Dimensional analysis — Buckingham π theorem: Visual walkthrough

Step 1 — What "dimensions" even are (a bag of three counters)

WHAT. We take a quantity like speed and strip away its number, keeping only its recipe of $M, L, T$ .

WHY. The whole theorem is about units cancelling. To see cancellation we must first see units as things you can add and subtract exponents of — a bookkeeping we can put on a picture.

PICTURE. In the figure, each quantity is three little counters. Speed $= L T^{- 1}$ has the $L$ -counter at $+ 1$ , the $T$ -counter at $- 1$ , the $M$ -counter at $0$ . Force $F = M L T^{- 2}$ lights all three up.

Step 2 — A quantity is an arrow of exponents

WHAT. We plot each quantity as one point/arrow in an $M$ – $L$ – $T$ space.

WHY. Once quantities are arrows, "combining quantities" becomes "adding arrows," and "dimensionless" becomes "the arrow lands at the origin." Geometry now does the reasoning for us. This is the single trick that makes the whole theorem visual.

PICTURE. Density $ρ = M L^{- 3}$ is the arrow to $(1, - 3, 0)$ ; viscosity $μ = M L^{- 1} T^{- 1}$ is $(1, - 1, - 1)$ . When you multiply quantities you add their arrows; when you raise to a power you stretch the arrow.

Step 3 — "Dimensionless" means the arrows cancel to the origin

WHAT. We look for a combination $q_{1}^{a_{1}} q_{2}^{a_{2}} \dots$ whose exponent-arrows sum to zero.

WHY. This is the object the theorem counts. A dimensionless group survives any change of units — metres to miles, grams to pounds — untouched. That invariance (Dimensional homogeneity) is why physics can be written with them.

PICTURE. Take $ρ, v, d$ and $F$ with unknown powers $a, b, c, 1$ . Each is an arrow; we scale and add them. We are asking: which scalings make the four arrows nose-to-tail return to the origin?

Step 4 — Stack the arrows: the dimension matrix

WHAT. We line up all $n$ arrows side by side as columns of one $k \times n$ table of exponents.

WHY. The zero-arrow condition from Step 3, for all variables at once, is a single tidy statement: $D a = 0$ , where $a$ holds the unknown powers. One matrix now encodes every possible dimensionless group. We can finally count them.

PICTURE. For the sphere-drag variables $F, d, v, ρ, μ$ the grid is $3 \times 5$ . Each column is a stalk of three numbers; the whole thing is a numeric fingerprint of the problem.

Step 5 — Rank: how many independent directions the arrows really fill

WHAT. We ask how many of the $M, L, T$ rows carry new information versus being copies/combinations of the others.

WHY. Every independent row is one hard constraint on $a$ — one equation the powers must obey. The number of real constraints is the rank, not merely "how many rows we wrote." (This is the fix to Mistake D in the parent: use $k = rank (D)$ , not "3 because the universe has 3.")

PICTURE. The three constraint equations are three planes through the origin in the space of powers $a$ . If all three planes are truly different (rank 3), they slice space down hard. If two coincide (a dependent dimension), you have fewer real planes.

Step 6 — Counting the survivors: $p = n - rank (D)$

WHAT. We count the free directions left after the rank-many planes have done their slicing.

WHY. This is the punchline. We start with $n$ free powers (one per variable). Each independent constraint (there are $rank (D)$ of them) pins down one power in terms of the rest. What remains free is exactly: $p = knobs we started with n - independent constraints rank (D) = n - k .$ This is the rank–nullity fact: (number of unknowns) − (rank) = (free parameters). Each free parameter is one independent $π$ group.

PICTURE. Picture a room. The full space of powers is $n$ -dimensional. The constraints are a wall of dimension $k$ you're pushed against. What's left to move freely — the floor space you can still walk on — has dimension $p = n - k$ . For sphere drag: $5 - 3 = 2$ .

Step 7 — Edge cases: when a dimension can't be balanced (and rank drops)

Case A — a lonely dimension (the pendulum's mass). In the pendulum problem the variables are $T, L, g, m$ , with only $m$ carrying $M$ . In the dimension matrix the $M$ -row is $(0, 0, 0, 1)$ — mass is the only nonzero entry. For the group to be dimensionless we need $a_{m} = 0$ : mass cannot appear. That's why $T = 2 π L / g$ has no $m$ . (Parent Mistake A.)

Case B — fake repeaters (rank collapse). If you pick repeating variables that are secretly dependent — like speed $v$ , diameter $d$ , and time $t$ where $t = d / v$ — their columns are not independent, so the rank drops below the number of columns you chose. Two constraint-planes coincide, the null space grows, and your "π groups" won't be independent. (Parent Mistake B.) Fix: the repeaters' own sub-matrix must be invertible.

WHAT / WHY. We inspect the two ways the picture degenerates: a dimension with only one carrier (forces a power to zero) and a linearly dependent variable set (drops the rank).

PICTURE. Left panel: the $M$ -row is a single spike on $m$ → its power is forced to $0$ , drawn as a padlocked axis. Right panel: three arrows that are coplanar (dependent) — they can't fill 3D, so a whole plane of extra solutions appears.

The one-picture summary

Recall Feynman retelling — the walkthrough in plain words

Every quantity is secretly a recipe: so many pinches of mass, so many of length, so many of time. Write those pinch-counts as an arrow. Multiplying quantities just adds the arrows; a unitless combo is one whose arrows perfectly cancel back to the starting point — zero pinches of everything. Now line up all your variable-arrows as columns of a grid. Asking "which combos cancel?" is asking the grid one clean question, and linear algebra answers it: you began with $n$ dials, each independent dimension jams one dial, and $k$ dimensions really bite, so $n - k$ dials stay free to turn. Those free dials are your $π$ groups. If a dimension shows up in only one variable, that variable's dial is jammed at zero — it can't appear (goodbye, pendulum mass). If two of your chosen "building block" variables are secretly the same recipe, one dial was never independent, the grid loses a bite, and your groups double up. Count honestly with the rank, and the number of knobs the universe left you is exactly $p = n - k$ .

Recall Quick self-check

Why does each independent dimension remove exactly one free power? ::: It is one independent equation $D a = 0$ row; rank–nullity says free powers $= n - (independent rows) = n - k$ . In the sphere-drag matrix, what is $rank (D)$ and $p$ ? ::: $rank (D) = 3$ , so $p = 5 - 3 = 2$ (drag coefficient and Reynolds number). What geometric event is "a dimensionless group"? ::: The scaled exponent-arrows summing nose-to-tail back to the origin $(0, 0, 0)$ . Why can't the pendulum's mass appear? ::: The $M$ -row is nonzero only in the $m$ column, forcing that power to $0$ .

Connections

2.2.26 Dimensional analysis — Buckingham π theorem (Hinglish) — the parent topic this walkthrough derives
Reynolds number — the second $π$ group falling out of Step 3–4
Drag force and drag coefficient — $π_{1} = F / ρ v^{2} d^{2}$ is exactly $C_{D}$
Dimensional homogeneity — why the zero-arrow condition is the whole game
Fundamental and derived units — where the $M, L, T$ axes come from
Model testing and similarity — matching the $p$ groups between model and prototype
Navier–Stokes equations — non-dimensionalising them produces $R e$ rigorously