Classical mechanics predicts motion by combining a model of matter, geometry, forces or energies, constraints, initial conditions, and a reference frame. Its equations govern particles, rigid bodies, mechanisms, vibrations, fluids, solids, robots, wafer stages, rotating equipment, and many process tools whenever quantum, relativistic, and molecular fluctuations can be coarse-grained. A trustworthy solution must state the system boundary, degrees of freedom, constitutive assumptions, and measurement comparison rather than presenting equations without a physical contract.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif">
<rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">A mechanics model closes motion with forces and constraints</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">Geometry defines degrees of freedom; balance laws determine acceleration</text><rect x="30" y="100" width="165" height="275" rx="12" fill="#161b22" stroke="#58a6ff" stroke-width="2"/><text x="112" y="132" fill="#79c0ff" font-size="14" font-weight="700" text-anchor="middle">System</text><text x="112" y="185" fill="#c9d1d9" font-size="11" text-anchor="middle">particles and bodies</text><text x="112" y="225" fill="#c9d1d9" font-size="11" text-anchor="middle">coordinates and frames</text><text x="112" y="265" fill="#c9d1d9" font-size="11" text-anchor="middle">mass and inertia</text><text x="112" y="325" fill="#8b98a5" font-size="10" text-anchor="middle">q, q̇, material fields</text><path d="M200 235 H250" stroke="#3fb950" stroke-width="4"/><polygon points="250,235 238,228 238,242" fill="#3fb950"/><rect x="255" y="100" width="250" height="275" rx="12" fill="#161b22" stroke="#3fb950" stroke-width="2"/><text x="380" y="132" fill="#7ee787" font-size="14" font-weight="700" text-anchor="middle">Dynamics</text><text x="380" y="195" fill="#e6edf3" font-size="17" text-anchor="middle">d p / dt = ΣF</text><text x="380" y="240" fill="#e6edf3" font-size="17" text-anchor="middle">d L / dt = Στ</text><text x="380" y="285" fill="#e6edf3" font-size="17" text-anchor="middle">dE / dt = power</text><text x="380" y="335" fill="#8b98a5" font-size="10" text-anchor="middle">constraints and constitutive laws close equations</text><path d="M510 235 H560" stroke="#d29922" stroke-width="4"/><polygon points="560,235 548,228 548,242" fill="#d29922"/><rect x="565" y="100" width="165" height="275" rx="12" fill="#161b22" stroke="#d29922" stroke-width="2"/><text x="647" y="132" fill="#e3b341" font-size="14" font-weight="700" text-anchor="middle">Prediction</text><text x="647" y="185" fill="#c9d1d9" font-size="11" text-anchor="middle">trajectory and load</text><text x="647" y="225" fill="#c9d1d9" font-size="11" text-anchor="middle">stress and vibration</text><text x="647" y="265" fill="#c9d1d9" font-size="11" text-anchor="middle">stability and energy</text><text x="647" y="325" fill="#8b98a5" font-size="10" text-anchor="middle">compare through instrument model</text><text x="380" y="425" fill="#c9d1d9" font-size="11" text-anchor="middle">Every arrow carries assumptions that must survive verification and experiment.</text></svg>

Position becomes motion only after a reference frame and clock are chosen. A particle trajectory $r(t)$ gives velocity $v=dr/dt$ and acceleration $a=dv/dt$ in one frame. Coordinates may be Cartesian, polar, cylindrical, generalized, or attached to moving hardware. Vector motion is independent of coordinate notation, but components and derivatives are not. A sensor reports position through calibration, sampling, filtering, and frame alignment, so measured acceleration is not merely a second numerical derivative of noisy displacement.

Degrees of freedom count independent configuration variables after constraints. A free particle in three-dimensional space has three translational degrees of freedom, while a free rigid body has three translational and three rotational degrees. Joints, contacts, symmetry, prescribed motion, and incompressibility reduce or relate them. Redundant coordinates can simplify geometry but require constraint equations and reaction forces. Incorrect degree counting produces singular mass matrices, impossible initial conditions, or missing modes before any numerical solver is involved.

Newton’s first law defines the privileged role of inertial frames. In an inertial frame a body with zero net force maintains constant velocity. Frames moving at constant velocity relative to an inertial frame are also inertial in Galilean mechanics. Accelerating or rotating frames require apparent forces if Newton’s second law is retained in its familiar form. A laboratory fixed to Earth is approximately inertial for many short, local experiments but Coriolis and centrifugal effects matter for precision stages, long trajectories, fluids, and navigation.

Newton’s second law balances momentum rather than merely mass times acceleration. The general particle statement is $F_{ext}=dp/dt$. For constant mass and nonrelativistic velocity it reduces to $F=ma$. Variable-mass systems require a clearly chosen control system and momentum flux; inserting a changing mass into $ma$ alone can be wrong. Force is an interaction model inferred through deformation, fields, momentum exchange, or calibrated transducers. A free-body diagram must include only forces acting on the chosen system.

Newton’s third law depends on how the interacting system is partitioned. Pairwise contact or central forces often appear equal and opposite, supporting cancellation of internal forces in total momentum balance. Electromagnetic systems can store momentum in fields, delayed interactions need broader accounting, and constraint forces may be distributed over contacts. Momentum conservation is the safer system-level statement. When reaction forces do not appear equal in a partial model, inspect omitted field, fluid, support, or actuator momentum before declaring a law violated.

Kinematics constrains possible motion before dynamics selects one. Geometry relates positions, velocities, and accelerations independent of mass and force. Rolling without slip connects translation and rotation; linkage closure relates joint angles; a rigid-body velocity field has translation plus angular velocity cross position. Differentiating constraints can introduce hidden consistency conditions. Numerical drift may violate a position constraint even when velocity constraints appear satisfied, motivating stabilization or coordinate reduction.

Work converts force along motion into energy transfer. Differential work is $dW=F\cdot dr$, so only the force component along displacement contributes. Kinetic energy $T=mv^2/2$ changes by net work for a constant-mass particle. Power is $P=F\cdot v$ plus torque-rotation contributions for extended systems. Forces can do zero work while changing momentum direction, as in ideal centripetal constraint forces. Actuator electrical power, mechanical shaft power, stored energy, dissipation, and heat must not be conflated.

