math.wiki
Topology › First algebraic topology › Brouwer's fixed point theorem
Sign in

Brouwer's fixed point theorem

Lay a plan of a round ice rink anywhere on the rink itself, turned, shrunk or even crumpled. Must some spot of the plan lie exactly over the spot of the ice it shows?

Where you are 5 in First algebraic topology

Why this exists

A continuous map of [a, b] into itself fixes a point: the graph of h(x) - x starts at or above 0 and ends at or below it, and the intermediate value theorem does the rest (Exercise 9 of Extreme and intermediate value theorems, proved). In the disk there is no “between”, so that argument has nothing to hold. What replaces it is the count of turns from The fundamental group of the circle: a map of the disk with no fixed point would push the whole disk onto its rim while holding the rim still, and that would unwind a loop that winds once. The result is the first existence theorem that topology proves and analysis cannot easily reach, and through it equilibria, stationary states and positive eigenvectors are shown to exist without being computed.

Definition

“Map” means continuous map. \lVert x \rVert is the Euclidean norm and x \cdot y the dot product of Dot product, length and orthogonality in n dimensions.

Disks and spheres. For n \ge 1, D^n = \{x \in \mathbb{R}^n : \lVert x \rVert \le 1\} and S^{n-1} = \{x \in \mathbb{R}^n : \lVert x \rVert = 1\}. So D^1 = [-1, 1] with S^0 = \{-1, 1\}, and D^2 is the closed disk with rim S^1. D^n is compact by the Heine-Borel theorem (Compactness in products and the Heine-Borel theorem) and convex: for a, b \in D^n and t \in I, \lVert (1 - t)a + tb \rVert \le (1 - t)\lVert a \rVert + t\lVert b \rVert, which is at most 1.

Fixed points. A point x \in X is a fixed point of h : X \to X if h(x) = x. A space X has the fixed point property if every map X \to X has a fixed point.

Retracts. As in Homotopy, a retraction of X onto A \subseteq X is a map r : X \to A with r(a) = a for every a \in A; A is a retract of X if there is one.

Nearest points. For K \subseteq \mathbb{R}^2 and x \in \mathbb{R}^2, a point q \in K is nearest to x if \lVert x - q \rVert \le \lVert x - k \rVert for every k \in K. When there is exactly one, it is written p_K(x).

Intuition

Picture each point x of the disk carried to h(x). If no point stayed put, every x would have a direction, from h(x) toward x. Follow the ray from h(x) through x until it leaves the disk, at a point r(x) of the rim. Nearby points move to nearby points, so nearby rays are close and r is continuous. For x already on the rim, h(x) lies in the disk, so the ray leaves exactly at x: r(x) = x. Then r would be a retraction of D^2 onto S^1. The loop once around the rim shrinks to a point inside the disk, and r would carry that shrinking into the rim itself, making a loop of degree 1 nullhomotopic in S^1. That cannot happen, so some point does stay put.

x
h(x)
r(x)
p
D^2
Figure 1If h fixed no point, the ray from h(x) through x would reach the edge at r(x), and r would retract D^2 onto S^1. Drawn for the map and the point \displaystyle a = (\frac{1}{2}, 0) of Example 2; p is that map’s fixed point.

Results

Theorem 1 (passing the property on). Let X have the fixed point property. (a) Every space homeomorphic to X has it. (b) Every retract of X has it.

Theorem 2 (no retraction onto the rim). (a) S^0 is not a retract of D^1. (b) S^1 is not a retract of D^2.

Theorem 3 (Brouwer’s fixed point theorem in dimensions one and two). Every map D^1 \to D^1 and every map D^2 \to D^2 has a fixed point.

Theorem 4 (compact convex sets). Let K \subseteq \mathbb{R}^2 be nonempty, compact and convex. (a) Each x \in \mathbb{R}^2 has exactly one nearest point p_K(x) in K, and \lVert p_K(x) - p_K(y) \rVert \le \lVert x - y \rVert. (b) K, and every space homeomorphic to K, has the fixed point property. This covers closed squares, triangles and half disks, and closed intervals [a, b] \approx [a, b] \times \{0\}.

Theorem 5 (all dimensions). For every n \ge 1, every map D^n \to D^n has a fixed point, and S^{n-1} is not a retract of D^n.

