The Lagrangian formalism

Constraints

The motion of a particle may be restricted in an arbitrary way via contraints, for example, a pendulum attached to a string of length \(L\) has a contraint

\[x^2 + y^2 = L^2\]

In general a contraint can be characterised as a force, in the case of the pendulum this is tension.

A holonomic constraint has the form

\[g(\vec{x}_1, \vec{x}_2, \dots, \vec{x}_n, t) = 0\]

If the constraint has time-dependence it is rheonomous, otherwise it is scleronomous. Not every contraint is holonomic; the constraint that a particle may not enter a sphere is an inequality, and thus non-holonomic, for example.

Generalised Coordinates

Generalised coordinates are chosen to simplify the description of a system, and are linearly independent. These are a set of independent quantities \(\set{q_i}\), such that

\[\label{eq:11} \vec{x}_i = \vec{x}_i(\set{q_i}, t)\]

The number of degrees of freedom, \(f\), for a system is equal to the product of the system’s dimensionality, \(D\), and its number of particles, \(N\), less the number of constraints, i.e.

\[f = DN - k\]

and this results in a system requiring \(f\) generalised coordinates to fully characterise it.

Virtual Work \(\star\)

A virtual displacement of a system is a change in a system’s configuration as a result of any arbitrary infinitessimal change of its coordinates, \(\delta \vec{r}_i\).

The virtual work of a force \(\vec{F}_i\) is \(\vec{F}_i \vdot \delta \vec{r}_i\). If the system is in equilibrium then \(\vec{F}_i = 0\), so the virtual work also vanishes. For the whole system

\[\sum \vec{F}_i \vdot \delta \vec{r}_i = 0\]

We can express a force in the form \(\vec{F}_i = \vec{F}_i^{\rm (a)} +\vec{f}_i\), the sum of an applied force and a constraint, so

\[ \begin{align}\begin{aligned} \label{eq:12} \sum \vec{F}_i^{\rm (a)} \vdot \delta \vec{r}_i + \sum_i \vec{f}_i \vdot \delta \vec{r}_i =0\\If the constraint forces produce no net virtual work (excluding e.g.\end{aligned}\end{align} \]

sliding friction),

\[ \begin{align}\begin{aligned} \label{eq:13} \sum \vec{F}_i^{\rm (a)} \vdot \delta \vec{r}_i = 0\\Which is the principle of virtual work.\end{aligned}\end{align} \]

Thanks to the constraints \(\delta \vec{r}_i\) are not independent, so a form for the general motion of the system in general coordinates is required.

D’Alembert’s Principle

Take the equation of motion,

\[ \begin{align}\begin{aligned}\vec{F}_i = \dot{\vec{p}}_i\\which can be expressed\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\vec{F}_i - \dot{\vec{p}}_i = 0\\Then\end{aligned}\end{align} \]
\[\begin{split}\begin{aligned} \label{eq:14} \sum (\vec{F}_i - \dot{\vec{p}}_i) \vdot \delta \vec{r}_i &=0 \\ \sum_i (\vec{F}_i^{\rm (a)} - \dot{\vec{p}}_i ) \vdot \delta \vec{r}_i + \sum \vec{f}_i \vdot \delta \vec{r}_i &= 0 \\ \sum_i (\vec{F}_i^{\rm (a)} - \dot{\vec{p}}_i ) \vdot \delta \vec{r}_i &= 0\end{aligned}\end{split}\]

which is D’Alembert’s Principle. To convert this to generalised coordinates we use a transformation

\[ \begin{align}\begin{aligned} \label{eq:15} \delta \vec{r}_i = \sum_j \pdv{\vec{r}_i}{q_j} \delta q_j\\using the summation convention, and defining\end{aligned}\end{align} \]

\(\Lambda^j_i = \pdv{\vec{r}_i}{q_j}\),

\[ \begin{align}\begin{aligned}\delta \vec{r}_i = \Lambda^j_i \delta q_j\\Then the virtual work becomes\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:16} \sum_i \vec{F}_i \vdot \delta \vec{r}_i = \sum_{i,j} \vec{F}_i \vdot \Lambda^j_i \delta q_j = \sum_j Q_j \delta q_j\\with :math:`Q_j` the components of a virtual force,\end{aligned}\end{align} \]
\[Q_j = \sum_i \vec{F}_i \Lambda^i_j\]

