d'Alembert's Principle

In 1743, a twenty-five-year-old Parisian who had been abandoned as a baby on the steps of a church published a book with an audacious thesis: there is no such thing as a dynamics problem. Every problem of motion, Jean le Rond d'Alembert argued in his Traité de dynamique, can be converted into a statics problem — a problem of equilibrium, the kind mechanicians had known how to solve since Archimedes — by one act of bookkeeping. Take Newton's law $\mathbf{F} = \dot{\mathbf{p}}$ , move the momentum term across the equals sign, and read $\mathbf{F} - \dot{\mathbf{p}} = \mathbf{0}$ as a balance: the applied forces plus a fictitious inertial force $-\dot{\mathbf{p}}$ hold every particle in equilibrium at every instant. On a single free particle the move is a triviality. Its power appears the moment constraints do. A constrained system — a pendulum, a rolling wheel, a linkage, a bead on a spinning hoop — hides its dynamics behind a crowd of unknown constraint forces: tensions, normal reactions, rod stresses, one vector unknown per contact. But statics already owns the perfect weapon against exactly that crowd: the principle of virtual work, which tests equilibrium using probes that ideal constraint forces cannot touch. Weld the inertial force onto the virtual-work principle and the constraint forces vanish from dynamics the way they vanished from statics: $\sum_i (\mathbf{F}_i^{(a)} - \dot{\mathbf{p}}_i)\cdot\delta\mathbf{r}_i = 0$ . That single line is d'Alembert's principle — the exact, constraint-free projection of Newton's laws, and the engine room of analytical mechanics. Feed it generalized coordinates and it hands you Lagrange's equations; keep the forces general and it handles friction and drag that no action principle can touch. Every multibody simulator and humanoid-robot controller running today is, at bottom, executing this 1743 idea at kilohertz rates.

Key takeaways

d'Alembert's principle rewrites Newton's law $\mathbf{F}_i = \dot{\mathbf{p}}_i$ as a balance $\mathbf{F}_i - \dot{\mathbf{p}}_i = \mathbf{0}$ : adding the inertial force $-\dot{\mathbf{p}}_i = -m_i\ddot{\mathbf{r}}_i$ puts every particle in "dynamic equilibrium" at every instant.
Dotting that balance into virtual displacements and using the ideal-constraint condition kills every constraint force: $\sum_i(\mathbf{F}_i^{(a)}-\dot{\mathbf{p}}_i)\cdot\delta\mathbf{r}_i=0$ — tensions, normals, and rod stresses never appear.
Geometrically, the principle is Newton's equation projected onto the directions the constraints allow; the constraint forces are exactly the normal components that the projection discards.
It is equivalent to Newton's laws plus the postulate that constraint forces do no virtual work — no new physics, but a radical reorganization that yields exactly one equation per degree of freedom.
Expressed in generalized coordinates it becomes $\sum_j\big[Q_j - \big(\text{inertia in the } q\text{'s}\big)\big]\,\delta q_j = 0$ — one substitution away from the Euler–Lagrange equations.
It is a differential principle (true instant by instant, no trajectory comparison) and strictly more general than Hamilton's action principle: friction and drag enter as generalized forces with no potential required.
Key idea: add the reversed inertial force $-\dot{\mathbf p}_i$ to each particle so the system is in dynamic equilibrium, then demand that the total (applied + inertial) force do no virtual work — ideal constraint forces drop out automatically, and projecting onto generalized coordinates yields the Euler–Lagrange equations.

Where this lives today

Every physics engine that animates a robot, a vehicle chassis, or a game character is a d'Alembert machine. Multibody dynamics codes — the simulators used to train and validate today's humanoid robots, and the solvers inside CAD suites — assemble exactly equation (5): project the Newton–Euler equations of every link onto the joint coordinates, let the joint reaction forces cancel because they do no virtual work, and integrate the $n$ surviving equations. The modern closed form of this projection, the Udwadia–Kalaba equation (1992), computes the constrained acceleration directly from the unconstrained one plus the constraint geometry — d'Alembert's principle as a single matrix formula, executed millions of times per second across the robotics industry. When a 2026 whole-body controller decides in under a millisecond how sixty joint torques should respond to a shove, the reason the computation is tractable at all is that the sixty equations it solves contain no bearing reactions: d'Alembert deleted them three centuries in advance. The same projection idea, applied to continua, underlies the "weak form" of the equations of motion on which dynamic finite-element analysis is built — crash simulation, earthquake response of buildings, vibration of turbine blades. The test functions of the weak form are virtual displacements by another name, and the statement "internal constraint stresses drop out of the weighted residual" is equation (1) in engineering dress. From a braking bus to a crash-test supercomputer, the trick is the one d'Alembert sold in 1743: add $-\dot{\mathbf{p}}$ , pretend it is statics, and let the geometry pay its own enforcers.