Why it is true

Theorem 1. Proof. (a) Let \varphi : X \to Y be a homeomorphism and g : Y \to Y a map. Then \varphi^{-1} \circ g \circ \varphi : X \to X is a map (Theorem 3(d) of Continuous maps and homeomorphisms) with a fixed point x, and applying \varphi to \varphi^{-1}(g(\varphi(x))) = x gives g(\varphi(x)) = \varphi(x).

(b) Let r : X \to A be a retraction and g : A \to A a map. Then \iota \circ g \circ r : X \to X has a fixed point x. So x = g(r(x)) \in A, hence r(x) = x and g(x) = g(r(x)) = x. \square

Theorem 2. Proof. (a) A retraction r : D^1 \to S^0 fixes -1 and 1, so it is onto, and S^0 = r(D^1) would be connected (Theorems 3 and 4 of Connectedness). But \{-1\} and \{1\} are disjoint nonempty open sets of the subspace S^0, a separation.

(b) Suppose r : D^2 \to S^1 is a retraction and let x_0 = (1, 0). Then r \circ \iota = \mathrm{id}_{S^1}, so by Theorem 5 of The fundamental group

r_* \circ \iota_* = \mathrm{id}_{\pi_1(S^1, x_0)}.

As D^2 is convex, \pi_1(D^2, x_0) = \{e\} (Theorem 6 there), so \iota_* sends every class to e, and so does r_* \circ \iota_*. But [\omega_1] \neq e, since its degree is 1 (Theorem 4 of The fundamental group of the circle). \square

Theorem 3. Proof. Dimension one: suppose h : D^1 \to D^1 has no fixed point. Then \displaystyle r(x) = \frac{x - h(x)}{|x - h(x)|} is a map D^1 \to S^0. As h(1) \le 1 and h(1) \neq 1, 1 - h(1) > 0 and r(1) = 1; likewise r(-1) = -1. So r is a retraction, contradicting Theorem 2(a).

Dimension two: suppose h : D^2 \to D^2 has no fixed point, and put \displaystyle u(x) = \frac{x - h(x)}{\lVert x - h(x) \rVert}, a map D^2 \to S^1 (Theorem 6(b) of Subspace and product topologies). The point x + t u(x), t \ge 0, lies on S^1 exactly when

t^2 + 2\,(x \cdot u(x))\, t + \lVert x \rVert^2 - 1 = 0.

The constant term is \lVert x \rVert^2 - 1 \le 0, so the discriminant is at least 0 and the larger root

t(x) = -\,x \cdot u(x) + \sqrt{(x \cdot u(x))^2 + 1 - \lVert x \rVert^2}

is at least -x \cdot u(x) + |x \cdot u(x)| \ge 0. Built from u and x by sums, products and the continuous square root on [0, \infty), t is a map D^2 \to [0, \infty), and so is r(x) = x + t(x)u(x), a map D^2 \to S^1: the point where the ray from h(x) through x leaves D^2. For x \in S^1, the Cauchy-Schwarz inequality gives x \cdot h(x) \le \lVert x \rVert \lVert h(x) \rVert \le 1, so

x \cdot u(x) = \frac{1 - x \cdot h(x)}{\lVert x - h(x) \rVert} \ge 0,

hence t(x) = -x \cdot u(x) + |x \cdot u(x)| = 0 and r(x) = x. So r retracts D^2 onto S^1, contradicting Theorem 2(b). \square

Theorem 4. Proof. (a) k \mapsto \lVert x - k \rVert is continuous on the nonempty compact K, so it has a least value (Theorem 7 of Compactness and open covers): a nearest point exists. Let q be nearest to x and k \in K. For 0 < s \le 1, q + s(k - q) \in K by convexity, so \lVert x - q - s(k - q) \rVert^2 \ge \lVert x - q \rVert^2, that is -2s\,(x - q) \cdot (k - q) + s^2 \lVert k - q \rVert^2 \ge 0. Divide by s and let s \to 0:

(x - q) \cdot (k - q) \le 0 \quad \text{for every } k \in K. \tag{$*$}

Now let q be nearest to x and q' nearest to y. By (*), (x - q) \cdot (q' - q) \le 0 and (y - q') \cdot (q - q') \le 0. Adding gives (x - y) \cdot (q' - q) + \lVert q' - q \rVert^2 \le 0, so by Cauchy-Schwarz

\lVert q' - q \rVert^2 \le (y - x) \cdot (q' - q) \le \lVert x - y \rVert \lVert q' - q \rVert,

and \lVert q' - q \rVert \le \lVert x - y \rVert. With y = x this gives q' = q: the nearest point is unique.