We also have the reversed effective force in equation ,

\[\label{eq:17} \sum \dot{\vec{p}}_i \vdot \delta \vec{r}_i = \sum m_i \ddot{\vec{r}}_i \vdot \delta \vec{r}_i = \sum_{i,j} m_i \ddot{\vec{r}}_i \Lambda^j_i \delta q_j\]

Then

\[ \begin{align}\begin{aligned}\begin{split} \begin{aligned} \sum m_i \ddot{\vec{r}}_i \vdot \Lambda_i^j &= \sum_i \qty[ \dv{t} \qty(m_i \dot{\vec{r}} \vdot \Lambda_i^j ) - m_i \dot{\vec{r}} \vdot \dv{t} \qty( \Lambda^j_i ) ] \\ &= \sum_i \qty[ \dv{t} \qty( m_i \vec{v}_i \vdot \pdv{\vec{v}_i}{\dot{q}_j} ) - m_i \vec{v}_i \pdv{\vec{v}_i}{q_j}]\end{aligned}\end{split}\\Since\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\begin{split} \begin{aligned} \dv{t} \pdv{\vec{r}_i}{q_j} &= \pdv{\vec{v}}{q_j} \\ \pdv{\vec{v}_i}{\dot{{q}}_j} &= \pdv{\vec{r}_i}{q_j}\end{aligned}\end{split}\\Then equation can be expanded to\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\begin{split} \begin{aligned} \sum_j & \bigg\{ \dv{t} \qty[ \pdv{\dot{q}_j} \qty( \sum_i \half m_i v_i^2) ] \\ & \quad- \pdv{q_j} \qty( \sum_i \half m_i v_i^2 ) - Q_j \bigg\} \delta q_j\end{aligned}\end{split}\\Then\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:18} \sum \qty{ \qty[ \dv{t} \qty( \pdv{T}{\dot{q}_j} ) - \pdv{T}{q_j} ] - Q_j } \delta q_j = 0\\For :math:`T` the kinetic energy, and so,\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:19} \dv{t} \qty( \pdv{T}{\dot{q}_j} ) - \pdv{T}{q_j} = Q_j\\When the forces are produced by a potential,\end{aligned}\end{align} \]

\(\vec{F}_i = - \nabla_i V\),

\[ \begin{align}\begin{aligned}Q_j = - \sum_i \nabla_i V \vdot \pdv{\vec{r}_i}{q_j} = - \pdv{V}{q_i}\\so we now have\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:20} \dv{t} \qty( \pdv{T}{\dot{q}_j} ) - \pdv{(T-V)}{q_i} = 0\\and defining a function :math:`L = T - V`, the *Lagrangian*, and noting\end{aligned}\end{align} \]

that \(\pdv{V}{\dot{q}_j} = 0\), we get

\[\label{eq:21} \dv{t} \qty( \pdv{L}{\dot{q}_j} ) - \pdv{L}{q_j} = 0\]

which are Lagrange’s equations.

Velocity-dependent Potentials \(\star\)

Suppose there is no potential \(V\) to generate the generalised forces, but they can instead be found from a function \(U(q_j, \dot{q_j})\) by

\[ \begin{align}\begin{aligned} \label{eq:22} Q_j = - \pdv{U}{q_j} + \dv{t}\qty( \pdv{U}{\dot{q}_j} )\\The Lagrangian is now\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}L = T-U\\and :math:`U` is a “generalised potential”. Such a potential is of\end{aligned}\end{align} \]

importance in electromagnetism.

