Search Knowledge

© 2026 LIBREUNI PROJECT

Introduction to Calculus/Limits and derivatives
Page 1 of 12

Limits and Continuity

A rigorous foundation for calculus using the epsilon-delta definition of limits, continuity, and fundamental limit theorems.

Lesson briefLimits and derivatives · 3 h · 3 outcomes

Required knowledge

  • Real-number inequalities, triangle inequality, and absolute-value manipulation
  • Least upper bound (completeness) property of the real numbers

Learning outcomes

  • Construct and verify epsilon-delta proofs for polynomial, rational, and root limits
  • Apply the Squeeze Theorem, Intermediate Value Theorem, and Extreme Value Theorem with explicit hypotheses
  • Distinguish removable, jump, and essential discontinuities from one-sided limit behavior

Assessment target

Construct an epsilon-delta proof for a non-linear limit and diagnose which theorem hypothesis fails on a non-compact or discontinuous counterexample.

Scope boundary

  • Metric-space topology, Moore-Smith nets, and multivariable limits are deferred to real analysis.

How fast is a falling stone travelling at a single instant? Average velocity over a time interval is easy — divide distance by duration — but at one instant both quantities are zero, and 0/00/0 is undefined. The same tension appears everywhere in calculus: tangent slopes, areas, infinite sums. Each is a limit: the value a quantity approaches as the input approaches a target.

This lesson develops the limit concept precisely enough to prove things about it, then builds the two structural theorems of continuous functions — the Intermediate Value Theorem and the Extreme Value Theorem — that the rest of calculus depends on.

Prerequisites. Inequalities, absolute value, and the completeness of R\mathbb{R} (every nonempty bounded set has a supremum). No derivative or integral machinery is used.

The Formal Definition of a Limit

Intuitively, lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L means we can make f(x)f(x) arbitrarily close to LL by making xx sufficiently close to cc The obstacle in making this precise is circularity: “sufficiently close” must not depend on the value f(c)f(c) which may not even exist. The resolution, due to Weierstrass, is to quantify the two tolerances and order them: for every output tolerance ϵ\epsilon there exists an input tolerance δ\delta — and δ\delta must be chosen before seeing any particular xx

Definition 1.1.1Limit of a Function

Let f:D→Rf: D \to \mathbb{R} with D⊆RD \subseteq \mathbb{R} and let cc be a limit point of DD (every neighborhood of cc contains points of DD other than cc Then

lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L

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

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

Three boundary cases deserve explicit attention:

  • The condition 0<∣x−c∣0 < |x - c| excludes x=cx = c. The limit ignores what happens at the point — ff need not even be defined there. This is exactly what lets tangent slopes and instantaneous velocity make sense.
  • ”cc is a limit point” prevents vacuous truth: if cc were isolated in DD some δ\delta-neighborhood would contain no other xx and the implication would hold trivially for any LL
  • δ\delta may depend on ϵ\epsilon (and on cc but a single δ\delta must work for all xx in the window simultaneously — the condition is uniform in xx

The definition is best internalized as a two-player game. Your opponent picks ϵ\epsilon shrinking the target zone L±ϵL \pm \epsilon on the yy-axis; you must respond with δ\delta a window c±δc \pm \delta on the xx-axis, so that every input from the window (except cc lands inside the target zone. You win if you have a response to every ϵ>0\epsilon > 0 [Spivak (4th ed.), Ch. 5 “Limits”, §“Delta–Epsilon Proofs”]

Worked Example 1: a linear function

Claim. lim⁡x→2(3x−1)=5\lim_{x \to 2} (3x - 1) = 5

Given ϵ>0\epsilon > 0 we must force ∣f(x)−5∣<ϵ|f(x) - 5| < \epsilon by bounding ∣x−2∣|x - 2| Start from the quantity we control:

∣f(x)−5∣=∣(3x−1)−5∣=∣3x−6∣=3 ∣x−2∣.|f(x) - 5| = |(3x - 1) - 5| = |3x - 6| = 3\,|x - 2|.

So ∣f(x)−5∣<ϵ|f(x) - 5| < \epsilon whenever ∣x−2∣<ϵ/3|x - 2| < \epsilon/3 Choose δ=ϵ/3\delta = \epsilon/3 Then 0<∣x−2∣<δ0 < |x - 2| < \delta gives ∣f(x)−5∣=3∣x−2∣<3δ=ϵ|f(x) - 5| = 3|x - 2| < 3\delta = \epsilon for every xx at once. The limit is proved.

Concretely: if the opponent demands ϵ=0.3\epsilon = 0.3 respond with δ=0.1\delta = 0.1 any xx within 0.10.1 of 22 gives f(x)f(x) within 0.30.3 of 55

Verification Check

0 / 4 answered

For f(x)=3x−1f(x)=3x-1 at c=2c=2 with target tolerance ϵ=0.30\epsilon = 0.30: evaluate which proposed δ\delta values guarantee the bound.

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
δ=3.00\delta = 3.00 is a valid choice satisfying the tolerance condition ∣f(x)−5∣<0.30|f(x)-5| < 0.30.
2
δ=0.05\delta = 0.05 is a valid choice satisfying the tolerance condition ∣f(x)−5∣<0.30|f(x)-5| < 0.30.
3
δ=0.30\delta = 0.30 is a valid choice satisfying the tolerance condition ∣f(x)−5∣<0.30|f(x)-5| < 0.30.
4
δ=0.10\delta = 0.10 is a valid choice satisfying the tolerance condition ∣f(x)−5∣<0.30|f(x)-5| < 0.30.

