The Extreme Value Theorem

Consider a hiker walking along a mountain trail described by a continuous function of position. It seems obvious the hiker reaches a highest point and a lowest point somewhere on the trail. But this "obvious" fact is false without the right hypotheses. Detecting exactly which hypotheses are needed is one of the sharper lessons in analysis. The Extreme Value Theorem (EVT) says that a continuous real-valued function on a closed, bounded interval [a,b][a,b] is bounded. Moreover, it actually attains a maximum value and a minimum value somewhere in [a,b][a,b]. This theorem matters because it is the rigorous foundation underneath an enormous swath of applied mathematics. Every time a calculus student sets a derivative to zero to find "the" maximum of a function on an interval, they are silently invoking the EVT. They need it to know a maximum exists in the first place. Optimization, physics (least-action principles, energy minimization), economics (utility maximization on constraint sets), and numerical analysis all lean on existence-of-extrema results. These results trace back to this theorem. The big picture is that continuity alone is not enough; the domain must be compact. In the setting of R\mathbb{R}, this means closed and bounded, by the Heine-Borel Theorem. The theorem is a beautiful illustration of how a topological property of the domain (compactness) interacts with an analytic property of the function (continuity). Together they produce a strong conclusion (existence of extrema). Understanding why each hypothesis is load-bearing is as important as the proof itself. This includes counterexamples where continuity fails, or the interval is open, or unbounded. This article gives the complete statement and proof of the EVT using the Bolzano-Weierstrass Theorem and the Sequential Characterization of Continuity. It works through two asymmetric examples, and exhibits sharp counterexamples for each hypothesis. Finally, it points toward its natural sequel: Uniform Continuity of continuous functions on compact sets.

Key takeaways
  • A continuous function on a closed, bounded interval [a,b][a,b] is always bounded.
  • Stronger: it attains both a maximum and a minimum value at some point of [a,b][a,b].
  • The proof combines Compactness of closed bounded intervals (via Bolzano–Weierstrass) with the Sequential Characterization of Continuity.
  • Every hypothesis is essential: drop continuity, closedness, or boundedness, and the conclusion can fail.
  • The theorem justifies routine calculus practice: solving f′(x)=0f'(x)=0 presumes a max/min exists to search for.
  • Key idea: compactness of the domain converts "continuity" into "existence of extrema" by trapping any maximizing sequence into a convergent subsequence whose limit is the maximizer.

Where this lives today

The Extreme Value Theorem is the quiet justification behind every optimization procedure that first asks "does a solution exist?" before searching for one. In calculus pedagogy, it licenses the standard technique of comparing critical-point values and endpoint values to find absolute extrema on a closed interval — without EVT, that comparison could be comparing numbers that don't actually bound anything. In numerical optimization and machine learning, existence theorems in this spirit justify that loss functions restricted to compact parameter sets (e.g., a ball of bounded weights) actually have a global minimizer for gradient-based methods to converge toward, rather than an infimum that recedes forever. In economics, utility or profit functions maximized over compact budget or feasible sets rely on EVT (or its multivariable/topological generalizations) to guarantee optimal choices exist, underwriting existence proofs in general equilibrium theory. In physics, variational principles (least action, minimal energy configurations) implicitly assume compactness-type conditions to guarantee that an extremal path or configuration is actually realized, not just approached in a limit. In numerical analysis and root-finding, EVT combined with the Intermediate Value Theorem underlies error-bound arguments, since a continuous error function on a compact domain has a genuine worst case to bound against. Finally, this theorem is the direct stepping stone to Uniform Continuity: the same compactness machinery (sequences, subsequential limits staying in the domain) that proves extrema are attained also proves that continuity on [a,b][a,b] automatically upgrades to uniform continuity, a strictly stronger and highly useful property for integration theory.

:::quiz[Why does f(x) = 1/x fail to be bounded on (0,1]?]

  • Because f is discontinuous somewhere in (0,1]
  • Because (0,1] is unbounded
  • Because (0,1] is missing its limit point 0, where f would blow up ✓
  • Because f is not differentiable at some point

Ask the tutor: Can you show me step by step why a maximizing sequence for a continuous function on (0,1) can fail to have its limit inside the domain?

DefinitionBoundedness, Maximum, and Attainment

Let S⊆RS \subseteq \mathbb{R} and let f:S→Rf: S \to \mathbb{R}.

