The second law, statistically

State the second law of thermodynamics the way most textbooks do and it sounds like a decree: 'the entropy of an isolated system never decreases.' Stated that way it's easy to imagine some cosmic rule-enforcer forbidding processes that would lower entropy. Nothing of the kind is actually happening. Given everything built in this unit so far — macrostates, microstates, and Boltzmann's S = k_B ln W — the second law is not a separate physical law at all; it's a direct probabilistic consequence of the counting argument from the counting-microstates lesson. An isolated system left to itself wanders freely among its accessible microstates, all equally likely, and since the overwhelming majority of those microstates belong to the single macrostate of highest multiplicity (the equilibrium macrostate, by definition), the system is — with a probability indistinguishable from certainty for any macroscopic system — found evolving toward, and then remaining in, that macrostate. 'Entropy increases' just restates 'the system moves toward the macrostate with vastly more microstates.'

The word 'overwhelming' is doing real, quantifiable work here, and it's worth seeing why. Multiplicities for macroscopic systems are combinatorial — binomial coefficients like C(N, N/2) for the two-box gas — and the relative width of that distribution around its peak scales as 1/√N, a standard result for binomial statistics. For the 100-particle toy system from the counting-microstates lesson, √N ≈ 10, so fluctuations of order 10% away from the 50/50 macrostate are not rare. But push N up to a real gas sample — a mole, N ≈ 6×10²³ — and 1/√N shrinks to about 1.3×10⁻¹², meaning fluctuations away from equilibrium are confined to about a trillionth of a percent. That's why the second law feels absolute at human scales: it isn't that entropy-decreasing fluctuations are impossible, it's that for any system with Avogadro-scale particle counts they're so relatively improbable that 'never observed' is a fair description of something that is, in principle, merely fantastically unlikely.

This reframing also resolves an old worry, sometimes called Loschmidt's paradox, raised against Boltzmann in his own lifetime: if the microscopic laws of motion are time-reversible (as the arrow-of-time lesson established), how can a statistical argument built purely from those laws produce an irreversible-looking result? The resolution is that reversibility of the underlying dynamics and irreversibility of the observed behavior don't conflict — a low-entropy configuration can, by the time-symmetry of the microscopic laws, evolve toward an even lower-entropy one exactly as often as toward a higher-entropy one. But low-entropy macrostates are astronomically rare compared to high-entropy ones among all possible microstates, so starting from a low-entropy macrostate — as every real experiment does, having been deliberately prepared into some special, non-equilibrium configuration — the overwhelmingly likely direction to evolve is toward higher entropy, not because the laws prefer that direction, but because there is so much more of the entropy-increasing directions to go to. Formally, for any process inside an isolated system (or, more usefully, a system plus every environment it exchanges heat with — 'the universe' in the thermodynamic sense), EQ 1.3 below is the second law's precise statement, with equality only in the idealized limit of a perfectly reversible process.