Worked Example 2: a nonlinear function

Claim. lim⁡x→3x2=9\lim_{x \to 3} x^2 = 9

Here ∣f(x)−L∣=∣x2−9∣=∣x−3∣ ∣x+3∣|f(x) - L| = |x^2 - 9| = |x - 3|\,|x + 3| a variable factor multiplies the quantity we control. The trick is to freeze it. Impose the preliminary restriction δ≤1\delta \le 1 then ∣x−3∣<1|x - 3| < 1 means 2<x<42 < x < 4 so ∣x+3∣<7|x + 3| < 7 Now

∣x−3∣<δ  ⟹  ∣x2−9∣=∣x−3∣ ∣x+3∣<7 δ.|x - 3| < \delta \implies |x^2 - 9| = |x-3|\,|x+3| < 7\,\delta.

Choose δ=min⁡(1,ϵ/7)\delta = \min(1, \epsilon/7) Then δ≤1\delta \le 1 activates the bound ∣x+3∣<7|x+3| < 7 and 7δ≤ϵ7\delta \le \epsilon finishes: ∣x2−9∣<ϵ|x^2 - 9| < \epsilon for every xx with 0<∣x−3∣<δ0 < |x-3| < \delta

Compare the two proofs. In Example 1 the response was proportional δ=ϵ/3\delta = \epsilon/3 in Example 2 it is not, because the sensitivity of x2x^2 to changes in xx varies along the window. The further cc is from 00 the smaller the working δ\delta for the same ϵ\epsilon — the definition captures local sensitivity, not a global property of the formula.

Epsilon-delta verification for $\lim_{x \to 3} x^2 = 9$ with $\varepsilon = 1$ and preliminary bound $\delta = \frac{1}{7}$. Epsilon-delta verification for the limit of x squared as x approaches 3 equals 9, illustrating the tolerance band and delta window. c=3 L=9 c-δ c+δ c-δ* c+δ*
Figure 1.1.1: Epsilon-delta verification for lim⁡x→3x2=9\lim_{x \to 3} x^2 = 9 with ε=1\varepsilon = 1 and preliminary bound δ=17\delta = \frac{1}{7}.

Verification Check

0 / 4 answered

In proving lim⁡x→3x2=9\lim_{x \to 3} x^2 = 9, evaluate each statement concerning the preliminary bound δ≤1\delta \le 1:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
It guarantees that xx never approaches within zero distance of origin x=0x=0.
2
It is required because the definition of limit forbids any radius exceeding one.
3
It decouples the local scale so δ=min⁡(1,ϵ/7)\delta = \min(1, \epsilon/7) controls error.
4
It restricts xx to the open interval (2,4)(2, 4), bounding ∣x+3∣<7|x+3| < 7.

The geometry of this game, for the linear example: the horizontal ϵ\epsilon-band around L=5L = 5 pulls back through the line to a vertical δ\delta-window around c=2c = 2 and δ=ϵ/3\delta = \epsilon/3 is the exact threshold.

Interactive exploration

Move the ε-band

Change ε and watch the corresponding δ-window shrink or expand.

ε = 0.45, δ = 0.15
ε-bandδ-window
Move the ε-band
ε-bandδ-window

For f(x)=3x−1 at c=2, the δ-window pulls back the ε-band around L=5.

Fundamental Limit Theorems

Computing limits directly from the ϵ\epsilon-δ\delta definition for every new function would be unusable. The algebra of limits lets us assemble complicated limits from simple ones, provided the pieces exist and the operations are defined.

Theorem 1.1.2Algebra of Limits

Builds on:Definition 1.1.1

Suppose lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L and lim⁡x→cg(x)=M\lim_{x \to c} g(x) = M and cc is a limit point of the relevant domains. Then:

  1. lim⁡x→c(f+g)(x)=L+M\lim_{x \to c} (f + g)(x) = L + M
  2. lim⁡x→c(f⋅g)(x)=L⋅M\lim_{x \to c} (f \cdot g)(x) = L \cdot M
  3. if additionally M≠0M \ne 0 then lim⁡x→c(f/g)(x)=L/M\lim_{x \to c} (f/g)(x) = L/M
ProofExpand proof

We prove the sum rule; the product and quotient rules follow the same pattern with an extra preliminary bound (Problem 1.P.1).

Let ϵ>0\epsilon > 0

  1. Since lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L there exists δ1>0\delta_1 > 0 such that 0<∣x−c∣<δ1  ⟹  ∣f(x)−L∣<ϵ/20 < |x - c| < \delta_1 \implies |f(x) - L| < \epsilon/2
  2. Since lim⁡x→cg(x)=M\lim_{x \to c} g(x) = M there exists δ2>0\delta_2 > 0 such that 0<∣x−c∣<δ2  ⟹  ∣g(x)−M∣<ϵ/20 < |x - c| < \delta_2 \implies |g(x) - M| < \epsilon/2
  3. Set δ=min⁡(δ1,δ2)\delta = \min(\delta_1, \delta_2) For 0<∣x−c∣<δ0 < |x - c| < \delta both bounds hold, so the triangle inequality gives:

∣(f+g)(x)−(L+M)∣≤∣f(x)−L∣+∣g(x)−M∣<ϵ/2+ϵ/2=ϵ.|(f+g)(x) - (L+M)| \le |f(x) - L| + |g(x) - M| < \epsilon/2 + \epsilon/2 = \epsilon.