  • ff is bounded above on SS if there exists M∈RM \in \mathbb{R} such that f(x)≤Mf(x) \le M for all x∈Sx \in S. Similarly bounded below with a lower bound mm. ff is bounded if it is bounded both above and below.
  • ff attains a maximum on SS if there exists x∗∈Sx^* \in S such that f(x∗)≥f(x)f(x^*) \ge f(x) for all x∈Sx \in S; we then write f(x∗)=sup⁡x∈Sf(x)=max⁡x∈Sf(x)f(x^*) = \sup_{x \in S} f(x) = \max_{x\in S} f(x). Attaining a minimum is defined symmetrically, with x∗∈Sx_* \in S such that f(x∗)≤f(x)f(x_*) \le f(x) for all x∈Sx\in S.
  • The Extreme Value Theorem asserts: if f:[a,b]→Rf:[a,b]\to \mathbb{R} is continuous on the closed bounded interval [a,b][a,b] (with a<ba<b real numbers), then ff is bounded on [a,b][a,b], and there exist points x∗,x∗∈[a,b]x^*, x_* \in [a,b] with
f(x∗)≤f(x)≤f(x∗)for all x∈[a,b].f(x_*) \le f(x) \le f(x^*) \quad \text{for all } x \in [a,b].

Intuition

Picture the graph of a continuous function drawn without lifting the pen, over a finite closed segment [a,b][a,b] of the xx-axis. Because the domain is closed and bounded, the curve cannot "run off to infinity" as xx approaches an endpoint, and it cannot approach a boundary point that is excluded from the domain (there is none — aa and bb themselves belong to [a,b][a,b]). Continuity prevents the curve from having sudden jumps or removable gaps that let it "sneak past" a value without ever equaling it. The mechanism behind the proof is a compactness argument: take a sequence of points xn∈[a,b]x_n \in [a,b] whose function values f(xn)f(x_n) climb toward the supremum of ff on [a,b][a,b] (such a sequence always exists by definition of supremum). Because [a,b][a,b] is bounded, Bolzano–Weierstrass guarantees a convergent subsequence xnk→x∗x_{n_k} \to x^*, and because [a,b][a,b] is closed, the limit x∗x^* stays inside [a,b][a,b]. Continuity of ff then forces f(xnk)→f(x∗)f(x_{n_k}) \to f(x^*), pinning down that the supremum is actually the value f(x∗)f(x^*) — a genuine maximum, not just a limiting value approached but never reached. The two figures below show the contrast: a continuous function on a closed interval reaching a rooftop and a floor, versus the classic broken example — a continuous function on a half-open interval whose graph climbs forever without ever reaching a top, because the point where it "would" peak has been removed from the domain.

Figure
A continuous function on a closed interval attains its max and min
x y a b local dip max f(x*) also high min f(x_*) domain [a,b] is closed and bounded — both extremes are reached
Figure
Removing an endpoint breaks attainment of the maximum
x y 0 1 (excluded) endpoint 0 included open circle: sup never attained f(x)=x on [0,1): sup f = 1 but no maximizer exists in the domain
TheoremExtreme Value Theorem

Let a,b∈Ra, b \in \mathbb{R} with a<ba < b, and let f:[a,b]→Rf : [a,b] \to \mathbb{R} be continuous on [a,b][a,b]. Then: 1. ff is bounded on [a,b][a,b]: there exist m,M∈Rm, M \in \mathbb{R} such that m≤f(x)≤Mm \le f(x) \le M for all x∈[a,b]x \in [a,b]. 2. ff attains a maximum on [a,b][a,b]: there exists x∗∈[a,b]x^* \in [a,b] with f(x∗)=sup⁡x∈[a,b]f(x)f(x^*) = \sup_{x\in[a,b]} f(x). 3. ff attains a minimum on [a,b][a,b]: there exists x∗∈[a,b]x_* \in [a,b] with f(x∗)=inf⁡x∈[a,b]f(x)f(x_*) = \inf_{x\in[a,b]} f(x).

LemmaBoundedness Lemma

Let f:[a,b]→Rf:[a,b] \to \mathbb{R} be continuous on the closed bounded interval [a,b][a,b]. Then ff is bounded above on [a,b][a,b] (and, by an identical argument applied to −f-f, bounded below).

