Sources and References

Primary references — H.S.M. Coxeter, Introduction to Geometry (2nd ed., Wiley, 1969); Coxeter, Regular Polytopes (3rd ed., Dover, 1973) Supplementary texts — Robin Hartshorne, Geometry: Euclid and Beyond (Springer, 2000); David Hilbert, Foundations of Geometry (trans. Townsend); Thomas Heath (trans.), The Thirteen Books of Euclid’s Elements Online resources — MIT OCW 18.783, Stanford lecture notes on discrete geometry, Keith Ball’s survey on sphere packing (arXiv), Henry Cohn’s notes on sphere packing and the kissing problem

Chapter 1: Axioms

1.1 Euclid’s Postulates

Geometry, as a deductive science, begins with Euclid’s Elements (c. 300 BCE), one of the most influential mathematical texts ever written. The Elements comprise thirteen books covering plane geometry, number theory, and solid geometry, and were the dominant mathematical textbook in the Western world for over two thousand years. Euclid sought to derive all geometric truths from a minimal collection of self-evident starting points — a feat of intellectual architecture that set the template for axiomatic reasoning in all of mathematics. He identified five postulates and five “common notions” (general logical principles). The five postulates, paraphrased in modern language, are:

1. Any two distinct points can be joined by a straight line segment.
2. Any straight line segment can be extended indefinitely in a straight line.
3. Given any point and any distance, a circle can be drawn with that point as center and that distance as radius.
4. All right angles are equal to one another.
5. (The Parallel Postulate) If a straight line crossing two other straight lines makes the interior angles on one side sum to less than two right angles, then the two lines, if extended, meet on that side.

The first four postulates are short, local, and intuitively obvious. The fifth is conspicuously different: it is longer, it makes a claim about what happens when lines are extended arbitrarily far (“at infinity”), and it feels less like an axiom and more like a theorem waiting to be proved. This discomfort was noticed by Euclid himself — he delays using the fifth postulate as long as possible in the Elements, proving everything he can from the first four. This rhetorical choice is striking: Euclid seems to have sensed that the fifth postulate is on a different footing from the others.

The Parallel Postulate — visualized:

         α         β
   ------+----+------
          \  /
           \/
           /\
          /  \
    -----+----+------

  If α + β < 180°, the lines meet on that side.
  (If α + β = 180°, the lines are parallel — never meeting.)
  (If α + β > 180°, the lines meet on the OTHER side.)

The fifth postulate has a conspicuously different character from the other four. Its complexity led mathematicians for two millennia to attempt proving it from the first four alone — the great logical project of “proving the parallel postulate.” The eventual realization that this is impossible — and that coherent geometries exist in which the fifth postulate fails — gave birth to non-Euclidean geometry in the 19th century. The story of this discovery is one of the great intellectual dramas in the history of mathematics.

Despite its foundational importance, Euclid’s system has gaps by modern standards. He silently invokes facts about the continuity of lines and circles (for instance, that two circles with overlapping centers must intersect), and his notion of “betweenness” is never formally axiomatized. The constructions in Book I assume that intersection points always exist, a claim that cannot be deduced from his five postulates alone without additional continuity axioms. These gaps were not flaws of carelessness — Euclid was writing for an audience that took geometric intuition for granted — but they became important when, two millennia later, mathematicians tried to build the subject on purely logical foundations.

1.1.1 Two Thousand Years of Failed Proofs

The parallel postulate generated an extraordinary body of mathematics precisely because so many brilliant mathematicians were convinced it must be provable from the other four. The general strategy was to assume the negation of the fifth postulate and derive a contradiction. The problem was that the derivations often produced results that seemed bizarre — but not technically contradictory.

Among the notable attempts:

Proclus (c. 450 CE) argued, in his commentary on the Elements, that two lines at unequal distances from a third must eventually meet. His argument was circular: it assumed a version of the parallel postulate in order to prove it.

Ibn al-Haytham (c. 1000 CE) attempted a proof by considering a line segment that moves perpendicular to a fixed line. He claimed the endpoint traces a straight line — which is exactly the parallel postulate in disguise.

Saccheri (1733) wrote Euclides ab Omni Naevo Vindicatus (“Euclid Freed of Every Flaw”), making the most systematic attempt yet. He considered three cases for the angle sum of a quadrilateral: the “hypothesis of the right angle” (Euclidean geometry), the “hypothesis of the obtuse angle,” and the “hypothesis of the acute angle.” He correctly eliminated the obtuse angle hypothesis (it is inconsistent with Euclid’s other axioms). He spent most of the book trying to eliminate the acute angle hypothesis — and produced dozens of theorems that are in fact correct in hyperbolic geometry. He was so close to discovering non-Euclidean geometry that the great Coxeter later wrote that “Saccheri unconsciously discovered non-Euclidean geometry.”

Lambert (1766) pushed further, computing that in the “acute angle” geometry, the angle defect of a triangle is proportional to its area — a beautiful formula that is actually correct in hyperbolic geometry.

Legendre (1794–1823) made multiple attempts across decades, each time discovering a subtle error. His Éléments de Géométrie went through twelve editions with successive, flawed “proofs.”

The resolution came in the 1820s. Gauss had privately concluded by around 1817 that the parallel postulate was independent, but he published nothing, reportedly fearing ridicule. Bolyai (Hungary, 1832) and Lobachevsky (Russia, 1830) independently published explicit constructions of consistent non-Euclidean geometries — what we now call hyperbolic geometry. In this geometry, through any point not on a given line, there are infinitely many lines parallel to the given line. Gauss wrote to Bolyai’s father that he was “astonished and delighted” but had known this for years — a letter that (understandably) infuriated the younger Bolyai.

The rigorous confirmation that non-Euclidean geometry is consistent — that is, that it cannot be logically contradicted from within — came when Beltrami (1868) and Klein (1871) constructed explicit models of hyperbolic geometry within Euclidean geometry. If Euclidean geometry is consistent, so is hyperbolic geometry; thus the fifth postulate is genuinely independent of the first four.

1.1.2 A First-Principles Example: Proving the Base Angles Theorem

To appreciate the character of Euclidean reasoning, here is an example of a proof from Euclid’s first principles (Book I, Proposition 5: “In isosceles triangles the angles at the base equal one another”). This is traditionally called the pons asinorum (“bridge of asses”), supposedly because students who could not cross this bridge of difficulty could make no further progress.

Euclid’s original proof was considerably more elaborate, extending the equal sides and using auxiliary triangles. The modern proof via SAS is more efficient but relies on Hilbert’s congruence axioms, which Euclid did not state explicitly. This illustrates the trade-off in axiom systems: more powerful axioms enable shorter proofs but require more justification at the foundational level.

1.2 Hilbert’s Axiom System

David Hilbert, in his 1899 Grundlagen der Geometrie (Foundations of Geometry), provided a rigorous reformulation of Euclidean geometry. Hilbert was responding to the realization that Euclid’s system, while mathematically productive for two millennia, was not logically complete by the standards of 19th-century rigor. The discovery of non-Euclidean geometry had sharpened everyone’s sense that geometric axioms needed to be stated with complete precision. Hilbert’s goal was to exhibit a set of axioms from which all of Euclidean geometry — and only Euclidean geometry — could be derived by pure logical deduction, with no appeal to intuition or diagrams.

Rather than appealing to intuition or physical diagrams, Hilbert’s system builds the theory from undefined primitive notions — points, lines, and planes — along with three undefined relations: incidence (a point lies on a line), betweenness (a point lies between two others), and congruence (of segments and angles). Crucially, the “meanings” of these terms are entirely determined by the axioms — Hilbert famously remarked that one could replace “points, lines, and planes” with “tables, chairs, and beer mugs” and the proofs would still be valid. This radical separation of form from content is the hallmark of modern axiomatic mathematics.

From these undefined terms, Hilbert derived all of classical Euclidean geometry. His axioms fall into five groups:

Axioms of Incidence (I1–I8). Two distinct points determine a unique line. Three non-collinear points determine a unique plane. A line lying in a plane meets any other line in that plane at most once. And so on.

Axioms of Order (B1–B4). These formalize betweenness. If \(B\) lies between \(A\) and \(C\), then \(A\), \(B\), \(C\) are distinct collinear points and \(B\) lies between \(C\) and \(A\) as well. Given two points on a line, there exists a third point between them (density), and also a point such that one of the original points lies between the new point and the other (no endpoints). The crucial axiom here is Pasch’s axiom: a line entering a triangle through one side must exit through one of the other two sides. This captures the intuitive fact that lines do not “jump over” triangles — something Euclid implicitly assumed but never stated.

Axioms of Congruence (C1–C6). These axioms allow us to “copy” segments and angles. Congruence of segments is an equivalence relation compatible with concatenation. Similarly for angles. The most important is C6: SAS congruence — if two sides and the included angle of one triangle are congruent to the corresponding parts of another triangle, the triangles are congruent. In classical treatments, SAS is often stated as a theorem proved by “superposition” (physically picking up one triangle and placing it on the other), but Hilbert recognized that superposition is not a legitimate logical operation and elevated SAS to the status of an axiom.

The Axiom of Parallelism (P). Through a point not on a given line, there exists exactly one line parallel to the given line. This is Hilbert’s version of Euclid’s fifth postulate, sometimes called Playfair’s axiom (after the Scottish mathematician John Playfair, 1748–1819, who popularized this equivalent formulation). Dropping this axiom entirely yields a system compatible with both Euclidean and hyperbolic geometry (called absolute geometry or neutral geometry). Replacing it with “there are no parallel lines” (every pair of lines meets) yields elliptic geometry (modeled on the sphere with antipodal points identified).

Axioms of Continuity (Archimedes and Completeness). The Archimedean axiom states that given any two segments, some finite multiple of the smaller exceeds the larger — ruling out infinitely small or infinitely large segments. The completeness axiom (or Dedekind’s axiom in some formulations) ensures that the geometric line is isomorphic to \(\mathbb{R}\) and that circles and lines intersect when expected. Without these axioms, one could construct “non-Archimedean” planes over hyperreal number fields — geometrically valid but containing infinitely small segments.

This means that Euclidean geometry, properly axiomatized, has essentially one model. The parallel postulate, however, can be replaced by alternatives that yield hyperbolic or elliptic geometry — each internally consistent. Hilbert’s work is the template for all modern axiomatic theories: give a list of axioms, prove the list is consistent (by exhibiting a model), and then derive theorems by pure logical deduction.

              C
             / \
            /   \
           /     \
          /   ℓ   \
         /  ----   \
        A-----D-----B
              ↑
         ℓ enters here; it must exit through AC or BC

1.3 Straightedge and Compass Constructions

The ancient Greek geometers were interested not merely in existence proofs but in explicit constructions using idealized tools. The two allowed tools are: a straightedge (an unmarked ruler, capable only of drawing the unique line through two given points — notably not capable of measuring distances) and a compass (capable of drawing a circle given a center and a point on the circumference, or equivalently given a center and a radius equal to a given segment). The restriction to these two tools — and only these — was a philosophical as well as a practical choice: the Greeks prized constructions that required no measurement, only the primitive operations of joining points and finding intersections.

A point is constructible if it can be obtained from a given set of points by finitely many straightedge-and-compass operations. These operations are:

1. Draw the line through two constructed points and take its intersection with an already-constructed line.
2. Draw a circle with center at a constructed point and radius equal to the distance between two constructed points, and take its intersection with another constructed circle or line.

Starting from just two points (conventionally taken at distance 1 apart, establishing the unit length), one generates a progressively richer set of points. The algebraic content of these operations is captured by the following classical result due to Pierre Wantzel, a French mathematician who, at age 23, resolved two of antiquity’s most famous open problems in a single paper.

The key insight is that each straightedge-or-compass operation introduces at most one new “square root.” Intersecting two lines gives a rational combination of the coordinates of the known points. Intersecting a line with a circle, or two circles with each other, gives at most a quadratic extension of the current field. Therefore, after \(n\) operations, the coordinates of any constructed point lie in a field of degree at most \(2^n\) over \(\mathbb{Q}\). Conversely, any number in a tower of quadratic extensions can be constructed step by step.

