Chapter 8 · Classical Mechanics

8.4 Small Oscillations

Near a stable equilibrium every conservative system is a set of independent harmonic oscillators: expand $T$ and $V$ to second order, solve $\det(\mathsf V-\omega^2\mathsf T)=0$, and read off the normal frequencies, normal modes and normal coordinates — for coupled pendulums, the double pendulum, a triatomic molecule and a chain of springs.

Pull a guitar string, tap a wine glass, let an earthquake shake a building, shine infrared light on a carbon dioxide molecule: in each case a system sitting at a stable equilibrium is disturbed slightly and vibrates. The vibration looks complicated — every part moves, energy sloshes from one part to another — but it is always a superposition of a few very simple motions, the normal modes, in which every coordinate oscillates at one common frequency, in step, with fixed amplitude ratios. Finding those modes is a linear-algebra problem: two quadratic forms, the kinetic energy and the potential energy, are diagonalised simultaneously.

This section develops the theory in four steps: when an equilibrium is stable (the Lagrange–Dirichlet theorem); how $T$ and $V$ are replaced by quadratic forms near it; the generalised eigenvalue problem $\det(\mathsf V-\omega^2\mathsf T)=0$ and the normal coordinates that decouple the motion; and the standard examples — coupled pendulums and beats, the double pendulum, the linear triatomic molecule and spring–mass chains. It uses Lagrange's equations (§8.1) throughout and the spectral theorem for real symmetric matrices (Unit 1, linear algebra). CSIR NET examines this material through quick frequency computations ("the normal frequencies of the system with Lagrangian … are"), statements about normal modes to accept or reject, and stability questions about equilibria of a given potential.

Equilibrium and Stability

We work with a natural system: $n$ degrees of freedom, generalised coordinates $q=(q_1,\dots,q_n)$, and a Lagrangian $L=T-V$ with

$$T=\tfrac12\sum_{j,k}a_{jk}(q)\,\dot q_j\dot q_k,\qquad V=V(q),$$

where the matrix $a(q)=(a_{jk}(q))$ is symmetric and positive definite and $a$, $V$ are smooth. This covers every scleronomic system of particles and rigid bodies under conservative forces (§8.1); the total energy $E=T+V$ is then a constant of the motion.

📘 Definition 8.20 — Equilibrium configuration
A point $q^0$ is an equilibrium configuration if the constant function $q(t)\equiv q^0$ is a solution of Lagrange's equations; that is, the system placed at $q^0$ at rest stays there.
📐 Theorem 8.40 — Equilibria are critical points of $V$
For a natural system, $q^0$ is an equilibrium configuration if and only if $$\frac{\partial V}{\partial q_j}(q^0)=0\qquad(j=1,\dots,n).$$

Proof. Lagrange's equations read $$\frac{d}{dt}\Big(\sum_ka_{jk}\dot q_k\Big)-\frac12\sum_{k,l}\frac{\partial a_{kl}}{\partial q_j}\dot q_k\dot q_l+\frac{\partial V}{\partial q_j}=0 .$$ If $q\equiv q^0$ then $\dot q=\ddot q=0$ and the equations reduce to $\partial V/\partial q_j(q^0)=0$. Conversely, if the gradient vanishes at $q^0$, the constant function satisfies every equation. (Since $a$ is invertible, the equations can be solved for $\ddot q$ and form a smooth first-order system in $(q,\dot q)$; by the Picard–Lindelöf theorem (§1.1) the constant solution is then the only solution starting from $q^0$ at rest.) ∎

📘 Definition 8.21 — Stability (in the sense of Lyapunov)
An equilibrium $q^0$ is stable if for every $\varepsilon\gt0$ there is $\delta\gt0$ such that every motion with $|q(0)-q^0|\lt\delta$ and $|\dot q(0)|\lt\delta$ satisfies $|q(t)-q^0|\lt\varepsilon$ and $|\dot q(t)|\lt\varepsilon$ for all $t\ge0$. Otherwise it is unstable. (A conservative system can never be asymptotically stable: energy conservation forbids motions starting with $E\gt V(q^0)$ from settling at $q^0$.)
📐 Theorem 8.41 — Lagrange–Dirichlet theorem
If $V$ has a strict local minimum at $q^0$, then $q^0$ is a stable equilibrium. In particular this holds if $\nabla V(q^0)=0$ and the Hessian matrix of $V$ at $q^0$ is positive definite.

Proof. Translate so that $q^0=0$ and $V(0)=0$. Fix $\varepsilon\gt0$, small enough that $V(q)\gt0$ for $0\lt|q|\le\varepsilon$; on the compact ball $|q|\le\varepsilon$ the continuous matrix $a(q)$ has smallest eigenvalue at least some $c\gt0$, so $T\ge\tfrac c2|\dot q|^2$ there. Let $m=\min_{|q|=\varepsilon}V(q)\gt0$ (a continuous positive function on a compact sphere) and $\eta=\min\big(m,\tfrac c2\varepsilon^2\big)$. The energy $E(q,\dot q)=T+V$ is continuous with $E(0,0)=0$, so there is $\delta\in(0,\varepsilon)$ such that $|q|\lt\delta$, $|\dot q|\lt\delta$ imply $E\lt\eta$.

Take such initial data. Energy is conserved, so $E\lt\eta$ for all $t$. If $|q(t)|$ ever reached $\varepsilon$, then at the first such time $E\ge V(q(t))\ge m\ge\eta$ (as $T\ge0$) — impossible. So $|q(t)|\lt\varepsilon$ for all $t$; and then $V\ge0$ gives $\tfrac c2|\dot q|^2\le T\le E\lt\tfrac c2\varepsilon^2$, i.e. $|\dot q(t)|\lt\varepsilon$. If the Hessian is positive definite at a critical point, Taylor's theorem shows that the minimum is strict. ∎

The converse needs care. If the Hessian at $q^0$ has a negative eigenvalue, $q^0$ is unstable (a theorem of Lyapunov; Exercise I.5 proves the one-dimensional case of a strict maximum). But when the Hessian is only semidefinite, anything can happen: $V=x^4$ is stable, $V=-x^4$ and $V=x^3$ are unstable, and Painlevé's example $V(x)=e^{-1/x^2}\sin(1/x)$ is stable without being a minimum (Exercise I.8). Lagrange–Dirichlet is a sufficient condition, not a characterisation.

✏️ Example 1 — A bead on a rotating hoop
A smooth circular hoop of radius $a$ rotates with constant angular velocity $\Omega$ about its vertical diameter; a bead of mass $m$ slides on it. With $\theta$ the angle of the bead from the lowest point, its Lagrangian is (§8.1) $$L=\tfrac12ma^2\big(\dot\theta^2+\Omega^2\sin^2\theta\big)+mga\cos\theta .$$ Find the equilibria (relative to the hoop), decide their stability, and find the frequencies of small oscillations. Evaluate for $a=0.2\ \text{m}$, $\Omega=10\ \text{rad/s}$, $g=9.8\ \text{m/s}^2$.
Effective potential $L=T_2-U$ with $T_2=\tfrac12ma^2\dot\theta^2$ and $$U(\theta)=-mga\cos\theta-\tfrac12ma^2\Omega^2\sin^2\theta .$$ $L$ has no explicit $t$, so the Jacobi integral $T_2+U$ is conserved (§8.1), and the whole theory above applies with $U$ in place of $V$.
Equilibria $U'(\theta)=ma\sin\theta\,(g-a\Omega^2\cos\theta)=0$: $\theta=0$, $\theta=\pi$, and, if $\Omega^2\gt g/a$, the pair $\pm\theta^*$ with $\cos\theta^*=\dfrac{g}{a\Omega^2}$.
Stability $U''(\theta)=ma\big[g\cos\theta-a\Omega^2\cos2\theta\big]$, and the frequency is $\omega^2=U''/(ma^2)$. At $\theta=0$: $\omega^2=\tfrac ga-\Omega^2$ — stable iff $\Omega^2\lt g/a$. At $\theta=\pi$: $\omega^2=-\tfrac ga-\Omega^2\lt0$ — always unstable. At $\theta^*$: $U''=ma^2\Omega^2\sin^2\theta^*\gt0$ — stable, with $$\omega^2=\Omega^2\sin^2\theta^*=\Omega^2-\frac{g^2}{a^2\Omega^2}.$$
Numbers $g/a=49\ \text{s}^{-2}\lt\Omega^2=100\ \text{s}^{-2}$: the bottom is unstable ($\omega^2=-51$). $\cos\theta^*=0.49$, $\theta^*\approx60.7^\circ$, and $\omega=\sqrt{100-24.01}\approx8.72\ \text{rad/s}$. As $\Omega$ increases through $\sqrt{g/a}=7\ \text{rad/s}$, the bottom loses stability and two new stable equilibria branch off — a pitchfork bifurcation.

The Quadratic Approximation

Near an equilibrium we put $q=q^0+\eta$ with $\eta$ small and keep only the terms of lowest order. Two symmetric matrices carry all the information.

📘 Definition 8.22 — Mass and stiffness matrices
At an equilibrium $q^0$ the mass (inertia) matrix and the stiffness (potential) matrix are the $n\times n$ symmetric matrices $$\mathsf T=\big(a_{jk}(q^0)\big),\qquad \mathsf V=\Big(\frac{\partial^2V}{\partial q_j\partial q_k}(q^0)\Big).$$ $\mathsf T$ is positive definite; $\mathsf V$ is the Hessian of $V$ at $q^0$.
📐 Theorem 8.42 — The linearised (small-oscillation) equations
Up to an additive constant and terms of order three in $(\eta,\dot\eta)$, the Lagrangian near $q^0$ is $$L_2=\tfrac12\,\dot\eta^T\mathsf T\,\dot\eta-\tfrac12\,\eta^T\mathsf V\,\eta ,$$ and Lagrange's equations of $L_2$, $$\mathsf T\,\ddot\eta+\mathsf V\,\eta=0,$$ are exactly the linearisation at $(q^0,0)$ of the full equations of motion.