(b) K is bounded, so K \subseteq E = \{x : \lVert x \rVert \le \rho\} for some \rho > 0. The scaling \displaystyle x \mapsto \frac{x}{\rho} is a homeomorphism E \to D^2, so E has the fixed point property (Theorems 3 and 1(a)). By (a), p_K is continuous (take \delta = \varepsilon in Theorem 4 of Continuous maps and homeomorphisms), and p_K(k) = k for k \in K, so p_K restricted to E retracts E onto K. Theorem 1(b) and then 1(a) finish. \square

Theorem 5 is beyond this lesson for n \ge 3 (for n = 1, 2 it is Theorems 2 and 3). For n \ge 3 the sphere S^{n-1} is simply connected, so \pi_1 cannot see the obstruction; the proof replaces it with homology (Hatcher, Algebraic Topology, Corollary 2.15). Given Theorem 5, the dimension-two proof of Theorem 3 and the proof of Theorem 4 run word for word in \mathbb{R}^n.

Worked examples

Example 1

A fixed point on the interval

Let \displaystyle h(x) = \frac{1 - x^2}{2} on D^1. (a) Show that h maps D^1 into D^1. (b) Find its fixed point. (c) Where h(x) \neq x, find \displaystyle r(x) = \frac{x - h(x)}{|x - h(x)|}, and say why it is not a retraction.

Solution

(a) For x \in [-1, 1], x^2 \in [0, 1], so \displaystyle h(x) \in [0, \frac{1}{2}].

(b) h(x) = x means x^2 + 2x - 1 = 0, so \displaystyle x = -1 \pm \sqrt{2}. Only \displaystyle c = \sqrt{2} - 1 lies in D^1.

(c) \displaystyle x - h(x) = \frac{x^2 + 2x - 1}{2}, which factors as \displaystyle \frac{(x - c)(x + 1 + \sqrt{2})}{2} with the second factor positive on D^1. So r(x) = -1 for x < c and r(x) = 1 for x > c. It fixes \pm 1, but it is defined only on D^1 \setminus \{c\} and jumps at c. The proof of Theorem 3 shows the jump has to be somewhere: it sits at a fixed point.

Example 2

A projection on an ice rink

The rink at a skating club is a disk of radius 15 meters; with its center as origin and 15 meters as the unit, the ice is D^2. Kofi projects the rink’s floor plan onto the ice, turned a quarter turn counterclockwise, at half size, with its center moved to \displaystyle (\frac{1}{4}, \frac{1}{4}). The spot of the plan showing the point (x, y) lands at \displaystyle h(x, y) = (\frac{1}{4} - \frac{y}{2}, \frac{1}{4} + \frac{x}{2}). (a) Show that h maps D^2 into D^2. (b) Find the spot of ice the plan shows in its true place, its distance from the center in meters, and show it is the only one. (c) For \displaystyle a = (\frac{1}{2}, 0), find where the ray from h(a) through a leaves the rink.

Solution

(a) \displaystyle h(x, y) = (\frac{1}{4}, \frac{1}{4}) + \frac{1}{2}(-y, x) and \lVert (-y, x) \rVert = \lVert (x, y) \rVert \le 1, so by the triangle inequality \displaystyle \lVert h(x, y) \rVert \le \frac{\sqrt{2}}{4} + \frac{1}{2} < 1.

(b) Brouwer’s theorem says a fixed point exists; here it can be found. Substituting \displaystyle y = \frac{1}{4} + \frac{x}{2} into \displaystyle x = \frac{1}{4} - \frac{y}{2} gives \displaystyle \frac{5x}{4} = \frac{1}{8}, so \displaystyle p = (\frac{1}{10}, \frac{3}{10}), at \displaystyle \frac{\sqrt{10}}{10} \cdot 15 = \frac{3\sqrt{10}}{2} \approx 4.74 meters from the center. For any two points, \displaystyle h(a) - h(b) = \frac{1}{2}(-(a_2 - b_2), a_1 - b_1), so \displaystyle \lVert h(a) - h(b) \rVert = \frac{1}{2}\lVert a - b \rVert; two fixed points a \neq b would give \displaystyle \lVert a - b \rVert = \frac{1}{2}\lVert a - b \rVert.