Thus lim⁡x→c(f+g)(x)=L+M\lim_{x \to c} (f+g)(x) = L + M

∎

The quotient rule’s hypothesis M≠0M \ne 0 is essential and is the most common source of wrong limits. The classic lim⁡x→0sin⁡xx\lim_{x \to 0} \frac{\sin x}{x} cannot be evaluated by the quotient rule: lim⁡x→0sin⁡x=0\lim_{x \to 0} \sin x = 0 and lim⁡x→0x=0\lim_{x \to 0} x = 0 so rule 3 does not apply. The correct value is 11 but it must be proved by squeezing, not by algebra [Spivak (4th ed.), Ch. 15, §“The Derivatives of Trigonometric Functions”]. The algebra of limits never resolves a 0/00/0 form; that is precisely the case calculus exists to handle.

Verification Check

0 / 4 answered

A proposed calculation posits lim⁡x→0sin⁡xx=lim⁡x→0sin⁡xlim⁡x→0x=00\lim_{x\to 0}\frac{\sin x}{x} = \frac{\lim_{x\to 0}\sin x}{\lim_{x\to 0} x} = \frac{0}{0} and concludes the limit does not exist. Evaluate each claim about this calculation:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
The quotient rule is universally valid for any real numerator and denominator.
2
The quotient rule cannot be applied because denominator limit is zero.
3
An indeterminate form 0/00/0 does not imply that the limit diverges.
4
The calculation fails because numerator limit lim⁡x→0sin⁡x\lim_{x \to 0} \sin x is undefined.
Theorem 1.1.3Squeeze Theorem

Builds on:Definition 1.1.1

Let II be an open interval containing cc except possibly at cc itself. Let f,g,hf, g, h be functions defined on II If for all x∈I∖{c}x \in I \setminus \{c\}

g(x)≤f(x)≤h(x)g(x) \le f(x) \le h(x)

and lim⁡x→cg(x)=lim⁡x→ch(x)=L\lim_{x \to c} g(x) = \lim_{x \to c} h(x) = L then:

lim⁡x→cf(x)=L.\lim_{x \to c} f(x) = L.

ProofExpand proof

Let ϵ>0\epsilon > 0 be given.

  1. Since lim⁡x→cg(x)=L\lim_{x \to c} g(x) = L there exists δ1>0\delta_1 > 0 such that for all 0<∣x−c∣<δ10 < |x - c| < \delta_1
∣g(x)−L∣<ϵ  ⟺  L−ϵ<g(x)<L+ϵ.|g(x) - L| < \epsilon \iff L - \epsilon < g(x) < L + \epsilon.
  1. Since lim⁡x→ch(x)=L\lim_{x \to c} h(x) = L there exists δ2>0\delta_2 > 0 such that for all 0<∣x−c∣<δ20 < |x - c| < \delta_2
∣h(x)−L∣<ϵ  ⟺  L−ϵ<h(x)<L+ϵ.|h(x) - L| < \epsilon \iff L - \epsilon < h(x) < L + \epsilon.
  1. Choose δ=min⁡(δ1,δ2)>0\delta = \min(\delta_1, \delta_2) > 0 For all xx satisfying 0<∣x−c∣<δ0 < |x - c| < \delta both inequalities hold simultaneously. Combining with the sandwiching hypothesis g(x)≤f(x)≤h(x)g(x) \le f(x) \le h(x)
L−ϵ<g(x)≤f(x)≤h(x)<L+ϵ.L - \epsilon < g(x) \le f(x) \le h(x) < L + \epsilon.

Extracting the outer bounds yields:

L−ϵ<f(x)<L+ϵ  ⟺  ∣f(x)−L∣<ϵ.L - \epsilon < f(x) < L + \epsilon \iff |f(x) - L| < \epsilon.

This establishes that lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L

∎

The Squeeze Theorem is the standard tool for functions that oscillate too fast to control directly. A classic application: lim⁡x→0xsin⁡(1/x)=0\lim_{x \to 0} x \sin(1/x) = 0 (and similarly lim⁡x→0x2sin⁡(1/x)=0\lim_{x \to 0} x^2 \sin(1/x) = 0 Since −1≤sin⁡(1/x)≤1-1 \le \sin(1/x) \le 1 for all x≠0x \ne 0 multiplying by ∣x∣>0|x| > 0 yields −∣x∣≤xsin⁡(1/x)≤∣x∣-|x| \le x \sin(1/x) \le |x| (or −x2≤x2sin⁡(1/x)≤x2-x^2 \le x^2 \sin(1/x) \le x^2 Both bounds tend to 00 as x→0x \to 0 so the middle function does too — regardless of how fast it oscillates inside the envelope. The theorem transfers existence, not rates: it says nothing about how fast ff approaches LL

Contrast with a function the squeeze cannot rescue: sin⁡(1/x)\sin(1/x) itself has no limit at 00 — no envelope tighter than [−1,1][-1, 1] is available, and the values cycle through all of [−1,1][-1, 1] in every neighborhood of 00 (see the counterexample below).