Proof

Technique: proof by contradiction using Bolzano–Weierstrass and sequential continuity. Suppose, for contradiction, that ff is not bounded above on [a,b][a,b]. Then for every n∈Nn \in \mathbb{N}, the number nn fails to be an upper bound for ff on [a,b][a,b]; that is, there exists a point xn∈[a,b]x_n \in [a,b] with

f(xn)>n.f(x_n) > n.

This produces a sequence (xn)n=1∞(x_n)_{n=1}^{\infty} with xn∈[a,b]x_n \in [a,b] for every nn. Since a≤xn≤ba \le x_n \le b for all nn, the sequence (xn)(x_n) is bounded. By the Compactness of closed bounded intervals (equivalently, the Bolzano–Weierstrass theorem applied to a bounded sequence of reals), (xn)(x_n) has a convergent subsequence (xnk)k=1∞(x_{n_k})_{k=1}^{\infty} with

xnk→cx_{n_k} \to c

for some c∈Rc \in \mathbb{R}. Because a≤xnk≤ba \le x_{n_k} \le b for every kk, and limits preserve non-strict inequalities, we get a≤c≤ba \le c \le b, i.e. c∈[a,b]c \in [a,b] — this is precisely where closedness of the interval is used: the limit of points in [a,b][a,b] cannot escape [a,b][a,b]. Since ff is continuous at c∈[a,b]c \in [a,b], the Sequential Characterization of Continuity gives

f(xnk)→f(c)as k→∞.f(x_{n_k}) \to f(c) \quad \text{as } k \to \infty.

In particular, the sequence (f(xnk))k=1∞(f(x_{n_k}))_{k=1}^\infty is convergent, hence bounded: there exists B∈RB \in \mathbb{R} with f(xnk)≤Bf(x_{n_k}) \le B for all kk. But by construction, f(xnk)>nkf(x_{n_k}) > n_k for every kk, and nk→∞n_k \to \infty as k→∞k \to \infty (since (nk)(n_k) is a strictly increasing sequence of positive integers, being the index sequence of a subsequence). Hence f(xnk)→∞f(x_{n_k}) \to \infty, contradicting that (f(xnk))(f(x_{n_k})) is bounded by BB. This contradiction shows the assumption was false: ff must be bounded above on [a,b][a,b]. For the lower bound: apply the argument just given to the function g=−fg = -f, which is continuous on [a,b][a,b] because it is a scalar multiple of a continuous function (a fact immediate from the algebra of limits, not assumed circularly here — continuity of −f-f follows directly from the ε\varepsilon–δ\delta or sequential definition of continuity of ff). Boundedness above of g=−fg = -f gives −f(x)≤M′-f(x) \le M' for all xx, i.e. f(x)≥−M′f(x) \ge -M' for all x∈[a,b]x \in [a,b], which is a lower bound. ■\blacksquare

∎
Proof

Proof of the Extreme Value Theorem (attainment of maximum), technique: supremum + sequential compactness argument. (1) Boundedness is the Boundedness Lemma just proved. (2) Attainment of the maximum. By part (1), the set

E={f(x):x∈[a,b]}E = \{ f(x) : x \in [a,b] \}

is a nonempty (since a∈[a,b]a \in [a,b]) subset of R\mathbb{R} that is bounded above. By the Least Upper Bound Property of R\mathbb{R}, EE has a finite supremum; set

M=sup⁡x∈[a,b]f(x)=sup⁡E.M = \sup_{x \in [a,b]} f(x) = \sup E.

We must produce x∗∈[a,b]x^* \in [a,b] with f(x∗)=Mf(x^*) = M. By the defining property of supremum, for every n∈Nn \in \mathbb{N} the number M−1nM - \frac{1}{n} is not an upper bound for EE (since MM is the least upper bound). Hence there exists xn∈[a,b]x_n \in [a,b] with

M−1n<f(xn)≤M.M - \frac{1}{n} < f(x_n) \le M.

This defines a sequence (xn)(x_n) in [a,b][a,b], and by the squeeze theorem applied to the inequality above,

f(xn)→Mas n→∞.(∗)f(x_n) \to M \quad \text{as } n \to \infty. \tag{$\ast$}

Since (xn)(x_n) is a sequence in the bounded set [a,b][a,b], by Compactness of closed bounded intervals there is a convergent subsequence xnk→x∗x_{n_k} \to x^*, and since [a,b][a,b] is closed, x∗∈[a,b]x^* \in [a,b] exactly as in the Boundedness Lemma's proof. By continuity of ff at x∗x^* and the Sequential Characterization of Continuity,

f(xnk)→f(x∗)as k→∞.f(x_{n_k}) \to f(x^*) \quad \text{as } k \to \infty.

