Intuition The big pictureFor a system that does periodic motion (a pendulum swinging, a planet orbiting), the orbit in phase space is a closed loop (or a torus). Instead of tracking the messy ( q , p ) (q,p) ( q , p )
How big is the loop? → captured by the action J J J constant of the motion.Where along the loop are we? → captured by the angle θ \theta θ constant rate θ = ω t + θ 0 \theta = \omega t + \theta_0 θ = ω t + θ 0
So integrable systems are ones we can transform into coordinates where everything trivially marches in a straight line . The hard part (finding the frequencies) reduces to computing areas.
Solving Hamilton's equations q ˙ = ∂ H / ∂ p \dot q = \partial H/\partial p q ˙ = ∂ H / ∂ p p ˙ = − ∂ H / ∂ q \dot p = -\partial H/\partial q p ˙ = − ∂ H / ∂ q canonical transformation ( q , p ) → ( θ , J ) (q,p)\to(\theta,J) ( q , p ) → ( θ , J )
The new Hamiltonian depends on J J J only : H = H ( J ) H = H(J) H = H ( J )
Then J ˙ = − ∂ H / ∂ θ = 0 \dot J = -\partial H/\partial\theta = 0 J ˙ = − ∂ H / ∂ θ = 0 θ ˙ = ∂ H / ∂ J = ω ( J ) \dot\theta = \partial H/\partial J = \omega(J) θ ˙ = ∂ H / ∂ J = ω ( J ) one number per degree of freedom , the frequency ω \omega ω without ever solving the full time evolution . That's the 80/20 payoff.
Definition Action variableFor a 1-DOF system with periodic motion, the action is the phase-space area enclosed by one cycle, divided by 2 π 2\pi 2 π J ≡ 1 2 π ∮ p d q J \equiv \frac{1}{2\pi}\oint p\,dq J ≡ 2 π 1 ∮ p d q ∮ \oint ∮ one complete period of the bounded motion. J J J
Definition Angle variableIts canonical conjugate = = θ = = ==\theta== == θ == 2 π 2\pi 2 π θ ( t ) = ω t + θ 0 , ω = ∂ H ∂ J . \theta(t) = \omega t + \theta_0,\qquad \omega = \frac{\partial H}{\partial J}. θ ( t ) = ω t + θ 0 , ω = ∂ J ∂ H .
Recall Self-derivation checkWhy is ∮ d θ = 2 π \oint d\theta = 2\pi ∮ d θ = 2 π θ = ∂ W / ∂ J \theta=\partial W/\partial J θ = ∂ W / ∂ J ∮ d θ = d d J ∮ ∂ W ∂ q d q = d d J ∮ p d q = d d J ( 2 π J ) = 2 π . \oint d\theta = \frac{d}{dJ}\oint \frac{\partial W}{\partial q}dq = \frac{d}{dJ}\oint p\,dq = \frac{d}{dJ}(2\pi J)=2\pi. ∮ d θ = dJ d ∮ ∂ q ∂ W d q = dJ d ∮ p d q = dJ d ( 2 π J ) = 2 π . why J J J 2 π 2\pi 2 π
H = p 2 2 m + 1 2 m ω 0 2 q 2 = E H = \frac{p^2}{2m} + \frac12 m\omega_0^2 q^2 = E H = 2 m p 2 + 2 1 m ω 0 2 q 2 = E Step A — get p ( q , E ) p(q,E) p ( q , E ) p = ± 2 m E − m 2 ω 0 2 q 2 p = \pm\sqrt{2m E - m^2\omega_0^2 q^2} p = ± 2 m E − m 2 ω 0 2 q 2 Why? Just rearrange H = E H=E H = E p p p
Step B — the phase orbit is an ellipse with semi-axes a = 2 E / ( m ω 0 2 ) a=\sqrt{2E/(m\omega_0^2)} a = 2 E / ( m ω 0 2 ) q q q b = 2 m E b=\sqrt{2mE} b = 2 m E p p p Why? p 2 2 m E + q 2 2 E / m ω 0 2 = 1 \frac{p^2}{2mE}+\frac{q^2}{2E/m\omega_0^2}=1 2 m E p 2 + 2 E / m ω 0 2 q 2 = 1
Step C — area / 2 π 2\pi 2 π ∮ p d q = area = π a b = π 2 E m ω 0 2 2 m E = 2 π E ω 0 . \oint p\,dq = \text{area} = \pi a b = \pi\sqrt{\frac{2E}{m\omega_0^2}}\sqrt{2mE} = \frac{2\pi E}{\omega_0}. ∮ p d q = area = π ab = π m ω 0 2 2 E 2 m E = ω 0 2 π E . J = 1 2 π ⋅ 2 π E ω 0 = E ω 0 ⇒ E = ω 0 J . J = \frac{1}{2\pi}\cdot\frac{2\pi E}{\omega_0} = \frac{E}{\omega_0}\;\Rightarrow\; E = \omega_0 J. J = 2 π 1 ⋅ ω 0 2 π E = ω 0 E ⇒ E = ω 0 J . Why use ellipse area? Because ∮ p d q \oint p\,dq ∮ p d q is the enclosed area — no integration needed.
