D'Alembert's Principle

Why it matters

Newton's second law, F = ma, is complete but clumsy for systems that are wired together. Every rigid rod, taut string, smooth rail, or contact point introduces an unknown constraint force — a tension, a normal force, a reaction at a hinge — that you must carry through every equation even though you usually don't care about its value. A double pendulum has four constraint forces you'd rather never see; a bead sliding on a wire hides a normal force that points in a direction that keeps changing.

D'Alembert's Principle is the trick that makes all of that vanish. By moving the ma term to the force side and calling −ma an inertial force, Newton's law becomes a statement of equilibrium: the applied forces and the inertial forces balance. And equilibrium problems have a powerful tool — the principle of virtual work, which says that in equilibrium the total work done by all forces during any allowed virtual displacement is zero. Because ideal constraint forces do no virtual work, they never appear. What remains is one clean scalar equation per degree of freedom, expressed in whatever generalized coordinates are natural to the problem. This is the doorway to Lagrangian mechanics and, through it, to Hamiltonian mechanics, field theory, and Noether's theorem.

The governing equation, symbol by symbol

For a system of N particles, D'Alembert's Principle in the virtual-work form is:

Σ_i ( F_i − m_i·a_i ) · δr_i = 0     for all virtual displacements δr_i

Every symbol matters:

Symbol	Meaning	Units (SI)
F_i	Net applied (impressed) force on particle i — gravity, springs, thrust, etc. Does not include constraint forces.	newton (N)
m_i	Mass of particle i	kilogram (kg)
a_i	Acceleration of particle i, a_i = d²r_i/dt², measured in an inertial frame	m/s²
−m_i·a_i	The inertial force (d'Alembert force / reversed effective force)	newton (N)
δr_i	Virtual displacement of particle i — instantaneous (δt = 0), infinitesimal, consistent with the constraints	metre (m)
·	Vector dot product (so each term is a virtual work, a scalar)	joule (J)

The term −m_i·a_i is the reversed effective force (the classical name for the inertial force). Note that F_i − m_i·a_i using the applied force alone is not zero for each particle — it equals minus the constraint force, since Newton's second law gives zero only for the full force (applied + constraint): (F_i + C_i) − m_i·a_i = 0. D'Alembert's insight is that when you dot the applied-force version with a virtual displacement and sum, the constraint contribution cancels, leaving a relation among the applied forces and the motion alone.

How it works, step by step

Step 1 — Split the force. On each particle the true total force is the applied force plus the constraint force: F_i^total = F_i + C_i. Newton's law says F_i^total − m_i·a_i = 0, i.e. (F_i + C_i) − m_i·a_i = 0 for every particle. This is exact but full of unknown C_i.

Step 2 — Dot with a virtual displacement and sum. Take the scalar product of each particle's equation with its virtual displacement δr_i and add them up:

Σ_i ( F_i + C_i − m_i·a_i ) · δr_i = 0

Step 3 — Kill the constraint forces. For ideal constraints the reaction C_i is perpendicular to the allowed motion (a normal force is normal to the surface, a rod tension is along the rod which cannot stretch, a rail reaction is transverse to the rail). A virtual displacement δr_i lies within the allowed motion, so C_i·δr_i = 0 for each particle and Σ_i C_i·δr_i = 0. This is the principle of virtual work for constraint forces. What survives is D'Alembert's Principle:

Σ_i ( F_i − m_i·a_i ) · δr_i = 0

Step 4 — Switch to generalized coordinates. If the configuration is described by n independent generalized coordinates q_1 … q_n (with n = degrees of freedom), then r_i = r_i(q_1,…,q_n,t) and

δr_i = Σ_j (∂r_i/∂q_j) δq_j

The δq_j are now independent. Substituting and collecting terms, the coefficient of each δq_j must vanish separately.

Step 5 — Read off the Lagrange equations. Using the two "d'Alembert identities" that convert Σ m_i·a_i·(∂r_i/∂q_j) into kinetic-energy derivatives, the surviving relation becomes, for each j:

d/dt ( ∂T/∂q̇_j ) − ∂T/∂q_j = Q_j

where T is the total kinetic energy and Q_j = Σ_i F_i·(∂r_i/∂q_j) is the generalized force conjugate to q_j. If the applied forces are conservative, F_i = −∇_i V, then Q_j = −∂V/∂q_j and, defining the Lagrangian L = T − V, we obtain the Euler–Lagrange equations:

d/dt ( ∂L/∂q̇_j ) − ∂L/∂q_j = 0

Not a single constraint force ever appeared. That is the whole point.