Conservative forces admit a potential energy. If $F=-\nabla V$ in a simply connected configuration region, work between endpoints is path independent and mechanical energy $T+V$ is conserved when the potential is time independent and no nonconservative work enters. Friction, drag, hysteresis, active control, and time-dependent fields generally break that simple conservation. A locally curl-free force may still have global topology issues. Potential zero is arbitrary, while potential differences and gradients are physical.

Linear momentum conservation follows from isolation and translational symmetry. Summing particle momentum cancels suitable internal forces, leaving $dP/dt=F_{external}$. Center-of-mass motion obeys $M a_{CM}=F_{external}$ for constant total mass. Impulse $J=\int Fdt$ changes momentum and handles short impacts without resolving every force detail. In manufacturing equipment, cable forces, air bearings, reaction frames, floor coupling, and moving fluids determine whether the chosen stage is genuinely isolated.

Angular momentum requires an origin and a system boundary. For a particle $L_O=r\times p$, and its rate equals external torque about a fixed inertial origin under standard conditions. For a rigid body, angular momentum is related to angular velocity through an inertia tensor and need not be parallel to it. Choosing a moving point adds transport terms. Gyroscopic reactions, rotor imbalance, wafer spin, and robot motion are easily misread when scalar moment-of-inertia intuition replaces the vector balance.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif">
<rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Free-body diagrams depend on the chosen boundary</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">Internal interactions cancel only after both partners enter the system</text><rect x="70" y="130" width="250" height="220" rx="12" fill="#161b22" stroke="#58a6ff" stroke-width="2"/><text x="195" y="162" fill="#79c0ff" font-size="14" font-weight="700" text-anchor="middle">Single stage boundary</text><rect x="140" y="225" width="110" height="55" rx="7" fill="#1f6feb" stroke="#79c0ff" stroke-width="2"/><text x="195" y="258" fill="#fff" font-size="11" text-anchor="middle">moving stage</text><path d="M195 220 V180" stroke="#3fb950" stroke-width="4"/><polygon points="195,180 188,192 202,192" fill="#3fb950"/><text x="220" y="198" fill="#7ee787" font-size="10">actuator</text><path d="M135 252 H95" stroke="#f85149" stroke-width="4"/><text x="115" y="242" fill="#ff7b72" font-size="10">drag</text><path d="M325 240 H425" stroke="#d29922" stroke-width="4"/><polygon points="425,240 413,233 413,247" fill="#d29922"/><rect x="440" y="130" width="250" height="220" rx="12" fill="#161b22" stroke="#3fb950" stroke-width="2"/><text x="565" y="162" fill="#7ee787" font-size="14" font-weight="700" text-anchor="middle">Stage plus actuator</text><rect x="500" y="225" width="130" height="55" rx="7" fill="#238636" stroke="#7ee787" stroke-width="2"/><text x="565" y="250" fill="#fff" font-size="11" text-anchor="middle">combined system</text><text x="565" y="268" fill="#fff" font-size="9" text-anchor="middle">internal actuator pair cancels</text><path d="M495 252 H465" stroke="#f85149" stroke-width="4"/><path d="M635 252 H665" stroke="#a371f7" stroke-width="4"/><text x="565" y="325" fill="#8b98a5" font-size="10" text-anchor="middle">supports and environment remain external</text><text x="380" y="420" fill="#c9d1d9" font-size="11" text-anchor="middle">Changing the boundary changes the force inventory but not physical motion.</text></svg>

Conservation laws are strongest when derived from symmetry. Noether’s theorem connects continuous symmetries of the action to conserved quantities: time-translation invariance to energy, spatial translation to momentum, and rotation to angular momentum. This formulation clarifies when a conservation law fails because a support, drive, or external field breaks the symmetry. Numerical methods can preserve or drift invariants depending on discretization. Conservation residuals provide verification checks even when exact conservation is physically broken by known inputs.

Constraints separate admissible motion from reaction forces. Holonomic constraints can be written as relations among coordinates and time, while nonholonomic constraints involve velocities and may not integrate to configuration relations. Ideal constraint forces do no virtual work in allowed variations, enabling elimination through generalized coordinates or Lagrange multipliers. Frictional contact, backlash, compliance, and actuator saturation are not ideal constraints. Their forces require constitutive or complementarity models and can create nonsmooth transitions.

Generalized coordinates should follow configuration geometry. Coordinates $q_i$ may be angles, lengths, modal amplitudes, or any independent parameters of configuration. Generalized velocity need not be a physical Cartesian velocity, and generalized force is defined through virtual work $\delta W=\sum_i Q_i\delta q_i$. A smart coordinate choice embeds constraints and reduces equations; a poor one introduces singularities or unnecessary multipliers. Coordinate charts can fail globally for rotations, so quaternions or multiple charts may be preferable.

D’Alembert’s principle converts dynamics into virtual-work balance. Adding inertial forces to applied forces yields zero virtual work for admissible variations, forming a bridge from Newtonian vector balance to analytical mechanics. Reaction forces of ideal constraints disappear from the reduced equations because their virtual work is zero. The principle does not erase physical reactions; they can be recovered through multipliers or balance equations. Using it with dissipative or nonideal constraints requires explicit generalized forces.

Hamilton’s principle selects the path with stationary action. For Lagrangian $L(q,\dot q,t)=T-V$ in a conservative system, the physical path makes $S=\int Ldt$ stationary under endpoint-fixed variations. Stationary does not always mean minimum. The Euler–Lagrange equations $d(\partial L/\partial\dot q_i)/dt-\partial L/\partial q_i=Q_i^{nc}$ generate equations of motion. The variational form handles coordinates and constraints elegantly, but it relies on a correct kinetic energy, potential, and nonconservative-force model.