This theorem immediately resolves three classical problems that puzzled antiquity for two thousand years:

• Doubling the cube (Delian problem): Given a cube with side length 1 and volume 1, construct a cube with volume 2. This requires a side of length \(2^{1/3}\). The minimal polynomial of \(2^{1/3}\) over \(\mathbb{Q}\) is \(x^3 - 2\), which is irreducible of degree 3. Since 3 is not a power of 2, \(2^{1/3}\) does not lie in any tower of quadratic extensions of \(\mathbb{Q}\). Hence doubling the cube is impossible with straightedge and compass alone.

• Trisecting an arbitrary angle: To trisect 60°, we need \(\cos 20°\). The triple angle formula gives \(\cos 60° = 4\cos^3 20° - 3\cos 20° = 1/2\), so \(\cos 20°\) satisfies \(8x^3 - 6x - 1 = 0\). This polynomial is irreducible over \(\mathbb{Q}\) (it has no rational roots), so \(\cos 20°\) lies in a cubic extension of \(\mathbb{Q}\), not a tower of quadratic extensions. Hence trisecting a general angle is impossible.

• Squaring the circle: Construct a square with the same area as a given unit circle, i.e., construct \(\sqrt{\pi}\). Since \(\pi\) is transcendental (Lindemann, 1882), \(\sqrt{\pi}\) cannot lie in any algebraic extension of \(\mathbb{Q}\), let alone a tower of quadratic extensions. Hence squaring the circle is impossible.

1.3.1 A Worked Construction: Bisecting an Angle

On the positive side, many constructions are possible. Here is the classic construction for bisecting an angle \(\angle BAC\), which was known to Euclid (Book I, Proposition 9).

Given: An angle \(\angle BAC\) with vertex at \(A\).

Construction:

1. With center \(A\), draw a circle of arbitrary radius. Let it meet ray \(AB\) at \(D\) and ray \(AC\) at \(E\).
2. With center \(D\) and radius \(DE\), draw a circle. With center \(E\) and radius \(DE\), draw another circle. Let these two circles meet at point \(F\) (on the interior side of the angle).
3. Draw ray \(AF\). This is the angle bisector.

          B
         /
        /   D
       /   .
      A-----------E--------C
       \       .
        \   F (intersection of the two circles)
         \
          (bisector AF)

Why it works: \(AD = AE\) (both radii of the first circle). \(DF = EF\) (both radii of the equal circles of step 2). \(AF = AF\) (common side). By SSS congruence, \(\triangle ADF \cong \triangle AEF\). Therefore \(\angle DAF = \angle EAF\), so \(AF\) bisects \(\angle DAE = \angle BAC\). \(\square\)

On the positive side, Gauss proved that the regular \(n\)-gon is constructible if and only if \(n = 2^a p_1 p_2 \cdots p_k\) where the \(p_i\) are distinct Fermat primes — primes of the form \(2^{2^m} + 1\). The known Fermat primes are 3, 5, 17, 257, and 65537. Gauss himself constructed the regular 17-gon at age 18, and this achievement was reportedly what convinced him to pursue mathematics rather than philology. Whether there are infinitely many Fermat primes is an open problem.

1.4 Origami Axioms: The Huzita-Hatori System

Paper folding — origami — gives a different and strictly more powerful construction system than straightedge and compass. The ancient Chinese and Japanese paper-folding traditions were primarily artistic, but their mathematical power was not recognized until the 20th century. The Huzita-Hatori axioms (named after Humiaki Huzita, who identified six axioms in 1991, and Koshiro Hatori, who added the seventh in 2003) describe the seven fundamental fold operations that are physically realizable with a flat sheet of paper:

H1. Given two points, there is a fold passing through both. H2. Given two points, there is a fold that places one point onto the other. H3. Given two lines, there is a fold that places one line onto the other. H4. Given a point and a line, there is a fold perpendicular to the line passing through the point. H5. Given two points \(P_1, P_2\) and a line \(\ell\), there is a fold that places \(P_1\) onto \(\ell\) and passes through \(P_2\). H6. Given two points \(P_1, P_2\) and two lines \(\ell_1, \ell_2\), there is a fold that simultaneously places \(P_1\) onto \(\ell_1\) and \(P_2\) onto \(\ell_2\). H7. Given a point \(P\) and two lines \(\ell_1, \ell_2\), there is a fold that places \(P\) onto \(\ell_1\) and is perpendicular to \(\ell_2\).

Huzita-Hatori fold operations (schematic):

H1: fold through two points   H2: fold mapping P1 → P2
   P1 ●                          P1 ●              ● P2
      |  fold line                    \            /
   P2 ●                               \  fold    /
                                       \--------/
                                        crease = perpendicular bisector

H4: perpendicular to ℓ         H6: "the cubic fold"
   ℓ: ————————————               P1 ●      ● P2
      |                             |          |
      ↓ fold drops P ⊥ ℓ           ℓ1         ℓ2
      P ●                       places P1→ℓ1 and P2→ℓ2 simultaneously

Algebraically, H1–H5 reproduce everything constructible by straightedge and compass (quadratic extensions). The sixth axiom, H6, is especially powerful. It corresponds geometrically to finding a common tangent to two parabolas — an operation that can solve general cubic and quartic equations. Algebraically, the origami-constructible numbers include all numbers obtainable by a tower of field extensions of degree 2 or 3 over \(\mathbb{Q}\). This means that the cube can be doubled and a general angle can be trisected by origami, since the relevant cubics are solvable in this setting. For example, to trisect angle \(\theta\), one can reduce it to finding a root of \(4x^3 - 3x - \cos\theta = 0\) (from the triple angle formula), and H6 performs exactly the cubic-resolving fold.

Chapter 2: Conic Sections

2.1 Classical Definitions

The conic sections — ellipse, parabola, and hyperbola — arise as the intersection of a plane with a right circular cone. This is their oldest definition, dating to Apollonius of Perga (c. 262–190 BCE), one of the greatest Greek mathematicians. Apollonius’s eight-volume treatise Conics established most of the classical theory and introduced the terminology “ellipse,” “parabola,” and “hyperbola” (from Greek words meaning “deficiency,” “equality,” and “excess” — referring to the relationship between a certain square and a rectangle in the focal construction). The Conics were so comprehensive that the subject lay essentially dormant for fifteen centuries.

Depending on the angle that the cutting plane makes with the cone’s axis, the intersection is:

• An ellipse if the plane meets the cone in a closed curve (the plane cuts all generators on one nappe, at an angle steeper than the generators themselves).
• A parabola if the plane is parallel to exactly one generator of the cone.
• A hyperbola if the plane cuts both nappes of the cone.

Degenerate cases (point, line, pair of lines) occur when the plane passes through the apex. The three non-degenerate cases are distinguished by a single continuous parameter — the angle of the cutting plane — which explains their deep unity. From the projective point of view (Section 2.4), they are literally the same curve viewed from different perspectives.

Conic sections from a cone:

         *         ← apex
        /|\
       / | \
      /  |  \    ← plane cuts only upper nappe
     /   |   \      at angle steeper than generators
    *----+----*   → ELLIPSE (closed curve)
   /     |     \
  /      |      \   ← plane parallel to one generator
 *-------+-------*  → PARABOLA (unbounded, one branch)
/        |        \
*--------+--------* ← plane cuts both nappes
          |          → HYPERBOLA (two branches)

          cone axis

2.2 Focus-Directrix Definition

The cone-section definition is beautiful but not the most useful for computation. A unified and analytically powerful description of all three non-degenerate conics uses the focus-directrix property. Fix a point \(F\) (the focus) and a line \(\ell\) (the directrix) not passing through \(F\), and fix a positive constant \(e\) called the eccentricity. The conic is then the locus of points \(P\) such that

where \(d(P, \ell)\) denotes the perpendicular distance from \(P\) to \(\ell\). This definition makes it transparent why the three cases are distinguished by a single number: the eccentricity \(e\) controls whether the focus is closer to, equal to, or farther from the directrix than from the curve.

Setting up coordinates with \(F\) at the origin and \(\ell\) the vertical line \(x = -d\), the focus-directrix condition yields the polar equation of a conic with focus at the origin:

This single formula encompasses all three types by varying \(e\). It is the natural form for celestial mechanics: Kepler’s first law states that planetary orbits are ellipses with the Sun at one focus, and the polar equation shows directly how the shape (eccentricity) controls the orbit’s geometry. Earth’s orbit has \(e \approx 0.0167\) (nearly circular); Halley’s comet has \(e \approx 0.967\) (highly elliptical); a comet on a parabolic orbit (\(e = 1\)) escapes the solar system after one pass.

2.2.1 Deriving the Standard Ellipse Equation

The derivation of the standard Cartesian form of the ellipse from the focus-directrix definition illustrates the power of coordinates. Let the ellipse have eccentricity \(e < 1\), focus \(F = (c, 0)\) (with \(c > 0\)), and the corresponding directrix \(\ell: x = a/e\) (where \(a > 0\) is the semi-major axis, to be determined). The focus-directrix condition says: for a point \(P = (x, y)\) on the ellipse,

Squaring both sides (both sides are positive since \(x < a/e\)):

Expanding the left side: \(x^2 - 2cx + c^2 + y^2 = a^2 - 2aex + e^2 x^2\). Rearranging:

Setting \(c = ae\) (so that the relation \(e = c/a\) holds by definition), the middle term vanishes:

Dividing by \(a^2(1-e^2)\) and setting \(b^2 = a^2(1-e^2) = a^2 - c^2\):

This is the standard form of the ellipse with semi-major axis \(a\) along the \(x\)-axis and semi-minor axis \(b\) along the \(y\)-axis. The ellipse has two foci at \((\pm c, 0)\) and two directrices at \(x = \pm a/e\).

         (0,4)
          *
         / \
        /   \
(-5,0)*---F1---+---F2---*(5,0)
    (-3,0)     (3,0)
        \   /
         \ /
          *
         (0,-4)

  a = 5, b = 4, c = 3, e = 3/5
  Sum of distances from any point to F1 and F2 = 2a = 10

ELLIPSE: x²/25 + y²/16 = 1          PARABOLA: y² = 8x
     (semi-major a=5, semi-minor b=4)       (focus at (2,0))

        * * *                              |
      *       *                            |     *
     *  F1  F2  *                          |   *
     *         *                    -------+--*------
     *  F1  F2  *                          |   *
      *       *                            |     *
        * * *                              focus (2,0)
   directrix    directrix                  directrix: x=-2

HYPERBOLA: x²/9 - y²/16 = 1         CIRCLE: x²+y²=25
     (a=3, b=4, c=5, e=5/3)                (special ellipse, e=0)

  *\        /*                               * * *
  * \  F1  / *                           *       *
   \ \    / /                           *    F    *
    \ \  / /                             *       *
     F2  F1                               * * *
    / /  \ \
   / /    \ \
  * /  F2  \ *
  */        \*

2.3 Reflective Properties

Each conic has a remarkable reflective property that has practical applications in optics and engineering. These properties are beautiful theorems, but they also have an elegant unified explanation: they all follow from the optical principle that light travels in straight lines and reflects at equal angles, combined with the focal characterizations of the conics.

These reflective properties are not merely aesthetically pleasing — they are foundational to entire technologies. The parabola property is the basis of parabolic reflectors used in satellite dishes, telescope mirrors, and flashlight reflectors: parallel incoming rays (from a distant star or signal) all converge to the focus, which is why parabolic dishes have such a concentration of signal. The ellipse property underlies elliptical billiard tables: a ball shot from one focus will always pass through the other focus after one bounce, regardless of direction. The hyperbola property is used in Cassegrain reflectors (a type of telescope combining a parabolic primary mirror with a hyperbolic secondary mirror).

Proof sketch (parabola). Let \(P = (x_0, y_0)\) be a point on the parabola \(y^2 = 4px\). The focus is at \(F = (p, 0)\) and the directrix is \(x = -p\). The tangent line at \(P\) has slope \(dy/dx = 2p/y_0\) (by implicit differentiation). The line \(PF\) has slope \((y_0 - 0)/(x_0 - p) = y_0/(x_0 - p)\). The angle that \(PF\) makes with the tangent equals the angle that the horizontal (parallel to axis) makes with the tangent — this can be verified using the angle bisector formula and the fact that, for a point on the parabola, the distance to the focus equals the distance to the directrix: \(PF = x_0 + p = d(P, \ell)\). The equality of distances is the essence of the equal-angle law.

2.4 Second-Degree Curves and Projective Unification

In Cartesian coordinates, any non-degenerate conic is the zero set of a quadratic polynomial:

The type is determined by the discriminant \(\Delta = B^2 - 4AC\):

• \(\Delta < 0\): ellipse (or circle if \(A = C\) and \(B = 0\)).
• \(\Delta = 0\): parabola.
• \(\Delta > 0\): hyperbola.

