Topology Primer for EXKNOTS

A self-contained introduction to every topological concept used in the EXKNOTS puzzle series. Written for someone with no formal topology training. If you can tie your shoes and follow an argument, you can read this document.

---

1. What Is Topology? #

Plain-language definition #

Topology is the branch of mathematics that studies properties of shapes that survive stretching, bending, and deforming — but not cutting or gluing. A topologist does not care about distances, angles, or curvature. She cares about connectivity: which parts of a shape are joined to which, how many holes pass through it, whether a loop can be shrunk to a point.

The standard joke is that a topologist cannot tell the difference between a coffee cup and a donut. Here is why that joke is precise:

Both objects have exactly one hole — one passage through which you could thread a string. The cup's hole is the handle; the donut's hole is the obvious one through the middle. A topologist can continuously deform one shape into the other by stretching the cup's body until it merges with the handle, leaving a torus. No cuts, no gluing. This deformation is called a homeomorphism, and it means the two shapes are topologically identical.

Invariants vs. geometry #

Geometry cares about measurements: lengths, angles, curvature. Topology cares about invariants — quantities that do not change under continuous deformation. If you stretch a rubber band into an oval, the circumference changes (geometry), but the fact that it is a single closed loop with no self-crossings does not change (topology).

Every concept in this primer is a topological invariant or is built from one. The puzzles in the EXKNOTS series exploit the gap between what your eyes measure (geometry) and what actually matters (topology). A cord that wraps impressively around a bar may have zero linking number. A twisted strip that looks more complex may actually have a simpler boundary. The recurring lesson: ignore the geometry, find the invariant.

Physical intuition #

When you pick up an EXKNOTS puzzle, you are holding a physical theorem. The metal, wood, and cord embody topological relationships. Your hands perform continuous deformations — sliding a cord, rotating a ring, threading a loop. You cannot cut or glue. You are, physically, doing topology.

Rigorous statement #

A topological space is a set X together with a collection of subsets (called open sets) satisfying certain axioms (unions of open sets are open, finite intersections of open sets are open, the empty set and X are open). A homeomorphism is a continuous bijection whose inverse is also continuous. Two spaces are topologically equivalent (homeomorphic) if a homeomorphism exists between them. A topological invariant is any property preserved by homeomorphisms.

---

2. Curves: Open vs. Closed #

Plain-language definition #

A closed curve (loop) is a curve whose endpoints meet — it has no free ends. A rubber band is a closed curve. An open curve (arc) has two distinct endpoints. A piece of string with two loose ends is an open curve.

This distinction is the single most important idea in the EXKNOTS series. Whether a cord forms a loop or an arc changes what is topologically possible.

Why endpoints change everything #

A closed loop around a post is trapped. It must be cut or the post must be broken. But an open arc draped over the same post can always be slid off one end. The arc's free endpoints provide escape routes that a loop does not have.

Which puzzles use this #

• Puzzle 1, The Gatekeeper: The cord is an arc (both ends fixed to the U-bar), not a loop. This is why the ring can be freed — the cord never truly encircles the bar. Solvers who assume the cord is a closed loop will believe the puzzle is unsolvable.

• Puzzle 8, The Ferryman's Knot: The cord wraps around a post in a trefoil-like pattern with three crossings. If the cord were a closed loop, it would be a genuine trefoil knot — permanently knotted. But the cord is an open arc (one end tied to a ring on the post, the other to a hook in the base). Each wrap can be individually lifted over the finial because the arc's endpoint slides along the post's axis.

Physical intuition #

When you encounter an EXKNOTS puzzle with cord, the very first question to ask is: "Is this cord a loop or an arc?" Trace the cord from one end to the other. If the ends meet (or are spliced together), it is a loop and linking matters. If the ends are separate (tied to different points, or one is free), it is an arc and you may have more freedom than you think.

Rigorous statement #

An arc is the image of a continuous injection from the closed interval [0, 1] into three-dimensional space. A loop (simple closed curve) is the image of a continuous injection from the circle S^1 into three-dimensional space. The fundamental difference: arcs are contractible (can be continuously shrunk to a point), while loops may or may not be contractible depending on the ambient space. In particular, a loop linked with another curve cannot be separated by isotopy, but an arc in the complement of a straight line can always be unlinked.

---

3. Linking Number #

Plain-language definition #

The linking number measures how many times two closed curves wind around each other. It is an integer: positive, negative, or zero. If the linking number is zero, the two curves can be pulled apart. If it is nonzero, they are genuinely linked.

The key insight: crossings have signs. A crossing where curve A passes over curve B from left to right contributes +1. A crossing where A passes over B from right to left contributes -1. The linking number is the sum of all crossing signs, divided by 2.

Worked example: +1 and -1 cancellation #

Consider Puzzle 3, The Prisoner's Ring. A cord loop drapes over a crossbar, creating two visible crossings:

Computing a crossing sign

Here is the recipe the figure is applying:

1. Orient both curves. Draw a travel arrow on the cord and a travel arrow on the crossbar. (Which directions you pick does not matter, as long as you keep them fixed while you count.)
2. At each crossing, look only at the two arrows. Pivot the under-strand's arrow about the crossing point, through the smaller angle (always less than 180 degrees), until it points the same way as the over-strand's arrow.
3. Read off the sign. If that pivot was clockwise, the crossing is +1. If it was counterclockwise, the crossing is -1. (This is the right-hand rule in disguise — equivalently, rotating the over-strand's arrow counterclockwise onto the under-strand's arrow marks the crossing +1.)

Apply it to the figure. At the left crossing, the crossbar's arrow points right and the cord's arrow points down-and-right: aligning the crossbar's arrow with the cord's arrow is a short clockwise pivot, so the sign is +1. At the right crossing, the crossbar's arrow points left and the cord's arrow points down-and-left: the pivot is counterclockwise, so the sign is -1. Sum of signs: +1 + (-1) = 0, and half of that is still 0.

The two crossings cancel perfectly. Despite the visual impression that the cord "wraps around" the crossbar, the linking number is zero. The cord and crossbar can be separated.

Why zero means separable #

The linking number is a topological invariant. If two closed curves have linking number zero, there exists a continuous deformation (isotopy) that pulls them apart without cutting either curve. This is a theorem, not a guess. It means: if you compute linking number zero, a solution exists, period. The puzzle becomes finding the physical moves that realize the mathematical separation.

(A technical caveat: linking number zero is necessary but not sufficient for unlinking in general — there exist links with linking number zero that cannot be separated, such as the Whitehead link. However, for the simple two-component situations in EXKNOTS Puzzles 1 and 3, linking number zero does guarantee separability.)

Which puzzles use this #

• Puzzle 1, The Gatekeeper: The cord's arc has linking number 0 with the U-bar because it is an open arc, not a closed loop. The ring slides free because there is no genuine linking.

• Puzzle 3, The Prisoner's Ring: The cord loop's two crossings with the crossbar have opposite signs, yielding linking number 0. The cord can be freed from the crossbar by pulling a bight over the crossbar's end, after which the ring slides off.

Physical intuition #

When you see a cord wrapped around a bar and your instinct says "it's locked," stop and count crossings. Assign each crossing a sign. If the signs cancel to zero, the lock is an illusion — the cord can be freed. The visual complexity of the wrap is geometric noise; the linking number is the topological signal.

Rigorous statement #

Given two disjoint oriented closed curves C1 and C2 in R^3, project them onto a plane to obtain a link diagram. At each crossing where C1 passes over C2, assign +1 or -1 according to the right-hand rule (the sign of the crossing in the oriented diagram). The linking number lk(C1, C2) is half the sum of these signed crossings. It is an isotopy invariant of the link. Equivalently, lk(C1, C2) equals the degree of the Gauss map (C1 x C2) -> S^2 defined by (x, y) -> (y - x)/|y - x|.

---

4. Genus and Handles #

Plain-language definition #

The genus of a surface is the number of "handles" attached to a sphere. A handle is a tube connecting two points on the surface — it creates a hole you can stick your finger through.

• Genus 0: a sphere (no holes, no handles)
• Genus 1: a torus, i.e., a donut (one hole, one handle)
• Genus 2: a two-holed torus (two holes, two handles)

Think of "handles" literally: a coffee mug has one handle, so it has genus

1. A pot with two handles on its sides has genus 2.

How a hole creates a topological handle #

Take a sphere. Cut out two small disks. Connect the two holes with a tube (a cylinder). You have added one handle, raising the genus by one. The tube provides a new path through the surface — a loop that goes "into" one hole and "out" the other. This loop cannot be shrunk to a point on the surface. Each such irreducible loop corresponds to one generator of the surface's fundamental group.

Which puzzles use this #

• Puzzle 2, Shepherd's Yoke: The wooden paddle with a single hole is a genus-1 object (topologically equivalent to a solid torus). The cord loop is threaded through the handle. The solution exploits the handle: by pushing a bight of cord back through the hole and stretching it over the paddle's edge, you pass the paddle's body through the loop — the handle becomes the escape route, not the trap.

• Puzzle 11, Genus Trap: The acrylic block with two non-intersecting through-tunnels is a genus-2 handlebody. Each tunnel is a handle. The fundamental group of this handlebody is the free group on two generators, F(a, b), where generator a corresponds to a path through Tunnel A and generator b corresponds to a path through Tunnel B. The cord's path encodes the word aba^{-1} in this group. (See Section 6 for details.)

Physical intuition #

When you see a rigid object with holes, count the holes. That is the genus. Each hole is a "handle" that a flexible cord can exploit. Holes are not traps — they are passages. The topology of the object is determined not by its overall shape (which is geometry) but by how many independent paths pass through it.

Rigorous statement #

The genus of a closed orientable surface is the number of handles in a connected sum decomposition with copies of the torus T^2. Equivalently, for a compact orientable surface S without boundary, the genus g satisfies the Euler characteristic formula chi(S) = 2 - 2g. A handlebody of genus g is a 3-manifold homeomorphic to a closed regular neighborhood of a wedge of g circles embedded in R^3. Its boundary is a closed orientable surface of genus g, and its fundamental group is the free group F_g on g generators.

---

5. Orientability and the Mobius Band #

Plain-language definition #

A surface is orientable if it has two distinct sides — an "inside" and an "outside," or a "top" and a "bottom." A sphere is orientable: you can paint the outside blue and the inside red, and the two colors never meet. A cylinder is orientable: the inside surface and the outside surface are separate.

A surface is non-orientable if it has only one side. The most famous example is the Mobius band (also spelled Mobius strip): take a rectangular strip of paper, give it a half-twist (180 degrees), and glue the short edges together.

One-sided vs. two-sided: the edge count #

The edge count tells the story. Compare an ordinary band (cylinder) with a Mobius band:

The ordinary band has two separate edges (top and bottom) and two sides (inner and outer). A cord wedged between the two edges is trapped — it cannot cross from one edge to the other without leaving the band.

The Mobius band has only one edge. If you start tracing along what appears to be the "top" edge, the half-twist carries you to what appears to be the "bottom" edge, and you arrive back at the start after going around twice. Because there is only one edge, a cord that seems to be trapped between "two edges" is actually free to slide along the single continuous boundary until it escapes.

