Thermodynamics described the bulk behavior of heat, work, and entropy with almost no microscopic content. Statistical mechanics — developed in the 1870s by Ludwig Boltzmann in Vienna and J. Willard Gibbs in New Haven — supplied the missing layer: macroscopic properties of matter are statistical averages over a vast number of microscopic configurations. A gas in a box has 10²³ molecules; you cannot track them all, but you can compute statistics, and the statistics determine temperature, pressure, and entropy. The idea was so radical that Boltzmann's contemporaries called him a romantic — atoms themselves were not yet established physics. Boltzmann took his own life in 1906, exhausted by the defense; vindication followed when Jean Perrin's experiments (1908–09), confirming Einstein's 1905 analysis of Brownian motion, gave the first direct evidence that atoms were real.
Consider a system with many microscopic degrees of freedom — a gas of molecules, a magnet of spins, a crystal of vibrating atoms. The system at any instant occupies one of an astronomically large number of microstates, each specifying the position and momentum (or spin direction, or vibration phase) of every constituent. The macrostate — what we observe at human scales — is a coarser description: temperature, pressure, energy, magnetization. Each macrostate is compatible with an enormous number of microstates. Boltzmann's central insight: the macrostate observed is overwhelmingly the one with the most microstates — not because the others are forbidden, but because they are vastly less probable in any uniform sampling of the phase space. The Boltzmann entropy formula, S = k_B · ln W (etched on his tombstone in Vienna), states that entropy is the logarithm of the number of microstates W compatible with the macrostate. The Boltzmann distribution, P(state) ∝ e^(−E/k_BT), says that at thermal equilibrium at temperature T the probability of finding the system in a state of energy E is exponentially suppressed in E/k_BT. The partition function Z = Σ e^(−E/k_BT) encodes the entire thermodynamics: free energy, internal energy, entropy, heat capacity all derive from Z by differentiation. Different statistics apply at the quantum level: Maxwell-Boltzmann for distinguishable classical particles, Fermi-Dirac for fermions (electrons, with Pauli exclusion), Bose-Einstein for bosons (photons, allowing condensation). Equilibrium is the macrostate of maximum entropy given the constraints. Phase transitions emerge as singular behavior in the partition function as parameters cross critical values.
Condensed matter physics — the largest subfield of contemporary physics — is essentially applied statistical mechanics: superconductors, superfluids, spin glasses, Bose-Einstein condensates, quantum Hall systems, topological insulators. Materials science designs alloys and polymers using stat-mech free-energy calculations. Chemical kinetics uses Boltzmann distributions to compute reaction rates (the Arrhenius equation). Biology: protein folding, RNA structure prediction, evolutionary fitness landscapes are fundamentally stat-mech problems. Markov Chain Monte Carlo methods in machine learning derive conceptually from stat-mech sampling techniques. The Ising model is the workhorse of computational physics, and modern energy-based models — Hopfield networks, Boltzmann machines, and the diffusion models behind today's image generators — are stat-mech objects.