betaFeedback

WorksMathematics

Archimedes' Angle Measure and The Fundamental Theorem of Trigonometry

Abstract

We aim to understand a distant conversation about a trigonometric principle. Although foundational to calculus, this principle and the basic concept of angle measure are often treated tacitly in both elementary and advanced accounts. The primordial idea of counting rotations makes one aware of the integers. In the same vein, Archimedes' measure of partial rotations hints at the completeness of the real numbers. By a less well-known implication, these ideas also lead to the complex numbers and to the Fundamental Theorem of Trigonometry. What does it all mean? And is it true?

Introduction

A café conversation

The place was noisy and memories are fuzzy. Maybe the details were different. Maybe the conversation never took place at all... No matter, we’ll try to recount what we remember and interpret what we recount.

A café conversation

Figure 1.

S: “What is the Fundamental Theorem of Trigonometry?”

S: “Is it sin⁡2θ+cos⁡2θ=1\sin^2\theta+\cos^2\theta=1?”

T: “That’s famous and precedes the Fundamental Theorem, but it’s a corollary.”

S: “How about eiθ=cos⁡θ+isin⁡θe^{i\theta}=\cos\theta+i\sin\theta?”

T: “You’re getting warmer, though the coffee is getting colder. This too is a corollary, less distant in time from the Fundamental Theorem.”

A server comes and freshens the interlocutors’ coffee.

T: (Pauses actively) “You know, counting rotations is a primordial idea. (Pauses again.) It makes us aware of the integers. By the same token, Archimedes’ measure of partial rotations leads to the completeness of the real numbers.”

T: “. . . eh, . . ., though less well-known, Archimedes’ angle measure also unearths the notion of complex numbers, paving the way to the Fundamental Theorem of Trigonometry.”

(T pauses to sip some coffee.)

T: “Okay, here it is.”

Theorem (Fundamental Theorem of Trigonometry). There is an additive non-constant continuous map P(θ)P(\theta) of the real line to the unit circle, unique up to rescaling. This map is surjective.

Stepping back, seeking perspective

By additivity we mean addition of real numbers corresponds to addition of angles, or equivalently to multiplication on the unit circle (13).

Early in the development of trigonometric calculus we come across the identity

lim⁡θ→0sin⁡θθ=1.(1)\lim_{\theta\to0}\frac{\sin\theta}{\theta}=1. \tag*{(1)}

But what exactly is θ\theta?

The limit (1) and its cosine companion are central in developing the derivatives of trigonometric functions. Yet a number of standard textbooks give circular arguments in justifying (1), as [16] observes. A mystifying feature of today’s calculus books is their lack of discussion of what exactly is meant by the measure θ\theta of an angle. The subject is assumed to have been treated somehow in secondary school. However, this is rarely so. Secondary school curricula typically follow the G.D. Birkhoff axioms [1], and in particular the Birkhoff Protractor Postulate, which assumes the existence of angle measure, bypassing this issue by design. Major texts such as [7] and [12] are based on the synthetic axiomatization of Hilbert and on a modified set of Birkhoff axioms, respectively.

The notion of solid angle may get an entire chapter in an advanced monograph [2]. Yet in pre-calculus or early portions of calculus treatments, instead of developing angle measure within the framework of just-learned Cartesian geometry, books often go outside this framework by appealing to pictures not only for motivation, but also for justification. As a consequence, one forfeits the opportunity to present angle measure as a basic paradigm of calculus.

Typically, calculus books state θ\theta is the length of the subtended arc along the unit circle, use this to define sin⁡θ\sin\theta, then go on to derive (1) by sandwiching a sector’s area between the areas of inscribed and circumscribed triangles. At the beginning of Cartesian geometry, none of this has yet been defined analytically [16].

The book [19] comes close to defining θ\theta, as it discusses the history of the subject at length. The books [9], [10] ignore the issue completely, although the material here would fit perfectly there. Whatever definition one takes for the measure of the angle, it should be additive: When angles are stacked, their measures should add (Figure 1).

Two illustrations of additive angle stacking

Figure 1 (PDF p. 2). Additivity.

Additive angle measure θ\theta was used implicitly by Archimedes (287 – 212 BCE), as part of his work [8] leading to his estimate of the half-circumference of the unit circle,

22371<π<227.(2)\frac{223}{71}<\pi<\frac{22}{7}. \tag*{(2)}

By contrast, Hipparchus (190 – 120 BCE) [20] and Ptolemy (100 – 170 CE) [21] used chord measure θ1\theta_1, which is equivalent to θ\theta but not additive, to build their trigonometric tables. Aryabhata (476 – 550 CE) presented the first additive trigonometric tables [18]. His use of ardha-jya for half-chords, shortened to jya or jiva, morphed into the Arabic jaib (“pocket”) which translates into the Latin sinus [14].

The path below is in three stages. We first introduce angle stacking and show how it leads naturally toward the complex-number multiplication law. We then follow Archimedes in constructing additive angle measure and proving its existence and uniqueness. Finally, we recover trigonometry from angle measure and establish the Fundamental Theorem of Trigonometry.

Geometry—angles, and how to stack them

Working in the Cartesian plane, we will take points to be ordered pairs of numbers, say P=(x,y)P=(x,y), P′=(x′,y′)P'=(x',y'), P′′=(x′′,y′′)P''=(x'',y''). Points may be added, P+P′:=(x+x′,y+y′)P+P':=(x+x',y+y'), and scaled by numbers tt, tP:=(tx,ty)tP:=(tx,ty).

An angle is a pair of rays emanating from a point, the vertex of the angle. If the vertex is the origin O=(0,0)O=(0,0), then an angle is determined by the intersections of its rays with the unit circle. If the points of intersection are denoted NN and PP, angles are well-defined even when P=±NP=\pm N. Notice that angles so-defined are unoriented—if we swap rays we obtain the same angle. If we choose an ordering for the rays then we obtain the notion of oriented angle. We will be somewhat cavalier about the oriented/unoriented dichotomy of angles, except where precision is required.