On the other hand, (f(xnk))k(f(x_{n_k}))_k is a subsequence of (f(xn))n(f(x_n))_n, and by (∗)(\ast) the full sequence f(xn)→Mf(x_n) \to M; every subsequence of a convergent sequence converges to the same limit, so

f(xnk)→M.f(x_{n_k}) \to M.

A sequence of real numbers has a unique limit (uniqueness of limits in R\mathbb{R}), so comparing the two computed limits of (f(xnk))k(f(x_{n_k}))_k forces

f(x∗)=M=sup⁡x∈[a,b]f(x).f(x^*) = M = \sup_{x\in[a,b]} f(x).

Since MM is an upper bound for ff on [a,b][a,b] and f(x∗)=Mf(x^*) = M, we conclude f(x)≤f(x∗)f(x) \le f(x^*) for all x∈[a,b]x \in [a,b]; that is, ff attains its maximum at x∗∈[a,b]x^* \in [a,b]. (3) Attainment of the minimum. Apply part (2) to g=−fg = -f, which is continuous on [a,b][a,b] (as noted above). Part (2) gives a point x∗∈[a,b]x_* \in [a,b] with g(x∗)≥g(x)g(x_*) \ge g(x) for all x∈[a,b]x \in [a,b], i.e. −f(x∗)≥−f(x)-f(x_*) \ge -f(x), i.e. f(x∗)≤f(x)f(x_*) \le f(x) for all x∈[a,b]x \in [a,b]. Thus x∗x_* is a minimizer of ff on [a,b][a,b], and

f(x∗)=inf⁡x∈[a,b]f(x)f(x_*) = \inf_{x \in [a,b]} f(x)

by the same supremum/infimum duality used in part (2) (infimum of ff equals −sup⁡(−f)-\sup(-f), a standard property of sup⁡\sup/inf⁡\inf under negation). ■\blacksquare

∎

Worked examples

Example 1 — a rational function on an asymmetric closed interval. Let f(x)=x3−3x+1x+3f(x) = \dfrac{x^3 - 3x + 1}{x+3} on the closed bounded interval [−1,4][-1, 4]. Step 1: check hypotheses. The denominator x+3x+3 vanishes only at x=−3x=-3, which does not lie in [−1,4][-1,4]. Since ff is a quotient of polynomials (continuous everywhere by the algebra of limits) with nonvanishing denominator on [−1,4][-1,4], ff is continuous on [−1,4][-1,4]. The interval [−1,4][-1,4] is closed and bounded. The EVT applies: ff is bounded on [−1,4][-1,4] and attains a max and min there. Step 2: locate candidates. Extrema occur either at endpoints or at interior critical points where f′(x)=0f'(x) = 0 (since ff is differentiable on the open interval (−1,4)(-1,4), by Fermat's Theorem on Critical Points, attained interior extrema must be critical points — but note this fact is a consequence, used only to search efficiently; existence of the extremum is already guaranteed by EVT regardless of differentiability). Compute using the quotient rule:

f′(x)=(3x2−3)(x+3)−(x3−3x+1)(1)(x+3)2=3x3+9x2−3x−9−x3+3x−1(x+3)2=2x3+9x2−10(x+3)2.f'(x) = \frac{(3x^2-3)(x+3) - (x^3-3x+1)(1)}{(x+3)^2} = \frac{3x^3+9x^2-3x-9 - x^3+3x-1}{(x+3)^2} = \frac{2x^3 + 9x^2 - 10}{(x+3)^2}.

Set the numerator to zero: 2x3+9x2−10=02x^3+9x^2-10=0. Testing x=1x=1: 2+9−10=1≠02+9-10=1\ne0. Testing x=−1x = -1: −2+9−10=−3≠0-2+9-10=-3 \ne 0. A numerical check shows a root near x≈0.98x \approx 0.98: at x=0.98x=0.98, 2(0.941)+9(0.9604)−10≈1.882+8.644−10=0.5262(0.941)+9(0.9604)-10 \approx 1.882+8.644-10=0.526; at x=0.9x=0.9, 2(0.729)+9(0.81)−10=1.458+7.29−10=−1.2522(0.729)+9(0.81)-10=1.458+7.29-10=-1.252. So the root lies in (0.9,0.98)(0.9, 0.98); refine at x=0.95x=0.95: 2(0.857)+9(0.9025)−10=1.715+8.123−10=−0.1622(0.857)+9(0.9025)-10=1.715+8.123-10=-0.162; at x=0.97x=0.97: 2(0.9127)+9(0.9409)−10=1.825+8.468−10=0.2932(0.9127)+9(0.9409)-10=1.825+8.468-10=0.293. So the critical point is near xc≈0.96x_c \approx 0.96. Step 3: evaluate.