The Lagrangian formulation exposes coupled mechanics systematically. For multiple bodies, write position and orientation as functions of generalized coordinates, build total kinetic and potential energy, add dissipation or applied generalized forces, and differentiate. Mass matrices, gyroscopic terms, stiffness, and forcing emerge without drawing every internal reaction. Symbolic expressions can become large and hide sign errors. Verify by comparing Newton–Euler balances, checking energy, testing simple configurations, and confirming that the mass matrix is symmetric positive definite for independent coordinates.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif"><rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Three equivalent views organize the same motion</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">Choose the representation that makes forces, constraints, or phase geometry simplest</text><rect x="35" y="125" width="205" height="230" rx="12" fill="#161b22" stroke="#58a6ff" stroke-width="2"/><text x="138" y="158" fill="#79c0ff" font-size="15" font-weight="700" text-anchor="middle">Newton–Euler</text><text x="138" y="215" fill="#e6edf3" font-size="17" text-anchor="middle">F = dp/dt</text><text x="138" y="250" fill="#c9d1d9" font-size="11" text-anchor="middle">vectors, forces, reactions</text><text x="138" y="285" fill="#8b98a5" font-size="10" text-anchor="middle">best for free bodies</text><rect x="278" y="125" width="205" height="230" rx="12" fill="#161b22" stroke="#3fb950" stroke-width="2"/><text x="380" y="158" fill="#7ee787" font-size="15" font-weight="700" text-anchor="middle">Lagrange</text><text x="380" y="215" fill="#e6edf3" font-size="16" text-anchor="middle">dL/dq̇ − ∂L/∂q = Q</text><text x="380" y="250" fill="#c9d1d9" font-size="11" text-anchor="middle">energy, coordinates, constraints</text><text x="380" y="285" fill="#8b98a5" font-size="10" text-anchor="middle">best for coupled mechanisms</text><rect x="520" y="125" width="205" height="230" rx="12" fill="#161b22" stroke="#a371f7" stroke-width="2"/><text x="623" y="158" fill="#d2a8ff" font-size="15" font-weight="700" text-anchor="middle">Hamilton</text><text x="623" y="215" fill="#e6edf3" font-size="16" text-anchor="middle">q̇ = ∂H/∂p</text><text x="623" y="240" fill="#e6edf3" font-size="16" text-anchor="middle">ṗ = −∂H/∂q</text><text x="623" y="275" fill="#c9d1d9" font-size="11" text-anchor="middle">phase space and invariants</text><path d="M240 240 H275" stroke="#d29922" stroke-width="3"/><path d="M485 240 H517" stroke="#d29922" stroke-width="3"/><text x="380" y="420" fill="#c9d1d9" font-size="11" text-anchor="middle">Equivalence is a powerful cross-check; convenience is not a change of physics.</text></svg>

Hamiltonian mechanics evolves states in phase space. Canonical momentum is $p_i=\partial L/\partial\dot q_i$, and a regular Legendre transform gives $H(q,p,t)=\sum_i p_i\dot q_i-L$. Hamilton’s equations are $\dot q_i=\partial H/\partial p_i$ and $\dot p_i=-\partial H/\partial q_i$. In many conservative systems $H$ equals total energy, but this is not automatic for time-dependent coordinates or unusual velocity-dependent potentials. The paired first-order equations reveal geometry and support structure-preserving integration.

Poisson brackets encode evolution and canonical structure. For observables $A(q,p)$ and $B(q,p)$, the Poisson bracket $\{A,B\}=\sum_i(\partial A/\partial q_i\,\partial B/\partial p_i-\partial A/\partial p_i\,\partial B/\partial q_i)$. Evolution obeys $dA/dt=\{A,H\}+\partial A/\partial t$. A quantity with zero bracket with the Hamiltonian is conserved when it has no explicit time dependence. Canonical transformations preserve these brackets, allowing coordinates chosen around invariants, actions, or perturbations.

Symplectic geometry constrains faithful numerical evolution. Hamiltonian flow preserves phase-space volume by Liouville’s theorem and preserves a symplectic two-form more strongly. A generic high-order time integrator may have small local error yet create secular energy drift over long runs. Symplectic schemes usually keep a nearby modified Hamiltonian and bounded energy error, which is valuable for orbital, molecular, and undamped vibration simulations. Dissipative and controlled systems require extensions rather than pretending their flow is Hamiltonian.

Central forces reduce three-dimensional motion to an effective radial problem. A force depending only on distance and pointing along the radius conserves angular momentum, fixing motion to a plane. The radial coordinate experiences the physical potential plus a centrifugal effective term. Kepler orbits, Rutherford scattering, and simplified bearing or particle trajectories share this reduction. Real equipment adds noncentral contact, drag, fields, and control, so symmetry-derived invariants should be tested rather than assumed.

Rigid-body orientation is more subtle than particle position. A rigid body preserves distances among its material points, while its attitude belongs to the rotation group rather than ordinary vector space. Euler angles are intuitive but possess coordinate singularities; rotation matrices are redundant but geometric; unit quaternions are compact but require normalization and identify opposite signs. Angular velocity is the instantaneous generator of rotation and depends on whether its components are expressed in body or spatial axes.

The inertia tensor connects mass distribution to rotational response. About a selected point, $I=\int(r^2\mathbf{1}-rr^T)dm$ is symmetric and has orthogonal principal axes. Rotational kinetic energy is $T_r=\omega^TI\omega/2$, and angular momentum is $L=I\omega$ when both use compatible components about a fixed point or center of mass. Products of inertia matter away from principal axes. A payload moved a few centimeters can alter robot or stage dynamics substantially because inertia weights distance squared.

