Intuition The ONE core idea
A canonical transformation is just relabelling every point of "the space of all possible states" so that the machine that predicts the future (Hamilton's equations) still runs by the exact same rules. A single scalar recipe — the generating function — tells you how to relabel without breaking those rules.
Before you can read the parent note, you must be fluent in the words it throws around: phase space , coordinate , momentum , Hamiltonian , partial derivative , total time derivative , variation , new variables ( Q , P ) , generating function , Legendre transform , and Poisson bracket . This page builds each from nothing, in an order where every symbol is earned before it is used.
Definition Generalized coordinate
q
q is one number (or a list of numbers q 1 , q 2 , … ) that completely pins down where a system is at one instant. For a bead on a wire it is the distance along the wire; for a pendulum it is the swing angle.
Picture a single number line. The dot sitting on it at position q is the whole configuration.
Intuition Why "generalized"?
The word means we don't insist on x , y , z . A pendulum's natural coordinate is an angle , not a length. We pick whatever number is most convenient — that freedom is the whole point of this topic.
Definition Conjugate momentum
p
p is a second number that says how the system is changing — loosely, its "amount of motion" in the direction of q . For a plain mass, p = m q ˙ (mass times velocity).
Here q ˙ (read "q -dot") is our first piece of notation:
Definition The dot = rate of change in time
q ˙ means d t d q : how fast q changes each second. If q is measured in metres, q ˙ is in metres per second. Two dots q ¨ = acceleration.
Intuition Units of momentum (do this once, remember forever)
If q is in metres (m) and mass m is in kilograms (kg), then p = m q ˙ has units
[ p ] = kg ⋅ s m = kg ⋅ m ⋅ s − 1 .
So position and its conjugate momentum carry different units — a fact that matters the moment you start mixing q and p in one transformation.
Position alone can't predict the future: a pendulum at the bottom might be moving left, right, or resting. You need both where it is and how it moves. So plot them together.
Intuition What to notice in the figure above
The violet oval is one system's entire history — a closed loop means the motion repeats. The single magenta dot is the state right now : its horizontal position reads off q , its height reads off p . The orange arrow shows the dot creeping along the loop as time ticks. Read the picture as "the future is a walk along this curve."
A plane (or higher-dimensional space) whose horizontal axis is q and whose vertical axis is p . One dot = one complete state of the system, right now. As time passes, the dot traces a curve — the system's history.
Newton needed x and v ; Hamilton packages them as one point ( q , p ) . The genius of this topic is that a canonical transformation moves us to new axes on this same plane — a relabelling of the dots, not a change of physics. Those new axes get their own names, which we introduce next.
Definition New canonical variables
Q , P
Capital Q i = Q i ( q , p , t ) and P i = P i ( q , p , t ) are brand-new coordinates and momenta — each one a formula built from the old ( q , p ) (and possibly time t ). They label the same phase-space dots with different numbers, exactly like laying a new grid over an old map.
Intuition Why capitals, and why bother?
Capital letters are just a bookkeeping habit: lower-case = old labels, upper-case = new labels. We bother because a clever regrid can make the motion boring — a straight line instead of a tangled curve. The parent note's whole payoff (solving the oscillator with no differential equation) comes from choosing ( Q , P ) well.
Q must still be a length and P a momentum."
Why it feels right: the letters echo q , p . Fix: after a transformation, Q might be an angle and P an action (energy over frequency). The words "coordinate" and "momentum" survive only in the sense that they play the roles q and p played in Hamilton's equations — nothing about units is guaranteed.
H ( q , p , t )
A single number computed from the current state ( q , p ) — almost always the total energy (kinetic + potential) written in terms of q and p . Example: H = 2 m p 2 + 2 1 m ω 2 q 2 for a spring.
Intuition Picture: a height map over phase space
Imagine phase space as a floor and H as a height above every point — hills and valleys. The state-dot slides around this landscape, and H tells it which way to go. After a canonical transformation we get a new landscape, traditionally called K ( Q , P , t ) — the same physics seen over the new grid.