This algebraic classification corresponds precisely to the geometric one. The discriminant measures, in effect, whether the conic meets the “line at infinity” in real or complex points — a fact made precise by projective geometry.

From the perspective of projective geometry, all non-degenerate conics are equivalent under projective transformations. In the projective plane \(\mathbb{P}^2\), an ellipse and a hyperbola differ only in how they meet the line at infinity: the ellipse meets it in two complex conjugate points, the hyperbola in two real points, and the parabola is tangent to the line at infinity at one real point. This unification is one of the great insights of 19th-century geometry, due in part to Poncelet and later to Cayley and Klein. Geometrically, the difference between an ellipse and a hyperbola is “merely” whether two invisible points (at infinity) are real or complex — a distinction that disappears over the complex numbers.

Chapter 3: Polytopes

3.1 Regular Polygons and Schläfli Symbols

A polygon is a closed planar figure bounded by straight line segments. A polygon is regular if all its sides are equal in length and all its interior angles are equal. The regular polygon with \(n\) sides is denoted \(\{n\}\) in the Schläfli symbol notation (introduced by Ludwig Schläfli, Swiss geometer, 1814–1895), and has interior angle

This formula has a simple derivation: the sum of interior angles of any convex \(n\)-gon is \((n-2) \cdot 180°\) (divide the polygon into \(n-2\) triangles from one vertex), so each angle in the regular case is \((n-2) \cdot 180°/n\).

For \(n \geq 3\), regular \(n\)-gons exist for every integer \(n\). The interior angle increases from \(60°\) (equilateral triangle \(\{3\}\)) toward \(180°\) as \(n \to \infty\). The regular hexagon \(\{6\}\) has interior angle exactly \(120°\), which is why hexagons tile the plane: three hexagons meet at each vertex, contributing \(3 \times 120° = 360°\), exactly filling the angle around a point.

3.2 Star Polygons

The Schläfli symbol can be extended to star polygons \(\{n/k\}\), where \(n\) and \(k\) are integers with \(1 < k < n/2\) and \(\gcd(n, k) = 1\). The star polygon \(\{n/k\}\) is formed by connecting every \(k\)-th vertex of a regular \(n\)-gon. These are not convex polygons but self-intersecting ones, and they form the faces of the star polyhedra (Kepler-Poinsot solids) and appear in Islamic geometric art.

If \(\gcd(n, k) = d > 1\), the result is a compound of \(d\) copies of \(\{n/d\}\), not a single connected star polygon. The figure \(\{6/2\}\), for instance, is the compound of two triangles.

Star polygons:

{5/2} pentagram:        {8/3} octagram:         {7/3} heptagram:
       *                    *   *                    *
      / \                  /|\ /|\                  / \
     /   \                / | X | \                /   \
    *     *              *  |/ \|  *              *     *
     \   /                \ |   | /                \   /
      \ /                  \|   |/                  \ /
       *                    *   *                    *
   (every 2nd vertex     (every 3rd vertex        (every 3rd vertex
    of a pentagon)         of an octagon)           of a heptagon)

3.3 Regular Polyhedra: The Platonic Solids

Moving to three dimensions, a polyhedron is a solid bounded by flat polygonal faces. A polyhedron is regular (a Platonic solid) if all its faces are congruent regular polygons, and the same number of faces meet at each vertex. The Schläfli symbol \(\{p, q\}\) denotes the regular polyhedron whose faces are \(\{p\}\) and in which \(q\) faces meet at each vertex.

The condition for a convex regular polyhedron is that the sum of face angles at each vertex must be less than \(360°\), giving \(q \cdot \alpha_p < 360°\), i.e.,

Rearranging: \(\frac{1}{p} + \frac{1}{q} > \frac{1}{2}\). The integer solutions with \(p, q \geq 3\) are exactly five. This is why there are exactly five Platonic solids: the inequality has a finite number of integer solutions, each corresponding to a different solid.

These five solids were known to the ancient Greeks and celebrated by Plato in the Timaeus (c. 360 BCE), where he associated them with the classical elements: tetrahedron with fire, cube with earth, octahedron with air, icosahedron with water, and dodecahedron with the cosmos (or “ether”). This identification was more philosophical than scientific, but it reflects the deep cultural significance these five solids held. Theaetetus (c. 417–369 BCE) is generally credited with the first rigorous proof that there are exactly five.

3.3.1 Worked Example: Vertex Angle Sums and the Five Platonic Solids

Here is the vertex-angle argument in full detail. At each vertex of a Platonic solid \(\{p, q\}\), exactly \(q\) regular \(p\)-gons meet. The face angle of a regular \(p\)-gon at each vertex is \(\alpha_p = (p-2) \cdot 180°/p\). For the solid to be convex, these face angles must sum to less than \(360°\):

With \(p, q \geq 3\), the solutions are:

• \(p=3, q=3\): \(1/3 + 1/3 = 2/3 > 1/2\). ✓ Three equilateral triangles per vertex. Each vertex angle sum: \(3 \times 60° = 180° < 360°\). → Tetrahedron.
• \(p=3, q=4\): \(1/3 + 1/4 = 7/12 > 1/2\). ✓ Four equilateral triangles per vertex. Sum: \(4 \times 60° = 240° < 360°\). → Octahedron.
• \(p=3, q=5\): \(1/3 + 1/5 = 8/15 > 1/2\). ✓ Five equilateral triangles per vertex. Sum: \(5 \times 60° = 300° < 360°\). → Icosahedron.
• \(p=3, q=6\): \(1/3 + 1/6 = 1/2\). ✗ (Not strictly greater.) Six triangles: \(6 \times 60° = 360°\), which gives a flat tessellation (the triangular tiling), not a polyhedron.
• \(p=4, q=3\): \(1/4 + 1/3 = 7/12 > 1/2\). ✓ Three squares per vertex. Sum: \(3 \times 90° = 270° < 360°\). → Cube.
• \(p=4, q=4\): \(1/4 + 1/4 = 1/2\). ✗ Four squares: \(4 \times 90° = 360°\), the square tiling.
• \(p=5, q=3\): \(1/5 + 1/3 = 8/15 > 1/2\). ✓ Three pentagons per vertex. Sum: \(3 \times 108° = 324° < 360°\). → Dodecahedron.
• \(p=5, q=4\): \(1/5 + 1/4 = 9/20 < 1/2\). ✗ (Violates the inequality.)
• \(p=6, q=3\): \(1/6 + 1/3 = 1/2\). ✗ Hexagonal tiling.
• All larger \(p\) or \(q\): give \(1/p + 1/q \leq 1/2\). ✗

So there are exactly five convex regular polyhedra. No more, no fewer.

3.3.2 ASCII Art of the Five Platonic Solids

TETRAHEDRON {3,3}         CUBE {4,3}             OCTAHEDRON {3,4}
  4 faces, 4 verts         6 faces, 8 verts        8 faces, 6 verts
  6 edges                  12 edges                12 edges

        A                   +-------+                   *
       /|\                 /|      /|                  /|\
      / | \               / |     / |                 / | \
     /  |  \             +-------+  |                *  |  *
    B---+---C            |  +- - -| -+              /|  *  |\
     \  |  /             | /     | /               / | / \ | \
      \ | /              |/      |/               *--*-----*--*
       \|/               +-------+                 \ | \ / | /
        D                                           \|  *  |/
                                                     *  |  *
                                                      \ | /
                                                       \|/
                                                        *

DODECAHEDRON {5,3}        ICOSAHEDRON {3,5}
  12 faces, 20 verts        20 faces, 12 verts
  30 edges                  30 edges

    (top view)              (top view)
       * * *                    *
      *     *                  /|\
     *       *                * | *
    * *     * *              /|\ /|\
   *   * * *   *            * | X | *
    *         *              \|/ \|/
     *       *                * | *
      *     *                  \|/
       * * *                    *

  Regular pentagons         Equilateral triangles
  at every face             at every face

3.3.3 Euler Characteristic of Each Platonic Solid

3.4 Star Polyhedra: Kepler-Poinsot Solids

By relaxing the convexity condition and allowing star polygons as faces or allowing faces to penetrate through each other (but maintaining regularity in the combinatorial sense), four additional regular polyhedra appear. These are the Kepler-Poinsot polyhedra, discovered partly by Johannes Kepler (1571–1630), who found two of them in his 1619 Harmonices Mundi, and completed by Louis Poinsot (1809), who rediscovered them and added two more. Cauchy subsequently proved in 1811 that these four are the only regular star polyhedra.

In the Kepler-Poinsot polyhedra, the faces either are star polygons (pentagrammic faces) or meet in a star-polygon pattern around each vertex. They are not convex and have faces that intersect each other. Their Euler characteristic is not always 2.

Kepler-Poinsot Solids (schematic views):

Small Stellated        Great Dodecahedron       Great Stellated
Dodecahedron           {5, 5/2}                 Dodecahedron
{5/2, 5}                                        {5/2, 3}

    *                      *                        *
   /|\                    /|\                      /|\
  * | *                  * | *                    * | *
 /|\ /|\                /|\ /|\                  /|*\ /|
* |X| * |              * | * | *                * /|\ * |
 \|/ \|/                \|/ \|/                  *| X |*
  *   *                  *   *                   *\|/*
   \ /                    \ /                     \|/
    *                      *                       *
(pentagram spikes       (indented,               (star-shaped
 pointing out from       faces are                spikes)
 a dodecahedron)         pentagons)

The small stellated dodecahedron looks like a dodecahedron with a pentagrammic pyramid built on each face — twelve “spiky” faces piercing through each other. Despite their apparent complexity, these four solids satisfy a generalized Euler formula \(V - E + F = \chi\) with \(\chi\) depending on how one counts the genus of the underlying surface.

3.5 Regular Polytopes in Four Dimensions

In four dimensions, a regular 4-polytope (or regular polychoron) has regular polyhedral cells, with the same number of cells meeting at each edge, and the same number of faces at each vertex. Remarkably, there are six convex regular 4-polytopes, compared to five in three dimensions.

The 24-cell \(\{3,4,3\}\) is the most exceptional: it has no analogue in any other dimension, is self-dual, and its vertices can be taken as the 24 unit vectors \(\pm e_i \pm e_j\) in \(\mathbb{R}^4\) (the root system \(D_4\) rescaled). The 120-cell is the four-dimensional analogue of the dodecahedron, and the 600-cell is the analogue of the icosahedron — they are dual to each other.

In dimensions \(n \geq 5\), only three regular convex polytopes exist: the simplex \(\{3^{n-1}\}\), the hypercube \(\{4, 3^{n-2}\}\), and the cross-polytope \(\{3^{n-2}, 4\}\). The richness of dimensions 3 and 4 is a special phenomenon with no known deep explanation — it simply happens that the Diophantine inequality \(1/p + 1/q > 1/2\) (and its four-dimensional analogue) has more solutions in low dimensions.

3.5.1 Intuiting the Tesseract: A 4D Shadow

How does one visualize a four-dimensional polytope? The key insight is that we see 3D objects as 2D projections (shadows), so we should expect 4D objects to appear as 3D projections. Just as a cube’s wireframe projection onto a plane shows a small square inside a large square connected by four diagonal edges (a “picture of a cube”), the tesseract’s projection onto 3D space shows a small cube inside a large cube connected by eight edges.

Tesseract (hypercube) projection onto 2D:

       5---------6
      /|        /|
     / |       / |
    8---------7  |
    |  1------|--2
    | /       | /
    |/        |/
    4---------3

  Outer cube: vertices 1,2,3,4,5,6,7,8
  Inner cube: same vertices after 4D projection
  (In reality, the "inner" cube and "outer" cube
   are the same size in 4D — perspective makes
   one appear smaller)

  The tesseract has:
  - 16 vertices
  - 32 edges
  - 24 square faces
  - 8 cubic cells
  - V - E + F - C = 16 - 32 + 24 - 8 = 0
    (Euler characteristic for 4D sphere = 0)

Chapter 4: Polyhedra

4.1 Euler’s Formula

The most fundamental result about convex polyhedra is Euler’s formula, relating the number of vertices \(V\), edges \(E\), and faces \(F\):

Euler announced this formula in 1758, though Descartes had observed something equivalent a century earlier. It is a genuinely non-obvious fact: why should these three counts, which seem to depend on the shape of the polyhedron in complicated ways, combine to always give the same number? The answer lies in topology: the formula is really a statement about the sphere (any convex polyhedron is topologically a sphere), and it follows from properties of planar graphs.

