nondimensionalization
/ non-dih-MEN-shun-al-ization /
Before solving a model, a seasoned modeler does something that looks like cleaning up but is really deep: they strip out the units. Instead of measuring time in seconds and length in metres, they re-measure everything in units natural to the problem itself — time in 'how long the process takes', length in 'how big the system is'. The equation, rewritten in these natural units, often collapses from a clutter of physical constants to a clean form with just one or two pure numbers left. This rescaling is nondimensionalization.
Concretely, you replace each variable by itself divided by a characteristic scale: write t = T tau, where T is a chosen time scale and tau is the new dimensionless time, and similarly for the other variables. Substitute into the differential equation, and the chain rule sweeps the scales out front; by choosing the scales cleverly you can make several constants equal 1 and vanish. A messy equation like m x'' + c x' + k x = 0, with three physical constants, becomes the tidy x'' + 2 zeta x' + x = 0, in which the only survivor is the damping ratio zeta — a single dimensionless number that controls everything.
The payoff is enormous. First, it reveals which combinations of parameters actually matter: the behaviour depends not on all the constants separately but on a few dimensionless groups (like the Reynolds number, the damping ratio, or R0), so two systems with the same group behave identically however different their raw numbers. Second, it tells you which terms are big and which are negligible, justifying approximations honestly. Third, it cuts the number of parameters you must explore, turning a forbidding study into a manageable one — solve it once in dimensionless form and rescale to any specific case.
A caution to keep straight: nondimensionalization rearranges and simplifies, but it never changes the actual physics or loses information — it is a reversible change of variables, and you can always scale back to real units. What it CAN do is mislead if you pick the wrong characteristic scales, making a genuinely large term look small. The art is choosing scales that match the regime you care about; the algebra is easy, the judgement is everything.
The logistic equation P' = r P (1 - P/K) carries two constants, r and K. Let p = P/K (population as a fraction of carrying capacity) and tau = r t (time in units of the growth rate). The equation becomes simply dp/dtau = p(1 - p) — every logistic model in the world, with all its r and K, reduces to this one universal dimensionless equation.
Two parameters vanish into the scales; one universal dimensionless equation remains.
Nondimensionalization loses no information — it is a reversible change of variables, not an approximation. But choosing the wrong characteristic scales can disguise a large term as small; the algebra is routine, picking scales that fit the regime of interest is the real skill.