Euler’s rigid-body equations include gyroscopic coupling. In body principal axes, $I_1\dot\omega_1+(I_3-I_2)\omega_2\omega_3=\tau_1$ with cyclic counterparts. The cross terms arise because the basis rotates even if angular momentum is inertially fixed. They explain precession, nutation, reaction torque, and intermediate-axis instability. Rotor and wafer-spindle models need imbalance, bearing stiffness, damping, and drive torque in addition to ideal rigid-body terms.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif"><rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Rigid-body response couples geometry, inertia, and torque</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">Angular momentum and angular velocity align only on a principal axis</text><ellipse cx="365" cy="235" rx="165" ry="80" transform="rotate(-18 365 235)" fill="#1f6feb" fill-opacity="0.35" stroke="#58a6ff" stroke-width="3"/><circle cx="365" cy="235" r="7" fill="#fff"/><path d="M365 235 L590 145" stroke="#3fb950" stroke-width="4"/><polygon points="590,145 575,145 581,158" fill="#3fb950"/><text x="602" y="144" fill="#7ee787" font-size="14">ω</text><path d="M365 235 L540 90" stroke="#a371f7" stroke-width="4"/><polygon points="540,90 526,95 537,105" fill="#a371f7"/><text x="550" y="88" fill="#d2a8ff" font-size="14">L = Iω</text><path d="M365 235 L250 95" stroke="#d29922" stroke-width="4"/><polygon points="250,95 252,111 264,101" fill="#d29922"/><text x="215" y="88" fill="#e3b341" font-size="14">τ = dL/dt</text><path d="M365 235 L365 390" stroke="#8b98a5" stroke-width="2" stroke-dasharray="6 5"/><text x="375" y="405" fill="#8b98a5" font-size="11">principal axis</text><path d="M100 320 C140 380 220 400 285 365" fill="none" stroke="#f85149" stroke-width="3"/><polygon points="285,365 272,364 278,376" fill="#f85149"/><text x="120" y="405" fill="#ff7b72" font-size="11">precession under applied torque</text><text x="380" y="445" fill="#c9d1d9" font-size="11" text-anchor="middle">Mass location controls inertia; inertia controls acceleration and reaction load.</text></svg>

Gyroscopic effects redirect torque across axes. A rapidly spinning rotor resists changes to its angular-momentum direction, so frame rotation generates reactions proportional to spin and precession rates. These effects can stabilize, destabilize, or couple otherwise separate axes. In vacuum pumps, spindles, flywheels, and scanning stages, gyro terms may shift resonances and control margins. Direction signs should come from a consistent frame derivation, not a memorized right-hand-rule sketch.

Impact is governed by impulse, contact geometry, and energy loss. Integrating momentum balance across a short collision relates impulse to the velocity jump. A coefficient of restitution closes a simple normal-impact model but is an empirical aggregate, not a universal material constant; it changes with speed, shape, temperature, and deformation. Oblique contact also needs friction and possibly spin. Compliant contact models resolve finite force histories, while rigid impact models accept discontinuous velocity.

Friction is a constitutive law with regimes, memory, and uncertainty. Coulomb friction distinguishes sticking from sliding and bounds tangential force during stick, but real contacts exhibit presliding displacement, Stribeck behavior, rate dependence, adhesion, wear, and thermal drift. Static and kinetic coefficients alone cannot predict nanometer stages or precision robot joints. Friction identification must match surface preparation, normal load, velocity range, environment, and measurement bandwidth.

The harmonic oscillator is the local language of stable mechanical systems. Near a stable equilibrium, smooth potential energy is approximately quadratic, giving $m\ddot x+kx=0$ and natural frequency $\omega_n=\sqrt{k/m}$. Many nonlinear systems therefore look harmonic at small amplitude. The approximation fails when clearance, geometric nonlinearity, material nonlinearity, or large rotation changes stiffness. Natural frequency is a property of the model boundary and constraints, not of a component in isolation.

Damping controls decay and resonance without being a single physical mechanism. Viscous damping gives $m\ddot x+c\dot x+kx=f(t)$ and damping ratio $\zeta=c/(2\sqrt{km})$. Under-, critical-, and over-damped responses describe mathematical regimes. Real energy loss may arise from fluid shear, material hysteresis, joints, eddy currents, or active control and need not be proportional to velocity. A fitted viscous coefficient is often local to frequency and amplitude.

Forced response distinguishes resonance from instability. Sinusoidal forcing yields a frequency-response function whose amplitude and phase depend on frequency, damping, and observation point. Near a lightly damped mode, dynamic amplification can be large while remaining bounded. Instability instead means perturbations grow in the unforced or feedback-coupled dynamics. Swept-sine tests, impulse responses, and operating spectra answer different questions and must use sufficient settling and resolution.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif"><rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Resonance is shaped by mode, damping, and measurement point</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">A response peak is not automatically an unstable system</text><path d="M80 380 H700 M80 380 V95" stroke="#8b98a5" stroke-width="2"/><text x="390" y="430" fill="#c9d1d9" font-size="12" text-anchor="middle">forcing frequency</text><text x="26" y="230" fill="#c9d1d9" font-size="12" transform="rotate(-90 26 230)" text-anchor="middle">response magnitude</text><path d="M85 355 C180 350 250 330 310 210 C340 145 365 105 390 175 C420 280 465 335 700 350" fill="none" stroke="#f85149" stroke-width="4"/><path d="M85 355 C210 350 270 330 330 275 C365 245 410 280 450 320 C510 348 590 352 700 354" fill="none" stroke="#3fb950" stroke-width="4"/><line x1="365" y1="92" x2="365" y2="380" stroke="#d29922" stroke-width="2" stroke-dasharray="6 5"/><text x="365" y="82" fill="#e3b341" font-size="11" text-anchor="middle">modal frequency</text><line x1="510" y1="120" x2="560" y2="120" stroke="#f85149" stroke-width="4"/><text x="570" y="124" fill="#ff7b72" font-size="11">light damping</text><line x1="510" y1="150" x2="560" y2="150" stroke="#3fb950" stroke-width="4"/><text x="570" y="154" fill="#7ee787" font-size="11">higher damping</text><rect x="115" y="115" width="145" height="75" rx="8" fill="#161b22" stroke="#58a6ff"/><text x="188" y="143" fill="#79c0ff" font-size="11" text-anchor="middle">input → structure → sensor</text><text x="188" y="169" fill="#8b98a5" font-size="10" text-anchor="middle">all three affect the curve</text><text x="380" y="455" fill="#c9d1d9" font-size="11" text-anchor="middle">Test configuration is part of every frequency-response result.</text></svg>

Multiple degrees of freedom create mode shapes as well as frequencies. Linearized structural dynamics has $M\ddot q+C\dot q+Kq=f$. With suitable symmetric $M$ and $K$, the undamped eigenproblem $K\phi=\omega^2M\phi$ yields mass-orthogonal modes. A natural frequency without its mode shape is incomplete because participation depends on actuator and sensor locations. Boundary conditions, payload, cables, joints, and fluid loading can shift both.