Euler's formula — the tree argument:

  Polyhedron → remove one face → flatten to plane
  
  Cube flattened:              Spanning tree of cube:
  
  *---*---*---*               *---*   *   *
  |   |   |   |               |            
  *---*---*---*               *   *   *   *
  |   |   |   |                   |        
  *---*---*---*               *   *---*   *
  
  V=8, E=12, F=6              V=8, E=7 (=V-1), F=1
  8-12+6=2 ✓                  8-7+1=2 ✓

  Then add each non-tree edge:
  each addition creates exactly 1 new face.
  E increases by 1, F increases by 1.
  V-E+F stays constant = 2.

The quantity \(V - E + F\) is the Euler characteristic \(\chi\). For a sphere (or any simply connected surface), \(\chi = 2\). For a torus, \(\chi = 0\), and indeed one can embed a graph on a torus such that \(V - E + F = 0\). The Euler characteristic is a topological invariant, unchanged by continuous deformation. This is why Euler’s formula holds for all convex polyhedra — they are all topologically spheres, regardless of their specific combinatorial structure.

Euler’s formula has powerful corollaries. For instance, combining it with the inequalities \(3F \leq 2E\) (each face has at least 3 edges, each edge borders 2 faces) and \(3V \leq 2E\) (each vertex has degree at least 3), one can prove that every convex polyhedron has at least one face with 5 or fewer sides, and that the only convex polyhedra with all triangular faces are those with \(V \leq 12\).

4.1.1 The Stereographic Projection Perspective

Another clean proof of Euler’s formula uses stereographic projection. Take the polyhedron and place it so that no vertex or edge lies on the top of the circumscribed sphere. Project from the north pole of the sphere onto the equatorial plane.

Stereographic projection:

     North pole N
         *
        /|\
       / | \
      /  |  \
     /   |   \    ← project from N through surface of sphere
    /    |    \      onto the equatorial plane
   *-----+-----*  equatorial plane
    \    |    /
     \   |   /
      \  |  \
       \ |  /
        \|/
         * South pole
         
  A point P on the sphere maps to P' = intersection
  of line NP with the equatorial plane.
  
  The polyhedron's edges become a planar graph.
  One face (containing N) becomes the outer face.
  Now apply Euler's formula for planar graphs.

4.2 Platonic Solids Revisited

Using Euler’s formula, we can prove rigorously that there are exactly five Platonic solids. Suppose we have a regular polyhedron with Schläfli symbol \(\{p, q\}\). Then:

Substituting into Euler’s formula:

For \(E\) to be a positive integer, the denominator \(2p + 2q - pq\) must be positive, giving \(\frac{1}{p} + \frac{1}{q} > \frac{1}{2}\). This has exactly five solutions in integers \(p, q \geq 3\): \((3,3)\), \((4,3)\), \((3,4)\), \((5,3)\), \((3,5)\), corresponding to the five Platonic solids.

Truncated Tetrahedron:

         (triangle)
           * * *
          *     *
         *       *
        * *     * *
       *   * * *   *
      *   (hex)    *
      *             *
       *           *
        * * * * * *
       (hexagonal face at bottom)
       
  V=12, E=18, F=8 (4 triangles + 4 hexagons)
  12 - 18 + 8 = 2 ✓

4.3 Duality

Two polyhedra are dual to each other if one can be obtained from the other by swapping vertices and faces, with edges corresponding to edges. More precisely, the dual of a polyhedron \(P\) is a polyhedron \(P^*\) whose vertices correspond to faces of \(P\), whose faces correspond to vertices of \(P\), and whose edges correspond to edges of \(P\) (connecting a vertex of \(P^*\) to another whenever the corresponding faces of \(P\) share an edge).

Duality is a deep symmetry: it shows that the distinction between “vertices” and “faces” is arbitrary from the combinatorial point of view. A theorem about vertices of \(P\) automatically gives a theorem about faces of \(P^*\). The same abstract combinatorial structure underlies both.

For Platonic solids with Schläfli symbol \(\{p, q\}\), the dual has Schläfli symbol \(\{q, p\}\). Thus:

• The tetrahedron \(\{3,3\}\) is self-dual.
• The cube \(\{4,3\}\) and octahedron \(\{3,4\}\) are dual.
• The dodecahedron \(\{5,3\}\) and icosahedron \(\{3,5\}\) are dual.

Cube-Octahedron duality:

CUBE {4,3}                  OCTAHEDRON {3,4}
  8 vertices                  6 vertices
  12 edges                    12 edges
  6 faces                     8 faces

  +---------+                      *
  |         |                     /|\
  |         |                    / | \
  |         |                   /  |  \
  +----+----+                  *---+---*
  |    |    |                   \  |  /
  |    |    |                    \ | /
  +---------+                     \|/
                                    *

  Place a point at center of    (cube's 6 face centers
  each cube face → 6 points      become octahedron's
  → connect neighbors             6 vertices)
  → octahedron appears!

Notice the symmetry: cube has 8V, 12E, 6F. Octahedron has 6V, 12E, 8F. The V and F have swapped, while E stays the same. This is always true for dual pairs.

4.4 Archimedean Solids

Beyond the Platonic solids, the next most symmetric class of convex polyhedra are the Archimedean solids: convex polyhedra whose faces are regular polygons (possibly of more than one type), and whose symmetry group acts transitively on the vertices (every vertex looks the same). There are 13 Archimedean solids, discovered by Archimedes (c. 287–212 BCE) and later catalogued systematically by Kepler in his 1619 Harmonices Mundi.

Archimedes’ original treatise on these solids is lost; our knowledge that he discovered them comes from Pappus of Alexandria (c. 320 CE), who listed all 13 and attributed them to Archimedes. Kepler rediscovered and proved completeness of the list using the Schläfli vertex-type classification.

Examples include:

• Truncated tetrahedron \(\{3,6,6\}\) (vertex type \(3.6.6\)): obtained by cutting the corners of a tetrahedron. Faces: 4 triangles, 4 hexagons.
• Cuboctahedron (vertex type \(3.4.3.4\)): the common dual of cube and octahedron. It sits halfway between them. Faces: 8 triangles, 6 squares.
• Truncated icosahedron (vertex type \(5.6.6\)): the soccer ball pattern, with 12 pentagons and 20 hexagons. This is also the structure of the Buckminsterfullerene (C\(_{60}\)) molecule.
• Snub cube (vertex type \(3.3.3.3.4\)): a chiral Archimedean solid with no improper symmetries. It comes in left-handed and right-handed versions.

Truncated Icosahedron (soccer ball):

        (pentagon)
         * * *
        *     *
       *   *   *
      * * (hex) * *
     *   *     *   *
    *     * * *     *
     *   *     *   *
      * * (hex) * *
       *   *   *
        *     *
         * * *
     (pentagon below)

  12 pentagons (black patches on a soccer ball)
  20 hexagons (white patches)
  V=60, E=90, F=32
  60 - 90 + 32 = 2 ✓

Each Archimedean solid also has a dual, called a Catalan solid, whose faces are congruent but not regular. For example, the dual of the cuboctahedron is the rhombic dodecahedron (12 rhombic faces, closely related to the FCC crystal structure), and the dual of the truncated icosahedron is the pentakis dodecahedron (60 triangular faces, appearing in certain virus capsid structures).

Chapter 5: Symmetries, Tessellations, and Wallpaper Groups

5.1 Isometries of the Euclidean Plane

An isometry of \(\mathbb{R}^2\) is a distance-preserving bijection — a map \(\phi: \mathbb{R}^2 \to \mathbb{R}^2\) such that \(d(\phi(P), \phi(Q)) = d(P, Q)\) for all points \(P\) and \(Q\). The set of all isometries forms a group under composition. Understanding isometries is fundamental to understanding symmetry, because the symmetries of any geometric figure form a subgroup of the isometry group.

Every isometry of the plane preserves distances and therefore maps lines to lines, angles to angles, and circles to circles. It can be orientation-preserving (like a rotation or translation) or orientation-reversing (like a reflection or glide reflection). The classification of all plane isometries is a beautiful and complete theorem.

• Translations \(T_{\mathbf{v}}\): \(P \mapsto P + \mathbf{v}\). Orientation-preserving, no fixed points (if \(\mathbf{v} \neq 0\)).
• Rotations \(R_{C,\theta}\): rotation by angle \(\theta\) about center \(C\). Orientation-preserving, fixed point \(C\) (if \(\theta \neq 0\)).
• Reflections \(M_\ell\): reflection across line \(\ell\). Orientation-reversing, fixed set is \(\ell\).
• Glide reflections: composition of reflection and translation along the reflection line. Orientation-reversing, no fixed points.

The orientation-preserving isometries (translations and rotations) form a subgroup called the group of direct isometries or motions. The full isometry group is \(\text{Isom}(\mathbb{R}^2) = \mathbb{R}^2 \rtimes O(2)\), the semidirect product of translations with the orthogonal group.

The four isometry types:

TRANSLATION by v:      ROTATION by θ about C:
  P → P + v               P → R_{C,θ}(P)

  * . . . . . .*          *
  P             P'           *   *
                              C← rotation center
                           *   *
                          *

REFLECTION across ℓ:   GLIDE REFLECTION:
  P → M_ℓ(P)             reflect then translate along ℓ

  P                        P
   *                        *     
    \                        \      
ℓ: -+-------               --+-------→ v
    |                        |
   *                        *
   P' (mirror image)        P'' (reflected then slid)

5.1.1 Composing Isometries

Isometries combine in non-obvious ways. Here are some key composition rules:

Parallel reflections → translation:

  ℓ₁       ℓ₂
  |         |        P
  |         |         *
  |         |    *    |    *
  |  d  |  d  |   P' |     P''
  |<--->|<--->|       |
  
  Reflect P in ℓ₁ → P'. Reflect P' in ℓ₂ → P''.
  P → P'' is a translation by 2d to the right.

Intersecting reflections → rotation:

      ℓ₁  ℓ₂
       \ /
    *   X   ← center of rotation
   P   / \
      /   \
     *     *
     P'    P''
     
  Reflect P in ℓ₁ → P'. Reflect P' in ℓ₂ → P''.
  P → P'' is a rotation by 2θ about the intersection.

5.2 Symmetry Groups and the Crystallographic Restriction

A discrete symmetry group of the plane is a group of isometries that acts properly discontinuously: every compact set is moved off itself by all but finitely many group elements. Such groups arise naturally in crystallography, where one studies repeating patterns in materials. The mathematical theory of these groups was developed in the late 19th century, motivated by the classification of crystal structures.

If a symmetry group contains translations in two linearly independent directions, it is called a wallpaper group (or plane crystallographic group). The set of all translation vectors in such a group forms a lattice \(\Lambda \subset \mathbb{R}^2\). Geometrically, a lattice is an infinite grid of parallelograms tiling the plane.

Why pentagons cannot tile the plane:

   * * *         * * *
  *     *       *     *
 *       *     *       *
  *     *   GAP  *     *
   * * *   ↑↑↑   * * *
           (a gap always appears;
            5×72°=360° works for rotation
            but not for tiling)

This rules out, for example, fivefold symmetry in crystalline materials — a fact that made the discovery of quasicrystals by Dan Shechtman in 1984 so surprising. Shechtman observed a diffraction pattern with clear tenfold (hence fivefold) symmetry in a rapidly cooled aluminum-manganese alloy, an observation so unexpected that his colleagues initially dismissed it as error. He was eventually awarded the 2011 Nobel Prize in Chemistry. Quasicrystals exhibit fivefold diffraction patterns because they are not periodic but quasiperiodic — see Section 5.5.

5.3 The 17 Wallpaper Groups

The wallpaper groups are the 17 distinct discrete symmetry groups of the plane that contain translations in two independent directions. That there are exactly 17 was proven independently by Fedorov (1891) and Schoenflies (1891). They are classified by their rotational symmetry, the presence of reflections and glide reflections, and the type of lattice.

The 17 groups are denoted using the crystallographic notation (also called Hermann-Mauguin or IUCr notation): \(p1\), \(p2\), \(pm\), \(pg\), \(cm\), \(p2mm\), \(p2mg\), \(p2gg\), \(c2mm\), \(p4\), \(p4mm\), \(p4gm\), \(p3\), \(p3m1\), \(p31m\), \(p6\), \(p6mm\).