The squeeze theorem: $-|x| \le x \sin(1/x) \le |x|$, trapping the oscillating function at the origin. Graph illustrating the squeeze theorem with x sin(1/x) bounded above by |x| and below by -|x| on [-0.4, 0.4].
Figure 1.1.2: The squeeze theorem: −∣x∣≤xsin⁡(1/x)≤∣x∣-|x| \le x \sin(1/x) \le |x|, trapping the oscillating function at the origin.

Continuity

Continuity formalises “no gaps or jumps” — the input–output contract that makes the two structural theorems below possible.

Definition 1.1.4Continuity

A function f:D→Rf: D \to \mathbb{R} is continuous at a point c∈Dc \in D if:

lim⁡x→cf(x)=f(c).\lim_{x \to c} f(x) = f(c).

The function is continuous on an interval I⊆DI \subseteq D if it is continuous at every c∈Ic \in I

Unpacking the definition: continuity at cc requires (i) f(c)f(c) is defined, (ii) lim⁡x→cf(x)\lim_{x \to c} f(x) exists, and (iii) the two are equal. Each requirement fails in a different, characteristic way, giving a taxonomy of discontinuities:

TypeDefinition failureCanonical exampleContinuous extension possible?
Removable(i) or (iii): limit exists but f(c)f(c) is undefined or differsf(x)=(x2−1)/(x−1)f(x) = (x^2-1)/(x-1) at c=1c=1 (limit =2= 2 or sin⁡(x)/x\sin(x)/x at c=0c=0 (limit =1= 1Yes: (re)define f(c)=lim⁡x→cf(x)f(c) = \lim_{x \to c} f(x)
Jump(ii): one-sided limits exist but differf(x)=sgn⁡(x)f(x) = \operatorname{sgn}(x) at c=0c=0 limits ∓1\mp 1No
Essential(ii): limit does not exist, even one-sidedf(x)=sin⁡(1/x)f(x) = \sin(1/x) at c=0c=0No
Three types of discontinuity: removable (hole at $(1, 2)$), jump (finite gap at $x = 0$), and essential (infinite oscillation as $x \to 0$). Comparison of three discontinuity types: removable at (1, 2), jump discontinuity at x = 0, and essential oscillation of sin(1/x). x=1
Figure 1.1.3: Three types of discontinuity: removable (hole at (1,2)(1, 2)), jump (finite gap at x=0x = 0), and essential (infinite oscillation as x→0x \to 0).

Sums and products of continuous functions are continuous (transfer the Algebra of Limits termwise); quotients are continuous only where the denominator is nonzero — the quotient-rule hypothesis again.

Counterexample: essential discontinuity

Consider f(x)=sin⁡(1/x)f(x) = \sin(1/x) for x≠0x \ne 0 The limit as x→0x \to 0 does not exist, so no definition of f(0)f(0) can make ff continuous — the discontinuity is essential.

Verification Check

0 / 4 answered

Consider f(x)=sin⁡(1/x)f(x)=\sin(1/x) on R∖{0}\mathbb{R}\setminus\{0\}. Evaluate each claim regarding its limiting behavior as x→0x \to 0:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
Sequences xn=1/(π/2+2πn)x_n = 1/(\pi/2 + 2\pi n) and yn=1/(3π/2+2πn)y_n = 1/(3\pi/2 + 2\pi n) converge to different limits.
2
The limit fails to exist because values grow unboundedly towards positive or negative infinity.
3
The limit fails solely because the function lacks a designated evaluation value at x=0x = 0.
4
Every punctured neighborhood (−δ,δ)∖{0}(-\delta, \delta) \setminus \{0\} contains values spanning the full range [−1,1][-1, 1].

Compare with x2sin⁡(1/x)x^2 \sin(1/x) multiplying the wild factor by x2x^2 tames it enough for the Squeeze Theorem, and the extension F(0)=0F(0) = 0 is continuous. The boundary between “tameable” and “untameable” oscillation is exactly what a squeeze envelope can control — a distinction that returns when derivatives are defined.

The Intermediate Value Theorem (IVT)

Continuous functions on closed intervals cannot skip values. This is the theorem that guarantees roots exist before we compute them.

Theorem 1.1.5Intermediate Value Theorem (IVT)

Builds on:Definition 1.1.4

If f:[a,b]→Rf: [a, b] \to \mathbb{R} is continuous and uu is any number between f(a)f(a) and f(b)f(b) then there exists at least one c∈(a,b)c \in (a, b) such that:

f(c)=u.f(c) = u.

The proof [Spivak (4th ed.), Ch. 8 “Least Upper Bound”, Theorem 1; Rudin, PMA (3rd ed.), Ch. 4, Thm 4.23] constructs c=sup⁡{x∈[a,b]:f(x)<u}c = \sup\{x \in [a,b] : f(x) < u\} — a set that is nonempty and bounded by completeness — and shows continuity forces f(c)=uf(c) = u exactly. Completeness is not decoration: on Q\mathbb{Q} the theorem fails f(x)=x2f(x) = x^2 is continuous on [1,2]∩Q[1,2] \cap \mathbb{Q} takes endpoint values 11 and 44 but no rational has square 22

Both hypotheses are load-bearing:

  • Discontinuity: f(x)=sgn⁡(x)f(x) = \operatorname{sgn}(x) on [−1,1][-1, 1] jumps from −1-1 to 11 and never takes the value u=0u = 0 between its endpoint values.
  • Endpoint values: the theorem cannot even be stated without f(a)f(a) f(b)f(b) — this is where “closed interval” enters.