Strip diagram: how the twist changes the boundary #

Which puzzle uses this #

• Puzzle 4, Mobius Snare: A cord loop with a ring is threaded around a leather Mobius band. On an ordinary (untwisted) band, the cord would be trapped between the two edges — genuinely inescapable. But the Mobius band's single edge means the cord can travel continuously from one "face" to the other by following the half-twist. The cord slides along the surface, through the twist, and eventually off the single edge. The half-twist is the solution, not the complication.

Physical intuition #

When you encounter a strip or band in a puzzle, check for twists. An even number of half-twists (0, 2, 4, ...) gives you an orientable band with two edges — a cord between the edges is trapped. An odd number of half-twists (1, 3, 5, ...) gives you a non-orientable band with one edge — a cord can reach any point by following the surface, and escape is possible.

The twist looks like it adds complexity. It does the opposite: it removes a boundary, simplifying the topology.

Rigorous statement #

A surface is orientable if it admits a consistent choice of normal vector at every point — equivalently, if it does not contain a Mobius band as a subspace. The Mobius band is the quotient space [0,1] x [0,1] / ((0, y) ~ (1, 1-y)). It is a compact non-orientable surface with one boundary component (a single closed curve). The boundary of the Mobius band is homeomorphic to S^1 and has a specific embedding in R^3 that wraps twice around the band's core circle. The Euler characteristic of the Mobius band is 0.

---

6. The Fundamental Group #

Plain-language definition #

The fundamental group of a space captures the different ways you can walk in a loop starting and ending at the same point. Two loops are considered "the same" if one can be continuously deformed into the other (without cutting). The fundamental group collects all the genuinely different loop classes and gives them a group structure: you can compose loops (walk one, then the other) and reverse them (walk one backwards).

Generators and relations #

In many spaces, you can describe every possible loop using a small set of building blocks called generators. Each generator is a basic loop that cannot be simplified. Every other loop can be written as a sequence (word) of generators and their inverses.

For example, in a space with two generators a and b:

• The word ab means "walk loop a, then walk loop b."
• The word a^{-1} means "walk loop a backwards."
• The word aba^{-1} means "walk a, then b, then a backwards."

The free group F(a, b) #

When the generators obey NO relations (no equations between them other than the trivial ones like aa^{-1} = identity), the fundamental group is called a free group. The free group on two generators, F(a, b), is the fundamental group of a genus-2 handlebody (a solid block with two tunnels).

In the figure, generator a (red) circles the left post and generator b (blue) circles the right post — the same colors the Genus Trap diagrams use for Tunnel A and Tunnel B. A loop that encircles neither post (green) shrinks to a point: it is the identity.

In a free group, the only way a word simplifies is by cancellation of adjacent inverses:

• aa^{-1} = identity (walking a forward then backward returns you to the start)
• b^{-1}b = identity (same idea)
• aba^{-1} does NOT simplify — the b is "shielded" between a and a^{-1}, preventing any cancellation

This last point is crucial. In an abelian (commutative) group, aba^{-1} would simplify to b (because you could swap the order). But the free group is non-abelian: the order of generators matters, and you cannot move b past a or a^{-1}.

Worked example: aba^{-1} in the genus-2 handlebody (Puzzle 11) #

Puzzle 11, Genus Trap, has an acrylic block with two tunnels:

• Tunnel A (left-to-right) contributes generator a
• Tunnel B (front-to-back) contributes generator b

The cord's path:

1. Through Tunnel A, left to right = a
2. Through Tunnel B, front to back = b
3. Through Tunnel A, right to left = a^{-1}

The cord encodes the word aba^{-1}.

Why aba^{-1} is not the identity: In the free group F(a, b), the only simplification rule is cancellation of adjacent inverse pairs. In aba^{-1}, the adjacent pairs are (a, b) and (b, a^{-1}). Neither pair consists of a generator and its inverse. The b in the middle blocks the a and a^{-1} from meeting and canceling. Therefore aba^{-1} cannot be reduced. It is a non-trivial element, meaning the cord is genuinely tangled with the block's topology.

Why aa^{-1} = identity: If you remove the b (physically, by rerouting the cord so it no longer passes through Tunnel B), you are left with aa^{-1}. Here, a and a^{-1} are adjacent and cancel immediately, giving the identity element. The cord is now topologically trivial — it is not tangled with anything, and the rings slide off.

The solution: Pull the cord segment that passes through Tunnel B back out and reroute it through Tunnel A. This transforms the word from aba^{-1} to a(aa^{-1})a^{-1} = aa^{-1} = identity. The rings are free.

Which puzzles use this #

• Puzzle 7, Devil's Pitchfork: The configuration space of the ring on the three-pronged fork has a non-trivial fundamental group. The solution requires the cord to trace a specific non-contractible loop (over the center prong) before the ring can be transferred. The loop represents a non-trivial element of the fundamental group of the configuration space.

• Puzzle 11, Genus Trap: The full worked example above. The free group F(a, b) is the fundamental group of the genus-2 handlebody, and the cord's path is a word in this group.

Physical intuition #

Think of the fundamental group as an accounting system for tangles. Each tunnel or handle you thread through is a "letter" in a word. Threading through and then back (same tunnel, opposite direction) cancels the letter. Threading through different tunnels stacks up letters that cannot cancel with each other. To free a cord, you must manipulate it until the word reduces to nothing — the identity element.

When you are stuck on Puzzle 11, write down the word. Every move you make with the cord changes the word. If the word is getting longer, you are going the wrong direction. If adjacent inverses appear, you are making progress.

Rigorous statement #

The fundamental group pi_1(X, x_0) of a topological space X at a basepoint x_0 is the set of homotopy classes of loops based at x_0, equipped with the operation of concatenation. A free group F(S) on a generating set S is a group where every element has a unique reduced representation as a finite word in S and S^{-1} (letters and their formal inverses), with the only relation being cancellation of adjacent inverse pairs. For a genus-g handlebody H_g, pi_1(H_g) is isomorphic to F_g, the free group on g generators. This is because H_g deformation-retracts onto a wedge of g circles.

---

7. Borromean and Brunnian Links #

Plain-language definition #

Borromean rings are three closed curves that are mutually linked — the three of them hold together — but no two of them are linked to each other. Remove any one ring, and the other two fall apart.

This is deeply counterintuitive. We expect linking to be a pairwise relationship: A is linked to B, B is linked to C, and that is why the three hold together. Borromean rings violate this expectation. The linking is a purely three-body phenomenon that cannot be reduced to pairwise interactions.

Classic Borromean rings diagram #

In the diagram, the three rings are interlocked in a cyclic over-under pattern:

• Ring A passes over Ring B
• Ring B passes over Ring C
• Ring C passes over Ring A

No two rings are linked (linking number = 0 for every pair), but the three together cannot be pulled apart.

Brunnian links: the general case #

A Brunnian link is a link of n components where removing any single component makes the remaining n-1 components completely unlinked. Borromean rings are the simplest Brunnian link (n = 3). Brunnian links exist for any n >= 3.

Milnor's invariant: higher-order linking #

If linking number (a pairwise invariant) is zero for every pair, what detects the Borromean property? The answer is Milnor's invariant, a higher-order linking invariant. While the ordinary linking number counts how two curves wind around each other, Milnor's invariant captures how three (or more) curves interact collectively.

For Borromean rings:

• Pairwise linking numbers: lk(A,B) = lk(B,C) = lk(A,C) = 0
• Milnor's triple linking number: mu(A,B,C) = +/-1 (nonzero)

The nonzero Milnor invariant is the mathematical certificate that the three rings are collectively linked despite being pairwise unlinked.

Which puzzle uses this #

• Puzzle 6, Trinity Lock: Three identical steel ovals must be assembled into the Borromean configuration. The puzzle is an assembly challenge: the solver must weave all three simultaneously because no two are ever linked at any intermediate stage. The natural instinct — "connect two first, then add the third" — fails because there is no pairwise connection to build on. The Borromean property forces a fundamentally non-incremental approach.

Physical intuition #

Hold three rubber bands. Try to arrange them so they hold together as a cluster, but any one you remove lets the other two fall apart. You will discover that you cannot assemble them incrementally — you must weave all three at once, maintaining the cyclic over-under pattern throughout. This feeling of "I cannot build this step by step" is your hands discovering that the Borromean property is irreducibly collective.

Rigorous statement #

A Borromean link is a 3-component link L = L_1 ∪ L_2 ∪ L_3 in S^3 such that each 2-component sublink is the unlink, but L itself is not the unlink. More generally, a Brunnian link of n components is an n-component link such that every proper sublink is trivial. The non- triviality of Borromean rings is detected by Milnor's mu-bar invariant mu_bar(1,2,3), a higher-order invariant derived from the lower central series of the link group. For the standard Borromean rings, |mu_bar(1,2,3)| = 1.

---

8. Configuration Spaces #

Plain-language definition #

A configuration space is the space of all possible states of a system. Each point in the configuration space represents one specific arrangement of all the parts. As you manipulate a puzzle, the state traces a path through the configuration space.

A solution to a puzzle is a path in the configuration space from the starting state to the goal state. If no such path exists, the puzzle is unsolvable. If the path exists but must navigate around obstacles (holes, barriers), the configuration space has non-trivial topology.

Concrete example #

Consider a ring on a prong. The ring's state is determined by its height on the prong and its rotation. The configuration space might be a simple line segment (just height) — or it might be more complex if the ring is connected to a cord that constrains its movement. Barriers in the configuration space (ball-stops, cord length limits) create holes and walls that the state-path must navigate around.

flowchart LR
    subgraph Physical["Physical Space"]
        direction TB
        BS["O ball-stop"] --- Post["| post"] --- Base["+ base"]
        Post -.- Note1["ring slides\nup and down"]
    end
    subgraph Config["Configuration Space"]
        direction TB
        Goal(("goal\n(ring on right prong)"))
        Path["path must wind through\nnon-trivial topology"]
        Start(("start\n(ring on left prong)"))
        Goal ~~~ Path ~~~ Start
    end

Non-trivial topology of the state space #

The key idea: the configuration space itself can have holes, handles, and non-contractible loops — the very same topological features described elsewhere in this primer. When the configuration space has a non-trivial fundamental group, certain sequences of moves (loops in the configuration space) cannot be contracted to a point. This means some rearrangements require the system to pass through specific intermediate states — there are no shortcuts.

Which puzzle uses this #

• Puzzle 7, Devil's Pitchfork: A ring sits on the left prong of a three-pronged fork. A cord connects it to the base of the center prong. The goal is to move the ring to the right prong. The configuration space of this system has a non-trivial fundamental group because of how the cord, the ring, and the three prongs interact. The solution requires first reconfiguring the cord (looping it over the shorter center prong) to change the topology of the accessible configuration space, and only then transferring the ring. The solver must change the constraints before moving the constrained object — a meta-level insight that arises directly from the configuration space's topology.

Physical intuition #

When a puzzle feels like it "should" be solvable but every direct approach fails, you may be encountering non-trivial configuration-space topology. The accessible states are not simply connected — there are holes in the space of possibilities. You must navigate around these holes, which sometimes means making moves that feel like going backward (they are actually navigating around an obstacle in the configuration space).