The physical picture

Stand in a bus that brakes hard. In the ground frame the story is plain Newton: friction from the floor decelerates your feet while your torso tries to keep going. But your body runs a different accounting: you feel pushed forward, as if a force $-m\mathbf{a}$ had materialized, and you lean back against it exactly as you would lean against a steady wind. Your nervous system is doing d'Alembert's bookkeeping: adding the inertial force to the real ones and solving a statics problem — stay balanced — at every instant. d'Alembert's principle is the license to do this everywhere: append $-\dot{\mathbf{p}}$ to the force ledger and every dynamics problem becomes an equilibrium problem, solvable with the frozen-time probes of virtual work.

Figure

Dynamic equilibrium — adding the inertial force closes the force polygon at every instant

Why does this buy anything? Because statics owns the one tool dynamics lacked: a systematic way to never meet a constraint force. An ideal constraint force is perpendicular to everything the constraint allows; a virtual displacement is, by construction, along what the constraint allows. In statics that orthogonality proved a lever balances without knowing the pivot force. d'Alembert noticed that the argument never used the word "rest" — it needs only a balance of forces, and $\mathbf{F} - \dot{\mathbf{p}} = \mathbf{0}$ is a balance. So the entire virtual-work machinery lifts to moving systems verbatim, and the unknown tensions and normals exit dynamics forever. There is a second, deeper way to see the same move — as a projection. Picture the system's configuration as a point sliding on the constraint surface inside configuration space. Newton's equation is a vector statement in the full space; it has components along the surface (the directions the system can actually explore) and components perpendicular to it (the directions the constraints forbid). The perpendicular components are pure constraint-force bookkeeping — whatever the geometry needs, the geometry supplies. The tangential components are the real, autonomous dynamics. d'Alembert's principle is exactly the statement: project Newton onto the tangent directions and keep everything; what you discarded was never dynamics at all.

Figure

The projection picture — Newton lives in the full space, dynamics lives tangent to the constraint surface

One more framing, the practical one. Think of the generalized coordinates as the genuine knobs of the system — the pendulum's angle, the Atwood machine's one descent coordinate. d'Alembert's principle says: wiggle each knob virtually and demand that applied plus inertial forces do zero net work for that wiggle. Each knob yields one equation; $n$ knobs yield exactly $n$ equations of motion, with not one wasted line on the forces that merely enforce the geometry. Compare the Newtonian route through the same problems — the free-body craft of tensions and normals — and you will feel the compression immediately: the wedge problem below is two lines of physics instead of eight equations in six unknowns.

The principle, stated and earned

First the cast of characters, inherited from statics. Split every force on particle $i$ into applied and constraint parts, $\mathbf{F}_i = \mathbf{F}_i^{(a)} + \mathbf{R}_i$ , and let $\delta\mathbf{r}_i$ denote virtual displacements — infinitesimal repositionings consistent with the constraints at frozen time, $\delta t = 0$ . The constraints are ideal: their forces do no net virtual work,

\sum_i \mathbf{R}_i \cdot \delta \mathbf{r}_i = 0 \qquad (1)

where: $\mathbf{R}_i$ — the constraint (reaction) force on particle $i$ : normal reactions, string tensions, rigid-body internal forces, rolling-contact friction. This is the same load-bearing hypothesis as in statics — the previous article proves it case by case for the standard inventory — and it is a genuine physical postulate about which constraints we admit, not a theorem of Newton's laws. Sliding friction fails it and must be booked as applied. Now the principle itself:

\sum_i \left( \mathbf F_i^{(a)} - \dot{\mathbf p}_i \right)\cdot \delta\mathbf r_i = 0 \qquad (2)

where: $\dot{\mathbf{p}}_i = m_i\ddot{\mathbf{r}}_i$ — the rate of change of momentum, so $-\dot{\mathbf{p}}_i$ is the inertial force; the sum runs over all particles, and the statement holds for every admissible set of virtual displacements. Read it physically: the applied forces, minus what the motion itself "uses up", must have no component along any direction the constraints allow. At rest ( $\dot{\mathbf{p}}_i = \mathbf{0}$ ) it collapses to the static principle of virtual work; for a single free particle ( $\delta\mathbf{r}$ arbitrary) it collapses to Newton's second law. Everything between those poles — every constrained, moving system — is where it earns its keep.

Derivation AFrom Newton's laws to d'Alembert's principle and back

We are deriving equation (2) from Newton's second law plus the ideal-constraint postulate (1), and — the half that textbooks mumble — checking the converse, so that the principle is an honest equivalence and can serve as a foundation rather than a corollary. The forward strategy is three lines of algebra; the reverse direction is where the meaning lives: d'Alembert's principle is Newton's equation with exactly the constraint-force information projected out, no more and no less. Watch what is postulated versus what is derived: the workless-constraint condition is an input about the kinds of contacts we model, not a consequence of $\mathbf{F} = m\mathbf{a}$ .