In what follows, we make the normalizing assumption that N=(1,0)N=(1,0). Readers needn’t worry—generality will not be lost. With this normalization, an angle is determined by a single point PP on the unit circle, together with ordering. We need to choose an ordering for the elements of {N,P}\{N,P\}, or eschew orientation.

In the Cartesian plane, angles may be stacked, leading to complex multiplication and division. When stacking two unoriented angles, it’s not clear whether we are adding or subtracting them since, for ∠α\angle\alpha unoriented, ∠α=−∠α\angle\alpha=-\angle\alpha, so ∠α+∠β\angle\alpha+\angle\beta actually stands for ±∠α±∠β\pm\angle\alpha\pm\angle\beta. This is already manifested in ancient tables, such as Ptolemy’s [21], where both sums and differences of angles are tabulated.

Let PP and P′P' be points on the unit circle corresponding to two given angles. Then the angles may be stacked (Figure 2), resulting in the angle corresponding to a point P′′P'' on the unit circle. The illustration seems to suggest subliminal orientation, denoting a sum and not a difference. But not to worry, the geometry and the algebra will ally to bring out both possibilities together.

Figure 2. Angle stacking.

To make this precise, draw the circle with center Q=x′PQ=x'P and radius ∣y′∣|y'|. Note that in the figure, x′x' is only slightly smaller than 1, the length of the hypotenuse of triangle PP. The small gap is significant, as QQ and PP play distinct roles.

When x′y′≠0x'y'\ne0, we claim this circle intersects the unit circle at two points, both labelled P′′P'', for beneficial ambiguity.1

P′′=(xx′−yy′,x′y+xy′)P′′=(xx′+yy′,x′y−xy′).(3)P^{\prime\prime}=(xx^{\prime}-yy^{\prime},x^{\prime}y+xy^{\prime})\\ P^{\prime\prime}=(xx^{\prime}+yy^{\prime},x^{\prime}y-xy^{\prime}). \tag*{(3)}

Since we obtain two points P′′P^{\prime\prime}, angle stacking in fact results in two angles, which may remind the reader of the complex product and complex quotient of PP and P′P^{\prime}. Although we do not develop complex numbers here, we’ll speak of the real and imaginary parts of a point in the plane.

To derive (3), let ⟨P,P′⟩=xx′+yy′\langle P,P^{\prime}\rangle=xx^{\prime}+yy^{\prime}, ∣P∣2=⟨P,P⟩|P|^{2}=\langle P,P\rangle, and P⊥=(−y,x)P^{\perp}=(-y,x). Then PP is on the unit circle if and only if ∣P∣2=1|P|^{2}=1, and P,P′P,P^{\prime} on the unit circle satisfy ⟨P,P′⟩=0\langle P,P^{\prime}\rangle=0 if and only if P′=±P⊥P^{\prime}=\pm P^{\perp}.

Since P′′P^{\prime\prime} is on the circle of center QQ and radius ∣y′∣|y^{\prime}|, we may write P′′=Q+y′RP^{\prime\prime}=Q+y^{\prime}R for some point RR on the unit circle. Then

1=∣P′′∣2=∣x′P+y′R∣2=x′2∣P∣2+y′2∣R∣2+2x′y′⟨P,R⟩=(x′2+y′2)+2x′y′⟨P,R⟩=1+2x′y′⟨P,R⟩,\begin{aligned} 1=|P^{\prime\prime}|^{2}&=|x^{\prime}P+y^{\prime}R|^{2}\\ &=x^{\prime2}|P|^{2}+y^{\prime2}|R|^{2}+2x^{\prime}y^{\prime}\langle P,R\rangle\\ &=(x^{\prime2}+y^{\prime2})+2x^{\prime}y^{\prime}\langle P,R\rangle\\ &=1+2x^{\prime}y^{\prime}\langle P,R\rangle, \end{aligned}

which occurs if and only if ⟨P,R⟩=0\langle P,R\rangle=0, i.e., if and only if R=±P⊥R=\pm P^{\perp}. This yields P′′=x′P±y′P⊥P^{\prime\prime}=x^{\prime}P\pm y^{\prime}P^{\perp}, which is (3). That establishes the claim.

Given any two points P=(x,y)P=(x,y), P′=(x′,y′)P^{\prime}=(x^{\prime},y^{\prime}) on the unit circle, we can define their product PP′PP^{\prime} and quotient P/P′P/P^{\prime} by (3). One can check directly that this multiplication rule is commutative and associative, with NN as an identity element: for any point RR on the unit circle,

PP′=P′P,P(P′R)=(PP′)R,PN=NP=P.PP^{\prime}=P^{\prime}P,\qquad P(P^{\prime}R)=(PP^{\prime})R,\qquad PN=NP=P.

(Just copy and paste from an introduction to complex numbers, but don’t tell.) If P‾:=(x,−y)\overline{P}:=(x,-y), then PP‾=NP\overline{P}=N for any point PP on the unit circle, and P/P′=PP′‾P/P^{\prime}=P\overline{P^{\prime}} is the Hermitian product.

In the next section, we define an additive angle measure θ(P)\theta(P), assigning real numbers to points PP in the punctured circle {x≠−1}\{x\ne-1\}. To take into account both cases in (3), we require

θ(P)=−θ(P‾),x≠−1,(4)\theta(P)=-\theta(\overline{P}),\qquad x\ne-1, \tag*{(4)}

and consequently, θ(N)=0\theta(N)=0. We then interpret Figure 1 by

θ(PP′)=θ(P)+θ(P′),x>0, x′>0.(5)\theta(PP^{\prime})=\theta(P)+\theta(P^{\prime}),\qquad x>0,\ x^{\prime}>0. \tag*{(5)}

It is in this form that we shall establish additivity of θ=θ(P)\theta=\theta(P).

