Skip to content

Chapter 5 — Trigonometric Functions of Real Numbers

5.2 - Unit Circle and Wrapping Function

With the total length of the circle already constructed in §5.1, what gives meaning to “wrapping” the real line onto it is missing: knowing that each partial arc length corresponds to a single point, and that every point corresponds to some length.


§5.1 constructed π\pi by means of the length of the complete circle. This section repeats the same method, restricted to a portion of that circle —the arc traveled from (1,0)(1,0), counterclockwise, to any other point—.

The inscribed polygons of §5.1 were constructed starting from a square with vertices (1,0)(1,0), (0,1)(0,1), (−1,0)(-1,0), (0,−1)(0,-1), and doubling sides by means of the arc midpoint —the point CC on the ray OMOM, with MM the midpoint of the chord—. That same procedure, applied repeatedly starting only from (1,0)(1,0) and (0,1)(0,1) —without completing the rest of the square—, generates a sequence of vertices V0=(1,0),V1,V2,…V_0=(1,0), V_1, V_2, \ldots that subdivides, ever more finely, the arc of the first quadrant.

Symmetry of the bisection. Each application of the arc midpoint bisects that arc into two parts of exactly equal length —not only approximately—: the reflection about the line OMOM exchanges the two sub-arcs, and since that reflection is an isometry —it preserves all distances, hence it preserves perimeters of inscribed and circumscribed polygons approximating each half—, both halves have the same length.

Consequently, if level kk of subdivision produces 2k2^k equally spaced vertices along the quadrant —V0,V1,…,V2kV_0,V_1,\ldots,V_{2^k}, with V2k=(0,1)V_{2^k}=(0,1)—, the arc length from (1,0)(1,0) to VjV_j is, by symmetry, exactly

ℓ(Vj)=j2k⋅π2\ell(V_j)=\frac{j}{2^k}\cdot\frac\pi2

—a jj-th part of 2k2^k of the quarter circle, whose total length is π/2\pi/2 by the definition of §5.1—.


With arc length defined at the vertices of the subdivision, it remains to verify that those vertices, and the length that separates them, behave consistently with the order in which they appear along the arc. It is the content of §2.

Proposition (monotonicity). For j<j′j<j', ℓ(Vj)<ℓ(Vj′)\ell(V_j)<\ell(V_{j'}), and VjV_j has xx-coordinate strictly greater than that of Vj′V_{j'}.

Proof. The first claim is immediate from the formula of §1: ℓ(Vj)=jπ/2k+1<j′π/2k+1=ℓ(Vj′)\ell(V_j)=j\pi/2^{k+1}<j'\pi/2^{k+1}=\ell(V_{j'}). For the second, it suffices to verify it for consecutive vertices —j′=j+1j'=j+1— and proceed by induction on the levels of bisection. Let A,BA,B be two points of the first quadrant on the circle with xA>xBx_A>x_B —inductive hypothesis—, and CC the midpoint of the arc ABAB, with M=(A+B)/2M=(A+B)/2. Since A,BA,B are both in the first quadrant, MM —a convex combination of AA and BB— has xx-coordinate strictly between xBx_B and xAx_A; and C=M/∣M∣C=M/|M|, the radial projection of MM onto the circle, preserves that relative position —scaling a vector by a positive factor does not change on which side of a line through the origin it lies, and the lines x=xAx=x_A and x=xBx=x_B seen as cones from the origin separate MM in the same angular region as AA and BB—. Hence xB<xC<xAx_B<x_C<x_A. Since the initial square already satisfies x(1,0)=1>0=x(0,1)x_{(1,0)}=1>0=x_{(0,1)}, the induction on the levels of bisection gives the monotonicity at each level. ■\blacksquare

Let now Lk(t)L_k(t) and Uk(t)U_k(t), for a given t∈[0,π/2]t\in[0,\pi/2] and each level kk, be the consecutive vertices of level kk such that ℓ(Lk(t))≤t<ℓ(Uk(t))\ell(L_k(t))\le t<\ell(U_k(t)) —they exist and are unique by the monotonicity just proved, and because {jπ/2k+1}\{j\pi/2^{k+1}\} is a complete partition of [0,π/2][0,\pi/2] into 2k2^k equal subintervals—. By construction, Lk+1(t)L_{k+1}(t) coincides with Lk(t)L_k(t) or advances to the new midpoint inserted between Lk(t)L_k(t) and Uk(t)U_k(t), never moving back; similarly, Uk(t)U_k(t) never advances. Let xk:=x(Lk(t))x_k:=x(L_k(t)), Xk:=x(Uk(t))X_k:=x(U_k(t)).