Modal superposition compresses dynamics when retained modes span the response. Expressing displacement as $q=\Phi\eta$ can decouple an undamped linear model and reduce computation. Truncation misses high-frequency flexibility, residual stiffness, and local stress. Closely spaced modes, nonproportional damping, strong nonlinearities, or changing configuration weaken simple modal models. Reduced-order validation must cover the inputs, outputs, bandwidth, and operating configurations for which it will be used.

Wave motion transports disturbance through distributed inertia and stiffness. Strings, rods, plates, acoustic volumes, and elastic solids possess fields rather than a finite list of exact coordinates. Wave speed follows constitutive and inertial properties; boundaries reflect, transmit, scatter, and form standing waves. Dispersion means different frequencies propagate at different phase or group speeds. A finite mechanical assembly approximates a continuum with increasingly dense modes as frequency rises.

Stability asks what nearby trajectories do, not whether one trajectory looks quiet. Linearizing $\dot x=f(x)$ around an equilibrium gives a Jacobian whose eigenvalues classify local behavior under standard conditions. Negative real parts indicate asymptotic decay for continuous-time linear systems; imaginary eigenvalues require nonlinear or energy analysis. Lyapunov functions can establish stability without solving trajectories. Transient nonnormal amplification can still be large even when all eigenvalues predict eventual decay.

Nonlinearity makes frequency and superposition amplitude dependent. Geometric stiffening, softening springs, backlash, saturation, friction, impact, and nonlinear fluid forces generate harmonics, subharmonics, jumps, internal resonance, and multiple attractors. A Duffing oscillator already exhibits amplitude-dependent resonance and hysteresis. Linearization remains valuable locally, but extrapolation across load or clearance changes can be dangerous. Continuation and bifurcation analysis map solution branches more reliably than isolated time runs.

Chaos is deterministic sensitivity rather than random forcing. Nonlinear systems with enough state dimension can have trajectories that diverge exponentially from nearly identical initial conditions while remaining bounded. Poincaré sections, Lyapunov exponents, and invariant sets distinguish chaos from broadband noise. Long-term point prediction becomes impossible beyond a horizon, but statistical or geometric predictions may remain useful. Numerical error, model uncertainty, and measurement noise must be separated from intrinsic sensitivity.

Coriolis and centrifugal terms arise from differentiating in rotating frames. If a frame rotates with angular velocity $\Omega$, acceleration includes relative, Euler, Coriolis $2\Omega\times v_{rel}$, and centrifugal $\Omega\times(\Omega\times r)$ terms. They are bookkeeping for using a noninertial coordinate system, not new pair interactions. Their scale can be negligible in a benchtop translation yet decisive in rotating-fluid, spindle, planetary, and precision metrology problems.

Continuum mechanics replaces discrete particles with fields after a scale test. Density, velocity, displacement, temperature, and stress are treated as smooth functions when the observation scale is large relative to molecular spacing and representative heterogeneity. The continuum hypothesis works extraordinarily well down to many microdevices, but rarefied gas, atomically thin films, granular matter, and nanoscale interfaces may need slip, stochastic, kinetic, or discrete models. A Knudsen or size-effect estimate should precede automatic use of bulk laws.

Material and spatial descriptions answer different tracking questions. A Lagrangian material description follows each material label through the motion $x=\chi(X,t)$, while an Eulerian spatial description observes fields at fixed locations. Solids often favor material coordinates and fluids spatial coordinates, though either is possible. The material derivative $D()/Dt=\partial()/\partial t+v\cdot\nabla()$ connects them and explains why a steady velocity field can still accelerate a moving parcel.

Deformation separates translation and rotation from genuine shape change. The deformation gradient $F=\partial x/\partial X$ maps material line elements, and its determinant $J$ gives local volume ratio. Polar decomposition $F=RU$ separates rotation from stretch. Small-strain theory uses $\varepsilon=(\nabla u+\nabla u^T)/2$ when displacement gradients are small; large rotations invalidate it even if local stretches are modest. Strain is dimensionless geometry, not a force or material property.

Stress represents internal force transmission across imagined surfaces. Cauchy’s stress tensor maps a surface normal to traction $t=\sigma n$. Balance of angular momentum makes ordinary Cauchy stress symmetric when body couples are absent. Normal and shear components change with plane orientation, while principal stresses are tensor invariants. Wafer bow, film delamination, chuck contact, and package failure depend on stress distributions and interface tractions rather than a single average value.

Balance laws constrain every constitutive model. Local mass balance, linear momentum $\rho Dv/Dt=\nabla\cdot\sigma+\rho b$, angular momentum, and energy apply across materials within their assumptions. They do not specify how stress depends on deformation, rate, history, or temperature. That closure is a constitutive law. A simulation can solve its discrete equations accurately and still be physically wrong because its material closure or boundary flux is wrong.

Elasticity stores recoverable deformation energy. Linear isotropic elasticity relates stress and strain through Young’s modulus and Poisson ratio, equivalently two independent elastic constants. Hooke’s law is a local small-strain approximation, not a statement that all materials are linear springs. Crystals are anisotropic, thin films can be textured, porous layers are effective media, and temperature or prestress can change tangent stiffness. Energy positivity imposes constraints on admissible constants.

Plasticity makes deformation history part of the state. When a yield criterion is reached, irreversible strain evolves through a flow rule and hardening law. Yield strength is not fracture strength, and unloading can be elastic around a plastically changed configuration. Residual stress and springback therefore persist after external load removal. Thin metal films, bonded stacks, contacts, and thermal cycling may require anisotropic, rate-dependent, or cyclic plasticity rather than a single bilinear curve.