The first symbol (p or c) indicates a primitive or centered lattice. The number indicates the maximal rotational order. Additional letters indicate reflections (m) or glide reflections (g).

Selected Wallpaper Groups — Fundamental Domains and Symmetries:

p1 (no symmetry beyond translation):    p4 (4-fold rotation):
  . . . . . .                           + + + + +
  . . . . . .                           + + + + +
  . . . . . .                           + + + + +
  (just translate the tile)             (90° rotation centers at +)

p2 (180° rotation):                     p6 (6-fold rotation):
  * * * * * *                           ◆ ◆ ◆ ◆ ◆
  * * * * * *                          ◆ ◆ ◆ ◆ ◆
  (rotation centers at *)              ◆ ◆ ◆ ◆ ◆
                                        (60° rotation centers)

pm (reflections in parallel lines):     p6mm (highest symmetry group):
  | | | | | |                           * - * - * - *
  | | | | | |                           |\ /|\ /|\ /|
  | | | | | |                           | X | X | X |
  (mirror lines vertical)              |/ \|/ \|/ \|
                                        * - * - * - *
                                        (reflections + 6-fold rotation)

5.4 Tessellations by Regular Polygons

A tessellation (or tiling) of the plane is a covering of \(\mathbb{R}^2\) by polygons (tiles) with no overlaps and no gaps. The simplest tessellations use a single type of regular polygon — these are the regular tessellations.

For a regular tessellation by \(\{p\}\) with \(q\) polygons meeting at each vertex, the angle condition requires that \(q\) copies of the interior angle of \(\{p\}\) exactly fill a full turn:

This simplifies to \(\frac{1}{p} + \frac{1}{q} = \frac{1}{2}\), which has exactly three solutions in integers \(p, q \geq 3\):

• \((p, q) = (3, 6)\): triangular tiling \(\{3, 6\}\).
• \((p, q) = (4, 4)\): square tiling \(\{4, 4\}\).
• \((p, q) = (6, 3)\): hexagonal tiling \(\{6, 3\}\).

Notice the parallel with the Platonic solid condition: replacing \(> 1/2\) (strict inequality, giving a curved surface) with \(= 1/2\) (equality, giving a flat plane) yields exactly the regular tessellations. And replacing \(< 1/2\) gives the hyperbolic tessellations, which tile the hyperbolic plane. The three cases — sphere, plane, hyperbolic plane — correspond to positive, zero, and negative curvature.

The Three Regular Tessellations:

TRIANGULAR {3,6}:          SQUARE {4,4}:         HEXAGONAL {6,3}:
  *-*-*-*-*                 +-+-+-+-+               * * * * *
 /|/|/|/|/|                 | | | | |              * * * * * *
*-*-*-*-*                   +-+-+-+-+               * * * * *
 \|\|\|\|\                  | | | | |              * * * * * *
  *-*-*-*-*                 +-+-+-+-+               * * * * *

6 triangles per vertex    4 squares per vertex    3 hexagons per vertex
6×60°=360° ✓              4×90°=360° ✓            3×120°=360° ✓

Semi-regular (Archimedean) tessellations allow more than one type of regular polygon but require all vertices to be equivalent. There are exactly 8 such tessellations (plus the 3 regular ones), including the snub square tiling and the trihexagonal tiling \((3.4.6.4)\).

Trihexagonal tiling 3.6.3.6:

      * * *
     *     *
    *       *
   * *     * *
  *   * * *   *
  *   *   *   *
  *   * * *   *
   * *     * *
    *       *
     *     *
      * * *

  At each vertex: triangle, hexagon, triangle, hexagon
  Angles: 60° + 120° + 60° + 120° = 360° ✓

5.5 Penrose Tilings and Quasiperiodicity

Penrose tilings provide striking examples of non-periodic tilings with local fivefold symmetry. Developed by Roger Penrose in 1974, they use two types of rhombus — a “thick” rhombus with angles \(72°\) and \(108°\), and a “thin” rhombus with angles \(36°\) and \(144°\) — assembled according to matching rules that force aperiodicity.

Penrose Tiling tiles:

Thick rhombus:             Thin rhombus:
   *                           *
  / \                         / \
 /   \                       /   \
*     *  (72° angles)       *     *  (36° angles)
 \   /                       \   /
  \ /                         \ /
   *                           *

  These two shapes, with matching rules
  (arrows on edges must match), tile the
  plane only APERIODICALLY.

Why are Penrose tilings quasiperiodic? The key is the inflation rule: each thick rhombus can be divided into smaller thick and thin rhombuses in a definite pattern (scaled by the golden ratio \(\phi = (1+\sqrt{5})/2\)), and similarly for the thin rhombus. Iterating this inflation generates larger and larger patches of a valid Penrose tiling. The golden ratio appears because the tiling has approximate fivefold symmetry, and \(\cos 72° = (\sqrt{5}-1)/4\) involves \(\sqrt{5}\).

Three remarkable properties of Penrose tilings:

1. Aperiodicity: No Penrose tiling has any translational symmetry. There is no vector \(\mathbf{v} \neq 0\) such that translating the tiling by \(\mathbf{v}\) gives the same tiling.

2. Local isomorphism: Every finite patch that appears in one Penrose tiling appears in every other Penrose tiling (with the same inflation rules). In fact, every finite patch appears infinitely often.

3. Fivefold quasisymmetry: The diffraction pattern of a Penrose tiling has perfect tenfold (hence fivefold) symmetry, matching the electron diffraction patterns of quasicrystals. This is why Shechtman’s discovery (Section 5.2) was eventually understood through the mathematics of Penrose tilings.

The connection to number theory is deep: Penrose tilings can be understood as two-dimensional “cut-and-project” sections of a five-dimensional periodic lattice. The fivefold symmetry that is forbidden in the plane can exist in \(\mathbb{R}^5\), and the tiling is a shadow of that higher-dimensional periodicity.

Chapter 6: Sphere Packings and the Kissing Problem

6.1 Sphere Packing Fundamentals

How efficiently can identical spheres be packed in space? This deceptively simple question has occupied mathematicians for over four centuries and, in dimensions 8 and 24, was finally resolved in 2016 by Maryna Viazovska — one of the most spectacular recent results in mathematics.

Formally, a sphere packing in \(\mathbb{R}^n\) is a collection of non-overlapping unit spheres (balls of radius 1) in \(n\)-dimensional Euclidean space. The centers form a discrete set \(S \subset \mathbb{R}^n\) with \(\|x - y\| \geq 2\) for all distinct \(x, y \in S\).

A lattice packing uses the translates of a lattice \(\Lambda \subset \mathbb{R}^n\) as sphere centers; every center is of the form \(Mv\) for a fixed matrix \(M\) and \(v \in \mathbb{Z}^n\). Lattice packings are easier to analyze but need not be optimal. In dimensions 8 and 24, however, the optimal packing turns out to be a lattice packing — a remarkable fact.

6.1.1 Worked Example: Density of the Hexagonal Lattice

Hexagonal circle packing:

  O   O   O   O   O
   O   O   O   O
  O   O   O   O   O
   O   O   O   O
  O   O   O   O   O

Each circle touches exactly 6 neighbors.
The centers form a triangular lattice.
Density ≈ 90.69% of area covered.

Fundamental domain (one parallelogram):
     *
    / \
   /   \    area = √3/2
  /     \
 *-------*
  (contains exactly 1 center)

6.2 Densest Packings in Low Dimensions

In dimension 1, spheres are intervals, and the unique packing achieves density 1. In higher dimensions, the optimal density decreases.

In dimension 2, the densest packing of circles is achieved by the hexagonal lattice (also called the triangular lattice), where each circle touches exactly six others. The density is

This was proven by Axel Thue (1892, with a gap) and rigorously by László Fejes Tóth (1940).

In dimension 3, the two most natural packings are:

• Face-centered cubic (FCC): spheres centered at all points with integer coordinates summing to an even number, or equivalently the lattice generated by \((1,1,0)\), \((1,0,1)\), \((0,1,1)\). Each sphere touches 12 others.
• Hexagonal close packing (HCP): a hexagonal layer arrangement alternating in an ABABAB pattern. Also achieves 12 touching neighbors.

Both achieve density

FCC vs HCP sphere packing layers:

FCC (ABC stacking):          HCP (AB stacking):

Layer C:  O   O   O          Layer B:  O   O   O
          (shifted by 1/3)             (shifted by 1/2)
Layer B:    O   O   O        Layer A:    O   O   O
          (shifted by 1/3)             (same as B, repeated)
Layer A:  O   O   O          
(base)                       
                             
FCC: offset by 1/3 each layer  HCP: offset by 1/2, alternates
Both give density π/(3√2) ≈ 74.05%

                FCC cross-section:
                
                * * *
               * * * *
              * * * * *
               * * * *   ← layers A, B, C, A, B, C, ...
                * * *

6.3 Kepler’s Conjecture and Hales’s Theorem

In 1611, Johannes Kepler stated in Strena seu de Nive Sexangula (“A New Year’s Gift of Six-Cornered Snow”) that no packing of equal spheres in \(\mathbb{R}^3\) has a greater density than the FCC or HCP arrangement. This became known as Kepler’s conjecture. The paper was motivated by the question of why snowflakes have hexagonal symmetry — Kepler traced it all the way back to sphere packing, making a remarkably prescient connection between macroscopic crystal form and microscopic atomic arrangement.

For nearly four centuries, the problem resisted all attempts at proof. In 1998, Thomas Hales announced a proof, and it was published in 2005.

Hales’s proof is a computer-assisted verification involving a reduction to a finite but large number of cases, each verified by computer. The proof reduced the problem to showing that a certain linear programming problem had a specific optimal value, then verified this by exhaustive search over several thousand cases. A fully formal proof (project Flyspeck, completed 2014) was subsequently certified in the Isabelle/HOL proof assistant, making it one of the most rigorously verified major mathematical results. The Flyspeck project, involving many collaborators over ten years, represents a new paradigm in mathematical proof: human-guided, machine-verified.

6.4 The Kissing Number Problem

A related question: how many unit spheres can simultaneously touch a given central unit sphere? This is the kissing number problem. If all spheres have radius 1 and the central sphere is centered at the origin, then the kissing number \(\kappa_n\) is the maximum number of points on the sphere \(S^{n-1}\) of radius 2 such that any two points are at distance at least 2 from each other. Equivalently, it is the maximum number of non-overlapping unit spheres that can touch a central unit sphere.

The term “kissing number” comes from the spheres “kissing” (touching) the central sphere without overlapping each other.

In dimension 2, the answer is clearly \(\kappa_2 = 6\) — six circles fit around a central circle in the hexagonal arrangement, and simple angle-counting shows that 7 cannot fit.

In dimension 3, the question was famously debated by Isaac Newton (who argued \(\kappa_3 = 12\)) and David Gregory (who thought 13 might be possible), in 1694. Their dispute was remarkable: Newton was right, but he could not prove it, and neither could Gregory. The reason Gregory thought 13 might work is that the 12 spheres in the icosahedral arrangement leave small gaps — the 12 touching spheres are not jammed, and can be moved around without separating from the central sphere. This “slack” makes it non-obvious that a 13th sphere cannot be squeezed in. The correct answer \(\kappa_3 = 12\) was not rigorously proven until Schütte and van der Waerden in 1953, nearly three centuries after Newton and Gregory’s debate.

The 12 touching spheres can be arranged in the FCC configuration (vertices of a cuboctahedron) but also in many other configurations, since the constraint is not tight — the 12 touching spheres in the icosahedral arrangement, for instance, leave small gaps between neighbors, and one cannot fit a 13th sphere without rearranging.

Kissing number in 3D:

   Cuboctahedron arrangement (FCC):    Icosahedral arrangement:
   
        *   *                              *   *
       / \ / \                            / \ / \
      *---*---*                          *---*---*
      |   |   |    12 spheres            |   |   |   12 spheres
      *---*---*    touch center          *---*---*   touch center
       \ / \ /                            \ / \ /
        *   *                              *   *

   In FCC: 12 touching spheres are        In icosahedral: same count,
   the vertices of a cuboctahedron.       but small gaps between neighbors.
   They are NOT all equidistant.          Slack means the arrangement
                                          is not rigid.