Step 1. Write Newton's second law for each particle, with the force split into applied and constraint parts, and move everything to one side:

\mathbf F_i^{(a)} + \mathbf R_i - \dot{\mathbf p}_i = \mathbf{0}

Why this rearrangement is not empty: it converts a law of motion into a balance — each particle is in equilibrium under three "forces", one of them the inertial force $-\dot{\mathbf{p}}_i$ . Statics reasoning now applies verbatim, because statics never cared what the balanced vectors were made of.

Step 2. Dot each balance into the particle's virtual displacement $\delta\mathbf{r}_i$ and sum over the system:

\sum_i \big(\mathbf F_i^{(a)} + \mathbf R_i - \dot{\mathbf p}_i\big)\cdot\delta\mathbf r_i = 0

Why this is trivially true so far: each bracket is the zero vector, so its dot product with anything vanishes. No information has been used yet — the next step is where physics enters.

Step 3. Invoke the ideal-constraint postulate (1) to delete the reaction sum, leaving the principle:

\sum_i \big(\mathbf F_i^{(a)} - \dot{\mathbf p}_i\big)\cdot\delta\mathbf r_i = 0

Why the constraint forces are gone for good: not because each $\mathbf{R}_i\cdot\delta\mathbf{r}_i$ vanishes separately (a rod's two ends trade work), but because the total does — the geometry's enforcers form a closed economy that the probes cannot tax.

Step 4. Now the converse. Suppose (2) holds for all admissible probes, with ideal constraints. Define the residual $\mathbf{G}_i \equiv \mathbf{F}_i^{(a)} - \dot{\mathbf{p}}_i$ ; then $\sum_i \mathbf{G}_i\cdot\delta\mathbf{r}_i = 0$ says the residual, viewed as one big vector in $3N$ -dimensional configuration space, is orthogonal to every tangent direction of the constraint surface. A vector orthogonal to all tangents is a normal vector — and normal vectors are precisely what ideal constraint forces are. So there exist constraint forces $\mathbf{R}_i = -\mathbf{G}_i$ (workless by construction) with $\mathbf{F}_i^{(a)} + \mathbf{R}_i = \dot{\mathbf{p}}_i$ : Newton's laws, reassembled, with the reactions determined after the fact by the geometry. Why this settles the principle's status: it is exactly equivalent to Newton plus ideal constraints — no new physics, but a lossless repackaging in which the $n$ genuine equations of motion and the constraint-force bookkeeping are cleanly separated instead of entangled. *Verified: matches Goldstein §1.4 (eqs. 1.43–1.45) and Lanczos ch. IV; the converse argument is Lanczos's "the constraint force is the normal component" reading of the principle.*

The projection language in Step 4 deserves to be made completely concrete, because it is the version of the principle you can see.

Go deeper · Derivation B · 1 minThe projection picture on a single constraint surface

Derivation BThe projection picture on a single constraint surface

The goal is to show, for one particle confined to a smooth (possibly moving) surface, that d'Alembert's principle is literally the tangential projection of Newton's equation — and that the discarded normal component is a formula for the constraint force, recoverable on demand. The strategy: decompose Newton's vector equation in the local tangent–normal frame of the surface and match components against what (2) asserts. The subtlety to watch: the probes $\delta\mathbf{r}$ span only the frozen-time tangent plane, which is what makes the argument survive moving surfaces.

Step 1. Let the particle live on the smooth surface $g(\mathbf{r},t)=0$ , with unit normal $\hat{\mathbf{n}} = \nabla g/|\nabla g|$ . The reaction of a frictionless surface is purely normal, $\mathbf{R} = R\,\hat{\mathbf{n}}$ , with $R$ unknown. Newton reads

m\ddot{\mathbf r} = \mathbf F^{(a)} + R\,\hat{\mathbf n}

Step 2. Resolve into the frozen tangent plane and the normal. Any virtual displacement satisfies $\hat{\mathbf{n}}\cdot\delta\mathbf{r} = 0$ and, conversely, every frozen-tangent vector is an admissible probe — the probes span the tangent plane exactly. d'Alembert's principle $(\mathbf F^{(a)} - m\ddot{\mathbf r})\cdot\delta\mathbf r = 0$ for all such probes therefore says precisely:

\big(\mathbf F^{(a)} - m\ddot{\mathbf r}\big)_{\parallel} = \mathbf{0}

Why nothing about the normal direction is claimed: the probes cannot look there. The principle is silent about normal components by design — that silence is the elimination of the constraint force.