Viscoelasticity couples memory, time scale, and temperature. Springs and dashpots produce idealized relaxation and creep, while hereditary integrals or internal variables represent broader spectra. A material can appear glassy at high frequency and compliant at low frequency. Time-temperature superposition may shift response across frequency but must be validated. Polymers, adhesives, seals, damping layers, and photoresist can transmit slowly evolving loads that an elastic model misses.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif"><rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Continuum prediction needs balance plus constitutive closure</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">Conservation is universal; material response and boundaries are model choices</text><rect x="35" y="105" width="200" height="260" rx="12" fill="#161b22" stroke="#58a6ff" stroke-width="2"/><text x="135" y="138" fill="#79c0ff" font-size="14" font-weight="700" text-anchor="middle">Kinematics</text><text x="135" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">motion and deformation</text><text x="135" y="230" fill="#e6edf3" font-size="16" text-anchor="middle">x = χ(X,t)</text><text x="135" y="270" fill="#e6edf3" font-size="16" text-anchor="middle">F = ∂x/∂X</text><text x="135" y="320" fill="#8b98a5" font-size="10" text-anchor="middle">geometry only</text><rect x="280" y="105" width="200" height="260" rx="12" fill="#161b22" stroke="#3fb950" stroke-width="2"/><text x="380" y="138" fill="#7ee787" font-size="14" font-weight="700" text-anchor="middle">Balance laws</text><text x="380" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">mass, momentum, energy</text><text x="380" y="235" fill="#e6edf3" font-size="15" text-anchor="middle">ρDv/Dt = ∇·σ + ρb</text><text x="380" y="290" fill="#8b98a5" font-size="10" text-anchor="middle">necessary but not closed</text><rect x="525" y="105" width="200" height="260" rx="12" fill="#161b22" stroke="#d29922" stroke-width="2"/><text x="625" y="138" fill="#e3b341" font-size="14" font-weight="700" text-anchor="middle">Constitutive law</text><text x="625" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">stress, flux, memory</text><text x="625" y="230" fill="#e6edf3" font-size="14" text-anchor="middle">σ = C : ε + …</text><text x="625" y="270" fill="#c9d1d9" font-size="11" text-anchor="middle">elastic, plastic, viscous</text><text x="625" y="320" fill="#8b98a5" font-size="10" text-anchor="middle">identified and validated</text><path d="M235 235 H277 M480 235 H522" stroke="#a371f7" stroke-width="4"/><text x="380" y="415" fill="#c9d1d9" font-size="11" text-anchor="middle">Boundary and initial conditions complete the physical problem.</text></svg>

Fracture requires an energy or crack-tip criterion beyond peak stress. A crack concentrates fields, making nominal stress inadequate. Linear elastic fracture mechanics relates stress intensity and energy release rate to crack growth when its assumptions hold. Ductile damage, fatigue, interfaces, and small structures may require cohesive zones or other models. Defect population and environment make failure probabilistic, so validation specimens should reproduce geometry, processing, and loading mode.

Fluid mechanics applies momentum balance with fluid constitutive behavior. For a Newtonian fluid, viscous stress is proportional to rate of deformation; combined with mass and momentum balance this yields the Navier–Stokes equations. Incompressibility means material volume preservation, not necessarily constant pressure or zero velocity divergence in every approximate setting. Non-Newtonian slurries, polymers, and process chemicals need viscosity models that depend on shear rate, history, or microstructure.

Reynolds number compares inertia with viscosity. $Re=\rho UL/\mu$ helps classify dynamically similar flows, but its characteristic velocity and length must match the phenomenon. Low Reynolds number suppresses inertial memory and often makes flow reversible; high Reynolds number enables separation and turbulence but does not guarantee either. Microchannels can have low $Re$ yet meaningful entrance, rarefaction, electrokinetic, or surface effects.

Boundary layers concentrate gradients near surfaces. At high Reynolds number, viscosity may be weak in the bulk but essential in a thin no-slip layer that determines drag, separation, heat transfer, and contamination transport. A boundary-layer approximation follows from scale analysis, not from setting viscosity to zero everywhere. Surface roughness, pressure gradients, suction, and transition alter its behavior. Mesh resolution must capture wall-normal gradients or use a validated wall model.

Pressure is a constraint field in incompressible flow. It adjusts so that momentum evolution remains compatible with incompressibility and boundary conditions. Pressure is not generally prescribed independently at every boundary, and only differences matter in many formulations. Projection algorithms solve a Poisson equation to enforce divergence-free velocity. Pressure loads on chamber walls, wafers, seals, and particles can couple fluid prediction back to structural deformation.

Turbulence is a multiscale transport problem rather than extra random viscosity. Fluctuating eddies transfer momentum and energy across scales until viscosity dissipates it. Direct numerical simulation resolves all relevant scales at immense cost; large-eddy simulation filters smaller scales; Reynolds-averaged models close statistics. Each predicts different observables and carries closure uncertainty. A colorful instantaneous flow image is not validation of pressure drop, mixing, residence time, or particle deposition.

Dimensional analysis reveals controlling groups before computation. Buckingham’s Pi theorem expresses a dimensionally consistent relationship through independent nondimensional groups. Reynolds, Mach, Knudsen, Strouhal, Froude, and Cauchy numbers compare mechanisms. Scaling a chamber, robot, or test coupon preserves behavior only if the governing groups and boundary conditions remain similar. Unit checking catches many errors but cannot prove that the chosen physical variables are complete.

Coupled fields exchange power through shared variables. Fluid–structure interaction transfers traction and velocity; thermoelasticity transfers temperature, strain, and heat; electromechanics transfers fields, force, and current. One-way coupling is justified only when feedback is demonstrably small. Partitioned solvers can suffer added-mass or time-lag instability, while monolithic solvers cost more but enforce coupling strongly. Interface interpolation should conserve force, moment, and energy to appropriate accuracy.

The finite element method converts weak balance into algebra. Multiplying a governing equation by test functions and integrating by parts produces a weak or virtual-work form that reduces derivative requirements and exposes natural boundary conditions. The domain is partitioned into elements with interpolation functions, leading to mass, damping, stiffness, and load arrays. Rayleigh and Ritz energy ideas foreshadow this structure. Element choice, quadrature, mesh quality, and constraints determine whether the discrete space can represent the physics.

Mesh convergence must target a quantity of interest. Displacement may converge while peak contact stress, film curvature, or eigenfrequency remains inaccurate. Refinement should compare a defined output across systematically smaller elements, with singularities interpreted rather than chased to infinity. Polynomial-order refinement, adaptive error estimates, and local submodels can be more efficient than uniform refinement. A converged discretization proves only that the chosen equations were solved consistently.