The Electromagnetic Vector Potential

Consider the force on a charge,

\[ \begin{align}\begin{aligned} \label{eq:23} \vec{F} = q \qty[ \vec{E} + \vec{v} \cp \vec{B}]\\for :math:`\vec{E} = \vec{E}(x,y,z,t)` and\end{aligned}\end{align} \]

\(\vec{B} = \vec{B}(x,y,z,t)\) being continuous functions of time and position. These can be derived from a scalar potential and a vector potential, respectively \(\phi(t,x,y,z)\) and \(\vec{A}(t,x,y,z)\):

\[\begin{split}\begin{aligned} \vec{E} & = - \nabla \phi - \pdv{\vec{A}}{t} \\ \vec{B} & = \nabla \cp \vec{A} \end{aligned}\end{split}\]

The force can then be derived from the potential \(U\),

\[ \begin{align}\begin{aligned} \label{eq:24} U = q \phi - q \vec{A} \vdot \vec{v}\\and the Lagrangian is then\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:25} L = \half m v^2 - q \phi + q \vec{A} \vdot \vec{v}\\This can then be used to derive equation .\end{aligned}\end{align} \]

Dissipation \(\star\)

If not all of the forces in a system can be derived from a potential then

\[ \begin{align}\begin{aligned} \label{eq:26} \dv{t} \qty( \pdv{L}{\dot{q}_j} ) - \pdv{L}{q_j} = Q_j\\This happens in the case of friction, where there is a force\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}F_{{\rm f}x} = - k_x v_x\\Such a force can be considered by the dissipation function,\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:27} \mathcal{F} = \half \sum_i \qty( k_x v_{ix}^2 + k_y v_{iy}^2 + k_z v_{iz}^2 )\\where\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\vec{F}_{{\rm f}x} = - \pdv{\mathcal{F}}{v_x} \implies \vec{F}~{f} = - \nabla_v \mathcal{F}\\The Lagrange equations then become\end{aligned}\end{align} \]
\[\label{eq:28} \dv{t} \pdv{L}{\dot{q}_j} - \pdv{L}{q_j} + \pdv{\mathcal{F}}{\dot{q}_j} = 0\]

Hamilton’s Principle

It is possible to derive the Lagrange equations for a system from an integrated perspective of the motion, using Hamilton’s Principle (the principle of least action):

The motion of a system from a time \(t_1\) to \(t_2\) is such that the line integral

\[I = \int_{t_1}^{t_2} L \dd{t}\]

for \(L = T - V\) has a stationary value for the actual path of the motion.

We define the action as the integral from Hamilton’s Principle,

\[\label{eq:29} S( \set{q_i}, \set{\dot{q}_i} ) = \int_{t_0}^{t_1} \dd{t} L(\set{q_i}, \set{\dot{q}_i}, t)\]

To derive the Lagrange equations introduce a small perturbation to \(q_i\),

\[q_i(t) \to q_i(t) + \delta q_i(t)\]

The trajectory has fixed end-points, so \(\delta q_i(t_0) = \delta q_i(t_1) = 0\), so

\[\dot{q}_i \to q_i + \delta \dot{q}_i, \quad \delta \dot{q}_i = \dv{t} \delta q_i\]

This perturbs the action,

\[ \begin{align}\begin{aligned}S \to S + \var{S} = \int_{t_0}^{t_1} \dd{t} L(q_i+\var{q_i}, \dot{q}_i + \var{\dot{q}_i}, t)\\Using Taylor’s theorem,\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:30} S + \var{S} = \int_{t_0}^{t_1} \dd{t} \qty[ L + \sum_i \qty( \pdv{L}{q_i} \var{q_i} + \pdv{L}{\dot{q}_i} \var{\dot{q}_i} ) ] + \cdots\\Then\end{aligned}\end{align} \]
\[\begin{split}\begin{aligned} \label{eq:31} \var{S} &= \sum_i\int_{t_0}^{t_1} \dd{t} \qty[ \pdv{L}{q_i} \var{q_i} + \pdv{L}{\dot{q}_i} \var{\dot{q}_i} ] \\ &= \sum_i \qty{ \underbracket{\eval{ \pdv{L}{\dot{q}_i} \var{q_i} }_{t_0}^{t_1}}_{=0} + \underbracket{\int_{t_0}^{t_1} \dd{t} \qty[\pdv{L}{q_i} - \dv{t} \pdv{L}{\dot{q}_i} ] \delta q_i }_{\text{Must be zero to extremise action}} } \\ \therefore \dv{t} \pdv{L}{\dot{q}_i}&= \pdv{L}{q_i}\end{aligned}\end{split}\]

The variational approach to finding the Lagrangian allows easy extension to systems which are not normally the domain of dynamics, for example the descriptions of electrical circuits.