H depends on several inputs (q , p , maybe t ). To ask "how does H change if I nudge only p ?" we need a derivative that holds the others fixed.
Definition Partial derivative
∂ p ∂ H
The slope of H as you increase p by a tiny amount while keeping q and t frozen . The curly ∂ (not straight d ) is the flag that says "other variables held constant."
Intuition What to notice in the figure above
The violet bowl is the energy surface H sitting over the q , p floor. The magenta curve is what you get by slicing the bowl along a line of fixed q and letting only p vary. The steepness of that single curve is ∂ H / ∂ p — a slope measured along one axis only, not across the whole surface.
Intuition Why partial and not ordinary
d ?
Phase space has more than one direction. "The slope" is meaningless until you say along which axis . ∂ / ∂ q = slope walking east; ∂ / ∂ p = slope walking north. This is exactly the notation Hamilton's equations q ˙ = ∂ H / ∂ p , p ˙ = − ∂ H / ∂ q are built from.
The parent topic's whole quest is: change ( q , p ) → ( Q , P ) so these two equations keep their exact shape , with a possibly new landscape K :
Q ˙ = ∂ P ∂ K , P ˙ = − ∂ Q ∂ K .
A transformation that pulls this off is called canonical . See Hamilton's equations for the full derivation.
Here is the object the entire parent note revolves around, stated plainly so nothing later is a surprise.
Definition Generating function
F
A single scalar function — one recipe — from which the whole transformation ( q , p ) → ( Q , P ) is squeezed out by taking derivatives. You never guess the new variables; you write down one F and derive them.
Intuition What the four "types" mean (the numbering scheme)
F needs to depend on one old and one new variable per degree of freedom. There are exactly four ways to pick that pair, so there are four types :
Type 1: F 1 ( q , Q ) — old coordinate, new coordinate.
Type 2: F 2 ( q , P ) — old coordinate, new momentum.
Type 3: F 3 ( p , Q ) — old momentum, new coordinate.
Type 4: F 4 ( p , P ) — old momentum, new momentum.
The subscript on F is just this catalogue number. We'll meet F 1 (the "old q , new Q " recipe) again in the next two sections; the others are built from it by the swap trick below.
The master relation contains d t d F . This differs from the partial one, and mixing them up wrecks everything.
Definition Total vs partial time derivative
If F = F 1 ( q , Q , t ) and q , Q themselves change with time, then
d t d F = ∂ q ∂ F q ˙ + ∂ Q ∂ F Q ˙ + ∂ t ∂ F .
The total d / d t counts all the ways time sneaks in — directly through t , and indirectly through the moving variables. The partial ∂ F / ∂ t counts only the direct route (variables frozen).
d F / d t with ∂ F / ∂ t
They agree only if q and Q don't move — which they always do. The rule K = H + ∂ F / ∂ t from Section 8 uses the partial one; the master relation uses the total one. Keeping them straight is the whole game.
Definition The variation symbol
δ
δ means "imagine a slightly different trajectory between the same start and end points, and ask how a quantity changes." δ ∫ ( … ) d t = 0 says the true path makes the integral stationary (a valley or ridge among all nearby paths).
Intuition What to notice in the figure above
The navy line is the true path from a pinned start dot to a pinned end dot. The dashed magenta/orange/violet curves are nearby "imagined" paths sharing those same two endpoints. Because all of them start and end together, anything of the form ∫ d t d F d t = F ( end ) − F ( start ) is identical for every path — so it cannot affect which path wins.
Intuition Why endpoints must be pinned (this is why
F is allowed)
The reason two integrands may differ by d F / d t without changing the motion is exactly that ∫ d t d F d t = F ( end ) − F ( start ) , and with fixed endpoints its variation is zero . That is why a generating function is legal — the master relation of Section 8 rests entirely on this fact.
To turn a Type-1 recipe F 1 ( q , Q ) into a Type-2 recipe F 2 ( q , P ) we must trade the new coordinate Q for the new momentum P . That trade is a Legendre transform, and it is worth seeing step by step.