(c) \displaystyle h(a) = (\frac{1}{4}, \frac{1}{2}), \displaystyle a - h(a) = (\frac{1}{4}, -\frac{1}{2}) of length \displaystyle \frac{\sqrt{5}}{4}, so \displaystyle u(a) = (\frac{1}{\sqrt{5}}, -\frac{2}{\sqrt{5}}) and \displaystyle a \cdot u(a) = \frac{\sqrt{5}}{10}. With \displaystyle \lVert a \rVert^2 = \frac{1}{4}, the formula of Theorem 3 gives \displaystyle t(a) = -\frac{\sqrt{5}}{10} + \sqrt{\frac{1}{20} + \frac{3}{4}}, which is \displaystyle \frac{3\sqrt{5}}{10}, and \displaystyle r(a) = a + t(a)u(a) = (\frac{4}{5}, -\frac{3}{5}), which lies on S^1 as \displaystyle \frac{16}{25} + \frac{9}{25} = 1.

p
Figure 2Example 2: an arrow from each sample point x toward h(x), drawn at a quarter of its length. Every point moves except \displaystyle p = (\frac{1}{10}, \frac{3}{10}).
Example 3

Three spaces without the property

Find a map with no fixed point on (a) the open disk B = \{x \in \mathbb{R}^2 : \lVert x \rVert < 1\}, (b) the annulus A = \{x \in \mathbb{R}^2 : 1 \le \lVert x \rVert \le 2\}, (c) \mathbb{R}^2. Say which hypothesis of Theorem 4 fails, and why none is homeomorphic to D^2.

Solution

(a) \displaystyle g(x) = \frac{1}{2}(x + (1, 0)) has \displaystyle \lVert g(x) \rVert \le \frac{1}{2}(\lVert x \rVert + 1) < 1, so it maps B into B. Its only fixed point in \mathbb{R}^2 is (1, 0) \notin B. B is not closed, so not compact (Compactness in products and the Heine-Borel theorem).

(b) The quarter turn g(x, y) = (-y, x) keeps the norm, so it maps A onto A. g(x, y) = (x, y) forces x = -y and y = x, so x = y = 0, which is not in A. A is compact but not convex: (1, 0) and (-1, 0) lie in A, their midpoint 0 does not.

(c) g(x) = x + (1, 0) moves every point. \mathbb{R}^2 is not bounded.

By Theorems 3 and 1(a), any space homeomorphic to D^2 has the fixed point property, and these three do not.

Example 4

A map that pushes the rim inward

Let f : D^2 \to \mathbb{R}^2 be a map with \lVert f(x) \rVert \le 1 for every x \in S^1. Prove that f has a fixed point, though f may carry points of the open disk outside D^2.

Solution

Proof. Let \displaystyle \rho(y) = \frac{y}{m(y)} with m(y) = \max(1, \lVert y \rVert), which equals \displaystyle \frac{1}{2}(1 + \lVert y \rVert + |1 - \lVert y \rVert|) and so is continuous and at least 1. So \rho : \mathbb{R}^2 \to D^2 is a map fixing D^2, a retraction. By Theorem 3, \rho \circ f : D^2 \to D^2 has a fixed point a. If \lVert f(a) \rVert \le 1, then a = \rho(f(a)) = f(a). Otherwise \displaystyle a = \frac{f(a)}{\lVert f(a) \rVert} has norm 1, so a \in S^1 and the hypothesis gives \lVert f(a) \rVert \le 1, a contradiction. So f(a) = a. \square

The hypothesis on the rim is needed: f(x) = x + (3, 0) has no fixed point.

Example 5

A positive eigenvector

Let

A = \begin{pmatrix} 2 & 1 & 1 \\ 1 & 2 & 3 \\ 5 & 2 & 4 \end{pmatrix}

and let \Delta be the set of (a, b, c) \in \mathbb{R}^3 with a, b, c \ge 0 and a + b + c = 1. Prove from Theorem 4 that A has an eigenvector in \Delta with positive entries and a positive eigenvalue, then find it.

Solution

For v = (a, b, c) \in \Delta, every entry of Av is positive, since every entry of A is positive and v \ge 0 is not 0. The sum of the entries of Av is \sigma(v) = 8a + 5b + 8c > 0 (the column sums of A are 8, 5, 8). So \displaystyle f(v) = \frac{Av}{\sigma(v)} is a map \Delta \to \Delta (Theorem 6(b) of Subspace and product topologies).