Verification Check

0 / 4 answered

A continuous function on [−2,2][-2, 2] satisfies f(−2)=3f(-2) = 3 and f(2)=−1f(2) = -1. Evaluate each statement:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
The function can avoid crossing zero by oscillating an even number of times across the interval.
2
The theorem fails to apply because the initial endpoint value exceeds the final endpoint value.
3
The intermediate value property guarantees that every real value between −1-1 and 33 is attained.
4
By the Intermediate Value Theorem, there exists at least one root c∈(−2,2)c \in (-2, 2) where f(c)=0f(c) = 0.

The IVT also gives a numerical algorithm. If ff is continuous and f(a)f(a) f(b)f(b) have opposite signs, the sign change survives on exactly one half of [a,b][a,b] after bisecting at m=(a+b)/2m = (a+b)/2 and that half must contain a root. Repeating halves the error each step: after nn bisections the root is pinned to within (b−a)/2n(b-a)/2^n so a tolerance τ\tau is guaranteed by any n≥log⁡2((b−a)/τ)n \ge \log_2((b-a)/\tau) [Burden & Faires, Numerical Analysis (9th ed.), §2.1 “The Bisection Method”]

python

Interactive Lab

1def bisection_method(f, a, b, tol=1e-5):
2 if f(a) * f(b) >= 0:
3 raise ValueError("IVT condition not met: f(a) and f(b) must have opposite signs.")
4 
5 while (b - a) / 2.0 > tol:
6 midpoint = (a + b) / 2.0
7 if f(midpoint) == 0:
8 return midpoint
9 elif f(a) * f(midpoint) < 0:
10 b = midpoint
11 else:
12 a = midpoint
13 return (a + b) / 2.0
14 
15# Find the root of f(x) = x^3 - x - 2 on [1, 2].
16# Check: f(1) = -2 < 0 < f(2) = 4, so the IVT guarantees a root in (1, 2).
17f = lambda x: x**3 - x - 2
18root = bisection_method(f, 1, 2)
19print(f"Calculated Root: {root:.6f}")
20print(f"f(root) = {f(root):.6f}")
21 

Verify the error bound. The loop exits when (b−a)/2≤10−5(b-a)/2 \le 10^{-5} since the interval starts with length 11 and each pass halves it, after nn passes the length is 2−n2^{-n} and 2−17<10−52^{-17} < 10^{-5} So the printed root is within 10−510^{-5} of a true root of x3−x−2x^3 - x - 2 — a claim about the real root, justified by the IVT plus halving arithmetic, not by inspecting the output.

The Intermediate Value Theorem guarantees a root $r \approx 1.5214$ in $[1, 2]$ because $f(1) = -2 < 0 < 4 = f(2)$. Intermediate Value Theorem guaranteed root of the cubic polynomial x cubed minus x minus 2 on the interval [1, 2]. r ≈ 1.521
Figure 1.1.4: The Intermediate Value Theorem guarantees a root r≈1.5214r \approx 1.5214 in [1,2][1, 2] because f(1)=−2<0<4=f(2)f(1) = -2 < 0 < 4 = f(2).

The Extreme Value Theorem (EVT)

Continuity on a closed, bounded interval does more than prevent skipped values — it keeps ff bounded and forces the bounds to be attained. [Spivak (4th ed.), Ch. 7 “Three Hard Theorems”, Thms 1–2; Rudin, PMA (3rd ed.), Ch. 4, Thms 4.16–4.17]

Theorem 1.1.6Extreme Value Theorem (EVT)

Builds on:Definition 1.1.4

If f:[a,b]→Rf: [a, b] \to \mathbb{R} is continuous, then ff attains a maximum and a minimum: there exist points xmin⁡,xmax⁡∈[a,b]x_{\min}, x_{\max} \in [a, b] such that

f(xmin⁡)≤f(x)≤f(xmax⁡)for all x∈[a,b].f(x_{\min}) \le f(x) \le f(x_{\max}) \quad \text{for all } x \in [a, b].

The theorem has three hypotheses — continuity, closedness, boundedness of the interval — and each is falsifiable by a concrete counterexample:

  • Open or half-open interval fails. f(x)=xf(x) = x on (0,1)(0, 1) is continuous and bounded 0<f<10 < f < 1 but no point attains 00 or 11 even more dramatically, f(x)=1/xf(x) = 1/x on (0,1](0, 1] is continuous on its domain but blows up as x→0+x \to 0^+ so it is not even bounded above (shown below).
  • Unbounded interval fails. f(x)=xf(x) = x on [0,∞)[0, \infty) is continuous but unbounded above.
  • Discontinuity fails. Define f(x)=xf(x) = x on the open part (0,1)(0, 1) and f(0)=f(1)=−1f(0) = f(1) = -1 This ff is bounded on a closed interval, but its values approach 11 from below without ever attaining it: the jump at the endpoints dodges the supremum, so no maximum exists.

Like the IVT, the EVT is a completeness theorem: its proof [Spivak (4th ed.), Ch. 8; Rudin, PMA, Ch. 4, Thm 4.16] shows the set f([a,b])f([a,b]) is bounded and that sup⁡f([a,b])\sup f([a,b]) is a value of ff — both steps requiring least upper bounds.

