Constraints and Lagrange Multipliers

The Idea in Plain Language

Imagine a bead sliding on a frictionless wire. The wire exerts a force on the bead to keep it on the path, but we do not know the magnitude or direction of this force at any given moment. In Newtonian mechanics, we would have to guess the force vector and solve a complex system of equations. In the Lagrangian approach, we treat the constraint as a mathematical condition that the position must satisfy. We introduce a new variable, the multiplier, which acts like a "price tag" for violating the constraint. By adding this term to the equations of motion, we effectively let the mathematics determine the exact force required to keep the system on the constraint surface. The multiplier itself is directly proportional to the constraint force, allowing us to extract it after solving the equations.

Formulating the Problem

Consider a system with \( n \) degrees of freedom described by generalized coordinates \( q_1, \dots, q_n \). Suppose there are \( m \) independent holonomic constraints, expressed as functions \( f_\alpha(q, t) = 0 \) for \( \alpha = 1, \dots, m \). The standard Lagrangian \( L = T - V \) describes the dynamics if the system were free. However, the actual motion is restricted to the submanifold defined by the constraints.

To incorporate these restrictions, we modify the action principle. Instead of requiring the action \( S = \int L \, dt \) to be stationary, we require the augmented action \( S^* \) to be stationary, where we add the constraints multiplied by time-dependent functions \( \lambda_\alpha(t) \):

The functions \( \lambda_\alpha(t) \) are the Lagrange multipliers. They are not known a priori; they are determined by the equations of motion along with the coordinates \( q_i(t) \).

Deriving the Equations of Motion

We apply the Euler-Lagrange equation to the augmented Lagrangian \( L^* = L + \sum_{\alpha} \lambda_\alpha f_\alpha \). The variation with respect to the coordinates \( q_i \) yields:

Since \( \lambda_\alpha \) does not depend on \( \dot{q}_i \), the first term is simply \( \frac{d}{dt} \left( \frac{\partial L}{\partial \dot{q}_i} \right) \). The second term splits into the partial derivative of \( L \) and the partial derivative of the constraint term:

The left-hand side is the standard Euler-Lagrange expression for the free system. The right-hand side represents the generalized forces of constraint, denoted \( Q_i^c \). Thus, the equation of motion becomes:

In addition to these \( n \) equations, we must satisfy the \( m \) constraint equations \( f_\alpha(q, t) = 0 \). This gives a total of \( n + m \) equations for the \( n + m \) unknowns \( q_i(t) \) and \( \lambda_\alpha(t) \).

Interpreting the Multipliers as Forces

The physical interpretation of \( \lambda_\alpha \) depends on the choice of coordinates. If the coordinates \( q_i \) are Cartesian coordinates \( x, y, z \), the generalized force \( Q_i^c \) is simply the component of the constraint force vector \( \mathbf{F}^c \) in the \( i \)-th direction.

For a single constraint \( f(x, y, z, t) = 0 \), the constraint force is normal to the surface defined by \( f = 0 \). The gradient \( \nabla f \) points in the direction of the steepest increase of \( f \), which is perpendicular to the surface. The equation for the force components is:

This can be written in vector form as:

Here, \( \lambda \) scales the gradient vector to provide the correct magnitude of the force. If the constraint is a rigid rod of length \( l \) attached to the origin, \( f = x^2 + y^2 + z^2 - l^2 = 0 \). Then \( \nabla f = 2x \mathbf{i} + 2y \mathbf{j} + 2z \mathbf{k} = 2\mathbf{r} \). The force is \( \mathbf{F}^c = 2\lambda \mathbf{r} \), which is directed radially, as expected for a central constraint force.