\Delta is homeomorphic to the triangle T = \{(a, b) : a, b \ge 0,\ a + b \le 1\} by (a, b, c) \mapsto (a, b), with continuous inverse (a, b) \mapsto (a, b, 1 - a - b). T is closed and bounded, so compact, and convex as an intersection of three half-planes. By Theorem 4, f has a fixed point v, so Av = \sigma(v)v: an eigenvector (Eigenvalues and eigenvectors) with eigenvalue \sigma(v) > 0 and entries \displaystyle \frac{Av}{\sigma(v)} > 0.

To find it, expand the characteristic polynomial:

\begin{aligned} \det(\lambda I - A) &= \lambda^3 - 8\lambda^2 + 8\lambda - 7 \\ &= (\lambda - 7)(\lambda^2 - \lambda + 1). \end{aligned}

Here \lambda^2 - \lambda + 1 has no real root. So \sigma(v) = 7, and (A - 7I)(1, 2, 3) = 0 gives \displaystyle v = (\frac{1}{6}, \frac{1}{3}, \frac{1}{2}). Check: \displaystyle \sigma(v) = \frac{8}{6} + \frac{5}{3} + \frac{8}{2} = 7.

Boundaries

  • Every hypothesis is used. Example 3 drops compactness, the absence of a hole, and boundedness in turn. Continuity is needed too: \displaystyle h(x) = -\frac{1}{2} for x \ge 0 and \displaystyle h(x) = \frac{1}{2} for x < 0 maps D^1 into itself with no fixed point.
  • Existence only. A fixed point need not be unique (Exercises 3 and 4), and the proof, by contradiction, says nothing about where it is. Constructive proofs, for instance through Sperner’s lemma on triangulations, are beyond this lesson.
  • Not about convexity. The property passes to every space homeomorphic to D^2 and to every retract of one (Theorem 1); it fails on S^1, S^2 and T^2 (Exercise 10).
  • Dimension three and up need homology (Theorem 5); \pi_1 cannot tell D^3 from its rim.

Where this goes next

The argument here, a map that would unwind a loop of degree 1, also proves the The fundamental theorem of algebra, and a refinement proves the Borsuk-Ulam theorem in dimension two (Hatcher, section 1.1). Example 5 is the pattern for Markov chains: the same map on a simplex gives a stationary distribution, and existence proofs for equilibria in economics and game theory run in the same spirit. Classification of surfaces turns from maps to spaces, sorting the closed surfaces.

Exercises

Fixed points of particular maps

  1. Find every fixed point of \displaystyle h(x) = \frac{x^3 + 1}{2} on D^1.
  2. Show that \displaystyle h(x, y) = (\frac{y}{2}, \frac{1 - x}{2}) maps D^2 into D^2 and find its fixed point.
  3. Show that h(x, y) = (x^2 - y^2, 2xy) maps D^2 into D^2 and find every fixed point.
  4. Find the fixed points of the rotation R_\theta(x, y) = (x\cos\theta - y\sin\theta, x\sin\theta + y\cos\theta) of D^2 for 0 < \theta < 2\pi, and of the reflection (x, y) \mapsto (x, -y).
  5. Yuna reruns Kofi’s show from Example 2 with the plan turned a half turn, at a third of the size, its center moved to \displaystyle (\frac{1}{2}, 0): \displaystyle h(x, y) = (\frac{1}{2} - \frac{x}{3}, -\frac{y}{3}). Show that h maps D^2 into D^2, find the spot shown in its true place, and its distance from the center in meters.
  6. Show that \displaystyle h(x, y) = (\frac{1}{2}\cos y, \frac{1}{2}\sin x) maps D^2 into D^2 and has exactly one fixed point, and find it to three decimal places.

Rays and nearest points

  1. Let h be the constant map \displaystyle h(x) = (0, \frac{1}{2}) on D^2. Find where the ray from h(a) through a leaves D^2 for \displaystyle a = (\frac{1}{2}, \frac{1}{2}) and for \displaystyle a = (0, -\frac{1}{2}), and say why there is no ray at \displaystyle a = (0, \frac{1}{2}).
  2. Let Q = [0, 1] \times [0, 1] and T = \{(a, b) : a, b \ge 0,\ a + b \le 1\}. Find \displaystyle p_Q(3, -\frac{1}{2}), \displaystyle p_Q(\frac{1}{2}, 2), p_T(1, 1) and p_T(2, -1), and compare \lVert p_T(1, 1) - p_T(2, -1) \rVert with \lVert (1, 1) - (2, -1) \rVert.
  3. Let A = \begin{pmatrix} 1 & 2 \\ 8 & 1 \end{pmatrix} and \displaystyle f(v) = \frac{Av}{\sigma(v)} on the segment \{(a, 1 - a) : a \in I\}, where \sigma(v) is the sum of the entries of Av. Find \sigma as a function of a, the fixed point of f, and its eigenvalue.