Canonical Momentum

Recall the Lagrangian for a particle moving in one dimension,

\[ \begin{align}\begin{aligned}L = \half m \dot{x}^2 - V(x)\\from which\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\pdv{L}{x} = m \dot{x}\\which is the particle’s momentum. Given a general Lagrangian the\end{aligned}\end{align} \]

quantity

\[ \begin{align}\begin{aligned} \label{eq:32} p_i = \pdv{L}{\dot{q}_i}\\is the generalised momentum, or the canonically conjugate momentum to\end{aligned}\end{align} \]

\(q_i\).

Symmetries and Conservation

If the Lagrangian of a system doesn’t explicity contain a coordinate \(q_i\), (but may contain \(\dot{q}_j\)) it is called cyclic, so

\[ \begin{align}\begin{aligned} \label{eq:33} \pdv{L}{q_i} = 0 \implies \dv{t}\pdv{L}{\dot{q}_i} = \dot{p}_i = 0\\Therefore, the momentum conjugate to a cyclic coordinate is conserved.\end{aligned}\end{align} \]

The Energy Function

Consider the time derivative of \(L\),

\[ \begin{align}\begin{aligned}\begin{split} \begin{aligned} \label{eq:34} \dv{L}{t} &= \sum_i \pdv{L}{q_i} \dv{q_i}{t} + \sum_i \pdv{L}{\dot{q}_i} \dv{\dot{q}_i}{t} + \pdv{L}{t} \\ &= \sum_j \dv{t} \qty(\pdv{L}{\dot{q}_j}) \dot{q}_j + \sum_j \pdv{L}{\dot{q}_j} \dv{\dot{q}_j}{t} +\pdv{L}{t}\end{aligned}\end{split}\\Thus\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned}\begin{split} \begin{aligned} \dv{t} \qty( \sum_j \dot{q}_j \pdv{L}{\dot{q}_j} - L ) + \pdv{L}{t} &= 0 \\ \dv{H}{t} &= - \pdv{L}{t}\end{aligned}\end{split}\\For\end{aligned}\end{align} \]
\[ \begin{align}\begin{aligned} \label{eq:35} h = \sum_i \dot{q}_i \pdv{L}{\dot{q}_i} - L\\defined as the “Energy function”, which is physically, if not\end{aligned}\end{align} \]

mathematically, identical to the Hamiltonian. Thus

\[ \begin{align}\begin{aligned} \label{eq:92} \dv{h}{t} = - \pdv{L}{t}\\If the Lagrangian is not explicitly a function of time then :math:`h`\end{aligned}\end{align} \]

is conserved. This is one of the first integrals of the motion, and is Jacobi’s integral.

Under some circumstances \(h\) is the total energy of a system; recall that the total kinetic energy of a system can be expressed

\[ \begin{align}\begin{aligned} \label{eq:93} T = T_0 + T_1 + T_2\\with :math:`T_0` a function of only the generalised coordinates,\end{aligned}\end{align} \]

\(T_1(q, \dot{q})\) is linear in the generalised velocities, and \(T_2(q, \dot{q})\) is quadratic in \(\dot{q}\). For a wide range of systems we can similarly decompose the Lagrangian,

\[ \begin{align}\begin{aligned} \label{eq:94} L = L_0 + L_1 + L_2\\If a function :math:`f` is homogeneous and of degree :math:`n` in the\end{aligned}\end{align} \]
variables \(x_i\), then :math:`

sum_i x_i pdv*{f}{x_i} = nf

` applied to the function \(h\),

\[ \begin{align}\begin{aligned} \label{eq:96} h = 2 L_2 + L_1 - L = L_2 - L_0\\If the transformations to generalised coordinates are time independent\end{aligned}\end{align} \]

\(T = T_2\), and then if the potential doesn’t depend on generalised velocity, \(L_0 = -V\), so

\[\label{eq:97} h = T + V = E\]