The law of mass action: turning chemistry into an ODE
Picture a flask of molecules jostling in solution. A reaction happens when the right partners collide, so its speed should scale with how often they meet — and that frequency scales with how crowded each reactant is. This is the law of mass action: the rate of a reaction is proportional to the product of the concentrations of its reactants. Write a concentration in square brackets, like [A], and the chemical kinetics model practically writes itself as a system of ODEs.
Take the simplest case, A decaying into B. Each A has a fixed chance per second of transforming, so the disappearance rate is proportional to how much A is present: [A]' = -k [A]. You have met this equation a hundred times by now — it is plain exponential decay, the same separable equation that governs radioactive nuclei and cooling coffee. Its solution is [A](t) = [A]0 e^(-k t), and the chemist's half-life ln(2)/k is exactly the physicist's half-life wearing a lab coat.
The order of a reaction is just the total power of concentrations in its rate law, and it changes the equation's character completely. A first-order step gives the linear A' = -k A; a second-order step like A + A → product gives the nonlinear A' = -k A^2, whose solution 1/[A] = 1/[A]0 + k t decays much more slowly, only like 1/t. The shape of the rate law, read straight off the chemistry, is the difference between gentle exponential fade and stubborn algebraic tail.
Chains, equilibria, and a reaction that refuses to settle
Most real chemistry is a chain: A → B → C, where B is born from A and consumed into C. That gives a coupled pair, [A]' = -k1 [A] and [B]' = k1 [A] - k2 [B] — exactly the first-order system you learned to solve with eigenvalues. The middle species B is the interesting one: it rises while A is plentiful, then falls once A runs low, tracing a hump with a single peak. Hold that picture, because in a moment B will become a drug in your blood and that hump will be the dose you actually feel.
When a reaction is reversible, A ⇌ B, the forward and backward rates fight each other: [A]' = -kf [A] + kr [B]. The system stops changing not when everything has reacted but when the two rates balance — an equilibrium where kf [A] = kr [B], giving the equilibrium constant K = kf/kr = [B]/[A]. This is a sink in disguise: from any starting mix the concentrations slide toward that balance point and stay, exactly the stable-equilibrium behaviour you studied on the phase line.
Compartments: where chemistry becomes medicine
Now swap the flask for a body. Pharmacokinetics asks what the body does to a drug — how it is absorbed, distributed, and eliminated — and it answers with a compartment model. The trick is to pretend the body is a small number of well-mixed tanks. The simplest is one compartment: the bloodstream, with a drug amount A(t), losing drug to the kidneys and liver at a rate proportional to how much is there. That is again A' = -k A, the same first-order decay, because elimination machinery clears a fixed fraction per hour.
A tablet, though, does not appear in the blood all at once; it dissolves in the gut and seeps across. So we use two compartments — gut G and blood C — coupled exactly like the A → B → C chain above, with an absorption rate ka and an elimination rate ke. This is the classic pharmacokinetics model, and its blood-concentration curve is the same single-peaked hump we saw for the intermediate species B.
G' = -ka G, G(0) = D (whole dose in gut) C' = ka G - ke C, C(0) = 0 (none in blood yet) solution (ka != ke): C(t) = (D ka)/(ka - ke) * ( e^(-ke t) - e^(-ka t) ) rises from 0, peaks at t_max = ln(ka/ke)/(ka - ke), then fades like e^(-ke t)
Read that solution like a story. Early on the e^(-ka t) term is large and negative, so C climbs as the drug is absorbed; later that fast term dies and only e^(-ke t) survives, so C decays on the slow elimination timescale set by the half-life ln(2)/ke. The peak in between is the dose you actually feel; push it too high and you risk toxicity, let it sag too low and the drug stops working. The whole art of dosing — how much, how often — is reading this curve.
Steady state, repeated doses, and infusions
Real patients take a pill every eight hours, not once. Each dose adds a fresh hump on top of whatever is left from the last, and because the equation is linear, superposition lets you simply add the responses. The drug accumulates dose by dose, but not forever: elimination grows in step with the rising level until, after a few half-lives, what you add each interval equals what you clear. The blood level then plateaus around a steady state, oscillating gently between a peak and a trough.
A steady intravenous drip makes the same idea cleaner still. Now the input is a constant rate R, so C' = R - ke C — a first-order linear equation with constant forcing. Its solution is the textbook transient-plus-steady split: a transient e^(-ke t) that fades and a steady plateau C_ss = R/ke that remains. The drug climbs toward R/ke and levels off, and the speed of that approach is set entirely by the time constant 1/ke — about four time constants to get within a couple of percent of the target.
What the models hide, and how we test them
Be honest about the fictions. A body is not a stirred tank, so the well-mixed compartment is a deliberate cartoon; some drugs need a third compartment for slow-equilibrating tissues like fat, and even then it is an approximation. The mass-action rate constants k are not constants of nature either — they hide temperature, pH, and enzymes, and at high doses elimination machinery saturates so that first-order clearance breaks down into the nonlinear Michaelis-Menten kinetics. Knowing the limits of a model is not pessimism; it is the part of modeling that keeps it safe.
So how do we trust any of it? We measure. Draw blood at several times, then fit ka, ke, and the dose-scaling to the data — model fitting and validation. A neat diagnostic falls straight out of the math: once absorption is done, ln C(t) versus t is a straight line whose slope is -ke, so plotting the tail on a log scale reads the elimination rate off by eye. When the fitted curve tracks fresh measurements it did not see, the model has earned a little trust — and only a little.
Step back and admire the unity. A flask of molecules, a tablet in the gut, a radioactive sample, an RC circuit charging up — all of them are the same handful of first-order linear ODEs wearing different costumes. That is the quiet payoff of this whole ladder: once you recognize A' = -k A and its coupled cousins, you can read chemistry and medicine as fluently as you read a falling body, because underneath they were never really different equations at all.