D'Alembert's Principle and Generalized Forces

Statement

For a system of N particles subject to ideal (workless) constraints, the total virtual work of the applied forces together with the inertial "reversed effective forces" vanishes for every virtual displacement consistent with the instantaneous constraints: Σ_i (F_i^(a) − ṗ_i) · δr_i = 0. Projected onto generalized coordinates q_j, this is equivalent to d/dt(∂T/∂q̇_j) − ∂T/∂q_j = Q_j, where Q_j = Σ_i F_i^(a) · ∂r_i/∂q_j is the generalized force.

Why it matters

D'Alembert's principle is the bridge from vectorial Newtonian mechanics to the analytical mechanics of Lagrange. Its decisive move is to eliminate the unknown constraint forces from the equations of motion entirely: because ideal constraints do no virtual work, they simply drop out of the balance, leaving only quantities we can actually compute — applied forces and kinetic energy.

It also reframes dynamics as a statement of virtual work, i.e. a variational-flavoured condition, rather than a set of vector equations. This is the conceptual seed of the whole Lagrangian and Hamiltonian program, and it lets us choose coordinates adapted to the constraints (angles, arc-lengths, mode amplitudes) instead of Cartesian components tied together by awkward reaction forces.

Assumptions

Derivation

Result

Limiting cases

• Statics (all q̇_j=q̈_j=0): T≡0, so the equations reduce to Q_j=0 — the principle of virtual work for equilibrium. D'Alembert is virtual work with inertia added.
• No constraints, Cartesian coordinates (q_j→x,y,z): ∂r_i/∂q_j are unit basis vectors, Q_j=F_j, T=½m|v|², and each equation is exactly mr̈=F. Newton is recovered as a special case.
• Conservative applied forces: Q_j=−∂V/∂q_j with V=V(q,t); the equations become d/dt(∂L/∂q̇_j)−∂L/∂q_j=0, L=T−V.
• Single free particle, one coordinate: reduces to the familiar 1-D Newton equation, confirming the projection introduces nothing new when there is nothing to constrain.

Breaks when

Failure modes

• Confusing δr with dr. Students use the actual trajectory displacement (which includes the constraint's own motion in time) rather than the frozen-time virtual displacement, so constraint work fails to cancel.
• Including constraint forces in Q_j. Adding tension, normal reactions, or rod forces into the generalized force double-counts what D'Alembert was designed to remove; only genuinely applied forces (gravity, springs, drives) enter Q_j.
• Forgetting the −∂T/∂q_j term. In curvilinear coordinates T depends on q (e.g. T=½m(ṙ²+r²θ̇²)); dropping ∂T/∂q loses the centrifugal/Coriolis pseudo-forces.
• Setting each bracket to zero for non-holonomic systems. The independence of δq_j is only guaranteed for holonomic constraints; applying step 13 to a rolling constraint gives wrong equations.
• Miscounting degrees of freedom. Using more coordinates than independent DOF (without multipliers) makes the δq_j dependent and the resulting equations inconsistent.
• Differentiating T before substituting the constraint. One must express T in the independent q,q̇ first; substituting constraints into T prematurely can silently drop terms.

Discussion

D'Alembert's genius was to convert a dynamics problem into a statics problem. By reversing the inertial force −ṗ_i and treating it on the same footing as an applied force, every instant of motion is recast as an equilibrium under the combined system of applied, constraint, and inertial forces. The principle of virtual work — already trusted for statics — then applies, and the ideal-constraint hypothesis does the real work: it annihilates precisely the forces we can never easily compute. What survives is a scalar (energy) statement, not a set of coupled vector equations, and scalars transform trivially under changes of coordinate. This is why we may adopt angles, arc-lengths, or normal-mode amplitudes freely.

The connection to the rest of analytical mechanics is direct. For conservative forces the principle becomes the Euler–Lagrange equations and, by Legendre transform, Hamilton's equations. It is closely related to, but logically distinct from, Hamilton's principle δ∫L dt=0: Hamilton's principle is an integral (global, over a path) variational statement, whereas D'Alembert's is a differential (local, at each instant) one. One can derive Hamilton's principle from D'Alembert's by integrating over time and discarding endpoint terms, which is why some texts call D'Alembert's principle the "central equation" of mechanics.

A subtlety worth stressing is the geometric status of the virtual displacement. The set {δr_i} lives in the tangent space to the constraint surface at fixed t; the vectors ∂r_i/∂q_j are a basis for that tangent space. D'Alembert's principle asserts that the vector (F^(a)−ṗ) is orthogonal to this tangent space, i.e. it lies purely in the space spanned by the constraint reactions (the normal bundle). For non-holonomic systems the admissible virtual displacements form a distribution that is not the tangent space of any surface, and orthogonality to it no longer permits component-wise vanishing — hence the multipliers. This is the precise geometric reason holonomic and non-holonomic constraints behave so differently, a distinction obscured if one works only with components.