Failure of the Extreme Value Theorem on $(0, 1]$: $f(x) = \frac{1}{x}$ is continuous but fails to attain a maximum as $x \to 0^+$. Failure of the Extreme Value Theorem for the reciprocal function 1/x on the non-compact open interval (0, 1]. x=1
Figure 1.1.5: Failure of the Extreme Value Theorem on (0,1](0, 1]: f(x)=1xf(x) = \frac{1}{x} is continuous but fails to attain a maximum as x→0+x \to 0^+.

Verification Check

0 / 4 answered

For f(x)=x2f(x) = x^2 on the open interval (−1,1)(-1, 1), evaluate each statement:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
The function attains an absolute minimum value 00 at the interior point x=0x = 0.
2
The Extreme Value Theorem guarantees that both absolute extrema must be attained on this interval.
3
The function attains an absolute maximum value 11 at the boundary endpoints x=±1x = \pm 1.
4
The function possesses a least upper bound sup⁡f=1\sup f = 1 that is never attained as a maximum.

The IVT and EVT are complementary guarantees: the IVT says the image f([a,b])f([a,b]) has no gaps between attained values, the EVT says the image is bounded and contains its own endpoints. Together they make the image of a compact interval a closed, bounded — and, for the derivative-theorems to come, decisive — object.

Section Exercises

Deep Practice

Solve these problems to test and consolidate your understanding of epsilon-delta limits, theorem hypotheses, and continuity contracts. Complete worked solutions and verification criteria are available for each problem.

Exercise 1.1.1: Linear Epsilon-Delta Verification

Verificationapplication

Prove directly from the formal definition of a limit (Definition 1.1.1) that lim⁡x→1(2x+3)=5\lim_{x \to 1} (2x + 3) = 5

HintExpand

Compute ∣f(x)−5∣|f(x) - 5| in terms of ∣x−1∣|x - 1| first; the constant 33 is eliminated before choosing δ\delta.

Worked solutionExpand

Given ϵ>0\epsilon > 0 choose δ=ϵ/2\delta = \epsilon/2 Then 0<∣x−1∣<δ0 < |x - 1| < \delta gives

∣(2x+3)−5∣=∣2x−2∣=2 ∣x−1∣<2δ=ϵ.|(2x + 3) - 5| = |2x - 2| = 2\,|x - 1| < 2\delta = \epsilon.

Verification criteria. A complete derivation requires: (a) an explicit δ\delta depending on ϵ\epsilon (b) the implication stated for all xx with 0<∣x−1∣<δ0 < |x-1| < \delta not just one; (c) the chain of equalities ending in a quantity <ϵ< \epsilon A response that verifies the limit only numerically (e.g., plugging in x=1.01x = 1.01 is invalid.

Exercise 1.1.2: Quadratic Sensitivity and Preliminary Bound

Constructionapplication

Prove from Definition 1.1.1 that lim⁡x→2x2=4\lim_{x \to 2} x^2 = 4 giving δ\delta explicitly in terms of ϵ\epsilon

HintExpand

Factor ∣x2−4∣=∣x−2∣ ∣x+2∣|x^2 - 4| = |x-2|\,|x+2| and freeze the variable factor with a preliminary bound δ≤1\delta \le 1, as in Worked Example 2.

Worked solutionExpand

Given ϵ>0\epsilon > 0 choose δ=min⁡(1,ϵ/5)\delta = \min(1, \epsilon/5) If ∣x−2∣<δ≤1|x - 2| < \delta \le 1 then 1<x<31 < x < 3 so ∣x+2∣<5|x + 2| < 5 and

∣x2−4∣=∣x−2∣ ∣x+2∣<5 δ≤5⋅ϵ5=ϵ.|x^2 - 4| = |x-2|\,|x+2| < 5\,\delta \le 5 \cdot \frac{\epsilon}{5} = \epsilon.

Verification criteria. Any correct δ\delta (e.g., min⁡(1,ϵ/5)\min(1, \epsilon/5) is valid; a response of δ=ϵ/5\delta = \epsilon/5 without the preliminary bound is a defect, because ∣x+2∣<5|x+2| < 5 was used without justification — that inequality needs ∣x−2∣<1|x-2| < 1 A response of δ=ϵ/4\delta = \epsilon/4 is falsifiable: with ϵ=0.4\epsilon = 0.4 take x=2.1x = 2.1 then ∣x2−4∣=0.41>0.4|x^2 - 4| = 0.41 > 0.4

Exercise 1.1.3: Oscillating Squeeze vs Removable Discontinuity

Descriptiontransfer

Define f(x)=xsin⁡(1/x)f(x) = x \sin(1/x) for x≠0x \ne 0 Show that lim⁡x→0f(x)=0\lim_{x \to 0} f(x) = 0 and state where ff can be extended continuously. Then contrast: does sin⁡(1/x)\sin(1/x) have a limit at 00

HintExpand

Bound ∣xsin⁡(1/x)∣|x \sin(1/x)| by a quantity that ignores the sine, then apply the Squeeze Theorem (Theorem 1.1.3).

Worked solutionExpand
Show squeeze diagram(Click to reveal)
The squeeze theorem: $-|x| \le x \sin(1/x) \le |x|$, trapping the oscillating function at the origin. Graph illustrating the squeeze theorem with x sin(1/x) bounded above by |x| and below by -|x| on [-0.4, 0.4].
Figure 1.1.2: The squeeze theorem: −∣x∣≤xsin⁡(1/x)≤∣x∣-|x| \le x \sin(1/x) \le |x|, trapping the oscillating function at the origin.