Note that (5) implies

θ(P2)=2θ(P),x>0.(6)\theta(P^{2})=2\theta(P),\qquad x>0. \tag*{(6)}

For P=(x,y)P=(x,y) on the punctured unit circle x≠−1x\ne-1, write

P=(1+x2+2x,y2+2x).(7)\sqrt{P}=\left(\frac{1+x}{\sqrt{2+2x}},\frac{y}{\sqrt{2+2x}}\right). \tag*{(7)}

Then P↦PP\mapsto\sqrt{P} maps the punctured unit circle continuously and bijectively onto the right-half unit circle x>0x>0, with inverse P↦P2P\mapsto P^{2}, and the imaginary parts of PP and P\sqrt{P} have the same sign.

Since (PP′)2=PP′(\sqrt{P}\sqrt{P^{\prime}})^{2}=PP^{\prime}, we have

PP′=PP′,(8)\sqrt{PP^{\prime}}=\sqrt{P}\sqrt{P^{\prime}}, \tag*{(8)}

up to sign, whenever both sides of (8) are defined. (This is one instance where we benefit from beneficial ambiguity.) When the imaginary parts of PP, P′P^{\prime}, and PP′PP^{\prime} have the same sign, the imaginary parts of the sides of (8) have the same sign. Hence (8) is correct as written, when the imaginary parts of PP, P′P^{\prime}, and PP′PP^{\prime} have the same sign.

Measurement

Enter Archimedes

Archimedes interpreted angle measure θ\theta as the length of the arc along the unit circle subtended by the angle (Figure 4). In his treatises On the sphere and the cylinder and Measurement of a circle [8], Archimedes grappled with the ideas underlying the length, area, and volume of curves, surfaces, and solids.

Figure 4. Elbow Bounds.

Archimedes not only computed the circumference and area of a circle, the surface area of a sphere, and the volume of a ball, he also isolated a priori properties such measures ought to have. This led to a plethora of ideas including, explicitly or implicitly, the definition of a convex curve, the Archimedes Convexity Postulate, the additivity of angle measure (§ 2), the bisection method (§ 3.2), and the completeness of the real number line.

The Archimedes Convexity Postulate (ACP) [4], [5], states:

Archimedes Convexity Postulate

Figure 3. Archimedes Convexity Postulate (ACP).

If two convex curves have common endpoints A,BA,B along a base segment and one curve lies below the other then the lower curve has smaller length than the upper curve.

Let P=(x,y)P=(x,y) be a point on the unit circle first quadrant. In Figure 4, θ\theta is the as-yet-undefined length of the arc. We will use the ACP to derive the following estimates:

y<θ<1−x+y,x,y>0(Elbow Bounds).(9)y<\theta<1-x+y,\qquad x,y>0\qquad\textit{(Elbow Bounds)}. \tag*{(9)}

In Figure 4, applying the ACP (with base PNPN) to the curve θ\theta (the lower curve) and the elbow consisting of 1−x1-x and yy (the upper curve), we obtain

θ<1−x+y;\theta< 1-x+y;

applying the postulate to the chord PNPN and the arc θ\theta we obtain PN<θPN<\theta. Then, by the Pythagorean theorem, y<PNy<PN. Putting these together we have (9).

This is all we require of the ACP and its direct role ends here: the postulate serves to certify that the Elbow Bounds capture what arc length ought to satisfy. Below we derive the existence and uniqueness of an additive angle measure θ\theta satisfying the Elbow Bounds.

Archimedean bisection: chord estimates

Archimedean bisection is based on Figure 5, a starting point for the method of exhaustion of Eudoxus and Euclid. Figure 5, Figure 6, and Figure 7 are in Archimedes’ On the sphere and the cylinder (see the figures of Propositions 1–6 therein) and his Measurement of a circle.

Figure 5. Inscribed and circumscribed chords; lower diagram enlarged.

Figure 6. Bisection.

Figure 7. First Cut.

In Figure 5, let NN, PP, P1P_1 be three points on the upper-half unit circle, y>0y>0. Let θ1\theta_1, θ1∗\theta_1^* be the inscribed and circumscribed chord-length sums on the top, and let θ1\theta_1, θ2\theta_2, θ2∗\theta_2^*, θ1∗\theta_1^* be the inscribed and circumscribed chord-length sums on the bottom. By the triangle inequality, or by the ACP applied to the four polygonal convex curves joining the endpoints NN and PP,

θ1<θ2<θ2∗<θ1∗,y>0.\theta_1 < \theta_2 < \theta_2^* < \theta_1^*, \qquad y>0.

By incrementally adding points on the arc, partitioning subarcs between adjoining points, we obtain inscribed and circumscribed polygonal approximations of the arc whose lengths are arbitrarily close to each other. This is Proposition 3 in On the sphere and the cylinder. By adding points repeatedly to the arc, we will establish the Archimedes sequences

θ1<θ2<θ3<⋯<θ3∗<θ2∗<θ1∗,y>0.(10)\theta_1 < \theta_2 < \theta_3 < \cdots< \theta_3^* < \theta_2^* < \theta_1^*, \qquad y>0. \tag*{(10)}

This is for any point PP in the upper-half circle, while the Elbow Bounds are only for PP in the unit circle first quadrant.

To obtain (10) assume the points P=(x,y)P=(x,y), P1P_1, P2P_2, P3,…P_3,\ldots are generated by bisection (Figure 6): P1P_1 bisects the arc from NN to PP, and each PnP_n bisects the arc from NN to Pn−1P_{n-1}. Thus PnP_n cuts off 1/2n1/2^n of the arc from NN to PP, and the corresponding chords NPnNP_n shrink accordingly. In particular, with Pn=(xn,yn)P_n=(x_n,y_n), the sequences xnx_n and yny_n are increasing and decreasing respectively.

We now focus on the nn-th bisection. In Figure 7, let MnM_n be the midpoint of the chord Pn−1NP_{n-1}N, and let PnP_n and TnT_n be the intersections of the line through OO and MnM_n with the unit circle and the vertical line through NN. Let Pn−1=(xn−1,yn−1)P_{n-1}=(x_{n-1},y_{n-1}), Rn−1=(1,yn−1)R_{n-1}=(1,y_{n-1}), Pn=(xn,yn)P_n=(x_n,y_n), Qn−1=(xn−1,0)Q_{n-1}=(x_{n-1},0), Qn=(xn,0)Q_n=(x_n,0), Tn=(1,yn/xn)T_n=(1,y_n/x_n). Then PnP_n is well-defined if yn−1≥0y_{n-1}\geq0, and computing MnM_n then Pn=Mn/∣Mn∣P_n=M_n/\lvert M_n\rvert shows Pn=Pn−1P_n=\sqrt{P_{n-1}}.