Which spaces have the property

  1. Say whether each space has the fixed point property, with a map that has no fixed point or a theorem: (a) [0, \infty); (b) the closed half disk \{(x, y) \in D^2 : y \ge 0\}; (c) the boundary \partial Q of Q = [-1, 1] \times [-1, 1]; (d) [0, 3] \times [0, 1]; (e) S^2; (f) T^2.
  2. Refute with a counterexample: “S^1 is a retract of no space that properly contains it.”

Proofs

  1. Let Q = [-1, 1] \times [-1, 1]. Prove from Theorem 4, without fundamental groups, that \partial Q is not a retract of Q.
  2. Prove that a space that is not connected does not have the fixed point property.
  3. Prove that for every map h : D^2 \to D^2 there is an x with h(x) = -x.
  4. Prove that if a map g : S^1 \to S^1 extends to a map G : D^2 \to S^1, then \deg g = 0. Deduce Theorem 2(b) again.
  5. Let X be Hausdorff and h : X \to X a map. Prove that the set of fixed points of h is closed.
Answers
  1. x^3 \in [-1, 1], so h(x) \in [0, 1]. h(x) = x means x^3 - 2x + 1 = (x - 1)(x^2 + x - 1) = 0, with roots 1 and \displaystyle \frac{-1 \pm \sqrt{5}}{2}; \displaystyle \frac{-1 - \sqrt{5}}{2} < -1. The fixed points are 1 and \displaystyle \frac{\sqrt{5} - 1}{2}.
  2. \displaystyle h(x, y) = (0, \frac{1}{2}) + \frac{1}{2}(y, -x), so \displaystyle \lVert h(x, y) \rVert \le \frac{1}{2} + \frac{1}{2} = 1. From \displaystyle x = \frac{y}{2} and \displaystyle y = \frac{1 - x}{2}, \displaystyle y = \frac{1}{2} - \frac{y}{4}, so the fixed point is \displaystyle (\frac{1}{5}, \frac{2}{5}).
  3. (x^2 - y^2)^2 + (2xy)^2 = (x^2 + y^2)^2 \le 1. A fixed point has y(2x - 1) = 0. If y = 0, x^2 = x; if \displaystyle x = \frac{1}{2}, \displaystyle y^2 = -\frac{1}{4}, impossible. The fixed points are (0, 0) and (1, 0).
  4. R_\theta(x, y) = (x, y) is a linear system with determinant (\cos\theta - 1)^2 + \sin^2\theta = 2 - 2\cos\theta > 0, so only (0, 0). The reflection fixes the segment [-1, 1] \times \{0\}.
  5. \displaystyle \lVert h(x, y) \rVert \le \frac{1}{2} + \frac{1}{3} = \frac{5}{6}. From \displaystyle x = \frac{1}{2} - \frac{x}{3} and \displaystyle y = -\frac{y}{3}, the spot is \displaystyle (\frac{3}{8}, 0), at \displaystyle \frac{3}{8} \cdot 15 = 5.625 meters from the center.
  6. \displaystyle \lVert h \rVert \le \sqrt{\frac{1}{4} + \frac{1}{4}} < 1. By the mean value theorem |\cos s - \cos s'| \le |s - s'| and |\sin s - \sin s'| \le |s - s'|, so \displaystyle \lVert h(a) - h(b) \rVert \le \frac{1}{2}\lVert a - b \rVert, and two fixed points would coincide, as in Example 2. Theorem 3 gives one; solving \displaystyle x = \frac{1}{2}\cos(\frac{1}{2}\sin x) numerically, it is (0.486, 0.234).
  7. For \displaystyle a = (\frac{1}{2}, \frac{1}{2}): u(a) = (1, 0), \displaystyle a \cdot u(a) = \frac{1}{2}, \displaystyle \lVert a \rVert^2 = \frac{1}{2}, so \displaystyle t(a) = -\frac{1}{2} + \sqrt{\frac{3}{4}} and \displaystyle r(a) = (\frac{\sqrt{3}}{2}, \frac{1}{2}). For \displaystyle a = (0, -\frac{1}{2}): u(a) = (0, -1), \displaystyle t(a) = -\frac{1}{2} + 1 = \frac{1}{2}, r(a) = (0, -1). At \displaystyle (0, \frac{1}{2}), h(a) = a: the fixed point, where a - h(a) = 0 has no direction.