Proof. By Taylor's theorem $a_{jk}(q^0+\eta)=a_{jk}(q^0)+O(|\eta|)$, so $T=\tfrac12\dot\eta^T\mathsf T\dot\eta+O(|\eta|\,|\dot\eta|^2)$, the remainder being of order three. Also $$\begin{aligned}V(q^0+\eta)&=V(q^0)+\nabla V(q^0)\cdot\eta\\&\quad+\tfrac12\eta^T\mathsf V\eta+O(|\eta|^3),\end{aligned}$$ and $\nabla V(q^0)=0$ by Theorem 8.40. Dropping the constant $V(q^0)$ gives $L_2$, whose Lagrange equations are $\mathsf T\ddot\eta+\mathsf V\eta=0$. In the full equations (proof of Theorem 8.40), the term $\sum_ka_{jk}(q)\ddot q_k$ has linear part $\sum_k\mathsf T_{jk}\ddot\eta_k$; the terms quadratic in $\dot q$ have no linear part; and $\partial V/\partial q_j(q^0+\eta)=\sum_k\mathsf V_{jk}\eta_k+O(|\eta|^2)$. So the linearisation is the same system. ∎

Two practical remarks. First, $\mathsf T$ is $a(q)$ evaluated at the equilibrium — the $q$-dependence of the kinetic energy never matters to this order (in the double pendulum, $\cos(\theta_1-\theta_2)$ in $T$ simply becomes $1$). Second, $V$ must be expanded honestly to second order: a first-order expansion of a spring's length, for example, misses the stiffness produced by its tension (Exercise II.6 and the ⚠️ box).

Normal Modes and Normal Coordinates

Try a solution in which all coordinates move together: $\eta(t)=\mathbf a\cos(\omega t+\varphi)$ with a constant vector $\mathbf a\ne\mathbf 0$. Substituting into $\mathsf T\ddot\eta+\mathsf V\eta=0$ gives $(\mathsf V-\omega^2\mathsf T)\mathbf a=\mathbf 0$, so $\omega^2$ must be a root of the secular equation $$\det(\mathsf V-\lambda\mathsf T)=0 .$$ This is a generalised eigenvalue problem. The next theorem is the whole theory of small oscillations in one statement.

📐 Theorem 8.43 — Simultaneous diagonalisation of $\mathsf T$ and $\mathsf V$
Let $\mathsf T$ be real symmetric positive definite and $\mathsf V$ real symmetric, both $n\times n$. Then:
  1. the $n$ roots $\lambda_1\le\dots\le\lambda_n$ of $\det(\mathsf V-\lambda\mathsf T)=0$ (with multiplicity) are real;
  2. there is a real nonsingular matrix $P=[\mathbf a_1\,\cdots\,\mathbf a_n]$ with $$P^T\mathsf T\,P=E,\qquad P^T\mathsf V\,P=\operatorname{diag}(\lambda_1,\dots,\lambda_n);$$ its columns satisfy $\mathsf V\mathbf a_k=\lambda_k\mathsf T\mathbf a_k$ and $\mathbf a_j^T\mathsf T\mathbf a_k=\delta_{jk}$ ($\mathsf T$-orthonormality);
  3. all $\lambda_k\gt0$ iff $\mathsf V$ is positive definite; all $\lambda_k\ge0$ iff $\mathsf V$ is positive semidefinite; and the number of $\lambda_k=0$ is $\dim\ker\mathsf V$.

Proof. Since $\mathsf T$ is positive definite it has a symmetric positive definite square root $\mathsf T^{1/2}$ (diagonalise $\mathsf T=Q_0DQ_0^T$ and put $\mathsf T^{1/2}=Q_0D^{1/2}Q_0^T$); write $R=\mathsf T^{1/2}$. Then $\mathsf V-\lambda\mathsf T=R\,(S-\lambda E)\,R$ with $$S=R^{-1}\mathsf V R^{-1},$$ a real symmetric matrix, so $\det(\mathsf V-\lambda\mathsf T)=(\det R)^2\det(S-\lambda E)$: the roots are the eigenvalues of $S$, which are real (spectral theorem). Choose an orthogonal $Q$ with $Q^TSQ=\Lambda=\operatorname{diag}(\lambda_k)$ and put $P=R^{-1}Q$. Then $P^T\mathsf TP=Q^TR^{-1}R^2R^{-1}Q=Q^TQ=E$ and $P^T\mathsf VP=Q^TSQ=\Lambda$. Multiplying $P^T\mathsf VP=\Lambda$ on the left by $P^{-T}=\mathsf TP$ (from the first identity) gives $\mathsf VP=\mathsf TP\Lambda$, i.e. $\mathsf V\mathbf a_k=\lambda_k\mathsf T\mathbf a_k$. For (3), with $\mathbf y=P^{-1}\mathbf x$ we have $\mathbf x^T\mathsf V\mathbf x=\mathbf y^T\Lambda\mathbf y=\sum\lambda_ky_k^2$, which is positive (nonnegative) for all $\mathbf x\ne\mathbf 0$ iff every $\lambda_k$ is; and $\ker\mathsf V=P\ker\Lambda$ has dimension equal to the number of zero $\lambda_k$. ∎

The reality and orthogonality can also be seen directly, exactly as for Sturm–Liouville eigenfunctions (§2.3): if $\mathsf V\mathbf a=\lambda\mathsf T\mathbf a$ with $\mathbf a$ complex, multiply by $\bar{\mathbf a}^T$; both $\bar{\mathbf a}^T\mathsf V\mathbf a$ and $\bar{\mathbf a}^T\mathsf T\mathbf a\gt0$ are real, so $\lambda$ is real. And if $\mathsf V\mathbf a_j=\lambda_j\mathsf T\mathbf a_j$, $\mathsf V\mathbf a_k=\lambda_k\mathsf T\mathbf a_k$, then $(\lambda_j-\lambda_k)\,\mathbf a_j^T\mathsf T\mathbf a_k=0$ by symmetry of both matrices.

📘 Definition 8.23 — Normal modes, normal frequencies, normal coordinates
In the setting of Theorem 8.43 with $\mathsf V$ positive definite:
  • $\omega_k=\sqrt{\lambda_k}$ are the normal frequencies (characteristic or eigen-frequencies);
  • the vector $\mathbf a_k$ (the mode shape) and the motion $\eta(t)=\mathbf a_k\cos(\omega_kt+\varphi)$ form the $k$-th normal mode;
  • the normal coordinates are $\xi=P^{-1}\eta=P^T\mathsf T\eta$, so that $\eta=\sum_k\xi_k\mathbf a_k$.
A root $\lambda_k=0$ gives a zero mode (a direction in which the system can move without restoring force).
📐 Theorem 8.44 — Decoupling: the general small oscillation
In normal coordinates the quadratic Lagrangian separates: $$L_2=\tfrac12\sum_{k=1}^n\big(\dot\xi_k^2-\lambda_k\xi_k^2\big),\qquad \ddot\xi_k+\lambda_k\xi_k=0 .$$ Hence every solution of $\mathsf T\ddot\eta+\mathsf V\eta=0$ is $\eta(t)=\sum_k\xi_k(t)\,\mathbf a_k$ with
  • $\xi_k=C_k\cos(\omega_kt+\varphi_k)$ if $\lambda_k=\omega_k^2\gt0$;
  • $\xi_k=\alpha_k+\beta_kt$ if $\lambda_k=0$;
  • $\xi_k=\alpha_ke^{\mu_kt}+\beta_ke^{-\mu_kt}$, $\mu_k=\sqrt{-\lambda_k}$, if $\lambda_k\lt0$.
The constants are fixed by $\xi_k(0)=\mathbf a_k^T\mathsf T\,\eta(0)$ and $\dot\xi_k(0)=\mathbf a_k^T\mathsf T\,\dot\eta(0)$. In particular every solution of the linearised equations is bounded if and only if $\mathsf V$ is positive definite.

Proof. Substitute $\eta=P\xi$: $\dot\eta^T\mathsf T\dot\eta=\dot\xi^TP^T\mathsf TP\dot\xi=\sum\dot\xi_k^2$ and $\eta^T\mathsf V\eta=\sum\lambda_k\xi_k^2$ by Theorem 8.43. The Lagrange equation for $\xi_k$ is $\ddot\xi_k+\lambda_k\xi_k=0$, whose solutions are listed. Since $P$ is invertible, $\xi=P^{-1}\eta=P^T\mathsf T\eta$ (because $P^T\mathsf TP=E$), which gives the initial values. If all $\lambda_k\gt0$, each $\xi_k$ is bounded, hence so is $\eta=P\xi$; if some $\lambda_k\le0$, the initial data $\eta(0)=\mathbf 0$, $\dot\eta(0)=\mathbf a_k$ give $\xi_k=t$ or $\sinh(\mu_kt)/\mu_k$, unbounded. ∎

Two consequences are used constantly. The frequencies do not depend on the choice of generalised coordinates: under a linear change $\eta=Q\eta'$, $\mathsf T'=Q^T\mathsf TQ$ and $\mathsf V'=Q^T\mathsf VQ$, so $\det(\mathsf V'-\lambda\mathsf T')=(\det Q)^2\det(\mathsf V-\lambda\mathsf T)$. And the sum and product of the $\omega_k^2$ are cheap checks: $\sum\omega_k^2=\operatorname{tr}(\mathsf T^{-1}\mathsf V)$ and $\prod\omega_k^2=\det\mathsf V/\det\mathsf T$.