Rigorous statement #

The configuration space C(X, n) of n particles on a space X is the subspace of X^n obtained by requiring all particles to be distinct (or, in constrained settings, by imposing mechanical constraints). More generally, for a mechanical system with parts P_1, ..., P_k subject to constraints, the configuration space is the submanifold of the product of all individual state spaces satisfying all constraint equations. The topology of this space (in particular, its fundamental group and higher homotopy groups) governs which state transitions are possible and which sequences of moves are topologically necessary.

---

9. Gray Codes and Recursive Complexity #

Plain-language definition #

A Gray code is a sequence of binary numbers in which consecutive entries differ by exactly one bit. The standard binary sequence 000, 001, 010, 011, ... has the problem that some consecutive entries differ by multiple bits (e.g., 011 to 100 changes all three bits). A Gray code avoids this: each step flips exactly one bit.

Standard binary: some consecutive pairs differ by multiple bits. Gray code: every consecutive pair differs by exactly one bit.

Why the sequence is optimal #

Gray codes are not just a convenient ordering — they are the unique minimal solution to certain sequential puzzles. The puzzle constraints determine which bit can be flipped at each step, and these constraints force the Gray code ordering. There is no shorter sequence. Any attempt to take a "shortcut" (flip multiple bits at once, or flip a different bit out of order) violates the physical constraints.

Binary recursion #

The Gray code has a beautiful recursive structure. To construct the n-bit Gray code from the (n-1)-bit Gray code:

1. Take the (n-1)-bit sequence
2. Write it forward, prefixing each entry with 0
3. Write it backward, prefixing each entry with 1
4. Concatenate

This recursion mirrors the structure of the Chinese Rings puzzle: to remove ring k, you must first set up a specific configuration of rings k+1 through n, which requires the same kind of recursive setup.

Which puzzle uses this #

• Puzzle 10, Ouroboros Chain: Six cord loops on posts, each threaded through its neighbor, with a shuttle bar through all of them. This is a reimagining of the Baguenaudier (Chinese Rings). Each loop is either ON (1) or OFF (0) the shuttle bar, giving a 6-bit state. The rules for which loop can be toggled at each step mirror the Gray code constraints exactly. The minimum solution requires 42 physical manipulations. No shortcuts exist — the recursive structure of the Gray code is the irreducible minimum.

Physical intuition #

The Ouroboros Chain teaches patience and trust. Each move is simple (toggle one loop on or off the shuttle bar), but the sequence is long and counterintuitive — you frequently "undo" progress by replacing loops you already removed. This feels wrong but is mathematically necessary: the recursive structure requires setting up configurations that look like backward steps but are actually preconditions for the next forward step.

If you are losing track, write down the state as a 6-bit binary number after each move. The pattern will become visible: you are walking through the reflected binary (Gray) code, one bit-flip at a time.

No shortcuts in topology #

The Ouroboros Chain embodies a deep principle: some topological puzzles have irreducible sequential complexity. There is no "aha!" moment, no single clever trick. The solution is an algorithm that must be followed completely. The topology of the state space (the Gray code graph) determines the minimum number of moves, and no amount of cleverness can reduce it.

Rigorous statement #

An n-bit Gray code is a Hamiltonian path on the n-dimensional hypercube graph Q_n (whose vertices are the 2^n binary strings of length n, with edges between strings differing in exactly one bit). The reflected binary Gray code (RBGC) is a specific recursive construction. The Baguenaudier (Chinese Rings) puzzle with n rings has a state graph isomorphic to a path graph whose vertex sequence is the RBGC. The minimum number of moves to reach the goal state (all rings off) from the initial state (all rings on) is (2^n + 1)/3 for n odd and (2^n - 1)/3 for n even (counting state transitions; physical manipulations may be higher).

---

10. Fiber Bundles and the Hopf Fibration #

Plain-language definition #

Imagine a space made up of a collection of identical "fibers" (think threads), all organized by a "base space." Each point in the base space has one fiber hanging over it. If you could separate the fibers cleanly — like untwisting a cable into its individual strands — the total space would just be the product of the base and the fiber. But in a non-trivial fiber bundle, the fibers are twisted together in a way that makes global separation impossible.

S^1, S^2, S^3: accessible definitions #

• S^1 (the 1-sphere): the circle. The set of all points at distance 1 from the origin in the plane (2D). It is a one-dimensional curve that closes on itself.

• S^2 (the 2-sphere): the ordinary sphere — the surface of a ball. The set of all points at distance 1 from the origin in 3D space. It is a two-dimensional surface. (Note: we mean only the surface, not the solid ball inside.)

• S^3 (the 3-sphere): a "sphere in 4D." The set of all points at distance 1 from the origin in 4-dimensional space. We cannot visualize S^3 directly (it lives in 4D), but we can reason about it mathematically. It is a three-dimensional manifold — locally it looks like ordinary 3D space, but globally it wraps around and closes on itself, just as S^2 (a 2D surface) closes on itself in 3D.

The Hopf map #

The Hopf fibration is a specific map h: S^3 -> S^2 that sends each point of the 3-sphere to a point on the 2-sphere. The key property: the preimage of each point on S^2 is a circle (S^1) inside S^3. These circles are the "fibers."

So the Hopf fibration decomposes S^3 into a family of circles, one for each point on S^2. These circles are not arbitrary — they are linked in a specific, beautiful way:

• Any two distinct fibers are linked (linking number 1)
• No fiber can be continuously deformed to a point in S^3 minus any other fiber
• The fibers twist around each other as you move across S^2

This structure makes S^3 a non-trivial fiber bundle: S^1 fibers over S^2 base, but the total space is NOT the simple product S^2 x S^1 (which would be a different space entirely). The twist is essential.

Why fibers twist: the 2:1 rotation #

The Hopf fibration arises naturally from the relationship between rotations in 3D and points on S^3. (The group of unit quaternions, which represents 3D rotations, is exactly S^3.) Moving along a Hopf fiber corresponds to rotating simultaneously about two orthogonal axes in a 2:1 ratio — one full rotation about one axis for every half rotation about the other.

This coupled rotation cannot be decomposed into sequential single-axis rotations. It is an irreducibly two-axis motion.

The belt trick / plate trick: an accessible analogy #

Hold a belt by one end, with the other end fixed. Give the belt a full twist (360 degrees). The belt is twisted and cannot be untwisted by any manipulation that keeps the ends fixed. Now give it another full twist in the same direction (720 degrees total). Remarkably, you can now untwist the belt by passing it around one end — the double twist is equivalent to no twist at all.

The belt trick is a physical demonstration that the space of 3D rotations is not simply connected: pi_1(SO(3)) = Z/2, realized by the double cover SU(2) -> SO(3). A 360-degree rotation traces a loop in SO(3) that cannot be contracted, but traversing it twice (720 degrees) gives a contractible loop. The connection to the Hopf fibration is close but not identical: SU(2) is the same 3-sphere S^3 the Hopf fibration lives on, but the fibration (S^3 -> S^2, with circle fibers) and the double cover (S^3 -> SO(3), with two-point fibers) are different maps — the belt trick illustrates the double cover, and serves as an analogy for the puzzle's coupled motion rather than an instance of the Hopf map.

The plate trick is similar: hold a plate flat on your palm. Rotate it 360 degrees (keeping it level) by twisting your arm — your arm is now twisted. Rotate it another 360 degrees in the same direction — and by looping your arm around, you can untwist back to the start. Filipino waiters use this daily.

Which puzzle uses this #

• Puzzle 12, The Hopf Paradox: A ring is trapped inside a cage made of two orthogonal great-circle hoops. The ring cannot be extracted by any sequence of single-axis rotations. At the pole (where the two hoops intersect), the ring must execute a corkscrew motion: simultaneous rotation and translation in a fixed ratio, corresponding to motion along a Hopf fiber. The solver must physically discover the coupled rotation — a genuinely new motor skill. The puzzle is unique in the series because understanding the solution conceptually is not sufficient; the coupled-rotation motor skill must also be developed.

Physical intuition #

When you hold the cage from Puzzle 12 and try to extract the ring at the pole junction, your hands will naturally attempt sequential moves: rotate the ring, then push it forward, then rotate again. This fails. The junction geometry requires both motions simultaneously. The moment you find the corkscrew — the smooth spiral that threads the ring between the two hoop wires — you have physically experienced a Hopf fiber. It feels like threading a nut onto a bolt: the rotation and the forward motion are coupled and cannot happen separately.

Rigorous statement #

A fiber bundle is a structure (E, B, F, pi) where E is the total space, B the base space, F the fiber, and pi: E -> B a continuous surjection such that each point b in B has a neighborhood U with pi^{-1}(U) homeomorphic to U x F (local triviality). The bundle is trivial if the total space is globally homeomorphic to B x F; otherwise it is non-trivial. The Hopf fibration is the fiber bundle h: S^3 -> S^2 with fiber S^1, defined in complex coordinates by h(z_1, z_2) = [z_1 : z_2] (viewing S^2 as CP^1). It is non-trivial: S^3 is not homeomorphic to S^2 x S^1. The Hopf fibration is classified by the generator of pi_3(S^2) = Z, discovered by Heinz Hopf in 1931.

---

11. Knot Theory Basics #

Plain-language definition #

Knot theory studies closed curves in three-dimensional space — specifically, how they are tangled. A knot is a single closed curve (loop) embedded in 3D. A link is a collection of two or more closed curves. Two knots (or links) are considered "the same" if one can be continuously deformed into the other without cutting.

Knots vs. links #

The unknot #

The unknot is the simplest knot: a closed curve with no crossings — a plain circle. Any knot that can be deformed into the unknot is called "trivially knotted" or just "unknotted."

Crossing number #

The crossing number of a knot is the minimum number of crossings in any diagram of the knot. The unknot has crossing number 0. The trefoil has crossing number 3. The figure-eight knot has crossing number 4.

Reidemeister moves #

Any deformation of a knot in 3D can be represented by a sequence of three local diagram changes called Reidemeister moves:

If two knot diagrams represent the same knot, there is a finite sequence of Reidemeister moves transforming one diagram into the other. These three moves are complete: they capture all possible continuous deformations.

Open knots vs. closed knots (the critical distinction) #

Classical knot theory studies closed curves. An open arc (with two free endpoints) is never really "knotted" in the classical sense — it can always be unknotted by sliding the tangles off the free ends. This is why the distinction between open and closed curves (Section 2) is so important for the EXKNOTS puzzles.

However, an open arc with constrained endpoints (e.g., one end tied to a post, the other to a hook) behaves differently from a free arc. The constraints limit which Reidemeister moves are available. In Puzzle 8, each wrap corresponds to a Type I Reidemeister move (twist removal), and the finial provides the mechanism for executing it.

Which puzzles use this #

• Puzzle 1, The Gatekeeper: The cord is an unknotted open arc. The solver must recognize that despite the visual wrapping, the cord has crossing number 0 with respect to the U-bar.