Common misconceptions. D'Alembert's principle is not merely "F=ma with the terms moved to one side" — the physics is the vanishing of constraint virtual work, which is an independent postulate about the nature of ideal constraints. Nor is the "inertial force" −ṗ a real force in any frame; it is a bookkeeping device (an effective force) that becomes a genuine fictitious force only after one commits to a non-inertial frame. Finally, the principle is not restricted to statics-like slow motion: it holds instant by instant for arbitrary accelerated motion.

Worked examples

Problems

1. A bead of mass m slides on a frictionless horizontal wire; a horizontal applied force F(t) acts along the wire. Using x as generalized coordinate, write Q_x and the equation of motion, and identify why the normal reaction is absent.
   Solution

   The wire exerts a normal (vertical) reaction f⊥δr (which is horizontal), so f·δr=0 — ideal constraint, dropped by D'Alembert. T=½m ẋ², so d/dt(∂T/∂ẋ)=m ẍ, ∂T/∂x=0. The only applied force along the wire is F(t): Q_x=F(t). Equation of motion: m ẍ = F(t). The normal reaction never appears because it is orthogonal to every admissible virtual displacement.
2. For the plane pendulum of Worked Example 2, take the small-angle limit and find the period for l=0.50 m. State the approximation used and give a numeric answer.
   Solution

   For small θ, sinθ≈θ, so θ̈=−(g/l)θ, simple harmonic with ω=√(g/l). Period T=2π√(l/g)=2π√(0.50/9.81)=2π√(0.05097)=2π(0.2258)=1.42 s. Approximation: sinθ≈θ (error <1% for θ<14°); the amplitude-dependence of the true period is thereby neglected.
3. A particle moves in a plane described by polar coordinates (r,θ) with no constraints, under a central applied force F=F_rê_r. Using T=½m(ṙ²+r²θ̇²), derive the two equations of motion and show angular momentum is conserved.
   Solution

   Generalized forces: with r=rê_r, ∂r/∂r=ê_r and ∂r/∂θ=rê_θ, so Q_r=F_r, Q_θ=0 (central force has no θ-component).
   r-equation: ∂T/∂ṙ=m ṙ, d/dt=m r̈; ∂T/∂r=m rθ̇². Thus m r̈ − m rθ̇² = F_r (the −m rθ̇² is the centrifugal term).
   θ-equation: ∂T/∂θ̇=m r²θ̇, d/dt(m r²θ̇)=0; ∂T/∂θ=0. So d/dt(m r²θ̇)=0 ⇒ L=m r²θ̇ is conserved. The vanishing of Q_θ is exactly the statement that a central force exerts no torque about the origin.
4. A block of mass m rests on a frictionless wedge of mass M that is free to slide on a frictionless floor; the incline angle is α. Using generalized coordinates X (wedge position) and s (block's displacement along the incline relative to the wedge), set up Q_X, Q_s and the kinetic energy, then write the two equations of motion. (Do not solve fully; reduce to two coupled equations.)
   Solution

   Let the incline descend to the right at angle α. Block position: x_b=X+s cosα, y_b=−s sinα. Velocities: ẋ_b=Ẋ+ṡcosα, ẏ_b=−ṡsinα.
   T=½M Ẋ²+½m[(Ẋ+ṡcosα)²+(ṡsinα)²]=½M Ẋ²+½m[Ẋ²+2Ẋṡcosα+ṡ²].
   Applied force is gravity on the block, (0,−mg). Q_X=(0,−mg)·∂r_b/∂X=(0,−mg)·(1,0)=0. Q_s=(0,−mg)·∂r_b/∂s=(0,−mg)·(cosα,−sinα)=mg sinα.
   X-equation: d/dt[(M+m)Ẋ+m ṡcosα]=0 ⇒ (M+m)Ẍ+m s̈cosα=0 (horizontal momentum conserved, since Q_X=0).
   s-equation: ∂T/∂ṡ=m(Ẋcosα+ṡ), ∂T/∂s=0, so m(Ẍcosα+s̈)=mg sinα. These are the two coupled equations; note the constraint (normal) force between block and wedge has dropped out.
5. State precisely, with one worked reason each, two physical situations in which the step "each bracket vanishes because the δq_j are independent" (Derivation step 13) is invalid, and say what replaces it.
   Solution

   (i) Non-holonomic constraint — e.g. a vertical disc rolling without slipping on a plane. The rolling condition relates velocities (ẋ=Rφ̇cosθ, etc.) but is not integrable to a relation among coordinates, so the δq_j satisfy differential relations Σ_ja_ajδq_j=0 and are not independent. Replacement: introduce Lagrange multipliers λ_a, giving d/dt(∂T/∂q̇_j)−∂T/∂q_j=Q_j+Σ_aλ_aa_aj. (ii) Redundant coordinates — if one uses n coordinates for a system with fewer independent DOF (a holonomic constraint g(q,t)=0 still imposed), then Σ_j(∂g/∂q_j)δq_j=0 makes the δq_j dependent. Replacement: again a multiplier λ times ∂g/∂q_j, or eliminate one coordinate using g=0 to restore independence. In both cases the naive "set each bracket to zero" gives wrong equations of motion.