Approximations make physics problems solvable, but a shorter equation is useful only when you can defend the assumptions behind it. This practical checklist shows how to identify the controlling dimensionless parameter, estimate the error, test limiting cases, and decide when a numerical comparison is necessary. It includes a reference table and worked examples from mechanics, electromagnetism, quantum physics, and thermodynamics.
Overview
An approximation replaces a complicated model with a simpler one. The goal is not to make a problem “easier” at any cost; it is to remove effects that are small compared with the accuracy you need. A good approximation therefore has three parts:
- A small or large dimensionless parameter: for example, an angle in radians, a speed ratio v/c, or a potential-to-kinetic-energy ratio.
- A stated regime: the conditions under which the neglected terms are genuinely smaller than the retained terms.
- An error check: an estimate, comparison with the exact expression, or numerical control calculation.
Dimensionless parameters matter because quantities with units cannot be compared directly. Before using an approximation, form ratios such as x/L, v/c, ka, U/E, or Kn. The ratio should describe the physical competition that determines whether a term is important.
There is no universal cutoff for “small.” A parameter of 0.01 is often safely small for a first-order estimate, while 0.2 may be acceptable for a rough result but not for precision work. Your required accuracy, measurement uncertainty, and proximity to a resonance or critical point all matter.
| Approximation | Mathematical form | Control parameter and usual regime | Leading error estimate |
|---|---|---|---|
| Small angle | sin θ ≈ θ; tan θ ≈ θ; cos θ ≈ 1 − θ²/2 | |θ| ≪ 1, with θ in radians | sin θ: about θ²/6 relative to θ; cos θ: omitted term about θ⁴/24 |
| Linearization | f(x₀+δx) ≈ f(x₀)+f′(x₀)δx | |δx|/L ≪ 1, where L is the function’s variation scale | Typically second order in δx; bounded by roughly |f″δx²/2| |
| Nonrelativistic limit | γ ≈ 1 + v²/(2c²) | β = v/c ≪ 1 | Relativistic kinetic energy correction beyond mv²/2 is about 3mv⁴/(8c²) |
| Weak perturbation | E ≈ E₀ + λE₁ | λ|E₁| is much smaller than level spacings or E₀ | Often second order in λ, but fails near degeneracy or resonance |
| Mean-field | Replace interactions by an average field | Many weakly correlated constituents or large coordination number | Correlation corrections are often of order 1/z or controlled by fluctuations |
| Born approximation | Scattering amplitude from the incident wave | Weak scattering; roughly |U|/E small over the interaction region | Next scattering events introduce higher powers of the interaction strength |
| Thermodynamic limit | Extensive quantities scale with N | N ≫ 1; boundary/N effects small | Surface corrections often scale like N−1/d for a d-dimensional sample |
How to estimate
Use this workflow before substituting numbers into a simplified formula.
- Write the full model first. Record the exact equation, governing law, or series expansion. This prevents an approximation from becoming an unstated assumption.
- Identify the competing effects. Ask what is being neglected: higher powers of an angle, magnetic forces, relativistic corrections, correlations, multiple scattering, or finite-size effects.
- Construct a dimensionless ratio. Divide the small effect by the dominant effect. Examples include θ, v/c, qV/E, ka, and U/ΔE.
- Determine the order of the first omitted term. If the retained result is first order and the next term is proportional to ε², the absolute error is usually of order ε² times the relevant scale. Do not confuse this estimate with a guaranteed bound.
- Compare with the required tolerance. A 1% approximation is not adequate if the measured quantity is known to 0.1%, but it may be more than adequate for a sketch or order-of-magnitude estimate.
- Check dimensions, limits, and conservation laws. A simplified result must still have correct units and sensible behavior as a parameter tends to zero or becomes large. For a broader plausibility method, see How to Check Whether a Physics Result Is Plausible.
- Run a numerical-control check when needed. Evaluate the exact and approximate expressions at representative parameter values. Near a resonance, turning point, phase transition, cancellation, or boundary layer, numerical comparison is especially important.
Dimensional analysis can expose a bad approximation, but it cannot prove that one is safe. A dimensionless ratio may be small while its coefficient is large, or several individually small terms may accumulate over a long time or distance.
Inputs and assumptions
Write down the inputs that control validity, not just the inputs needed to calculate the final answer. Include the characteristic length, time, energy, speed, temperature, particle number, or interaction strength. Also record the intended output accuracy.
For linearization, choose a local scale. If f(x) varies on length scale L, then the relevant parameter is approximately |δx|/L, not simply |δx|. For perturbation theory, compare the perturbing matrix element with the energy separation it couples to. For wave and scattering problems, identify the wavelength relative to the object size through ka = 2πa/λ.