📐 Theorem 8.45 — Rayleigh's principle
With $\mathsf T$ positive definite and $\mathsf V$ symmetric, define the Rayleigh quotient $R(\eta)=\dfrac{\eta^T\mathsf V\eta}{\eta^T\mathsf T\eta}$ for $\eta\ne\mathbf 0$. Then $$\lambda_1=\min_{\eta\ne\mathbf 0}R(\eta),\qquad \lambda_n=\max_{\eta\ne\mathbf 0}R(\eta),$$ attained at $\mathbf a_1$ and $\mathbf a_n$. More generally (Courant–Fischer) $\lambda_k=\min_{\dim W=k}\ \max_{\eta\in W\setminus\{0\}}R(\eta)$. Consequently, making a system stiffer ($\mathsf V\to\mathsf V+\mathsf W$ with $\mathsf W$ positive semidefinite) cannot lower any normal frequency, and making it heavier ($\mathsf T\to\mathsf T+\mathsf W$) cannot raise any.

Proof. With $\eta=P\xi$, $R(\eta)=\dfrac{\sum\lambda_k\xi_k^2}{\sum\xi_k^2}$, a weighted average of the $\lambda_k$ with weights $\xi_k^2\ge0$; it lies in $[\lambda_1,\lambda_n]$, with the ends attained at $\xi=\mathbf e_1$ and $\xi=\mathbf e_n$. The Courant–Fischer formula is the same argument applied to subspaces (as for symmetric matrices in Unit 1). If $\mathsf V$ is replaced by $\mathsf V+\mathsf W$, the quotient $R$ increases or stays the same at every $\eta$, so every min–max value does too; a heavier $\mathsf T$ decreases $R$ everywhere. ∎

Rayleigh's principle is a practical tool: a sensible guess for the shape of the slowest mode gives an upper bound for $\omega_1^2$ that is usually very accurate, because an error of order $\epsilon$ in the shape produces an error of order $\epsilon^2$ in $R$ (Example 5).

🧭 Strategy — Solving a small-oscillation problem
  1. Choose generalised coordinates and write $T$ and $V$ exactly.
  2. Find the equilibria from $\nabla V=0$ (for steadily rotating systems use the effective potential, as in Example 1).
  3. Read off $\mathsf T=a(q^0)$ and the Hessian $\mathsf V$; check positive definiteness of $\mathsf V$ for stability.
  4. Solve $\det(\mathsf V-\lambda\mathsf T)=0$; check $\sum\lambda_k=\operatorname{tr}(\mathsf T^{-1}\mathsf V)$ and $\prod\lambda_k=\det\mathsf V/\det\mathsf T$.
  5. For each $\lambda_k$ solve $(\mathsf V-\lambda_k\mathsf T)\mathbf a_k=\mathbf 0$; expect symmetric and antisymmetric modes in symmetric systems; check $\mathbf a_j^T\mathsf T\mathbf a_k=0$.
  6. Fit initial conditions with $\xi_k(0)=\mathbf a_k^T\mathsf T\eta(0)/\mathbf a_k^T\mathsf T\mathbf a_k$ (no normalisation needed then) and similarly for $\dot\xi_k(0)$.

Coupled Pendulums, Beats and the Double Pendulum

✏️ Example 2 — Two pendulums joined by a spring
Two identical simple pendulums (mass $m$, length $\ell$) hang side by side; their bobs are joined by a light spring of stiffness $k$, unstretched when both hang vertically. (a) Find the normal frequencies, modes and normal coordinates. (b) Pendulum 1 is pulled aside a distance $a$ and released from rest with pendulum 2 at rest at equilibrium. Describe the motion. (c) Take $\ell=1\ \text{m}$, $g=9.8\ \text{m/s}^2$, $k/m=1\ \text{s}^{-2}$.
Set-up Use the horizontal displacements $x_1,x_2$ of the bobs ($x_j\approx\ell\theta_j$). To second order $T=\tfrac12m(\dot x_1^2+\dot x_2^2)$, gravity contributes $mg\ell(1-\cos\theta_j)\approx\dfrac{mg}{2\ell}x_j^2$, and the spring $\tfrac12k(x_1-x_2)^2$. So with $\omega_0^2=g/\ell$: $$\mathsf T=mE,\qquad \mathsf V=\begin{pmatrix}m\omega_0^2+k&-k\\-k&m\omega_0^2+k\end{pmatrix}.$$
(a) Modes $\det(\mathsf V-\lambda\mathsf T)=(m\omega_0^2+k-\lambda m)^2-k^2$, so $\lambda m=m\omega_0^2+k\mp k$: $$\omega_1^2=\frac g\ell,\qquad \omega_2^2=\frac g\ell+\frac{2k}m .$$ Mode 1: $\mathbf a_1\propto(1,1)$ — the pendulums swing together, the spring is never stretched, so the frequency is that of a single pendulum. Mode 2: $\mathbf a_2\propto(1,-1)$ — they swing in opposition and the spring adds stiffness. Normal coordinates (up to normalisation): $\xi_1\propto x_1+x_2$, $\xi_2\propto x_1-x_2$.
(b) Beats From rest, $x_1+x_2=a\cos\omega_1t$ and $x_1-x_2=a\cos\omega_2t$. With $\bar\omega=\tfrac12(\omega_1+\omega_2)$ and $\Delta=\omega_2-\omega_1$, $$x_1=a\cos\tfrac{\Delta t}2\,\cos\bar\omega t,$$ $$x_2=a\sin\tfrac{\Delta t}2\,\sin\bar\omega t .$$ Each bob oscillates at the mean frequency $\bar\omega$ inside a slowly varying envelope. At $t=\pi/\Delta$ pendulum 1 is momentarily still and all the energy is in pendulum 2; at $t=2\pi/\Delta$ it is back.
(c) Numbers $\omega_1=\sqrt{9.8}\approx3.130\ \text{rad/s}$, $\omega_2=\sqrt{11.8}\approx3.435\ \text{rad/s}$, $\Delta\approx0.305\ \text{rad/s}$. The energy passes completely from one pendulum to the other in $\pi/\Delta\approx10.3\ \text{s}$, about five swings; the full cycle takes $2\pi/\Delta\approx20.6\ \text{s}$. Weaker springs give slower exchange: for $k/m=0.2\ \text{s}^{-2}$ the transfer takes about $50\ \text{s}$.
📘 Definition 8.24 — Beats
The superposition of two oscillations of nearly equal frequencies $\omega_1\lt\omega_2$, $$\cos\omega_1t+\cos\omega_2t=2\cos\tfrac{(\omega_2-\omega_1)t}{2}\,\cos\tfrac{(\omega_1+\omega_2)t}{2},$$ is a fast oscillation at the mean frequency whose amplitude varies slowly. The amplitude (or energy) envelope repeats with period $2\pi/(\omega_2-\omega_1)$; the number $\omega_2-\omega_1$ is the beat frequency. In weakly coupled identical oscillators the beats are the periodic transfer of energy from one oscillator to the other.
Interactive 1 — Coupled Pendulums: Two Normal Modes and the Beats Between Them

Two pendulums of length $1\ \text{m}$ ($g=9.8\ \text{m/s}^2$) coupled by a spring with $k/m$ set by the slider. Start them in mode 1, in mode 2, or with only pendulum 1 displaced. Left: the pendulums (swing exaggerated for visibility). Right: $x_1(t)$ (red) and $x_2(t)$ (blue) over the last $24\ \text{s}$. In a mode the two curves keep a fixed ratio; otherwise watch the energy swap at the beat frequency $\omega_2-\omega_1$.

k/m—
ω₁ (in phase)—
ω₂ (opposite)—
transfer time π/(ω₂ − ω₁)—
energy share E₁ : E₂—
t—
✏️ Example 3 — The double pendulum
A double pendulum has two equal masses $m$ and two light rods of equal length $\ell$; $\theta_1,\theta_2$ are the angles of the rods from the downward vertical. Find the normal frequencies and mode shapes, and check the orthogonality relations.
Exact $T$, $V$ (§8.1) $$T=\tfrac12m\ell^2\big[2\dot\theta_1^2+\dot\theta_2^2+2\dot\theta_1\dot\theta_2\cos(\theta_1-\theta_2)\big],$$ $$V=-mg\ell(2\cos\theta_1+\cos\theta_2).$$
Matrices At $\theta_1=\theta_2=0$ the cosine becomes $1$, and $\partial^2V/\partial\theta_1^2=2mg\ell$, $\partial^2V/\partial\theta_2^2=mg\ell$, mixed $0$: $$\mathsf T=m\ell^2\begin{pmatrix}2&1\\1&1\end{pmatrix},\quad \mathsf V=mg\ell\begin{pmatrix}2&0\\0&1\end{pmatrix}.$$
Frequencies With $\lambda=\omega^2$ and $\omega_0^2=g/\ell$: $\det(\mathsf V-\lambda\mathsf T)\propto2(\omega_0^2-\lambda)^2-\lambda^2=0$, so $\sqrt2(\omega_0^2-\lambda)=\pm\lambda$ and $$\omega^2=(2\mp\sqrt2)\,\frac g\ell,$$ i.e. $\omega_1\approx0.765\sqrt{g/\ell}$ and $\omega_2\approx1.848\sqrt{g/\ell}$. Check: $\operatorname{tr}(\mathsf T^{-1}\mathsf V)=4\omega_0^2$ and $\det\mathsf V/\det\mathsf T=2\omega_0^4$, matching $(2-\sqrt2)+(2+\sqrt2)=4$ and $(2-\sqrt2)(2+\sqrt2)=2$.
Modes The first row of $(\mathsf V-\lambda\mathsf T)\mathbf a=\mathbf 0$ gives $\theta_2/\theta_1=2(\omega_0^2-\lambda)/\lambda$: $+\sqrt2$ for the slow mode (rods swing together, the lower one further) and $-\sqrt2$ for the fast mode (they swing in opposition). Orthogonality: $(1,\sqrt2)\,\mathsf T\,(1,-\sqrt2)^T$ equals $m\ell^2\big[2-\sqrt2+\sqrt2-2\big]=0$, but the Euclidean dot product is $1-2=-1\ne0$. The ratio $\omega_2/\omega_1=\sqrt{3+2\sqrt2}=1+\sqrt2$ is irrational, so a general small motion of the double pendulum is never periodic.

