–δ\delta fact) with sum and product rules, skipping the capping argument entirely. But the algebra-of-limits rules themselves need to be proven, and their proofs use exactly this style of direct ε\varepsilon–δ\delta manipulation. So the shortcuts save work on any specific limit, but the underlying technique demonstrated here remains the engine powering the machinery, and it resurfaces whenever you must verify a limit the algebra rules don't cover (e.g. limits involving unusual or non-algebraic functions).

Before this

Where this leads