Skip to content

Chapter 5 — Trigonometric Functions of Real Numbers

5.1 - Construction of π

Before being able to “wrap” the real line onto a circle, it is necessary to know, precisely, what length that circle has. This book already constructed ee without taking anything for granted; π\pi will be no exception.


§1. The problem: what length does a circle have?

Section titled “§1. The problem: what length does a circle have?”

The following sections of this chapter will define the trigonometric functions by means of a “wrapping” function that associates, to each real number tt, a point of the unit circle x2+y2=1x^2+y^2=1 reached by traveling a distance tt along that circle, starting at (1,0)(1,0). For that definition to make sense —to know, for example, when the trip completes a turn and begins to repeat— a prior datum is needed: the total length of the unit circle.

That datum is, by definition, 2π2\pi. But “length of a curve” is not, yet, a concept that this book has defined with rigor —the general theory of arc length belongs to integral calculus—, and appealing to it here would be exactly the kind of concession that the book avoided with the derivative in Chapter 4. The way out, for the particular case of the circle, is the one found by Archimedes in the third century B.C.: to approximate the circle by regular inscribed and circumscribed polygons, whose lengths —finite sums of straight segments— are completely defined by the Pythagorean theorem, without any notion of curve length. It is, moreover, a construction that cannot rely on sine or cosine —that would be circular, since sine and cosine are not yet defined, and will be defined in §5.3 using π\pi—; everything that follows uses, exclusively, coordinates, distances and the Pythagorean theorem.

§2. Inscribed polygons: an increasing sequence of perimeters

Section titled “§2. Inscribed polygons: an increasing sequence of perimeters”

To avoid any implicit appeal to an angle or to a fraction of a turn —notions that depend on π\pi—, the starting polygon is constructed directly by coordinates.

The initial square. The four points (1,0)(1,0), (0,1)(0,1), (−1,0)(-1,0), (0,−1)(0,-1) satisfy x2+y2=1x^2+y^2=1, hence they are on the unit circle; by symmetry and the distance formula of §1.8 §2, they form a square. Its side is

s4=(1−0)2+(0−1)2=2s_4=\sqrt{(1-0)^2+(0-1)^2}=\sqrt2

and its perimeter, p4=42p_4=4\sqrt2.

The doubling lemma. Let a regular polygon of nn sides be inscribed in the unit circle, with side ss. Let A,BA,B be two consecutive vertices, MM the midpoint of the segment ABAB, and CC the point of the circle on the ray from the center OO passing through MM, on the side opposite to OO with respect to MM —the “midpoint of the arc” ABAB—. Since OMOM is the perpendicular bisector of ABAB —the triangle OABOAB is isosceles, with OA=OB=1OA=OB=1—, the point CC is, by construction, equidistant from AA and from BB; it will be the new vertex when doubling the number of sides.

Lemma. With a:=OM=1−(s/2)2a:=OM=\sqrt{1-(s/2)^2} —Pythagoras in the right triangle OMAOMA, with hypotenuse OA=1OA=1 and leg AM=s/2AM=s/2—, the new side s′=ACs'=AC satisfies

s′=2−4−s2s'=\sqrt{2-\sqrt{4-s^2}}

Proof. In the right triangle AMCAMC —right-angled at MM, since CC is on the line OMOM—, with legs AM=s/2AM=s/2 and MC=OC−OM=1−aMC=OC-OM=1-a, the Pythagorean theorem gives

s′2=(s2)2+(1−a)2=(s2)2+a2−2a+1s'^2=\left(\frac s2\right)^2+(1-a)^2=\left(\frac s2\right)^2+a^2-2a+1

Since (s2)2+a2=1\left(\dfrac s2\right)^2+a^2=1 —by the very definition of aa—, it follows that s′2=2−2a=2−21−(s/2)2=2−4−s2s'^2=2-2a=2-2\sqrt{1-(s/2)^2}=2-\sqrt{4-s^2}. ■\blacksquare

Proposition (strict growth). If pnp_n is the perimeter of the regular inscribed polygon of nn sides and p2np_{2n} that of the polygon of 2n2n sides obtained by doubling each side according to the Lemma, then p2n>pnp_{2n}>p_n.

Proof. For each side ABAB of the original polygon, the point CC of the Lemma does not belong to the segment ABAB —it is on the circle, strictly farther from the chord than any point of ABAB, except in the degenerate case s=0s=0—, so that AA, BB, CC form a genuine triangle. The strict triangle inequality gives AC+CB>ABAC+CB>AB, that is, the contribution of this side to the new perimeter exceeds its contribution to the original perimeter. Summing over the nn sides, p2n>pnp_{2n}>p_n. ■\blacksquare


With (pn)(p_n) increasing, a sequence is needed that bounds it from above —the role that (bn)(b_n) played for (an)(a_n) in §4.2—. The circumscribed polygons play exactly that role. It is the content of §3.