Molecules and Chains

✏️ Example 4 — The linear triatomic molecule
A symmetric linear molecule such as $\mathrm{CO_2}$ is modelled by masses $m,M,m$ on a line joined by two springs of stiffness $k$. Find the normal frequencies and modes of motion along the line.
Matrices With displacements $x_1,x_2,x_3$: $T=\tfrac12m(\dot x_1^2+\dot x_3^2)+\tfrac12M\dot x_2^2$ and $V=\tfrac12k\big[(x_2-x_1)^2+(x_3-x_2)^2\big]$, so $$\mathsf T=\operatorname{diag}(m,M,m),$$ $$\mathsf V=k\begin{pmatrix}1&-1&0\\-1&2&-1\\0&-1&1\end{pmatrix}.$$
Secular equation Expanding, $\det(\mathsf V-\lambda\mathsf T)$ equals $$-\lambda\,(k-\lambda m)\,\big(k(M+2m)-\lambda mM\big),$$ so $$\omega^2=0,\qquad \frac km,\qquad \frac km\Big(1+\frac{2m}M\Big).$$
Modes $\omega=0$: $(1,1,1)$, a rigid translation — a zero mode, since $\mathsf V(1,1,1)^T=\mathbf 0$. $\omega^2=k/m$: $(1,0,-1)$, the symmetric stretch: the centre atom stays still and the outer ones move oppositely. $\omega^2=\tfrac km(1+\tfrac{2m}M)$: $(1,-\tfrac{2m}M,1)$, the antisymmetric stretch: the outer atoms move together and the centre atom against them, keeping the centre of mass fixed ($m+M(-\tfrac{2m}M)+m=0$).
CO₂ With $m=16$, $M=12$ the model predicts $\omega_3/\omega_2=\sqrt{1+\tfrac{32}{12}}\approx1.91$. Spectroscopy gives about $1.76$ (wavenumbers near $2349$ and $1333\ \text{cm}^{-1}$): the springs are not independent. The symmetric stretch does not change the dipole moment and is invisible in infrared absorption; the antisymmetric stretch is strongly infrared-active and produces CO₂'s absorption band near $4.3\ \mu\text{m}$. (The bending mode, which a model of motion along the line leaves out, gives the band near $15\ \mu\text{m}$ that matters most for the greenhouse effect.)
📐 Theorem 8.46 — The uniform spring–mass chain
Let $N$ equal masses $m$ lie on a line between two fixed walls, joined to each other and to the walls by $N+1$ equal springs of stiffness $k$. The normal frequencies and modes of motion along the line are $$\omega_r=2\sqrt{\frac km}\,\sin\frac{r\pi}{2(N+1)},$$ $$a_r(j)=\sin\frac{jr\pi}{N+1}\qquad(j,r=1,\dots,N).$$

Proof. With displacements $\eta_1,\dots,\eta_N$ and $\eta_0=\eta_{N+1}=0$ (the walls), $V=\tfrac12k\sum_{j=0}^{N}(\eta_{j+1}-\eta_j)^2$ and $\mathsf T=mE$, so the equations are $m\ddot\eta_j=k(\eta_{j+1}-2\eta_j+\eta_{j-1})$ and the eigenproblem is $k(2a_j-a_{j+1}-a_{j-1})=\lambda m\,a_j$. Try $a_j=\sin j\phi$. Since $\sin(j+1)\phi+\sin(j-1)\phi=2\cos\phi\,\sin j\phi$, the equation holds for $1\le j\le N$ with $\lambda m=2k(1-\cos\phi)=4k\sin^2\tfrac\phi2$, provided the boundary values are right: $a_0=\sin0=0$ automatically, and $a_{N+1}=\sin(N+1)\phi=0$ requires $\phi=\tfrac{r\pi}{N+1}$. For $r=1,\dots,N$ the values $\lambda_r=\tfrac{4k}{m}\sin^2\tfrac{r\pi}{2(N+1)}$ are distinct (the sine is increasing on $(0,\tfrac\pi2)$) and the vectors $\mathbf a_r$ are nonzero, so these are all $N$ normal modes. ∎

Mode $r$ has $r-1$ interior sign changes — the discrete analogue of the $r$-th harmonic of a string. Letting $N\to\infty$ with total length $\ell=(N+1)h$, $m=\rho h$ and $k=\tau/h$ fixed in the right way gives $\omega_r\to\tfrac{r\pi}{\ell}\sqrt{\tau/\rho}$, the frequencies of the vibrating string of §4.3 (Exercise III.8).

✏️ Example 5 — Three masses between walls, and a Rayleigh estimate
For the chain of Theorem 8.46 with $N=3$, list the frequencies and modes and check them directly. Then estimate $\omega_1^2$ by Rayleigh's principle with the trial shapes $(1,1,1)$ and $(1,1.4,1)$.
Theorem 8.46 $\omega_r^2=\tfrac{4k}m\sin^2\tfrac{r\pi}8$: since $4\sin^2\tfrac\pi8=2-\sqrt2$, $4\sin^2\tfrac\pi4=2$, $4\sin^2\tfrac{3\pi}8=2+\sqrt2$, $$\omega^2=(2-\sqrt2)\tfrac km,\ \ 2\tfrac km,\ \ (2+\sqrt2)\tfrac km .$$ Modes $\big(\sin\tfrac{jr\pi}4\big)_j$: $(1,\sqrt2,1)$, $(1,0,-1)$, $(1,-\sqrt2,1)$.
Direct check $\mathsf V=k\begin{pmatrix}2&-1&0\\-1&2&-1\\0&-1&2\end{pmatrix}$. For $(1,\sqrt2,1)$: the first row gives $2-\sqrt2$ and the second $-1+2\sqrt2-1=\sqrt2(2-\sqrt2)$, so $\mathsf V\mathbf a=(2-\sqrt2)k\,\mathbf a$. Sum check: $\operatorname{tr}\mathsf V/m=6k/m$ and $(2-\sqrt2)+2+(2+\sqrt2)=6$; product: $\det\mathsf V=4k^3$ and $(2-\sqrt2)\cdot2\cdot(2+\sqrt2)=4$.
Rayleigh $R(\eta)=\dfrac{k\,\eta^T\!\left(\begin{smallmatrix}2&-1&0\\-1&2&-1\\0&-1&2\end{smallmatrix}\right)\eta}{m\,|\eta|^2}$. For $(1,1,1)$: numerator $2k$, denominator $3m$, $R=0.667\,k/m$. For $(1,1.4,1)$: numerator $k(2\cdot1+2\cdot1.96+2-4\cdot1.4)=2.32\,k$, denominator $3.96\,m$, $R\approx0.58586\,k/m$.
Comparison Exact: $(2-\sqrt2)\,k/m\approx0.58579\,k/m$. Both estimates are upper bounds (Theorem 8.45); the crude guess is $14\%$ high, while $(1,1.4,1)$ — within $1\%$ of the true shape $(1,1.414,1)$ — is within about $0.012\%$: the error in $R$ is quadratic in the error of the shape.
Interactive 2 — Mode Explorer: Chains, a Triatomic Molecule and the Double Pendulum

The lab solves $\det(\mathsf V-\lambda\mathsf T)=0$ numerically (Cholesky reduction plus Jacobi rotations — the proof of Theorem 8.43 as an algorithm) and animates the chosen normal mode. Top: the physical motion (exaggerated). Bottom: the displacements drawn as stems, with the mode shape as a dashed envelope. Right: the spectrum of normal frequencies; the selected one is highlighted. Choose "all modes mixed" to see a general — non-periodic — small oscillation.