Virtual work vs. real work — what δ actually means

The symbol δ is not the same as d. The distinction is the heart of the method:

Property	Real displacement dr	Virtual displacement δr
Time elapsed	dt > 0 (real time passes)	δt = 0 (frozen instant)
Obeys equations of motion?	Yes — it is the actual trajectory	No — it is an imagined variation
Consistency with constraints	Consistent, including how they move in time	Consistent with constraints as they are at that instant
For a moving (rheonomic) constraint	Includes the (∂r/∂t)dt drift of the constraint	Excludes it — the constraint is held fixed
Work done by ideal constraint force	Can be nonzero (e.g. a moving wall)	Always zero

This is why the principle uses δr, not dr: only virtual displacements guarantee that the constraint forces contribute nothing, because they freeze the constraints at the current instant. For time-independent (scleronomic) constraints the two agree in direction, but the conceptual distinction remains essential — it is exactly what lets d'Alembert dispose of the reaction forces cleanly.

Worked example — the Atwood machine in one line

Two masses m₁ and m₂ hang over an ideal, massless, frictionless pulley connected by an inextensible string of fixed length. The constraint (string length constant) means the two masses move oppositely with equal speed. Let the single generalized coordinate be q = the downward displacement of m₁ (so m₂ moves upward by the same q). Then:

• Position of m₁: descends by q ⇒ acceleration a₁ = q̈ (downward positive).
• Position of m₂: rises by q ⇒ acceleration a₂ = q̈ (upward), i.e. −q̈ downward.
• Virtual displacements: δr₁ = +δq (down), δr₂ = −δq (down) — the string forces (tensions) do zero net virtual work, which is precisely why we never write the tension.

Apply D'Alembert, using downward-positive and gravity g on each applied (weight) force:

(m₁·g − m₁·q̈)·δq  +  (m₂·g − m₂·(−q̈))·(−δq) = 0

⇒ [ m₁·g − m₁·q̈ − m₂·g − m₂·q̈ ] δq = 0

⇒ q̈ = (m₁ − m₂) / (m₁ + m₂) · g

The classic Atwood result a = (m₁ − m₂)g/(m₁ + m₂) falls out with the string tension never mentioned. A Newtonian treatment would require two free-body diagrams and elimination of the tension T. Compare with the fuller treatment on the Atwood machine and pulley systems pages. Applied to a bead on a rotating wire, a rolling disk, or a simple pendulum, the same three lines deliver the equation of motion.

A little history

Jean le Rond d'Alembert (1717–1783) — found as an infant on the steps of the church of Saint-Jean-le-Rond in Paris, later a co-editor with Diderot of the Encyclopédie — published his principle in the Traité de dynamique in 1743, when he was 26. His original statement was subtler than the modern slogan: he separated the impressed motion into the part that is "gained" (the actual, effective motion) and the part that is "lost," and asserted that the lost part is exactly what the constraints hold in equilibrium. It was Joseph-Louis Lagrange who, in the Mécanique analytique (1788), fused d'Alembert's principle with the principle of virtual work to produce the compact Σ(F − ma)·δr = 0 statement and the generalized-coordinate machinery that carries his name. The lineage runs directly: Bernoulli's virtual work → d'Alembert's dynamics-as-statics → Lagrange's analytical mechanics → Hamilton's principle and the whole edifice of modern theoretical physics.

Common misconceptions

• "−ma is a real force." It is not. There is no reaction partner (Newton's third law does not apply to it), and in a truly inertial frame it is a mathematical bookkeeping term. It only feels like a force when you sit in an accelerating frame, where it reappears as a genuine fictitious force.
• "D'Alembert's Principle needs a non-inertial frame." No — it is stated entirely in an inertial frame. The confusing overlap is that the fictitious forces of an accelerating frame have the same −ma form.
• "δr and dr are the same." Only for time-independent constraints do they point the same way. For moving constraints they differ, and the whole cancellation of constraint forces depends on using the frozen-instant δr.
• "It works with friction and other dissipative contacts." The clean cancellation requires ideal (workless) constraints. Sliding friction does virtual work and must be kept as an explicit generalized force Q_j; only its normal reaction drops out, not the friction itself.
• "You still have to compute the constraint forces." Not to get the motion — they cancel. If you actually want a tension or normal force, you reintroduce it afterward with a Lagrange multiplier or by isolating a subsystem.
• "It's just Newton's law rewritten, so it adds nothing." Logically it is equivalent to F = ma, but the reformulation is what makes constrained, many-body problems tractable and is the historical and conceptual bridge to Lagrangian and Hamiltonian mechanics.

Where D'Alembert's Principle shows up

• Analytical mechanics. The standard derivation of the Lagrange equations starts here; it is the load-bearing step between Newton and Lagrange.
• Multibody dynamics & robotics. Manipulator arms, vehicle suspensions, and linkages are modeled with generalized coordinates precisely to avoid solving for every joint reaction.
• Structural & earthquake engineering. Adding the inertial force −m·ü to the elastic and damping forces turns a vibrating structure into an "equilibrium" problem — the basis of the equation of motion Mü + Cu̇ + Ku = F.
• Rigid-body dynamics. Rolling without slipping, gyroscopes, and spinning tops are cleanest with generalized coordinates and the d'Alembert–Lagrange route.
• Continuum & field theory. The variational principles of elasticity and field theory descend from the same virtual-work idea extended to continuous systems.