f(−1)=−1+3+12=32=1.5,f(-1) = \frac{-1+3+1}{2} = \frac{3}{2} = 1.5,
f(4)=64−12+17=537≈7.571,f(4) = \frac{64-12+1}{7} = \frac{53}{7} \approx 7.571,
f(0.96)=(0.885)−2.88+13.96=−0.9953.96≈−0.2513.f(0.96) = \frac{(0.885) - 2.88 + 1}{3.96} = \frac{-0.995}{3.96} \approx -0.2513.

Comparing the three values 1.51.5, 7.5717.571, −0.2513-0.2513: the maximum value found is at x=4x=4 with f(4)=53/7≈7.571f(4) = 53/7 \approx 7.571, and the minimum is at the critical point xc≈0.96x_c \approx 0.96 with f(xc)≈−0.251f(x_c)\approx -0.251. The EVT guarantees these comparisons are conclusive — there is no need to check "infinity" or a boundary that isn't there, precisely because [−1,4][-1,4] is compact and ff is continuous throughout.

Example 2 — a piecewise function continuous despite the join, on [0,5][0,5]. Define

f(x)={x2+1,0≤x≤2,9−2x,2<x≤5.f(x) = \begin{cases} x^2 + 1, & 0 \le x \le 2, \\ 9 - 2x, & 2 < x \le 5. \end{cases}

Step 1: verify continuity, including at the seam x=2x=2. On [0,2)[0,2) and (2,5](2,5], ff is a polynomial, hence continuous. At x=2x=2: the left-hand piece gives f(2)=22+1=5f(2) = 2^2+1 = 5 using the formula valid at x=2x=2 (since the first case includes x=2x=2). The right-hand limit is

lim⁡x→2+(9−2x)=9−4=5.\lim_{x\to 2^+} (9-2x) = 9-4 = 5.

Since lim⁡x→2−f(x)=f(2)=5\lim_{x\to2^-} f(x) = f(2) = 5 (continuity of x2+1x^2+1 at 2 from the left) and lim⁡x→2+f(x)=5\lim_{x\to2^+}f(x)=5 matches, ff is continuous at x=2x=2, hence continuous on all of [0,5][0,5]. Step 2: apply EVT. [0,5][0,5] is closed and bounded, ff is continuous, so ff is bounded and attains max/min on [0,5][0,5]. Step 3: find them. On [0,2][0,2], f(x)=x2+1f(x)=x^2+1 is increasing (derivative 2x≥02x \ge 0), so its max on this piece is at x=2x=2: f(2)=5f(2)=5, and its min is at x=0x=0: f(0)=1f(0)=1. On (2,5](2,5], f(x)=9−2xf(x) = 9-2x is strictly decreasing (derivative −2<0-2 <0), so on this piece values decrease from just under f(2)=5f(2)=5 down to f(5)=9−10=−1f(5) = 9-10=-1. Comparing candidates from both pieces: f(0)=1f(0)=1, f(2)=5f(2)=5, f(5)=−1f(5)=-1. The global maximum is f(2)=5f(2)=5 (attained, since x=2x=2 is genuinely in the domain and f(2)=5f(2)=5 by direct substitution) and the global minimum is f(5)=−1f(5) = -1. Notice the maximum occurs exactly at the seam point x=2x=2 — a place where a naive analysis might worry about discontinuity, but continuity was verified explicitly, so the EVT's conclusion is airtight: the maximum of 55 is genuinely attained at x=2∈[0,5]x=2 \in [0,5], not merely approached.