Step D — frequency: ω = d E d J = ω 0 . \omega = \dfrac{dE}{dJ} = \omega_0. ω = dJ d E = ω 0 .
Bonus (quantum link): E = ω 0 J E=\omega_0 J E = ω 0 J J = ℏ ( n + 1 2 ) J=\hbar(n+\tfrac12) J = ℏ ( n + 2 1 ) E n = ℏ ω 0 ( n + 1 2 ) E_n=\hbar\omega_0(n+\tfrac12) E n = ℏ ω 0 ( n + 2 1 )
Worked example Particle bouncing between walls at0 0 0 L L L E = p 2 / 2 m E=p^2/2m E = p 2 /2 m Step A: p = 2 m E p=\sqrt{2mE} p = 2 m E Step B — one period = go right and come back: ∮ p d q = p ⋅ L + p ⋅ L = 2 p L \oint p\,dq = p\cdot L + p\cdot L = 2pL ∮ p d q = p ⋅ L + p ⋅ L = 2 p L Why 2 L 2L 2 L A full cycle traverses the box twice (out and back).
Step C: J = 1 2 π 2 p L = p L π = L 2 m E π J=\frac{1}{2\pi}\,2pL=\frac{pL}{\pi}=\frac{L\sqrt{2mE}}{\pi} J = 2 π 1 2 p L = π p L = π L 2 m E E = π 2 J 2 2 m L 2 E = \frac{\pi^2 J^2}{2mL^2} E = 2 m L 2 π 2 J 2 Step D: ω = d E d J = π 2 J m L 2 = π 2 m L 2 ⋅ L 2 m E π = π v L \omega=\frac{dE}{dJ}=\frac{\pi^2 J}{mL^2}=\frac{\pi^2}{mL^2}\cdot\frac{L\sqrt{2mE}}{\pi}=\frac{\pi v}{L} ω = dJ d E = m L 2 π 2 J = m L 2 π 2 ⋅ π L 2 m E = L π v v = p / m v=p/m v = p / m T = 2 π / ω = 2 L / v T=2\pi/\omega = 2L/v T = 2 π / ω = 2 L / v
Definition Liouville integrabilityA system with n n n integrable if it has n n n F 1 , … , F n F_1,\dots,F_n F 1 , … , F n in involution : { F i , F j } = 0 \{F_i,F_j\}=0 { F i , F j } = 0 i , j i,j i , j
Intuition What integrability buys youThe Liouville–Arnold theorem says: the bounded level set { F i = c i } \{F_i=c_i\} { F i = c i } n n n T n T^n T n ( θ i , J i ) (\theta_i, J_i) ( θ i , J i ) J ˙ i = 0 , θ ˙ i = ω i ( J ) = ∂ H ∂ J i . \dot J_i = 0,\qquad \dot\theta_i = \omega_i(J)= \frac{\partial H}{\partial J_i}. J ˙ i = 0 , θ ˙ i = ω i ( J ) = ∂ J i ∂ H . straight-line winding on a doughnut with n n n ω i \omega_i ω i
J J J p p p Why it feels right: J J J p p p q q q J = L J=L J = L Fix: J J J area-integral 1 2 π ∮ p d q \frac{1}{2\pi}\oint p\,dq 2 π 1 ∮ p d q constant of motion averaged over a cycle , not the instantaneous p p p
Common mistake Forgetting the cycle goes there-and-back.Why it feels right: you integrate p d q p\,dq p d q q m i n q_{min} q min q m a x q_{max} q ma x Fix: the contour ∮ \oint ∮ full closed loop ; for libration you traverse q m i n → q m a x → q m i n q_{min}\to q_{max}\to q_{min} q min → q ma x → q min q q q 2 π 2\pi 2 π
H = H ( J ) H=H(J) H = H ( J ) Why it feels right: E E E J J J Fix: H = H ( J ) H=H(J) H = H ( J ) after the canonical transformation removes θ \theta θ ( q , p ) (q,p) ( q , p ) H H H q q q
Common mistake Assuming every system is integrable.Why it feels right: low-dimensional examples (SHO, Kepler) all work. Fix: integrability requires n n n most systems (e.g., double pendulum, three-body) lack them and are chaotic. Integrable systems are the rare, beautiful exceptions.
Mnemonic Remember the recipe"AREA → INVERT → DIFFERENTIATE"
Action = Area (/2π), invert to get E(J), differentiate to get ω.
And for the contour: "Round trip = whole loop."
Recall Feynman: explain it to a 12-year-oldImagine a kid on a swing. The swing always traces the same loop back and forth — out, back, out, back. Two facts describe everything: (1) how big the swing is (that's the action — basically the size of the loop, and it stays the same), and (2) where the kid is right now in the swing (that's the angle , which ticks forward steadily like a clock hand). Once you know the loop's size, you can predict how many swings per minute just from the size — you don't have to watch the whole motion. Scientists love these "swing-like" systems because they're the only ones we can fully predict. Most real things (like a chaotic spinning top) aren't this nice.