parameter—
selected ω—
all ω (units below)—
mode shape—
Σω² = tr(T⁻¹V)—
✏️ Example 6 — Normal modes: prove or disprove
Prove or disprove: (a) the normal frequencies are the square roots of the eigenvalues of $\mathsf V$; (b) the matrix $\mathsf T^{-1}\mathsf V$ is symmetric; (c) if every diagonal entry of $\mathsf V$ is positive, the equilibrium is stable; (d) if $\mathsf V$ is positive definite, every small oscillation is bounded, but it need not be periodic.
(a) False Take $\mathsf T=\operatorname{diag}(1,4)$, $\mathsf V=E$ (two uncoupled oscillators with masses $1$ and $4$, equal springs). The eigenvalues of $\mathsf V$ are $1,1$, but $\det(\mathsf V-\lambda\mathsf T)=(1-\lambda)(1-4\lambda)$ gives $\omega=1$ and $\omega=\tfrac12$. Only when $\mathsf T$ is a multiple of $E$ do the two agree.
(b) False For the double pendulum (Example 3, units $m=\ell=g=1$), $\mathsf T^{-1}=\left(\begin{smallmatrix}1&-1\\-1&2\end{smallmatrix}\right)$ and $\mathsf T^{-1}\mathsf V=\left(\begin{smallmatrix}2&-1\\-2&2\end{smallmatrix}\right)$, not symmetric. It is, however, similar to the symmetric matrix $S=\mathsf T^{-1/2}\mathsf V\mathsf T^{-1/2}$, which is why its eigenvalues $2\pm\sqrt2$ are real.
(c) False $T=\tfrac12(\dot x^2+\dot y^2)$, $V=x^2+4xy+y^2$: $\mathsf V=\left(\begin{smallmatrix}2&4\\4&2\end{smallmatrix}\right)$ has positive diagonal but eigenvalues $6$ and $-2$; along $(1,-1)$ the potential is $-2s^2$, a saddle, and the motion $\eta=s(t)(1,-1)$ with $\ddot s=2s$ runs away. Positive diagonal entries say only that each coordinate alone is restored.
(d) True Boundedness is Theorem 8.44. Periodicity fails when two excited frequencies are incommensurable: for the double pendulum $\omega_2/\omega_1=1+\sqrt2$, and $\theta_1(t)=\cos\omega_1t+\cos\omega_2t$ is not periodic (a periodic sum would force $\omega_2/\omega_1\in\mathbb Q$). Such motion is quasi-periodic.
⚠️ Counterexamples and pitfalls
  1. Frequencies come from $\mathsf T^{-1}\mathsf V$, not from $\mathsf V$. With $\mathsf T=\operatorname{diag}(1,4)$, $\mathsf V=E$ the frequencies are $1$ and $\tfrac12$, not $1,1$.
  2. Modes are $\mathsf T$-orthogonal, not orthogonal. The double-pendulum modes $(1,\sqrt2)$ and $(1,-\sqrt2)$ have Euclidean dot product $-1$.
  3. Evaluate $\mathsf T$ at the equilibrium. Terms like $\cos(\theta_1-\theta_2)$ in $T$ become constants; keeping them, or "linearising" them, is a common source of wrong matrices.
  4. Expand $V$ to second order — including geometric terms. A mass at the centre of a string under tension $\tau$ (length $2\ell$), displaced sideways by $y$, has potential $\tau(\sqrt{\ell^2+y^2}-\ell)\cdot2\approx\tfrac{\tau}{\ell}y^2$: a "first-order" treatment of the length gives no restoring force at all.
  5. Zero Hessian is not instability. $V=x^4$ is stable, but its oscillations are not harmonic: the period is proportional to $1/(\text{amplitude})$ (Exercise I.2). Linear theory says nothing here.
  6. A minimum is sufficient, not necessary. Painlevé's $V=e^{-1/x^2}\sin(1/x)$ is stable at $0$ although $0$ is not a minimum; a fast "sleeping" top is stable upright although its potential is a maximum (gyroscopic stabilisation, not covered by Lagrange–Dirichlet).
  7. Positive on the axes does not mean positive definite. $x^2+4xy+y^2$ is positive on both axes and still a saddle. Check the eigenvalues or the leading minors of $\mathsf V$.
  8. Zero modes are not oscillations. A free molecule has $\omega=0$ for translation (and rotation): the linearised solution $\xi=\alpha+\beta t$ drifts. Count them: a free linear triatomic molecule moving along its axis has one zero mode and two vibrations.
  9. Degenerate frequencies make modes non-unique. For a spherical pendulum or the three-mass ring, any $\mathsf T$-orthonormal basis of the eigenspace is a valid set of modes.
  10. Rotating and rheonomic systems. For the bead on a rotating hoop use the effective potential $U=V-T_0$, and only $T_2$ in $\mathsf T$; using $V$ alone predicts the wrong equilibria.
  11. Complete energy transfer needs symmetry. For two different pendulums coupled weakly, the energy exchange in beats is only partial.

Exercises

I
Equilibrium and stability (one degree of freedom unless stated; $T=\tfrac12m\dot x^2$).
  • 1 For $V(x)=x^3-3x$, find the equilibria, decide their stability, and give the frequency of small oscillations where it exists.
  • 2 For $V(x)=x^4$ and $m=1$, show that $x=0$ is stable although $V''(0)=0$, and that the period of an oscillation of amplitude $A$ is $\tau=\dfrac{2\sqrt2}{A}\displaystyle\int_0^1\frac{du}{\sqrt{1-u^4}}\approx\frac{3.708}{A}$.
  • 3 In Example 1 take $\Omega^2=2g/a$. Find the stable equilibria and the frequency of small oscillations about them.
  • 4 A mass $m$ hangs from a vertical spring of stiffness $k$. Show that gravity shifts the equilibrium but does not change the frequency of small vertical oscillations.
  • 5 Prove that if $V$ has a strict local maximum at $x_0$ (one degree of freedom), the equilibrium $x_0$ is unstable.
  • 6 With $T=\tfrac12(\dot x^2+\dot y^2)$ and $V=x^2+4xy+y^2$, show that the origin is an unstable equilibrium although $V\gt0$ at every nonzero point of both coordinate axes. Find an explicit runaway solution.
  • 7 For $V(x)=(x^2-1)^2$ and $m=1$, find all equilibria and the frequencies of small oscillations about the stable ones.
  • 8 (Painlevé.) Let $V(x)=e^{-1/x^2}\sin(1/x)$ for $x\ne0$ and $V(0)=0$. Show that $0$ is an equilibrium that is stable but is not a local minimum of $V$.
1 $V'=3x^2-3=0$ at $x=\pm1$, and $V''=6x$. At $x=1$, $V''=6\gt0$: stable, $\omega=\sqrt{6/m}$. At $x=-1$, $V''=-6$: unstable (a strict maximum, Exercise I.5).
2 $x=0$ is a strict minimum of $x^4$, so Lagrange–Dirichlet applies. Energy: $\tfrac12\dot x^2+x^4=A^4$, so the quarter period is $\int_0^A\frac{dx}{\sqrt{2(A^4-x^4)}}$; with $x=Au$ this is $\frac1{\sqrt2A}\int_0^1\frac{du}{\sqrt{1-u^4}}$. Multiply by $4$; the integral is $1.31103$, so $\tau\approx3.708/A$. Large swings are fast and small ones slow — there is no single frequency.
3 $\cos\theta^*=\tfrac{g}{a\Omega^2}=\tfrac12$, so $\theta^*=\pm60^\circ$ (the bottom is unstable since $\Omega^2\gt g/a$). $\omega^2=\Omega^2\sin^2\theta^*=\tfrac{2g}a\cdot\tfrac34=\tfrac{3g}{2a}$.
4 With $z$ the downward extension, $V=\tfrac12kz^2-mgz$; $V'=0$ at $z_0=mg/k$ and $V''=k$, so $\omega=\sqrt{k/m}$, independent of $g$. In general a constant force only shifts the equilibrium of a linear spring.
5 Take $V(x_0)=0$ and an interval $J=[x_0-\varepsilon,x_0+\varepsilon]$ on which $V(x)\lt0$ for $x\ne x_0$ and $a(x)\le c$. Start at $x_0$ with speed $v_0\gt0$: then $E=\tfrac12a(x_0)v_0^2\gt0$, and while $x\in J$, $\tfrac12a(x)\dot x^2=E-V(x)\ge E$, so $|\dot x|\ge\sqrt{2E/c}$. The velocity never vanishes in $J$, so $x$ moves monotonically and leaves $J$ within time $\varepsilon/\sqrt{2E/c}$. However small $v_0$ is, the motion leaves $J$: unstable.
6 $\mathsf V=\left(\begin{smallmatrix}2&4\\4&2\end{smallmatrix}\right)$ has eigenvalues $6$ (along $(1,1)$) and $-2$ (along $(1,-1)$): a saddle. Here $V$ is exactly quadratic, so the equations $\ddot x=-2x-4y$, $\ddot y=-4x-2y$ are exact; $x=-y=s$ gives $\ddot s=2s$, e.g. $s=\epsilon\cosh(\sqrt2\,t)$, which starts arbitrarily close to rest at the origin and runs away.
7 $V'=4x(x^2-1)=0$ at $0,\pm1$; $V''=12x^2-4$. At $\pm1$: $V''=8$, stable, $\omega=2\sqrt2$. At $0$: $V''=-4$, unstable (the top of the barrier between the two wells).
8 $V$ is $C^\infty$ with all derivatives $0$ at $0$, so $0$ is an equilibrium. It is not a minimum: $V(x)\lt0$ where $\sin(1/x)\lt0$, at points arbitrarily close to $0$. Stability: given $\varepsilon$, pick $0\lt x_+\lt\varepsilon$ with $\sin(1/x_+)=1$ and $-\varepsilon\lt x_-\lt0$ with $\sin(1/x_-)=1$, so $h=\min V(x_\pm)\gt0$. If the initial point lies in $(x_-,x_+)$ and $E\lt h$, the motion can never reach $x_\pm$ (there $V\ge h\gt E$), so it stays in $(-\varepsilon,\varepsilon)$; and $\tfrac12m\dot x^2=E-V\le E+e^{-1/\varepsilon^2}$ keeps the speed small. Choosing $\delta$ so small that $E\lt h$ and $E$ is tiny completes the proof. The wells accumulating at $0$ trap the motion.
II
Two degrees of freedom.
  • 1 Two masses $m$ on a smooth line are joined to each other by a spring $k'$ and each to a fixed wall by a spring $k$ (wall–$k$–$m$–$k'$–$m$–$k$–wall). Find the normal frequencies and modes.
  • 2 A double pendulum has equal rod lengths $\ell$, upper mass $m_1=3m$ and lower mass $m_2=m$. Find the normal frequencies and mode shapes.
  • 3 For a double pendulum with equal lengths and general masses, show that $\omega^2=\dfrac{g/\ell}{1\pm\sqrt{m_2/(m_1+m_2)}}$, and deduce $\omega_1\lt\sqrt{g/\ell}\lt\omega_2$. What happens as $m_2/m_1\to0$?
  • 4 In Example 2 take $\ell=0.5\ \text{m}$, $g=9.81\ \text{m/s}^2$ and $k/m=0.5\ \text{s}^{-2}$. Find $\omega_1$, $\omega_2$ and the time for the energy to pass from one pendulum to the other.
  • 5 For weak coupling ($\omega_2-\omega_1\ll\omega_1$) in Example 2(b), show that the energies $E_j=\tfrac12m(\dot x_j^2+\omega_0^2x_j^2)$ are approximately $E\cos^2\tfrac{\Delta t}2$ and $E\sin^2\tfrac{\Delta t}2$.
  • 6 (Spring pendulum.) A mass $m$ hangs from a spring of stiffness $k$ and natural length $\ell_0$ and moves in a vertical plane. Using polar coordinates $(r,\theta)$, find the equilibrium and the normal frequencies. For which $k$ is the bouncing frequency exactly twice the swinging frequency?
  • 7 With $T=\tfrac12m(\dot x^2+\dot y^2)$ and $V=\tfrac12(k_1x^2+k_2y^2)+c\,xy$, find the normal frequencies and the exact condition for stability.
  • 8 Two masses $m_1,m_2$ joined by a spring $k$ move freely on a line. Find the normal frequencies and modes and interpret the zero mode.
  • 9 Show that for small oscillations of a spherical pendulum every horizontal direction is a normal mode, and give two different $\mathsf T$-orthonormal bases of modes.
1 $\mathsf T=mE$, $\mathsf V=\left(\begin{smallmatrix}k+k'&-k'\\-k'&k+k'\end{smallmatrix}\right)$. Modes $(1,1)$ with $\omega^2=k/m$ (the middle spring is idle) and $(1,-1)$ with $\omega^2=(k+2k')/m$.
2 $\mathsf T=m\ell^2\left(\begin{smallmatrix}4&1\\1&1\end{smallmatrix}\right)$, $\mathsf V=mg\ell\left(\begin{smallmatrix}4&0\\0&1\end{smallmatrix}\right)$. With $s=\lambda\ell/g$: $(4-4s)(1-s)-s^2=0$, i.e. $4(1-s)^2=s^2$, $s=\tfrac23$ or $2$. So $\omega^2=\tfrac{2g}{3\ell}$ and $\tfrac{2g}{\ell}$. From the first row, $\theta_2/\theta_1=4(1-s)/s$: $+2$ (slow, in phase) and $-2$ (fast, opposite).
3 With $M=m_1+m_2$ and $s=\lambda\ell/g$: $\det\propto M(1-s)^2-m_2s^2$, so $(1-s)\sqrt M=\pm\sqrt{m_2}\,s$ and $s=\dfrac1{1\pm r}$, $r=\sqrt{m_2/M}\in(0,1)$. Hence $s_1=\tfrac1{1+r}\lt1\lt\tfrac1{1-r}=s_2$. As $m_2/m_1\to0$, $r\to0$ and both frequencies tend to $\sqrt{g/\ell}$: a light lower bob barely disturbs the upper pendulum, and itself swings like a simple pendulum hung from it.
4 $\omega_1=\sqrt{19.62}\approx4.429\ \text{rad/s}$, $\omega_2=\sqrt{20.62}\approx4.541\ \text{rad/s}$, $\Delta\approx0.1115\ \text{rad/s}$; the transfer time is $\pi/\Delta\approx28.2\ \text{s}$.
5 $x_1=a\cos\tfrac{\Delta t}2\cos\bar\omega t$. When $\Delta\ll\bar\omega$ the envelope is nearly constant over a fast cycle, so $\dot x_1\approx-a\bar\omega\cos\tfrac{\Delta t}2\sin\bar\omega t$, and with $\bar\omega\approx\omega_0$, $E_1\approx\tfrac12ma^2\omega_0^2\cos^2\tfrac{\Delta t}2$. Similarly $E_2\approx\tfrac12ma^2\omega_0^2\sin^2\tfrac{\Delta t}2$. The total $\tfrac12ma^2\omega_0^2=E$ (the spring energy is of smaller order), shared out at the beat frequency.
6 $T=\tfrac12m(\dot r^2+r^2\dot\theta^2)$, $V=\tfrac12k(r-\ell_0)^2-mgr\cos\theta$. Equilibrium $\theta=0$, $r=\ell=\ell_0+mg/k$. There $\mathsf T=\operatorname{diag}(m,m\ell^2)$ and $\mathsf V=\operatorname{diag}(k,mg\ell)$: $\omega_r=\sqrt{k/m}$ (bouncing) and $\omega_\theta=\sqrt{g/\ell}$ (swinging), uncoupled at this order. $\omega_r=2\omega_\theta$ when $k/m=4g/\ell$; then the neglected cubic terms pump energy back and forth between bouncing and swinging (an autoparametric resonance).
7 $\omega^2=\dfrac{k_1+k_2\pm\sqrt{(k_1-k_2)^2+4c^2}}{2m}$. Stable iff both are positive iff $\mathsf V$ is positive definite iff $k_1\gt0$ and $k_1k_2\gt c^2$. If $k_1k_2\lt c^2$ the smaller root is negative (a saddle), as in Exercise I.6.
8 $\mathsf T=\operatorname{diag}(m_1,m_2)$, $\mathsf V=k\left(\begin{smallmatrix}1&-1\\-1&1\end{smallmatrix}\right)$: $\lambda=0$ with mode $(1,1)$ — uniform translation, the motion of the centre of mass ($\xi=\alpha+\beta t$); and $\lambda=k\big(\tfrac1{m_1}+\tfrac1{m_2}\big)=k/\mu$ ($\mu$ the reduced mass) with mode $(m_2,-m_1)$, which keeps the centre of mass fixed.
9 With horizontal displacements $(x,y)$ of the bob, $\mathsf T=mE$ and $\mathsf V=\tfrac{mg}\ell E$, so $\mathsf V-\lambda\mathsf T=0$ for $\lambda=g/\ell$: every vector is a mode. Bases: $\tfrac1{\sqrt m}(1,0),\tfrac1{\sqrt m}(0,1)$, or $\tfrac1{\sqrt{2m}}(1,1),\tfrac1{\sqrt{2m}}(1,-1)$. (A combination with a phase difference, such as an elliptical swing, is a superposition of two modes, not a mode.)
III
Several degrees of freedom.
  • 1 Use Theorem 8.46 to list the frequencies of the chains with $N=2$ and $N=4$, and check the case $N=2$ directly.
  • 2 For CO₂ ($m=16$, $M=12$ in atomic mass units), compute the ratio of the two vibrational frequencies of Example 4. Why does the model overestimate it?
  • 3 Three equal masses $m$ slide on a smooth circular ring and are joined in a closed loop by three equal springs $k$ (along the ring). Find the normal frequencies and describe the modes.
  • 4 A chain of $N$ masses $m$ is attached by springs $k$ to a wall at the left end only (the last mass is free on the right). Show that $\omega_r=2\sqrt{k/m}\,\sin\dfrac{(2r-1)\pi}{2(2N+1)}$, $r=1,\dots,N$, and check $N=1,2$.
  • 5 Estimate the slow frequency of the double pendulum of Example 3 by Rayleigh's principle with the trial vectors $(1,1)$ and $(1,1.5)$.
  • 6 Prove that adding a spring to a system cannot lower any normal frequency. Check it for the chain $N=2$ after a second spring $k$ is added between the first mass and the wall.
  • 7 Find $\mathsf T$-normalised normal coordinates for Example 2 and write $L_2$ in terms of them.
  • 8 In Theorem 8.46 let $N\to\infty$ with $(N+1)h=\ell$, $m=\rho h$ and $k=\tau/h$. Show that $\omega_r\to\dfrac{r\pi}{\ell}\sqrt{\tau/\rho}$ and that the modes tend to $\sin(r\pi x/\ell)$ — the vibrating string of §4.3.
1 $N=2$: $\omega^2=4\sin^2\tfrac\pi6\,\tfrac km=\tfrac km$ and $4\sin^2\tfrac\pi3\,\tfrac km=\tfrac{3k}m$; directly, $\mathsf V=k\left(\begin{smallmatrix}2&-1\\-1&2\end{smallmatrix}\right)$ has eigenvalues $k$, $3k$. $N=4$: $\omega^2/(k/m)=4\sin^2\tfrac{r\pi}{10}$, i.e. $\tfrac{3-\sqrt5}2\approx0.382$, $\tfrac{5-\sqrt5}2\approx1.382$, $\tfrac{3+\sqrt5}2\approx2.618$, $\tfrac{5+\sqrt5}2\approx3.618$.
2 $\sqrt{1+2m/M}=\sqrt{11/3}\approx1.915$, against about $1.76$ observed. Real bonds interact (stretching one changes the other's force constant), and the symmetric stretch is shifted by a resonance with an overtone of the bending mode; a two-independent-springs model cannot capture either.
3 $$V=\tfrac12k\big[(s_1-s_2)^2+(s_2-s_3)^2+(s_3-s_1)^2\big],$$ $\mathsf V=k\left(\begin{smallmatrix}2&-1&-1\\-1&2&-1\\-1&-1&2\end{smallmatrix}\right)$, $\mathsf T=mE$. Eigenvalues $0$ (mode $(1,1,1)$: the whole loop rotating rigidly) and $3k/m$ (double: every $(s_1,s_2,s_3)$ with $s_1+s_2+s_3=0$, e.g. $(1,-1,0)$ and $(1,1,-2)$).
4 The equations are those of Theorem 8.46 for $j\lt N$, and $m\ddot\eta_N=k(\eta_{N-1}-\eta_N)$ for the last mass, equivalent to the fictitious condition $\eta_{N+1}=\eta_N$. With $a_j=\sin j\phi$ the interior equations give $\lambda=\tfrac{4k}m\sin^2\tfrac\phi2$, and $\sin(N+1)\phi=\sin N\phi$ means $2\cos\tfrac{(2N+1)\phi}2\sin\tfrac\phi2=0$, so $\phi=\tfrac{(2r-1)\pi}{2N+1}$. $N=1$: $\omega=2\sqrt{k/m}\sin\tfrac\pi6=\sqrt{k/m}$, a mass on one spring. $N=2$: $\mathsf V=k\left(\begin{smallmatrix}2&-1\\-1&1\end{smallmatrix}\right)$ has eigenvalues $\tfrac{3\mp\sqrt5}2k$, equal to $4\sin^2\tfrac\pi{10}k$ and $4\sin^2\tfrac{3\pi}{10}k$.
5 In units $m=\ell=g=1$: for $(1,1)$, $\eta^T\mathsf V\eta=3$ and $\eta^T\mathsf T\eta=2+1+2=5$, so $R=0.600$; for $(1,1.5)$, $R=\tfrac{2+2.25}{2+2.25+3}=\tfrac{4.25}{7.25}\approx0.5862$. Exact: $2-\sqrt2\approx0.5858$. Both are upper bounds; the second is within $0.1\%$.
6 Adding a spring adds $\tfrac12\eta^T\mathsf W\eta$ with $\mathsf W$ positive semidefinite, so $R$ does not decrease at any $\eta$, and by Courant–Fischer (Theorem 8.45) no $\lambda_k$ decreases. Check: $\mathsf V=k\left(\begin{smallmatrix}3&-1\\-1&2\end{smallmatrix}\right)$ gives $\lambda=\tfrac{5\mp\sqrt5}2\tfrac km\approx1.382,\ 3.618$ (times $k/m$), versus $1$ and $3$ before.
7 $\mathbf a_1=\tfrac1{\sqrt{2m}}(1,1)$, $\mathbf a_2=\tfrac1{\sqrt{2m}}(1,-1)$ satisfy $\mathbf a_j^T\mathsf T\mathbf a_k=\delta_{jk}$. Then $\xi=P^T\mathsf T\eta$: $\xi_1=\sqrt{\tfrac m2}(x_1+x_2)$, $\xi_2=\sqrt{\tfrac m2}(x_1-x_2)$, and $$L_2=\tfrac12\big(\dot\xi_1^2-\omega_1^2\xi_1^2\big)+\tfrac12\big(\dot\xi_2^2-\omega_2^2\xi_2^2\big).$$ (Check: $\xi_1^2+\xi_2^2=m(x_1^2+x_2^2)$.)
8 $\sqrt{k/m}=\sqrt{\tau/\rho}/h$ and $\sin\tfrac{r\pi}{2(N+1)}=\sin\tfrac{r\pi h}{2\ell}\approx\tfrac{r\pi h}{2\ell}$ for fixed $r$, so $\omega_r\to\tfrac{r\pi}\ell\sqrt{\tau/\rho}$. The mass $j$ sits at $x_j=jh$, and $\sin\tfrac{jr\pi}{N+1}=\sin\tfrac{r\pi x_j}\ell$. High modes ($r$ comparable to $N$) do not converge: the chain's spectrum is bounded by $2\sqrt{k/m}$, the string's is not.
IV
Proofs, invariants and forcing.
  • 1 Prove directly that if $\mathsf T$ is positive definite and $\mathsf V$ positive semidefinite, every root of $\det(\mathsf V-\lambda\mathsf T)=0$ is $\ge0$, and that $\lambda=0$ occurs with multiplicity $\dim\ker\mathsf V$.
  • 2 Prove that two modes with distinct frequencies are $\mathsf T$-orthogonal and also $\mathsf V$-orthogonal.
  • 3 Show that $\mathsf T^{-1}\mathsf V$ is similar to a symmetric matrix, hence diagonalisable with real eigenvalues, though not symmetric in general.
  • 4 Prove that the normal frequencies do not change under a linear change of coordinates $\eta=Q\eta'$ ($Q$ invertible).
  • 5 A force $\mathbf F\cos pt$ acts on the system: $\mathsf T\ddot\eta+\mathsf V\eta=\mathbf F\cos pt$. Find the steady response in normal coordinates and show that resonance at $p=\omega_k$ occurs only if $\mathbf a_k^T\mathbf F\ne0$. Which mode of Example 2 cannot be excited by equal forces on both bobs?
  • 6 Show that $\sum\omega_k^2=\operatorname{tr}(\mathsf T^{-1}\mathsf V)$ and $\prod\omega_k^2=\det\mathsf V/\det\mathsf T$, and verify both for Examples 3 and 5.
1 If $\mathsf V\mathbf a=\lambda\mathsf T\mathbf a$ with $\mathbf a\ne\mathbf 0$ real (Theorem 8.43), then $\lambda=\dfrac{\mathbf a^T\mathsf V\mathbf a}{\mathbf a^T\mathsf T\mathbf a}\ge0$. The eigenvectors for $\lambda=0$ are exactly the nonzero vectors of $\ker\mathsf V$; since $P$ diagonalises, the number of zero $\lambda_k$ equals the dimension of that eigenspace (eigenspaces of $\mathsf T^{-1}\mathsf V$ have full dimension because it is diagonalisable).
2 $\mathbf a_k^T\mathsf V\mathbf a_j=\lambda_j\mathbf a_k^T\mathsf T\mathbf a_j$ and, transposing $\mathsf V\mathbf a_k=\lambda_k\mathsf T\mathbf a_k$, $\mathbf a_k^T\mathsf V\mathbf a_j=\lambda_k\mathbf a_k^T\mathsf T\mathbf a_j$. Subtracting, $(\lambda_j-\lambda_k)\mathbf a_k^T\mathsf T\mathbf a_j=0$, so $\mathbf a_k^T\mathsf T\mathbf a_j=0$, and then $\mathbf a_k^T\mathsf V\mathbf a_j=0$ too.
3 $$\mathsf T^{-1}\mathsf V=\mathsf T^{-1/2}\big(\mathsf T^{-1/2}\mathsf V\mathsf T^{-1/2}\big)\mathsf T^{1/2},$$ that is, $\mathsf T^{-1/2}S\,\mathsf T^{1/2}$ with $S$ symmetric. Example 6(b) shows it need not be symmetric.
4 $L_2$ in the new coordinates has $\mathsf T'=Q^T\mathsf TQ$, $\mathsf V'=Q^T\mathsf VQ$, so $\det(\mathsf V'-\lambda\mathsf T')=(\det Q)^2\det(\mathsf V-\lambda\mathsf T)$: the same roots. (The mode vectors transform as $\mathbf a'=Q^{-1}\mathbf a$.)
5 With $\eta=P\xi$ and $P^T\mathsf TP=E$: $\ddot\xi_k+\omega_k^2\xi_k=(\mathbf a_k^T\mathbf F)\cos pt$. For $p\ne\omega_k$ the steady solution is $\xi_k=\dfrac{\mathbf a_k^T\mathbf F}{\omega_k^2-p^2}\cos pt$; at $p=\omega_k$ the response grows like $t\sin\omega_kt$ unless $\mathbf a_k^T\mathbf F=0$. In Example 2, $\mathbf F=(f,f)$ gives $\mathbf a_2^T\mathbf F\propto f-f=0$: equal forces never excite the opposite-phase mode.
6 The $\lambda_k$ are the eigenvalues of $\mathsf T^{-1}\mathsf V$ (counted with multiplicity), so their sum is its trace and their product its determinant $\det\mathsf V/\det\mathsf T$. Example 3 ($m=\ell=g=1$): $\mathsf T^{-1}\mathsf V=\left(\begin{smallmatrix}2&-1\\-2&2\end{smallmatrix}\right)$, trace $4$, determinant $2$. Example 5: trace $6k/m$, determinant $4k^3/m^3$.
🎯 Exam corner CSIR Part BCSIR Part CGATE MA

E1. CSIR Part B Two identical simple pendulums of mass $m$ and length $\ell$ have their bobs connected by a light spring of stiffness $k$, unstretched at equilibrium. The ratio of the larger to the smaller normal frequency is (A) $\sqrt{1+\tfrac{2k\ell}{mg}}$; (B) $\sqrt{1+\tfrac{k\ell}{mg}}$; (C) $1+\tfrac{2k\ell}{mg}$; (D) $\sqrt{\tfrac{2k\ell}{mg}}$.

Modes In phase: $\omega_1^2=g/\ell$ (spring idle). Opposite: each bob feels the spring stretched by $2x$, so $\omega_2^2=g/\ell+2k/m$ (Example 2).
Ratio $\dfrac{\omega_2}{\omega_1}=\sqrt{\dfrac{g/\ell+2k/m}{g/\ell}}=\sqrt{1+\dfrac{2k\ell}{mg}}$. (B) forgets the factor $2$; (C) is the ratio of the squares; (D) drops gravity.
Answer (A).

E2. CSIR Part B GATE MA Three particles of masses $m$, $2m$, $m$ lie in that order on a straight line, joined by two identical springs of stiffness $k$. For motion along the line, the largest normal frequency is (A) $\sqrt{2k/m}$; (B) $\sqrt{3k/m}$; (C) $\sqrt{k/(2m)}$; (D) $\sqrt{3k/(2m)}$.

Example 4 The frequencies are $0$, $\sqrt{k/m}$ and $\sqrt{\tfrac km\big(1+\tfrac{2m}M\big)}$ with $M=2m$.
Compute $1+\tfrac{2m}{2m}=2$, so the largest is $\sqrt{2k/m}$ (the antisymmetric stretch $(1,-1,1)$). Check with the trace: $\operatorname{tr}(\mathsf T^{-1}\mathsf V)=\tfrac km+\tfrac{2k}{2m}+\tfrac km=\tfrac{3k}m$, and indeed $0+\tfrac km+\tfrac{2k}m=\tfrac{3k}m$.
Answer (A).

E3. CSIR Part B A system has Lagrangian $L=\tfrac12\dot x^2-\dot x\dot y+\dot y^2-\tfrac32x^2+xy-\tfrac52y^2$. Its normal frequencies are (A) $\sqrt2,\ \sqrt7$; (B) $2,\ 7$; (C) $\sqrt{4-\sqrt2},\ \sqrt{4+\sqrt2}$; (D) $1,\ \sqrt{14}$.

Matrices $T=\tfrac12\dot\eta^T\mathsf T\dot\eta$ and $V=\tfrac12\eta^T\mathsf V\eta$ with $$\mathsf T=\begin{pmatrix}1&-1\\-1&2\end{pmatrix},\quad \mathsf V=\begin{pmatrix}3&-1\\-1&5\end{pmatrix}.$$ $\mathsf T$ is positive definite ($\det=1$).
Secular equation $$(3-\lambda)(5-2\lambda)-(\lambda-1)^2=\lambda^2-9\lambda+14,$$ which factors as $(\lambda-2)(\lambda-7)$. So $\omega^2=2,7$ and $\omega=\sqrt2,\sqrt7$.
Distractors (B) gives $\omega^2$ instead of $\omega$; (C) uses the eigenvalues $4\pm\sqrt2$ of $\mathsf V$ alone (ignoring $\mathsf T$); (D) has the right product $\omega_1^2\omega_2^2=14$ but the wrong sum ($15\ne\operatorname{tr}(\mathsf T^{-1}\mathsf V)=9$).
Answer (A).

E4. CSIR Part C GATE MA For small oscillations $\mathsf T\ddot\eta+\mathsf V\eta=0$ about an equilibrium, with $\mathsf T$ symmetric positive definite and $\mathsf V$ symmetric, which of the following are true? (A) The squared normal frequencies are the eigenvalues of $\mathsf T^{-1}\mathsf V$. (B) Mode vectors belonging to distinct frequencies are orthogonal in the Euclidean sense. (C) If $\mathsf V$ has a negative eigenvalue, the linearised system has an unbounded solution. (D) The normal frequencies do not depend on the choice of generalised coordinates.

(A) True $\det(\mathsf V-\lambda\mathsf T)=\det\mathsf T\cdot\det(\mathsf T^{-1}\mathsf V-\lambda E)$.
(B) False They are $\mathsf T$-orthogonal. Double pendulum: $(1,\sqrt2)\cdot(1,-\sqrt2)=-1$ (Example 3).
(C) True By Theorem 8.43(3) some $\lambda_k\lt0$, and the normal coordinate $\xi_k$ can grow like $e^{\sqrt{-\lambda_k}\,t}$ (Theorem 8.44).
(D) True $\det(Q^T\mathsf VQ-\lambda Q^T\mathsf TQ)$ is $(\det Q)^2\det(\mathsf V-\lambda\mathsf T)$ for a linear change of coordinates; a nonlinear change acts on $(\mathsf T,\mathsf V)$ at the equilibrium through its Jacobian, with the same effect.
Answer (A), (C) and (D).

E5. CSIR Part C Three particles of mass $m$ on a line are joined to each other and to two fixed walls by four identical springs of stiffness $k$ (wall–$m$–$m$–$m$–wall). Which of the following are true? (A) In one normal mode the middle particle stays at rest. (B) The largest normal frequency is $2\sqrt{k/m}$. (C) The sum of the squares of the normal frequencies is $6k/m$. (D) The product of the squares of the normal frequencies is $4k^3/m^3$.

(A) True The mode $(1,0,-1)$ with $\omega^2=2k/m$ (Example 5).
(B) False The largest is $\sqrt{(2+\sqrt2)k/m}\approx1.848\sqrt{k/m}$; $2\sqrt{k/m}$ is only the limit of the chain's frequencies as $N\to\infty$.
(C) True $\operatorname{tr}(\mathsf V)/m=6k/m$.
(D) True $\det\mathsf V/\det\mathsf T=4k^3/m^3$, equal to $(2-\sqrt2)\cdot2\cdot(2+\sqrt2)\,(k/m)^3$.
Answer (A), (C) and (D).

E6. CSIR Part C A bead slides on a smooth vertical hoop of radius $a$ that rotates about its vertical diameter with $\Omega^2=2g/a$; $\theta$ is measured from the lowest point. Which of the following are true? (A) The lowest point $\theta=0$ is an unstable equilibrium. (B) $\theta=\pm\pi/3$ are stable equilibria. (C) Small oscillations about $\theta=\pi/3$ have angular frequency $\sqrt{g/a}$. (D) For sufficiently large $\Omega$ the highest point $\theta=\pi$ becomes stable.

(A) True $\omega^2=g/a-\Omega^2=-g/a\lt0$ (Example 1).
(B) True $\cos\theta^*=g/(a\Omega^2)=\tfrac12$, and $U''(\theta^*)=ma^2\Omega^2\sin^2\theta^*\gt0$.
(C) False $\omega^2=\Omega^2\sin^2\theta^*=\tfrac{2g}a\cdot\tfrac34=\tfrac{3g}{2a}$, so $\omega=\sqrt{3g/(2a)}$.
(D) False At $\theta=\pi$, $U''=-ma(g+a\Omega^2)\lt0$ for every $\Omega$: both gravity and the centrifugal term push the bead away from the top.
Answer (A) and (B).
🧑‍🏫 For the teacher

Suggested plan — 2 lectures × 60 min + 1 problem class.

  1. Stability and the quadratic approximation. Definitions 8.20–8.21, Theorem 8.40, the Lagrange–Dirichlet proof (draw the energy level below the "rim" $m$ of the potential well). Example 1 on the board, with the bifurcation diagram $\theta^*(\Omega)$. Theorem 8.42 and the two practical remarks. Then Theorem 8.43 — present the proof as "change variables so that $\mathsf T$ becomes $E$, then rotate": it is the spectral theorem in disguise.
  2. Normal modes in action. Theorem 8.44 and Example 2; project Interactive 1 (modes first, then "pendulum 1 displaced"; ask the class to predict the transfer time from $\pi/(\omega_2-\omega_1)$ before pressing Play). Example 3 with the orthogonality check; Example 4 and the CO₂ remark; Theorem 8.46 with Interactive 2 (chain, $N=5$: count the sign changes in each mode). Close with Rayleigh's principle and Example 5.
  3. Problem class. Exercises I.6, II.2, II.6, III.3, IV.5 and the Exam corner; insist on the trace and determinant checks.

Discussion questions. (1) Why is the in-phase mode of the coupled pendulums exactly as fast as a single pendulum, whatever the spring? (2) Interactive 2 shows a double pendulum's mixed motion never repeating. Is that the same as chaos? (No: the linearised motion is quasi-periodic; chaos appears only at large amplitude.) (3) Why does a molecule with $N$ atoms moving along a line have $N-1$ vibrational frequencies?

Misconceptions to watch for. Taking eigenvalues of $\mathsf V$ instead of solving $\det(\mathsf V-\lambda\mathsf T)=0$; expecting mode vectors to be Euclidean-orthogonal; linearising $T$ by keeping $\cos(\theta_1-\theta_2)$; testing stability by the signs of the diagonal entries; forgetting the factor $\tfrac12$ when reading $\mathsf V$ off $V$ (in E3, $-\tfrac32x^2$ in $L$ means $\mathsf V_{11}=3$); treating zero modes as instabilities.

Suggested homework. I(1, 2, 5, 8), II(1, 3, 4, 6, 8), III(1, 3, 5, 7), IV(2, 5, 6); Exam corner E3, E4, E6.

🔬 Research–Teaching NexusResearch-tutored · optional, about 60 minutes

Research question. How can a small mass on a spring, hung near the top of a skyscraper, stop the whole building from swaying at its natural frequency?

Why it matters. Tall buildings, footbridges and wind turbines have a lowest mode of vibration that wind or crowds can excite. Many carry a tuned mass damper: Taipei 101 has a steel pendulum of about 660 tonnes near its top, and London's Millennium Bridge was fitted with dampers after it swayed under pedestrians in 2000. The idea is pure small-oscillation theory: add one degree of freedom, tuned to the structure's frequency, and the single dangerous resonance splits into two harmless ones — with a "dead spot" exactly where the old resonance was.

What to do.

  1. Model the building as a mass $M$ on a spring $K$ (frequency $\omega_0=\sqrt{K/M}$) and the damper as a mass $m=\mu M$ on a spring $k$ attached to it, tuned so that $k/m=\omega_0^2$. Write $\mathsf T$ and $\mathsf V$ and show that the normal frequencies satisfy $$\frac{\omega_\pm^2}{\omega_0^2}=1+\frac\mu2\pm\sqrt{\mu+\frac{\mu^2}4}.$$
  2. Evaluate $\omega_\pm/\omega_0$ for $\mu=0.01,\ 0.05,\ 0.1$ (for $\mu=0.05$ the squares come out as exactly $0.8$ and $1.25$).
  3. Force the building with $F\cos pt$ and solve $(\mathsf V-p^2\mathsf T)\mathbf X=(F,0)^T$. Plot the building's amplitude $|X_1|$ against $p/\omega_0$ and show that $X_1=0$ at $p=\omega_0$ (anti-resonance). Use Exercise IV.5 to explain the two new peaks.
  4. Add a dashpot $c$ between the two masses (numerically) and see how the two peaks are flattened; experiment with the tuning ratio and the damping.

Tools: Python with NumPy (or a spreadsheet); Interactive 1 of this section as a mental model of two coupled oscillators.

Hand in: A one-page note with the frequency table, the response curve with and without damping, and a sentence on why the damper must be tuned to the building and not the other way round.

Going further. Read about Den Hartog's classical optimum for a damped absorber (tuning ratio $1/(1+\mu)$ and damping ratio $\sqrt{3\mu/(8(1+\mu)^3)}$), and about why footbridge dampers are tuned to the sideways mode excited by walking.

Instructor note: in units of $\omega_0$, $\mathsf T^{-1}\mathsf V=\left(\begin{smallmatrix}1+\mu&-\mu\\-1&1\end{smallmatrix}\right)$, characteristic polynomial $\lambda^2-(2+\mu)\lambda+1$. The ratios $\omega_\pm/\omega_0$ are $0.9512,\ 1.0512$ ($\mu=0.01$), $0.8944,\ 1.1180$ ($\mu=0.05$) and $0.8543,\ 1.1705$ ($\mu=0.1$). The anti-resonance follows from the second row of the forced system: at $p^2=k/m$ it reads $-kX_1=0$. The product $\omega_+\omega_-=\omega_0^2$ (the constant term $1$) is a nice check students can find themselves.

🧪 Self-check quiz

0/5 correct

Q1. With mass matrix $\mathsf T$ and stiffness matrix $\mathsf V$, the squared normal frequencies are the roots of

Q2. Two mode vectors $\mathbf a_1,\mathbf a_2$ with different frequencies always satisfy

Q3. The Lagrange–Dirichlet theorem says that an equilibrium of a natural system is stable if

Q4. A free linear triatomic molecule vibrating along its axis has how many zero-frequency modes?

Q5. Two identical weakly coupled pendulums have normal frequencies $\omega_1\lt\omega_2$. If one is released with the other at rest, the time for the energy to pass completely to the other is