From the Figures, it makes sense to define θn:=2n−1∣Pn−1N∣\theta_n:=2^{n-1}\lvert P_{n-1}N\rvert and θn∗:=2n∣TnN∣\theta_n^*:=2^n\lvert T_nN\rvert. Since MnM_n is the chord Pn−1NP_{n-1}N’s midpoint, θn=2n∣MnN∣\theta_n=2^n|M_nN|. Since the triangles OMnNOM_nN and OPnQnOP_nQ_n are congruent, ∣MnN∣=∣PnQn∣=yn|M_nN|=|P_nQ_n|=y_n, hence θn=2nyn\theta_n=2^ny_n, which implies θn∗=2nyn/xn\theta_n^*=2^ny_n/x_n. Thus, for n≥1n\geq1 and y>0y>0, we conclude

P=Pn2n,Pn=P2n,θn∗=θnxn,θn=2nyn.(11)P=P_n^{2^n},\qquad P_n=\sqrt[2^n]{P},\qquad\theta_n^*=\frac{\theta_n}{x_n},\qquad\theta_n=2^ny_n. \tag*{(11)}

Moreover 2n∣PnN∣>2n∣MnN∣2^n|P_nN|>2^n|M_nN| and

2n+1∣Tn+1N∣=2n(∣PnTn+1∣+∣Tn+1N∣)<2n(∣TnTn+1∣+∣Tn+1N∣)=2n∣TnN∣,2^{n+1}|T_{n+1}N|=2^n\left(|P_nT_{n+1}|+|T_{n+1}N|\right) <2^n\left(|T_nT_{n+1}|+|T_{n+1}N|\right)=2^n|T_nN|,

hence

θn<θn+1<θn+1∗<θn∗,\theta_n<\theta_{n+1}<\theta_{n+1}^*<\theta_n^*,

which implies (10).

Uniqueness of angle measure

3.3. Uniqueness of angle measure. There is at most one real-valued additive angle measure on the punctured unit circle satisfying the Elbow Bounds.

To establish uniqueness of angle measure, let α=α(P)\alpha=\alpha(P) and β=β(P)\beta=\beta(P) be real-valued additive angle measures satisfying the Elbow Bounds (9). By additivity (5), it is enough to show α=β\alpha=\beta for PP in the unit circle first quadrant. With PnP_n as above, let αn=α(Pn)\alpha_n=\alpha(P_n) and βn=β(Pn)\beta_n=\beta(P_n) be the corresponding angle measures. Replacing PP by Pn=(xn,yn)P_n=(x_n,y_n) in (9),

yn<αn<1−xn+yn,yn<βn<1−xn+yn.y_n<\alpha_n<1-x_n+y_n,\qquad y_n<\beta_n<1-x_n+y_n.

By additivity (5), 2nαn=α2^n\alpha_n=\alpha and 2nβn=β2^n\beta_n=\beta. This implies

∣α−β∣=2n∣αn−βn∣<2n(1−xn)=2nyn21+xn<2−nα2|\alpha-\beta|=2^n|\alpha_n-\beta_n|<2^n(1-x_n)=2^n\frac{y_n^2}{1+x_n}<2^{-n}\alpha^2

for n≥1n\geq1. Hence α=β\alpha=\beta, establishing uniqueness.

Remark. Note that we did not require continuity. Rigorous treatments of (1) assume a smoothness context from the start—via power series [17, 11], or a differential equation and arclength integral [6, 3]; continuity follows as a consequence of smoothness. Here it is derived from additivity and the Elbow Bounds.

Existence of angle measure

3.4. Existence of angle measure. There is at least one real-valued additive angle measure on the punctured unit circle satisfying the Elbow Bounds (9).

Turning to existence, the sequences in (10) are bounded. By (11), yn→0y_n\to0, hence xn→1x_n\to1, as n→∞n\to\infty. Since the real line is complete, the sequences in (10) have a common limit θ=θ(P)\theta=\theta(P) for PP in the upper-half circle y>0y>0. Extend θ(P)\theta(P) to the lower-half unit circle y<0y<0 by θ(P)=−θ(1/P)\theta(P)=-\theta(1/P), and set θ(N)=0\theta(N)=0. Then, by construction, (4) and (6) are valid. From Figure 7 with n=1n=1 and P0=PP_0=P,

y<∣PN∣=θ1<θ<θ1∗=∣PT1∣+∣T1N∣<∣PR0∣+∣R0T1∣+∣T1N∣=1−x+y,y<|PN|=\theta_1<\theta<\theta_1^*=|PT_1|+|T_1N| <|PR_0|+|R_0T_1|+|T_1N|=1-x+y,

verifying the elbow bounds (9).

Let P′′=PP′=(x′′,y′′)P''=PP'=(x'',y''), and let θ=θ(P)\theta=\theta(P), θ′=θ(P′)\theta'=\theta(P'), and θ′′=θ(P′′)\theta''=\theta(P''). Let Pn=(xn,yn)P_n=(x_n,y_n), Pn′=(xn′,yn′)P'_n=(x'_n,y'_n), Pn′′=(xn′′,yn′′)P''_n=(x''_n,y''_n), n≥1n\geq1, be the bisection sequences starting from PP, P′P', P′′P'' respectively, with θn\theta_n, θn′\theta'_n, θn′′\theta''_n, n≥1n\geq1, the corresponding chord-length sums. It remains to verify the additivity property (5).

If yy′y′′=0yy'y''=0, then (5) is immediate, hence we may assume yy′y′′≠0yy'y''\ne0. Then there are two cases, yy′>0yy'>0 and yy′<0yy'<0.