§3. Circumscribed polygons: the dual decreasing sequence

Section titled “§3. Circumscribed polygons: the dual decreasing sequence”

The circumscribed square. By the definition of a tangent line to a circle —§1.8—, the tangent at (1,0)(1,0) is perpendicular to the radius at that point, that is, the vertical line x=1x=1; similarly, the tangents at (0,1)(0,1), (−1,0)(-1,0), (0,−1)(0,-1) are y=1y=1, x=−1x=-1, y=−1y=-1. These four lines form a square of side 22 —from x=−1x=-1 to x=1x=1—, with perimeter P4=8P_4=8.

Lemma (equal tangent segments). If VV is a point exterior to the unit circle, and M1,M2M_1,M_2 are the points of tangency of the two tangent lines from VV, then VM1=VM2VM_1=VM_2.

Proof. The triangles OM1VOM_1V and OM2VOM_2V are right-angled at M1M_1 and M2M_2 respectively —tangent perpendicular to the radius—, with OM1=OM2=1OM_1=OM_2=1 and common hypotenuse OVOV. By Pythagoras, VM12=OV2−1=VM22VM_1^2=OV^2-1=VM_2^2, and since both are positive, VM1=VM2VM_1=VM_2. ■\blacksquare

Proposition (strict decrease). If PnP_n is the perimeter of the circumscribed polygon of nn sides —tangent at the same nn points as the vertices of the inscribed polygon of nn sides— and P2nP_{2n} the one obtained by doubling, tangent also at the new points CC of §2, then P2n<PnP_{2n}<P_n.

Proof. Consider a vertex VV of the circumscribed polygon of nn sides, where the tangents at two consecutive points M1,M2M_1,M_2 meet. The new point of tangency CC —midpoint of the arc M1M2M_1M_2, according to §2— determines a new tangent line, which cuts the segments VM1VM_1 and VM2VM_2 at points AA and BB respectively —by convexity, the new tangent separates VV from the arc—. The new perimeter replaces, locally, the path M1→V→M2M_1\to V\to M_2 —of length M1A+AV+VB+BM2M_1A+AV+VB+BM_2— by the path M1→A→B→M2M_1\to A\to B\to M_2 —of length M1A+AB+BM2M_1A+AB+BM_2, since AA, CC, BB are collinear—. By the strict triangle inequality in the triangle AVBAVB —non-degenerate, since VV does not belong to the new tangent—, AB<AV+VBAB<AV+VB. Hence the new path is strictly shorter than the previous one, for each vertex VV of the original polygon. Summing, P2n<PnP_{2n}<P_n. ■\blacksquare

Computation formula. With sns_n and an=1−(sn/2)2a_n=\sqrt{1-(s_n/2)^2} as in §2, the same triangle OMAOMA —now compared, by similarity of shared angle at OO, with the triangle OM′A′OM'A' formed by the radius to a point of tangency and the corresponding tangent— gives the ratio an/1=(sn/2)/(Tn/2)a_n/1=(s_n/2)/(T_n/2), where TnT_n is the side of the circumscribed polygon of nn sides; whence Tn=sn/anT_n=s_n/a_n, and

Pn=pnanP_n=\frac{p_n}{a_n}

With (pn)(p_n) increasing and (Pn)(P_n) decreasing, exactly what §4.2 §3 and §4 already did for ee is missing: the crossed bound, and the definition by means of supremum and infimum. It is the content of §4.

§4. The crossed bound and the definition of π

Section titled “§4. The crossed bound and the definition of π”

Proposition. For all n,mn,m of the form 4⋅2k4\cdot2^k, pn<Pmp_n<P_m.

Proof. Same index: since the circumscribed polygon strictly contains the inscribed one —each side of the inscribed one is a chord, and the corresponding circumscribed side is a tangent, farther from the center—, Pn>pnP_n>p_n; formally, Pn=pn/anP_n=p_n/a_n with 0<an<10<a_n<1 —since an=1−(sn/2)2<1a_n=\sqrt{1-(s_n/2)^2}<1 for sn>0s_n>0—, hence Pn=pn/an>pnP_n=p_n/a_n>p_n. Different indices: with k=max⁡(n,m)k=\max(n,m), (pj)(p_j) increasing gives pn≤pkp_n\le p_k, the equal-index case gives pk<Pkp_k<P_k, and (Pj)(P_j) decreasing gives Pk≤PmP_k\le P_m. Chaining, pn≤pk<Pk≤Pmp_n\le p_k<P_k\le P_m. ■\blacksquare