  8. In Q the squared distance (x - q_1)^2 + (y - q_2)^2 splits, so each coordinate is clamped to [0, 1]: \displaystyle p_Q(3, -\frac{1}{2}) = (1, 0) and \displaystyle p_Q(\frac{1}{2}, 2) = (\frac{1}{2}, 1). A point q \in T satisfying (*) is nearest, since \lVert x - k \rVert^2 - \lVert x - q \rVert^2 equals \lVert k - q \rVert^2 - 2(x - q) \cdot (k - q) and so is at least 0. For \displaystyle q = (\frac{1}{2}, \frac{1}{2}) and x = (1, 1): \displaystyle (x - q) \cdot (k - q) = \frac{1}{2}(k_1 + k_2 - 1) \le 0. For q = (1, 0) and x = (2, -1): (x - q) \cdot (k - q) = k_1 - 1 - k_2 \le 0. So \displaystyle p_T(1, 1) = (\frac{1}{2}, \frac{1}{2}) and p_T(2, -1) = (1, 0), at distance \displaystyle \frac{\sqrt{2}}{2} \le \sqrt{5}.
  9. Av = (2 - a, 1 + 7a), so \sigma = 3 + 6a > 0. f(v) = v means 2 - a = a(3 + 6a), that is 3a^2 + 2a - 1 = (3a - 1)(a + 1) = 0, so \displaystyle a = \frac{1}{3} and \displaystyle v = (\frac{1}{3}, \frac{2}{3}), with \displaystyle Av = (\frac{5}{3}, \frac{10}{3}) = 5v: eigenvalue 5.
  10. (a) No: x \mapsto x + 1. (b) Yes: D^2 and the half-plane y \ge 0 are convex and closed, and their intersection is bounded; Theorem 4. (c) No: (x, y) \mapsto (-y, x) keeps \max(|x|, |y|) = 1 and fixes only (0, 0) \notin \partial Q. (d) Yes: compact and convex, Theorem 4. (e) No: x \mapsto -x fixes only 0 \notin S^2. (f) No: (a, b) \mapsto (-a, b), as -a \neq a on S^1.
  11. \displaystyle r(x) = \frac{x}{\lVert x \rVert} is a map \mathbb{R}^2 \setminus \{0\} \to S^1 fixing S^1, so S^1 is a retract of \mathbb{R}^2 \setminus \{0\}. Theorem 2 needs the missing center.
  12. Proof. Suppose \rho : Q \to \partial Q is a retraction. \partial Q is symmetric under x \mapsto -x, so g(x) = -\rho(x) is a map Q \to \partial Q \subseteq Q. Q is compact and convex, so g(a) = a for some a (Theorem 4). Then a = -\rho(a) \in \partial Q, so \rho(a) = a and a = -a, giving a = 0 \notin \partial Q. \square
  13. Proof. Let U, V be a separation of X, with u \in U and v \in V. Let g = v on U and g = u on V. Each preimage g^{-1}(W) is \varnothing, U, V or X, all open, so g is a map. A point of U goes to V and a point of V to U, so g fixes nothing. \square
  14. Proof. -h is a map D^2 \to D^2, as \lVert -h(x) \rVert = \lVert h(x) \rVert. By Theorem 3, -h(a) = a for some a, that is h(a) = -a. \square
  15. Proof. H(x, t) = G((1 - t)x), for x \in S^1 and t \in I, is defined since (1 - t)x \in D^2, and is a homotopy from g to the constant map at G(0). So g is nullhomotopic and \deg g = 0 (Theorem 5(b) of The fundamental group of the circle). A retraction r : D^2 \to S^1 would extend \mathrm{id}_{S^1}, of degree 1 (Theorem 5(c) there). \square
  16. Proof. Let h(x) \neq x. Take disjoint open U \ni x and V \ni h(x). Then W = U \cap h^{-1}(V) is open and contains x. A fixed point y \in W would give y = h(y) \in U \cap V = \varnothing. So the complement of the fixed set is open. \square
math.wiki · updated 2026-11-18 · CC-BY-SA-4.0