If yy′>0yy'>0, by (4), (5) is valid for PP, P′P' if and only if (5) is valid for P‾,P′‾\overline{P},\overline{P'}. Hence we may assume PP, P′P' are in the first quadrant. It follows PnP_n, Pn′P'_n are in the first quadrant for n≥1n\geq1. By induction and (8), Pn′′=PnPn′P''_n=P_nP'_n. By (3), yn′′=xn′yn+xnyn′y''_n=x'_ny_n+x_ny'_n. By (11),

θn′′=xn′θn+xnθn′(12)\theta''_n=x'_n\theta_n+x_n\theta'_n \tag*{(12)}

(half-chord addition identity).

This identity is the Hipparchus–Ptolemy–Aryabhata chord addition alluded to in §1.2. Since xn→1x_n\to1, xn′→1x'_n\to1, we obtain θ′′=θ+θ′\theta''=\theta+\theta', which is (5).

If yy′<0yy'<0, then x′′=xx′−yy′>0x''=xx'-yy'>0. Since yy′<0yy'<0, y′′(−y)y''(-y) and y′′(−y′)y''(-y') have opposite signs. By switching the roles of PP and P′P' if necessary, we may assume y′′(−y)>0y''(-y)>0. Applying the first case to P′′P'' and P‾\overline{P},

θ(P′)=θ(P′′/P)=θ(P′′)+θ(P‾)=θ(PP′)−θ(P),\theta(P')=\theta(P''/P)=\theta(P'')+\theta(\overline{P})=\theta(PP')-\theta(P),

which is (5). This establishes existence.

Properties of the function theta

3.5. Properties of the function θ\theta. The function θ(P)\theta(P) is a nondecreasing function of yy, for PP in the right-half unit circle.

To derive this, let PP be in the unit circle first quadrant. From (7), y1y_1 is an increasing function of yy. Similarly, with y2y_2 playing the role of y1y_1, y2y_2 is an increasing function of y1y_1, hence an increasing function of yy. Continuing in this manner, yny_n, n≥1n\geq1, are increasing functions of yy. By (11), θn\theta_n, n≥1n\geq1, are increasing functions of yy. Passing to the limit, we obtain

θ(P)=θ(1−y2,y)\theta(P)=\theta\left(\sqrt{1-y^2},y\right)

is an odd function of yy on the right-half unit circle, the result follows.

The number pi

3.6. The number π\pi. Define π:=2θ(J)\pi:=2\theta(J) where JJ is the point (0,1)(0,1). Since 2θ1(J)=222\theta_1(J)=2\sqrt{2} and 2θ1∗(J)=42\theta_1^*(J)=4, 22<π<42\sqrt{2}<\pi<4. Archimedes achieved (2) by doubling a hexagon four times, resulting in a 96-gon. To achieve (2) using (10) one needs to calculate

2θ6=642−2+2+2+2+2,2\theta_6=64\sqrt{2-\sqrt{2+\sqrt{2+\sqrt{2+\sqrt{2+\sqrt{2}}}}}},

together with its circumscribed counterpart 2θ6∗=2θ6/x62\theta_6^*=2\theta_6/x_6. This corresponds to doubling an octagon four times, resulting in a 128-gon.

Plugging x=xn−1,y=yn−1x=x_{n-1},y=y_{n-1} into (7) we obtain

(xn,yn)=(1+xn−12+2xn−1,yn−12+2xn−1),(x_n,y_n)=\left(\frac{1+x_{n-1}}{\sqrt{2+2x_{n-1}}},\frac{y_{n-1}}{\sqrt{2+2x_{n-1}}}\right),

from which follows 2xn=2+2xn−12x_n=\sqrt{2+2x_{n-1}} and 2yn=yn−1/xn2y_n=y_{n-1}/x_n. By (11), this implies θn=θn−1/xn\theta_n=\theta_{n-1}/x_n, which telescopes as follows at P=JP=J:

θn=θ1x2⋯xn=θ1x1⋯xn,\theta_n=\frac{\theta_1}{x_2\cdots x_n}=\frac{\theta_1}{x_1\cdots x_n},

because θ1=2\theta_1=\sqrt{2} and x1=1/2x_1=1/\sqrt{2}. Since θn→π/2\theta_n\to\pi/2,

2π=x1x2x3⋯=222+222+2+22⋯ .\frac{2}{\pi}=x_1x_2x_3\cdots=\frac{\sqrt{2}}{2}\frac{\sqrt{2+\sqrt{2}}}{2}\frac{\sqrt{2+\sqrt{2+\sqrt{2}}}}{2}\cdots.

This is Viète’s product formula (1593), said to be “the oldest infinite product” [15 p. 136] and “the first time an infinite process was explicitly written as a mathematical formula” [13 p. 50].

By (9), and since θ(P)\theta(P) is an odd function of yy, θ(P)\theta(P) is continuous at P=NP=N and θ(P)≠0\theta(P)\ne0 when P≠NP\ne N. If P≠P′P\ne P' are in the right-half unit circle and close to each other, then P/P′≠NP/P' \ne N and close to NN. By (5),

θ(P/P′)=θ(P)−θ(P′),x>0,x′>0.\theta(P/P') = \theta(P) - \theta(P'), \qquad x > 0, x' > 0.

It follows that θ(P)\theta(P) is continuous and injective on the right-half unit circle. By the doubling identity (6), θ(P)\theta(P) is continuous and injective on the punctured unit circle.

Since θ(P)\theta(P) is an increasing function of yy and θ(±J)=±π/2\theta(\pm J) = \pm\pi/2, θ(P)\theta(P) maps the right-half unit circle into (−π/2,π/2)(-\pi/2,\pi/2). By the intermediate value theorem, θ(P)\theta(P) is a continuous bijection from the right-half unit circle onto (−π/2,π/2)(-\pi/2,\pi/2). By (6) again, θ(P)\theta(P) is a continuous bijection from the punctured unit circle P≠−NP \ne-N onto (−π,π)(-\pi,\pi) (Figure 8).