For x≠0x \ne 0 ∣f(x)∣=∣x∣ ∣sin⁡(1/x)∣≤∣x∣⋅1=∣x∣|f(x)| = |x|\,|\sin(1/x)| \le |x| \cdot 1 = |x| since ∣sin⁡t∣≤1|\sin t| \le 1 for all tt Both bound functions ∓∣x∣\mp|x| tend to 00 as x→0x \to 0 so the Squeeze Theorem (Theorem 1.1.3) gives lim⁡x→0f(x)=0\lim_{x \to 0} f(x) = 0 Since f(0)f(0) is undefined but the limit exists, the extension F(0):=0F(0) := 0 makes FF continuous at 00 (removable discontinuity). By contrast sin⁡(1/x)\sin(1/x) has no limit at 00 its values cycle through [−1,1][-1, 1] in every neighborhood, so no candidate LL survives ϵ=1/2\epsilon = 1/2

Verification criteria. Requires: (a) the explicit bound ∣sin⁡t∣≤1|\sin t| \le 1 invoked with the squeeze; (b) the conclusion stated for the extended function; (c) a falsifying ϵ\epsilon (any ϵ<1\epsilon < 1 works) for sin⁡(1/x)\sin(1/x) Claiming ”sin⁡(1/x)\sin(1/x) is undefined at 00 so no limit” is a failure: undefinedness at cc does not preclude a limit.

Exercise 1.1.4: Guaranteed vs Unguaranteed Intermediate Values

Example / Impossibletransfer

A continuous function gg on [−1,1][-1, 1] satisfies g(−1)=−0.5g(-1) = -0.5 and g(1)=0.5g(1) = 0.5 For each value uu decide whether the Intermediate Value Theorem (Theorem 1.1.5) guarantees a solution of g(x)=ug(x) = u on [−1,1][-1, 1] (a) u=0u = 0 (b) u=0.9u = 0.9 (c) u=−0.5u = -0.5

HintExpand

Check each uu against the hypothesis: is uu between g(−1)g(-1) and g(1)g(1)?

Worked solutionExpand
Show IVT root diagram(Click to reveal)
The Intermediate Value Theorem guarantees a root $r \approx 1.5214$ in $[1, 2]$ because $f(1) = -2 < 0 < 4 = f(2)$. Intermediate Value Theorem guaranteed root of the cubic polynomial x cubed minus x minus 2 on the interval [1, 2]. r ≈ 1.521
Figure 1.1.4: The Intermediate Value Theorem guarantees a root r≈1.5214r \approx 1.5214 in [1,2][1, 2] because f(1)=−2<0<4=f(2)f(1) = -2 < 0 < 4 = f(2).
  • (a) Yes: −0.5<0<0.5-0.5 < 0 < 0.5 so Theorem 1.1.5 guarantees some c∈(−1,1)c \in (-1, 1) with g(c)=0g(c) = 0
  • (b) No: 0.90.9 is not between g(−1)=−0.5g(-1) = -0.5 and g(1)=0.5g(1) = 0.5 The IVT is silent — a solution may or may not exist depending on gg (if g(x)=xg(x) = x none exists).
  • (c) Yes — but for a different reason: u=−0.5=g(−1)u = -0.5 = g(-1) is attained at the endpoint, not guaranteed inside (a,b)(a,b) The IVT as stated locates interior points for values strictly between f(a)f(a) and f(b)f(b) endpoint values are already in the image by definition.

Verification criteria. Each part scores independently. For (a): correct yes/no with the hypothesis quoted. For (b): a bare “no” without the reason (”uu not between the endpoint values”) is incomplete; a “no, because gg is discontinuous” is a failure gg is continuous by hypothesis). For (c): distinguishing guaranteed by IVT from true but not via IVT is the transfer step.

Exercise 1.1.5: Fixed-Point Existence via Intermediate Values

Proofsynthesis

Prove, using Theorem 1.1.5, that the equation x=cos⁡xx = \cos x has a solution in [0,1][0, 1] Define the auxiliary function yourself; you may use that cos⁡x\cos x is continuous on [0,1][0, 1] with cos⁡0=1\cos 0 = 1 and cos⁡1<1\cos 1 < 1 cos⁡1≈0.54\cos 1 \approx 0.54

HintExpand

Turn the fixed-point equation into a root-finding problem: move everything to one side and apply the IVT to the resulting continuous function.

Worked solutionExpand

Let h(x)=x−cos⁡xh(x) = x - \cos x As a difference of continuous functions xx and cos⁡x\cos x hh is continuous on [0,1][0, 1] At the endpoints: h(0)=0−1=−1<0h(0) = 0 - 1 = -1 < 0 and h(1)=1−cos⁡1>0h(1) = 1 - \cos 1 > 0 since cos⁡1<1\cos 1 < 1 So u=0u = 0 lies strictly between h(0)h(0) and h(1)h(1) and Theorem 1.1.5 guarantees some c∈(0,1)c \in (0, 1) with h(c)=0h(c) = 0 i.e., c=cos⁡cc = \cos c

Verification criteria. Requires: (a) an explicitly defined auxiliary function; (b) a stated continuity argument (sum/difference of continuous functions); (c) strict endpoint signs with justification cos⁡1<1\cos 1 < 1 not "≈\approx"); (d) the IVT invoked on u=0u = 0 between the endpoint values. Asserting “since cos⁡\cos is continuous it has a fixed point” without the sign change is a failure — continuity alone on a non-closed or badly-oriented domain does not give fixed points (compare x↦x+1x \mapsto x + 1

Exercise 1.1.6: Continuity of Modified Dirichlet Function

Descriptiontransfer

Let f:R→Rf: \mathbb{R} \to \mathbb{R} be defined by f(x)=xf(x) = x if xx is rational, and f(x)=0f(x) = 0 if xx is irrational. At which points is ff continuous according to Definition 1.1.4?