Step 3. Read off the two consequences. Tangentially, $m\ddot{\mathbf{r}}_{\parallel} = \mathbf{F}^{(a)}_{\parallel}$ — the genuine equation of motion, two scalar equations for a two-dimensional surface, exactly the number of degrees of freedom. Normally, Newton's remaining component defines the reaction:

R = m\,\ddot{\mathbf r}\cdot\hat{\mathbf n} - \mathbf F^{(a)}\cdot\hat{\mathbf n}

Why this is the "constraint force on demand" recipe: solve the tangential dynamics first, then substitute the known motion into the normal component. For a pendulum this yields the familiar $T = mg\cos\theta + mL\dot\theta^2$ — tension recovered after the fact, from a formalism that never needed it. The Lagrange-multiplier method automates exactly this recipe. *Verified: consistent with Taylor ch. 7 and Lanczos ch. IV; for the pendulum the recovered tension matches the Newtonian free-body result, including the $mL\dot\theta^2$ centripetal term.*

Into generalized coordinates: the gateway

Equation (2) is exact but still speaks particle language. Its destiny is to be rewritten in the system's own knobs. For a holonomic system with $n$ degrees of freedom, positions are a map $\mathbf{r}_i = \mathbf{r}_i(q_1,\dots,q_n,t)$ and the probes decompose as

\delta\mathbf r_i = \sum_{j=1}^{n} \frac{\partial \mathbf r_i}{\partial q_j}\,\delta q_j \qquad (3)

where: $\delta q_j$ — free, independent variations of the generalized coordinates; no $\partial/\partial t$ term appears because virtual displacements freeze time. The vectors $\partial\mathbf{r}_i/\partial q_j$ are the "influence coefficients" of knob $j$ — how particle $i$ moves per unit twist of that knob. Substituting (3) into the applied-force half of (2) sorts the work by knob and defines the generalized force:

Q_j = \sum_i \mathbf F_i^{(a)}\cdot \frac{\partial \mathbf r_i}{\partial q_j} \qquad (4)

where: $Q_j$ — the work the applied forces deliver per unit $\delta q_j$ : a torque if $q_j$ is an angle, a force if a length. This is the same object that ran statics; dynamics simply refuses to set it to zero.

Go deeper · Derivation C · 1 minOne substitution away from Lagrange

Derivation COne substitution away from Lagrange

We are performing the substitution that turns d'Alembert's principle into $n$ equations of motion in the generalized coordinates — and stopping deliberately one step short of the Euler–Lagrange equations, whose full derivation belongs to the next article. The strategy is bookkeeping: insert the probe decomposition (3) into (2), swap summation order, and use the independence of the $\delta q_j$ to split one scalar statement into $n$ . The step to watch is the last: independence of the probes is exactly where holonomicity pays — for nonholonomic systems the $\delta q_j$ are not all independent and this splitting needs multipliers.

Step 1. Insert (3) into the principle (2) and exchange the sums over particles and coordinates:

\sum_{j=1}^{n}\left[\; \underbrace{\sum_i \mathbf F_i^{(a)}\cdot\frac{\partial\mathbf r_i}{\partial q_j}}_{=\,Q_j} \;-\; \sum_i m_i\ddot{\mathbf r}_i\cdot\frac{\partial\mathbf r_i}{\partial q_j} \;\right]\delta q_j = 0

Why the regrouping is legitimate: both sums are finite; we are merely tallying the same virtual work knob by knob instead of particle by particle.

Step 2. Name the inertial projection $I_j \equiv \sum_i m_i\ddot{\mathbf{r}}_i\cdot(\partial\mathbf{r}_i/\partial q_j)$ — the "inertia felt by knob $j$ ". The principle now reads $\sum_j (Q_j - I_j)\,\delta q_j = 0$ for all admissible probes.

Step 3. Use independence. Because the constraints are holonomic and already absorbed into the coordinates, each $\delta q_j$ may be varied alone, with the others zero. Choosing the probes one at a time forces each coefficient to vanish separately:

I_j = Q_j \qquad (j = 1,\dots,n) \qquad (5)

where: this is already a complete, closed set of equations of motion — $n$ of them, one per degree of freedom, constraint forces absent. Why we stop here: the remaining work is a purely algebraic identity, $I_j = \frac{\mathrm{d}}{\mathrm{d}t}\big(\partial T/\partial\dot q_j\big) - \partial T/\partial q_j$ with $T$ the kinetic energy — the "cancellation of dots" lemma — which converts (5) into the Euler–Lagrange equations. That conversion, and everything the Lagrangian $L = T - V$ buys, is the next article's opening move. *Verified: matches Goldstein §1.4 (eqs. 1.45–1.53); applying (5) directly to the pendulum with $q = \theta$ gives $I_\theta = mL^2\ddot\theta$ , $Q_\theta = -mgL\sin\theta$ , hence $\ddot\theta = -(g/L)\sin\theta$ — the correct equation, no tension consulted.*