Hamiltonian mechanics — action-angle is a special canonical transformation.Canonical transformations and Generating functions — W ( q , J ) W(q,J) W ( q , J ) Hamilton–Jacobi equation — W W W Poisson brackets — involution { F i , F j } = 0 \{F_i,F_j\}=0 { F i , F j } = 0 Liouville–Arnold theorem — phase space tori.Adiabatic invariants — J J J Bohr–Sommerfeld quantization — ∮ p d q = 2 π ℏ ( n + 1 2 ) \oint p\,dq = 2\pi\hbar(n+\tfrac12) ∮ p d q = 2 π ℏ ( n + 2 1 ) KAM theorem — survival of tori under perturbation.
Define the action variable for 1-DOF periodic motion J = 1 2 π ∮ p d q J=\frac{1}{2\pi}\oint p\,dq J = 2 π 1 ∮ p d q , the phase-space area enclosed per cycle divided by
2 π 2\pi 2 π ; it is a constant of the motion.
How do you get the oscillation frequency from action-angle variables? ω = ∂ H / ∂ J = d E / d J \omega=\partial H/\partial J = dE/dJ ω = ∂ H / ∂ J = d E / dJ , by inverting
J ( E ) → E ( J ) J(E)\to E(J) J ( E ) → E ( J ) and differentiating.
For the harmonic oscillator, what is E E E J J J E = ω 0 J E=\omega_0 J E = ω 0 J , so
ω = d E / d J = ω 0 \omega=dE/dJ=\omega_0 ω = d E / dJ = ω 0 .
What does the angle variable do over one period? It advances by exactly
2 π 2\pi 2 π and evolves linearly:
θ = ω t + θ 0 \theta=\omega t+\theta_0 θ = ω t + θ 0 .
Why is J J J 1 / 2 π 1/2\pi 1/2 π So that
∮ d θ = 2 π \oint d\theta = 2\pi ∮ d θ = 2 π over one cycle, making
θ \theta θ a genuine angle.
What is Liouville integrability (n DOF)? Existence of
n n n independent conserved quantities in involution:
{ F i , F j } = 0 \{F_i,F_j\}=0 { F i , F j } = 0 .
What is the geometry of the bounded motion of an integrable system? It lies on an
n n n -dimensional invariant torus (Liouville–Arnold theorem); motion winds at constant frequencies.
Common factor-of-2 mistake in computing J J J Forgetting libration is a round trip
q m i n → q m a x → q m i n q_{min}\to q_{max}\to q_{min} q min → q ma x → q min , doubling the integral.
Which generating function builds action-angle variables? Hamilton's characteristic function
W ( q , J ) W(q,J) W ( q , J ) , a type-2 generator with
p = ∂ W / ∂ q p=\partial W/\partial q p = ∂ W / ∂ q ,
θ = ∂ W / ∂ J \theta=\partial W/\partial J θ = ∂ W / ∂ J .
How does action-angle connect to old quantum theory? Bohr–Sommerfeld:
J = ℏ ( n + 1 2 ) J=\hbar(n+\tfrac12) J = ℏ ( n + 2 1 ) , giving e.g. SHO levels
E n = ℏ ω 0 ( n + 1 2 ) E_n=\hbar\omega_0(n+\tfrac12) E n = ℏ ω 0 ( n + 2 1 ) .
Closed loop or torus in phase space
Intuition Hinglish mein samjhoDekho, jab koi system periodic motion karta hai — jaise pendulum jhoolta hai ya planet orbit karta hai — toh phase space (q vs p ka plot) me uska path ek band loop banata hai. Har baar wahi loop repeat hota hai. Ab idea simple hai: is loop ko describe karne ke liye sirf do cheez chahiye — (1) loop kitna bada hai, aur (2) abhi hum loop me kahan hai. Pehli cheez ko bolte hain action J = 1 2 π ∮ p d q J = \frac{1}{2\pi}\oint p\,dq J = 2 π 1 ∮ p d q area hai (divided by 2 π 2\pi 2 π angle θ \theta θ θ = ω t + θ 0 \theta=\omega t+\theta_0 θ = ω t + θ 0
Sabse mast baat: agar tum J J J E E E frequency seedha mil jaata hai — ω = d E / d J \omega = dE/dJ ω = d E / dJ = 2 π E / ω 0 =2\pi E/\omega_0 = 2 π E / ω 0 J = E / ω 0 J=E/\omega_0 J = E / ω 0 ω = d E / d J = ω 0 \omega=dE/dJ=\omega_0 ω = d E / dJ = ω 0
Multi-dimension me, agar system ke paas n n n integrable bolte hain. Aisa system ek torus (doughnut) ki surface par ghoomta hai, har direction me apni constant frequency ω i \omega_i ω i ∮ \oint ∮ J J J
