A group is not "a pile of things" — it is "a pile of actions." Take an equilateral triangle: what can you do to it so it looks unchanged? Rotate 120°, rotate 240°, do nothing, and flip across each of its three axes — exactly 6 actions. These 6 actions form a group.
Three intuitions are key: actions can be chained (rotate 120° then flip = another legal action, still among the 6); every action can be undone (the inverse of rotating 120° is rotating 240°); and there is a "do-nothing" action. The noun (the triangle) fools your eye; the verb — the operation applied to it — is what a group really studies. Moving attention from "the object" to "the actions that leave it unchanged" is the whole starting point of group theory.
A group $(G,\cdot)$ is a set $G$ with an operation $\cdot$ satisfying four axioms: closure — $a\cdot b$ is still in $G$; associativity — $(a\cdot b)\cdot c=a\cdot(b\cdot c)$; identity — some $e$ with $e\cdot a=a$; inverses — each $a$ has $a^{-1}$ with $a\cdot a^{-1}=e$.
Note there is no commutativity requirement: $a\cdot b\ne b\cdot a$ is the norm. Putting on socks then shoes is not the same as shoes then socks — the world where order matters is exactly where group theory lives.
Four seemingly meager axioms precisely capture the whole essence of "symmetry." More striking is Cayley's theorem: every group is isomorphic to some group of permutations — meaning any abstract group is, at heart, some concrete set of actions in disguise. This tells us: abstract algebra was not invented from thin air, but distilled to its essence from the single phenomenon of "reversible transformation." The fewer the axioms, the larger the world they hold.
All states of a Rubik's cube form a group (about $4.3\times10^{19}$ elements); "solving the cube" = finding a path back to the identity. Cryptography rests on groups: RSA lives on the multiplicative group mod $n$, elliptic-curve cryptography on the group of points on a curve. And in physics, which symmetries a system "has" directly determines which quantities it conserves — the group is the hinge connecting symmetry and conservation.
Rearranging $n$ things — that is shuffling. $S_n$ is the group of "all ways to shuffle $n$ cards." $S_3$ has $3!=6$ shuffles; $S_{52}$ (a deck of cards) has about $8\times10^{67}$. Each shuffle is a permutation.
Permutations have a lovely notation called cycles: $(1\,2\,3)$ means "1's spot goes to 2, 2 to 3, 3 back to 1." Any shuffle, however tangled, decomposes uniquely into disjoint cycles — like combing a knot into a few independent loops. This turns "arrangement" from a dizzying blur into something countable and computable.
$S_n$ is the group of all bijections (one-to-one correspondences) of the set $\{1,2,\dots,n\}$, with composition ("do one, then another") as the operation, and $|S_n|=n!$ elements. It is generally non-commutative: swapping (1 2) then (2 3) differs from the reverse order.
Cayley's theorem, restated: every finite group is a subgroup of some $S_n$. The most humble, concrete thing — "shuffling" — captures every abstract finite group; $S_n$ is "the mother of all finite groups." Deeper still: the internal structure of $S_n$ holds one of the most dramatic answers in the history of mathematics. Galois proved that for $n\ge5$, $S_n$ is "unsolvable," and hence that quintic and higher equations have no formula in radicals — a two-thousand-year puzzle about solving equations, its answer written into the group structure of shuffling.
Sorting algorithms essentially seek a path in $S_n$ from "disordered" to "ordered"; the number of inversions is the minimum count of adjacent swaps needed. Classical block ciphers scramble plaintext with permutations. The Rubik's cube group is a subgroup of $S_{48}$. And Galois theory — grown atop $S_n$ — today underpins coding theory and computer-algebra systems that solve polynomials.
A group can contain "smaller groups" — subgroups, like a self-consistent brick inside a larger structure. Lagrange's theorem states something very strong: with that brick, you can tile the entire group with no gaps and no overlaps, and every tile is the same size.
The consequence is immediately rigid: a group of 12 elements can only have subgroups of size dividing 12 — 1, 2, 3, 4, 6, 12 — and never a subgroup of 5 elements. The most primitive count, "size," pins down the possible internal structure. This is not an empirical pattern; it is necessity.
If a finite group $G$ has a subgroup $H$, then $|H|$ divides $|G|$, and the quotient $[G:H]=|G|/|H|$ is called the index of $H$. The key to the proof is cosets $gH=\{g\cdot h:h\in H\}$: one shows all cosets have equal size (each equal to $|H|$), are pairwise disjoint, and together make up $G$. So $G$ is cut into $[G:H]$ equal pieces — divisibility is just reading off this "tiling diagram."
This is a paragon of "the simple governing the complex": a pure counting constraint that, in one stroke, locks down every possible shape of the group. One corollary is beautiful to the extreme — every group of prime order is cyclic: if $|G|=p$ is prime, it has no subgroups but $\{e\}$ and itself, so its structure has no choice. From "infinitely many possibilities," a single divisibility carves out order.
Fermat's little theorem $a^{p-1}\equiv1\pmod p$ — a cornerstone of modern cryptography — is a direct corollary of Lagrange applied to "the multiplicative group mod $p$": that group has $p-1$ elements, so any element's "order" must divide $p-1$. The correctness of RSA, primality testing, and Diffie–Hellman key exchange all ultimately return to this divisibility constraint. An abstract counting theorem holds up billions of encryptions a day.
A group in itself is a set of abstract actions; a group action lets those actions actually "move" a concrete object — atoms in a crystal, points in a plane, quantum states. A representation goes further: it replaces each group element with a matrix, so that abstract multiplication becomes matrix multiplication you can genuinely compute.
Group actions bring a pair of twin concepts: the set of all positions a point can reach under every action is its orbit; the actions that pin that point in place form its stabilizer. They obey a delicate conservation law — the bigger the orbit, the smaller the stabilizer, and their product is exactly the size of the group.
A group action is a homomorphism $G\to\mathrm{Sym}(X)$, letting each group element play a permutation of the set $X$; a representation is a homomorphism $\rho:G\to GL(V)$, turning group elements into invertible matrices on a vector space. The core tool is the orbit–stabilizer theorem:
$\mathrm{Orb}(x)$ is everywhere $x$ can go under the action; $\mathrm{Stab}(x)$ is the elements fixing $x$. This is essentially a dynamic version of Lagrange's theorem — the group's size is split between "how far it can move" and "how firmly it is pinned."
Group actions let an abstract group first "touch" the real world, and representation theory translates group problems into linear algebra — the most computable branch of mathematics. How many ways a group can "essentially" act is captured completely by its irreducible representations. So "which symmetries does this system have" becomes equivalent to "which irreducible representations does its symmetry group have," turning a physics question fully algebraic.
This is group theory's most astonishing payoff. Crystallography: the symmetry groups that "leave a lattice invariant" in three-dimensional space, enumerated by pure group theory, number exactly 230 (the space groups) — not an experimental tally but a law computed by mathematics, obeyed by every crystal in nature without exception. Particle physics: Gell-Mann arranged known hadrons by the representations of SU(3), and they fell precisely onto the lattice points of hexagons and triangles (see figure); an empty slot predicted the then-unknown $\Omega^-$ particle, confirmed precisely by experiment years later. Quarks are the fundamental representation of SU(3); the entire Standard Model is, at heart, a play staged on the gauge symmetry group $SU(3)\times SU(2)\times U(1)$.