Which formulation answers which question

you want	use	notes
equations of motion, constraint-free	$\sum_i(\mathbf{F}_i^{(a)}-\dot{\mathbf{p}}_i)\cdot\delta\mathbf{r}_i=0$ , eq. (2)	one probe per degree of freedom
the same, in the system's knobs	$I_j = Q_j$ , eq. (5)	d'Alembert projected; pre-Lagrange form
a constraint force after all	normal projection (Derivation B) or Lagrange multipliers	solve motion first, substitute back
statics	drop $\dot{\mathbf{p}}_i$ : virtual work	the parent principle, recovered
friction or drag present	book them inside $Q_j$	no potential needed — see below
conservative forces, whole trajectories	Hamilton's principle / Euler–Lagrange	integral repackaging of this page

More general than the action principle

d'Alembert's principle is differential: a statement at each instant, comparing nothing to nothing, needing no knowledge of where the trajectory goes next. Hamilton's principle — stationary action — is integral: it compares whole candidate histories between fixed endpoints. For forces derivable from a potential the two are mathematically equivalent. But the differential form is strictly more accommodating, because a force needs no potential to enter equation (5) — it needs only a virtual-work tally:

Q_j = -\frac{\partial V}{\partial q_j} + Q_j^{(\mathrm{nc})} \qquad (6)

where: $Q_j^{(\mathrm{nc})}$ — the non-conservative residue: sliding friction, air drag $-b\dot{\mathbf{r}}$ , motor torques, feedback-control forces. Hamilton's action has no stationary formulation for such "non-monogenic" forces; d'Alembert simply adds their virtual work to the ledger and proceeds. A projectile with linear drag, a braking car, a robot joint under active control — all sit inside (5)–(6) without ceremony. This is why multibody dynamics engines and robotics simulators are built on the d'Alembert/projection formulation rather than on action principles: real machinery is full of forces that refuse to be potentials.

Limiting cases

Case	Result	Meaning
static equilibrium	$\dot{\mathbf p}_i = \mathbf{0} \Rightarrow \sum_i \mathbf F_i^{(a)}\cdot\delta\mathbf r_i = 0$	the principle of virtual work recovered — statics is dynamics with the inertial force asleep
single unconstrained particle	$\delta\mathbf r$ arbitrary $\Rightarrow \mathbf F = m\ddot{\mathbf r}$	Newton's second law recovered — with no constraints there is nothing to eliminate
conservative applied forces	$Q_j = -\partial V/\partial q_j \Rightarrow$ Euler–Lagrange with $L = T - V$	the gateway taken; the Lagrangian formalism begins one identity away
non-conservative forces	$I_j = -\partial V/\partial q_j + Q_j^{(\mathrm{nc})}$	drag and friction handled natively — beyond Hamilton's principle
rheonomic constraints	probes still freeze time $\Rightarrow$ reactions still workless	bead on a driven hoop: motor pumps real energy in, yet (2) holds instant by instant
nonholonomic velocity constraints	$\delta q_j$ no longer independent	Step 3 of the gateway fails; multiplier methods (Lagrange–d'Alembert) take over

Common trapThe inertial force is the same thing as a fictitious frame force

It is not, and the distinction protects you twice. The rotating-frame fictitious forces (centrifugal, Coriolis) arise from changing reference frames: a non-inertial observer must add them to make Newton work in their coordinates. d'Alembert's inertial force $-\dot{\mathbf{p}}$ involves no frame change at all — you remain in the inertial frame and simply move the momentum term across the equals sign, reading the equation of motion as a balance. The two can coincide numerically (stand in the braking bus: the frame force in bus coordinates equals the d'Alembert force of your ground-frame deceleration), which is exactly why they get conflated — but one is a coordinate transformation with physical consequences for measured accelerations, the other a bookkeeping convention inside a fixed frame. The second protection: d'Alembert's elimination works only under the ideal-constraint postulate. The moment a contact slides with friction, its force does virtual work, fails condition (1), and must be moved to the applied side of the ledger — forgetting this silently deletes real physics from your equations of motion.

Worked examples

1 · Atwood's machine without meeting the tension (confidence builder)

Masses $m_1 = 3.0\ \mathrm{kg}$ and $m_2 = 2.0\ \mathrm{kg}$ hang from an ideal string over a frictionless, massless pulley. Find the acceleration via d'Alembert's principle, then — for contrast — recover the tension the principle never needed.

Step 1. One knob. Let $q$ be the descent of $m_1$ ; inextensibility makes the rise of $m_2$ equal to $q$ , so $\ddot{x}_1 = \ddot{q}$ downward and $\ddot{x}_2 = \ddot{q}$ upward. The virtual probe: $\delta x_1 = \delta q$ down, $\delta x_2 = \delta q$ up. The string tension does zero net virtual work ( $+T\delta q - T\delta q$ ), and the pulley axle force acts on a point that never moves.

Step 2. Write d'Alembert's ledger (2), taking "down" as positive for each mass separately: mass 1 contributes $(m_1 g - m_1\ddot q)\,\delta q$ ; mass 2, whose descent is $-q$ , contributes $(m_2 g - m_2(-\ddot q))\,(-\delta q)$ . Sum and set to zero:

\big(m_1 g - m_1\ddot q\big)\,\delta q - \big(m_2 g + m_2\ddot q\big)\,\delta q = 0

Step 3. Solve: $\ddot q = \dfrac{(m_1 - m_2)\,g}{m_1 + m_2} = \dfrac{1.0 \times 9.81}{5.0} = \boxed{1.96\ \mathrm{m/s^2}}$ , with $m_1$ descending. Two lines, one unknown, no tension.

Step 4. The contrast. Newton's route writes two equations, $m_1 g - T = m_1 a$ and $T - m_2 g = m_2 a$ , and must carry $T$ through the algebra. If you want it (say, to size the string), the projection recipe hands it back: $T = m_1(g - a) = 3.0 \times (9.81 - 1.962) = \boxed{23.5\ \mathrm{N}}$ — note $T \neq (m_1{+}m_2)g/2$ ; the moving system carries less than the static average, $23.5\ \mathrm{N}$ against $24.5\ \mathrm{N}$ .

2 · Block on a free wedge — two knobs, zero constraint forces (exam level)

A block of mass $m = 2.0\ \mathrm{kg}$ slides on the frictionless face (angle $\alpha = 30°$ ) of a wedge of mass $M = 6.0\ \mathrm{kg}$ that itself slides on a frictionless floor. Find both accelerations. (The Newtonian route needs the normal forces on both surfaces; d'Alembert's needs neither.)

Step 1. Two knobs: $X$ = wedge position (positive right), $s$ = block's slide down the incline, measured relative to the wedge. The block's position is $\mathbf{r}_b = (X + s\cos\alpha,\; h - s\sin\alpha)$ , so its acceleration is $(\ddot X + \ddot s\cos\alpha,\; -\ddot s\sin\alpha)$ ; the wedge's is $(\ddot X, 0)$ . Influence vectors: $\partial\mathbf{r}_b/\partial X = (1,0)$ , $\partial\mathbf{r}_b/\partial s = (\cos\alpha, -\sin\alpha)$ , $\partial\mathbf{r}_w/\partial X = (1,0)$ , $\partial\mathbf{r}_w/\partial s = (0,0)$ .

Step 2. The $\delta X$ probe (slide everything sideways). Applied forces are the two weights, both vertical, so $Q_X = 0$ . The inertial projection: $I_X = m(\ddot X + \ddot s\cos\alpha) + M\ddot X$ . Equation (5):

(M+m)\,\ddot X + m\,\ddot s\cos\alpha = 0

This is horizontal momentum conservation, delivered automatically — the probe direction is a symmetry of the applied forces.

Step 3. The $\delta s$ probe (slide the block along the frozen wedge). Gravity's tally: $Q_s = (0,-mg)\cdot(\cos\alpha,-\sin\alpha) = mg\sin\alpha$ . Inertial projection: $I_s = m\big[(\ddot X + \ddot s\cos\alpha)\cos\alpha + (-\ddot s\sin\alpha)(-\sin\alpha)\big] = m(\ddot X\cos\alpha + \ddot s)$ . Equation (5):

\ddot s + \ddot X\cos\alpha = g\sin\alpha

Step 4. Solve the pair. Substituting $\ddot X = -\frac{m\cos\alpha}{M+m}\ddot s$ :

\ddot s = \frac{(M+m)\,g\sin\alpha}{M + m\sin^2\alpha} = \frac{8.0 \times 9.81 \times 0.5}{6.0 + 2.0 \times 0.25} = \frac{39.24}{6.5} = \boxed{6.04\ \mathrm{m/s^2}},

\ddot X = -\frac{m\,g\sin\alpha\cos\alpha}{M + m\sin^2\alpha} = -\frac{2.0 \times 9.81 \times 0.5 \times 0.866}{6.5} = \boxed{-1.31\ \mathrm{m/s^2}}

— the wedge recoils left as the block slides down-right.

Step 5. Audits. Fix the wedge ( $M \to \infty$ ): $\ddot s \to g\sin\alpha = 4.91\ \mathrm{m/s^2}$ — the freshman result, correctly smaller than our $6.04$ (a recoiling wedge steepens the block's fall). Horizontal momentum: block's horizontal acceleration is $\ddot X + \ddot s\cos\alpha = -1.307 + 6.037(0.866) = +3.92\ \mathrm{m/s^2}$ , and $m(3.92) + M(-1.307) = 7.84 - 7.84 = 0$ ✓. Neither normal force was ever named; the Newtonian solution of the same problem juggles both.