Verification Check

0 / 4 answered

Let f(x)=xf(x)=x for x∈Qx \in \mathbb{Q} and f(x)=0f(x)=0 for x∉Qx \notin \mathbb{Q}. Evaluate each statement about continuity:

Classify each option as correct or incorrect, or leave uncertain options unanswered (“I don't know”) and verify at any time.

1
The function is continuous at x=0x = 0 because ∣f(x)−f(0)∣≤∣x∣<ϵ|f(x) - f(0)| \le |x| < \epsilon whenever ∣x∣<ϵ|x| < \epsilon.
2
The function is continuous at all nonzero rational numbers because rational points map continuously to f(x)=xf(x) = x.
3
The function is discontinuous at every nonzero point c≠0c \ne 0 due to conflicting rational and irrational sequences.
4
The function fails to be continuous anywhere on the real line because rational and irrational numbers are everywhere dense.

Exercise 1.1.7: Logical Negation of the Limit Definition

Negationrecall

Formulate the precise logical negation of Definition 1.1.1 lim⁡x→cf(x)=L\lim_{x \to c} f(x) = L using quantifiers. Then apply your negated statement to prove that lim⁡x→0sgn⁡(x)≠0\lim_{x \to 0} \operatorname{sgn}(x) \ne 0

HintExpand

Flipping '∀ϵ>0,∃δ>0,…\forall \epsilon > 0, \exists \delta > 0, \dots' yields '∃ϵ0>0,∀δ>0,…\exists \epsilon_0 > 0, \forall \delta > 0, \dots'. Identify a specific ϵ0\epsilon_0 that sgn⁡(x)\operatorname{sgn}(x) always violates near 00.

Worked solutionExpand

Logical Negation: The statement lim⁡x→cf(x)≠L\lim_{x \to c} f(x) \ne L means: There exists ϵ0>0\epsilon_0 > 0 such that for every δ>0\delta > 0 there exists at least one x∈Dx \in D satisfying: 0<∣x−c∣<δand∣f(x)−L∣≥ϵ0.0 < |x - c| < \delta \quad \text{and} \quad |f(x) - L| \ge \epsilon_0.

Application to sgn⁡(x)\operatorname{sgn}(x) at c=0c = 0 with L=0L = 0 Choose ϵ0=1\epsilon_0 = 1 Let δ>0\delta > 0 be arbitrary. Select x=δ/2x = \delta/2 Then 0<∣x−0∣=δ/2<δ0 < |x - 0| = \delta/2 < \delta and x>0x > 0 so sgn⁡(x)=1\operatorname{sgn}(x) = 1 We evaluate the distance: ∣sgn⁡(x)−L∣=∣1−0∣=1≥ϵ0.|\operatorname{sgn}(x) - L| = |1 - 0| = 1 \ge \epsilon_0. Since this holds for every δ>0\delta > 0 the condition holds and lim⁡x→0sgn⁡(x)≠0\lim_{x \to 0} \operatorname{sgn}(x) \ne 0

Verification criteria. Requires: (a) correct quantifier inversion ∀ϵ→∃ϵ0\forall \epsilon \to \exists \epsilon_0 ∃δ→∀δ\exists \delta \to \forall \delta and the conditional P  ⟹  QP \implies Q negated to P∧¬QP \land \neg Q (b) an explicit witness ϵ0\epsilon_0 (any ϵ0≤1\epsilon_0 \le 1 (c) an explicit witness point xx as a function of δ\delta

References & Further Reading

  • Spivak, M. (2008). Calculus (4th ed.). Publish or Perish. — ϵ\epsilon-δ\delta proofs: Ch. 5 “Limits”, §“Delta–Epsilon Proofs”; algebra of limits: Ch. 5, §“Theorems on Limits”; continuity: Ch. 6 “Continuous Functions”; boundedness/IVT/EVT: Ch. 7 “Three Hard Theorems”, Thms 1–3; completeness proofs of IVT/EVT and uniform continuity (beyond this lesson’s scope): Ch. 8 “Least Upper Bound”.
  • Rudin, W. (1976). Principles of Mathematical Analysis (3rd ed.). McGraw-Hill. — limits and continuity: Ch. 4, §“Limits of Functions” and §“Continuous Functions”; sequential criterion for limits (Thm 4.12), EVT (Thms 4.16–4.17), IVT (Thm 4.23), and metric-space/compactness generalizations.
  • Burden, R. & Faires, J. (2011). Numerical Analysis (9th ed.). Cengage. §2.1 “The Bisection Method” — error bound and halving convergence.
  • Abbott, S. (2015). Understanding Analysis (2nd ed.). Springer. — Ch. 4 “Functional Limits and Continuity”: an alternate development of ϵ\epsilon-δ\delta limits (§4.2), the squeeze theorem, the IVT (§4.5), the EVT (§4.4), and uniform continuity on compact intervals.