6.5 The Exceptional Dimensions 8 and 24

The kissing number problem has a particularly beautiful solution in two special dimensions. What makes dimensions 8 and 24 special is the existence of exceptional lattices — the \(E_8\) lattice and the Leech lattice — which are extraordinarily symmetric and self-dual.

Dimension 8. The \(E_8\) lattice is a remarkable lattice in \(\mathbb{R}^8\). Its points are all vectors \((x_1, \ldots, x_8)\) where either all coordinates are integers or all are half-integers, and \(\sum x_i\) is even. The \(E_8\) lattice has 240 vectors of minimal length \(\sqrt{2}\) (or 240 of length 2, depending on normalization), meaning each sphere touches exactly 240 others.

The \(E_8\) lattice has an extraordinary symmetry group of order 696,729,600 — the Weyl group of the exceptional Lie algebra \(E_8\). It is the unique even unimodular lattice in \(\mathbb{R}^8\) (meaning the squared lengths of all lattice vectors are even, and the lattice equals its own dual). These algebraic properties are precisely what make it optimal for sphere packing.

Maryna Viazovska (Ukrainian mathematician, born 1984) proved this in a 21-page paper submitted to arXiv in March 2016. The proof was checked and confirmed within a week — remarkable for a result of this importance. Within a week of her \(E_8\) paper, she and four co-authors (Cohn, Kumar, Miller, Radchenko) extended the method to dimension 24. Viazovska received the Fields Medal in 2022 for this work.

Dimension 24. The Leech lattice \(\Lambda_{24}\) in \(\mathbb{R}^{24}\) is an even unimodular lattice with minimum distance \(\sqrt{4} = 2\) (equivalently, minimum squared length 4) and 196,560 vectors of that minimal length. This gives the kissing number in dimension 24:

6.5.1 The Linear Programming / Viazovska Method

These proofs use the linear programming method of Delsarte (1972) combined with the interpolation method developed by Viazovska (2016). The Delsarte bound says: if there exists a function \(f: \mathbb{R}^n \to \mathbb{R}\) (a “test function”) satisfying:

1. \(f(0) = 1\),
2. \(f(x) \leq 0\) for all \(|x| \geq 2r\) (where \(r\) is the sphere radius),
3. \(\hat{f}(\xi) \geq 0\) for all \(\xi\) (non-negative Fourier transform),

then the packing density satisfies \(\Delta \leq \hat{f}(0)\). The bound is sharp (achievable) if equality holds at the lattice’s actual minimal vectors.

The challenge Viazovska solved was constructing an explicit “magic function” \(f\) for which the Delsarte bound exactly equals the \(E_8\) or Leech lattice density. She found this function using the theory of modular forms — quasiperiodic functions on the complex upper half-plane that arise in number theory, algebraic geometry, and string theory. The remarkable fact is that the magic function can be written explicitly in terms of classical modular forms (Eisenstein series, the Jacobi theta function, the Ramanujan \(\Delta\)-function), and its key properties can be verified algebraically.

6.6 Known Kissing Numbers

The exact kissing number is known only for \(n = 1, 2, 3, 4, 8, 24\):

In dimension 4, the 24-cell \(\{3,4,3\}\) — whose vertices form the root system of \(D_4\) — realizes the kissing number 24. This connects the sphere packing problem back to the exceptional regular polytopes of Chapter 3: the 24-cell’s vertices are precisely the 24 shortest vectors of the \(D_4\) lattice.

The determination of kissing numbers in other dimensions (especially 5, 6, 7) remains one of the central open problems in discrete geometry. Upper and lower bounds are known but there are significant gaps. The Delsarte linear programming bound gives upper bounds, while explicit constructions of lattices or packings give lower bounds.

Known kissing numbers — visualization:

Dimension:  1    2    3    4       8         24
Kissing:    2    6   12   24     240     196,560
            |    |    |    |       |           |
            ●    ●    ●    ●       ●           ●

  Growth is NOT monotone or simple:
  - Low dimensions: slow growth (2, 6, 12, 24)
  - Dim 8 jumps to 240 due to E₈
  - Dim 24 jumps to 196,560 due to Leech lattice
  - Dims 5,6,7 are UNKNOWN exactly
  
Known bounds for unknown dimensions (approx):
  Dim 5: between 40 and 44
  Dim 6: between 72 and 78
  Dim 7: between 126 and 134
  Dim 9: between 306 and 364

Supplementary Material: Deeper Explorations

S1: The Geometry of the Parallel Postulate — Alternative Geometries

The preceding chapters developed Euclidean geometry as the unique model of Hilbert’s axiom system (Chapter 1). But what happens if we replace the parallel postulate with something else? The two main alternatives give hyperbolic geometry (infinitely many parallels through a given point) and elliptic geometry (no parallels at all). Understanding these alternatives illuminates Euclidean geometry by contrast.

S1.1 Hyperbolic Geometry: The Poincaré Disk Model

The most vivid model of hyperbolic geometry is the Poincaré disk model, introduced by Henri Poincaré (1882). The “plane” is the open unit disk \(\mathbb{D} = \{(x,y) : x^2 + y^2 < 1\}\). “Lines” are arcs of circles that meet the boundary circle \(\partial\mathbb{D}\) at right angles (including diameters, which are limiting cases of such circles). The distance between two points \(z, w \in \mathbb{D}\) (using complex number notation) is

Poincaré Disk Model of Hyperbolic Plane:

         ___________
        /           \
       /      L1     \
      |    /------\   |  ← L1 is a "line" (arc meeting
      |   /    P   \  |    the boundary at right angles)
      |  /     *    \ |
      | /  L2   \   / |  ← L2 and L3 both pass through P
      |/    \    \ /  |    and are parallel to L4
      *      \    *   *
      |\      \  /|   |
      | \  L3  \/  L4/    L4 does NOT meet L2 or L3
      |  \    /\  /  |    (they are parallel in the hyperbolic sense)
      |   \  /  \/   |
       \   \/   /   /
        \       \  /
         \______\/
         
  Through point P, there are INFINITELY MANY lines
  parallel to L4. This violates Euclid's fifth postulate.
  
  The boundary circle is the "circle at infinity" —
  points on it are infinitely far from any interior point.

In hyperbolic geometry, the angle sum of a triangle is less than \(180°\), and the defect \(180° - (\angle A + \angle B + \angle C)\) is proportional to the triangle’s area. This was computed by Lambert (1766) before non-Euclidean geometry was formalized. Larger triangles have smaller angle sums, and the maximum area of a hyperbolic triangle (a triangle with all three vertices on the boundary circle, having all angles zero) is \(\pi\) in the normalization where the curvature is \(-1\).

This formula has no Euclidean analogue (Euclidean triangles all have angle sum \(\pi\), so the right side would always be zero). It shows that in hyperbolic geometry, the angle sum of a triangle encodes its area — a remarkable coupling of angular and metric information that does not exist in Euclidean geometry.

S1.2 Elliptic Geometry: The Spherical Model

In elliptic geometry, there are no parallel lines — any two distinct lines meet. The simplest model is the sphere \(S^2\), with “lines” being great circles and “points” being pairs of antipodal points. (If we use actual points rather than antipodal pairs, we get spherical geometry, which is slightly different from elliptic geometry.)

Spherical / Elliptic Geometry:

         North Pole
             *
            /|\
           / | \
          /  |  \
         /   |   \
        *----|----*   ← equator (a "line" = great circle)
         \   |   /
          \  |  /
           \ | /
            \|/
             *
         South Pole
         
  Any two great circles ALWAYS INTERSECT (at two antipodal points).
  "Lines" never miss each other.
  
  Triangle on a sphere:
  
      *← North Pole
     /|\
    / | \
   /  |  \
  *---+---*  ← equator
  
  Angles at each corner: 90°, 90°, 90°
  Angle sum = 270° > 180°!
  
  On a sphere, angle sum of a triangle is GREATER than 180°,
  and the excess equals the area (times 1/R² for a sphere of radius R).

The quantity \(\alpha + \beta + \gamma - \pi\) is the spherical excess. For a triangle occupying 1/8 of the sphere (like the triangle from the North Pole to two points on the equator separated by 90°), the angles are \(90° + 90° + 90° = 270°\), excess \(= \pi/2\), and area \(= \pi R^2/2 = (1/8)(4\pi R^2)\). ✓

The three geometries — Euclidean, hyperbolic, elliptic — are distinguished by the sign of their curvature and by the Gauss-Bonnet theorem: angle sum of a triangle equals \(\pi\), less than \(\pi\), or greater than \(\pi\), respectively.

S2: Projective Geometry and Duality

Projective geometry is the geometry of the projective plane \(\mathbb{P}^2\), obtained from \(\mathbb{R}^2\) by adding a “line at infinity” — one extra point for each direction. The remarkable feature of projective geometry is its duality: any theorem about points and lines remains valid when “point” and “line” are interchanged.

S2.1 The Projective Plane

The real projective plane \(\mathbb{P}^2(\mathbb{R})\) is defined as the set of all lines through the origin in \(\mathbb{R}^3\). A “point” in \(\mathbb{P}^2\) corresponds to a line through the origin in \(\mathbb{R}^3\), represented by homogeneous coordinates \([X : Y : Z]\) (not all zero). A “line” in \(\mathbb{P}^2\) corresponds to a plane through the origin in \(\mathbb{R}^3\).

The ordinary Euclidean plane \(\mathbb{R}^2\) embeds into \(\mathbb{P}^2\) via \((x, y) \mapsto [x : y : 1]\). The “points at infinity” correspond to \([X : Y : 0]\) — directions in the plane, one for each family of parallel lines.

Projective Plane: adding points at infinity

         [0:1:0]  (vertical direction)
             *
             |
             |  ← "line at infinity" [0:0:1] = 0
             |    contains all [X:Y:0]
  [1:0:0]*---+---*  ordinary plane Z≠0
    (horiz)  |
             |
             *
         [0:-1:0]

  Parallel lines in R² (same direction) meet at ONE point at infinity.
  Two distinct families of parallels meet at DIFFERENT points at infinity.
  The "line at infinity" contains ALL points at infinity.
  
  In P², any two distinct lines meet in exactly one point.
  (No more parallel lines — they all meet at infinity!)

S2.2 Projective Duality and Poles and Polars

In the projective plane, there is a perfect symmetry between points and lines called projective duality. Every theorem about points and lines in \(\mathbb{P}^2\) has a “dual” theorem obtained by interchanging “point” and “line.”

For a conic section (like the unit circle \(x^2 + y^2 = 1\)), the polar line of a point \(P = (a, b)\) with respect to the conic is the line \(ax + by = 1\). This gives an explicit bijection between points and lines — the pole-polar duality. If \(P\) lies inside the conic, its polar line lies outside; if \(P\) lies on the conic, its polar line is the tangent at \(P\).

S2.3 Pascal’s Theorem and Its Dual

Brianchon discovered his theorem in 1806 not by independent geometric insight but by applying the principle of duality to Pascal’s theorem — an early and striking demonstration of the power of projective duality.

Pascal's Theorem (six points on a conic → Pascal line):

        B
       / \
      /   \   ← conic (ellipse)
     A     C
     |     |
     F     D
      \   /
       \ /
        E

  Extended sides meet at P, Q, R:
  P = (line AB) ∩ (line DE)
  Q = (line BC) ∩ (line EF)
  R = (line CD) ∩ (line FA)
  
  Claim: P, Q, R are COLLINEAR (on the Pascal line).
  This holds even when the conic degenerates to two lines!

S3: Detailed Proof Sketches and Additional Theorems

S3.1 The Angle Bisector Construction and Compass-Only Constructions

We saw in Section 1.3.1 how to bisect an angle with straightedge and compass. A surprising theorem, due to Georg Mohr (1672) and independently Lorenzo Mascheroni (1797), shows that the straightedge is not needed at all:

The key insight is that we can simulate drawing a line with a compass: while we cannot literally draw the line, we can find its intersection with a circle (which is what we actually need). The simulation uses the fact that two points determine a unique line, and intersecting that line with another circle can be done by a purely compass-based construction (using the two points as proxies for the line).

Conversely, the Poncelet-Steiner Theorem shows that if we are given a single fixed circle with its center marked, then the compass is not needed:

Together, these theorems show that the constructible points depend only on the set of operations abstractly, not on whether we use both tools. The single fixed circle serves as a “reference” that replaces all compass operations.