Proposition (squeezing bound). (xk)(x_k) is decreasing, (Xk)(X_k) is increasing, Xk≤xkX_k\le x_k for all kk, and xk−Xk→0x_k-X_k\to0 in the sense that, for every ε>0\varepsilon>0, there exists kk with xk−Xk<εx_k-X_k<\varepsilon.

Proof. Monotonicity of (xk)(x_k) and (Xk)(X_k): a direct consequence of the previous Proposition and of the fact that LkL_k does not move back, UkU_k does not advance. Xk≤xkX_k\le x_k: since Lk(t)L_k(t) precedes Uk(t)U_k(t) in the subdivision, x(Uk)<x(Lk)x(U_k)<x(L_k) by monotonicity. For the difference: the distance between Lk(t)L_k(t) and Uk(t)U_k(t) —two consecutive vertices of level kk— is, by Pythagoras, at most the length of the arc that separates them, π/2k+1\pi/2^{k+1} —a chord never exceeds the arc it subtends, since the chord is, itself, an inscribed approximation of a single segment—; and xk−Xkx_k-X_k, the difference of the xx-coordinates of two points, is at most that same distance. By the Archimedean Lemma —section 4.1, §4—, given ε>0\varepsilon>0 there exists kk with π/2k+1<ε\pi/2^{k+1}<\varepsilon. ■\blacksquare


With (xk)(x_k) decreasing and bounded and (Xk)(X_k) increasing and bounded, squeezing against each other, the supremum axiom produces a single candidate point. It remains to verify that this point has, indeed, arc length tt. It is the content of §3.

§3. Existence of W(t) in the first quadrant

Section titled “§3. Existence of W(t) in the first quadrant”

By the supremum axiom, (xk)(x_k) decreasing and bounded below by 00 has ξ:=inf⁡{xk}\xi:=\inf\{x_k\}; by the Lemma of existence of the infimum —or, symmetrically, the axiom itself applied to −Xk-X_k—, (Xk)(X_k) has sup⁡{Xk}\sup\{X_k\}. By the same approximation argument that closed the definition of ee in §4.2 §4 and that of π\pi in §5.1 §4 —bounding the difference and forcing it below any ε\varepsilon via the Archimedean Lemma, already verified in the Proposition of §2—, inf⁡{xk}=sup⁡{Xk}=:ξ\inf\{x_k\}=\sup\{X_k\}=:\xi.

Definition. W(t):=(ξ,1−ξ2)W(t):=(\xi,\sqrt{1-\xi^2}), for t∈[0,π/2]t\in[0,\pi/2] —the positive root, corresponding to the first quadrant—.

Proposition. ℓ(W(t))=t\ell(W(t))=t, where ℓ\ell is extended to any point of the arc by means of the same supremum/infimum of inscribed/circumscribed polygons of §1, now with final endpoint W(t)W(t) instead of a vertex of the subdivision.

Proof. For each kk, the point W(t)W(t) —with xx-coordinate equal to ξ∈[Xk,xk]\xi\in[X_k,x_k], since ξ\xi is the common value of infimum and supremum just proved— is located, by the monotonicity of §2, between Lk(t)L_k(t) and Uk(t)U_k(t) in the subdivision. Extending the inscribed path V0,…,Lk(t)V_0,\ldots,L_k(t) with the additional chord Lk(t) W(t)L_k(t)\,W(t) gives an inscribed approximation of the arc up to W(t)W(t) of length ℓ(Lk(t))+∣Lk(t)W(t)∣\ell(L_k(t))+|L_k(t)W(t)|; since ∣Lk(t)W(t)∣≤π/2k+1|L_k(t)W(t)|\le\pi/2^{k+1} —same bound as in §2, since W(t)W(t) is in the same sub-arc as Uk(t)U_k(t)—, it follows that ℓ(W(t))≥ℓ(Lk(t))\ell(W(t))\ge\ell(L_k(t)) taking supremum over this family of approximations, and ℓ(W(t))≤ℓ(Lk(t))+π/2k+1\ell(W(t))\le\ell(L_k(t))+\pi/2^{k+1} bounding from above with the dual argument —chords plus the tangent at W(t)W(t)—. Since ℓ(Lk(t))≤t<ℓ(Lk(t))+π/2k+1\ell(L_k(t))\le t<\ell(L_k(t))+\pi/2^{k+1} —definition of LkL_k—, both chains of inequalities, valid for all kk, force ℓ(W(t))=t\ell(W(t))=t by the Archimedean Lemma, exactly as it forced L=ℓL=\ell in the Proposition of §4.2 §4. ■\blacksquare