• Puzzle 3, The Prisoner's Ring: The cord loop and crossbar form a two-component link. The linking number (computed from crossing signs) is zero, so the link is trivial — the components can be separated.

• Puzzle 8, The Ferryman's Knot: The cord wraps around a post in a trefoil-like pattern (3 crossings). If the cord were a closed loop, this would be a genuine trefoil (crossing number 3, not unknottable). But the cord is an open arc on a fixed axis, and each crossing can be removed by a Type I Reidemeister move — lifting the cord loop over the finial. Three Reidemeister moves, and the cord hangs free.

Physical intuition #

When you look at a tangled cord in an EXKNOTS puzzle, draw the knot diagram: project the 3D arrangement onto a flat surface, marking each crossing as "over" or "under." Then check:

1. Is the curve closed or open?
2. If closed, what is the linking number with other components?
3. If open, can Reidemeister moves (executed via the puzzle's physical mechanisms) simplify the diagram?

Knot diagrams are tools, not abstract art. Draw them on paper. Mark the crossings. Count the signs. The diagram will tell you whether the puzzle is solvable and often suggest how.

Rigorous statement #

A knot is an embedding of S^1 into S^3 (or R^3), considered up to ambient isotopy. A link is an embedding of a disjoint union of copies of S^1. Two knots are equivalent if there is an ambient isotopy of S^3 carrying one to the other. Reidemeister's theorem (1927) states that two knot diagrams represent equivalent knots if and only if they are related by a finite sequence of Reidemeister moves (types I, II, III) and planar isotopy. The crossing number c(K) of a knot K is the minimum number of crossings over all diagrams of K. The unknot U satisfies c(U) = 0. The trefoil T satisfies c(T) = 3 and is the unique prime knot with this crossing number.

---

12. Chirality and Handedness #

Plain-language definition #

A knot is chiral if it is not equivalent to its mirror image. The simplest example: the trefoil knot comes in two versions — left-handed and right-handed — that are topologically distinct. No continuous deformation in 3D space can convert one into the other.

A knot that IS equivalent to its mirror image is called amphichiral (or achiral). The figure-eight knot is amphichiral — its mirror image can be deformed back to the original.

How to detect chirality #

At each crossing, follow the knot in a consistent direction. If the overpasses spiral clockwise, the trefoil is right-handed. If counterclockwise, left-handed. More formally, chirality can be detected by knot polynomials: the Jones polynomial of a chiral knot differs from the Jones polynomial of its mirror image.

Worked example: reading handedness crossing by crossing #

The crossing-sign rule from Section 3 turns "which hand is this?" into arithmetic. Orient the knot: pick a starting point and draw a travel arrow along the curve. Then visit each of the trefoil's three crossings and apply the sign rule — pivot the under-strand's arrow through the smaller angle onto the over-strand's arrow; clockwise means +1, counterclockwise means -1 (equivalently, as the figure's inset shows, over-strand arrow rotated counterclockwise onto the under-strand arrow means +1).

On the right-handed trefoil every crossing comes out +1: the read-off is

• • +, and the sum of signs — called the writhe of the diagram — is +3. Reflect the diagram in a mirror and every over-strand becomes an under-strand: each sign flips, the read-off is - - -, and the writhe is -3. Reversing the travel arrow does NOT change the answer: both arrows at a crossing reverse together, and the pivot direction survives. So the read-off is independent of the orientation you happened to choose.

One honest caveat: writhe is a property of the diagram, not of the knot — a Type I Reidemeister move adds a kink and changes the writhe by

1. But for reduced alternating diagrams like these minimal 3-crossing trefoil diagrams, the writhe is the same in every such diagram (one of the Tait conjectures, proved in the 1980s). So +3 versus -3 is a genuine, checkable difference between the two trefoils' standard diagrams — it is the number your hands compute when you trace a wire frame in Puzzle 5.

Detection by the Jones polynomial #

The Jones polynomial V_K(t) converts mirror reflection into an algebraic substitution: the mirror image m(K) satisfies V_{m(K)}(t) = V_K(t^{-1}). So if V_K is not symmetric under swapping t and t^{-1}, the knot cannot equal its mirror image. For the right-handed trefoil, V(t) = -t^4 + t^3 + t; substituting t -> t^{-1} gives -t^{-4} + t^{-3} + t^{-1}, the left-handed trefoil's polynomial. The two are different, so the trefoil is chiral — a three-line proof of a fact that resisted proof until Max Dehn's 1914 argument. The figure-eight knot's polynomial, V(t) = t^2 - t + 1 - t^{-1} + t^{-2}, is palindromic (unchanged by t -> t^{-1}), exactly as an amphichiral knot's must be.

Which puzzles use this #

• Puzzle 5, The Mirror Gate: Two trefoil frames — one left-handed, one right-handed — must be matched to mirror-image recesses. The solver must identify the handedness of each trefoil by examining its crossing pattern.

Physical intuition #

Hold both trefoils side by side. They look identical until you try to seat one in the other's recess — it simply does not fit. The physical mismatch at the crossing points IS the chirality. Your hands can feel the difference that your eyes initially miss. This is exactly the situation in chemistry, where chiral molecules (enantiomers) have identical properties in isolation but interact differently with other chiral structures.

Rigorous statement #

A knot K in S^3 is chiral if there is no ambient isotopy carrying K to its mirror image m(K), where m is any orientation-reversing homeomorphism of S^3 — concretely, reflection through a plane, (x, y, z) -> (x, y, -z). K is amphichiral if K and m(K) are ambient-isotopic. The trefoil is chiral (Dehn, 1914); the figure-eight knot is amphichiral — an explicit sequence of Reidemeister moves carries its mirror diagram back to the original. Mirror reflection reverses the sign of every crossing, so it negates the writhe of a diagram; by the Tait writhe theorem this already distinguishes the two trefoils' reduced alternating diagrams. In general, chirality is certified by the Jones polynomial via V_{m(K)}(t) = V_K(t^{-1}): asymmetry of V_K under t -> t^{-1} implies chirality (the converse fails — some chiral knots have symmetric Jones polynomials, so a symmetric V proves nothing).

---

13. Braid Groups #

Plain-language definition #

A braid on n strands is a set of n non-intersecting curves that may cross over and under each other. The braid group B_n is the set of all such braids, with "stacking" (composition) as the group operation.

For three strands (B_3), the two generators are:

• sigma_1: strand 1 crosses over strand 2
• sigma_2: strand 2 crosses over strand 3

These generators do NOT commute: sigma_1 * sigma_2 ≠ sigma_2 * sigma_1.

The Artin presentation #

Emil Artin showed in 1925 that B_n is completely described by n-1 generators and just two families of relations. The generators are sigma_1, ..., sigma_{n-1}, where sigma_i crosses strand i over strand i+1. The relations:

• Far commutation: sigma_i * sigma_j = sigma_j * sigma_i whenever |i - j| >= 2. Swaps involving disjoint pairs of strands do not interfere — swapping strands 1-2 and swapping strands 3-4 can happen in either order.
• Braid relation: sigma_i * sigma_{i+1} * sigma_i = sigma_{i+1} * sigma_i * sigma_{i+1}. Adjacent swaps that share a strand obey exactly one exchange law — the Yang-Baxter relation discussed next.

Everything true about braids follows from these two rules. Notice what is missing: there is no relation sigma_i * sigma_i = identity. Swapping the same pair of strands twice does not undo the swap — it adds a full twist to the cords. This is precisely where B_n parts ways with the group of mere permutations, and it is why B_n is an infinite group.

The Yang-Baxter relation #

The key algebraic relation in B_3 is: sigma_1 * sigma_2 * sigma_1 = sigma_2 * sigma_1 * sigma_2. This is the braid relation (Yang-Baxter equation). It says that two specific three-step sequences of swaps produce the same braid.

Worked example: sigma_2-then-sigma_1 vs sigma_1-then-sigma_2 #

Puzzle 13, The Braid Cage, starts with rings Blue - Yellow - Red on posts 1-2-3 and asks for Red - Blue - Yellow. Compare the two shortest candidate words, applying the left factor first (sigma_1 swaps the rings on posts 1 and 2; sigma_2 swaps posts 2 and 3):

The same two generators, applied in opposite orders, land the rings on different posts. This is non-commutativity at the coarsest possible level: the two words do not even agree as permutations, let alone as braids.

From braids to permutations #

Forgetting which strand crossed over which turns a braid into a plain permutation of the n positions. This forgetting map B_n -> S_n is a surjective group homomorphism: every permutation is achievable by some braid. Its kernel — the braids that permute nothing, yet are not trivial — is the pure braid group P_n. A non-trivial pure braid is exactly the Braid Cage's characteristic failure mode: every ring back on its target post, but the cords wound around each other. The rings see only the permutation; the cords remember the whole braid.

From braids to knots #

Joining the top of each strand to the bottom of the same position closes a braid into a knot or link. Torus knots arise this way: T(p, q) is the closure of the word (sigma_1 * sigma_2 * ... * sigma_{p-1})^q. The trefoil T(2, 3) is the closure of sigma_1^3 in B_2 — three identical swaps of two strands, sealed shut. Section 14 picks up this thread.

Which puzzles use this #

• Puzzle 13, The Braid Cage: Three rings on posts connected by cords. The cords record the history of swaps. Only braid-relation sequences leave the cords untangled. The solver feels non-commutativity directly — wrong swap orders tangle the cords.

Physical intuition #

When you swap two rings by lifting one over a post finial, the connecting cord records the swap as a braid generator. Swapping in a different order produces a different braid — and different braids leave the cords in different states (tangled vs. untangled). The order of operations has physical consequences.

Rigorous statement #

The braid group B_n is the group with the Artin presentation: generators sigma_1, ..., sigma_{n-1}, subject to sigma_i sigma_j = sigma_j sigma_i for |i - j| >= 2 and sigma_i sigma_{i+1} sigma_i = sigma_{i+1} sigma_i sigma_{i+1} for 1 <= i <= n-2. Equivalently, B_n is the fundamental group of the configuration space of n unordered points in the plane (Section 8 again: a braid is a loop of point configurations, traced out in time). Imposing the extra relations sigma_i^2 = 1 collapses B_n onto the symmetric group S_n; the quotient map B_n -> S_n sends each braid to its endpoint permutation, and its kernel is the pure braid group P_n, giving the short exact sequence 1 -> P_n -> B_n -> S_n -> 1. By Alexander's theorem (1923), every knot and link in R^3 is the closure of some braid.

---

14. Torus Knots #

Plain-language definition #

A (p,q) torus knot lies on the surface of a torus, winding p times around the central axis (the long way) and q times around the tube — each tube wrap passes through the hole. The curve closes up into a single connected knot when gcd(p,q) = 1.

Key examples:

• (1, q) for any q → unknot
• (2, 2) → two-component link
• (2, 3) → trefoil (simplest torus knot, 3 crossings)
• (2, 5) → Solomon's seal knot (5 crossings)
• (3, 4) → torus knot with 8 crossings

The rule: (p,q) with gcd(p,q) = 1 and p,q ≥ 2 produces a genuine knot.