Figure 8. Angle measure.

Trigonometry

The fundamental theorem of trigonometry

The basic result of trigonometry is the existence and uniqueness, up to scaling, of a continuous map P(θ)=(cos⁡θ,sin⁡θ)P(\theta) = (\cos\theta,\sin\theta) of the real line into the unit circle satisfying the addition formula

P(θ)P(θ′)=P(θ+θ′)(13)P(\theta)P(\theta') = P(\theta+\theta') \tag*{(13)}

for all θ,θ′\theta,\theta'. This form of the addition formula is implicit in the work of de Moivre. Ancient addition formulas are found in the Almagest [21], albeit only for chord lengths. Separating the addition formula into real and imaginary parts yields the usual trigonometric identities for sin⁡θ\sin\theta and cos⁡θ\cos\theta. We already attained a finite-resolution version of this in (12).

By (13), any such map satisfies P(0)=NP(0)=N and P(−θ)=P(θ)‾P(-\theta)=\overline{P(\theta)}. An immediate consequence of (13) is the doubling property

P(θ)=P(θ/2)2.(14)P(\theta)=P(\theta/2)^2. \tag*{(14)}

If P(θ)P(\theta) satisfies (13), then P(τθ)P(\tau\theta) satisfies (13), for any real τ\tau. Hence, at best, there is uniqueness up to rescaling. To show this is indeed the case, we need to attach a scale to each such map P(θ)P(\theta). This scale, or yardstick, is its least positive period.

Given a continuous map P(θ)P(\theta) satisfying (13) for all θ,θ′\theta,\theta', we say a real ω\omega is a period2 if P(ω)=NP(\omega)=N. Then 0 is a period, every integer multiple of a period is a period, and a limit of periods is a period.

Proof of existence. θ(P)\theta(P) is a continuous bijection from the punctured circle to (−π,π)(-\pi,\pi), with inverse P(θ)P(\theta). Restricting θ(P)\theta(P) to the right half-circle and using the yy coordinate as a parameter, θ(P)\theta(P) is nondecreasing (§3.5), hence θ(P)\theta(P) can be viewed as a continuous, monotone function onto (−π/2,π/2)(-\pi/2,\pi/2); injectivity shows the monotonicity is strict. The inverse of a continuous strictly monotone function is continuous and strictly monotone. Thus P(θ)P(\theta) is continuous on (−π/2,π/2)(-\pi/2,\pi/2). Using the doubling property, P(θ)P(\theta) is continuous on all of (−π,π)(-\pi,\pi).

Since θ(J)=π/2\theta(J)=\pi/2, P(π/2)=JP(\pi/2)=J. By (5), P(θ)P(\theta) satisfies (13) on (−π/2,π/2)(-\pi/2,\pi/2), hence (14) holds for θ\theta in (−π,π)(-\pi,\pi).

Fix α≥π\alpha\ge\pi, and suppose P(θ)P(\theta) is defined on (−α,α)(-\alpha,\alpha) and satisfies (13) on (−α/2,α/2)(-\alpha/2,\alpha/2). If P^(θ)\widehat{P}(\theta) extends P(θ)P(\theta) to (−2α,2α)(-2\alpha,2\alpha) and satisfies (13) on (−α,α)(-\alpha,\alpha), then P^(θ)=P^(θ/2)2=P(θ/2)2\widehat{P}(\theta)=\widehat{P}(\theta/2)^2=P(\theta/2)^2 on (−2α,2α)(-2\alpha,2\alpha), hence P^(θ)\widehat{P}(\theta) is uniquely determined. Conversely, define P^(θ):=P(θ/2)2\widehat{P}(\theta):=P(\theta/2)^2 on (−2α,2α)(-2\alpha,2\alpha). Then P^(θ)\widehat{P}(\theta) satisfies (13) on (−α,α)(-\alpha,\alpha), since

P^(θ)P^(θ′)=P(θ/2)2P(θ′/2)2=P((θ+θ′)/2)2=P^(θ+θ′),\begin{aligned} \widehat{P}(\theta)\widehat{P}(\theta') &= P(\theta/2)^2P(\theta'/2)^2 \\ &= P((\theta+\theta')/2)^2 = \widehat{P}(\theta+\theta'), \end{aligned}

for θ,θ′\theta,\theta' in (−α,α)(-\alpha,\alpha), and P^(θ)\widehat{P}(\theta) extends P(θ)P(\theta), since P^(θ)=P(θ/2)2=P(θ)\widehat{P}(\theta)=P(\theta/2)^2=P(\theta) on (−α,α)(-\alpha,\alpha).

Iterating this, P(θ)P(\theta) can be extended uniquely to (−2π,2π)(-2\pi,2\pi), (−4π,4π)(-4\pi,4\pi), (−8π,8π)(-8\pi,8\pi), …\ldots, with the extensions satisfying (13) on (−π,π)(-\pi,\pi), (−2π,2π)(-2\pi,2\pi), (−4π,4π)(-4\pi,4\pi), …\ldots, resulting in a continuous map P(θ)P(\theta) satisfying (13) for all reals. Since P(π)=P(π/2)2=J2=−NP(\pi)=P(\pi/2)^2=J^2=-N, P(θ)P(\theta) is surjective. Since θ(P)\theta(P) is bijective and P(2π)=P(π)2=NP(2\pi)=P(\pi)^2=N, 2π2\pi is the least positive period of P(θ)P(\theta). We write P(θ)=(cos⁡θ,sin⁡θ)P(\theta)=(\cos\theta,\sin\theta) for this extension.