State exclusions explicitly. Small-angle formulas require radians. The nonrelativistic approximation is not justified merely because an object is “ordinary”; its speed must be small compared with c. Mean-field methods can miss strong correlations, and the Born approximation can fail for a strongly reflecting or resonant potential. If the approximation changes the qualitative prediction, such as whether a trajectory is bound, treat it more cautiously than a small numerical correction.
Keep a control calculation reproducible. Save the parameter values, the exact expression or numerical method, the approximate expression, and the relative difference:
relative difference = |exact − approximate| / |exact|.
When the exact result is close to zero, use an absolute difference or compare both results with the uncertainty in the measurement.
Worked examples
1. Small-angle pendulum
For a pendulum of length L, the exact angular equation is θ̈ + (g/L)sin θ = 0. Replacing sin θ with θ gives simple harmonic motion and period T = 2π√(L/g). The control parameter is the maximum angle θmax in radians. At 0.10 rad, the first neglected term in sin θ is roughly θ³/6, or 0.017% of θ. At 0.5 rad, the relative error is about 2%, so the simple period is no longer a precision formula. For a larger amplitude, use the exact nonlinear equation or an amplitude-corrected period.
2. Linearizing a spring force
Suppose a force near equilibrium is F(x) = −kx − αx³. If the displacement is small, the cubic term is negligible when |αx²/k| ≪ 1. The linear model F ≈ −kx then has a fractional force correction of approximately |αx²/k|. Measure the largest expected displacement, not the displacement at one convenient instant. If the ratio reaches 0.1, the nonlinear correction can be about 10% and should be included in a controlled comparison.
3. Nonrelativistic kinetic energy
The exact kinetic energy is K = (γ−1)mc². Expanding in β = v/c gives K ≈ mv²/2, with the next correction 3mv⁴/(8c²). The correction relative to the Newtonian term is approximately 3β²/4. At v = 0.01c, this is about 0.0075%, usually negligible. At v = 0.3c, it is about 6.8%, so relativistic kinematics should be used. Momentum and energy should be treated consistently; mixing relativistic momentum with Newtonian energy can create a hidden approximation error.
4. Weak perturbation in quantum mechanics
Consider an unperturbed level separated from neighboring levels by ΔE, with a perturbing matrix element V. A first-order energy shift is credible when |V|/ΔE ≪ 1. If the ratio is 0.01, second-order effects are often of order 10−4 in the relevant scale. If the ratio approaches 0.3, or if two levels are nearly degenerate, ordinary nondegenerate perturbation theory is unsafe. Use degenerate perturbation theory or diagonalize the coupled subspace instead.
5. Born approximation for scattering
In a weak-potential scattering problem, the Born approximation calculates the scattered wave from the incident wave and ignores repeated scattering. A practical check compares the potential energy scale U with the incident kinetic energy E, while also considering the interaction range and wavelength. If |U|/E is small and no resonance is present, the approximation may be reasonable. A numerical solution of the radial Schrödinger equation is a useful control when the ratio is not clearly small or when the potential supports a near-bound state.
6. Ideal-gas and thermodynamic-limit reasoning
For a dilute gas, the ideal-gas approximation neglects interactions when the mean separation is large compared with the interaction range and the interaction energy is small compared with kBT. For quantum gases, the phase-space density also matters; wave-packet overlap can invalidate classical statistics. Separately, treating a finite system as macroscopic assumes boundary effects and fluctuations are small. Check particle number, density, temperature, and container dimensions rather than assuming that “large” automatically means thermodynamic.
7. Quasistatic approximation in thermodynamics
A process can be treated as quasistatic when the system relaxes to near-equilibrium much faster than the external parameters change. A useful control ratio is the relaxation time τrelax divided by the driving time τdrive. If τrelax/τdrive is much less than one, equilibrium formulas may apply along the path. If the ratio is not small, temperature, pressure, or chemical potential may not be well defined locally, and a nonequilibrium model is needed.
When to recalculate
Revisit an approximation whenever an input crosses its validity range or the purpose of the calculation changes. Recalculate if the speed, angle, amplitude, interaction strength, system size, wavelength, temperature, or required precision changes substantially. Also reassess after discovering a resonance, near-degeneracy, turning point, cancellation, sharp gradient, or long-time accumulation of error.
For homework and exam work, state the parameter and the expected error in one sentence: “Since θmax = 0.08 rad, the neglected small-angle correction is of order θ², so the linear model is adequate to about 0.1%.” For laboratory analysis, compare the approximation error with measurement uncertainty and include the control calculation in your report; the physics data analysis tutorial provides a useful framework. For simulations, test at least one parameter value inside the proposed regime and one near its boundary.
Finally, keep a short validity record: model used, dimensionless control parameter, estimated error, required tolerance, and exact or numerical check. That habit turns approximation from a guess into a defensible modeling decision.