S3.2 Napoleon’s Theorem

One of the most elegant theorems about equilateral triangles is Napoleon’s Theorem, whose attribution to Napoleon Bonaparte (1769–1821) is almost certainly apocryphal but too charming to abandon:

Napoleon's Theorem:

       B
      / \
     /   \
    A-----C   ← original triangle
    
Construct equilateral triangles on each side:

        *         *
       /|         |\
      / |  B      | \
     *  | / \     |  *
      \ |/   \    | /
       \A-----C  /
        *         *
        
The three NEW centroids (marked ★) form an equilateral triangle:

              ★
             / \
            /   \
           ★-----★   ← Napoleon's triangle (equilateral!)

The difference of the outer and inner Napoleon triangles’ areas equals the area of the original triangle — a beautiful quantitative version of the theorem.

S3.3 The Nine-Point Circle

Another gem of classical Euclidean geometry is the nine-point circle, discovered by Feuerbach (1822):

Nine-Point Circle:

               A
              /|\
             / | \  ← altitude from A, foot H_A
            /  |  \
           /   H_A  \
          /    |     \
  M_B----+--N₉-+----M_C   ← midpoints of sides
          \    |     /
           \   |    /
            \  |   /
             \ H  /  ← orthocenter H
              \|/
               B---M_A---C

  Nine-point center N₉ is the midpoint of OH
  (where O = circumcenter, H = orthocenter).
  
  Nine points on the circle:
    M_A, M_B, M_C (midpoints of BC, CA, AB)
    H_A, H_B, H_C (feet of altitudes)
    E_A, E_B, E_C (midpoints of AH, BH, CH)

The most remarkable part of Feuerbach’s theorem is an additional fact he proved: the nine-point circle is internally tangent to the incircle and externally tangent to each of the three excircles of the triangle. This is Feuerbach’s theorem, one of the most beautiful results in classical geometry.

S3.4 The Euler Line

In any (non-equilateral) triangle, the circumcenter \(O\), the centroid \(G\), and the orthocenter \(H\) are collinear. The line through them is the Euler line, and the distances satisfy \(OG : GH = 1 : 2\) (the centroid trisects the segment from circumcenter to orthocenter).

The Euler Line:

       A
      /|\
     / | \
    /  |  \
   /   H   \   ← orthocenter H
  / ---+--- \
 B--G--+--O--C
     centroid circumcenter
     
  O----G----H   (Euler line)
  |    |    |
  1    1    (OG = (1/3)OH, GH = (2/3)OH)
  
  Also on the Euler line: nine-point center N₉ = midpoint of OH.

The nine-point center \(N_9\) also lies on the Euler line, at the midpoint of \(OH\). In an equilateral triangle, \(O = G = H = N_9\) (all four coincide), which is why the theorem requires “non-equilateral.”

S4: Symmetry Groups — Deeper Structure

S4.1 The Point Groups: Symmetries Fixing a Point

Before wallpaper groups (which have translations), let us consider point groups — symmetry groups that fix a point. In the plane, the finite groups of isometries fixing the origin are:

• Cyclic group \(C_n\): generated by rotation by \(2\pi/n\). Order \(n\). Appears as the symmetry of a regular \(n\)-gon that has been “chiralized” (no reflections).
• Dihedral group \(D_n\) (or \(D_{2n}\) in some notations): generated by rotation by \(2\pi/n\) and a reflection. Order \(2n\). Appears as the full symmetry group of a regular \(n\)-gon.

This is often called the “Leonardo theorem” because da Vinci apparently knew (informally) the classification of symmetries of rosette patterns in his study of architecture. The rigorous proof uses the fact that any finite group of isometries fixing a point consists only of rotations and reflections (no translations, which would require infinite orbits), and that any combination of reflections generates a dihedral group.

Symmetry axes of a regular hexagon (D₆):

       *
      / \
  ---*   *---   ← 3 axes through opposite vertices
      \ /
       *       + 3 axes through opposite edge midpoints
       
  Total: 6 reflection axes + 6 rotational symmetries = D₆ of order 12

S4.2 Three-Dimensional Point Groups and Crystal Systems

In three dimensions, the finite groups of isometries are more complex. They include the rotation groups of the Platonic solids (tetrahedral, octahedral, icosahedral) and their extensions by reflections. These are classified into the crystallographic point groups, of which there are 32.

The cubic point group \(O_h\) (order 48) is the full symmetry group of both the cube and the octahedron (since they are dual). The icosahedral group \(I_h\) (order 120) is the full symmetry group of the dodecahedron and icosahedron — but it is NOT a crystallographic point group, because 5-fold symmetry is incompatible with a lattice (the crystallographic restriction).

S4.3 Space Groups: Combining Translations and Point Symmetries

The 230 space groups classify all possible discrete symmetry groups of \(\mathbb{R}^3\) that contain a lattice of translations. They combine the 32 crystallographic point groups with the 14 Bravais lattice types. The classification was completed by Fedorov, Schoenflies, and Barlow in 1890–1891.

Crystallographers use space groups to classify crystal structures: every crystal belongs to one of the 230 space groups. The Nacl (table salt) crystal has space group \(Fm\bar{3}m\) (number 225), one of the cubic space groups. Diamond has space group \(Fd\bar{3}m\) (number 227).

S5: Additional Examples in Sphere Packings

S5.1 The E₈ Root System in Detail

The \(E_8\) lattice can be described explicitly. Define the lattice \(\Lambda\) as the set of all vectors \((x_1, \ldots, x_8) \in \mathbb{R}^8\) such that:

• either all \(x_i \in \mathbb{Z}\) with \(\sum x_i \equiv 0 \pmod{2}\),
• or all \(x_i \in \mathbb{Z} + \frac{1}{2}\) with \(\sum x_i \equiv 0 \pmod{2}\).

The shortest vectors (minimal length \(\sqrt{2}\)) are the roots of the \(E_8\) root system. There are 240 of them, and they can be listed explicitly:

• 112 vectors of the form \(\pm e_i \pm e_j\) for \(1 \leq i < j \leq 8\): each with two nonzero coordinates of value \(\pm 1\).
• 128 vectors of the form \(\frac{1}{2}(\pm 1, \pm 1, \pm 1, \pm 1, \pm 1, \pm 1, \pm 1, \pm 1)\) with an even number of negative signs.

Projection of E₈ roots into 2D (the "E₈ Gosset polytope" projection):

  This shows the 240 roots projected onto a 2D plane.
  The actual roots live in 8D space.
  
  In projection:
  
                  * * *
              * *       * *
           * *    * * *    * *
          *    * *     * *    *
         *  * *    * *    * *  *
        * *    * *   * *    * * *
       *    * *   * * *   * *    *
        * *    * *   * *    * *
         *  * *    * *    * *  *
          *    * *     * *    *
           * * *   * * *   * *
               * *       * *
                  * * *
                  
  240 dots arranged with 8-fold symmetry visible in projection.
  (Actual symmetry group has order ~7×10⁸.)

S5.2 Random Packings and the Bravais Lattice Hierarchy

Not all sphere packings are lattice packings. Random packings arise naturally in physics when spheres are poured into a container and allowed to settle. Experiments and simulations show that random sphere packings in \(\mathbb{R}^3\) achieve a density of about 0.64 — the “random close packing” density — compared to 0.7405 for the optimal crystalline packing.

The gap between random (\(\approx 64\%\)) and crystalline (\(\approx 74\%\)) packings is significant. A key question in materials science is: why do some materials crystallize (achieving nearly optimal packing) while others form glasses (disordered packings)? The answer involves the kinetics of crystallization as well as the thermodynamics, and is still not fully understood.

S5.3 Sphere Packing Bounds via Linear Programming

The Delsarte (or linear programming) method provides upper bounds on sphere packing density. Here is the precise statement for sphere packings:

For the \(E_8\) lattice, Viazovska found a specific function \(f^*\) for which the bound is tight: \(\Delta = \pi^4/384\). The function \(f^*\) is constructed using the theory of quasi-modular forms and satisfies additional constraints (it has double zeros at the minimal vectors of \(E_8\) and their integer multiples) that force the bound to be achieved exactly by the \(E_8\) lattice.

S6: Classical Theorems in Triangle Geometry

This supplementary section collects additional classical results that illuminate the richness of Euclidean geometry within the triangle.

S6.1 Ceva’s Theorem

This theorem unifies many classical concurrency results:

• Medians (centroid): \(D, E, F\) are midpoints, so each ratio is 1, and \(1 \cdot 1 \cdot 1 = 1\). ✓
• Angle bisectors (incenter): By the angle bisector theorem, \(BD/DC = AB/AC\), etc. The product telescopes to 1. ✓
• Altitudes (orthocenter): One can verify the ratios using trigonometry. ✓

Ceva's Theorem — three concurrent cevians:

            A
           /|\
          / | \
         /  |  \
        /   P   \   ← P is the point of concurrency
       /   /|\   \
      /   / | \   \
     B---D--+--F---C
          ↑   ↑
          on BC  on AB (wait — rearranged below)
          
  D on BC, E on CA, F on AB.
  
  BD   CE   AF
  -- · -- · -- = 1  ⟺  AD, BE, CF concurrent.
  DC   EA   FB

S6.2 Menelaus’s Theorem

Ceva’s theorem and Menelaus’s theorem are “dual” to each other in a sense: Ceva’s condition gives concurrence (meeting at a point), while Menelaus’s gives collinearity (lying on a line). This duality is a precursor to projective duality.

S6.3 The Simson Line

The Simson line is one of the most beautiful theorems in elementary geometry. As \(P\) moves around the circumcircle, its Simson line rotates. If \(P\) moves by an arc of \(2\alpha\), the Simson line rotates by \(\alpha\).

Simson Line:

               P
              *|
             /  \  ← P on circumcircle
            /    \
  A--------*------*--------B
          Pa      Pb     ← feet of perpendiculars from P
           \     /
            \   /
             \ /
              *Pc  ← foot of perp from P to AB (extended)
              C
              
  Pa, Pb, Pc are COLLINEAR (on the Simson line of P).

The Simson line has a beautiful characterization in terms of the orthocenter: the Simson line of \(P\) bisects the segment from \(P\) to the orthocenter \(H\) of \(\triangle ABC\).

S6.4 Ptolemy’s Theorem

Ptolemy’s theorem has a powerful corollary: setting the quadrilateral to be a rectangle (with \(AC = BD =\) diagonal), it gives the Pythagorean theorem! And Ptolemy used his theorem as the basis for computing a table of chords (the ancient equivalent of a sine table), which was the cornerstone of Greek and Islamic trigonometry.

S7: From Euclid to Modern Geometry — A Historical Summary

The arc from Euclid’s Elements to Viazovska’s sphere packing proof spans 2300 years and represents one of humanity’s greatest intellectual achievements. Here is a compressed timeline of the key developments:

What unifies this timeline is the persistent interplay between geometric intuition and algebraic formalization. Euclid’s intuition guided two millennia of geometry, but only Hilbert’s formalization made it rigorous. The intuition that the FCC/HCP packing should be optimal guided four centuries of attempts, but only Hales’s computer proof (and Viazovska’s elegant analytic argument in higher dimensions) provided certainty. The intuition that fivefold symmetry is “impossible” for crystals guided crystallographers for a century, until Shechtman’s observation forced a fundamental reconception of what “crystal” means.

Modern geometry is characterized by its connections to algebra, topology, analysis, and number theory. The classification of wallpaper groups uses group theory. Euler’s formula is a topological invariant. Viazovska’s proof uses modular forms from number theory. The geometry of the 21st century increasingly looks like the intersection of all of mathematics.

The web of connections in this course:

  AXIOMS          CONICS              POLYTOPES
  (Ch. 1)         (Ch. 2)             (Ch. 3)
     |                |                   |
  Hilbert      Apollonius,          Euclid, Plato,
  Wantzel      Kepler orbits        Kepler-Poinsot
  Origami      Projective geom      4D polytopes
     |                |                   |
     +--------+--------+---------+--------+
                       |
              POLYHEDRA (Ch. 4)
              Euler's formula
              Duality, Archimedean
                       |
              SYMMETRIES (Ch. 5)
              17 wallpaper groups
              Crystallography
              Penrose tilings
                       |
              SPHERE PACKING (Ch. 6)
              Kepler → Hales
              E₈ → Viazovska
              Leech → Monster group

Each chapter feeds into the next: the Platonic solids of Chapter 3 reappear as the 24-cell in Chapter 6’s kissing number table; the conic sections of Chapter 2 reappear in Viazovska’s use of modular forms; the symmetry groups of Chapter 5 classify the crystal structures that Chapter 6 studies for packing efficiency. Euclidean geometry, properly understood, is not a static ancient subject but a living web of ideas connecting the most ancient and the most modern mathematics.