Parametrization on the standard torus #

Put the torus in standard position: a tube of radius r whose center circle has radius R. The (p,q) torus knot is traced by a point that circles the hole p times while circling the tube q times, both at constant rates:

x(t) = (R + r cos qt) cos pt, y(t) = (R + r cos qt) sin pt, z(t) = r sin qt, for t from 0 to 2 pi.

The curve lives entirely on the torus surface — it never tunnels through the solid part. A surprising symmetry: T(p,q) and T(q,p) are the same knot. The (2,3) curve and the (3,2) curve look different on the torus (their standard flat projections show 3 and 4 crossings respectively), yet an isotopy of the ambient space — turning the torus inside out through its own hole, so that hole-circles and tube-circles trade roles — carries one onto the other. Both are the trefoil.

Worked example: tracing the (2,3) winding #

Follow the cord of Puzzle 14, The Torus Winder, condensed to its topological skeleton:

1. Circle the central axis once, the long way around (first of p = 2), wrapping around the tube as you go — each tube wrap passes through the hole.
2. Circle the axis again (second of p = 2), still wrapping the tube.
3. Complete the third tube wrap (q = 3) — and the cord arrives back at its starting point, pointing the way it began.

The cord closes into a single curve because gcd(2,3) = 1: since 2 and 3 share no common factor, the strand cannot return to its start until it has spent all of its winding in both directions at once. Wind (2,2) instead and the curve closes early, after half the journey, leaving two separate loops — a link, and the ring escapes between them. Now count the crossings of the completed (2,3) winding: exactly 3, all of the same handedness. That matches the general formula below: the crossing number of T(p,q) is min(p(q-1), q(p-1)) = min(4, 3) = 3. The trefoil again — built this time by winding rather than knotting.

Two more numbers the winding determines #

The pair (p,q) does not just decide whether the curve is knotted; it determines the knot's key invariants by closed formulas:

• Genus (Section 16): (p-1)(q-1)/2. For (2,3): 1.
• Crossing number: min(p(q-1), q(p-1)). For (2,3): 3.
• Unknotting number (Section 17): (p-1)(q-1)/2 — the same number as the genus. For (2,3): 1.

Which puzzles use this #

• Puzzle 14, The Torus Winder: A cord must be wound around a torus following guide notches to create the (2,3) torus knot. Most windings fail to trap a sliding ring — only the correct (p,q) pair produces a genuine knot.

• Puzzle 17, The Satellite Trap: The internal tunnel of the torus shell follows a (2,3) torus knot path, forming the companion knot of the satellite structure.

Physical intuition #

Wind a cord around a donut-shaped ring. If you go around the central axis twice (the long way) while wrapping around the tube three times — each tube wrap passing through the hole — the cord crosses itself exactly three times and cannot be unwound without cutting. Change the numbers and the knot either simplifies to a circle or splits into multiple loops. The relationship between the two winding numbers is what determines knottedness.

Rigorous statement #

For integers p, q, the torus knot/link T(p,q) is the image of the curve t -> ((R + r cos qt) cos pt, (R + r cos qt) sin pt, r sin qt) on the standard torus in R^3 — equivalently, the image of t -> (e^{ipt}, e^{iqt}) on the Clifford torus in S^3. T(p,q) is ambient-isotopic to T(q,p). If gcd(p,q) = d > 1 the curve closes into a link of d parallel copies of T(p/d, q/d) rather than a knot; if gcd(p,q) = 1 it is a knot, and it is the unknot exactly when |p| <= 1 or |q| <= 1. For coprime p, q >= 2: the genus is g(T(p,q)) = (p-1)(q-1)/2, realized by Seifert's algorithm on the standard diagram (Section 16); the crossing number is c(T(p,q)) = min(p(q-1), q(p-1)) (Murasugi, 1991); and the unknotting number is u(T(p,q)) = (p-1)(q-1)/2 — the Milnor conjecture, proved by Kronheimer and Mrowka in 1993 (Section 17). T(p,q) is the closure of the braid word (sigma_1 sigma_2 ... sigma_{p-1})^q in B_p (Section 13).

---

15. Knot Coloring and Tricolorability #

Plain-language definition #

A Fox 3-coloring of a knot diagram assigns one of three colors to each arc (strand between consecutive undercrossings) such that at every crossing, the three meeting arcs are either all the same color or all different colors.

A knot is tricolorable if it admits a non-trivial Fox 3-coloring (one using more than one color). Tricolorability is a topological invariant — it is preserved under Reidemeister moves.

Key facts:

• The unknot is NOT tricolorable
• The trefoil IS tricolorable
• The figure-eight IS NOT tricolorable
• Since the unknot and trefoil differ in tricolorability, they are distinct knots

Worked example: checking the trefoil crossing by crossing #

Label the trefoil's three arcs a1, a2, a3 — an arc runs from one undercrossing to the next — and color them red, blue, and yellow, as in the left panel of the figure. In the standard diagram, each crossing brings together all three arcs: one passes over, and the other two are the under-strand's incoming and outgoing segments. Check the Fox rule at every crossing:

Every crossing passes, and the coloring uses more than one color, so the trefoil is tricolorable. The right panel of the figure shows why two colors cannot work: with only two colors in play, some crossing must see two arcs of one color and one arc of the other — neither "all same" nor "all different." That two-and-one configuration is exactly the forbidden one.

The arithmetic form of the rule #

Number the colors 0, 1, 2 and work mod 3. The Fox condition at a crossing with over-arc color o and under-arc colors u1, u2 becomes one linear equation:

2·o ≡ u1 + u2 (mod 3)

If all three colors are equal, both sides agree. If all three are different, then u1 + u2 is the sum of the two colors other than o, and since 0 + 1 + 2 ≡ 0 (mod 3), that sum is -o ≡ 2o (mod 3) — valid again. But two-same-one-different always fails. The whimsical-looking coloring rule is secretly a system of linear equations in mod-3 arithmetic: one equation per crossing, one unknown per arc.

Counting colorings #

Solve that system for the trefoil: there are exactly 9 colorings — 3 trivial ones (all arcs red, all arcs blue, all arcs yellow) plus 6 nontrivial ones, the 3! = 6 ways to give the three arcs three distinct colors. For the unknot — a diagram with a single arc and no crossings — only the 3 trivial colorings exist. The total count of 3-colorings is a knot invariant and is always a power of 3. Nine is bigger than three: the trefoil admits colorings the unknot cannot, so the trefoil is not the unknot. This is the simplest complete proof that a genuinely knotted curve exists.

Why Reidemeister moves preserve the count #

Sketch for a Type I move (adding a kink): the kink creates one new crossing at which the over-arc and one under-arc are the same strand, carrying some color x. The equation 2x ≡ x + u2 (mod 3) forces u2 ≡ x — the arc on the far side of the kink must wear the same color as the old one. So colorings of the kinked diagram correspond one-to-one with colorings of the plain diagram, and the count is unchanged. Type II and Type III moves yield to the same kind of bookkeeping, and together the three checks prove that the coloring count — in particular, tricolorability — is a topological invariant.

Which puzzles use this #

• Puzzle 15, The Tricolor Lock: The solver must find the valid Fox 3-coloring of a trefoil frame. Valid coloring reveals a physical passage (aligned notches) that frees a trapped ring. Invalid coloring leaves the ring trapped.

Physical intuition #

Color the three arcs of a trefoil with three distinct colors. At each crossing, check: are all three colors different? If yes at every crossing, the coloring is valid. This algebraic rule has a physical consequence in the puzzle: the colored sleeves have notches that align only under a valid coloring. The invariant is not just a number — it changes the puzzle's physical geometry.

Rigorous statement #

A Fox 3-coloring of a knot diagram D is a function c from the arcs of D to Z/3 satisfying 2c(o) - c(u1) - c(u2) ≡ 0 (mod 3) at every crossing, where o is the over-arc and u1, u2 the under-arcs. The solutions form a vector space over the field with three elements, so their number is 3^k with k >= 1 (the constant colorings always solve the system); this number is unchanged by all three Reidemeister moves and is therefore a knot invariant. K is tricolorable if and only if the solution space has dimension at least 2 — equivalently, a non-constant solution exists. Colorings correspond bijectively to homomorphisms from the knot group pi_1(S^3 \ K) to the symmetric group S_3 sending every meridian to a transposition: label the transpositions (12), (13), (23) with the three colors, and the Wirtinger relation at each crossing becomes exactly the Fox condition. Nontrivial colorings correspond to the surjective homomorphisms. The trefoil has 9 colorings; the unknot and the figure-eight have 3.

---

16. Seifert Surfaces #

Plain-language definition #

A Seifert surface for a knot is an orientable, connected surface whose boundary (edge) is the knot itself. Seifert's theorem (1934) guarantees that every knot bounds such a surface.

The Seifert algorithm constructs the surface explicitly:

1. At each crossing, smooth it into two parallel arcs
2. The smoothed arcs form simple closed curves (Seifert circles)
3. Fill each circle with a disk
4. Reconnect at crossings with half-twist bands

The genus of the minimal Seifert surface is a knot invariant: genus = (crossings - Seifert circles + 1) / 2.

Worked example: the algorithm on the trefoil, end to end #

Run the algorithm on the standard 3-crossing trefoil diagram — exactly what Puzzle 16, The Seifert Sail, has you do with physical panels:

1. Orient. Draw a travel arrow along the knot.
2. Smooth all 3 crossings. At each crossing, delete the crossing and reconnect the four loose ends in the only way that respects the arrows: incoming strands join outgoing strands without crossing.
3. Count the circles. The smoothed diagram falls apart into s = 2 Seifert circles — a small circle nested inside a larger one.
4. Cap with disks. Fill each circle with a disk; picture the two disks stacked at slightly different heights.
5. Reattach c = 3 bands. At each former crossing, join the two disks with a narrow band given a half-twist. The twist direction records the sign of the crossing it replaced.

The result is one connected surface whose boundary is precisely the original trefoil, and it is orientable: because the smoothing respected the travel arrows, the two disks can be given compatible rotation senses, and the half-twist bands join them consistently — the surface has a genuine front and back.

Now compute the genus from the counts. The Euler characteristic is chi = s - c = 2 - 3 = -1, and for a connected orientable surface with one boundary circle, chi = 1 - 2g, so g = 1. Equivalently, by the formula above: genus = (c - s + 1)/2 = (3 - 2 + 1)/2 = 1. The trefoil's spanning membrane is a torus with one puncture — one genuine handle, visible in the assembled Sail as the gap between the twisted bands. And since a knot of genus 0 would bound a disk and be the unknot, while Section 15 proved the trefoil is knotted, genus 1 is not just what this particular surface happens to have — it is the trefoil's true genus.

Which puzzles use this #

• Puzzle 16, The Seifert Sail: Three shaped panels are assembled inside a trefoil frame to physically construct a Seifert surface. Once built, a cord loop can be pushed across the surface and freed. The surface makes visible the theorem that every knot bounds an orientable surface.

Physical intuition #

The Seifert surface is a membrane spanning the interior of a knot. Imagine stretching a soap film inside a wire trefoil — the film would form a surface whose edge is the trefoil wire. This surface has a half-twist at each crossing (which is why soap films on trefoils look twisted). A cord linked with the wire can be pushed across this surface and freed, because the surface provides a continuous path from one side to the other.