With WW defined on [0,π/2][0,\pi/2], extending it to the three remaining quadrants and then to all of R\mathbb{R} does not require repeating the construction: it requires, only, reflecting and repeating with periodicity. It is the content of §4.

§4. Extension to the whole circle and to ℝ

Section titled “§4. Extension to the whole circle and to ℝ”

Extension to the four quadrants. The points (1,0)(1,0), (0,1)(0,1), (−1,0)(-1,0), (0,−1)(0,-1) divide the circle into four arcs, each of length π/2\pi/2 —by construction of the initial square in §5.1—. For t∈[π/2,π]t\in[\pi/2,\pi], define W(t)W(t) by repeating the construction of §3 on the arc from (0,1)(0,1) to (−1,0)(-1,0), with t−π/2∈[0,π/2]t-\pi/2\in[0,\pi/2] playing the role of the variable of §3; similarly for [π,3π/2][\pi,3\pi/2] and [3π/2,2π][3\pi/2,2\pi]. In each case, the Proposition of §3 —with the coordinates and signs adjusted to the corresponding quadrant— gives ℓ(W(t))=t\ell(W(t))=t.

Extension to R\mathbb{R} by periodicity. For arbitrary t∈Rt\in\mathbb{R}, the Archimedean Lemma gives an integer nn such that t−2nπ∈[0,2π)t-2n\pi\in[0,2\pi) —the “remainder” of dividing tt by 2π2\pi—; that integer is unique, by the same uniqueness of Euclidean division. Define

W(t):=W(t−2nπ)W(t):=W(t-2n\pi)

using the construction already made on [0,2π)[0,2\pi).

Proposition (periodicity and restricted injectivity). W(t+2π)=W(t)W(t+2\pi)=W(t) for all t∈Rt\in\mathbb{R}, and WW restricted to [0,2π)[0,2\pi) is injective.

Proof. The periodicity is immediate from the definition —t+2πt+2\pi and tt leave the same remainder modulo 2π2\pi—. The injectivity on [0,2π)[0,2\pi) follows from the strict monotonicity of ℓ\ell within each quadrant —§2— and from the fact that ℓ\ell distinguishes different quadrants by the signs of the coordinates, which are not repeated within a single turn. ■\blacksquare


With WW completely constructed, there remain to be computed, without any new tool, the four points that divide the circle into exact quadrants. It is the content of §5.

By direct construction —without additional computation—:

W(0)=(1,0)W ⁣(π2)=(0,1)W(π)=(−1,0)W ⁣(3π2)=(0,−1)W(0)=(1,0) \qquad W\!\left(\frac\pi2\right)=(0,1) \qquad W(\pi)=(-1,0) \qquad W\!\left(\frac{3\pi}2\right)=(0,-1)

These four values are not a computation: they are, literally, the four vertices of the square with which the construction of §5.1 started, and the reason why that square —and not a triangle or a hexagon— was chosen as the starting point.


This section translated, step by step, each piece of machinery already constructed in the book into a new geometric context: the Archimedean Lemma again bounded a difference until making it arbitrarily small, the supremum axiom again produced a value from two sequences that squeeze together, and the extension by periodicity —modulo 2π2\pi— used the same Euclidean division with remainder that had been familiar long before this chapter. The only thing genuinely new was geometric, not analytic: the points one works with are no longer numbers on a line, but positions on a curve, and it had to be verified, with care, that the notions of “closer” and “farther” inherited from the real line still make sense when the independent variable travels around a circle and returns, every 2π2\pi, over its own steps.

With W:R→W:\mathbb{R}\to unit circle completely constructed, the chapter has, at last, everything necessary to define sine and cosine as what they genuinely are: not relations between sides of a triangle, but the two coordinates of a point that moves, at constant speed in arc length, around the unit circle.


§5.3 defines the trigonometric functions of a real number from WW, and derives from that definition —does not postulate them— the fundamental identities.