<svg viewBox="0 0 760 470" xmlns="http://www.w3.org/2000/svg" font-family="-apple-system,BlinkMacSystemFont,Segoe UI,Roboto,sans-serif"><rect x="0" y="0" width="760" height="470" rx="18" fill="#0d1117"/><text x="380" y="34" fill="#e6edf3" font-size="21" font-weight="700" text-anchor="middle">Verification and validation answer different questions</text><text x="380" y="58" fill="#8b98a5" font-size="12" text-anchor="middle">A numerically exact answer to the wrong model remains wrong</text><rect x="45" y="105" width="195" height="250" rx="12" fill="#161b22" stroke="#58a6ff" stroke-width="2"/><text x="143" y="138" fill="#79c0ff" font-size="14" font-weight="700" text-anchor="middle">Physical model</text><text x="143" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">boundary and assumptions</text><text x="143" y="230" fill="#c9d1d9" font-size="11" text-anchor="middle">constitutive parameters</text><text x="143" y="270" fill="#c9d1d9" font-size="11" text-anchor="middle">initial and loading data</text><rect x="283" y="105" width="195" height="250" rx="12" fill="#161b22" stroke="#3fb950" stroke-width="2"/><text x="380" y="138" fill="#7ee787" font-size="14" font-weight="700" text-anchor="middle">Numerical model</text><text x="380" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">mesh and timestep</text><text x="380" y="230" fill="#c9d1d9" font-size="11" text-anchor="middle">solver and tolerances</text><text x="380" y="270" fill="#c9d1d9" font-size="11" text-anchor="middle">discrete conservation</text><rect x="520" y="105" width="195" height="250" rx="12" fill="#161b22" stroke="#d29922" stroke-width="2"/><text x="618" y="138" fill="#e3b341" font-size="14" font-weight="700" text-anchor="middle">Experiment</text><text x="618" y="190" fill="#c9d1d9" font-size="11" text-anchor="middle">instrument transfer</text><text x="618" y="230" fill="#c9d1d9" font-size="11" text-anchor="middle">uncertainty and repeats</text><text x="618" y="270" fill="#c9d1d9" font-size="11" text-anchor="middle">matched configuration</text><path d="M240 220 H280" stroke="#a371f7" stroke-width="4"/><text x="260" y="205" fill="#d2a8ff" font-size="10" text-anchor="middle">verify</text><path d="M480 220 H517" stroke="#f85149" stroke-width="4"/><text x="499" y="205" fill="#ff7b72" font-size="10" text-anchor="middle">validate</text><text x="380" y="410" fill="#c9d1d9" font-size="11" text-anchor="middle">Calibration estimates parameters; validation tests predictive adequacy.</text></svg>

Time integration trades accuracy, stability, and preserved structure. Explicit methods are simple and scalable but face timestep limits set by the fastest retained dynamics. Implicit methods permit larger stable steps for many stiff linear systems but require nonlinear solves and can hide unresolved transients. Newmark-family, Runge–Kutta, variational, and symplectic methods have different dissipation and invariant behavior. Stability does not imply accuracy; timestep convergence must use the output and spectrum of interest.

Constraint algorithms must prevent both drift and artificial work. Lagrange multipliers impose constraints and return reactions but create saddle-point systems. Penalty methods approximate constraints with high stiffness, introducing conditioning and timestep problems. Coordinate elimination is efficient when topology is simple; projection and stabilization correct drift. Contact adds changing active sets and complementarity. Monitor position, velocity, reaction, and energy consistency rather than accepting a solver’s convergence flag alone.

Model verification asks whether equations were solved correctly. Analytical limits, manufactured solutions, independent implementations, conservation residuals, order-of-accuracy studies, and mesh or timestep refinement expose coding and discretization errors. Verification uses known mathematical truth where possible. Comparing to experiment cannot isolate a numerical bug because parameter and model discrepancies coexist. Unit tests for transforms, inertia, elements, and load signs complement system benchmarks.

Model validation asks whether the equations represent reality well enough. Experiments should challenge intended predictions across the operating envelope, with inputs and outputs passed through the same geometry, filtering, timing, and uncertainty definitions. Tuning and testing on the same data exaggerates credibility. Calibration estimates parameters; validation evaluates held-out predictive performance. Validation is conditional on a use, range, and tolerance rather than a permanent badge.

Uncertainty separates variability from lack of knowledge. Manufacturing tolerances, material scatter, disturbance realizations, and environmental variation are aleatory descriptions, while uncertain model form or poorly measured parameters are epistemic. Probability distributions should reflect evidence, not convenience. Sensitivity analysis identifies dominant contributors, and uncertainty propagation turns inputs into prediction intervals. A narrow deterministic curve is not more precise when its assumptions are uncertain.

Experimental mechanics measures through a transfer function. Accelerometers, laser interferometers, strain gauges, load cells, pressure sensors, and cameras have bandwidth, noise, mounting effects, cross-axis sensitivity, and calibration uncertainty. Sampling can alias high-frequency motion; differentiation amplifies noise; filtering changes amplitude and phase. The model observable must match what the instrument actually returns. Sensor mass or cable stiffness can perturb small structures enough to invalidate the nominal boundary.

A wafer stage is a closed-loop mechanics system, not a free mass. Motors apply forces through structures whose flexible modes, air bearings, cables, metrology frames, and floor supports shape motion. Feedforward handles known inertia and friction; feedback rejects error within bandwidth but can excite modes or sensor resonances. Nanometer settling depends on modal damping, thermal drift, force ripple, quantization, and coordinate transforms. Stage performance must be evaluated at the wafer-relevant point, not only the encoder.

Vibration isolation works by frequency-dependent transmissibility. Below its resonance an isolator follows base motion; near resonance it can amplify; sufficiently above resonance it attenuates. More damping reduces the resonant peak but can worsen high-frequency transmission. Passive isolators cannot suppress quasi-static floor tilt, cable force, or internally generated reactions, while active systems add sensors, actuators, and control noise. Payload center of mass and rotational modes matter alongside vertical translation.

Robot handling combines multibody dynamics with compliant contact. Joint inertia varies with configuration, and rapid moves create Coriolis, centrifugal, gravity, and actuator-load coupling. End-effector placement also depends on link flexibility, backlash, calibration, and thermal expansion. Wafer pickup adds Bernoulli or vacuum forces, edge contact, slip risk, and fragile-body vibration. Trajectory shaping can reduce residual excitation without simply lowering peak speed.

