Friction, Normal Forces, and Constraints: Holonomic and Non-holonomic Systems

Intuition Beginner

Push a book across a table. It slows down. Something opposes its motion. That something is friction — a contact force between two surfaces, always directed opposite to the relative motion or the tendency toward motion.

There are two regimes. Static friction holds the book still when you push gently. It adjusts itself up to a maximum value, proportional to how hard the book presses against the table. Push hard enough and the book starts to slide. Kinetic friction takes over once the book is moving. Its magnitude is fixed (again proportional to the pressing force) and is generally slightly less than the maximum static friction ^{[source pending]}.

The "pressing force" is the normal force — the force the table exerts upward on the book, perpendicular to the surface. The name comes from the mathematical word for perpendicular. If the book weighs 10 newtons and sits on a flat table, the normal force is 10 newtons upward, exactly balancing gravity. On a tilted surface the normal force is smaller: only the component of gravity perpendicular to the surface.

Normal forces are constraint forces. They enforce a geometric restriction: the book cannot pass through the table. The table pushes back just hard enough to prevent penetration. The book's motion is constrained to remain on or above the table surface.

Constraints appear everywhere in mechanics. A bead threaded on a wire is constrained to follow the wire's shape. A pendulum bob is constrained to remain at a fixed distance from the pivot. A door on hinges is constrained to rotate about the hinge axis. In every case, the constraint generates a force — normal force, tension, hinge reaction — that maintains the geometric restriction ^{[source pending]}.

Consider a concrete example. Place a wooden block on a wooden ramp tilted at angle $\theta$. Three forces act: gravity straight down, the normal force perpendicular to the ramp surface, and friction along the ramp. If the block slides at constant speed, these three forces balance. The normal force equals the perpendicular component of gravity: $N=mg\cos \theta$. Friction equals the parallel component: $f_k=mg\sin \theta$. Increase the angle and eventually the block accelerates — the downhill pull of gravity exceeds the maximum friction the surface can provide.

The angle at which sliding just begins is the angle of repose. At this critical angle, the downhill component $mg\sin \theta$ exactly equals the maximum static friction ${\mu }_{s}mg\cos \theta$, giving $\tan \theta ={\mu }_s$. Measuring the angle of repose is one of the simplest ways to determine the coefficient of static friction ^{[source pending]}.

Friction dissipates mechanical energy as heat. A block sliding to a stop on a flat surface loses kinetic energy. The work done by friction is $W_{\mathrm{friction}}=-f_{k}d$ where $d$ is the distance traveled. This work does not return — it is not stored in any potential. The total mechanical energy $E=T+V$ decreases by exactly $f_{k}d$, converted to thermal energy in the two surfaces ^{[source pending]}.

Visual Beginner

Worked example Beginner

A block of mass $m=2$ kg sits on a ramp inclined at 30 degrees. The coefficient of static friction is ${\mu }_s=0.4$ and the coefficient of kinetic friction is ${\mu }_k=0.3$. Does the block slide? If so, what is its acceleration?

Step 1: Compare static friction with the downhill pull.

The downhill component of gravity is $mg\sin (30^{\circ })=2\times 9.8\times 0.5=9.8$ N. The normal force is $N=mg\cos (30^{\circ })=2\times 9.8\times 0.866=17.0$ N. The maximum static friction is ${\mu }_{s}N=0.4\times 17.0=6.8$ N.

Since 9.8 N exceeds 6.8 N, the block slides.

Step 2: Find the acceleration.

Once sliding, kinetic friction applies: $f_k={\mu }_{k}N=0.3\times 17.0=5.1$ N uphill. The net downhill force is $9.8-5.1=4.7$ N. By Newton's second law: $a=F/m=4.7/2=2.35$ m/s${}^2$ downhill.

Step 3: Energy check.

After the block slides 1 m downhill, it loses gravitational potential energy $\Delta V=-mg\sin \theta \cdot d=-9.8$ J and gains kinetic energy $\Delta T=\frac{1}{2}m{v}^2$. The work done by friction is $-f_{k}d=-5.1$ J. Energy balance: $\Delta T+\Delta V=-f_{k}d$, so $\Delta T=9.8-5.1=4.7$ J, consistent with $v^2=2ad=4.7$ giving $\Delta T=\frac{1}{2}(2)(4.7)=4.7$ J ^{[source pending]}.

Check your understanding Beginner

Formal definition Intermediate+

Coulomb friction law

Static friction. For two dry surfaces in contact with a normal force $N$, the static friction force satisfies the inequality

$$|f_s|\le {\mu }_{s}N,$$

where ${\mu }_s$ is the coefficient of static friction, a dimensionless number determined by the materials and surface conditions. The friction force takes whatever value and direction is needed to prevent relative motion, up to the maximum ${\mu }_{s}N$. Motion initiates when the net applied force along the contact surface exceeds ${\mu }_{s}N$.

Kinetic friction. Once the surfaces are sliding relative to each other,

$$f_k={\mu }_{k}N,$$

directed opposite to the relative velocity, where ${\mu }_k$ is the coefficient of kinetic friction. Typically ${\mu }_k<{\mu }_s$. The kinetic friction force is independent of the contact area and (to a good approximation) independent of the sliding speed ^{[source pending]}.

Angle of repose. For a block on an inclined plane, the critical angle at which sliding begins satisfies

$$\tan {\theta }_{\mathrm{repose}}={\mu }_s.$$

This follows from balancing the downhill component $mg\sin \theta$ against maximum static friction ${\mu }_{s}mg\cos \theta$.

The Coulomb model is a macroscopic, phenomenological law. It does not follow from any fundamental force law. The microscopic physics involves interatomic forces at asperity contacts, surface contamination, and deformation — topics in tribology, the science of friction, wear, and lubrication ^{[source pending]}. The Coulomb model works well for dry (unlubricated) surfaces and is the standard engineering approximation.

Normal force and constraint forces

The normal force $\mathbf{N}$ is a reaction force perpendicular to a surface that prevents interpenetration. More generally, constraint forces are forces that enforce geometric restrictions on the motion. They are determined not by a force law (like gravity or electromagnetism) but by the requirement that the constraint be satisfied at all times.

For a particle constrained to remain on a surface $f(\mathbf{r})=0$, the constraint force is

$${\mathbf{F}}^{(c)}=\lambda \nabla f,$$

where $\lambda$ is a Lagrange multiplier determined by the dynamics. The constraint force is always normal to the constraint surface — perpendicular to the surface at the point of contact ^{[source pending]}.

Holonomic constraints

A constraint is holonomic if it can be expressed as an equation relating only the coordinates and possibly time, but not the velocities or their higher derivatives:

$$\boxed{f(q^1,q^2,\dots ,q^n,t)=0.}$$

Each holonomic constraint reduces the number of independent coordinates by one. A system of $N$ particles in three dimensions has $3N$ coordinates; $k$ independent holonomic constraints leave $3N-k$ degrees of freedom.

Examples. A particle constrained to a surface: $f(\mathbf{r})=0$. A rigid rod of length $\ell$ connecting two particles: $|{\mathbf{r}}_1-{\mathbf{r}}_2{|}^2-{\ell }^2=0$. A bead constrained to a wire: two constraint equations reducing three coordinates to one. A simple pendulum: $x^2+y^2-{\ell }^2=0$, reducing two coordinates to one angle $\theta$.

Holonomic constraints are classified by their time dependence:

• Scleronomic (time-independent): $f(q)=0$. A bead on a fixed wire, a pendulum with fixed length.
• Rheonomic (time-dependent): $f(q,t)=0$. A bead on a moving wire, a pendulum whose length changes in time.

They are also classified by their inequality structure:

• Bilateral: $f(q,t)=0$ (the constraint holds as an equality). A bead on a wire can move in either direction.
• Unilateral: $f(q,t)\ge 0$ (the constraint is an inequality). A block on a table: $y\ge 0$ (the block can lift off but cannot penetrate). Unilateral constraints require separate treatment because the constraint may become active or inactive during the motion ^{[source pending]}.

Non-holonomic constraints

A constraint that cannot be integrated to the form $f(q,t)=0$ is non-holonomic. The standard form is

$$g(q^1,\dots ,q^n,{\dot{q}}^1,\dots ,{\dot{q}}^n,t)=0,$$

where $g$ depends on velocities in an essential way — no integration can eliminate the velocity dependence.

The canonical example is a disk rolling without slipping on a plane. Let $(x,y)$ be the position of the contact point, $\varphi$ the rotation angle of the disk, and $\theta$ the heading angle. The rolling condition is

$$dx-R\cos \theta d\varphi =0,dy-R\sin \theta d\varphi =0.$$

These relate differentials of position to the rotation angle and heading, but cannot be integrated to relations among coordinates alone. The disk can reach any position and orientation on the plane — the constraint restricts velocities without reducing the accessible configuration space ^{[source pending]}.

Non-holonomic constraints do not reduce the number of degrees of freedom. They restrict velocities but not configurations. This is the fundamental distinction from holonomic constraints, which restrict both.

Virtual displacement and virtual work

A virtual displacement $\delta {\mathbf{r}}_i$ is an infinitesimal change in the coordinates consistent with the constraints at a fixed instant in time (a "frozen-time" displacement). For a holonomic constraint $f(q,t)=0$, virtual displacements satisfy

$$\sum _{i=1}^N\frac{\partial f}{\partial {\mathbf{r}}_i}\cdot \delta {\mathbf{r}}_i=0.$$

The word "virtual" distinguishes these from real displacements: a real displacement $d{\mathbf{r}}_i$ occurs over a time interval $dt$ and may violate a time-dependent constraint's frozen-time condition.

The virtual work of a force ${\mathbf{F}}_i$ on the $i$-th particle is

$$\delta W=\sum _{i=1}^N{\mathbf{F}}_i\cdot \delta {\mathbf{r}}_i.$$

Constraint forces (normal forces, tensions, smooth-wall reactions) do zero virtual work for ideal constraints: ${\sum }_i{\mathbf{F}}_i^{(c)}\cdot \delta {\mathbf{r}}_i=0$. This is the principle of virtual work: constraint forces are perpendicular to the constraint surface and therefore perpendicular to virtual displacements along it ^{[source pending]}.

Friction is the exception. Kinetic friction does negative virtual work — it opposes motion and dissipates energy. This is why friction is classified as a non-ideal constraint force.

D'Alembert's principle

Decompose the total force on particle $i$ as ${\mathbf{F}}_i={\mathbf{F}}_i^{(a)}+{\mathbf{F}}_i^{(c)}$, where ${\mathbf{F}}_i^{(a)}$ are applied forces (gravity, springs, external pushes) and ${\mathbf{F}}_i^{(c)}$ are constraint forces. D'Alembert's principle states:

$$\boxed{\sum _{i=1}^N\left({\mathbf{F}}_i^{(a)}-m_i{\mathbf{a}}_i\right)\cdot \delta {\mathbf{r}}_i=0}$$

for all virtual displacements $\delta {\mathbf{r}}_i$ consistent with the constraints. The underlying assumption is that constraint forces do zero virtual work: ${\sum }_i{\mathbf{F}}_i^{(c)}\cdot \delta {\mathbf{r}}_i=0$ ^{[source pending]}.

D'Alembert's principle is the bridge from Newtonian mechanics to the Lagrangian formulation. It eliminates constraint forces from the equations of motion by projecting onto the unconstrained directions. This is the physical content that the Euler-Lagrange equations 09.02.02 encode through the choice of generalised coordinates adapted to the constraints.

Classification summary

Constraint type	Equation form	Examples	DOF effect
Holonomic scleronomic	$f(q)=0$	Bead on fixed wire, pendulum	Reduced by 1 per constraint
Holonomic rheonomic	$f(q,t)=0$	Bead on moving wire	Reduced by 1 per constraint
Non-holonomic	$g(q,\dot{q},t)=0$ (non-integrable)	Rolling disk, skate	Not reduced
Unilateral	$f(q,t)\ge 0$	Block on table	Conditional

Key derivation Intermediate+

Key result (Constraint forces via Lagrange multipliers). For a system of $N$ particles subject to holonomic constraints $f_{\alpha }(q^1,\dots ,q^{3N},t)=0$ for $\alpha =1,\dots ,k$, the constraint forces are

$$\boxed{{\mathbf{F}}_i^{(c)}=\sum _{\alpha =1}^k{\lambda }_{\alpha }\frac{\partial f_{\alpha }}{\partial {\mathbf{r}}_i},}$$

where the Lagrange multipliers ${\lambda }_{\alpha }$ are determined by requiring the constraints to be satisfied.

Derivation. The constraint $f_{\alpha }=0$ implies $\delta f_{\alpha }=0$ for virtual displacements: ${\sum }_i(\partial f_{\alpha }/\partial {\mathbf{r}}_i)\cdot \delta {\mathbf{r}}_i=0$. These are $k$ linear constraints on the $\delta {\mathbf{r}}_i$. Introduce $k$ multipliers ${\lambda }_{\alpha }$ and add ${\sum }_{\alpha }{\lambda }_{\alpha }\delta f_{\alpha }=0$ to D'Alembert's principle:

$$\sum _{i=1}^N\left({\mathbf{F}}_i^{(a)}-m_i{\mathbf{a}}_i+\sum _{\alpha =1}^k{\lambda }_{\alpha }\frac{\partial f_{\alpha }}{\partial {\mathbf{r}}_i}\right)\cdot \delta {\mathbf{r}}_i=0.$$

Choose the ${\lambda }_{\alpha }$ so that the coefficient of each constrained $\delta {\mathbf{r}}_i$ component vanishes. The remaining $3N-k$ free components are independently variable, so their coefficients must also vanish. This yields $3N$ equations of motion with the constraint forces identified ^{[source pending]}.

Worked example: the Atwood machine

Two blocks of masses $m_1$ and $m_2$ ($m_1>m_2$) are connected by a light inextensible string over a frictionless pulley. The string constraint is $y_1+y_2=\text{const}$, or equivalently ${\dot{y}}_1=-{\dot{y}}_2$. The constraint force is the string tension $T$.

Equations of motion:

$$m_1{\ddot{y}}_1=m_{1}g-T,m_2{\ddot{y}}_2=m_{2}g-T.$$

Using the constraint ${\ddot{y}}_1=-{\ddot{y}}_2$:

$$m_1{\ddot{y}}_1=m_{1}g-T,-m_2{\ddot{y}}_1=m_{2}g-T.$$

Subtracting: $(m_1+m_2){\ddot{y}}_1=(m_1-m_2)g$, giving

$${\ddot{y}}_1=\frac{(m_1-m_2)g}{m_1+m_2},T=\frac{2m_1{m}_{2}g}{m_1+m_2}.$$

The constraint (inextensible string) has eliminated one degree of freedom, reducing two coordinates to one ^{[source pending]}.

Worked example: block on an accelerating surface

A block of mass $m$ sits on a flat-topped cart that accelerates horizontally with acceleration $a$. The coefficient of static friction between block and cart is ${\mu }_s$. The block does not slip. What is the friction force, and what is the maximum $a$ before slipping?

In the inertial frame, the only horizontal force on the block is static friction $f_s$, which must accelerate the block at the same rate as the cart: $f_s=ma$. The constraint "no slipping" determines the friction force.

The maximum friction is ${\mu }_{s}N={\mu }_{s}mg$ (the normal force equals the weight because there is no vertical acceleration). The block slips when $ma>{\mu }_{s}mg$, i.e., when $a>{\mu }_{s}g$. For ${\mu }_s=0.5$, the critical acceleration is $a=4.9$ m/s${}^2$ ^{[source pending]}.

Work-energy theorem with friction

For a particle moving under conservative forces plus kinetic friction, the work-energy theorem becomes

$$\Delta T+\Delta V=-f_{k}d,$$

where $d$ is the distance traveled along the friction surface. The mechanical energy $E=T+V$ is not conserved — it decreases by $f_{k}d$, which is the energy dissipated as heat. This is the mechanical statement of the first law of thermodynamics for a frictional process: the energy lost from the mechanical sector appears as thermal energy in the materials ^{[source pending]}.

The rate of energy dissipation is

$$\frac{dE}{dt}=-f_k|v|=-{\mu }_{k}N|v|,$$

where $v$ is the speed. This power dissipation is always negative — friction always removes mechanical energy.

Bridge. D'Alembert's principle builds toward the Lagrangian formulation 09.02.03, where constraint forces are handled via Lagrange multipliers or eliminated entirely by choosing generalised coordinates adapted to the constraints; the foundational reason is that constraint forces do zero virtual work, which is exactly the property that makes the Euler-Lagrange equations independent of the constraint forces when the right coordinates are chosen. The energy-dissipation result above connects to the thermodynamic treatment of irreversible processes, where friction is the prototypical dissipative mechanism converting mechanical work to heat. This result generalises to field-theoretic constraints in continuum mechanics 09.07.01, and the central insight — that constraints reduce the configuration space to a submanifold — is dual to the geometric treatment of constrained motion on manifolds in the Hamiltonian framework.

Exercises Intermediate+

Lean formalization Intermediate+

The Coulomb friction model — with its discontinuous dependence on velocity sign and its inequality constraints ($|f_s|\le {\mu }_{s}N$) — is not represented in Mathlib. The classification of constraints (holonomic, non-holonomic, scleronomic, rheonomic) would require building a type-class hierarchy on top of Mathlib's smooth-manifold machinery. D'Alembert's principle, as a statement about virtual work vanishing, requires integrating the tangent-space geometry of constraint submanifolds with force-balance equations, which is a meaningful formalisation project but has not been undertaken.

Advanced results Master

From D'Alembert to the Euler-Lagrange equations

D'Alembert's principle is not merely a computational tool — it is the conceptual bridge from Newtonian mechanics to the Lagrangian formulation. When the constraints are holonomic, the virtual displacements can be parameterised by independent generalised coordinates $q^1,\dots ,q^n$ (where $n=3N-k$ is the number of degrees of freedom). Expressing the positions as ${\mathbf{r}}_i={\mathbf{r}}_i(q^1,\dots ,q^n,t)$ and substituting into D'Alembert's principle:

$$\sum _{j=1}^n\left[\sum _{i=1}^N\left({\mathbf{F}}_i^{(a)}\cdot \frac{\partial {\mathbf{r}}_i}{\partial q^j}\right)-\frac{d}{dt}\frac{\partial T}{\partial {\dot{q}}^j}+\frac{\partial T}{\partial q^j}\right]\delta q^j=0.$$

Since the $\delta q^j$ are independent, each coefficient must vanish. Defining the generalised force $Q_j={\sum }_i{\mathbf{F}}_i^{(a)}\cdot \partial {\mathbf{r}}_i/\partial q^j$ and the kinetic energy $T$:

$$\frac{d}{dt}\frac{\partial T}{\partial {\dot{q}}^j}-\frac{\partial T}{\partial q^j}=Q_j.$$

If the applied forces are conservative (${\mathbf{F}}_i^{(a)}=-{\nabla }_{i}V$), then $Q_j=-\partial V/\partial q^j$ and the equations become the Euler-Lagrange equations 09.02.02:

$$\frac{d}{dt}\frac{\partial L}{\partial {\dot{q}}^j}-\frac{\partial L}{\partial q^j}=0,$$

where $L=T-V$ is the Lagrangian. The entire derivation rests on D'Alembert's principle and the workless nature of constraint forces ^{[source pending]}.

Non-holonomic constraints and the Lagrange-d'Alembert equations

For non-holonomic constraints written as $A_{\alpha j}{\dot{q}}^j=0$ ($\alpha =1,\dots ,m$), the Euler-Lagrange equations are modified to the Lagrange-d'Alembert equations:

$$\frac{d}{dt}\frac{\partial L}{\partial {\dot{q}}^j}-\frac{\partial L}{\partial q^j}=\sum _{\alpha =1}^m{\mu }_{\alpha }{A}_{\alpha j},$$

where the ${\mu }_{\alpha }$ are Lagrange multipliers for the non-holonomic constraints. Unlike the holonomic case, these multipliers cannot be absorbed into a redefined Lagrangian because the constraints do not reduce the configuration space ^{[source pending]}.

The geometry of constraint forces

On a configuration manifold $Q$, a holonomic constraint $f(q)=0$ defines a submanifold $S=f^{-1}(0)\subset Q$. The constraint force at each point is proportional to $\nabla f$, lying in the normal bundle of $S$. Virtual displacements lie in the tangent bundle $TS$. The vanishing of virtual work is the orthogonality ${\mathbf{F}}^{(c)}\perp T_{q}S$ at each point ^{[source pending]}.

For non-holonomic constraints, the constraint defines a distribution $D\subset TQ$ (a sub-bundle of the tangent bundle). The constraint forces annihilate $D$, and the dynamics is restricted to $D$ without reducing the configuration space. The Frobenius theorem states that $D$ is integrable (i.e., comes from a submanifold) if and only if it is closed under the Lie bracket — this is the geometric content of the holonomic/non-holonomic distinction ^{[source pending]}.

Thermodynamic aspects of friction

Friction is the prototypical dissipative mechanism. At the macroscopic level, the work done by friction converts mechanical energy into thermal energy. The rate of dissipation for a sliding block is

$$P_{\mathrm{diss}}={\mu }_{k}N|v|.$$

At the microscopic level, this energy goes into exciting lattice vibrations (phonons) in both surfaces, increasing their internal energy. The temperature rise at the sliding interface can be substantial: in brake systems, surface temperatures can reach several hundred degrees Celsius ^{[source pending]}.

The irreversibility of friction is fundamental. While the Coulomb friction law $f_k={\mu }_{k}N$ is time-reversal symmetric as a mathematical equation, the physical process is not: friction always converts organised kinetic energy into disordered thermal energy, never the reverse. This asymmetry is the mechanical manifestation of the second law of thermodynamics — the entropy of the universe increases by $\Delta S=f_{k}d/T_{\mathrm{interface}}$ for each increment of sliding distance $d$ ^{[source pending]}.

Tribology: the microscopic physics of friction

The Coulomb model treats friction coefficients as empirical constants. Modern tribology explains their origin. Real surfaces are not flat at the microscopic scale — they contact only at asperity peaks. The real contact area $A_{\mathrm{real}}$ is typically orders of magnitude smaller than the apparent contact area, and is approximately proportional to the normal load:

$$A_{\mathrm{real}}\approx \frac{N}{H},$$

where $H$ is the hardness of the softer material. The friction force is then proportional to the real contact area: $f\approx \tau \cdot A_{\mathrm{real}}$ where $\tau$ is the shear strength of the junction. This explains why Coulomb friction is proportional to normal force and approximately independent of apparent contact area — the Amontons-Coulomb laws ^{[source pending]}.

Leonardo da Vinci first studied friction systematically around 1493, observing that friction doubles when the weight doubles and that friction does not depend on the contact area. Guillaume Amontons rediscovered these laws in 1699. Charles-Augustin de Coulomb's 1781 experiments distinguished static from kinetic friction and established the coefficient model still used today ^{[source pending]}.

Modern developments include Stribeck curves (friction coefficient as a function of sliding speed in lubricated contacts), boundary lubrication theory, and atomic-scale friction studies using atomic force microscopy. These reveal that friction at the nanoscale can differ qualitatively from the macroscopic Coulomb law — for example, atomic-scale stick-slip oscillations produce friction even for a single-atom contact.

Non-holonomic control problems

Non-holonomic constraints play a central role in robotics and control theory. A car, a bicycle, and a mobile robot with wheels are all subject to rolling-without-slipping constraints. These systems are underactuated: they have fewer control inputs than configuration variables. The non-holonomic constraint prevents arbitrary instantaneous motion — a car cannot move sideways — but the system can reach any configuration through appropriate manoeuvres (parallel parking being the familiar example).

The controllability of non-holonomic systems is governed by the Chow-Rashevskii theorem: if the constraint distribution $D$ and its iterated Lie brackets span the entire tangent space at each point, then any configuration is reachable. The rolling disk satisfies this condition: the two rolling constraints and their Lie brackets generate all directions in the four-dimensional configuration space $(x,y,\theta ,\varphi )$ ^{[source pending]}.

This connects to optimal control theory, where non-holonomic path planning problems (shortest paths for a car-like robot) are solved using techniques from sub-Riemannian geometry — the constraint distribution defines a sub-Riemannian metric, and geodesics of this metric are the optimal paths.

Synthesis. The classification of constraints into holonomic and non-holonomic is the foundational reason the Lagrangian formalism can eliminate constraint forces entirely in the holonomic case and must retain Lagrange multipliers in the non-holonomic case. The central insight is that holonomic constraints define submanifolds of configuration space, and D'Alembert's principle — the vanishing of virtual work — is the statement that constraint forces are normal to these submanifolds. Putting these together with the Frobenius integrability condition, the passage from Newtonian force-balance to the Euler-Lagrange equations is a single conceptual step enabled by the geometry of constraints. The thermodynamic irreversibility of friction connects this mechanical analysis to the second law, and the non-holonomic control problems of robotics demonstrate that the mathematical structure of constraints has practical consequences far beyond the classical problems that motivated the original classification. This result generalises to infinite-dimensional systems in continuum mechanics 09.07.01 and appears again in the Hamiltonian treatment of constrained systems 09.02.03.

Connections Master

• 09.01.01 Kinematics provides the position, velocity, and acceleration descriptions that the friction and constraint-force equations act upon; this unit adds the forces that real surfaces and geometric restrictions impose on kinematic motion.
• 09.01.02 Newton's laws supply the framework $m\mathbf{a}=\mathbf{F}$ into which friction and constraint forces are inserted; D'Alembert's principle is a reformulation of Newton's second law that eliminates constraint forces by projecting onto unconstrained directions.
• 09.02.01 The action principle replaces Newton's force-balance equations with a variational formulation; this unit's treatment of constraint forces as forces that do no work along permissible displacements provides the physical content that the action principle encodes through the choice of generalised coordinates.
• 09.02.02 The Euler-Lagrange equations are derived from D'Alembert's principle by choosing generalised coordinates adapted to the constraints; this unit provides the Newtonian starting point that the Lagrangian formalism generalises.
• 09.02.03 Constrained Lagrangian mechanics treats holonomic constraints via Lagrange multipliers, extending this unit's treatment to compute constraint forces explicitly rather than eliminating them.
• 09.07.01 Continuum mechanics imposes field-theoretic constraints (incompressibility, rigidity) that are the infinite-dimensional analogues of the holonomic constraints studied here; D'Alembert's principle extends to field theories as the principle of virtual work.
• 11.01.02 The thermodynamic treatment of frictional dissipation connects the mechanical energy balance $\Delta E=-f_{k}d$ to the first law of thermodynamics; the entropy increase from frictional heating is the mechanical manifestation of the second law.

Historical & philosophical context Master

Leonardo da Vinci conducted the first systematic studies of friction around 1493, recording in his notebooks that friction doubles when weight doubles and is independent of contact area. These observations remained unpublished and were independently rediscovered by Guillaume Amontons in 1699, who established what are now called Amontons' laws: friction is proportional to the normal load and independent of the apparent contact area ^{[source pending]}.

Charles-Augustin de Coulomb published his experimental study of friction in 1781 (Theorie des machines simples), distinguishing static from kinetic friction and introducing the coefficient model (${\mu }_s$ and ${\mu }_k$) that bears his name. Coulomb's experiments covered a wide range of materials and conditions, and his model remains the standard engineering approximation more than two centuries later ^{[source pending]}.

The modern science of tribology — from the Greek tribos (rubbing) — emerged from the work of Bowden and Tabor in the mid-twentieth century, who established the asperity-contact theory of friction that explains the Amontons-Coulomb laws microscopically. The 1966 "Jost Report" to the UK government coined the term "tribology" and estimated that friction and wear cost industrial economies approximately 1–2% of GDP ^{[source pending]}.

D'Alembert's principle appears in his Traite de dynamique (1743), where he proposed replacing the Newtonian force balance with a principle of virtual work. The key idea — that equilibrium plus the "lost force" $-m\mathbf{a}$ gives a virtual-work equation — allowed the calculus of variations to be applied to dynamics. This was the direct precursor to Lagrange's Mechanique analytique (1788), which built the entire Lagrangian formalism on D'Alembert's foundation ^{[source pending]}.

The distinction between holonomic and non-holonomic constraints was clarified by Heinrich Hertz in The Principles of Mechanics (1894), who recognised that non-holonomic constraints represent a fundamentally different type of restriction — one that constrains velocities without constraining configurations. The Frobenius theorem (1877, fully developed by Frobenius in 1877 and applied to mechanics by subsequent authors) provides the mathematical criterion for integrability, connecting the physics of constraints to the geometry of distributions on manifolds ^{[source pending]}.

A philosophical point: constraint forces illustrate that not all forces in mechanics arise from force laws. The normal force and string tension are determined by the requirement that constraints be satisfied, not by independent laws of nature. This distinction — between forces given by laws (gravity, electromagnetism) and forces determined by constraints — is a structural feature of Newtonian mechanics that persists in the Lagrangian and Hamiltonian frameworks, where it appears as the distinction between the Lagrangian (encoding force laws) and the constraint equations (restricting configuration space).

Bibliography Master

• da Vinci, L., Codex Madrid I (c. 1493), unpublished notebooks on friction experiments.
• Amontons, G., "De la resistance causee dans les machines," Memoires de l'Academie Royale des Sciences (1699), 206–222.
• Coulomb, C.-A., "Essai sur une application des regles de maximis et minimis a quelques problemes de statique relatifs a l'architecture," Memoires de Mathematique et de Physique 7 (1776/1781), 343–382.
• d'Alembert, J. le R., Traite de dynamique (David l'aine, Paris, 1743).
• Hertz, H., The Principles of Mechanics Presented in a New Form (Macmillan, 1894; Dover reprint, 1956).
• Taylor, J. R., Classical Mechanics (University Science Books, 2005), Ch. 2, Ch. 4, Ch. 9.
• Goldstein, H., Poole, C. P. & Safko, J. L., Classical Mechanics, 3rd ed. (Pearson, 2002), Ch. 1.