Rigorous statement #

Seifert's theorem (1934): every oriented knot or link in S^3 bounds a compact, connected, orientable surface embedded in S^3 — a Seifert surface — and Seifert's algorithm constructs one from any diagram. For a knot diagram with c crossings whose smoothing produces s Seifert circles, the resulting surface has Euler characteristic chi = s - c and genus (c - s + 1)/2. The genus g(K) of a knot K is the minimum genus over all Seifert surfaces for K. It satisfies g(K) = 0 if and only if K is the unknot — a knot bounding an embedded disk is trivial — and it is additive under connected sum: g(K1 # K2) = g(K1) + g(K2). For torus knots, Seifert's algorithm applied to the standard diagram realizes the minimum, giving g(T(p,q)) = (p-1)(q-1)/2 (Section 14).

---

17. Unknotting Number #

Plain-language definition #

The unknotting number u(K) of a knot K is the minimum number of crossing changes needed to convert K into the unknot. A crossing change swaps which strand goes over and which goes under at a single crossing.

Key values:

• Unknot: u = 0
• Trefoil: u = 1
• Figure-eight: u = 1
• Cinquefoil (5_1): u = 2

The unknotting number is a topological invariant that measures how far a knot is from being trivial.

Worked example: predict-then-flip on the minimal diagram #

Puzzle 9, The Crossing Number, makes u(figure-eight) = 1 physical — with a generosity the definition does not promise. The frame realizes the standard 4-crossing diagram, labeled A through D as in the figure, each crossing flippable by inverting its pin. On this diagram, every one of the four flips produces the unknot.

Why: the minimal diagram has exactly two bigon faces — an outer shell whose corners are crossings A and C, and a central eye whose corners are B and D — and every crossing is a corner of exactly one bigon. In the alternating diagram each bigon is a locked clasp. Flip either of its corners and one arc of that bigon rides over at both ends: a Reidemeister II move slides it off, both crossings vanish, and the two surviving crossings degenerate into kinks that Reidemeister I untwists. Any pin you pick is a corner of some clasp, so any pin frees the ring.

This over-delivery is itself instructive:

• What u = 1 actually promises is that some single flip in some diagram unknots the knot. This particular diagram happens to realize the minimum at every crossing — a fact about the diagram, not a law about crossing changes (on the cinquefoil, u = 2, so no single flip of any diagram works at all).
• No flip on this diagram produces a trefoil — all four land on the unknot — and that is not a shortcoming of the diagram. A 2012 theorem of Kawauchi shows that no single crossing change in any diagram of the figure-eight yields a trefoil: the Gordian distance between the two knots is 2. Crossing-change distance between knots, like unknotting number itself, is a question about the knots, not about the picture you happen to be looking at. For a diagram where the choice of crossing genuinely decides the destination, look one knot up the table at 5_2: flip a clasp crossing of its minimal diagram and the knot dissolves to the unknot, but flip a crossing in its twist region and two twists cancel, leaving a trefoil.
• Because the flip always succeeds, the puzzle's real test is foresight: predict which clasp opens and trace the R-II-then-R-I dissolution before touching the pin. Flip first and the frame unknots anyway — teaching nothing.

Why unknotting number is hard #

The definition hides a quantifier over infinitely many diagrams: u(K) is the minimum over ALL diagrams of K and all choices of crossings. The optimal crossing change may not be visible in the diagram in front of you — you might have to deform the knot into a completely different diagram before the decisive crossing even exists. This is why no simple algorithm computes u, why lower bounds like the signature bound below are precious, and why some knots with only ten crossings still have unknown unknotting numbers.

Which puzzles use this #

• Puzzle 9, The Crossing Number: A figure-eight knot frame with 4 flippable crossing pins. Flipping any one of the four converts this minimal diagram to the unknot (u = 1 realized four ways); the solver's task is to predict, before flipping, which clasp opens and which Reidemeister moves dissolve the frame.

Physical intuition #

Each crossing pin controls which strand is on top at that point. Flip a pin and the knot type changes instantly, though the steel never moves; slide the ring and it glides past what used to be a catch point. On this frame any pin frees the ring, so the test is of foresight rather than luck: say which moves the flip will unlock before you pull the pin. The unknotting number tells you how many flips are needed — for the figure-eight, exactly one. In general it does not tell you which crossing, or even which diagram, realizes them.

Rigorous statement #

The unknotting number u(K) is the minimum, over all diagrams of K and all finite sequences of crossing changes and ambient isotopies transforming K into the unknot, of the number of crossing changes used. u(K) = 0 if and only if K is the unknot. The signature bound (Murasugi, 1965): |sigma(K)| / 2 <= u(K), where sigma(K) is the knot signature. The trefoil has |sigma| = 2, forcing u >= 1 and hence u = 1; the figure-eight has sigma = 0, so the bound says nothing — yet u = 1. Lower bounds can be blind. For torus knots the answer is complete: u(T(p,q)) = (p-1)(q-1)/2, conjectured by Milnor and proved by Kronheimer and Mrowka (1993) using gauge theory. Examples: u(T(2,3)) = 1 (trefoil) and u(T(2,5)) = 2 (cinquefoil), matching the table above — and for torus knots the unknotting number coincides with the genus (Sections 14 and 16).

---

18. Satellite Knots and JSJ Decomposition #

Plain-language definition #

A satellite knot is a knot that can be decomposed into two layers:

• The companion: a non-trivial knot embedded as the core curve of a solid torus
• The pattern: a knot or link inside the solid torus that wraps around the companion

The satellite knot is the result of replacing the companion's tubular neighborhood with the pattern's structure. It is more complex than either component alone.

JSJ decomposition #

The Jaco-Shalen-Johannson (JSJ) decomposition theorem states that every compact, orientable, irreducible 3-manifold has a unique decomposition along incompressible tori into Seifert-fibered and hyperbolic pieces. For knot complements, this means satellite knots decompose uniquely into companion and pattern components.

The practical consequence: a satellite knot's properties can be analyzed by studying each layer independently.

Worked example: decomposing the Satellite Trap #

Apply the two-layer analysis to Puzzle 17 itself:

• Companion: the trefoil tunnel. The tunnel molded into the torus shell follows a (2,3) torus-knot path. It is rigid, embedded in the shell, and nothing a solver does can change it.
• Pattern: the cord. The cord's route — through the tunnel, out a port, across the surface, back in — is a curve inside the solid torus that surrounds the tunnel. Pulling bights of cord through the ports changes the pattern, and only the pattern.
• The JSJ wall: the shell itself. Topologically, the boundary between the layers is the incompressible torus around the companion; physically, it is the acrylic wall your fingers stop at.

Now place the two rings, one per layer:

• The outer ring encircles an external arc of cord. Its linking is entirely with the pattern — the layer on the accessible side of the wall — so pattern moves (pull a bight through a port, pass it over the ring, feed it back) can and do free it.
• The inner ring rides a section of cord inside the tunnel. Its linking is with the companion layer, on the far side of the wall. The companion is a genuine trefoil that no surface manipulation can alter, so the inner ring is trapped permanently.

The decomposition is not just a description — it is the solving strategy. It tells you in advance which sub-problem is solvable (the pattern) and which is provably hopeless (the companion), so you stop wasting moves on the wrong ring.

Which puzzles use this #

• Puzzle 17, The Satellite Trap: A torus shell contains a trefoil-knotted tunnel (companion). A cord threads through the tunnel and connects externally (pattern). Two rings are trapped by different layers: the outer ring is linked only with the pattern (and can be freed by rerouting the external cord), while the inner ring is linked with the companion (and is permanently trapped).

Physical intuition #

Think of a satellite knot as a knot within a knot, like Russian nesting dolls. The outer structure (pattern) can be manipulated without affecting the inner structure (companion). In the puzzle, you can reroute the cord where it exits the torus (changing the pattern) without being able to change the trefoil tunnel inside (the companion). One ring lives in the pattern layer and can be freed; the other lives in the companion layer and cannot.

Rigorous statement #

Let V = S^1 x D^2 be the standard solid torus and let P (the pattern) be a knot in V that is essential: P is not contained in any 3-ball inside V and is not isotopic to the core circle S^1 x {0}. Let e: V -> S^3 be an embedding carrying the core to a non-trivial knot C (the companion). Then K = e(P) is a satellite knot. The image torus e(dV) is essential (incompressible and not boundary-parallel) in the complement of K, and conversely a knot is a satellite if and only if its complement contains an essential torus. The JSJ decomposition theorem (Jaco-Shalen and Johannson, 1979): every compact, orientable, irreducible 3-manifold contains a minimal family of disjoint essential tori, unique up to isotopy, that cuts it into pieces each of which is Seifert-fibered or atoroidal (and the atoroidal pieces are hyperbolic, by Thurston's geometrization of Haken manifolds). For the Satellite Trap, the shell wall is the JSJ torus: the piece beyond it is the trefoil complement, which is Seifert-fibered — as befits a torus knot (Section 14) — while the piece containing the cord is the pattern space V minus P.

---

19. The Whitehead Link and Higher-Order Linking #

Plain-language definition #

Section 3 gave you the linking number and, with it, a promise: if two closed curves have linking number zero, some continuous motion pulls them apart. That promise came with a footnote. This section is the footnote made solid.

The Whitehead link is a pair of closed loops whose linking number is exactly zero, and which nevertheless cannot be separated by any amount of sliding, stretching, or bending. The winding of one loop around the other genuinely cancels — count it and you get zero. What holds them together is a clasp: a place where the loops hook through each other in a pattern the linking number is blind to.

So the inference "linking number zero, therefore separable" is false in general. Zero linking is necessary for two loops to come apart — a nonzero value proves they are stuck (Section 3) — but it is not sufficient. A zero says only that the door is not provably locked. It does not say the door opens.

Worked example: recomputing lk = 0 #

Take the standard diagram in the figure: a rigid ring and a cord loop woven through it twice — once front to back, once back to front — with the cord's two hanging lobes meeting in a clasp below. Orient both curves and sign each ring-cord crossing by the recipe of Section 3.

The two passes through the ring's aperture contribute four crossings. The front-to-back pass pierces the ring's spanning disk one way and shows up as two crossings of the same sign, +1 and +1. The back-to-front pass pierces the other way: two crossings of sign -1 and -1. Summing,

(+1) + (+1) + (-1) + (-1) = 0, so lk = 1/2 x 0 = 0.

(The factor of 1/2 is the same one from Section 3: between two closed loops every threading registers as two crossings, so the signed crossing sum is always twice the linking number.) The clasp is a crossing of the cord with itself, not with the ring, so it never enters the linking-number sum at all. The count is honestly, provably zero.

What the count cannot see: the clasp #

Here is where the analysis of Section 3 runs out. The +1 pass and the -1 pass would, on their own, happily cancel — pull one back out through the aperture, then the other, and the cord lifts away. But the clasp ties the cord's two lobes together, and neither lobe can retract without first passing through the other. The cancellation that the arithmetic describes exists on paper and cannot be performed in space. The clasp remembers exactly what the linking number forgets.