S8: Worked Problem Sets with Full Solutions

This section provides worked problems at the level of the course, illustrating the main techniques and theorems.

S8.1 Problems on Constructibility

S8.2 Problems on Conic Sections

Parabolic orbit with Sun at focus:

                Comet at θ=90°:
              r=2 AU
             *
            /
           /
Sun *------*-----------  (direction θ=0)
  (focus)  perihelion
           r=1 AU

Orbit: r = 2/(1 + cosθ)
The semi-latus rectum p=2 = distance at θ=90°.
The comet escapes to infinity as θ → 180°.

S8.3 Problems on Euler’s Formula and Polyhedra

S8.4 Problems on Wallpaper Groups and Symmetries

p6 fundamental domain:         p6mm fundamental domain:

      *                               *
     / \                             /|\
    /   \  ← 60° sector             / | \
   /     \                          /  |  \
  *-------*                        *---*---*
  
  No reflection axes.             6 reflection axes through
  Only 6-fold rotation.           center of hex cell.

S9: Geometry and Physics — Connections

S9.1 Platonic Solids and the Elements

Plato’s assignment of the five solids to the elements in the Timaeus (c. 360 BCE) was not arbitrary. He chose based on geometric properties: the tetrahedron to fire because it is the sharpest; the cube to earth because it packs perfectly; the octahedron to air because it can spin freely held between two fingers; the icosahedron to water because it is the most spherical; and the dodecahedron to the cosmos, because it most closely approximates the sphere.

While this scheme was not scientific (it was not predictive or falsifiable), it represented the first attempt to connect abstract mathematical structures to physical reality — a mode of thinking that would culminate in Kepler’s solar system model (1596, the Mysterium Cosmographicum) and, much later, in the Standard Model of particle physics, where symmetry groups classify fundamental particles.

Kepler, building on Plato, proposed that the distances of the six known planets from the Sun were determined by nested Platonic solids: Saturn outside a cube, cube outside a tetrahedron, tetrahedron outside a dodecahedron, dodecahedron outside an icosahedron, icosahedron outside an octahedron, with the Sun at the center. The model gave rough agreement with known planetary distances and convinced Kepler to pursue mathematical astronomy. The model was abandoned when Kepler discovered his three laws of planetary motion — and realised the orbits are ellipses (Chapter 2), not circles.

S9.2 The Shape of Space

What is the global shape of the universe? General relativity allows the universe to have any of several topologies: it could be flat (Euclidean), positively curved (spherical), negatively curved (hyperbolic), or various compact manifolds. The three geometries of Chapter S1 — Euclidean, spherical, and hyperbolic — are the three geometries of constant curvature, and one of them (or a mixture) describes the large-scale geometry of spacetime.

Current cosmological measurements (particularly of the cosmic microwave background) suggest that the universe is spatially flat to within measurement precision — the angles of enormous triangles sum to \(180°\). But topology allows a flat universe to be compact: the three-torus \(T^3 = \mathbb{R}^3/\mathbb{Z}^3\) is flat but finite, like a three-dimensional video game where leaving one edge brings you back at the opposite edge.

The Euler characteristic connects to this: a closed 3-manifold’s topology constrains its possible symmetries and curvature. The tools developed in this course — symmetry groups, Euler characteristics, fundamental domains — are the building blocks of modern 3-manifold topology.

S9.3 Quasicrystals and Penrose Tilings Revisited

When Dan Shechtman observed tenfold symmetry in electron diffraction in 1984, the standard crystallographic dogma (crystallographic restriction theorem, Section 5.2) said this was impossible. The community’s response ranged from skepticism to outright dismissal — Linus Pauling famously said “There is no such thing as quasicrystals, only quasi-scientists.”

The mathematical resolution came through Penrose tilings (Section 5.5): quasicrystals are three-dimensional analogs of Penrose tilings. They are quasiperiodic — they can be described as projections of periodic structures in higher-dimensional space (specifically, as two-dimensional slices of five-dimensional periodic lattices, projected to three dimensions). The fivefold symmetry that is forbidden in two or three dimensions can exist in five dimensions and survive the projection.

Shechtman was awarded the Nobel Prize in Chemistry in 2011, and quasicrystals have since been found in many materials, including a natural mineral (icosahedrite, discovered in a Siberian meteorite) and in ancient Iraqi metalwork. The crystallographic restriction theorem is not wrong — it correctly characterizes periodic crystals — but Nature is not limited to periodic structures.

Quasicrystal vs. Crystal diffraction:

Crystal (periodic):          Quasicrystal:
  
  * * * * *                    *
  * * * * *                  *   *
  * * * * *                *       *  ← tenfold
  * * * * *                  *   *     symmetry!
  * * * * *                    *
  
  Diffraction spots:           Diffraction spots:
  4-fold or 6-fold             10-fold symmetric
  symmetric                    arrangement
  
  Lattice translation T:       No translation T:
  f(r + T) = f(r)              f(r + T) ≠ f(r) for any T≠0
  (periodic)                   (aperiodic, quasiperiodic)

S10: Further Reading and Open Problems

Open Problems in the Topics of This Course

The following major open problems connect directly to the material in this course:

1. Sphere packing in dimensions 5–7, 9, etc. Only in dimensions 1, 2, 3, 8, and 24 is the optimal sphere packing density known. In all other dimensions, there are gaps between upper bounds (from linear programming) and lower bounds (from explicit lattice packings). Dimensions 5, 6, and 7 are the next natural targets, but no exceptional lattices are known in these dimensions.

2. Kissing numbers in dimensions 5–7. The exact kissing number is unknown for \(n = 5, 6, 7, 9, 10, \ldots\). For \(n = 5\), the best known bounds are \(40 \leq \kappa_5 \leq 44\). Closing this gap seems to require fundamentally new methods.

3. Are there infinitely many Fermat primes? The known Fermat primes (3, 5, 17, 257, 65537) are the only values of \(2^{2^m} + 1\) known to be prime. No new Fermat prime has been found despite computer searches to enormous values of \(m\). If there are only finitely many Fermat primes, then only finitely many regular \(n\)-gons (for odd \(n\)) are constructible.

4. The Erdős-Ko-Rado problem for spheres. How many unit spheres can be packed in a ball of radius \(R\) in \(\mathbb{R}^n\)? This finite version of sphere packing has connections to coding theory and combinatorics.

5. Quasicrystal mathematics. The precise mathematical theory of quasicrystals — their diffraction spectra, their classification, the “cut-and-project” method — is still being developed. The connection to modular forms and automorphic forms (used by Viazovska) suggests deeper structure.

S11: Nets, Unfoldings, and Geometric Constructions

S11.1 Nets of Polyhedra

A net of a polyhedron is a planar figure obtained by cutting along some edges and unfolding the surface flat. Not every assignment of cuts produces a valid net (the faces might overlap), and the same polyhedron can have many different nets.

Tetrahedron net (one of 2 distinct nets):

       /\
      /  \
     /    \
    /------\
   /\  /\  /\
  /  \/  \/  \
 (4 equilateral triangles unfolded)

Cube net (one of 11 distinct nets — the "cross"):

    +--+
    |  |  ← top face
+--++--++--+--+
|  ||  ||  |  |  ← middle strip (4 faces)
+--++--++--+--+
    |  |  ← bottom face
    +--+

Octahedron net (one of 11 distinct nets):

  /\/\/\/\
 /\/\/\/\/\
(8 triangles in a zig-zag strip)

Dodecahedron net (one of 43380 distinct nets):

       * * *
      *  P  *        P = pentagon
    * * * * * *
   * P * * P * *     (12 pentagons unfolded;
   * * * * * * *      one net has a "cap" of
      *  P  *         6 pentagons on top and bottom)
       * * *
       
Icosahedron net (one of 43380 distinct nets):
  
  /\/\/\/\/\/\/\/\/\/\
 /\/\/\/\/\/\/\/\/\/\/\
(20 triangles in a double row)

An open problem is whether every convex polyhedron has at least one non-overlapping net. For Platonic and Archimedean solids, nets exist and have been catalogued. For general convex polyhedra, the question is still open (Shephard’s conjecture, 1975).

S11.2 The Rhombic Dodecahedron and Space-Filling

The rhombic dodecahedron (the dual of the cuboctahedron) has a special property: it tiles three-dimensional space without gaps. Its 12 rhombic faces have diagonals in ratio \(\sqrt{2}:1\), and the solid fills space in the same way that the cube fills space, but with a different shape.

Rhombic dodecahedron — space-filling:

  Side view:               Top view:
  
    *                          *   *
   / \                        / \ / \
  *   *                      *   *   *
  |   |                       \ / \ /
  *   *                        *   *
   \ /                        / \ / \
    *                        *   *   *
                              \ / \ /
                               *   *
                               
  Copies fit together perfectly — no gaps.
  This is the natural Voronoi cell of the FCC lattice:
  the region of space closer to a given FCC lattice point
  than to any other has the shape of a rhombic dodecahedron.

The rhombic dodecahedron is the Wigner-Seitz cell (or Voronoi cell) of the FCC lattice. This means it plays a fundamental role in the physics of FCC metals: the “first Brillouin zone” (in reciprocal space) for an FCC crystal is the Wigner-Seitz cell of the BCC lattice, which is a truncated octahedron — another Archimedean solid.

S11.3 The 24-Cell as a Self-Dual 4-Polytope

(the 24 vectors \(\pm e_i \pm e_j\) in \(\mathbb{R}^4\)), where \(e_1, e_2, e_3, e_4\) are the standard basis vectors. Each vertex is connected by edges to exactly 8 others (at distance \(\sqrt{2}\)), and each vertex is surrounded by 6 octahedral cells.

The 24-cell is related to the quaternion group: its vertices, when viewed as unit quaternions \(\pm 1, \pm i, \pm j, \pm k, (\pm 1 \pm i \pm j \pm k)/2\), form the binary tetrahedral group — a group of order 24. The 24-cell thus has an algebraic interpretation that the other regular 4-polytopes lack.

24-cell cross-sections as hyperplane sweeps:

Sweep direction ↓ :

 Layer 1:  * (single vertex)
 
 Layer 2:  * * *     (small triangle of vertices)
          * * * *
          
 Layer 3:  * * * * *  (larger cross-section)
          * * * * * *
          * * * * *
          
  (middle section is a cuboctahedron — 12 vertices,
   24 edges, 8 triangular + 6 square faces)
   
 Layer 4:  (mirror of Layer 3)
 
 Layer 5:  * (single vertex, the "antipode")

S12: Geometric Inequalities

Geometry is not only about equalities (like Euler’s formula) but also about geometric inequalities — bounds on areas, volumes, and other quantities. Some of the most beautiful results in this category connect geometry to symmetry.

S12.1 The Isoperimetric Inequality

with equality for the sphere.

S12.2 The Brunn-Minkowski Inequality

A powerful generalization relates volumes of sets and their “Minkowski sums”:

The Brunn-Minkowski inequality is one of the deepest results in convex geometry. It implies the isoperimetric inequality (take \(L = \epsilon B\) where \(B\) is the unit ball and let \(\epsilon \to 0\)). It also implies the AM-GM inequality (as a special case with \(n = 1\)).

The connection to sphere packing is through the Rogers bound: any sphere packing in \(\mathbb{R}^n\) has density at most \(n \cdot 2^{-n/2+1}\) (roughly), a bound derived using covering arguments related to Brunn-Minkowski.

S12.3 The Cauchy-Schwarz Inequality and Geometry

The Cauchy-Schwarz inequality \(|\langle u, v\rangle| \leq \|u\|\|v\|\) has a beautiful geometric interpretation: the cosine of the angle between vectors \(u\) and \(v\) in \(\mathbb{R}^n\) satisfies \(|\cos\theta| \leq 1\). This is both the definition of angle in higher dimensions and the constraint that makes geometric angle measure consistent.

For conic sections: the angle between two intersecting conics at a point can be computed using the gradient vectors of their defining equations, via Cauchy-Schwarz — this is essentially what the “angle of intersection” means analytically.

Summary of Key Formulas and Results

For convenient reference, here are the most important formulas from the course: