8.1 Generalised Coordinates and Lagrange’s Equations
Describe a constrained system by as many numbers as it has freedoms, let D'Alembert's principle discard the forces of constraint, and obtain the equations of motion from one scalar function $L=T-V$: $\dfrac{d}{dt}\dfrac{\partial L}{\partial\dot q_j}-\dfrac{\partial L}{\partial q_j}=0$.
Newton's second law $m\ddot{\mathbf r}=\mathbf F$ is complete in principle and awkward in practice. A bead sliding on a spinning hoop, a pendulum hanging from a pendulum, a mass on a string over a pulley: in each case the forces that keep the system on its track — the normal reaction of the wire, the tension of the rod, the pull of the string — are unknown until the motion is known, and they enter Newton's equations as extra unknowns that must be eliminated by hand. Lagrange's idea (in his Mécanique analytique of 1788, a book famous for containing not a single figure) was to choose coordinates in which the constraints are satisfied automatically, and to write the dynamics so that the workless constraint forces never appear. The result is a recipe of startling economy: write the kinetic energy $T$ and the potential energy $V$ in any convenient coordinates $q_1,\dots,q_n$, form $L=T-V$, and the equations of motion are $\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}=\frac{\partial L}{\partial q_j}$ — the same form in Cartesian, polar, angular or any other coordinates.
Lagrange's equations are also the Euler–Lagrange equations of the action functional $\int L\,dt$ (§6.1); we return to that variational point of view in §8.2. In this section we derive them from Newton's laws through virtual work and D'Alembert's principle, so that every hypothesis is visible: holonomic constraints, ideal (workless) constraints, and forces derived from a potential. We then exploit the structure of $L$: a coordinate that does not appear in $L$ hands us a conserved momentum, and a Lagrangian without explicit time dependence hands us the energy integral. CSIR NET asks for exactly these things — the number of degrees of freedom, the type of a constraint, the Lagrangian or equation of motion of a standard system, a conserved quantity, an equilibrium and its stability — in Part B, and asks which statements about constraints, conservation laws and equivalent Lagrangians are true in Part C.
Constraints, Degrees of Freedom and Generalised Coordinates
Consider $N$ particles with masses $m_i$ and position vectors $\mathbf r_i\in\mathbb R^3$, $i=1,\dots,N$. Unrestricted, the system needs $3N$ numbers to fix its configuration. Constraints are restrictions on the possible positions or velocities, imposed by rods, strings, surfaces, wires, hinges and rolling contacts.
- A constraint is holonomic if it can be written as an equation between the coordinates and possibly time, $$f(\mathbf r_1,\dots,\mathbf r_N,t)=0 .$$ Every other constraint is non-holonomic: an inequality ($|\mathbf r|\ge a$, a gas molecule in a box) or a relation between velocities that cannot be integrated to a relation between coordinates (a coin rolling on a table).
- A constraint is scleronomic (rigid) if it does not involve time explicitly, and rheonomic (flowing) if it does: a bead on a wire rotating at a prescribed rate, or a pendulum whose support is driven, obeys a rheonomic constraint.
- A constraint given as a linear relation between differentials, $\sum_k a_k(\mathbf x,t)\,dx_k+a_t(\mathbf x,t)\,dt=0$, is called a Pfaffian constraint. It is holonomic exactly when it is integrable: some factor $\mu\ne0$ makes the left side the differential $dF$ of a function, so that the constraint is $F(\mathbf x,t)=\text{const}$.
The two classifications are independent: a particle on a fixed sphere $|\mathbf r|=a$ is holonomic and scleronomic; on a sphere whose radius grows, $|\mathbf r|=a+ct$, it is holonomic and rheonomic; a disc rolling on a table is non-holonomic and scleronomic.
Suppose the $N$ particles are subject to $k$ independent holonomic constraints $f_1=\dots=f_k=0$ (independent means that the gradients of $f_1,\dots,f_k$ with respect to the $3N$ coordinates are linearly independent). The number $n=3N-k$ is the number of degrees of freedom. Generalised coordinates are $n$ real parameters $q_1,\dots,q_n$ such that
$$\mathbf r_i=\mathbf r_i(q_1,\dots,q_n,t),\qquad i=1,\dots,N,$$with $\mathbf r_i$ of class $C^2$, every choice of $q$ (in an open set) gives a configuration satisfying all the constraints, every nearby allowed configuration arises from exactly one $q$, and the $3N\times n$ Jacobian matrix $\big(\partial\mathbf r_i/\partial q_j\big)$ has rank $n$. The derivatives $\dot q_j$ are the generalised velocities, and the $n$-dimensional space of the $q$'s is the configuration space.
For a non-holonomic system with Pfaffian constraints, the number of generalised coordinates (needed to fix the configuration) exceeds the number of independent velocities (the degrees of freedom of the motion) by the number of non-integrable constraints.
Generalised coordinates need not be lengths, need not have the same units, and need not be Cartesian: angles, arc lengths, the distance of a bead along a wire, the amount of string unwound from a pulley are all typical. By the implicit function theorem (Unit 1, multivariable calculus), $k$ independent holonomic constraints can always be solved locally for $k$ of the $3N$ coordinates in terms of the other $3N-k$, so generalised coordinates always exist locally; finding convenient ones is the art.
Proof. Suppose $\mu\mathbf A=\nabla F$. Then $\mathbf A=\mu^{-1}\nabla F$ and, by the product rule for the curl and $\nabla\times\nabla F=0$, $$\nabla\times\mathbf A=\nabla(\mu^{-1})\times\nabla F .$$ Therefore $\mathbf A\cdot(\nabla\times\mathbf A)=\mu^{-1}\,\nabla F\cdot\big(\nabla(\mu^{-1})\times\nabla F\big)=0$, because a scalar triple product with a repeated vector vanishes. The converse is the three-dimensional case of Frobenius' integrability theorem, which we quote. In two variables the constraint says $dy/dx=-A_1/A_2$ (or $dx/dy=-A_2/A_1$), a first-order ODE; by the existence theorem of §1.1 its solution curves fill a neighbourhood, they are the level curves of some $C^1$ function $F$ with $\nabla F\ne0$, and then $\nabla F$ is parallel to $\mathbf A$, i.e. $\nabla F=\mu\mathbf A$ — this is the integrating factor of §1.1. ∎
When the constraint involves time, $\sum a_k\,dq_k+a_t\,dt=0$, treat $t$ as one more variable; the same criterion applies in the enlarged space. The rolling coin is the standard non-integrable example: a disc of radius $a$ rolling upright without slipping on a horizontal plane has configuration $(x,y,\theta,\varphi)$ — the contact point, the heading angle and the angle of rotation — and the rolling conditions $$\dot x=a\dot\varphi\cos\theta,\qquad \dot y=a\dot\varphi\sin\theta$$ relate velocities only. They cannot be integrated: by rolling along a suitable closed loop the coin can return to the same contact point and heading with any rotation angle $\varphi$ we like, so no relation $F(x,y,\theta,\varphi)=0$ holds. The coin needs four coordinates but has only two independent velocities ($\dot\theta$ and $\dot\varphi$). By contrast, a cylinder rolling along a straight line obeys $\dot x=a\dot\varphi$, which integrates to $x=a\varphi+c$: a holonomic constraint.
Proof. By the chain rule $\dot{\mathbf r}_i=\sum_j\frac{\partial\mathbf r_i}{\partial q_j}\dot q_j+\frac{\partial\mathbf r_i}{\partial t}$. Squaring and summing with weights $\tfrac12m_i$ gives the three groups of terms: products of two velocity terms ($T_2$), cross terms between a velocity term and $\partial\mathbf r_i/\partial t$ (twice $\tfrac12$, giving $T_1$), and the square of $\partial\mathbf r_i/\partial t$ ($T_0$). Symmetry of $a_{jk}$ is clear. For $\xi\in\mathbb R^n$, $$\sum_{j,k}a_{jk}\xi_j\xi_k=\sum_i m_i\Big|\sum_j\xi_j\frac{\partial\mathbf r_i}{\partial q_j}\Big|^2\ge0,$$ with equality only if $\sum_j\xi_j\,\partial\mathbf r_i/\partial q_j=0$ for every $i$, i.e. only if $\xi$ is in the kernel of the Jacobian $\big(\partial\mathbf r_i/\partial q_j\big)$. That Jacobian has rank $n$, so its kernel is $\{0\}$: the form is positive definite. ∎
Two familiar cases: a particle in plane polar coordinates has $T=\tfrac12m(\dot r^2+r^2\dot\theta^2)$, so $(a_{jk})=\mathrm{diag}(m,mr^2)$, positive definite for $r\gt0$ (at $r=0$ the polar coordinates fail to be generalised coordinates — the rank drops). A bead at distance $r$ along a straight wire turning in a horizontal plane at constant angular speed $\omega$ about one of its points has $\mathbf r=r(\cos\omega t,\sin\omega t,0)$ and $$T=\tfrac12m\dot r^2+\tfrac12m\omega^2r^2,\qquad T_2=\tfrac12m\dot r^2,\quad T_1=0,\quad T_0=\tfrac12m\omega^2r^2 .$$
Virtual Work and D'Alembert's Principle
The constraint forces are the obstacle in Newton's equations. The key observation is that, for most constraints met in practice, they do no work in any displacement that is compatible with the constraints as they stand at a frozen instant.
A virtual displacement of a holonomic system at time $t$ is an infinitesimal change of configuration consistent with the constraints at that fixed time:
$$\delta\mathbf r_i=\sum_{j=1}^n\frac{\partial\mathbf r_i}{\partial q_j}\,\delta q_j\qquad(\delta t=0),$$with $\delta q_1,\dots,\delta q_n$ arbitrary. (An actual displacement in time $dt$ is $d\mathbf r_i=\sum_j\frac{\partial\mathbf r_i}{\partial q_j}dq_j+\frac{\partial\mathbf r_i}{\partial t}dt$; the two agree only for scleronomic constraints.) Write the total force on particle $i$ as $\mathbf F_i+\mathbf R_i$, the applied force plus the constraint force. The constraints are ideal (workless) if
$$\sum_{i=1}^N\mathbf R_i\cdot\delta\mathbf r_i=0\quad\text{for every virtual displacement.}$$Smooth surfaces and wires (the reaction is normal to every virtual displacement, even when the surface moves), light inextensible strings and rigid rods (internal forces are equal, opposite and along the line joining the particles, and the virtual displacements keep that distance fixed), and rolling without slipping (the contact point has zero virtual velocity) are ideal. Sliding friction is not.
Proof. If the system stays at rest, every $\ddot{\mathbf r}_i=0$, so $\mathbf F_i+\mathbf R_i=0$ and $\sum\mathbf F_i\cdot\delta\mathbf r_i=-\sum\mathbf R_i\cdot\delta\mathbf r_i=0$. The equivalence with $Q_j=0$ is Definition 8.4 with the $\delta q_j$ independent. Conversely, if all $Q_j(q^0)=0$, then the constant path $q(t)\equiv q^0$ satisfies Lagrange's equations (Theorem 8.6 below; with scleronomic constraints and $\dot q=0$ every term involving $T$ vanishes), and by the uniqueness theorem for the initial value problem (§1.1, applied to Lagrange's equations in normal form — possible because $(a_{jk})$ is invertible, Theorem 8.2) it is the only motion starting at rest at $q^0$. ∎
A quick illustration. A uniform ladder $AB$ of length $2\ell$ and weight $W$ stands with $A$ on a smooth floor and $B$ against a smooth wall, making an angle $\alpha$ with the floor; a horizontal string pulls $A$ towards the wall with tension $S$. With the single coordinate $\alpha$, the foot is at distance $x=2\ell\cos\alpha$ from the wall and the centre at height $\ell\sin\alpha$. In a virtual change $\delta\alpha$ the string does work $-S\,\delta x=2\ell S\sin\alpha\,\delta\alpha$ and gravity does $-W\,\delta(\ell\sin\alpha)=-W\ell\cos\alpha\,\delta\alpha$; the smooth wall and floor do none. Setting the sum to zero, $S=\tfrac12W\cot\alpha$ — obtained without ever computing the two reactions.
Proof. Newton's law for particle $i$ is $m_i\ddot{\mathbf r}_i=\mathbf F_i+\mathbf R_i$, so $\sum(\mathbf F_i-m_i\ddot{\mathbf r}_i)\cdot\delta\mathbf r_i=-\sum\mathbf R_i\cdot\delta\mathbf r_i$, which vanishes by the definition of ideal constraints. ∎
D'Alembert's principle is "virtual work for dynamics": the reversed effective forces $-m_i\ddot{\mathbf r}_i$ are added to the applied forces and the system is treated as if in equilibrium. Its content lies in the hypothesis — the physical statement that the constraints are workless — and its power is that the unknown $\mathbf R_i$ have disappeared.
Lagrange's Equations
D'Alembert's principle still uses Cartesian vectors. To express it in generalised coordinates we need two identities about the functions $\mathbf r_i(q,t)$ and their time derivatives $\dot{\mathbf r}_i(q,\dot q,t)=\sum_j\frac{\partial\mathbf r_i}{\partial q_j}\dot q_j+\frac{\partial\mathbf r_i}{\partial t}$, in which $q$ and $\dot q$ are treated as independent variables.
Proof. (a) $\dot{\mathbf r}_i$ is affine in $\dot q$ with coefficient $\partial\mathbf r_i/\partial q_j$ in front of $\dot q_j$. (b) Along a motion, the chain rule gives $$\frac{d}{dt}\frac{\partial\mathbf r_i}{\partial q_j}=\sum_k\frac{\partial^2\mathbf r_i}{\partial q_k\,\partial q_j}\dot q_k+\frac{\partial^2\mathbf r_i}{\partial t\,\partial q_j},$$ while differentiating the formula for $\dot{\mathbf r}_i$ with respect to $q_j$ (holding $\dot q$ fixed) gives $$\frac{\partial\dot{\mathbf r}_i}{\partial q_j}=\sum_k\frac{\partial^2\mathbf r_i}{\partial q_j\,\partial q_k}\dot q_k+\frac{\partial^2\mathbf r_i}{\partial q_j\,\partial t}.$$ The mixed partial derivatives are equal because $\mathbf r_i$ is $C^2$ (Schwarz's theorem). ∎
Let a system of $N$ particles have $n$ degrees of freedom, holonomic ideal constraints and $C^2$ generalised coordinates $q_1,\dots,q_n$. Then along every motion
$$\frac{d}{dt}\frac{\partial T}{\partial\dot q_j}-\frac{\partial T}{\partial q_j}=Q_j,\qquad j=1,\dots,n .$$If moreover the applied forces derive from a potential $V(q,t)$, then with the Lagrangian $L(q,\dot q,t)=T-V$,
$$\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}-\frac{\partial L}{\partial q_j}=0,\qquad j=1,\dots,n .$$Proof. Insert $\delta\mathbf r_i=\sum_j\frac{\partial\mathbf r_i}{\partial q_j}\delta q_j$ into D'Alembert's principle. The applied forces contribute $\sum_jQ_j\,\delta q_j$ (Definition 8.4). For the inertial term, using the product rule and then both parts of Theorem 8.5,
$$\sum_i m_i\ddot{\mathbf r}_i\cdot\frac{\partial\mathbf r_i}{\partial q_j}=\frac{d}{dt}\Big(\sum_i m_i\dot{\mathbf r}_i\cdot\frac{\partial\mathbf r_i}{\partial q_j}\Big)-\sum_i m_i\dot{\mathbf r}_i\cdot\frac{d}{dt}\frac{\partial\mathbf r_i}{\partial q_j}$$ $$=\frac{d}{dt}\Big(\sum_i m_i\dot{\mathbf r}_i\cdot\frac{\partial\dot{\mathbf r}_i}{\partial\dot q_j}\Big)-\sum_i m_i\dot{\mathbf r}_i\cdot\frac{\partial\dot{\mathbf r}_i}{\partial q_j}=\frac{d}{dt}\frac{\partial T}{\partial\dot q_j}-\frac{\partial T}{\partial q_j},$$because $\partial T/\partial\dot q_j=\sum_im_i\dot{\mathbf r}_i\cdot\partial\dot{\mathbf r}_i/\partial\dot q_j$ and similarly for $\partial T/\partial q_j$. D'Alembert's principle therefore reads
$$\sum_{j=1}^n\Big(Q_j-\frac{d}{dt}\frac{\partial T}{\partial\dot q_j}+\frac{\partial T}{\partial q_j}\Big)\delta q_j=0 .$$Since the constraints are holonomic, the $\delta q_j$ are independent and arbitrary; choosing $\delta q$ to be each coordinate vector in turn shows that every bracket vanishes. In the conservative case $Q_j=-\partial V/\partial q_j$, and since $V$ does not depend on $\dot q$, $\partial T/\partial\dot q_j=\partial L/\partial\dot q_j$; moving $-\partial V/\partial q_j$ to the left gives the second form. ∎
Two remarks complete the picture. If some forces are conservative and others are not, put the first into $L$ and keep the rest on the right: $\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}-\frac{\partial L}{\partial q_j}=Q_j'$. And the second form still holds for a velocity-dependent potential $U(q,\dot q,t)$ whenever $Q_j=\frac{d}{dt}\frac{\partial U}{\partial\dot q_j}-\frac{\partial U}{\partial q_j}$; the important case is a charge $e$ in electromagnetic potentials $\phi,\mathbf A$, for which $L=\tfrac12m|\dot{\mathbf r}|^2-e\phi+e\,\dot{\mathbf r}\cdot\mathbf A$ reproduces the Lorentz force $e(\mathbf E+\dot{\mathbf r}\times\mathbf B)$ (Exercise II.9).
Proof. (a) Linearity of the Lagrange expression $\mathcal E_j(L)=\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}-\frac{\partial L}{\partial q_j}$ reduces the claim to $\mathcal E_j(dF/dt)=0$. Now $\frac{\partial}{\partial\dot q_j}\frac{dF}{dt}=\frac{\partial F}{\partial q_j}$, so $\frac{d}{dt}\frac{\partial}{\partial\dot q_j}\frac{dF}{dt}=\frac{d}{dt}\frac{\partial F}{\partial q_j}$; and $\frac{\partial}{\partial q_j}\frac{dF}{dt}=\frac{d}{dt}\frac{\partial F}{\partial q_j}$ by the same interchange of mixed partials as in Theorem 8.5(b). The difference is zero. (b) This is Theorem 8.5 again, now applied to the map $Q\mapsto q$: exactly the computation in the proof of Theorem 8.6 shows $\mathcal E_k^{Q}(\tilde L)=\sum_j\frac{\partial q_j}{\partial Q_k}\,\mathcal E_j^{q}(L)$, and the matrix $\big(\partial q_j/\partial Q_k\big)$ is invertible, so one set of equations holds iff the other does. (Alternatively, both sets are the Euler–Lagrange equations of the same action integral — §8.2.) ∎
Part (a) is a favourite of Part C. The converse is false in general: two Lagrangians can produce equivalent equations without being related as in (a) — for one free particle, $L=\dot q^2$ and $L=\dot q^4$ both force $\ddot q=0$ (the second gives $12\dot q^2\ddot q=0$), and $L=e^{\gamma t}\big(\tfrac12\dot q^2-\tfrac12\omega^2q^2\big)$ gives the damped equation $\ddot q+\gamma\dot q+\omega^2q=0$ (Exercise III.7).
Proof (sketch). Now the virtual displacements are restricted to the subspace $K=\{\delta q:\sum_ja_{lj}\delta q_j=0,\ l=1,\dots,m\}$. The computation in the proof of Theorem 8.6 shows that the vector $\mathcal E=(\mathcal E_1(L),\dots,\mathcal E_n(L))$ satisfies $\mathcal E\cdot\delta q=0$ for all $\delta q\in K$. By linear algebra, a linear functional vanishing on the common kernel of the functionals $a_l=(a_{l1},\dots,a_{ln})$ is a linear combination of them: $\mathcal E=\sum_l\lambda_la_l$. ∎
Multipliers are also the way to compute a constraint force for a holonomic constraint: keep a redundant coordinate and impose the constraint with a multiplier. For a pendulum written in polar coordinates $(r,\theta)$ with the constraint $\dot r=0$, the $r$-equation $m\ddot r-mr\dot\theta^2-mg\cos\theta=\lambda$ with $r=\ell$ gives $\lambda=-(m\ell\dot\theta^2+mg\cos\theta)$: minus the string tension.
- Count and choose. Count the degrees of freedom $n=3N-k$ (a rigid body counts 6, a rigid body with a fixed point 3, a rigid body in plane motion 3). Pick $n$ coordinates that make the constraints automatic — angles for rods and hoops, distance along a wire, length of string paid out.
- Write positions, then velocities. Express the Cartesian position of every mass (and the angular velocity of every rigid body) in terms of $q$ and $t$; differentiate. Mistakes are almost always made here, not in the Lagrange equations — moving supports and rotating wires contribute $\partial\mathbf r/\partial t$ terms.
- Form $T$ and $V$. $T=\tfrac12\sum m_i|\dot{\mathbf r}_i|^2$ (plus $\tfrac12I\omega^2$ for rotating bodies); $V$ from gravity ($mgz$), springs ($\tfrac12k(\text{extension})^2$), central potentials. Then $L=T-V$.
- Differentiate mechanically. For each $j$ compute $\partial L/\partial\dot q_j$, its total time derivative, and $\partial L/\partial q_j$. Before solving, harvest the free first integrals: a coordinate absent from $L$ (cyclic) gives a conserved momentum; no explicit $t$ gives the energy integral (Theorems 8.9 and 8.11).
- Check. Units; the limit of a known case (pulley mass $\to0$, rotation $\to0$); equilibrium positions from $\partial V_{\text{eff}}/\partial q=0$.
Cyclic Coordinates, Conservation Laws and the Energy Integral
Lagrange's equations are second-order ODEs, and the best thing one can do with them before solving is to find first integrals: functions of $(q,\dot q,t)$ that stay constant along every motion. Each first integral lowers the order of the problem by one. Two of them come almost free from the shape of $L$.
For a particle in Cartesian coordinates $p_x=m\dot x$ is the linear momentum; for an angle $\theta$ the conjugate momentum is an angular momentum ($p_\theta=mr^2\dot\theta$ in polar coordinates); in a magnetic field $p_x=m\dot x+eA_x$ is not the mechanical momentum at all.
Proof. Lagrange's equation for $q_j$ is $\dfrac{dp_j}{dt}=\dfrac{\partial L}{\partial q_j}=0$. ∎
Whether a coordinate is cyclic depends on the coordinates chosen: for a central potential $V(r)$, the polar angle $\theta$ is cyclic in $L=\tfrac12m(\dot r^2+r^2\dot\theta^2)-V(r)$, while neither $x$ nor $y$ is cyclic in Cartesian coordinates — yet the conserved quantity, angular momentum, is the same. The next theorem explains conservation laws without reference to a particular coordinate.
Proof. Along a motion, using $\dot p_j=\partial L/\partial q_j$ (Lagrange's equations), $$\frac{dI}{dt}=\sum_j\dot p_jK_j+\sum_jp_j\sum_k\frac{\partial K_j}{\partial q_k}\dot q_k=\sum_j\frac{\partial L}{\partial q_j}K_j+\sum_j\frac{\partial L}{\partial\dot q_j}\sum_k\frac{\partial K_j}{\partial q_k}\dot q_k=0 .$$ For a translation of all particles by $\varepsilon\mathbf n$ in Cartesian coordinates, $K$ assigns $\mathbf n$ to every particle and $I=\sum_i m_i\dot{\mathbf r}_i\cdot\mathbf n$. For an infinitesimal rotation $\mathbf r_i\mapsto\mathbf r_i+\varepsilon\,\mathbf n\times\mathbf r_i$, $I=\sum_i m_i\dot{\mathbf r}_i\cdot(\mathbf n\times\mathbf r_i)=\mathbf n\cdot\sum_i\mathbf r_i\times m_i\dot{\mathbf r}_i$. ∎
If $q_j$ is cyclic, the translation $q_j\mapsto q_j+\varepsilon$ (that is, $K=e_j$) is a symmetry, and Theorem 8.10 returns Theorem 8.9. Conversely, near any point where $K\ne0$ one can choose coordinates in which the flow of $K$ is a translation of a single coordinate (straightening the flow, as in the theory of first integrals of §3.1), and in those coordinates that coordinate is cyclic — which is why one chooses polar coordinates for central forces.
(a) Along every solution of Lagrange's equations $\dfrac{dh}{dt}=-\dfrac{\partial L}{\partial t}$. Hence if $L$ does not contain $t$ explicitly, $h$ is a constant of the motion.
(b) If $L=T-V$ with $T=T_2+T_1+T_0$ as in Theorem 8.2 and $V=V(q,t)$, then $h=T_2-T_0+V$.
(c) In particular, if the constraints are scleronomic and $V=V(q)$, then $h=T+V=E$ and the total mechanical energy is conserved.
Proof. (a) By the chain rule along a motion, $$\frac{dL}{dt}=\sum_j\frac{\partial L}{\partial q_j}\dot q_j+\sum_j\frac{\partial L}{\partial\dot q_j}\ddot q_j+\frac{\partial L}{\partial t},$$ so $$\frac{dh}{dt}=\sum_j\ddot q_j\frac{\partial L}{\partial\dot q_j}+\sum_j\dot q_j\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}-\frac{dL}{dt}=\sum_j\dot q_j\Big(\frac{d}{dt}\frac{\partial L}{\partial\dot q_j}-\frac{\partial L}{\partial q_j}\Big)-\frac{\partial L}{\partial t}=-\frac{\partial L}{\partial t}.$$ (b) $T_k$ is homogeneous of degree $k$ in $\dot q$, so by Euler's theorem on homogeneous functions $\sum_j\dot q_j\,\partial T_k/\partial\dot q_j=kT_k$. Since $V$ does not involve $\dot q$, $\sum_j\dot q_j\,\partial L/\partial\dot q_j=2T_2+T_1$, and $h=2T_2+T_1-(T_2+T_1+T_0-V)=T_2-T_0+V$. (c) Scleronomic coordinates give $T_1=T_0=0$ (Theorem 8.2), so $h=T+V$, and $L$ has no explicit $t$. ∎
The two conditions in (c) — "no explicit time in $L$" (which makes $h$ conserved) and "scleronomic, velocity-independent potential" (which makes $h$ equal to the energy) — are logically independent, and CSIR Part C exploits exactly this. For the rotating hoop of Example 4, $L$ has no explicit $t$, so $$h=\tfrac12ma^2\dot\theta^2-\tfrac12ma^2\omega^2\sin^2\theta-mga\cos\theta$$ is conserved; but the energy $$E=T+V=\tfrac12ma^2\dot\theta^2+\tfrac12ma^2\omega^2\sin^2\theta-mga\cos\theta$$ equals $h+ma^2\omega^2\sin^2\theta$, which changes whenever $\theta$ does. The difference is the work done by the motor that keeps $\omega$ constant (the hoop pushes the bead sideways when it moves up or down). With additional non-potential generalised forces $Q_j'$, the computation in (a) gives $\dfrac{dh}{dt}=\sum_jQ_j'\dot q_j-\dfrac{\partial L}{\partial t}$; for linear friction $Q_j'=-c_j\dot q_j$ this is the rate of dissipation $-\sum c_j\dot q_j^2\le0$.
- Rheonomic is not non-holonomic. A bead on a rotating wire has a time-dependent but holonomic constraint ($n=1$). Non-holonomic means an inequality or a non-integrable velocity relation.
- A velocity constraint may still be holonomic. $\dot x=a\dot\varphi$ (rolling on a line) and $y\,dx-x\,dy=0$ integrate to coordinate relations; the rolling coin and $dz-y\,dx=0$ do not (Theorem 8.1).
- Exactness is not the test. $y\,dx-x\,dy$ is not exact, yet the constraint is holonomic (integrating factor $1/y^2$). Use $\mathbf A\cdot(\nabla\times\mathbf A)=0$.
- Virtual work of a moving constraint is zero; actual work need not be. The wire of a rotating bead does no work in a virtual displacement (along the wire, time frozen) but does positive work in the actual motion: the kinetic energy of the bead grows.
- Sliding friction is not ideal. D'Alembert's principle, and hence Lagrange's equations in the form $\mathcal E_j(L)=0$, fail for it; friction must be added as a non-potential force $Q_j'$.
- "No explicit $t$ in $L$" gives $h$, not $T+V$. For the rotating hoop $h=T_2-T_0+V$ is conserved and $T+V$ is not. Conversely, a spring with time-dependent stiffness $k(t)$ in fixed coordinates has $h=T+V$ but $h$ is not conserved.
- The effective potential is not the potential energy. Centrifugal terms $-\tfrac12ma^2\omega^2\sin^2\theta$ (rotating hoop) or $+\ell^2/2mr^2$ (central force) come from the kinetic energy.
- Equivalent Lagrangians. $cL+dF(q,t)/dt$ gives the same equations, but $L+tq\dot q$ does not ($tq\dot q$ is a total derivative plus $-\tfrac12q^2$); and different Lagrangians can give the same equations without being related this way ($\dot q^2$ and $\dot q^4$).
- Generalised momentum is not always $m\times$velocity. In a magnetic field $p=m\dot{\mathbf r}+e\mathbf A$; for an angle it is an angular momentum; for the hoop $p_\theta=ma^2\dot\theta$ is only the meridian part of the angular momentum.
- Circular orbits need not be stable. In $V=-k/r^n$ circular orbits exist at every radius but are stable only for $n\lt2$ (Exercise IV.6); at $n=2$ the effective potential has no minimum at all.
- Kepler's second law holds for every central force. Constancy of $r^2\dot\theta$ is conservation of $p_\theta$; only the first and third laws are special to the inverse square.
Exercises
- 1 Find the number of degrees of freedom of: (a) two particles joined by a light rigid rod and constrained to move in a fixed plane; (b) a rigid body free to turn about a fixed point; (c) three particles with all three mutual distances fixed, one of them constrained to a fixed plane; (d) a rigid body moving parallel to a fixed plane (plane motion).
- 2 Classify (holonomic or not; scleronomic or rheonomic): (a) a particle on the surface of a sphere of radius $a+ct$ centred at the origin; (b) a particle constrained to stay outside the sphere $|\mathbf r|=a$; (c) a pendulum whose length is shortened at a constant rate $\ell=\ell_0-vt$; (d) a sphere rolling without slipping on a fixed rough plane.
- 3 Decide which constraints are holonomic and integrate those that are: (a) $z\,dx+x\,dy=0$; (b) $2xz\,dx+2yz\,dy-(x^2+y^2)\,dz=0$.
- 4 Write $T$ for a particle in spherical polar coordinates $(r,\theta,\varphi)$, identify the matrix $(a_{jk})$, and find where it fails to be positive definite.
- 5 A particle moves in a plane; describe it by coordinates $(X,Y)$ in a frame rotating with constant angular velocity $\omega$ about the origin ($x=X\cos\omega t-Y\sin\omega t$, $y=X\sin\omega t+Y\cos\omega t$). Find $T_2$, $T_1$ and $T_0$.
- 6 Use virtual work to find the tension $S$ in the horizontal string that holds the ladder of the text (length $2\ell$, weight $W$, smooth wall and floor) at angle $\alpha=60^\circ$ to the floor, for $W=300\ \text{N}$.
- 7 A rhombus $ABCD$ of four equal light rods, smoothly jointed, hangs from $A$; a weight $W$ hangs from $C$, and a light rod $BD$ keeps the angle $BAD$ equal to $2\alpha$. Use virtual work to show that the rod $BD$ is in compression with thrust $W\tan\alpha$.
- 8 Explain why the reaction of a moving smooth surface still satisfies $\sum\mathbf R\cdot\delta\mathbf r=0$, and give an example in which the same reaction does non-zero actual work.
- 1 A simple pendulum (length $\ell$, mass $m$) hangs from a support that moves horizontally as $x_s=A\cos\Omega t$. Show that $\ell\ddot\theta+g\sin\theta=A\Omega^2\cos\Omega t\cos\theta$.
- 2 The same pendulum hangs from a support moving vertically as $y_s=A\cos\Omega t$. Show that $\ell\ddot\theta+\big(g-A\Omega^2\cos\Omega t\big)\sin\theta=0$, so that the support acts like a periodically modulated gravity.
- 3 A mass $m$ hangs from a spring (stiffness $k$, natural length $\ell_0$) whose other end is fixed, and swings in a vertical plane (spring pendulum). Write $L$ in polar coordinates $(r,\theta)$ and derive both equations of motion.
- 4 A particle slides on the smooth inside of a right circular cone with vertical axis and vertex downward, semi-vertical angle $\alpha$. Using the distance $\rho$ from the axis and the azimuth $\varphi$, find $L$, the conserved momentum, and the angular velocity of a horizontal circular motion at distance $\rho_0$ from the axis.
- 5 A double Atwood machine: over a fixed pulley passes a string carrying $m_1=4m$ on one side and a light pulley on the other; over the light pulley passes a string carrying $m_2=m$ and $m_3=3m$. All pulleys are light and smooth. Find the accelerations of the three masses.
- 6 A bead is released from rest (relative to the wire) at distance $r_0$ from $O$ on a smooth straight horizontal wire that rotates about a vertical axis through $O$ with constant angular velocity $\omega$. Find $r(t)$ and the rate at which the wire does work on the bead.
- 7 A smooth straight wire through $O$ makes a fixed angle $\alpha$ with the upward vertical and rotates about the vertical through $O$ at constant $\omega$. Find the equation of motion of a bead on it, the position of relative equilibrium, and decide its stability.
- 8 Use Binet's equation to find the central force under which a particle can describe (a) the cardioid $r=a(1+\cos\theta)$; (b) the equiangular spiral $r=ae^{c\theta}$.
- 9 Show that $L=\tfrac12m(\dot x^2+\dot y^2+\dot z^2)+e\,\dot{\mathbf r}\cdot\mathbf A$ with $\mathbf A=\tfrac12B(-y,x,0)$ gives the motion of a charge in the uniform field $\mathbf B=B\mathbf k$, and find the angular frequency of the circular motion.
- 1 For a particle in a central potential $V(r)$ in spherical coordinates, which coordinates are cyclic and what are the conserved momenta? Find a third conserved quantity.
- 2 $L=\tfrac12m(\dot x_1^2+\dot x_2^2)-V(x_1-x_2)$. Find a conserved momentum by changing to the coordinates $X=\tfrac12(x_1+x_2)$, $\xi=x_1-x_2$, and confirm it with Theorem 8.10.
- 3 Show by Theorem 8.10 that for $L=\tfrac12m(\dot x^2+\dot y^2)-V(x^2+y^2)$ the quantity $m(x\dot y-y\dot x)$ is conserved.
- 4 For the bead on the rotating horizontal wire (II.6) compute $h$ and verify that it is conserved while $T$ is not.
- 5 For the rotating hoop, compute $\dfrac{dE}{dt}$ where $E=T+V$, and verify that it equals $ma^2\omega^2\frac{d}{dt}\sin^2\theta$.
- 6 $L=\tfrac12m\dot x^2-\tfrac12k(t)x^2$. Is $h$ conserved? Is $h=T+V$?
- 7 Show that $L=e^{\gamma t}\big(\tfrac12m\dot q^2-\tfrac12kq^2\big)$ gives the damped oscillator $m\ddot q+m\gamma\dot q+kq=0$. Is the corresponding $h$ conserved?
- 8 In the field $V=-k/r$ show that circular orbits are stable, and find the period of small radial oscillations about a circular orbit of radius $r_0$; compare with the orbital period.
- 1 Prove that if $T=\tfrac12\sum a_{jk}(q)\dot q_j\dot q_k$ with $(a_{jk})$ positive definite and $V=V(q)$, Lagrange's equations can be written in the normal form $\ddot q=\Phi(q,\dot q)$, so that the initial value problem with given $q(0),\dot q(0)$ has a unique local solution when $a_{jk},V$ are $C^2$.
- 2 Let $L(q,\dot q,t)$ (one degree of freedom, $C^2$) have Lagrange expression $\frac{d}{dt}\frac{\partial L}{\partial\dot q}-\frac{\partial L}{\partial q}$ identically zero, for all functions $q(t)$. Prove that $L=a(q,t)\dot q+b(q,t)$ with $\partial a/\partial t=\partial b/\partial q$, and deduce that $L=dF/dt$ for some $F(q,t)$.
- 3 Prove that the motion of a particle in a central field lies in a fixed plane through the centre.
- 4 Give an example of a scleronomic system with $V=V(q,t)$ for which $h=T+V$ but $h$ is not conserved, and one with a rheonomic constraint for which $h$ is conserved but $h\ne T+V$.
- 5 Show that, in a central field, a pericentre and the next apocentre are separated by the apsidal angle $\displaystyle\Delta\theta=\int_{r_{\min}}^{r_{\max}}\frac{\ell\,dr}{r^2\sqrt{2m\big(E-V_{\text{eff}}(r)\big)}}$, and evaluate it for $V=-k/r$.
- 6 Prove that a circular orbit of radius $r_0$ under an attractive central force $F(r)\lt0$ is stable if $3F(r_0)+r_0F'(r_0)\lt0$ and unstable if $3F(r_0)+r_0F'(r_0)\gt0$. Deduce that for $F=-kr^{\nu}$ circular orbits are stable iff $\nu\gt-3$.
E1. CSIR Part B GATE MA A bead slides on a smooth circular hoop of radius $a$ that rotates with constant angular velocity $\omega$ about its vertical diameter, where $\omega^2=2g/a$. The angle $\theta$ (measured from the lowest point) of a stable position of relative equilibrium of the bead is (A) $0$; (B) $\pi/6$; (C) $\pi/3$; (D) $\pi/2$.
E2. CSIR Part B Three particles move in space so that all three mutual distances stay fixed, and in addition one of the particles is constrained to move on a fixed smooth sphere. The number of degrees of freedom is (A) $4$; (B) $5$; (C) $6$; (D) $8$.
E3. CSIR Part B For $L=\tfrac12m(\dot x^2+\dot y^2)+\alpha(x\dot y-y\dot x)-\tfrac12k(x^2+y^2)$, with constants $\alpha,k\gt0$, which quantity is conserved? (In polar coordinates $x=r\cos\theta$, $y=r\sin\theta$.) (A) $mr^2\dot\theta$; (B) $mr^2\dot\theta+\alpha r^2$; (C) $mr^2\dot\theta+2\alpha r^2$; (D) $\tfrac12m\dot r^2+\tfrac12kr^2$.
E4. CSIR Part C A bead can slide on a smooth straight horizontal wire that rotates with constant angular velocity $\omega$ about a vertical axis through a point $O$ of the wire. It is released from rest relative to the wire at distance $r_0$ from $O$. Which of the following are true? (A) $r(t)=r_0\cosh\omega t$. (B) The kinetic energy of the bead is conserved. (C) $\dot r^2-\omega^2r^2$ is constant. (D) The reaction of the wire does no work on the bead.
E5. CSIR Part C Which of the following Lagrangians lead to the equation of motion $\ddot q+q=0$? (A) $\tfrac12\dot q^2-\tfrac12q^2+3q^2\dot q$; (B) $\tfrac12q^2-\tfrac12\dot q^2$; (C) $\tfrac12\dot q^2-\tfrac12q^2+t\dot q$; (D) $\tfrac12(\dot q+q)^2-q^2$.
E6. CSIR Part C A particle moves in the central potential $V(r)=-k/r^n$ with $k\gt0$, $n\gt0$. Which of the following are true? (A) For every $r_0\gt0$ there is a circular orbit of radius $r_0$. (B) Circular orbits are stable if and only if $n\lt2$. (C) For $n=2$ circular orbits are stable. (D) The areal velocity $\tfrac12r^2\dot\theta$ is constant only for $n=1$.
Suggested plan — 3 lectures × 60 min.
- Constraints and coordinates. Definitions 8.1–8.2 with Example 1 done as a quick-fire class poll; the rolling coin and Theorem 8.1 with Example 2 (students compute the triple products). Theorem 8.2: draw the decomposition $T_2+T_1+T_0$ for the rotating wire. End with the ladder by virtual work (Theorem 8.3) — let students first try it with Newton's laws and count the unknowns.
- From D'Alembert to Lagrange. Definitions 8.3–8.4, Theorem 8.4, the lemma 8.5 and the full proof of Theorem 8.6 on the board (this proof is worth writing out completely once). Examples 3 and 5; project Interactive 1 and do Example 4 live while the class predicts what happens at $\omega=\omega_c$.
- First integrals. Theorems 8.9–8.11 with proofs; Example 6 with Interactive 2 (ask: why does the perturbed orbit not close, although $p_\theta$ and $E$ are conserved?). Theorem 8.7 and Example 7. Finish with Exam-corner E4 and E5.
Discussion questions. (1) The hoop's reaction does no virtual work but changes the bead's energy. Where exactly does the energy come from, and which term of $T$ records it? (2) Why can we not obtain Lagrange's equations for a rolling coin by substituting the rolling conditions into $L$? (3) If two Lagrangians give the same equations, must they differ by $cL+dF/dt$?
Misconceptions to watch for. Confusing rheonomic with non-holonomic; using actual instead of virtual displacements when checking that a constraint is workless; forgetting the $\partial\mathbf r/\partial t$ terms of a moving support; equating "no explicit $t$" with "energy conserved"; calling $\ell^2/2mr^2$ a potential energy; substituting a conserved momentum into $L$ before differentiating (this gives the wrong sign of the centrifugal term — see the Routhian in §8.2).
Suggested homework. I(1, 3, 5, 6), II(1, 3, 4, 6, 8), III(1, 2, 4, 6, 8), IV(2, 6); Exam corner E2, E3, E6.
Research question. How does the bead on a spinning hoop "choose" its new equilibrium as the rotation rate crosses $\omega_c=\sqrt{g/a}$, and what universal law governs the angle and the oscillation frequency near the threshold?
Why it matters. The same pitchfork bifurcation controls James Watt's centrifugal governor (Scotland, 1788), which regulated steam engines by letting flyballs rise as the speed increased; it is the simplest mechanical model of a phase transition (Landau's theory of symmetry breaking), and a close cousin of the instabilities studied in rotating machinery, MEMS gyroscopes and centrifuges today.
What to do.
- Write $\omega=\omega_c(1+\delta)$. Using $\cos\theta_*=g/(a\omega^2)$, compute $\theta_*$ for $\delta=0.01,0.02,0.05,0.1,0.2$ and plot $\log\theta_*$ against $\log\delta$. Estimate the slope and compare with the prediction $\theta_*\approx2\sqrt\delta$ obtained by expanding the cosine.
- Compute the small-oscillation frequency about the stable equilibrium on both sides of the threshold, $\sqrt{\omega_c^2-\omega^2}$ for $\omega\lt\omega_c$ and $\omega\sin\theta_*$ for $\omega\gt\omega_c$, and plot it against $\omega$. What happens at $\omega_c$ ("critical slowing down")?
- Use Interactive 1 (or a few lines of RK4) to release the bead at $\theta=0.01$ with $\delta=0.05$, and measure the time it takes to leave the neighbourhood of the bottom. Relate it to the growth rate $\sqrt{\omega^2-\omega_c^2}$.
- Add a small imperfection — a hoop axis tilted by a small angle, or a little friction — and describe how the symmetric pitchfork becomes an "imperfect" one in which one side is preferred.
Tools: Python or a spreadsheet; Interactive 1 for qualitative checks.
Hand in: A one-page report with the log–log plot and slope, the frequency graph, and two sentences on the imperfect case.
Going further. Read about the stability analysis of Watt's governor by J. C. Maxwell (1868, "On governors") and I. A. Vyshnegradsky (1877), often regarded as the birth of control theory.
🧪 Self-check quiz
Q1. A particle is constrained to the surface of a sphere whose radius increases uniformly with time. The constraint is
Q2. The number of degrees of freedom of a rigid body that can only turn about a fixed point is
Q3. If the coordinate $q_k$ does not appear in the Lagrangian, then
Q4. If $L(q,\dot q)$ has no explicit time dependence, the quantity that is always conserved is
Q5. Which Lagrangian gives the same equation of motion as $L=\tfrac12\dot q^2-\tfrac12q^2$?