Rotating process hardware couples imbalance, bearings, and fluid forces. A mass eccentricity produces synchronous forcing that grows with spin speed squared. Bearings contribute speed- and temperature-dependent stiffness and damping; seals and fluids add cross-coupled forces; gyro terms split forward and backward whirl. Campbell diagrams track modes against rotational speed. Passing a critical speed safely requires transient and stability analysis, not only a static balance specification.

Film stress converts nanometer layers into wafer-scale curvature. Intrinsic growth stress, thermal-expansion mismatch, phase change, and gradients create membrane loads and bending. Stoney-type relations infer average thin-film biaxial stress from curvature under restrictive thickness, uniformity, and substrate assumptions. Patterning redistributes stress, multilayers interact, and anisotropic wafers complicate inference. Curvature measurement is therefore an inverse mechanics problem with model and metrology uncertainty.

Chucking and contact mechanics govern wafer shape and particle risk. Electrostatic, vacuum, mechanical, or edge-grip chucks impose distributed pressure and constraint. Wafer thickness variation, backside particles, surface roughness, and chuck flatness convert force into local bending and contact stress. More holding force can reduce slip yet print defects or increase bow. Contact compliance and friction must be coupled to thermal and handling loads when overlay or breakage margins are tight.

Gas delivery and chamber flow connect mechanics to process uniformity. Pressure-driven viscous flow sets residence time, species transport, wall shear, and particle trajectories. At low pressure, increasing Knudsen number invalidates no-slip continuum assumptions and eventually requires kinetic descriptions. Showerhead jets, pumping asymmetry, wafer rotation, buoyancy, and thermal gradients break simple symmetry. Flow validation should target measured pressure, conductance, velocity proxies, or deposition outcomes rather than streamline appearance.

Plasma-facing mechanics includes momentum flux and evolving surfaces. Ion and neutral bombardment transmit pressure and can sputter, heat, charge, or erode components. Particle trajectories in electromagnetic fields remain classical over many equipment scales, but their distribution and collisions require plasma or kinetic closures. Erosion changes geometry and hence fields and flow over maintenance cycles. Treating the wall as rigid and permanent can miss drift in matching, contamination, or uniformity.

MEMS inhabit classical mechanics with strong surface and scale effects. Beams, plates, proof masses, resonators, and switches follow elasticity and dynamics, while electrostatic forces, squeeze-film damping, adhesion, residual stress, and fabrication variation dominate behavior. Pull-in is a nonlinear instability rather than simple force balance. Thermal noise may set a measurement floor even though the device motion is classically modeled. Continuum validity and size-dependent properties must be checked at the smallest dimensions.

Thermomechanics converts temperature fields into deformation and stress. Free thermal strain is approximately $\alpha\Delta T$ locally, but constraints turn incompatible expansion into stress. Spatial gradients bend wafers, stages, optics, and chamber parts; transient heat flow creates lag and drift. Multimaterial assemblies need temperature-dependent properties, interfaces, and assembly history. A uniform-temperature compensation cannot correct local gradients or metrology-frame distortion it does not observe.

Classical mechanics has clear domain limits without becoming obsolete. Relativity replaces Galilean kinematics near light speed or in precision spacetime problems. Quantum mechanics governs microscopic states, quantization, tunneling, and measurement. Statistical mechanics connects microscopic populations to thermodynamic and transport laws. Classical equations nevertheless remain the effective description of most equipment motion, continuum fields, orbital motion, and mean trajectories when scale separation and decoherence justify them.

The same physical system can be represented at different levels, but each representation carries a different state, closure, and validation burden.

Model choice should follow the decision and dominant scales. Begin with the required output, tolerance, bandwidth, geometry, and operating range. Estimate dimensionless ratios and characteristic times, then choose particle, rigid-body, flexible-body, continuum, fluid, or coupled-field detail. Add complexity only when a neglected mechanism can change the decision. A simple model with quantified error can be more useful than an elaborate model whose parameters cannot be measured.

flowchart TD
    A[Define system boundary, decision, and observable] --> B[Choose reference frame and degrees of freedom]
    B --> C{Can bodies be treated as rigid?}
    C -->|Yes| D[Use particle or Newton–Euler multibody balance]
    C -->|No| E{Solid, fluid, or coupled fields?}
    E -->|Solid| F[Choose elasticity, plasticity, viscoelasticity, contact, or fracture]
    E -->|Fluid| G[Check Reynolds, Mach, and Knudsen regimes]
    E -->|Coupled| H[Define conservative interface variables and feedback]
    D --> I[State forces, constraints, initial conditions, and controls]
    F --> I
    G --> I
    H --> I
    I --> J[Verify units, balances, limits, mesh, and timestep]
    J --> K[Validate matched observables with uncertainty]
    K --> L{Prediction adequate for intended use?}
    L -->|No| M[Revise boundary, closure, parameters, or resolution]
    M --> B
    L -->|Yes| N[Use within validated envelope and monitor drift]

A reliable workflow closes a traceable loop from assumptions to evidence. Document why the system boundary excludes each interaction, how coordinates map to hardware, which conservation laws are exact or broken, where parameters came from, and what numerical studies establish convergence. Compare predictions with an independent measurement through its instrument model and uncertainty. When disagreement appears, test boundary, input, closure, discretization, and measurement hypotheses separately instead of tuning the nearest coefficient.

Historical formalisms are complementary tools rather than competing truths. Galileo clarified inertial motion; Newton organized force and momentum; Euler extended rotation and continua; D’Alembert and Lagrange used virtual work and generalized coordinates; Hamilton exposed phase-space structure; Poisson encoded canonical algebra; Cauchy formalized stress; Navier and Stokes closed viscous momentum; Reynolds exposed flow scaling; Hooke characterized elasticity; Noether connected symmetry to conservation; Rayleigh and Ritz made energy approximation practical. Their ideas survive because each exposes a reusable structure.

Classical intuition improves when conservation replaces formula hunting. Ask what crosses the boundary, what is stored, what symmetry removes a dependence, and what constitutive rule closes the balance. Force, impulse, work, torque, stress, and pressure are related transfers but are not interchangeable. A trajectory is the consequence of the complete model, not the starting explanation. Read classical mechanics through a system-boundary-and-conservation lens rather than a force-formula-and-trajectory lens.