CounterexampleWhy closedness of the interval cannot be dropped

Let f(x)=1xf(x) = \dfrac{1}{x} for x∈(0,1]x \in (0,1] — note the domain is bounded but not closed (it is missing its left endpoint 00). ff is continuous on (0,1](0,1]: it is a quotient of continuous functions with nonvanishing denominator on this domain. But ff is not bounded above on (0,1](0,1]: given any M>0M>0, choosing x=1M+1∈(0,1]x = \frac{1}{M+1} \in (0,1] gives f(x)=M+1>Mf(x) = M+1 > M. So no finite upper bound works, and in particular no maximum can be attained. What breaks in the proof: constructing a maximizing sequence xn=1/n∈(0,1]x_n = 1/n \in (0,1] with f(xn)=n→∞f(x_n) = n \to \infty, the sequence (xn)(x_n) does converge (to 00), consistent with boundedness of the domain and Bolzano–Weierstrass — but its limit 00 is not in the domain (0,1](0,1]. The step "c∈[a,b]c \in [a,b] because the interval is closed" fails, so continuity of ff at the (nonexistent, out-of-domain) limit point can never be invoked to control f(xn)f(x_n). This is exactly the gap that closedness plugs. A second, milder illustration on the same domain: even boundedness of ff alone can fail. Even if we replace ff with a bounded but non-attaining function, e.g. g(x)=xg(x) = x on (0,1)(0,1) (open, bounded interval, missing both endpoints), gg is continuous and bounded (0<g(x)<10 < g(x) < 1), yet attains neither a max nor a min: sup⁡g=1\sup g = 1 and inf⁡g=0\inf g = 0 are both unreached, since 1,0∉(0,1)1, 0 \notin (0,1). This shows that even when boundedness of the function survives, attainment can still fail if the domain is not closed.

Scope, edge cases & common traps

SituationWhat happensWhy
ff continuous on closed bounded [a,b][a,b]max & min both attainedEVT applies directly
ff continuous on open (a,b)(a,b)may be unbounded, or bounded without attaining extremadomain not closed; limit points can escape
ff continuous on unbounded closed set, e.g. [0,∞)[0,\infty)may be unbounded (e.g. f(x)=xf(x)=x)domain not bounded; Bolzano–Weierstrass fails
ff discontinuous at one point of [a,b][a,b]can fail to attain max/min even though boundedsequential continuity argument breaks at the jump
ff continuous on closed bounded [a,b][a,b] but non-differentiableEVT still gives attained extremaEVT needs only continuity, not differentiability
[a,b][a,b] replaced by any compact subset K⊂RK\subset\mathbb{R} (e.g. finite union of closed bounded intervals)max & min still attainedproof only used sequential compactness of KK, not that KK is an interval

The mathematician behind it

Origins · 1 minKarl Weierstrass and how this was discovered
KW
Karl Weierstrass
1815–1897 · German analysis, rigorization of calculus

Karl Weierstrass was a German mathematician often called "the father of modern analysis." Working largely outside the traditional university track for much of his early career (he taught secondary school for over a decade before his mathematical talent was recognized), Weierstrass later became a professor in Berlin, where he built much of the rigorous ε\varepsilon–δ\delta apparatus that underlies continuity, limits, and uniform convergence as they are taught today. The Extreme Value Theorem is frequently attributed jointly to Weierstrass and Bernard Bolzano, and is sometimes called the Bolzano–Weierstrass Theorem in older or European texts (distinct from, but historically entangled with, the Bolzano–Weierstrass Theorem on convergent subsequences used in the proof above). Weierstrass insisted that geometric intuition be backed by airtight sequential and ε\varepsilon–δ\delta arguments — the very style of proof given here.

History
- 1817 — Bernard Bolzano gives an early rigorous treatment of continuity and intermediate values, laying groundwork for compactness-flavored arguments.
- 1860s — Karl Weierstrass, lecturing in Berlin, develops the ε\varepsilon–δ\delta formalism and proves boundedness/attainment results for continuous functions on closed intervals.
- 1872 — Weierstrass's construction of a continuous nowhere-differentiable function sharpens the community's sense that continuity alone, without further structure like compactness of the domain, is a subtle hypothesis.
- 1880s–1890s — Heine and Borel formalize the covering-based notion of compactness, eventually shown equivalent to "closed and bounded" for subsets of R\mathbb{R} (Heine–Borel theorem), giving the modern topological lens for the EVT.
- Early 1900s — The EVT is generalized from closed intervals to arbitrary compact metric and topological spaces, becoming a cornerstone statement in general topology and real analysis textbooks.
Does the Extreme Value Theorem require f to be differentiable?