The linking number is a first-order invariant: it reports the net number of times one curve passes through the other. It says nothing about how those passes are arranged relative to one another — and that arrangement is the whole content of the Whitehead link.

Higher-order linking #

What, then, certifies that the Whitehead link is genuinely stuck? The answer is a higher-order linking invariant, and it is the two-component cousin of the invariant that rescued the Borromean rings in Section 7.

Recall the pattern there: three rings, every pairwise linking number zero, yet inseparable as a trio — and the certificate was Milnor's invariant mu_bar(1,2,3), which measures a purely three-body linking that no pairwise count can detect. The Whitehead link distills that warning down to two components. Its ordinary (first-order) linking number vanishes, and its non-triviality is witnessed instead by a second-order Milnor invariant, mu_bar(1,1,2,2) — one with a repeated index, encoding how the two passes of one component clasp back through the other. Equivalently, this is the Sato-Levine invariant, defined precisely for two-component links of linking number zero, which evaluates to +/-1 on the Whitehead link.

Both the Borromean rings (Section 7) and the Whitehead link are therefore members of the same family: links that are trivial to every lower-order test and are held together only by higher-order linking. Milnor's invariants form a graded ladder of such tests, and these links live one rung above the linking number.

Which puzzle uses this #

• Puzzle 18, The Whitehead Waltz: Two near-identical stations each present a ring with a cord woven through it twice and clasped below. Both compute to linking number zero. One is a genuine Whitehead link (the clasp hooks) and can never be freed; the other has that single clasp crossing flipped, making it an unlink in disguise, and its cord lifts off. The solver must resist the reflex "lk = 0, therefore free," identify the true clasp, and prove the point by freeing only the loop that can be freed.

Physical intuition #

Hold the trapped station and try to lift the cord. It always feels one clever move from freedom — there is slack everywhere, the count says zero, the loop swings loosely. But every attempt to retract one pass drags the other lobe into the clasp, and the hook cinches against the ring wire. You can slide the clasp anywhere along the cord, yet it is always somewhere, and it always arrives at the aperture exactly when the cord tries to leave. That inescapable catch, felt in the hands, is higher-order linking — the thing the arithmetic could not see.

Rigorous statement #

The Whitehead link W is a two-component link in S^3 with linking number lk(W) = 0 that is nonetheless non-split: no embedded 2-sphere in S^3 separates its two components, so no ambient isotopy pulls them apart. Its non-triviality is detected by the first non-vanishing Milnor mu-bar invariant mu_bar(1,1,2,2) = +/-1 — a second-order invariant drawn from the lower central series of the link group — equivalently by the Sato-Levine invariant, defined for two-component links of linking number zero, which equals +/-1 on W. The Whitehead link also has unlinking number 1 (a single crossing change at the clasp converts it to the two-component unlink), and its complement S^3 \ W is a hyperbolic 3-manifold of finite volume — one of the standard examples of low-dimensional topology, widely used in Dehn-surgery constructions. The Borromean rings (Section 7) are the three-component member of the same Brunnian/higher-order family: all lower-order linking data vanish, and a Milnor invariant supplies the certificate.

---

20. Rational Tangles and Continued Fractions #

Plain-language definition #

A tangle, in John Conway's sense, is what you get by pinning the four ends of two arcs to the boundary of a disk (think: a square frame with a cord end fixed at each corner) and letting the arcs cross each other freely inside. The simplest tangle is two horizontal parallel strands with no crossings at all — the 0 tangle.

From the 0 tangle you can build endlessly with just two operations:

• TWIST (T): grab the two right-hand ends and cross one over the other, adding a single crossing.
• ROTATE (R): turn the entire frame a quarter-turn.

Any tangle you can reach from the 0 tangle using only these two moves is called rational. (Not every tangle is — but every tangle in Puzzle 19 is, because T and R are the only moves the frame allows.) Conway's discovery is that each rational tangle carries a single number, its fraction, an element of the rationals Q together with one extra value, infinity — and that this number is a complete record of the tangle.

The two moves are arithmetic #

Assign the 0 tangle the number 0. Then the two physical moves act on the fraction as pure arithmetic:

• T (twist): x -> x + 1. A right-hand twist adds one positive crossing, and adds 1 to the fraction.
• R (rotate): x -> -1/x. The quarter-turn moves no strand relative to any other; it only changes which pair of ends faces the twisting edge, and its effect on the number is to send it to its negative reciprocal.

One caution about the extra value: the fraction infinity is the tangle of two vertical parallel strands. R swaps 0 and infinity, so of the two crossingless tangles only one — the horizontal one, fraction 0 — is the goal.

Conway's theorem: a complete invariant #

Conway's theorem (1970): two rational tangles are ambient-isotopic, with their four endpoints held fixed, if and only if their fractions are equal.

This is the hinge of the entire arc, so it is worth stating against the backdrop of Section 19. The linking number is an incomplete invariant: it is lossy, collapsing many genuinely different configurations onto the same integer, which is exactly why lk = 0 could promise nothing about the Whitehead link. The tangle fraction is the opposite — a complete invariant. Equal fractions do not merely fail to distinguish two tangles; they guarantee the tangles are the same. An incomplete invariant can only ever prove impossibility ("these differ, so no motion connects them"); a complete one hands you an algorithm ("drive the fraction to the target and the tangles must coincide"). Half a meter of cord and any pile of crossings compress, losslessly, into one rational number.

The compass: continued fractions #

Where does the fraction come from? Write it as a continued fraction. For the puzzle's starting tangle,

3/2 = 1 + 1/2,

which reads directly off the diagram as one right twist stacked on top of two vertical twists. In general a twist/rotate word encodes, and its fraction equals, a continued fraction [a_n, ..., a_2, a_1] built from the counts of successive twist regions.

Unwinding a continued fraction from the outside in is nothing but Euclid's algorithm on the numerator and denominator, and that is what dictates the solving rule. Read the fraction and obey a compass:

• Positive? ROTATE.
• Negative? TWIST.
• Zero? Stop — solved.

Each rotate-then-twist block performs one division step of Euclid's algorithm, and because Euclid's algorithm always terminates, the descent is guaranteed to reach 0 — walking up from the negative integers one unit at a time, never overshooting.

Worked example: the descent from 3/2 to 0 #

Start at 3/2 and apply the compass, verifying each step against the two laws (T adds 1; R sends x to -1/x):

1. 3/2 is positive -> R: -1/(3/2) = -2/3.
2. -2/3 is negative -> T: -2/3 + 1 = 1/3.
3. 1/3 is positive -> R: -1/(1/3) = -3.
4. -3 is negative -> T: -3 + 1 = -2.
5. -2 is negative -> T: -2 + 1 = -1.
6. -1 is negative -> T: -1 + 1 = 0. Solved.

Six moves, R T R T T T, each one forced. Notice that the picture does not simplify monotonically — step 3 turns a tidy 1/3 into three stacked crossings (-3), which looks like a setback. But the number is marching toward zero, and distance-to-solved is measured by continued-fraction depth, not by crossing count or by absolute value. Judge the tangle by its fraction and the descent is a straight line down.

Which puzzle uses this #

• Puzzle 19, The Tangle Dance: Two cords span a square frame, with the two legal moves — T and R — embossed on its edges. The frame is preset to fraction 3/2, and the objective is to return it to 0. Because the fraction is a complete invariant, solving is not a search but a computation: run the compass, write down the ledger above, and execute six forced moves.

Physical intuition #

A twist is honest labor — the cords visibly wrap and the tangle thickens. A rotate is eerie: you turn the frame like a steering wheel and nothing between your hands changes, yet the number naming the state flips from 3/2 to -2/3, and a move that was making things worse a moment ago is suddenly the move that helps. The most important operation looks like doing nothing, because the thing it changes is not the cords but their relationship to the twisting edge. At the very end, the last twist lands a crossing against its mirror image, a Reidemeister II pair annihilates, and the cords fall parallel. You feel the zero.

Rigorous statement #

A rational tangle is a tangle obtained from the 0 tangle (two boundary-parallel arcs in a 3-ball with four marked boundary points) by a finite sequence of the generating operations twist and rotate; equivalently, it is a tangle whose double branched cover is a solid torus. To each rational tangle Conway assigns a fraction F in Q ∪ {infinity}, computed as the continued fraction determined by its twist/rotate word. Conway's classification theorem (1970): two rational tangles are ambient-isotopic rel their four endpoints if and only if they have the same fraction. The generators act on the fraction by T: x -> x + 1 and R: x -> -1/x, so the fraction of a word in T and R equals the continued fraction that word encodes; R is its own inverse (-1/(-1/x) = x), while T has infinite order. Because the compass rule realizes the Euclidean algorithm on (numerator, denominator), every rational tangle descends to the 0 tangle in finitely many moves. In contrast to the linking number of Section 19, which is a lossy (incomplete) invariant, the tangle fraction is a complete invariant of rational tangles: it determines the isotopy class exactly.

---

21. Connected Sums and Composite Knots #

Plain-language definition #

Tie a knot in a closed loop of cord, then tie a second knot farther along the same loop, and you have formed the connected sum of the two, written K1 # K2. Each factor keeps its own stretch of cord; the two are joined by a short two-strand neck. Slice through that neck with an imaginary sphere — the sum sphere — and you separate the loop into two pieces, one factor sealed inside the sphere and the other outside, with the loop puncturing the sphere at exactly two points.

This is the knot-theoretic echo of prime factorization. A knot is prime if it is not the connected sum of two nontrivial knots — it cannot be cut along any sum sphere into simpler pieces. A knot that is a nontrivial connected sum is composite. The trefoil, the figure-eight, and every torus knot are prime; the square and granny knots of Puzzle 20 are composite, each built from two trefoils.

Square versus granny #

Two composites, both made of two trefoils, can differ only in the handedness of their factors:

• Square knot = left trefoil # right trefoil (a mirror-image pair).
• Granny knot = left trefoil # left trefoil (two of the same hand).

If your shoelaces keep slipping, you are tying the granny where you wanted the square. Topologically the two are genuinely different knots — but proving it takes the right invariant, because almost every invariant in this primer cannot tell them apart.

Additive and multiplicative invariants #

Connected sum interacts cleanly with the standard invariants, and how it interacts is the key:

• Genus is additive: g(K1 # K2) = g(K1) + g(K2) (Section 16). Two genus-1 trefoils give genus 2, for both square and granny.
• Signature is additive: sigma(K1 # K2) = sigma(K1) + sigma(K2). This is the one that will separate them.
• Crossing number is additive for alternating knots such as these (3 + 3 = 6); additivity in general is a famous open problem.
• The Alexander polynomial is multiplicative: the polynomial of a sum is the product of the factors' polynomials. Each trefoil contributes t - 1 + t^{-1}, so both composites have Alexander polynomial (t - 1 + t^{-1})^2.
• The number of Fox 3-colorings is multiplicative (up to the shared trivial colorings): each trefoil has 9, and each composite has 9 x 9 / 3 = 27 (Section 15). Mirroring never changes a coloring count.