Proof of uniqueness, step 1. Let P(θ)P(\theta) be any non-constant continuous map P(θ)P(\theta) of the real line into the unit circle satisfying (13). Then P(θ)P(\theta) has a positive period:

Write P(θ)=(x(θ),y(θ))P(\theta)=(x(\theta),y(\theta)). Since 0 is a period, x(0)=1x(0)=1. Since P(θ)P(\theta) is non-constant, there is an α≠0\alpha\ne0 with x(α)<1x(\alpha)<1. Since cos⁡0=1\cos0=1, by continuity, there is an n≥1n\ge1 with x(α)<cos⁡(2π/n)<1x(\alpha)<\cos(2\pi/n)<1. By the intermediate value theorem, there is a β≠0\beta\ne0 with x(β)=cos⁡(2π/n)x(\beta)=\cos(2\pi/n), hence one of y(±β)=±y(β)y(\pm\beta)=\pm y(\beta) equals sin⁡(2π/n)\sin(2\pi/n). By (13), since x(α)x(\alpha) is even,

P(±nβ)=P(±β)n=(x(±β),y(±β))n=(cos⁡(2π/n),sin⁡(2π/n))n=(cos⁡(2π),sin⁡(2π))=N,\begin{aligned} P(\pm n\beta) &= P(\pm\beta)^n = (x(\pm\beta),y(\pm\beta))^n \\ &= (\cos(2\pi/n),\sin(2\pi/n))^n \\ &= (\cos(2\pi),\sin(2\pi)) = N, \end{aligned}

hence one of ±nβ\pm n\beta is a positive period.

Proof of uniqueness, step 2. P(θ)P(\theta) has a least positive period ω\omega:

Since a limit of periods is a period, the infimum ω\omega of all positive periods is a period. For every real θ\theta and every period β>0\beta>0, there is some integer nn for which nβ≤θ<(n+1)βn\beta\le\theta< (n+1)\beta. Hence for every period β>0\beta> 0, every real θ\theta lies within distance β\beta of some period. If ω=0\omega= 0, there would be arbitrarily small positive periods β\beta. This would imply every real θ\theta is the limit of some sequence of periods, and thus P(θ)≡NP(\theta) \equiv N. Since identically constant functions are disallowed, ω>0\omega> 0.

Proof of uniqueness, step 3. By rescaling P(θ)P(\theta) to P(ωθ/2π)P(\omega\theta/2\pi), we may assume ω=2π\omega= 2\pi. Then P(π)=−NP(\pi) = -N, so P(π/2)=JP(\pi/2) = J or P(−π/2)=JP(-\pi/2) = J. By rescaling P(θ)P(\theta) to P(−θ)P(-\theta) if necessary, we may assume P(π/2)=JP(\pi/2) = J. Then P(θ)P(\theta) is uniquely determined:

Since 2π2\pi is the least positive period, P(θ)≠±NP(\theta) \ne\pm N for 0<∣θ∣<π0 < |\theta| < \pi. In particular, P(θ)\sqrt{P(\theta)} is defined for −π<θ<π-\pi< \theta< \pi. Since (13) implies (14),

P(θ/2)=P(θ),−π<θ<π,(15)P(\theta/2) = \sqrt{P(\theta)}, \qquad-\pi< \theta< \pi, \tag*{(15)}

up to sign (beneficial ambiguity again). We claim (15) is correct as written, with no sign adjustment. This is immediate when θ=0\theta= 0, so assume θ≠0\theta\ne0. If the imaginary parts of P(θ)P(\theta) and P(θ/2)P(\theta/2) have opposite signs, then, by the intermediate value theorem, for some θ′\theta' between θ\theta and θ/2\theta/2, we have P(θ′)=NP(\theta') = N or P(θ′)=−NP(\theta') = -N. Since this can’t happen, the imaginary parts of P(θ)P(\theta) and P(θ/2)P(\theta/2) must have the same sign. It follows that the imaginary parts of P(θ)\sqrt{P(\theta)} and P(θ/2)P(\theta/2) have the same sign, establishing the claim. Using (13) and (15) repeatedly, P(θ)P(\theta) satisfies

P(θπ/2)=Jθ,J=(0,1),P(\theta\pi/2) = J^\theta, \qquad J = (0,1),

for all dyadic rationals θ=k/2n\theta= k/2^n. Since the dyadic rationals are dense and P(θ)P(\theta) is continuous, this determines P(θ)P(\theta), completing the proof of the fundamental theorem.

Trigonometric derivatives

Inserting P=(cos⁡θ,sin⁡θ)P = (\cos\theta,\sin\theta) in the Elbow Bounds (9) leads to (1) as follows. The left inequality gives sin⁡θ<θ\sin\theta< \theta, or

sin⁡θθ<1.\frac{\sin\theta}{\theta} < 1.

The right inequality gives θ<1−cos⁡θ+sin⁡θ\theta< 1-\cos\theta+\sin\theta, or

0<1−sin⁡θθ<1−cos⁡θθ=1−cos⁡θθ⋅1+cos⁡θ1+cos⁡θ=sin⁡2θθ(1+cos⁡θ)<θsin⁡θ⋅sin⁡2θθ(1+cos⁡θ)=sin⁡θ1+cos⁡θ,0 < 1-\frac{\sin\theta}{\theta} < \frac{1-\cos\theta}{\theta} = \frac{1-\cos\theta}{\theta}\cdot\frac{1+\cos\theta}{1+\cos\theta} = \frac{\sin^2\theta}{\theta(1+\cos\theta)} < \frac{\theta}{\sin\theta}\cdot\frac{\sin^2\theta}{\theta(1+\cos\theta)} = \frac{\sin\theta}{1+\cos\theta},

and this last expression tends to zero as θ→0\theta\to0. The reasoning here depends on the Elbow Bounds (9), hence requires 0<θ<π/20 < \theta< \pi/2 and proves the right hand limit portion of (1); that sin⁡θ/θ\sin\theta/\theta is an even function verifies (1) in full. As a byproduct, we also obtain

lim⁡θ→01−cos⁡θθ=0.\lim_{\theta\to0}\frac{1-\cos\theta}{\theta}=0.

With this limit and (1) established, the derivatives of trigonometric functions can be computed in the standard way.

Afterwards

S: “Well, that was quite a journey you took us through. Would you regard the angle measure construction as a furthering of the G.D. Birkhoff postulates?”