By the supremum axiom, there exists L:=sup⁡{pn}L:=\sup\{p_n\}; by the Lemma of existence of the infimum —section 4.1, §7—, there exists ℓ:=inf⁡{Pn}\ell:=\inf\{P_n\}, and L≤ℓL\le\ell by the previous Proposition. The equality is obtained by bounding Pn−pnP_n-p_n: since Pn=pn/anP_n=p_n/a_n,

Pn−pn=pn(1an−1)=pn⋅1−ananP_n-p_n=p_n\left(\frac1{a_n}-1\right)=p_n\cdot\frac{1-a_n}{a_n}

and since an→1a_n\to1 —because sn→0s_n\to0, according to the doubling Proposition of §2 applied repeatedly— and pnp_n is bounded —by P4=8P_4=8—, this difference can be forced below any ε>0\varepsilon>0, with the same type of Archimedean argument that closed the Proposition of §4.2 §4. We conclude L=ℓL=\ell.

Definition. π:=L2=12sup⁡{pn}=12inf⁡{Pn}\pi:=\dfrac{L}{2}=\dfrac12\sup\{p_n\}=\dfrac12\inf\{P_n\}.


With π\pi defined without ambiguity, a couple of iterations of the method already suffice to bound it between two concrete numbers. It is the content of §5.

With s4=2s_4=\sqrt2, the doubling Lemma of §2 gives s8=2−4−2=2−2s_8=\sqrt{2-\sqrt{4-2}}=\sqrt{2-\sqrt2}.

Bound of 2\sqrt2. 1.4142=1.999396<21.414^2=1.999396<2 and 1.4152=2.002225>21.415^2=2.002225>2, whence 1.414<2<1.4151.414<\sqrt2<1.415.

Bound of s8s_8. From the above, 0.585<2−2<0.5860.585<2-\sqrt2<0.586. Since 0.7642=0.583696<0.5850.764^2=0.583696<0.585 and 0.7662=0.586756>0.5860.766^2=0.586756>0.586, it follows that 0.764<s8<0.7660.764<s_8<0.766.

Bound of p8/2p_8/2. p8=8s8p_8=8s_8, so that 3.056<p8/2<3.0643.056<p_8/2<3.064. In particular, π>3.05\pi>3.05.

Bound of a8a_8 and of P8/2P_8/2. With s8<0.766s_8<0.766, (s8/2)2<0.147(s_8/2)^2<0.147, whence a8=1−(s8/2)2>0.853a_8=\sqrt{1-(s_8/2)^2}>\sqrt{0.853}; since 0.9232=0.851929<0.8530.923^2=0.851929<0.853, it follows that a8>0.923a_8>0.923. Hence P8/2=(p8/2)/a8<3.064/0.923<3.33P_8/2=(p_8/2)/a_8<3.064/0.923<3.33.

Conclusion. 3.05<π<3.333.05<\pi<3.33.


The complete method of this section —inscribed and circumscribed perimeters, doubling of sides, the crossed bound— appears, practically unchanged, in the treatise On the Measurement of the Circle by Archimedes of Syracuse, written around the year 250 B.C. It is, of the three historical precedents that this chapter kept citing —Napier in 1614, Newton in 1687, Euler in 1748—, the oldest by almost eighteen centuries, and the only one that did not need to wait for any later development of analysis in order to be, in its own terms, fully rigorous: Archimedes did not have the completeness axiom under that name, but his argument that the two sequences “squeeze together” leaving no margin is, in essence, exactly the one formalized in §4.

It is worth, before continuing with the unit circle proper, situating the importance of this method beyond having produced a number. Its structure —improving an approximation by doubling the computational effort at each step, with an error bound that is reduced in a controlled and demonstrable way— is the same principle that sustains, two thousand years later, a good part of computational numerical analysis: every iterative algorithm that refines a solution step by step, from the bisection method to the numerical methods that solve differential equations without a closed formula, shares with Archimedes the same underlying logic, although none of them needs to double polygons any longer. And the number itself that this method produced turns out, once the trigonometric functions are defined in the following sections, to be inseparable from every periodic phenomenon that science describes —from a sound wave to a planetary orbit—; but that ubiquity cannot yet be shown here, because it depends on tools —sine, cosine, and their periodic behavior— that this chapter is only beginning to construct.


With π\pi constructed and bounded, the chapter can, at last, define the unit circle and the wrapping function that translates real numbers into points of that circle. That is the content of §5.2.