Origins · 1 minJean le Rond d'Alembert and how this was discovered

Jean le Rond d'Alembert

1717–1783 · Paris, Académie des sciences and the Encyclopédie

d'Alembert was found as a newborn on the steps of the Saint-Jean-le-Rond chapel in Paris — the abandoned, illegitimate son of a salon hostess and an artillery officer — and took his name from the church. Raised by a glazier's wife whom he honored as his mother all his life, he was elected to the Académie des sciences at 23 and published the Traité de dynamique at 25, its famous principle reducing dynamics to equilibrium and, in his own program, reducing mechanics to geometry — he distrusted "force" as a metaphysical notion and wanted motion built from kinematic first principles. The decade that followed was a supernova: the wave equation for the vibrating string (1747, the first PDE ever solved), the ratio test for series, the d'Alembert paradox of fluid mechanics, and an early, flawed-but-first proof attempt of the fundamental theorem of algebra. From 1751 he co-edited the Encyclopédie with Diderot, writing its mathematical articles and its manifesto-like Preliminary Discourse — making him simultaneously a founder of analytical mechanics and of the French Enlightenment's greatest publishing project. Famously combative (lifelong feuds with Clairaut and later Euler over priority), he mentored the young Lagrange, arranged his Berlin appointment, and lived to see his instant-by-instant principle become, in Lagrange's hands, the foundation of the Mécanique analytique.

History

1687

— Newton's Principia states the laws of motion; constrained systems must be handled force by force, reaction by reaction.

1717

— Johann Bernoulli's letter to Varignon states the principle of virtual velocities: statics without constraint forces.

1743

— d'Alembert's Traité de dynamique adjoins the inertial force to the ledger, converting every dynamics problem into an equilibrium problem.

1788

— Lagrange's Mécanique analytique fuses Bernoulli's and d'Alembert's principles and derives the equations now bearing his name — analytical mechanics is born.

1834

— Hamilton recasts the differential principle as a statement about whole trajectories: the principle of stationary action, exactly equivalent for potential forces and narrower for the rest.

Going further

References. Goldstein, Poole & Safko, Classical Mechanics, 3rd ed., §1.4 · Landau & Lifshitz, Mechanics, 3rd ed., §1–2 · Morin, Introduction to Classical Mechanics, Ch. 6 · Taylor, Classical Mechanics, Ch. 7 · Lanczos, The Variational Principles of Mechanics, Ch. IV

Provenance. Equations cross-checked against Goldstein §1.4 (d'Alembert's principle and Lagrange's equations, Eqs. 1.45–1.61), Landau–Lifshitz Mechanics §1–2, Taylor Ch. 7, and Lanczos Ch. IV. A verifier should recompute: (1) the inertial-term identity using the two 'dot-cancellation' lemmas; (2) the Atwood result $\ddot q=(m_1-m_2)g/(m_1+m_2)$ ; (3) the rotating-hoop equation $mR\ddot\theta=mR\omega^2\sin\theta\cos\theta-mg\sin\theta$ . The historical attribution (d'Alembert, 1743, Traité de dynamique) is standard. No physical constants are quoted; all results are symbolic and textbook-standard. Worked-example numbers on this page ( $\ddot q = 1.96\ \mathrm{m/s^2}$ and $T = 23.5\ \mathrm{N}$ for Atwood; $\ddot s = 6.04\ \mathrm{m/s^2}$ , $\ddot X = -1.31\ \mathrm{m/s^2}$ for the wedge) are illustrative arithmetic recomputed from the boxed formulas.

Builds on: Virtual Displacements and the Principle of Virtual Work · Constraints and Their Classification · Generalized Coordinates and Configuration Space · Degrees of Freedom. Leads to: The Euler-Lagrange Equations · The Lagrangian Function · Generalized Forces and Generalized Momenta · Hamilton's Principle of Stationary Action.

Is the inertial force −ma a real force or not?

Not in the inertial frame where d'Alembert works — no agent exerts it, it has no third-law partner, and a force-plate under the accelerating body would not register it. It is a term of Newton's equation relocated across the equals sign and read as if it were a force, because the reading unlocks statics machinery. This is different from the centrifugal and Coriolis forces of a rotating frame: those are corrections a non-inertial observer must add for Newton's law to hold in their coordinates, and they have measurable consequences in that frame (your accelerometer literally reads them). d'Alembert never changes frames. The numerical coincidence in simple cases — braking-bus lurch computed either way — is what fuels the confusion the warning box flags.

If d'Alembert's principle is just Newton rearranged, why does it deserve the name "principle"?

Two honest reasons. First, it is Newton plus a genuine physical postulate: that constraint forces do no net virtual work. That postulate selects the class of contacts we model (smooth, rigid, inextensible, rolling) and is falsifiable — sliding friction violates it. Second, its content is organizational in a way that changes what is computable: it delivers exactly $n$ equations for $n$ degrees of freedom with the reactions pre-eliminated, where raw Newton delivers $3N$ equations tangled with unknown reactions. Historically it also carried a philosophical program — d'Alembert wanted mechanics grounded without "forces" as primitive metaphysics. Lagrange treated it as the axiom of mechanics and derived everything else from it, which is the strongest possible endorsement of the name.

The principle eliminated the tension — but I actually need the tension to size my cable. Am I stuck?