T: “In a sense, yes. By analogy, an abstract definition of the real numbers is proposed, as a complete ordered field, and then Richard Dedekind comes along and builds a model with his Dedekind Cuts. Historically, the cuts came first, and then the abstract, unique description, though with precursors to the abstraction far earlier on. The Birkhoff postulates include the Protractor Axiom, and we have built a model for a protractor, following Archimedes.”

S: “Yes, I’ve heard of models in mathematics. This gives me a clearer idea of what they’re about.”

T: “Speaking of mathematical models, my first college mathematics professor was Michael Morley. Only later did I learn that he was a distinguished model theorist. Interestingly, in the Real Analysis class that he taught, he had us read on our own about the Dedekind Cuts model in the first chapter of Rudin’s Principles book—Baby Rudin we called it, and started the course assuming everything therein.”

S: “The limit lim⁡θ→0sin⁡θθ\lim_{\theta\to0}\frac{\sin\theta}{\theta} reminds me of my freshman year. We covered it in my calculus class and it seemed routine. But my roommate, who was taking the same class with a different professor kept coming back with stories. When the limit was introduced, he said that his professor ‘couldn’t do it’. He knew that his professor was very smart, maybe even internationally distinguished, so getting stuck on a basic limit seemed out of character. That professor came back the next day and said he still couldn’t do it. He looked it up in the book and said that the book was wrong; he tried other books and they were also wrong. I think they ended up just taking the value of the limit as an axiom.

Would you suggest including the angle measure development in a calculus course?”

T: “Well, I wouldn’t subject them to the whole story, but I would mention the sublime challenge of defining arc length and that this limit is intimately related to it. I would also show the Archimedes Convexity Postulate and its Elbow Bounds specialization. I think that students would appreciate the ancient historical span of calculus concepts, and the exposition challenges that even current textbooks face.”

S: “You had also mentioned that rotations lead to the concept of the integers. I remember being introduced to negative numbers in grade school via the concept on debt. In hindsight, this is a rather abstract notion, not easily incorporated into arithmetic. Full rotations, on the other hand, forward and back, seem more intuitive and were surely known to ancient people long before debt. After all, they saw rotational motion in celestial bodies and in bird maneuvers.”

T: “Yes, quite so. And partial rotations are implicit in Archimedes’ work, with limits of arcs hinting at the completeness of the real numbers.”

The server refreshes the interlocutors’ coffee once more.

T: “You know, the ambiguity of angle stacking brings back memories. Years ago, in another university, when I had an administrative responsibility I was called to a high level meeting with an academic officer. Sliding off the agenda he offered: ‘What are you teaching our students? You make them use obscure concepts like absolute value. In Engineering we have a much better construct: plus/minus, ±\pm. This remark didn’t go over well at the time. Looking back, though, I see that he was anticipating beneficial ambiguity.”

S: “Was Archimedes an engineer?”

Acknowledgements

The authors thank Sheldon Axler and Daniel Klain for helpful remarks.

The cafe photograph comes from Wikipedia under a Creative Commons Attribution-Share Alike 3.0 Unported license: commons.wikimedia.org/wiki/File:Cafe_Tortoni_La_mesa_que_Borges_acostumbraba_ocupar.jpg

The AI tools of Gemini, ChatGPT and Claude were used to improve the tikz diagrams and as sounding boards for writing style.

University of Massachusetts, Boston

Email address: [email protected]

Temple University, Philadelphia

Email address: [email protected]

References

  1. [1]G. D. Birkhoff, A Set of Postulates for Plane Geometry, Ann. Math. 33 (1932) 329–345.
  2. [2]M. Beck and S. Robbins, Computing the Continuous Discretely: Integer-Point Enumeration in Polyhedra, Springer (2007).
  3. [3]R. Courant and F. John, Introduction to Calculus and Analysis, I, Interscience Publishers (1965).
  4. [4]L. Ehrenpreis, personal communication.
  5. [5]P. M. Gruber and J. M. Wills, Convexity and Its Applications, Birkhäuser Basel (1983).
  6. [6]G. H. Hardy, A Course of Pure Mathematics, 10th ed., Cambridge University Press, Cambridge, 1993.
  7. [7]R. Hartshorne, Geometry: Euclid and Beyond, Springer (2000).
  8. [8]T. L. Heath, The Works of Archimedes, Cambridge University Press (1897).
  9. [9]O. Hijab, Introduction to Calculus and Classical Analysis, Springer (2016).
  10. [10]S. Lang, Basic Mathematics, Springer (1988).
  11. [11]Edmund Landau, Differential and Integral Calculus, AMS Chelsea Publishing, Providence, RI, 1956. Translated by Melvin Hausner and Martin Davis.
  12. [12]John M. Lee, Axiomatic geometry, Pure and Applied Undergraduate Texts, vol. 21, American Mathematical Society, Providence, RI, 2013.
  13. [13]E. Maor, Trigonometric Delights, Princeton University Press, 1998.
  14. [14]E. Maor and R. W. Barnard, Trigonometry, Encyclopaedia Britannica (2023), www.britannica.com/science/trigonometry.
  15. [15]T. J. Osler, A simple geometric method of estimating the error in using Vieta’s product for π, Int. J. Math. Educ. Sci. Technol. 38 (2007), no. 1, 136–142, DOI 10.1080/00207390601002799.DOI
  16. [16]F. Richman, A Circular Argument, The College Math. Journal 24:2, 160–162 (1993).DOI
  17. [17]Walter Rudin, Real and Complex Analysis, third ed., McGraw-Hill Book Co., New York, 1987.
  18. [18]K. V. Sarma and Kripa Shankar Shukla, Aryabhatiya of Aryabhata, India NSA (1976).
  19. [19]O. Toeplitz, The Calculus, A Genetic Approach, University of Chicago Press (1963).
  20. [20]G. J. Toomer, The Chord Table of Hipparchus. Centaurus 18 (1973) 6–28.
  21. [21]G. J. Toomer, Ptolemy’s Almagest, Princeton University Press (1998).

Paper details

vlnc:W_y7erbmcu
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Gemini (multiple versions), ChatGPT, Claude

View statement
Looking for related papers…
Contents