No. The EVT requires only that ff be continuous on [a,b][a,b]; differentiability plays no role in the proof, which uses sequences, subsequential limits (Bolzano–Weierstrass), and the sequential characterization of continuity. Differentiability becomes useful only after EVT guarantees an extremum exists, because Fermat's Theorem on Critical Points then lets you search for it efficiently among critical points and endpoints. Many continuous functions with attained extrema, such as f(x)=∣x∣f(x)=|x| on [−1,1][-1,1], fail to be differentiable at the very point where an extremum occurs (here, the minimum at x=0x=0), and EVT applies to them without any trouble.

Can the Extreme Value Theorem fail if f is continuous but the interval is unbounded, like [0, infinity)?

Yes. Take f(x)=xf(x) = x on [0,∞)[0,\infty): this set is closed, ff is continuous, but the domain is unbounded, and indeed ff has no upper bound and attains no maximum (though it does attain a minimum, f(0)=0f(0)=0, showing partial conclusions can survive dropping one hypothesis). The proof's use of Bolzano–Weierstrass crucially needs the maximizing sequence to be bounded to extract a convergent subsequence; on an unbounded domain, a maximizing sequence like xn=nx_n=n has no convergent subsequence in R\mathbb{R}, so the argument breaks down entirely.

Is the Extreme Value Theorem true for functions of several variables?

Yes, in a suitably generalized form: if f:K→Rf: K \to \mathbb{R} is continuous and K⊆RnK \subseteq \mathbb{R}^n is compact (closed and bounded, by Heine–Borel in Rn\mathbb{R}^n), then ff attains a maximum and minimum on KK. The proof is essentially identical, replacing the one-dimensional Bolzano–Weierstrass theorem with its Rn\mathbb{R}^n analogue for bounded sequences of vectors. This generalization is exactly what underlies constrained optimization theory in multivariable calculus and beyond.

What is the difference between the supremum of f and the maximum of f?

The supremum sup⁡x∈Sf(x)\sup_{x\in S} f(x) is always well-defined as a real number whenever ff is bounded above on SS (by completeness of R\mathbb{R}), regardless of whether any point of SS actually achieves that value. The maximum, by contrast, requires a specific point x∗∈Sx^* \in S with f(x∗)f(x^*) equal to that supremum. The EVT's real content is proving these coincide when ff is continuous and S=[a,b]S=[a,b] is closed and bounded — outside that setting (e.g. g(x)=xg(x)=x on (0,1)(0,1)), the supremum can exist as a number while no maximum exists as an attained value.

Why does the theorem need BOTH boundedness and closedness of the interval, rather than just one?

Each hypothesis blocks a different failure mode. Boundedness (of the domain) is what feeds Bolzano–Weierstrass, guaranteeing some convergent subsequence exists at all; without it (e.g. f(x)=xf(x)=x on [0,∞)[0,\infty)) a maximizing sequence can run off to infinity with no subsequential limit whatsoever. Closedness is what guarantees that once a subsequential limit exists, it actually lies back inside the domain, so continuity can be invoked there; without it (e.g. f(x)=1/xf(x)=1/x on (0,1](0,1]) the limit point 00 exists as a real number but sits outside the domain, so the continuity argument has nothing to grab onto. Both gaps are independently fatal, which is why the hypothesis is stated as "closed and bounded," i.e. compact.

Does the theorem say the maximum and minimum are unique?

No, and it should not be read that way. The EVT guarantees existence of at least one maximizing point x∗x^* and at least one minimizing point x∗x_*, but says nothing about uniqueness. For instance f(x)=cos⁡(x)f(x) = \cos(x) on [0,4π][0, 4\pi] attains its maximum value 11 at multiple points (x=0,2π,4πx=0, 2\pi, 4\pi), and the EVT is fully satisfied by any one of them witnessing the attained maximum.

How does the Extreme Value Theorem connect to Uniform Continuity?

Both results exploit the same compactness mechanism on [a,b][a,b]: sequences in the domain always have subsequences converging back into the domain. The EVT uses this to pin down a maximizing point; the theorem that continuous functions on closed bounded intervals are automatically uniformly continuous uses an almost identical proof-by-contradiction with paired sequences xn,ynx_n, y_n with ∣xn−yn∣→0|x_n-y_n|\to 0 but ∣f(xn)−f(yn)∣|f(x_n)-f(y_n)| bounded away from 00, again extracting a convergent subsequence to derive a contradiction with ordinary continuity. Studying the EVT proof carefully is the best preparation for understanding Uniform Continuity rigorously.

Before this

Where this leads