Definition Legendre swap — step by step
Goal: replace a variable Q by the slope P = ∂ F / ∂ Q as the thing we hold in our hand.
Start from the product rule for differentials: d ( P Q ) = P d Q + Q d P . (This is just the derivative of a product, written with d 's.)
Rearrange to isolate the d Q term: P d Q = d ( P Q ) − Q d P .
Read it: the left side is written in terms of d Q ; the right side is written in terms of d P (plus a clean boundary piece d ( P Q ) ). So a term "in d Q " has been converted into a term "in d P " .
Absorb the boundary piece d ( P Q ) into the definition of a new generator, F 2 = F 1 + P Q . What remains is naturally a function of P instead of Q .
That is the entire swap: a variable (Q ) is exchanged for its conjugate slope (P ), at the cost of one bookkeeping term P Q .
Intuition Why this exact trick appears in the topic
F 1 naturally lives in ( q , Q ) ; some problems are easier if the generator lives in ( q , P ) . Steps 1–4 are how you legally move between the four types without changing any physics. See Legendre transformation for the deep version; the very same trick turns the Lagrangian into the Hamiltonian.
Definition Poisson bracket (one degree of freedom)
For two functions f , g of phase space,
{ f , g } = ∂ q ∂ f ∂ p ∂ g − ∂ p ∂ f ∂ q ∂ g .
It measures how the two quantities intertwine in phase space.
Intuition Why it matters as a check
The generating-function machinery guarantees these brackets automatically, but you can also test a raw map ( q , p ) → ( Q , P ) directly with them — no generator needed. Full story in Poisson brackets .
Master relation with dF over dt
Canonical transformations
This flows straight into Symplectic geometry , Liouville's theorem , the Hamilton–Jacobi equation and Action–angle variables once the parent topic is mastered.
What does q ˙ mean, and in what units if q is metres? d q / d t , the time rate of change; units metres per second.
If q is in metres and mass in kilograms, what are the units of p = m q ˙ ? Kilogram-metres per second, kg ⋅ m ⋅ s − 1 — different units from q .
What single point in phase space represents? One complete state of the system: both where it is (q ) and how it moves (p ).
What are Q and P ? New coordinates and momenta, each a formula in the old ( q , p , t ) — a fresh grid on the same phase space, not guaranteed to keep the old units.
What is the physical meaning of H in most problems? The total energy (kinetic + potential) written as a function of q and p .
Why write ∂ H / ∂ p with a curly ∂ instead of d ? Because H has several inputs; ∂ means "vary p only, freeze q and t ."
State Hamilton's two equations. q ˙ = ∂ H / ∂ p and p ˙ = − ∂ H / ∂ q .
What is the generating function F , in one sentence? A single scalar recipe from which the whole transformation ( q , p ) → ( Q , P ) is obtained by differentiation.
State the master relation that all canonical transformations obey. ∑ p i q ˙ i − H = ∑ P i Q ˙ i − K + d t d F .
How does the new Hamiltonian K relate to H ? K = H + ∂ F / ∂ t ; equal only if F has no explicit time.
What are the four generator types keyed on? Which old/new pair F depends on: F 1 ( q , Q ) , F 2 ( q , P ) , F 3 ( p , Q ) , F 4 ( p , P ) .
How does d F / d t differ from ∂ F / ∂ t ? Total d F / d t adds the indirect changes through moving q , Q ; partial ∂ F / ∂ t counts only the explicit t dependence.
What algebraic identity powers the Legendre swap between generator types? P d Q = d ( P Q ) − Q d P , so a d Q term becomes a d P term plus a boundary piece.
Write the multi-dimensional Poisson bracket { f , g } . ∑ k ( ∂ q k ∂ f ∂ p k ∂ g − ∂ p k ∂ f ∂ q k ∂ g ) .
What fundamental brackets certify that ( q , p ) → ( Q , P ) is canonical? { Q i , P j } = δ ij and { Q i , Q j } = { P i , P j } = 0 .