No — elimination is reversible on demand, and the projection picture shows how. d'Alembert splits Newton's equation into a tangential part (the dynamics, which you solve first) and a normal part (which defines the constraint force once the motion is known). Practically: solve $I_j = Q_j$ for the motion, then substitute the now-known accelerations back into Newton's equation for the body of interest and read off the reaction. For the Atwood machine: $a = 1.96\ \mathrm{m/s^2}$ first, then $T = m_1(g-a) = 23.5\ \mathrm{N}$ . The systematic version of this back-substitution is the method of Lagrange multipliers, where the multiplier attached to a constraint is its force. You pay for a reaction only when you ask for it.

How can the principle hold on a motor-driven hoop where the constraint is clearly pumping energy into the bead?

Because the principle makes no energy claim — it is an instant-by-instant statement about virtual work, and virtual displacements freeze the hoop. The motor does its real work through the $(\partial\mathbf{r}/\partial t)\,\mathrm{d}t$ part of the real displacement, a term that virtual probes amputate by definition. So both facts coexist: the bead's energy grows (real work by the normal force, nonzero), and equation (2) holds at every instant (virtual work of the same force, zero), correctly generating the equation of motion $mR\ddot\theta = mR\omega^2\sin\theta\cos\theta - mg\sin\theta$ . Energy accounting for such rheonomic systems is genuinely subtle — the conserved quantity, when one exists, is the Jacobi integral rather than $T+V$ — and that story belongs to the energy-function article downstream.

Where exactly does sliding friction break the machinery, and what is the standard repair?

It breaks the ideal-constraint postulate, equation (1) — nothing else. Kinetic friction acts along the contact surface, i.e. along admissible virtual displacements, so its virtual work is nonzero (and negative) and it cannot be left on the constraint side of the ledger. The repair is reclassification: keep the normal component of the contact as an ideal constraint (still workless), move the frictional component into the applied forces, model it constitutively ( $|\mathbf{f}| = \mu N$ , direction opposing slip), and let it enter through the generalized force $Q_j^{(\mathrm{nc})}$ . The equations stay valid; you simply carry one more applied force — and note the wrinkle that $N$ itself may need the multiplier method, since friction's magnitude references a constraint force the formalism normally hides.

What really goes wrong for nonholonomic constraints like a rolling sphere?

The gateway derivation's last step. Rolling-without-slipping on a surface constrains velocities in a way that cannot be integrated into a constraint on positions, so you cannot choose independent coordinates that absorb it — the system explores more configurations than its velocity freedoms suggest. The probes $\delta q_j$ then satisfy linear side-conditions and are not independently variable, so "each coefficient vanishes separately" fails. d'Alembert's principle itself, equation (2), survives untouched — the rolling contact is still workless — but extracting equations of motion requires either Lagrange multipliers or the Lagrange–d'Alembert equations built for this case. That machinery, and the classic rolling-disk examples, get their own article; the classification of which constraints integrate and which do not is the constraints article's central theme.

Why is being a "differential" principle an advantage over Hamilton's "integral" principle — isn't the action more fundamental?

They are different tools, and fundamentality depends on the job. The action principle's strengths are structural: symmetries, conservation laws, quantization all speak its language. But it buys that elegance with a hypothesis — forces must be monogenic (derivable from a potential, possibly velocity-dependent). d'Alembert's differential form needs only a virtual-work tally per instant, so friction, drag, contact impulses, and control torques enter natively; that is why engineering dynamics and robotics simulators are d'Alembertian, not Hamiltonian. There is also no lookahead: the differential principle never compares trajectories, so it composes naturally with time-stepping numerics. The mature view: d'Alembert is the general-purpose foundation; Hamilton's principle is the beautiful special case you use whenever the forces allow it.

Why does kinetic energy suddenly show up when the inertial term is rewritten?

Because the projection $\sum_i m_i\ddot{\mathbf{r}}_i\cdot(\partial\mathbf{r}_i/\partial q_j)$ is, by an exact algebraic identity, expressible as $\frac{\mathrm{d}}{\mathrm{d}t}(\partial T/\partial\dot q_j) - \partial T/\partial q_j$ with $T = \tfrac12\sum m_i\dot{\mathbf{r}}_i^2$ written in the $q$ 's. The identity rests on two lemmas — the "cancellation of dots" $\partial\dot{\mathbf{r}}_i/\partial\dot q_j = \partial\mathbf{r}_i/\partial q_j$ and the interchange of $\mathrm{d}/\mathrm{d}t$ with $\partial/\partial q_j$ — and its meaning is that $T(q,\dot q)$ encodes the entire inertia of the system as seen through the coordinates, including every centripetal- and Coriolis-like coupling that curvilinear knobs generate. That one function replaces all the mass-times-acceleration bookkeeping is the miracle that makes the Lagrangian formalism possible; the next article opens by proving the identity in full.