The invariant scoreboard #

Line up the series' whole toolkit against the two suspects:

Four of the five tests are stone silent. The signature sigma is the one that speaks. It is an integer read off a knot's Seifert surface (Section 16); it adds under connected sum and, crucially, flips sign under mirroring. Fix the convention that the right-handed trefoil (all-positive crossings) has sigma = -2, so its mirror the left-handed trefoil has sigma = +2. Then:

• sigma(square) = sigma(L) + sigma(R) = (+2) + (-2) = 0. The mirror factors cancel. (Any amphichiral knot must have signature 0.)
• sigma(granny) = sigma(L) + sigma(L) = (+2) + (+2) = +4. Same hands add up.

Since 0 does not equal +4, the square and granny knots are provably distinct. The same sign-flip-under-mirroring settles each composite's relationship to its own reflection: mirror(square) = mirror(L # R) = R # L = square, so the square knot is amphichiral — equal to its own mirror image. But mirror(granny) = R # R is the other granny, so the granny knot is chiral, coming in mirror twins with signatures +4 and -4.

Worked example: reading the scoreboard #

The discipline the scoreboard teaches is that an invariant is a question you ask a knot, and the skill is choosing a question that separates your suspects. Ask "how many crossings?" and square and granny answer identically. Ask "what is your genus? your Alexander polynomial? are you tricolorable?" — identical, identical, identical. Every one of those questions is blind to handedness, and the difference between square and granny is entirely a matter of handedness. Only the signature is built to see a mirror, and it is built to survive connected sum (it adds), so it survives to deliver the verdict: 0 versus +4.

Which puzzle uses this #

• Puzzle 20, The Granny's Downfall: Two closed six-crossing loops — one a square knot, one a granny — must be matched against a mold carved with the square knot's diagram. Only the square loop seats fully, because a seated loop physically realizes that diagram and therefore is that knot. The granny cannot seat: its wrong-handed second clump would have to be re-handed, which would drop its signature from +4 to 0 and turn it into a square knot — and no manipulation of a closed loop changes its knot type.

Physical intuition #

Pull either loop from opposite sides and the two knotted clumps slide apart until a clean two-strand neck connects them — your hands are holding the sum sphere's equator. Seating the square loop feels like slotting a key. The granny starts identically, then its second clump refuses: all three of its crossings insist on lying the wrong way over the mold's bridges. Flip that clump to fix it and nothing improves — the flip is a symmetry of the trefoil, so the same three crossings refuse in the same way. What you keep failing to seat is not one stray crossing but a whole wrong-handed clump, a conserved lump of three crossings. You are pushing, with your hands, on the +4 that the signature measured.

Rigorous statement #

The connected sum K1 # K2 of two oriented knots is formed by removing a small arc from each and splicing the four ends together respecting orientation; the operation is well-defined and, on oriented knots, commutative and associative, with the unknot as unit. A knot is prime if it is not the connected sum of two nontrivial knots. Schubert's theorem (1949): every knot factors as a connected sum of prime knots, uniquely up to order — the oriented knots form a free commutative monoid on the set of prime knots, the exact analogue of the fundamental theorem of arithmetic. Under connected sum the genus is additive, g(K1 # K2) = g(K1) + g(K2) (Section 16); the signature is additive, sigma(K1 # K2) = sigma(K1) + sigma(K2); and the Alexander polynomial is multiplicative, Delta_{K1 # K2}(t) = Delta_{K1}(t) Delta_{K2}(t). The signature also negates under mirroring, sigma(mK) = -sigma(K). With sigma(right trefoil) = -2 and sigma(left trefoil) = +2, these rules give sigma(square) = 0 and sigma(granny) = +4, so the square and granny knots are distinct; and since the square knot equals its own mirror it is amphichiral, while the granny is chiral.

---

22. Glossary #

Concise definitions of every technical term used across the EXKNOTS puzzle files, listed alphabetically.

Arc — An open curve with two distinct endpoints. Unlike a loop, an arc can always be unknotted in free space. In the EXKNOTS puzzles, arcs arise when a cord has two separate attachment points (Puzzles 1, 8).

Bight — A U-shaped fold or loop of cord, created by doubling the cord back on itself without crossing the ends. Used as a manipulation technique in Puzzles 2, 3, 7, 11, and 17 to thread cord through holes or over obstacles.

Borromean rings — A specific 3-component link in which the three components are mutually linked but no two are linked to each other. The simplest non-trivial Brunnian link. Puzzle 6 (Trinity Lock) is built on this structure.

Brunnian link — A link of n components such that removing any single component makes the remaining components completely unlinked. Borromean rings are the case n = 3. Named after Hermann Brunn (1892).

Composite knot — A nontrivial connected sum of two or more prime knots, the knot-theoretic analogue of a composite number. The square and granny knots (Puzzle 20) are composite: each is a sum of two trefoils.

Configuration space — The space of all possible states of a mechanical system. Each point represents one arrangement of all parts. The topology of this space governs which transitions between states are possible (Puzzle 7).

Connected sum — The knot formed by splicing two knots into one loop along a sum sphere, written K1 # K2. Genus and signature add under connected sum, the Alexander polynomial multiplies, and factorization into primes is unique (Schubert). The basis of Puzzle 20.

Crossing number — The minimum number of crossings in any planar diagram of a knot or link. The unknot has crossing number 0; the trefoil has crossing number 3.

Fiber bundle — A space that is locally a product of a base space and a fiber, but may be globally twisted. The Hopf fibration is the central example in EXKNOTS (Puzzle 12).

Free group — A group whose generators satisfy no relations other than the trivial cancellation of a generator with its inverse. The free group F(a, b) on two generators is the fundamental group of a genus-2 handlebody (Puzzle 11).

Fundamental group — The group of homotopy classes of loops at a basepoint in a topological space. Captures the distinct ways to walk in a closed path. Denoted pi_1(X). Used in Puzzles 7 and 11.

Generator — A basic element of a group from which all other elements can be built by composition and inversion. In the fundamental group of a genus-g handlebody, there are g generators, one for each tunnel/handle.

Genus — The number of handles on a surface. A sphere has genus 0, a torus has genus 1, a two-holed torus has genus 2. For a handlebody, the genus equals the number of through-tunnels (Puzzles 2, 11).

Gray code — A binary numbering system in which consecutive values differ by exactly one bit. Also called reflected binary code. Governs the solution sequence of the Ouroboros Chain (Puzzle 10).

Handlebody — A solid body with through-tunnels. A genus-g handlebody has g tunnels and its fundamental group is the free group on g generators. The acrylic block in Puzzle 11 is a genus-2 handlebody.

Homeomorphism — A continuous bijection whose inverse is also continuous. Two spaces related by a homeomorphism are topologically identical — they have the same topological invariants.

Homotopy — A continuous deformation of one map (or loop, or path) into another. Two loops are homotopic if one can be continuously deformed into the other. Homotopy is the equivalence relation underlying the fundamental group.

Hopf fibration — The map h: S^3 -> S^2 whose fibers are circles (S^1). Discovered by Heinz Hopf in 1931. Decomposes the 3-sphere into a family of linked circles. Underlies the coupled-rotation mechanism of Puzzle 12.

Identity — The neutral element of a group. In the fundamental group, the identity is the class of loops that can be contracted to a point. In the free group F(a, b), a word reduces to the identity only when all generators cancel via adjacent inverse pairs.

Linking number — An integer invariant measuring how many times two closed curves wind around each other. Computed by summing signed crossings and dividing by 2. A linking number of 0 is necessary (though not always sufficient) for the curves to be separable (Puzzles 1, 3).

Milnor invariant — A higher-order linking invariant that detects collective linking when the pairwise linking numbers all vanish. The triple invariant mu-bar(1,2,3) certifies the Borromean rings (Puzzle 6); a repeated-index form mu-bar(1,1,2,2), equivalently the Sato-Levine invariant, certifies the Whitehead link (Puzzle 18).

Mobius band — A non-orientable surface with one edge and one side, formed by joining a rectangular strip with a single half-twist. The single boundary component is the key to Puzzle 4 (Mobius Snare).

Orientability — A surface is orientable if it has two distinct sides (a consistent notion of "clockwise" at every point). Non-orientable surfaces, like the Mobius band, have only one side.

Prime knot — A nontrivial knot that is not the connected sum of two nontrivial knots, the analogue of a prime number. Trefoils, the figure-eight, and torus knots are prime; every knot factors uniquely into primes (Schubert).

Rational tangle — A tangle of two arcs in a disk, built from the trivial tangle by twisting adjacent ends and rotating the disk, and classified completely by its Conway fraction in Q ∪ {infinity} (Puzzle 19).

Reidemeister moves — Three types of local diagram changes (twist, poke, slide) that generate all equivalences between knot diagrams. Any continuous deformation of a knot in 3D can be decomposed into a sequence of these three moves (Puzzles 1, 3, 5, and 8).

S^1 — The 1-sphere; the circle. The set of points at unit distance from the origin in the plane. The fiber in the Hopf fibration.

S^2 — The 2-sphere; the ordinary sphere surface. The set of points at unit distance from the origin in 3D. The base space in the Hopf fibration.

S^3 — The 3-sphere; a three-dimensional manifold living in 4D space. The set of points at unit distance from the origin in R^4. The total space of the Hopf fibration. Locally looks like R^3 but is compact and closed.

Signature — An integer knot invariant sigma(K) computed from a Seifert surface; it adds under connected sum and negates under mirroring. It separates the square knot (sigma = 0) from the granny knot (sigma = +4) in Puzzle 20.

Tangle fraction — The rational number (in Q ∪ {infinity}) that Conway assigns to a rational tangle. A complete invariant: two rational tangles are isotopic if and only if their fractions are equal (Puzzle 19).

Topological invariant — Any property of a topological space that is preserved under homeomorphism (or, for knots and links, under ambient isotopy). Examples: genus, linking number, crossing number, fundamental group. Invariants are what topology actually measures.

Trefoil — The simplest non-trivial knot, with crossing number 3. It cannot be unknotted. The visual pattern in Puzzle 8 (The Ferryman's Knot) resembles a trefoil, but the cord is an open arc rather than a closed loop, so it is not a true trefoil.

Unknot — A knot equivalent to a simple circle — no crossings, no tangles. A closed curve that can be deformed into a round circle. Crossing number 0. The cord in Puzzle 1 (The Gatekeeper) is topologically an unknot (or rather, an unknotted arc).

Whitehead link — A two-component link with linking number zero that is nonetheless non-split, held together by a clasp. Its non-triviality is certified by a higher-order (Milnor / Sato-Levine) invariant, not by the linking number. The trap in Puzzle 18.

Word — In group theory, a finite sequence of generators and their inverses. For example, aba^{-1} is a word in the free group F(a, b). The word represents an element of the group. In EXKNOTS, a cord's path through tunnels encodes a word in the fundamental group, and the puzzle is solved when the word reduces to the identity (Puzzle 11).

---

This primer covers every topological concept used in the twenty EXKNOTS puzzles. For construction details, solution walkthroughs, and physical specifications, see the individual puzzle files in the puzzles/ directory.